The Routledge Handbook of Philosophy and Implicit Cognition [1 ed.] 9780367857189, 9781003014584

Citation preview

THE ROUTLEDGE HANDBOOK OF PHILOSOPHY AND IMPLICIT COGNITION

Humans think of ourselves as acting according to reasons that we can typically articulate and acknowledge, though we may be reluctant to do so. Yet some of our actions do not fit this mold – they seem to arise from motives and thoughts that appear outside of our control and our self-awareness. Rather than treating such cases as outliers, theorists now treat significant parts of the mind as operating implicitly or ‘behind the scenes’. Mental faculties like reasoning, language, and memory seem to involve this sort of implicit cognition, and many of the structures we use to understand one another seem infused with biases, perceptions, and stereotypes that have implicit features. The Routledge Handbook of Philosophy and Implicit Cognition is an outstanding guide and reference source to this important topic. Composed of more than thirty chapters by an international team of contributors, the Handbook is divided into eight clear parts: • Defining Features? Identifying Implicitness Among Cognate Notions • The Nature and Limits of Implicit Processing • Ways of Perceiving, Knowing, Believing • Language • Agency and Control • Social Cognition • Memory • Learning and Reasoning The Routledge Handbook of Philosophy and Implicit Cognition is essential reading for students and researchers in philosophy of psychology, moral psychology, and philosophy of mind, and will also be of interest to those in related disciplines such as psychology, neuroscience, and linguistics. J. Robert Thompson is Associate Professor of Philosophy at Mississippi State University, USA. He studies implicit phenomena as they arise within the fields of developmental psychology, psycholinguistics, and the philosophy of language.

ROUTLEDGE HANDBOOKS IN PHILOSOPHY

Routledge Handbooks in Philosophy are state-of-the-art surveys of emerging, newly refreshed, and important fields in philosophy, providing accessible yet thorough assessments of key problems, themes, thinkers, and recent developments in research. All chapters for each volume are specially commissioned, and written by leading scholars in the field. Carefully edited and organized, Routledge Handbooks in Philosophy provide indispensable reference tools for students and researchers seeking a comprehensive overview of new and exciting topics in philosophy. They are also valuable teaching resources as accompaniments to textbooks, anthologies, and research-orientated publications. Also available: THE ROUTLEDGE HANDBOOK OF PHILOSOPHY OF FRIENDSHIP Edited by Diane Jeske THE ROUTLEDGE HANDBOOK OF INDIAN BUDDHIST PHILOSOPHY Edited by William Edelglass, Pierre-Julien Harter and Sara McClintock THE ROUTLEDGE HANDBOOK OF BODILY AWARENESS Edited by Adrian J.T. Alsmith and Matthew R. Longo THE ROUTLEDGE HANDBOOK OF AUTONOMY Edited by Ben Colburn THE ROUTLEDGE HANDBOOK OF THE PHILOSOPHY AND PSYCHOLOGY OF FORGIVENESS Edited by Glen Pettigrove and Robert Enright For more information about this series, please visit: www.routledge.com/RoutledgeHandbooks-in-Philosophy/book-series/RHP

THE ROUTLEDGE HANDBOOK OF PHILOSOPHY AND IMPLICIT COGNITION

Edited by J. Robert Thompson

Cover image: © Getty Images First published 2023 by Routledge 2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN and by Routledge 605 Third Avenue, New York, NY 10158 Routledge is an imprint of the Taylor & Francis Group, an informa business © 2023 selection and editorial matter J. Robert Thompson; individual chapters, the contributors The right of J. Robert Thompson to be identified as the author of the editorial material, and of the authors for their individual chapters, has been asserted in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988. All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any information storage or retrieval system, without permission in writing from the publishers. Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library Library of Congress Cataloging-in-Publication Data Names: Thompson, J. Robert, editor. Title: The Routledge handbook of philosophy of implicit cognition / edited by J. Robert Thompson. Description: 1 Edition. | New York, NY : Routledge, 2023. | Series: Routledge handbooks in philosophy | Includes bibliographical references and index. Identifiers: LCCN 2022036668 (print) | LCCN 2022036669 (ebook) | ISBN 9780367857189 (hardback) | ISBN 9781032423708 (paperback) | ISBN 9781003014584 (ebook) Subjects: LCSH: Cognition. | Implicit memory. | Implicit learning. | Psychology—Philosophy. Classification: LCC BF311 .R6678 2023 (print) | LCC BF311 (ebook) | DDC 153—dc23/eng/20221017 LC record available at https://lccn.loc.gov/2022036668 LC ebook record available at https://lccn.loc.gov/2022036669 ISBN: 978-0-367-85718-9 (hbk) ISBN: 978-1-032-42370-8 (pbk) ISBN: 978-1-003-01458-4 (ebk) DOI: 10.4324/9781003014584 Typeset in Bembo by Apex CoVantage, LLC

CONTENTS

Acknowledgementsix Notes on Contributors x

Introduction: In Search of the Implicit J. Robert Thompson

PART 1

1

Defining Features? Identifying Implicitness Among Cognate Notions

31

  1 Implicit Mental Representation William Ramsey

33

  2 Measuring and Modeling Implicit Cognition Samuel A.W. Klein and Jeffrey W. Sherman

44

  3 Implicit Cognition and Unconscious Mentality Tim Crane and J. Robert Thompson

56

  4 Implicit Cognition in Relation to the Conceptual/Nonconceptual Distinction69 José Luis Bermúdez and Arnon Cahen   5 The Fragmented Mind: Personal and Subpersonal Approaches to Implicit Mental States Zoe Drayson   6 The Levels Metaphor and the Implicit/Explicit Distinction Judith Carlisle v

79 90

Contents PART 2

The Nature and Limits of Implicit Processing

103

  7 Implicit Cognition, Dual Process Theory, and Moral Judgment Charlie Blunden, Paul Rehren, and Hanno Sauer

105

  8 Implicit Bias and Processing Ema Sullivan-Bissett

115

  9 Predictive Processing, Implicit and Explicit Paweł Gładziejewski

127

10 Cognitive Penetration and Implicit Cognition Lucas Battich and Ophelia Deroy

144

PART 3

Ways of Perceiving, Knowing, Believing

153

11 Helmholtz on Unconscious Inference in Experience Lydia Patton

155

12 Husserl on Habit, Horizons, and Background Dermot Moran

168

13 Polanyi and Tacit Knowledge Stephen Turner

182

14 Tacit Knowledge Tim Thornton

191

15 Collective and Distributed Knowledge: Studies of Expertise and Experience Harry Collins

202

16 Implicit Beliefs Joseph Bendaña

215

17 Implicit Self-Knowledge Kristina Musholt

226

vi

Contents PART 4

Language235 18 Chomsky, Cognizing, and Tacit Knowledge John Collins

237

19 Language Processing: Making It Implicit? David Pereplyotchik

247

20 Implicit Knowledge in Pragmatic Inference Chris Cummins and Albertyna Paciorek

259

PART 5

Agency and Control

269

21 Implicit Mechanisms in Action and in the Experience of Agency Sofia Bonicalzi

271

22 Implicit Cognition and Addiction: Selected Recent Findings and Theory Reinout W. Wiers and Alan W. Stacy

282

23 Phenomenology, Psychopathology, and Pre-Reflective Experience Anthony Vincent Fernandez

300

PART 6

Social Cognition

311

24 Race and the Implicit Aspects of Embodied Social Interaction Jasper St. Bernard and Shaun Gallagher

313

25 Implicit Social Cognition Shannon Spaulding

324

26 The Development of Implicit Theory of Mind Hannes Rakoczy

336

PART 7

Memory351 27 Implicit Memory Sarah K. Robins

353

vii

Contents

28 Memory During Failures of Recall: Information That Is Forgotten Is Not Gone Anne M. Cleary PART 8

362

Learning and Reasoning

375

29 Implicit Reasoning Thomas Sturm and Uljana Feest

377

30 Implicit Knowledge of (Parts of ) Logic, and How to Make It Explicit Keith Stenning and Michiel van Lambalgen

389

31 What Is It Like to Learn Implicitly? Arnaud Destrebecqz

402

Index416

viii

ACKNOWLEDGEMENTS

I would like to thank Adam Johnson and the staff at Routledge for their assistance and patience in bringing this volume into print. The pressures of the COVID-19 pandemic delayed and complicated the volume in numerous ways, some of which are described in the Introduction. In addition to some referees who chose to remain anonymous, I  would like to thank the following individuals who reviewed material for the Handbook: Nicholas Allot, Kristin Andrews, Jacob Berger, John Bickle, Dan Burston, Scott Edgar, Susan Erck, Bart Geurts, Dick Grandy, Walter Gulick, Liz Irvine, Alistair Isaac, Greg Johnson, Koray Karaca, Alex Kiefer, Peter Koenigs, Anastasia Kozyreva, Salvador Mascarenhas, Doug McConnell, Alex Miller, Bart Moffatt, Myrto Mylopoulos, Joe Neisser, Albert Newen, Manuel Rodeiro, Schirmer dos Santos, Bennett Schwartz, John Schwenkler, Shawn Simpson, Joulia Smortchkova, Mason Westfall, and Evan Westra. I also want to acknowledge the contributors to the volume who weighed in at various points in its development. Steven Gross provided me with very helpful feedback early in the process of editing this volume, and I want to express my appreciation to him for his help with this project and for his kind assistance at other points as I emerged into the profession. Not everyone was able to respond to the challenges of the pandemic with a sense of responsibility and compassion. So, I am deeply grateful for those who were able to, especially my wife and fellow academic, Devon Brenner. JRT

ix

NOTES ON CONTRIBUTORS

Lucas Battich is a postdoctoral fellow at the Institut Jean Nicod, École Normale Supérieure in Paris. He works on perception and social cognition, combining tools from philosophy and experimental psychology. Joseph Bendaña is a Ph.D. student at the CUNY Graduate Center. His research focuses on cognitive architecture. José Luis Bermúdez is Professor of Philosophy and Samuel Rhea Gammon Professor of Liberal Arts at Texas A&M University. Charlie Blunden is a Ph.D. student in philosophy at Utrecht University. He is interested in topics in empirically informed political and moral philosophy and has published on business ethics, the ethics of mandatory vaccination, and moral progress. Sofia Bonicalzi is Assistant Professor of Moral Philosophy at Roma Tre University and Associate Researcher in the Cognition, Value and Behavior research group (Ludwig Maximilian University of Munich). She works on moral philosophy, philosophy of action, philosophy of cognitive neuroscience, and moral psychology. Arnon Cahen is a research associate and teaching fellow in the Brain and Cognitive Sciences Department at the Hebrew University of Jerusalem, Israel. Judith Carlisle is a Ph.D. student in the Philosophy, Neuroscience, and Psychology program at Washington University in Saint Louis. Her research sits at the intersections of philosophy of science and philosophy of mind. Anne M. Cleary is Professor of Cognitive Psychology at Colorado State University. She studies aspects of human memory and metacognition with a particular focus on how memory can be exhibited during retrieval failure and how this ability can be capitalized on in real-world settings. Harry Collins is Distinguished Research Professor in the School of Social Sciences at Cardiff University and a Fellow of the British Academy. He studies the nature of tacit knowledge and the sociology of gravitational wave physics.

x

Notes on Contributors

John Collins is Professor of Philosophy at the University of East Anglia. He mainly researches in the philosophy of language, with an accent toward linguistics, philosophy of mind, and the concept of truth. Tim Crane is Professor of Philosophy and Pro-Rector for Teaching and Learning at the Central European University in Vienna. He studies the nature of consciousness, intentionality, and the place of mind in nature. Chris Cummins is Reader in Linguistics and English Language at the University of Edinburgh, UK. His research interests include implicature, presupposition, and number, with a particular focus on how quantitative information can be conveyed so as to be accurately understood. Ophelia Deroy holds the Chair for Philosophy of Mind and Neuroscience at the LudwigMaximilian University of Munich. She also directs the Center for Research in Experimental Aesthetics, CREATE, at the Institute of Philosophy, University of London. Arnaud Destrebecqz is an experimental psychologist at the Centre for Research in Cognitive Neuroscience at the Université libre de Bruxelles. His primary area of expertise is in the basic mechanisms of learning. He has studied this issue in depth in adults as well as children and infants using behavioral and physiological measures. Zoe Drayson is Associate Professor of Philosophy at the University of California, Davis. Her research explores the epistemic, ontological, and semantic commitments of different explanatory frameworks in the mind sciences. Uljana Feest is Professor of Philosophy at the Leibniz University of Hannover, where she holds the chair for Philosophy of Social Science and Social Philosophy. She specializes in the philosophy and history of the behavioral sciences. Anthony Vincent Fernandez is Assistant Professor of Applied Philosophy at the Danish Institute for Advanced Study and the Department of Sports Science and Clinical Biomechanics, University of Southern Denmark. His research is on applications of philosophical phenomenology across a variety of disciplines, especially areas of health care, such as psychiatry and nursing. Shaun Gallagher is Lillian and Morrie Moss Professor of Excellence in Philosophy at the University of Memphis and Professorial Fellow at the School of Liberal Arts, University of Wollongong (AU). His areas of research include phenomenology and the cognitive sciences, especially topics related to embodiment, self, agency and intersubjectivity, hermeneutics, and the philosophy of time. Paweł Gładziejewski is a philosopher of cognitive science working at the Department of Cognitive Science at Nicolaus Copernicus University in Toruń, Poland. He is mainly interested in questions regarding the explanatory role of mental representations and the epistemological consequences of Bayesian models of perception. Samuel A.W. Klein is a Ph.D. candidate in psychology at the University of California, Davis. He examines how contexts impact the cognitive mechanisms underlying attitudes and social judgment. Dermot Moran holds the Joseph Chair in Catholic Philosophy at Boston College. His research includes medieval Christian philosophy (especially John Scottus Eriugena) and contemporary European philosophy (especially phenomenology, Edmund Husserl, and Maurice Merleau-Ponty).

xi

Notes on Contributors

Kristina Musholt is Professor of Cognitive Anthropology in the Department of Philosophy at the University of Leipzig. She works in philosophy of mind and philosophy of cognitive science, and her research focuses on self-consciousness, social cognition, normativity, and the relation between conceptual and nonconceptual forms of representation. Albertyna Paciorek is Assistant Professor in Psycholinguistics at the University of Warsaw, Poland. Her research focuses on implicit cognition, in particular implicit learning in second language acquisition. Lydia Patton is Professor of Philosophy at Virginia Tech. She is a historian and philosopher of science, whose research has appeared in journals including Studies in History and Philosophy of Modern Physics, Synthese, The Monist, History and Philosophy of Logic, and Historia Mathematica. David Pereplyotchik is Associate Professor in the Philosophy Department at Kent State University. He is the author of Psychosyntax: The Nature of Grammar and Its Place in the Mind (Springer 2017) and various articles on consciousness, language, and mental representation. Hannes Rakoczy is Professor of Cognitive Development at the University of Göttingen. His research interests lie in developmental and comparative cognitive science. He investigates how cognitive capacities develop in human ontogeny, how human thought compares to the ways nonhuman primates think, and what this implies about the foundations and the evolution of cognition. William Ramsey is Professor of Philosophy at the University of Nevada-Las Vegas. He works in the philosophy of cognitive science, focusing on a variety of topics including naturalistic accounts of mental representation. His main contribution in this area is his book Representation Reconsidered (Cambridge University Press 2007). Paul Rehren is a Ph.D. student in philosophy at Utrecht University. He is interested in various topics at the intersection of (moral) philosophy and psychology. He has published on moral framing effects, the psychology of punishment, moral progress, and the role of moral psychology for moral philosophy. Sarah K. Robins is Associate Professor of Philosophy at the University of Kansas. Her research is focused primarily on memory, through which she writes on a range of issues in philosophy of mind, psychology, and neuroscience. Hanno Sauer is Associate Professor of Philosophy at Utrecht University. His main research interests include metaethics, moral psychology, and the nature of moral progress. His most recent books are Debunking Arguments in Ethics (Cambridge University Press 2018) and Moral Thinking, Fast and Slow (Routledge 2018). Jeffrey W. Sherman is Professor of Psychology at the University of California, Davis. His research investigates the cognitive processes underlying social perception, attitude formation, and attitude change. Shannon Spaulding is Associate Professor of Philosophy at Oklahoma State University. Her current research projects are on motivated empathy, the relationship between trust and perspective taking, and cognitive structure of implicit bias. Alan W. Stacy is Professor in the School of Community and Global Health, Claremont Graduate University.

xii

Notes on Contributors

Jasper St. Bernard is a Ph.D. candidate in philosophy at The University of Memphis. His primary research interests are in social and political philosophy, particularly 18th- and 19thcentury African American philosophy. He is currently working on a dissertation examining the anti-lynching work of Ida B. Wells-Barnett. The dissertation will be looking through the lens of Wells-Barnett’s work at the intersection of race, gender, myth, and politics and the law. Keith Stenning is Emeritus Professor at Edinburgh University. Thomas Sturm is ICREA Research Professor at the Universitat Autònoma de Barcelona, Spain. He specializes in the philosophy of Kant, the history and philosophy of the cognitive sciences, and theories of rationality. He is the author of Kant und die Wissenschaften vom Menschen (Mentis, 2009), co-author of How Reason Almost Lost Its Mind: The Strange Career of Cold War Rationality (Chicago, 2013), and articles in Studies in History and Philosophy of Science, Synthese, and Kant-Studien. Ema Sullivan-Bissett is Reader in Philosophy at the University of Birmingham, UK. She works primarily on issues in the philosophy of mind and psychology, in particular, belief, delusion, and implicit bias. J. Robert Thompson is Associate Professor of Philosophy at Mississippi State University. He studies implicit phenomena as they arise within the fields of developmental psychology, psycholinguistics, and the philosophy of language. Tim Thornton is Professor of Philosophy and Mental Health in the School of Nursing at the University of Central Lancashire in the UK. His books include Wittgenstein on Language and Thought (EUP 1998), John McDowell (Routlege 2019), and the co-authored Tacit Knowledge (Acumen 2013). Stephen Turner is Distinguished University Professor in the Department of Philosophy, University of South Florida. His books related to this topic include Sociological Explanation as Translation; The Social Theory of Practices: Tradition, Tacit Knowledge, and Presuppositions; Brains/Practices/ Relativism: Social Theory after Cognitive Science; Understanding the Tacit; and Cognitive Science and the Social: A Primer. Michiel van Lambalgen is Emeritus Professor of Logic and Cognitive Science at the University of Amsterdam. He believes that the tools of modern mathematical logic have much to offer to cognitive science. Reinout W. Wiers is Professor of Developmental Psychopathology at the University of Amsterdam. He is co-chair of the interdisciplinary Centre for Urban Mental Health at the same university.

xiii

INTRODUCTION In Search of the Implicit J. Robert Thompson

1. Introduction Life might be simpler if the roots of human behavior were out in the open for all to see, though this openness would preclude a level of privacy that most people find valuable. There are several reasons why people find it helpful to shield these roots from others (and perhaps from themselves) – perhaps the roots reflect poorly upon that individual, or perhaps awareness of them would be upsetting to their audience or would prevent the individual from attaining some goal. The nature and extent of the shielding also varies. Some shield some bits of information but reveal others. Some readily reveal information, but it is only tangentially related to the actual roots. Others project information that is an outright misrepresentation of the roots. The Handbook is not directed toward cases where an agent purposefully misleads an audience as to the roots of her behavior. Instead, it will focus on cases where there is no intentional misrepresentation, but the actual roots of behavior nevertheless fail to be overt or open to others, and occasionally to the agent herself. The following chapters largely assume that human behavior can be given an etiological analysis. That is, it is useful to speak of behaviors as having causes and there is some sense in which these causes can be identified and form part of an explanation of why the behavior occurred. In this introduction, ‘roots’ is introduced as a semi-technical term. The roots of human behavior comprise a set of psychological causes that are picked out by a particular mentalistic idiom that includes terms for things like the following states: intentions, perceptions, goals, decisions, beliefs, desires, reasons; and the following processes: perceiving, remembering, reasoning, deliberating, thinking. This idiom makes up our folk psychology. The chapters in this volume also assume that a scientific psychology’s idiom is going to have some similarity to the folk idiom. Psychology textbooks commonly define the field accordingly as “the scientific study of behavior and mental processes” (Gleitman et al. 2010: 1) and include chapters on thinking, decision-making, rationality, learning, memory, emotion, and perception, inter alia. Some theorists have argued that the folk terms may not smoothly integrate into a scientific framework (Churchland 1981; Stich 1983). In fact, there is a sense in which much of this volume offers similar reasons to doubt a smooth integration, in that it describes states and processes that appear to have mostly escaped mention within the folk idiom. Nevertheless, the phenomena explored in this volume do arise at least in part within the folk idiom and offer DOI: 10.4324/9781003014584-1 1

J. Robert Thompson

theorists the opportunity to determine the extent to which these folk phenomena undermine or expand the account developed by scientific psychology. For the purposes of what follows, the roots of human behavior will be treated as a bundle of scientifically clarified and regimented mental states and processes, and it will remain to be seen how a thorough taxonomy of these roots fits with past and current folk psychological and scientific psychological accounts. Rather than treating these issues as belonging to any one discipline or subdiscipline, the fields that study these issues will be described generically as ‘cognitive science’. To summarize, human behavior has roots. The behavior and its roots, however, are not always overt1 or open to others, and this takes place not only in situations where an agent purposefully shields the behavior and/or its roots. Strikingly, the behavior and its roots may not be available to the agent who manifests them, and typical models for measuring and describing these elements don’t seem viable. The Handbook explores a range of phenomena that escape these models and can be grouped under the label of ‘implicit cognition’. Upcoming sections will gradually introduce the features of such implicitness.

2. Cases Humans seem especially interested in conversing about intentional actions. Questions like the following are commonly posed and answered: Why did you do that? (or often Why didn’t you do that?) Why did she do that? Why did I do that? These conversations persist at least partly because humans have found them to be valuable – they scratch some inquisitive itch. One reason humans find these to be valuable is that the conversations, pitched in the folk psychological idiom, address the pertinent questions and terminate the conversations. Eventually, sincere answers preclude further exploration. Such conversations overtly specify the roots of some behavior in themselves and in others. But humans also converse in this folk psychological idiom in order to reveal their mental states and processes to others even when not prompted to make sense of some past, current, or future behavior. It seems important for humans to report these states and processes to others, typically through a linguistic medium.2 Assuming that the behaviors in question arise neither as mere reflexes nor ex nihilo, some roots for them exist. Yet, the roots can reveal themselves in very different ways. From the perspective of the one undergoing the behavior, some appear to result from more deliberative roots, whereas others appear to result from roots that are more intuitive. Such appearances might be deceiving, but many deliberative activities have roots that are naturally identifiable and reportable, whereas many of the intuitive ones have roots that are not naturally identifiable and reportable, and any attempt to uncover these seems rather strained and relatively uninformative. There are cases where the thought processes involved in this sort of deliberation are put on display. In extreme versions of such overtness, items can be externalized and directly endorsed by the subject. One such case could be writing out the premises of a syllogism and then its conclusion, accompanied by a verbal recounting of the acceptability of the inference drawn and verbal assent to the statements therein. Another might be Charles Darwin’s infamous recounting of his decision to marry, with lists of pros and cons associated with his options and a direct

2

Introduction

weighing of his preferences. In such externalized cases, issues about identification and reporting don’t seem to arise. But, how much of human thought is captured by these overtly external forms? How much of this self-reporting gets at the actual roots? Deliberative processes lend themselves to externalization and self-report, but there are good reasons to doubt that all such processing can be externalized and self-reported. Moreover, most intuitive processes don’t lend themselves to this treatment. The Handbook explores areas that go uncaptured in self-reporting, some more deliberative, some more intuitive. This terrain remains relatively uncharted, but what have come to be known as discordant cases provide a fine introduction to these phenomena. In such cases, there is reason to expect two sets of roots are present at the same time. Even if we set aside cases of purposeful deception, it is clear that humans commonly say one thing while doing another.3 Assuming that both the verbal behavior and the nonverbal behavior have mental roots, the roots reported do not appear to be responsible for the nonverbal behavior. Hence, the real roots of the nonverbal behavior are not overt. Such cases can be analyzed with colorful examples. Gendler (2008) describes the behavior of someone who ventures onto a clear platform that extends over the Grand Canyon as a particularly salient case of discordance. Such a person can have a range of self-reported justified beliefs about the safety of such an apparatus (suppose one thoroughly researches the site, is an engineer, etc.) but nevertheless tremble with fear while walking out on the platform. This person’s verbal behavior and their nonverbal behavior are discordant. This mismatch between verbal and nonverbal behavior can also be found in cases involving confabulation. These cases arise quite frequently in various clinical syndromes (Hirstein 2005), but also among the general population. In a clinical setting, individuals who have lost their sight due to excessive alcohol consumption (Korsakoff’s syndrome) nevertheless report having just performed sighted activities, and do so with no indications of deception. In a nonclinical setting, researchers (Nisbett and Wilson 1977) studying decision-making found that when forced to decide among conspicuously undifferentiable products, shoppers expressed a notable position effect (a preference for items to their right). This went unnoticed by the shoppers, who when asked to explain their choice of the right-most item, generated any number of answers that were not clearly tethered to their circumstances. Moreover, when asked directly about a possible effect of the position of the article, virtually all subjects denied it, usually with a worried glance at the interviewer suggesting that they felt either that they had misunderstood the question or were dealing with a madman. (Nisbett and Wilson 1977: 244) Cases of such confabulation, then, involve a discordance between the actual roots of their behavior and the roots the confabulators report themselves as having, that is, they involve an honestly generated self-report that misrepresents the actual roots of the agent’s behavior. Studies in child development also provide insight into discordant behaviors by devising opportunities for different streams of behavior to emerge under controlled circumstances. Suppose that researchers are trying to determine whether a child has acquired a critical capacity. Experiments can be designed to differentiate between children who have and who have not acquired that capacity by how they answer a particular question, say about where an experimenter will look for an object. Those who have acquired the critical capacity will say that the experimenter will look in a box. Those who have not will say that he will look in a basket. Experiments in this tradition have found that under certain circumstances, a child verbally

3

J. Robert Thompson

reports that the experimenter will look in the basket, but before giving that answer, they direct their gaze to the box. Hence, their nonverbal behavior indicates that they have the capacity in question, but their verbal behavior indicates that they lack the capacity. Such discordance can help us explain possible intermediate stages of development and raises questions about which type of evidence best indicates a subject’s capacities. Examples can also be drawn from mainstream life in contemporary Western societies.4 Political discussions (in the United States, especially) have introduced implicit bias as a critical phenomenon to consider when addressing how people with different backgrounds are treated (Nosek et al. 2011). Discordance emerges in such cases when peoples’ verbal endorsements of some principle (say, that they view members of some race favorably) are not reflected in tasks that require them to put their conceptions to use. Although they may report viewing members of some group favorably, they find it easier to solve a task in which members of that group are associated with some unfavorable element and harder to solve the task when members of the group are associated with a favorable element. Discussions of racism and differential treatment have sought to use these sorts of results to explain how people who explicitly endorse anti-racist views can nevertheless behave in ways rooted in implicit biases. The existence of these “unseen” roots of behavior is puzzling, even disturbing. In some sense or other, the person seems to be of two minds about the situation, and both minds appear to have some role in generating behavior. What some find puzzling or disturbing about these cases is that humans view themselves as rational beings who have rather transparent access to their own minds, yet one of these two minds appears to operate somewhat covertly. The Handbook explores cases suggesting the presence of such covert processes. But, these processes are not always accompanied with discordant, overt roots. Indeed, it appears that a considerable amount of human behavior may be driven by covert processes. Hence, the chapters that follow focus on how these elements should be characterized, and how commonly they are needed to explain human activity.

3.  Initial Characterization of Implicit Phenomena The distinction between deliberative and intuitive thought is natural enough. As with any distinction, the relationship between the two affords a number of questions: Do the two types of thought interact at all, and if so, how? Is one more primary, basic, or fundamental than the other? When in conflict, can one overcome the other? What differences properly demarcate these types? Is there an actual (rather than merely apparent) difference between the two, and if so, is it a difference in kind, or something more akin to a difference in degree along a single dimension? Some cognitive scientists have aimed to explain these differences by positing separate processes or systems that are responsible for each type of cognition.5 For these theorists, ‘Type-1’ intuitive processes are said to be automatic and unconscious, and ‘Type-2’ deliberative processes are said to be controlled and conscious. These ‘Dual Systems’, ‘Dual Process’, or ‘Two Systems’ accounts abound and offer neat and straightforward explanations of all sorts of discordant behavior – in some sense, there are two minds involved, one driving each measured behavior. The existence of Type-1 processes is not just useful in explaining temporarily discordant two-minded phenomena. A number of empirical findings have challenged an overly intellectualist view of human thought that takes deliberation and other Type-2 characterizations to be the norm in explaining human behavior. In many instances, the roots of some behavior can be explained using solely Type-1 processes. For any human action, then, the question can be raised: Are its roots Type-1, Type-2, or some combination of the two? 4

Introduction

As aforementioned, some roots can be characterized as implicit cognition. To be clear, the Handbook does not equate implicit cognition with Type-1 processes or advocate adopting a Two Systems approach. Such an approach might turn out to be a fruitful way of exploring or explaining implicit cognition, or it might not. But, the definitional challenges facing those attempting to characterize Type-1 processes are similar to those facing those attempting to characterize implicit cognition and the features stressed in both attempts exhibit substantial overlap. In a sense, every chapter in the volume will contribute to a fuller understanding of how one should define or characterize ‘implicit’. Although most academic debates involve terminological disputes, the debates about implicitness can appear especially frustrating. This frustration has several sources. First, ‘implicit’ is often defined via direct contrast with ‘explicit’. Yet, too many theorists assume that ‘explicit’ is well-enough understood in cognitive science contexts to warrant defining ‘implicit’ simply as not explicit. Such contrast-class-ification may be appropriate in some contexts where the contrasting term is given sufficient definitional clarity (ideally identity conditions or necessary and jointly sufficient conditions) that can offer guidance in establishing membership in the class (and thereby, establish the contrast class). However, ‘explicit’ has not yet been given sufficient definitional clarity. Second, and relatedly, ‘explicit’ is rarely defined per se. Instead, it is characterized with a list of features. Yet, after providing a list, many theorists quickly shift to operationalize explicitness in terms of a single feature (sometimes this shift is acknowledged, sometimes not). To the extent that the features provided are coextensive, such a shift, acknowledged or not, would not be terribly problematic. Perhaps it simply provides a useful shorthand for a complicated phenomenon. But, as the chapters here will show, there are many reasons to doubt that these features are coextensive, so which feature is selected and utilized in an explanatory context matters. Third, some notion of consciousness is one of the most likely single features to play this privileged operational role in characterizing explicitness. Yet, a term within cognitive science with less definitional clarity than ‘conscious’ would be hard to identify. Hence, ‘implicit’ is often characterized as not explicit, ‘explicit’ is characterized as conscious or available to consciousness, and yet debates rage about how to define ‘conscious’. Thankfully, current debates about implicit cognition have entered something of a golden age of methodological reflection where many conceptual and experimental confounds are being discussed directly. Although debates are far from settled, a significant amount of energy has gone into identifying what kinds of clarifications and changes need to occur, if appeals to implicit cognition are going to be explanatorily powerful. A useful contribution to these methodological debates is Michael Brownstein’s The Implicit Mind (2018). In that work, he offers what should suffice as a working characterization of ‘implicit’ in introducing the rest of this volume. He writes, “At least three senses seem to be implied when we say that so-and-so’s feelings, thoughts, or skills are implicit in her behavior: Implicit=unarticulated (not verbalized); Implicit=unconscious (outside of self-awareness); Implicit=automatic (not under agent’s control)” (Brownstein 2018: 15–16). These three features of implicitness – unconscious, automatic, unarticulated – capture most of the uses of ‘implicit’ that appear in the volume, but there is no reason to view these as a list of necessary and jointly sufficient conditions for implicitness. Their centrality to implicitness and the relations they bear toward one another are very much up for debate, so the working conception won’t presuppose any particular claims Brownstein might have made about centrality or these relations. It will be his rough characterization of these features that will guide the discussion, eventually revealing some preliminary analysis of the challenges that arise in characterizing implicit phenomena. Readers should bear these challenges in mind while working through this volume.6 5

J. Robert Thompson

3.1  Outside of Awareness Implicit phenomena tend to occur outside of conscious self-awareness, and this is perhaps the most commonly identified feature of implicitness. Yet, as aforementioned, there is less consensus concerning the nature of consciousness than almost any other concept in philosophy, so the fact that implicit phenomena are defined in terms of the absence of such a disputed phenomenon make it challenging to assess this aspect of implicitness. Another challenge arises from the centrality of consciousness to the human self-conception. Although it has long been acknowledged that swaths of human cognition fall outside conscious self-awareness, there is no shared understanding of what this means for studying the mind in its conscious and unconscious aspects. For some theorists, the lack of awareness delegitimizes an alleged mental phenomenon: either the phenomenon is not mental because it is unavailable to awareness, or the phenomenon is treated with intense suspicion. From either perspective, often in underspecified ways, the phenomenon is taken to involve less legitimate, less real, inchoate features or entities whose detection requires a higher standard of evidence than what would be required for consciously accessible states. Other theorists concede that unconscious mental phenomena exist, but they insist these phenomena can be treated more or less like overt, explicit, conscious phenomena that have simply had conscious awareness removed from them: This ‘nothing special’ line of argument is a direct entailment of taking the stance that consciousness is primary and that the default position should be that conscious processes lie at the heart of human cognition. From this perspective, unconscious, implicit functions are dealt with derivatively and virtually all interesting cognitive functions are to be seen as dependent on conscious processes. (Reber 1993: 25) So, even when unconscious phenomena are widely acknowledged, the default expectation is for them to conform to the molds drawn from that which is present to our conscious awareness. In fact, such an absence in awareness of some phenomenon is often taken to be contingent or temporary – unconscious states or processes are ones that have simply not yet been consciously accessed. It is assumed that if provided the proper circumstances, subjects could make these phenomena explicit and report them verbally. Failures to evoke these reports indicate that an experimenter lacks sufficient imagination in devising a situation that will expose them. This default is problematic for several reasons. First, as the chapters that follow suggest, there are plenty of reasons to postulate implicit roots that possess very different characteristics than those postulated as the default. Second, theorists not only aim to show that implicit states and processes are distinct from explicit ones, but many (including Reber) will also argue for the primacy of the implicit. Once one begins to explore the range of implicit phenomena, they contend that it will become clear that implicit states and processes are the rightful default mode for human cognition. Hence, to the extent that the implicit involves unconscious states and processes, the primacy of the unconscious will similarly be revealed. At bottom, however, claims about primacy from either camp are premature. The extent to which conscious/explicit and unconscious/implicit processes are similar is an open empirical question, as is any claim about the primacy of one over the other. The ubiquity, representativeness, or primacy of the two sets of phenomena are very much not settled, but up for dispute and subsequent exploration. 6

Introduction

3.2 Automatic Earlier, automaticity was described as uncontrolled. But, things are much more complicated than this suggests. ‘Automaticity’ appears in these debates in a rather narrow psychological sense, commonly viewed through what Bargh identifies as the “four horsemen of automaticity”: efficiency, lack of awareness, lack of intentionality, and lack of controllability (Bargh 1994; for updated versions, see Payne 2012). Hence, more than mere controllability is involved, and one of automaticity’s classical features actually includes something like the feature of consciousness described in 3.1. Still, this cluster of features characterizes automaticity in a way that captures much of what makes implicit cognition distinctive.7 Efficiency in this context involves cognition that is fast and relatively unaffected by increases in overall cognitive load. For example, having someone count integers during a bout of automatic processing does not significantly disrupt task completion in terms of speed and accuracy. Automatic cognition involves a number of ways in which awareness could be lacking for a subject: their attention can be absent or drawn elsewhere at some relevant point in time; they can be unaware of some stimulus (as in subliminal perception); or they can be unaware of how a stimulus is impacting them (as in an emotional reaction) or their thought processes (as in activating associative patterns or connections that subsequently exert influence). ‘Intentionality’ is used in several ways by cognitive scientists, but in this context, automatic processes are intentional in the sense that the subject does not initiate the process through any standard sense of volition or control. The process is triggered or initiated regardless of tasks or goals the subject might possess. Finally, a lack of controllability arises to the extent that in automatic processes, the subject is unable to inhibit or cease the process midstream. Automatic processes are sometimes described as ballistic because they are initiated and run their course regardless of what other beliefs, plans, or goals might be present.8 Automaticity stands to explain much of what makes implicit processes implicit. In explicit cases, one can be aware of a stimulus and choose to draw inferences about it. One can refrain from drawing additional inferences or not, depending on current plans and goals. Finally, one can verbally report on how this was accomplished, and asking the subject in these cases to do something that draws upon working memory or requires them to inhibit other thoughts will disrupt the completion of the task. Implicit phenomena are distinct primarily because they take place quickly and quietly and regardless of an agent’s wishes. These same features limit integration with the rest of a subject’s cognitive life. There are many different ways that control can be lacking, and a fuller understanding of automaticity in all of its facets reveals a great deal of heterogeneity that impacts theorizing about implicitness.

3.3 Articulation When implicit phenomena are taken to be unarticulated, it is critical to determine what is being claimed: Is the phenomenon not articulated, but is articulable? Or is the phenomenon incapable of being articulated, that is, is it inarticulable? In the least committal sense, the articulable is simply that which can be verbally9 self-reported given whatever process humans use to report the roots of their behavior. The articulable is that which can be reported, whether it has been reported or not. The inarticulable, then, is not simply that which has not been reported (or not yet reported) but that which cannot be reported for some reason or other. There are many reasons why something cannot be reported in this context, but two are especially relevant. First, there may be a lack of access to the item. A lack of relevant connectivity 7

J. Robert Thompson

or some functional isolation may preclude its reporting.10 Second, the item may have characteristics that, even if sufficient motive, connectivity, and opportunity arose, prevent the item from being captured in the reporting medium. If we describe these two reasons in terms of isolation and expressibility, we can have the following possible pairings: isolated but expressible elements, isolated and inexpressible elements, nonisolated but inexpressible elements, and nonisolated and expressible elements. In discussing implicit phenomena, theorists need to be clear about which of these they have in mind when they describe them as unarticulated. The distinction between being isolated and being inexpressible is a critical divide not often acknowledged by those theorizing about the implicit. In particular, those who stress the inexpressibility of some phenomenon are often driven by a distrust of intellectualist resources that is not necessarily shared by those who stress isolation. Michael Polanyi’s (1966/2009) slogan about implicit phenomena – that in such cases, “we know more than we can say [or tell]” – cannot be fully appreciated without distinguishing the contributions of isolation and expressibility. For some, the slogan might simply suggest that there is too much to say, or that the knowledge is difficult to access. For many other theorists, the inarticulability involved in the most interesting or central cases of implicit phenomena is not merely due to isolation. The skill, understanding, or ability under discussion is not one that can be captured in explicit terms because there is something about its nature or content that resists expression in any linguistic medium, set of rules, or propositions. Perhaps one can show rather than tell how one accomplishes the task, or perhaps no attempt at expression could succeed. Hence, readers should bear in mind that some theorists positing implicit phenomena do so with the conviction that they are characterizing a range of roots of behavior – new skills, knowledge, understanding – that are essentially distinct from the explicit types in this very specific sense of being inexpressible. Other theorists do not express commitment to such inexpressibility – it may be the case that an implicit phenomenon can be articulated without substantial loss in a set of rules or propositions. Explanations that invoke implicit elements need to be understood as stemming from these different motivations that can trace back to views about the nature of the skill or knowledge involved. Hence, it is important to know whether a theorist views the domain in question as one that precludes explicitness. Issues of articulability are important in other ways as well. Many of the domains in which implicit cognition is postulated are situations where something cannot be said in a more definitive sense than those described earlier, in that the domains involve systems that are preverbal (young humans) or nonverbal (nonverbal humans, nonhuman animals, artificial systems, parts of human or animal systems, etc.). Although it might be tempting to see such systems as thereby limited to implicit elements, there are many who consider these as appropriate contexts for postulating explicit elements as well, noting that verbalization was never an essential feature required of explicit representations, processes, and the like. It may be that implicit elements often fail to be articulated, but that does not guarantee that systems where such articulation was never an option cannot be explicit. Fundamentally nonverbal systems and subsystems may involve both implicit and explicit cognitive machinery. It is at this point that the contents of the volume can begin to offer guidance into these debates. The three features of implicitness bring it into some relief, and they will reappear in various forms throughout the volume. That they can’t be the whole story, singly or jointly, is already clear. But, a richer exploration of how these features and other cognate notions relate to one another is needed, as is a better account of how to measure these somewhat covert phenomena. The next section will introduce the chapters in the volume, leaving the final sections to suggest some lessons that might be drawn from the volume as a whole.

8

Introduction

4.  Introducing the Chapters11 Part I Defining Features? Identifying Implicitness Among Cognate Notions The first two chapters further articulate the primary challenges facing any explanation of implicitness. The first is more metaphysical – what form or functions might such implicit cognitive processes possess? The second is more methodological – what resources can be brought to bear in detecting and analyzing processes that appear mostly covert? In addition to assuming that behaviors have causes, the chapters in this volume also engage with views in which these causes can be explained as representational states. Most will adopt such a perspective, while others will attempt to motivate their discussion in opposition to some specific version of this perspective. Hence, a natural and influential move within cognitive science is to account for implicitness by developing notions of implicit representation. William Ramsey (Ch. 1: Implicit Mental Representation) reviews attempts in which representations are described as being some combination of the following: unconscious, dormant, merely implied or entailed by explicit representations, and identifiable via the dispositional properties of the system’s functional architecture. Ramsey argues that the first two are indispensable features of implicit representations, but that the latter two should be dropped to avoid conceptual confusion and misunderstandings. Representations are not the only things classified as implicit: tasks, behaviors or outcomes, and the processes that bring about the outcomes are all implicated. Consider the infamous Implicit Association Task (IAT). In one version, researchers query subjects about what views they have about gender representation in professional fields and the subjects deliberately respond to the questions by assessing what they take their views to be and report them directly. The implicit elements resist such direct report and must be explored indirectly. Subjects are asked to sort cards that present terms representing gender he, she or field physics, literature. Their sorting behavior can be used to determine if the subject appears to associate gender and field and what association is present, in that it will be easier (and hence quicker) for them to sort cards into a pile that is consistent with their associations and harder to sort in a way that is inconsistent with those associations. If subjects more quickly sort she-literature than she-physics, this indicates that they hold an implicit association between the gendered term and the profession term that impacts their behavior, regardless of whether this matches their professed, directly reported attitude about such matters. Such tasks provide insight into what self-reporting cannot. Great strides have been made in devising novel measures of this covert processing, but the covertness also makes the measures and phenomena involved more difficult to isolate and interpret. The IAT is a paradigmatic implicit task, but in what sense are its elements taken to be implicit? Some treat the task itself as an implicit task because it involves an indirect measure. Others treat it to be an implicit task because it claims to measure an implicit psychological process – implicit mental associations – rather than an explicit process like deliberation among sentence-like, consciously accessible, and verbally expressible beliefs. Others describe it as implicit because the behavior appears rather automatic. Hence, not all implicit tasks are taken to be implicit in the same ways, and this has led to substantial confusion. Samuel A.W. Klein and Jeffrey W. Sherman (Ch. 2: Measuring and Modeling Implicit Cognition) diagnose this confusion about the nature of the tasks and processes in these debates, recommending more rigorous experimentation and the use

9

J. Robert Thompson

of formal process modeling, and in future research. Without these changes, theorists will continue to confuse things such as the conditions under which a process is instigated, the nature of the process that is instigated, and the measures that are used to provide evidence for the other two factors. Such confusion is often not spotted and this leads a number of theorists to misunderstand rival accounts and misidentify the type of evidence that will be relevant in deciding among various alternatives. The remaining chapters in Part I of this volume are devoted to unpacking what implicitness amounts to, comparing the Implicit/Explicit (I/E) distinction to a number of cognate distinctions: conscious/nonconscious, conceptual/nonconceptual, personal/subpersonal, high level/low level. Each of these chapters will explain the relevant similarities between the I/E distinction and the distinction at hand, noting the reasons why no straightforward identification between distinctions will suffice. In the end, separate consideration of each of these distinctions exposes important features about implicitness that are of independent interest for accounts of implicitness. As noted earlier, if there is one common or central feature used to identify a process as implicit, it is related to the fact that the element is not consciously accessible to the agent. Tim Crane and J. Robert Thompson (Ch. 3: Implicit Cognition and Unconscious Mentality) explain the different ways in which mentality might lack consciousness. Although appeals to unconscious mentality are commonplace in contemporary cognitive science, attempts to characterize it remain underdeveloped. Such attempts often struggle to articulate whether the unconscious realm differs from the conscious primarily in terms of the presence or absence of consciousness, or whether more profound differences between these two realms might exist. Crane and Thompson explain the pitfalls of using conscious mentality as a model for characterizing unconscious mentality, noting that the former’s focus on conscious sensory episodes makes it unsuited for characterizing the nonsensory states and processes that constitute most of what cognitive scientists postulate as unconscious machinery. In contrast to views that describe the unconscious in a deflationary way, as standard mentality that lacks consciousness, Crane and Thompson sketch several alternatives that allow for richer views about what makes unconscious representations unique. These alternatives allow a deeper understanding of cases of implicit cognition as well as provide possible solutions to a number of puzzles about representation within cognitive science. The distinction between conceptual and nonconceptual phenomena holds a central place in many philosophical systems and it seems likely to have some relevance to the implicit/explicit distinction. José Luis Bermúdez and Arnon Cahen (Ch. 4: Implicit Cognition in Relation to the Conceptual/Nonconceptual Distinction) examine two cases – linguistic competence and procedural knowledge – in order to show how considerations about nonconceptual content can illuminate the nature of implicit knowledge. The authors show that both cases seem to involve implicit knowledge because they posit states that count as knowledge that are nevertheless inarticulable. They argue that something like nonconceptual contents simultaneously account for both why these states count as knowledge and why they cannot be articulated. Attributing implicit states with nonconceptual contents violates common conceptions of the mind involving the introspective access of thoughts and their articulability. Zoe Drayson (Ch. 5: The Fragmented Mind: Personal and Subpersonal Approaches to Implicit Mental States) explains that distinguishing between personal and subpersonal explanations can help make sense of such violations. In cases where it seems useful to attribute implicit mental states, several options are available: one can take the subpersonal routes rampant in cognitive science and relax these common conceptions or normative strictures; one may offer personal-level explanations that still aim to maintain such strictures (she mentions relativizing mental state 10

Introduction

attributions to persons at a time or a context, or invoking practical modes of presentation); or, one can offer a mixture of both approaches in a full explanation of the phenomenon. Since there is no straightforward identification of the implicit with the subpersonal, whether one adopts a personal or subpersonal take in any scenario is going to depend on much broader methodological and metaphilosophical approaches. Judith Carlisle (Ch. 6: The Levels Metaphor and the Implicit/Explicit Distinction) offers an account of various ways in which levels are invoked in explanations within cognitive science (levels of organization, complexity, processing, sophistication, and Marr’s three levels). She considers whether drawing the I/E distinction might be done by identifying explicit phenomena as occurring at a higher level than implicit phenomena, but argues that this method of identification fails. The I/E distinction seems, in her view, to track aspects of consciousness and control most closely in debates about levels. Hence, she recommends that it would be better to focus on conscious control and availability rather than levels in the debates she addresses.

Part II The Nature and Limits of Implicit Processing Initial characterizations of the states and processes involved as either deliberative or intuitive need to be replaced with richer analyses of these elements – what they are like and the extent to which they interact. Charlie Blunden, Paul Rehren, and Hanno Sauer (Ch. 7: Implicit Cognition, Dual Process Theory, and Moral Judgment) elaborate the general motivation for positing more than one system or process. They introduce the issues such accounts have faced in demarcating distinct systems in an explanatorily powerful way. Worries about contrastclass-ification arise, worsened by the fact that most features under discussion manifest in different forms. It matters quite a bit which feature of consciousness or control is in play when one identifies its contrasting feature (which sense or version of the explicit feature is to be contrasted in articulating the implicit feature). The authors also describe the pernicious ‘alignment problem’ – decades of work describing multiple systems suggest that the features (described in detail later, in 5.1) do not cleave into two systems cleanly. That is, processes that are clearly of the Type-1 sort can (to varying extents) exhibit features from the Type-2 column, and vice versa. After introducing dual process accounts and these potential pitfalls, the authors focus on how implicit notions function in the context of moral psychology. They consider the extent to which something like implicit moral judgments can occur (say, an instant reaction to a morally salient event) and whether they are adequately explained as resulting from a Type-1 process. Many empirical results in moral psychology suggest that there is some type of judgment with some of the features of Type-1 processes, but the alignment problem arises in this context. The authors develop several approaches to explain the nature of moral judgment in the face of pressures to posit two (and even three) types of processes responsible for these judgments and explain how implicitness is best treated in these discussions. Ema Sullivan-Bissett (Ch. 8: Implicit Bias and Processing) also discusses two-systems accounts, but she does so in order to more fully examine the structure of implicit processing itself, especially the implicit processes involved in what are commonly referred to as ‘biases’. A prominent distinguishing feature of explicit thought is that it can involve propositional content and truth-preserving inference. Implicit thought, however, is often taken to involve ‘mere’ associations among its components. Indeed, many have viewed biases and the processes underlying them as involving such associations (the preceding discussion of the IAT affords such an interpretation). But, more recent analysis suggests that biases also exhibit propositional features. 11

J. Robert Thompson

Since the mental constructs involved in biases (and implicit cognition more generally) exhibit a fair amount of heterogeneity, a simple associationist rendering in not viable. Sullivan-Bissett explains that her own account, grounded in unconscious imaginings, is able to accommodate the heterogeneity better than alternatives. Several ‘revolutions’ in cognitive science have offered ways of situating notions like representation and inference in opposition to classical computational theories of cognition (e.g., connectionist, dynamical systems, ecological approaches). Each of these revolutions has embraced the idea that these new descriptions of cognition do away with explicit knowledge, representations, or deliberative reasoning in favor of something that can be characterized as merely implicit. Recent predictive processing models of cognition (PP) offer a framework with the potential for a similar sort of revolution. Paweł Gładziejewski (Ch. 9: Predictive Processing, Implicit and Explicit) offers a primer on PP and utilizes the I/E distinction to shed light on the theoretical commitments of PP. He proposes two realizations of the PP program. The first involves explicit representational structures that participate in implicit inferential processes, that is, processes that approximately accord to a (Bayes-)rational rule without the rule itself being represented in the system. The second involves ‘representations’ implicitly embodied in processing biases that do not undergo inferential updating. These appear to map fairly well onto two primary camps that have emerged in recent developments within PP: the ‘intellectualist’ and ‘radical’ readings of PP, respectively. If implicit notions clarify the debates along these lines, his analysis offers a way in which both readings of PP seem to be involved in cognition and each may get part of the story right. PP then offers both an account of what the I/E distinction is, and the I/E distinction allows one to clarify important issues within PP debates. Many theorists have recently turned their attention toward the extent to which beliefs, desires, intentions, and the like have some implicit influence on how we perceive objects and events. Lucas Battich and Ophelia Deroy (Ch. 10: Cognitive Penetration and Implicit Cognition) discuss this cognitive penetration of perception and its epistemic ramifications. If we are not aware of implicit forms of this influence, it seems we will be unaware that our perceptions involve some degree of misrepresentation as a result of those influences. These sorts of epistemic threats are relevant to questions about epistemic responsibility, which is standardly said to require something like awareness and control over such influence. In addition to consideration of how individuals might address such influences, the authors explore the role of social groups, both as sources of influence and correctives to such influence. They suggest that to the extent that there are implicit influences in perception, these influences have social significance.

Part III Ways of Perceiving, Knowing, Believing As indicated in Chapter 10, the elements involved in perception often take some effort to reveal, and the operations involved appear to have at least some similarities to thought. Lydia Patton (Ch. 11: Helmholtz on Unconscious Inference in Experience) explores Hermann von Helmholtz’s account of unconscious inferences in perception as a (19th-century) precursor to contemporary discussions of implicit cognition. Drawing from Kant and his own exploration of the physiology of perception, his account centers around perspective and inference, both in the experimenter and the perceiving subject, explaining how the world ultimately gets represented in relation to the subject’s possible experiences. His influence can certainly be felt in contemporary accounts of perception and mentality, including accounts of Predictive Processing 12

Introduction

(Chapter 9), but Patton brings out the features that make his explanatory apparatus both utterly unique and of considerable interest to those exploring implicit cognition. Phenomenology resists attempts to oversimplify perception or misconstrue it as akin to deliberation. Dermot Moran (Ch. 12: Husserl on Habit, Horizons, and Background) describes classical phenomenology and how its insights about the nature of experience inform approaches to understand implicit cognition. Many features emphasized by phenomenologists resonate with implicitness in the senses described earlier. Husserl identifies knowledge that is explicit in propositions that encapsulate judgements, but there is also a deeper, prior pre-predicative knowledge that is passively synthesized. Merleau-Ponty elaborates on Husserl’s accounts of this pre-theoretical, preconscious experience with an analysis of the habitual skillful knowledge implicit in embodiment. Moran argues that cognitive scientists and philosophers should heed phenomenology’s rich attention to experience, skill, embodiment, enculturation, and tradition, if they hope to give an adequate account of implicit cognition. The prior chapters complicate the relationships among perception, thought, embodiment, and skilled knowledge, as well as refine the ways we gain insight into these elements. This extends to our understanding of the types of knowledge that undergird these activities and the appearance of reasoning behind them. As self-identifying rational animals, we take intelligent intentional behavior to be the representative human activity. Reasoning and decision-making take center stage – playing chess, deciding what to cook, figuring out why a tool isn’t working, selecting a partner, traversing a blocked path or a stream – these are all the sorts of activities for which roots can be identified and examined. Yet, the preceding discussion suggests that there are “important cognitive lacunae between the (explicit) knowledge that we thought we used to make decisions and control choices, the (implicit) knowledge we actually used, and our differential capacities to articulate these kinds of knowledge” (Reber 1993: 14). How significant is the fact that sectors of cognition resist self-reporting? Deliberative processes lend themselves to externalization and self-report, but there are good reasons to doubt that all such processing can be externalized and self-reported. And there are reasons to think that nothing like deliberation is present among the roots of much intelligent behavior. That theorists might need correcting on how to categorize the roots of human behavior is hardly a new claim. As aforementioned, there has been a tendency to overestimate the prevalence of deliberation of the sort that is said to inform self-reports. Gilbert Ryle is often cited as providing an articulation and critique of this sort of intellectualism. He explains (1945: 9): I have been arguing in effect that ratiocination is not the general condition of rational behaviour but only one species of it. Yet the traditional associations of the word ‘rational’ are such that it is commonly assumed that behaviour can only be rational if the overt actions taken are escorted by internal operations of considering and acknowledging the reasons for taking them, i.e., if we preach to ourselves before we practise. “How else” (it would be urged) “could principles, rules, reasons, criteria, etc., govern performances, unless the agent thought of them while or before acting?” People equate rational behaviour with premeditated or reasoned behaviour, i.e., behaviour in which the agent internally persuades himself by arguments to do what he does. Among the premisses of these postulated internal arguments will be the formulae expressing the principles, rules, criteria or reasons which govern the resultant intelligent actions. This whole story now seems to me false in fact and refutable in logic. Ryle’s official doctrines and positive proposals are difficult to pin down.12 But, one can incorporate his attack on intellectualism into our examination of the real roots of human behavior. In so 13

J. Robert Thompson

doing, a strong version of intellectualism is challenged, one that claims that all rational, skilled, or intelligent behavior must have intellectualist roots. There are at least some compelling cases of such action in which no intellectualist roots or ‘escorts’ are present. Of specific relevance to the Handbook, Ryle notes that in such cases, it is not enough to insist that the apparently absent intellectualist roots are actually ‘implicitly’ available or understood (1945: 7–8): There is a not unfashionable shuffle which tries to circumvent these considerations by saying that the intelligent reasoner who has not been taught logic knows the logicians’ formulae ‘implicitly’ but not ‘explicitly’; or that the ordinary virtuous person has ‘implicit’ but not ‘explicit’ knowledge of the rules of right conduct; the skilful but untheoretical chess-player ‘implicitly’ acknowledges a lot of strategic and tactical maxims, though he never formulates them and might not recognise them if they were imparted to him by some Clausewitz of the game. This shuffle assumes that knowledge-how must be reducible to knowledge-that, while conceding that no operations of acknowledging-that need be actually found occurring. Part of what interests Ryle in these cases is the long-standing philosophical debate about whether declarative knowledge alone can motivate or account for some behavior. But for the purposes of the Handbook, this sort of attack allows us to address critical issues: whether all intelligent behaviors have escorts (roots); and to the extent that such escorts exist, what precisely they are like. One can formulate debates about intellectualism around questions over what type of escorts need to be present in the etiology of some act. In such debates, Ryle’s challenge remains – that intellectualists have things backwards: “In short the propositional acknowledgement of rules, reasons or principles is not the parent of the intelligent application of them; it is a step-child of that application” (1945: 9). Ryle aimed to do more than show that at least some roots were not intellectual or deliberative, but these types of attacks on intellectualism serve to motivate accounts that explore what might actually undergird these behaviors, if not some propositional acknowledgment or intellectual rehearsal. To the extent that the roots of some behavior resist capture in terms of deliberation, the need for a different type of knowledge is deepened. The generic distinction given earlier, between deliberative and intuitive, informs many views of human nature and intelligence. Although all theorists need to embrace both types of processes to account for human action, there are clear senses in which, for many theorists, humans are seen as either primarily driven by deliberative processes or primarily driven by intuitive processes. Human knowledge is cleaved along similar lines, with distinctions drawn between knowledge-that and knowledge-how (Ryle 1945, 1949), and between declarative knowledge and procedural knowledge (Anderson 1983). The study of the implicit cognition maps out this relatively novel territory: a sort of knowledge that is distinct from and not reducible to explicit knowledge; a sort of knowledge that resists articulation; a sort of behavior that is either not driven by (any?) escorts, or is driven by escorts that are not explicit; and a sort of behavior that is not mediated via propositional acknowledgment. Michael Polanyi offers an influential characterization of these sorts of anti-intellectual features, but he describes them as tacit rather than implicit.13 Polanyi’s views extend far beyond the slogan mentioned earlier, that in implicit contexts, we know more than we can tell, and much of the research into implicit phenomena begins with discussion of Polanyi or takes his accounts as its target. Stephen Turner (Ch. 13: Polanyi and Tacit Knowledge) captures Polanyi’s views in detail, highlighting his unique takes on skill, meaning, inarticulability, irreducibility, activity, and limited awareness. Many take his views to entail a subjectivism about tacit knowledge, others find his analysis of skill to be the most promising way of locating the skill as an objective 14

Introduction

object of study. Science studies in particular have adopted many of Polanyi’s insights about the importance of skills, expertise, and the centrality of scientific practices (rather than theories) that appear irreducible to rules. Tim Thornton (Ch. 14: Tacit Knowledge) offers an account of tacit knowledge that is situated in contrast to two claims from Polanyi: the idea that we know more than we can tell and the suggestion that knowledge is practical. As Thornton notes, the nature of the non-explicit is always in tension: the aspects that enable an item to be tacit appear to prevent it from counting as knowledge, and the aspects that enable an item to be known appear to prevent it from being tacit. Thornton develops a way of reconciling these tensions by equating tacit knowledge with conceptually structured, situation specific practical knowledge or know-how. He contrasts his account with prior work by Polanyi and Harry Collins. Harry Collins offers his own detailed accounts of tacit and implicit knowledge in other work (e.g., Collins (2010) but in his contribution to this volume (Ch. 15: Collective and Distributed Knowledge: Studies of Expertise and Experience), he focuses on ways in which not just individuals but communities can be seen as possessing shared knowledge of practices. Drawing upon Polanyi and Wittgenstein, he stresses that this tacit knowledge has relational, somatic, and collective features. Collins focuses on scientific communities to bring out how these practices are generated and transferred, drawing especially on his work embedded within the communities studying gravitational waves. Insofar as interaction is a key element in his account, he introduces readers to his method of using Turing-style imitation games to explore the nature of interactional expertise. This method allows a comparison of a group’s implicit and explicit understanding of another group, as measured by their success (or failure) in an imitation game. The preceding chapters in Part III analyze implicit elements that are present in perception, understood through perception, or involve ways of tacitly knowing or understanding. The final two chapters embrace more classically representationalist perspectives about how belief and selfknowledge might operate. Both aim to explain how representations might or might not help illuminate such possibilities. Joseph Bendaña (Ch. 16: Implicit Beliefs) begins his chapter by noting that, minimally, implicit beliefs are beliefs that are not explicit. But, given the fact that beliefs are characterized functionally, the very features that seem to enable the states to be implicit might preclude the states from being beliefs. Since implicit beliefs are attributed to satisfy some particular explanatory purpose, proponents of these states must account for the sense in which these states, minus the explicit features, can still meet those purposes. Bendaña motivates the need for such states – cases where beliefs are needed, but explicit ones will not do – and examines the reasons given by those who doubt their utility. Philosophers often invoke such states in order to defend an account of belief from some competitor, whereas psychologists often invoke them to explain nonverbal measures of attitudes or biases. These differing motivations for the states may signal that no univocal account will emerge, but Bendaña nevertheless explains how different approaches (different representations of belief-like states, treating the mind as though it consisted of separated epistemic domains) need to be considered in assessing the prospects for such accounts. Even if no univocal account can be given, there are plenty of reasons to postulate implicit beliefs of some sort or other. Philosophers often assume that certain instances of consciously controlled introspection reveal their own mental states. This would count as an explicit form of self-knowledge. Kristina Musholt (Ch. 17: Implicit Self-Knowledge) gives an overview of different types of knowledge that can reasonably be considered forms of implicit self-knowledge. She notes that philosophical discussions of self-knowledge focus on its immunity to error through misidentification and attempts to clarify what would characterize implicit self-knowledge. Theories of nonconceptual 15

J. Robert Thompson

self-consciousness and forms of implicit and explicit representation seem connected, and notions of know-how seem poised to address how an agent can self-attribute thoughts when they aren’t available or articulable to the agent. In the end, Musholt hopes that a refined understanding of implicit self-knowledge allows us to gain traction on a number of phenomena: language acquisition, rationality, conceptual abilities, theory of mind, self-reflection, metacognition, implicit bias, self-deception, and oppressive socialization. Moreover, it might lead to better insight into the possibility of gaining knowledge about one’s true self in cases of discordance.

Part IV Language Many take Ryle’s attacks on intellectualism to have established the impotence of explanations that capture an intellectual achievement solely in virtue of possessing of a body of declarative knowledge. He argued that possession of such knowledge could not suffice for executing that achievement. Such explanations would either need to be augmented with know-how in order to account for the action, or if no such know-how were added, the execution would either remain unaccounted for or be handled surreptitiously by some version of the homuncularist fallacy. As the cognitive revolution emerged in the second half of the 20th century, the arguments of Ryle against the intellectualist and homuncularist fallacies were turned from criticisms into research methodologies (Attneave 1960). Early work by Noam Chomsky (1966) and Jerry Fodor (1968a, 1968b) embrace intellectualism as a viable explanatory strategy. Their addition to the debate was to re-envision the cognitive roots of behavior as parts of systems that were encapsulated from general intelligence. As Fodor explains, “the anti-intellectualist arguments fail to go through because they confuse knowing that with being able to explain how” (1968b: 71). The knowledge involved in these explanations is not easily accessed by the agent who used it. Hence, there are things that we cannot say, but the fact that we cannot explain them (or state them) does not show that they fail to be declarative knowledge. What they fail to be is consciously accessible knowledge. Such approaches came to be widely accepted in some form or other (Cummins 1983; Dennett 1978; Lycan 1991) and debates turned to how this declarative knowledge operates within its nested systems and subsystems, with particular attention paid to whether the notion of representation could elucidate how such systems could explain cognition (Cummins 1989; Dennett 1978; Dretske 1981; Fodor 1975, 1981; Millikan 1984). Proponents of such proposals were impressed both with the arguments given against accounts of language use that depended on something akin to know-how as well as the alternatives offered in terms of a declarative knowledge of rules and representations of a speaker’s language. That is, they celebrated a breakthrough in understanding and explaining cognition that Fodor and others would attribute to Chomsky’s account of language acquisition and competence. The explanatory approach on offer was a cognitive account that involved a speaker possessing knowledge of their language that took the form of a ‘tacit’ theory. As mentioned earlier, this theory must have some tacit or implicit elements – if nothing else, it must be inaccessible to consciousness. This approach has become commonplace, and many cognitive scientists view their work as identifying an implicit theory that captures some domain – not only the linguistic domain but also social domains, mathematical domains, physical domains, biological domains – that humans possess and that allows them to reason about the elements in that domain. Cognitive science, in this intellectualist form, becomes the study of the implicit theories that allow cognizers to reason about their surroundings. 16

Introduction

Turning back to the linguistic domain, whether or not Chomsky held such a tacit-theory view, what sort of psychological states it committed him to ontologically, and during what periods in his career he may have seriously held such a view are analyzed in the chapter by John Collins (Ch. 18: Chomsky, Cognizing, and Tacit Knowledge). He helpfully explores the trajectory of Chomsky’s thinking about our ‘knowledge of language’ and how that meshes with the conception described earlier. Rather than seeing Chomsky as specifying the sorts of mental states that would constitute a form of knowledge that would explain linguistic behavior, Collins argues that Chomsky has long construed the knowledge at issue in a thin sense, more of a specification of the kind of phenomena at issue rather than a presupposition about the kind of states that explain linguistic behavior and in particular, the knowledge does not fit with either a ‘know that’ or ‘know how’ construal, as if linguistics posits tacit propositional knowledge or structures essentially involved in linguistic actions. This articulation of Chomsky’s commitments will seem surprising to many, but Collins sees Chomsky as having always provided a way of making clear what was implicit from the beginning. When it comes to understanding what Chomsky really believed and the upshot of his insight for implicit cognition, Collins insists that it was always the initial state of the language acquisition device that Chomsky was describing and that device always captured the sense in which anything about language was present implicitly. David Pereplyotchik (Ch. 19: Language Processing: Making It Implicit?) dives deeply into the domain of language processing in order to clarify the nature of implicit cognition. He identifies the properties of mental representations that lead psycholinguists to describe them as either implicit or explicit – being (non)conscious, (sub)personal, declarative or procedural, and occurrent or dispositional – and explains the extent to which these sorts of properties apply to various representations posited in psycholinguistics. Pereplyotchik focuses primarily on the existence of distinct representations required for parsing sentences in psycholinguistic accounts. He explains the sense in which these various representational states are required in our best psycholinguistic accounts, and the extent to which they should count as implicit. Using language requires a wealth of knowledge about how linguistic elements feature in one’s current socio-conversational circumstances. Chris Cummins and Albertyna Paciorek (Ch. 20: Implicit Knowledge in Pragmatic Inference) discuss four topics involving such pragmatic knowledge and consider whether the knowledge involved is implicit or explicit: metaphor interpretation, speech act recognition, quantity implicature (in the context of Rational Speech Act theory), and the expression of numerical quantity. For each topic, they contrast two accounts: one in which speakers convey determinate, introspectively accessible content to hearers, who explicitly reconstruct it via a discrete series of reasoning stages; and one in which the content communicated by the speaker need not be determinate, wherein the hearer’s process of meaning (re)construction makes use of implicit knowledge that could be acquired through the extraction of statistical regularities from the hearer’s linguistic experience. The authors argue that the accounts that rely on explicit knowledge struggle to explain how the precise meanings intended by the speaker could possibly be reconstructed under the conditions of normal interaction, whereas the accounts relying on implicit knowledge seem able to handle these challenges.

Part V Agency and Control Fully acknowledging the existence of implicit roots of behavior allows a richer analysis of human action than is available for researchers who aim to shoehorn their analyses into explicit deliberative frameworks. Sofia Bonicalzi (Ch. 21: Implicit Mechanisms in Action and 17

J. Robert Thompson

in the Experience of Agency) describes several explanations of action-related phenomena that utilize implicit mechanisms (the goal-directedness of actions, the correction of an ongoing action, and the experience of being an agent). She explains that there are several notions of implicitness in play in these debates: automaticity and inaccessibility to introspective awareness; nonverbalizable influences on behavior; and susceptibility to nonverbal (yet objective) measurement parameters. Successful explanations of agency carefully distinguish these notions and use them discriminately. Implicit cognition can be postulated to explain disruptions in agency as well. Addictive behaviors seem to involve a disruption for which such a postulation can be fruitful. Reinout W. Wiers and Alan W. Stacy (Ch. 22: Implicit Cognition and Addiction: Selected Recent Findings and Theory), editors of an important handbook on implicit cognition and addiction (Wiers and Stacy 2005), update developments in this field concerning theory, assessment, and clinical applications. They explore recent definitions of ‘implicit cognition’ and how they shed light on the phenomena of addiction, especially focusing on a loss of voluntary control. They describe implicit cognitive processes, focusing on the assessment of memory associations via reaction times and the strength of word associations for elements related to addiction. The authors discuss how using information gained from implicit measures can shape treatment outcomes and modify cognitive biases. Many psychopathologies also involve disruptions in agency, self-control, and self-awareness, many of which exhibit implicit profiles. Several such cases involve experiential features about agency and self-awareness that require a careful study of the sort that phenomenologists seek to offer. Anthony Vincent Fernandez (Ch. 23: Phenomenology, Psychopathology, and Pre-Reflective Experience) describes phenomenology and phenomenological psychopathology by clarifying the kind of implicit experiences that concern phenomenologists. He introduces the phenomenological concept of pre-reflective experience, focusing especially on its relation to the concept of implicit experience. Both classical phenomenologists and contemporary phenomenological psychopathologists have studied the structure of pre-reflective self-consciousness, which has been used by phenomenological psychopathologists to better understand the experience of schizophrenia. Fernandez assesses several of the methodological challenges that arise in this field of research, thereby facilitating critical engagement and collaboration between phenomenologists and researchers working across a variety of disciplines.

Part VI Social Cognition Humans often think about one another, but despite this feat’s centrality to the human experience, there is no consensus about how we accomplish it or even what to call it. ‘Social cognition’ serves as a general phrase for this type of thinking about others, but other theorists restrict their focus to a narrower phenomenon that is captured by ‘Theory of Mind’, where this is taken to be the capacity to understand, predict, and explain the intentional behavior of others by attributing mental states to these agents. There are myriad debates about the nature of these processes – what they are like, how they develop (in ontogeny and phylogeny), and the social significance gleaned from this research. Although discussions about how to measure and capture implicit social cognition appear in earlier chapters, the chapters in this section examine implicit social cognition more broadly and explain many challenges to situating its implicit features and their efficacy within wider theoretical contexts.

18

Introduction

As aforementioned, many see implicit cognition as an antidote to overly intellectualist accounts of human cognition. In their chapters, Fernandez and Moran stress the pre-reflective, embodied aspects of thought and perception, and Jasper St. Bernard and Shaun Gallagher (Ch. 24: Race and the Implicit Aspects of Embodied Social Interaction) stress similar features in social cognition. Much work in social cognition aims to uncover the ways implicit elements might lead to biased behaviors, but St. Bernard and Gallagher argue that because the literature on implicit bias overemphasizes mental states like beliefs and unconscious attitudes, and underemphasizes embodied processes and the dynamics of interaction, it has been difficult to understand some common examples of racial bias. They explain that bodily processes, many of which are best described as implicit, have a critical impact on the formation and manifestation of race-related attitudes that involve contextualized interactions, habit formation, and narrative practices. Embodied cognition approaches can shed light on claims about the automaticity and inaccessibility of implicit biases, and on the way that embodied interactions influence our perception of others. Perhaps most of our thinking about others occurs implicitly. Shannon Spaulding (Ch. 25: Implicit Social Cognition) explains that something like this this idea – that the cognitive processing involved in attributing mental states can occur non-consciously, in the absence of voluntary control, and may resist verbal articulation – plays a central role in defenses of general accounts of socio-cognitive mentalizing, the development of infants’ folk psychological abilities, and implicit bias. Although positing implicit social cognitive processes is common, she notes there is little effort to articulate what counts as implicit social cognition across all these cases. As a result, theorizing about implicit social cognition is extremely disparate across each of these sub-domains. Spaulding draws upon Brownstein’s (2018) account of implicit cognition to further articulate a fruitful general account of implicit social cognition that might apply across these sub-domains. Accounts of social cognition have been shaped by considering cases where these abilities are absent or lacking. For example, mature human capacities come into relief when contrasted both with what nonhuman animals lack and with what younger children initially lack, but subsequently develop. Hannes Rakoczy (Ch. 26: The Development of Implicit Theory of Mind) reviews the early development of this remarkable socio-cognitive capacity in humans, in particular the abilities dubbed ‘Theory of Mind’. The nature and timing of this development has become a hotly contested issue, in particular because it appears to be a form of implicit Theory of Mind that may develop early in ontogeny, may be evolutionarily more ancient, and may continue to function in automatic and unconscious ways across the lifespan. Yet, the evidence for such a capacity has been difficult to connect to theoretical accounts of this capacity. Rakoczy discusses whether there are robust findings supporting the existence of this implicit capacity and how such findings ought to be interpreted. He offers reasons to doubt that there are such robust findings, exploring the original findings, an apparent replication crisis, and future directions for research.

Part VII Memory Although many people may not have considered the possibility that memories might have implicit features, decades of research have explored the sense in which recall occurs unconsciously, but nevertheless impacts behavior. Sarah K. Robins (Ch. 27: Implicit Memory) clarifies that implicit memory is the use of memory without awareness of the activity of remembering and explains the distinct interests pursued by psychologists and philosophers regarding this

19

J. Robert Thompson

type of memory. Psychologists distinguish implicit recollection from explicit recall, but there is no consensus on whether any distinctive awareness accompanies explicit memory. Philosophers, in contrast, distinguish implicit and explicit memory storage. They view explicit storage as something that lacks memory traces, yet without these traces, it is not clear what options are available in explaining how implicit storage would differ. Robins concludes her chapter by considering alternative philosophical accounts of implicit memory that provide looser and more dynamic connections to explicit memory. Anne M. Cleary (Ch. 28: Memory During Failures of Recall) offers insight gained through examining how humans forget. In some cases of forgetting, information sits in memory but is unable to be accessed – what she describes as retrieval failure. Yet, even when this information fails to be consciously retrieved, these memories can nevertheless exert an influence on an agent. One way this influence can occur is through the sort of implicit memories described by Robins. Another involves metacognitive sensations of memory during retrieval failure, such as tipof-the-tongue states, déjà vu, or familiarity-detection. Cleary suggests several paths for future research that explore the possible relationships between the two, such as why the latter occurs with some sensation of awareness whereas the former appears to exist completely outside of awareness. The nature of these metacognitive feelings is especially interesting in that the extent to which some similar feelings might coincide with other implicit phenomena is rarely considered.

Part VIII Learning and Reasoning Deliberative and discursive reasoning involves inferential patterns of the sort that are typically identified as explicit and contrasted with implicit phenomena. Yet, there are many reasons to posit something like non-explicit reasoning and inference, that is, some version that takes place in unconscious, automatic, and unarticulated ways. There are similar reasons to posit something like implicit learning, that is, some version of learning that takes place automatically and without acknowledgment of what is being learned. If the roots of behavior, and our ability to come to understand them in ourselves and others, are to take implicit forms, then some sense must be made of the notion that humans can infer, reason, and learn implicitly. Thomas Sturm and Uljana Feest (Ch. 29: Implicit Reasoning) describe attempts to make sense of the notion of implicit reasoning. They note that there is an obvious need to posit something like implicit reasoning, yet at the same time, there appear to be three nonnegotiable assumptions in play: that reasoning must be explained in terms of inference, that inferences operate over propositions (or elements with propositional contents), and that the rules of inference have normative import. After unpacking how these assumptions might be seen as clashing not only with empiricist notions of mental processing but also with contemporary theories of Type-1 processes, they present a reading of the Type-1/Type-2 division, according to which there is no deep categorical or insuperable chasm between implicit and explicit processes. With this, they try to support their main thesis, i.e., “some implicit processes can . . . be instances of reasoning: because they can be made explicit and then also be evaluated critically just like standard, conscious, or deliberate instances of reasoning.” Keith Stenning and Michiel van Lambalgen (Ch. 30: Implicit Knowledge of (Parts of) Logic, and How to Make It Explicit) aim to expand conceptions of what implicit reasoning can be like. To accomplish this, they describe the flexible processing that takes place in understanding and reasoning about narratives. They note that such processing is often deemed to be insufficiently deliberative to count as reasoning, and that this insufficiency is often traced to 20

Introduction

a claim about a lack of formal structure in that reasoning. But, they argue that one can provide a formalism involving situation models that captures what is present implicitly in these processes, showing that implicitness can go hand in hand with intricate combinatorial structure. The authors then motivate their account by applying it to an interesting data point about the relationship between literacy and logical reasoning (the inferences (drawn and not drawn) during the processing of narratives by illiterate populations). In the end, this approach describes narrative processing as implicit, but nevertheless quite unlike the relatively structure-poor Type-1 processes appealed to in Two Systems approaches to reasoning. Hence, they expand our conceptions of what implicit processes can be like, providing an example of a structure-rich implicit reasoning process. Arnaud Destrebecqz (Ch. 31: What Is It Like to Learn Implicitly?) offers a brief overview of current debates over implicit learning (most fundamentally, learning that takes place in the absence of an intention to learn and results in acquiring knowledge that cannot be easily verbalized). When one begins to identify such cases as involving unconscious learning, there is still substantial debate over what evidence would show that such learning took place. He shows how researchers have modified their approaches in response to this state of disagreement. Although some researchers still aim to demonstrate that exclusively unconscious knowledge has been acquired, researchers like Pierre Perruchet have sought to show something more tractable – that some bit of learning has resulted from both conscious and unconscious processes. Destrebecqz explores how such results shed light on the nature of the processes that seem to match the unconscious and conscious categories developed in different account of consciousness. Such implicit phenomena appear to lack quantities of particular characteristics like strength, stability in time, and distinctiveness, that when augmented, render elements with a characteristic explicitness that makes them candidates for consciousness. It may be that learning in these mixed circumstances (conscious and unconscious processes) reveals potentially important general features of implicit and explicit phenomena.

5.  Lessons Learned, Looking Forward This introduction began by inquiring into the actual roots of human behavior. As the chapter descriptions in Section 4 convey, not all of these roots are helpfully viewed as explicit. Something like implicit cognition needs to be postulated in order to adequately explain human behavior. Since few attempts have been made to catalog the disparate treatments of implicitness that have appeared across the cognitive sciences, an initial hope in commissioning the Handbook was that it might offer a unifying, somewhat definitive account of implicitness. A theorist reviewing the project proposal for Routledge was rather pessimistic about the likelihood of such a unification, and now, after assembling this collection, it appears that some of this pessimism was warranted. Still, several lessons can be drawn about the causes contributing to this lack of a unified conception of implicitness. Such lessons can shape more fecund characterizations of implicitness for future research.

5.1  Contrast-Class-ification and the Terminological Morass Perhaps it should not be surprising that in past discussions of the I/E distinction, relatively little has been made explicit in characterizing implicitness. Much of this can be traced to the effects of contrast-class-ification, especially in its most rote forms (in which ‘un-’ or ‘non-’ is simply tacked on to a target term). If ‘implicit’ is characterized as non-explicit, then it seems to follow that all paradigmatic explicit features will be absent in the implicit phenomenon. Yet, to the extent that 21

J. Robert Thompson

these explicit features go unenumerated, then it is not even clear which features are expected to be absent in the implicit phenomenon. When the features happen to be enumerated, the explicit features often appear largely unanalyzed except via subsequent lists of features that are themselves often under-analyzed or the subject of additional contrast-class-ification. Table 0.1 is not intended to be exhaustive, but it aims to encompass many of the distinct purposes to which characterizations of the implicit are put. Notably, this table involves less

Table 0.1  Proposed Implicit and Explicit Features Implicit Manner of Processing Automatic Efficient Fast Parallel Inflexible Intuitive Associative operations Holistic Involves application of heuristics Manner of Content, Scope Associationist content Unconscious, preconscious, nonconscious Nonconceptual Not articulated/not articulable Belief-based, recognition-based, experiencebased decisions Pragmatic reasoning Interactional intelligence Prone to influence from tangential contextual associations, beliefs, and relations Basic emotions Hard constraints on the type of information processed Action-oriented/directed Often not connected to any single personally chosen goal or outcome Manner of Access/Connectivity Operates at subpersonal level Should appear via indirect measures, but not on direct measures Opaque to subject – no signature thoughts/ feelings produced by its operation Not available for elicited (verbal) judgments Hidden, masked, latent, covert (Relatively) Undemanding of cognitive capacity, working memory Independent of cognitive ability, general intelligence

Explicit Controlled, volitional Inefficient Slow Sequential, Serial Flexible Deliberative, reflective, analytical Rule-based operations Analytic Can involve application of valid rules Propositional content Conscious Conceptual Articulated/articulable Consequential, reason-based decisions Abstract, logical reasoning Analytic intelligence Able to disengage tangential associations and relations Complex emotions Soft constraints on the type of information processed Not always action-oriented When goal-directed, often connected to a single personally chosen goal or outcome May operate at the personal level Should appear via direct measures Some signature thoughts or feelings may be produced, even if the inside operations remain opaque Available for elicited (verbal) judgments Open, overt (Relatively) Demanding of cognitive capacity, working memory Correlated with cognitive ability, general intelligence

22

Introduction

Implicit

Explicit

Independent of language Independent of normative beliefs Informationally encapsulated May consist of a set of systems

Associated with language Influenced by normative beliefs Access to general knowledge Typically viewed as a single system

Manner of Activation/Cessation Resists strategic control Relatively automatic Triggered

Affords strategic control Never automatic (at most spontaneous)/deliberate Intentionally deployed

Manner of Modification Not easily altered Limited intervention possible, typically via nonrational means Impervious to verbal instruction Altered by training, tuning

Malleable Rational intervention possible Responsive to verbal instruction Altered by learning

Manner of Origin, Development Acquired via biology, exposure, and personal experience Early developing, under relatively tight environmental and genetic constraints Universal among humans Evolutionarily old/early Similar to (nonhuman) animal cognition/ conserved across species

Acquired via cultural and formal tuition Late developing, onset achieved through learning or conceptual development Variable (by culture and individual) Evolutionarily recent/late Distinctively human/unique to humans

Qualitative/Normative Characterizations Natural Intuitive Arational or irrational Quick and inflexible modularity Passive Biased responses Shallow Unfamiliar Unhesitating, immediate, undoubting Unintelligent, nonintelligent Perceives, apprehends, intuits

Normative Deliberative, reflective, analytical Rational Effortful, or simplified via downward modularization/ritualization Active Normative responses Deep Familiar Reflective, cautious Intelligent Judges, infers or reasons, knows

contrast-class-ification than is typical, so although not every contrast-class-ification has been eliminated, renderings from different theorists from different debates have been provided in order to flesh out some of the relevant contrasts being made between the implicit and explicit features.14 When viewing the more fleshed out features in the context of this fairly comprehensive list, it should be clear just how many distinct ways there are to manifest implicitness. A unified account is not ruled out by the distinctness of these features, but the features are suited for quite particular explanatory purposes. Hence, it should be abundantly clear that researchers who invoke implicit elements need to specify which of these features are operative in the explanation on offer. Generic appeals to implicitness will not do. 23

J. Robert Thompson

The alignment problem also reasserts itself. Since even prominent Two Systems theorists have reoriented their claims so as to weaken expectations about alignment (Evans and Stanovich 2013),15 it should not be overly contentious to allow for potential divergence among the features. Nothing in the volume proves that the items in a column cannot be coextensive, but one can no longer identify a single feature from one column and assume that the others come along for free.16 To be clear, the Handbook leaves open the possibility that Two Systems approaches might bear fruit, that ‘implicit cognition’ captures a separate type or style or system of thinking, but it devotes more attention to assessing the extent to which implicit features are present in cognition, whether or not they are joined by other implicit features. Embracing this more ecumenical approach should lead researchers to identify whichever features best capture the phenomenon. This may lead to unanticipated combinations of classic features or the need to characterize novel features that bear little resemblance to the classic ones. If there will most likely be cases of nonalignment to explain, it should become a priority to develop the resources to explain how a phenomenon can have some features from each column rather than operate with the assumption that the presence of any feature from one column indicates or warrants the presence of the rest of the features from that column. Once an expectation of alignment is set aside, theorists take on an additional burden of specifying which features are doing the explanatory work in any invocation of implicitness, and which ones are not. That is, theorists must be able to articulate what they don’t mean or include when characterizing some element as implicit. Although additional clarity enhances academic debates, there are more significant reasons to seek a clearer articulation of the dispensable and indispensable features of implicitness. Within the context of research into the implicit, there has been a marked tendency to treat any indicators of implicitness as a rather definitive reason for or against an account. But, once alignment is not a default expectation, certain such approaches need to be fundamentally reconsidered. It will no longer suffice for researchers to cast doubt on the legitimacy of some implicit phenomenon by showing that it lacks some feature (say, speed) or possesses some feature (controllability). Or, for example, suppose that research into implicit bias suggests more propositionality than once expected (e.g., Ch. 8). Researchers need frameworks that enable them to articulate whether this discovery shows that implicit bias isn’t legitimate, whether it isn’t implicit, or whether it is an example of implicitness that has features from both columns. Much more precision will be required both from those presenting the accounts and those challenging them, but more perspicuous explanations and experiments should result. In addition to issues in identifying which features count as implicit, a dizzying array of phenomena have been classified as ‘implicit’: tasks, measures, outcomes, conditions, states, attitudes, evaluations, judgments, processes, representations, and ways of understanding, inter alia. Yet, this list fails to include an entire range of implicit phenomena that involve a particular type of activity (e.g., implicitly remembering) rather than a state, entity, or situation (e.g., an implicit memory). This difference in what ‘implicit’ is taken to modify is rarely noted. Hence, it is often not clear whether the cognitive processing is what is being described as distinctly implicit, that is, some form of implicit cognition; or whether the processing itself is rather standard, but this cognitive activity operates over states that are themselves distinctly implicit, that is, some sort of implicit state; or whether some combination of the two is in play. The reader is encouraged to consider what significance the differences noted in Table 0.2 might suggest. As aforementioned, this noun/state, verb/activity difference is rarely acknowledged. It is not clear whether this amounts to a substantive difference or could serve as a basis for such a difference. Still, theorists should determine the extent to which these different articulations

24

Introduction Table 0.2  Implicit States and Implicit Activities Implicit + N

Implicitly + V

Representation Understanding Knowledge Grasp Process (-ing) Operation Inference Bias Memory Learning Belief Judgment Awareness Emotion Percept (-ion) Concept (-ion) Theory Cognition Measure (-ment) Stimulus Acquisition Simulation Attitude State Mechanism Capacity Skill Ability Task

Represent (-s)(-ed) Understand (-s), understood Know (-s)(-n) Grasp (-s) Process (-es)(-ed) Operate (-s)(-ed) Infer (-s)(-red) Bias (-es)(-ed) Remember (-s)(-ed) [retrieve, recall] Learn (-s)(-ed) Believe (-s)(-d) Judge (-s)(-d) Aware Feel (-s), felt Perceive (-s)(-d) Conceive (-s)(-d) Theorize (-s)(-d) Cognize (-s)(-d) Measure (-s)(-d) Stimulate (-s)(-d) Acquire (-s)(-d) Simulate (-s)(-d)

might matter to their accounts. As a first step, there appears to be some tendency for those who account for inarticulability in terms of expressibility to prefer the adverbial form modifying activities. This tendency likewise aligns with those who resist intellectualism, question the reducibility of know-how to know-that, and identify the mind in 4E terms (as embodied, enactive, extended, and embedded). These theorists seem keener on describing an agent as, for example, implicitly theorizing rather than as possessing an implicit theory; or knowing implicitly rather than possessing implicit knowledge.17 Every debate within the academy faces terminological and conceptual obstacles, but the debate over the I/E distinction and the viability of implicit cognition has suffered more than most. In the end, there may be too little unity among those who postulate implicit cognition to expect a unified account of implicitness. With such differing emphases and methodological commitments, there may be little that could hold these accounts together other than a rejection of the primacy of the explicit. It is perhaps too soon to tell.

25

J. Robert Thompson

5.2  The Ghosts of Intellectualism In many ways, the research described in the Handbook emerges from two distinct engagements with the limitations of intellectualism. Intellectualism appears largely successful in explaining a significant portion of conscious thought, but it founders in explaining what to do about the swaths of human thought that lack overtness.18 Out of this foundering, two very distinct motivations for implicit cognition arise. Intellectualists should concede that many intelligent behaviors cannot be explained in terms of overt deliberation.19 Hence, if intellectualism is to succeed as a viable explanatory approach, it will need to offer an account of implicit processes (in the Fodor-Chomsky sense) to undergird the myriad cases where it seems that overt explanations simply will not do. These processes lack some level of inferential integration and conscious awareness, but there is no commitment to the notion that they involve some knowledge that is fundamentally nondeclarative. Indeed, the representations utilized in these explanations are often thought to require a certain level of explicitness in terms of structure and content (Fodor 1975, 1981), but need to lack integration and awareness. On the other hand, anti-intellectualists take the foundering of intellectualism to warrant a shift in focus to uncover rich varieties of nondeclarative and nondeliberative thought. The processes that are postulated to explain the non-overt cases provide not a vindication of intellectualism but a pronounced alternative to it. Implicit processes are those that bring about intelligent behavior in an altogether distinctive manner. Many of the conflations and misunderstandings in debates over implicit cognition can be traced to failures in recognizing these very distinctive motivations and origins for postulating implicit cognition. Returning to Ryle’s terminology, the intellectualists aim to explain why the escorts of behavior aren’t easily reported or how apparent discordance arises. This is not because the items escorting one form of behavior are nondeclarative, but because they lack the level of inferential integration and conscious awareness that affords the reporting done on the other escorts. Alternatively, the anti-intellectualists take the lack of transparency into the nature of any escorts to suggest that either there are no mental escorts or that they must have inexpressible features. Both intellectualists and anti-intellectualists seek to appeal to states and processes that involve implicit elements, a fact that suggests that the term ‘implicit’ might be even more problematic than other conceptually fraught terms. Intellectualists view implicit phenomena as a salvation for intellectualism; anti-intellectualists view them as a definitive rival to intellectualism. Those researching the implicit need to keep these dual motivations in mind and clarify their positions accordingly. Once one gives up on robust alignment, theorists will need to be especially mindful of the arguments they can offer around expressibility. To the extent that intellectualists posit a range of implicit phenomena that are isolated but not necessarily inexpressible (the knowledge could count as knowledge-that), the anti-intellectualists will need to clarify what factors motivate their position. Arguments that draw upon a subject’s lack of phenomenological awareness of these intellectualist roots, escorts, knowledge structures, etc. seem of questionable significance for implicit cognition, so other considerations need to take center stage. It may be that many of the features that anti-intellectualists found problematic in classical versions of intellectualism (i.e., those that involve consciousness, reportability) will be absent in implicit renderings of intellectualism. But, if not, then at least the anti-intellectualist and intellectualist takes on implicit cognition can have clearer views about what the opposing stance takes to differentiate it from the other. Throughout this introduction and in the chapters that follow, it should become clear that the debate should be not whether implicit phenomena are needed at all but what features will allow us 26

Introduction

to address the explanatory challenges that motivate the need for such posits in the first place. If the behaviors of interest are neither mere reflexes nor appear ex nihilo, yet lack explicit reportable roots, what are the roots really like? How do they give rise to the behaviors in this essential, albeit covert manner? These questions will be addressed in the chapters that follow.

6. Coda The original plan for this volume included a few chapters that did not materialize due to the stresses and uncertainty stemming from the COVID-19 pandemic. Authors who committed to contribute chapters were, for a number of unique reasons, unable to submit them and the conditions of the pandemic prevented timely replacements. Although the topics selected for separate chapters are all covered to some extent in the chapters that appear, some of these topics would have been covered more expansively over multiple chapters. Readers may want to consult the following Routledge Handbooks and other resources for further information about several of these areas in which implicit cognition plays a role: The Routledge Handbook of Philosophy of Skill and Expertise (Edited by Ellen Fridland and Carlotta Pavese) The Routledge Handbook of Philosophy of Animal Minds (Edited by Kristin Andrews and Jacob Beck) The Routledge Handbook of the Computational Mind (Edited by Mark Sprevak and Matteo Colombo) The Routledge Handbook of Embodied Cognition (Edited by Lawrence Shapiro) Buckner, C. 2019. “Deep learning: a philosophical introduction”. Philosophy Compass, 14: e12625. Peters, M., Kentridge, R., Phillips, I. and Block, N. 2017. “Does unconscious perception really exist? Continuing the ASSC20 debate”. Neuroscience of Consciousness, 3: nix015.

Notes 1 The plan for this introduction is to introduce explicitness and implicitness gradually with the help of terms like ‘open’ and ‘overt’, ‘hidden’ and ‘covert’. At this point in the discussion, one can take ‘overt’ to mean something that is verbalized or reported out loud. As this volume will show, it is frustratingly difficult to pin down such notions because the sense of overtness that theorists will require can’t simply be what is actually verbalized or reported. It will need to cover cases where some thought could have been verbalized or reported, or was poised to be verbalized or reported. Basically, it is assumed that there is a basic form of verbal expression that accurately captures human thought. That is, there is a sort of thought process or structure that humans overtly verbalize and accurately report in such speech, and theorists can understand human thought by understanding what the overt expressions of thinking or thought would have looked like, if they had been generated. This assumption cannot really be articulated without quickly exposing its flaws and limitations, even though it is seemingly central to our folk conception of how we verbally capture basic human thought. The numerous odd ways in which cognitive science both depends upon and circumvents, ignores, or undermines this assumption drive many of the debates within this volume. 2 Evan Westra reminded me that speaking about mental states is considered taboo in some cultures (e.g., Robbins and Rumsey 2008 and that special issue of Anthropology Quarterly). One of the veins running throughout the Handbook is that we need to reconsider how complicated the relationship can be among what cognitive resources people actually use, what resources they take themselves and others to be using, and what they publicly or explicitly report themselves and others to be using. Such cultures offer intriguing insights into how much variation exists in how cultures treat the fidelity of mental state talk.

27

J. Robert Thompson 3 At least in English, this phrase suggests that the topic of behavior and discussion are related and at odds with one another. The phrase is not intended to simply identify cases where people speak while performing some nonverbal behavior. 4 This volume focuses on Western European cultures and theorists. For a start at how these debates might engage with non-Western cultures, see Buchtel and Norenzayan (2009) and for earlier Western accounts, see Crane and Patterson (2000) and Hendrix (2015). 5 For background on such approaches, as well as discussion of their pitfalls, see De Neys (2017), Evans and Frankish (2009). 6 Several engaging accounts of the I/E distinction are available (Brownstein 2018; Collins 2010; Davies 2015; Dienes and Perner 1999; Gascoigne and Thornton 2013), but Brownstein’s characterization will suffice for what’s discussed in the Handbook. 7 It is perhaps worth noting that these features seem most suited in applying to psychological processing rather than static or standing states. This differs from consciousness or articulability, which seem applicable to processes and states. 8 Things are even more complicated than described here. Several additional facets of automaticity have been identified (Mandelbaum 2015; Moors et al. 2010) 9 ‘Verbal’ is used here in an inclusive sense to include not just vocal and auditory phenomena, but any medium that is sufficiently linguistic, for example, sign language or written language, perhaps iconic or pictorial media. There are long-standing debates about what formats are capable of capturing whatever targets are identified (Fodor 1981; Haugeland 1998). 10 The sense in which processes or knowledge structures are isolated is sometimes described in terms of participation: it is said that they don’t appear to participate in many projects (Miller 1997) or are largely “harnessed to [a] single project” (Wright 1986: 227). In this sense, what is implicit cannot be cognitively integrated, will rarely or never be accessed by some other system for some other purpose, and will rarely (if ever) be subject to verbal report. 11 The following chapter descriptions fuse the editor’s terminology and the chapter authors’ own descriptions of their contributions. Text in italics mirrors the authors’ own wording but isn’t typically directly quoted from any single passage. 12 For background on Ryle and various misapprehensions about his work, see Tanney (2022). Some readers might assume Ryle would question whether any intelligent skills or actions have roots in the mental-etiological sense discussed here. But, what’s most critical to the current context is his insistence that not all intelligent skills and behavior have the sorts of roots under discussion, and his willingness to raise similar doubts for many sectors of skills and behavior. One needn’t take a stand on what his ‘official views’ might be, or how radical a position he might be able to defend to see the power behind his sort of challenge to a pervasive intellectualism. 13 Distinctions between ‘tacit’ and ‘implicit’ will not be stressed in this introduction. Some chapters will follow suit, whereas others will put the terms to different uses. 14 This table was assembled over many years, drawing from commonly available texts, lists, and tables (Apperly 2011; Brownstein 2018; Evans and Stanovich 2013). Much of this can also be found in the chapters in Evans and Frankish (2009) by Buchtel and Norenzayan, Carruthers, Evans, Evans and Frankish, and Samuels. 15 Evans and Stanovich (2013) distinguish between defining features and typical correlates to handle misalignment. The former (automaticity, freedom from working memory) are definitive and determine something’s implicitness, the others are simply features that typically or commonly appear with implicit phenomena. 16 This entire list has not been endorsed by any single theorist alongside an alignment claim, so it would not be fair to impugn any particular approach with the prospect that this particular list would fail to achieve near-perfect alignment. It is a collection drawn from multiple sources and different debates. As mentioned throughout, near-perfect alignment seems increasingly unlikely, but this is an open empirical question. 17 There are several approaches to studying mentality that attempt to replace a view that posits a SENSETHINK-ACT model with one that more closely matches a SENSE-ACT model (see suggested work in Section 6). Likewise, they reject accounts that paint cognition as passive or spectatorial. These perspectives are more likely to downplay or reject appeals to representational notions and deny that the roots or escorts have expressible content or intellectualist renderings. Hence, perhaps it is not surprising that these accounts might find the adverbial form more useful in capturing the implicit.

28

Introduction 18 At this point, ‘overt’ is used as it was earlier in the introduction to suggest the sort of thought that is either verbally reported or is of an internal format that is poised to be verbally reported in a way that genuinely reflects the nature of the underlying thought. 19 The ‘should’ here is used to suggest what a reasonable intellectualist ought to do when confronted with the problematic assumption discussed in note (1) and the prospect of actually having to construct overtly reportable intellectualist renderings (perhaps something akin to a verbal recipe for each behavior) that would not be overly strained. It could be that some intellectualists would resist such an admission and insist that all intelligent behaviors require an intellectualist account, strained appearances or not. There are few, if any, who might insist upon such a thing. A more likely philosophical insistence might involve the claim that an intellectualist account could be given. Nevertheless, it is very difficult to find careful articulations about the intended scope of intellectualism. Versions of the classical computational theory of mind suggest a very wide scope (Newell and Simon 1976), but many statements about scope are taken as something like a call to arms: why not see just how much of intelligent behavior can be accounted for in these terms? Intellectualism is not obligated to explain every behavior, but it had better not fail to explain whatever its core cases were supposed to be, or fail for Rylean reasons involving the incompleteness of any such declarative accounts.

References Anderson, J. 1983. The architecture of cognition. Cambridge, MA: Harvard University Press. Apperly, I. A. 2011. Mindreaders: the cognitive basis of “Theory of Mind”. New York: Psychology Press. Attneave, F. 1960. “In defense of homunculi”. In W. Rosenblith, ed., Sensory communication. Cambridge, MA: MIT Press: 777–782. Bargh, J. A. 1994. “The four horsemen of automaticity: awareness, intention, efficiency and control in social cognition”. In R. Wyer and T. Srull, eds., Handbook of social cognition. Hillsdale, NJ: Lawrence Erlbaum Associates: 1–40. Brownstein, M. 2018. The implicit mind: cognitive architecture, the self, and ethics. New York: Oxford University Press. Buchtel, E., and Norenzayan, A. 2009. “Thinking across cultures: implications for dual processes”. In J. S. B. T. Evans and K. Frankish, eds., In two minds: dual processes and beyond. New York: Oxford University Press: 217–238. Chomsky, N. 1966. Cartesian linguistics: a chapter in the history of rationalist thought. New York: Harper and Row. Churchland, P. M. 1981. “Eliminative materialism and the propositional attitudes”. The Journal of Philosophy, 78: 67–90. Collins, H. 2010. Tacit and explicit knowledge. Chicago: University of Chicago Press. Crane, T., and Patterson, S., eds. 2000. History of the mind-body problem. London: Routledge. Cummins, R. 1983. The nature of psychological explanation. Cambridge, MA: MIT Press. Cummins, R. 1989. Meaning and mental representation. Cambridge, MA: MIT Press. Davies, M. 2015. “Knowledge (explicit, implicit, and tacit): philosophical aspects”. In J. D. Wright, ed., International encyclopedia of social and behavioral sciences. 2nd edn. Oxford: Elsevier: 74–90. De Neys, W. ed. 2017. Dual process theory 2.0. London: Routledge. Dennett, D. 1978. Brainstorms. Cambridge, MA: MIT Press. Dienes, Z., and Perner, J. 1999. “A theory of implicit and explicit knowledge”. Behavioral and Brain Sciences, 22: 735–808. Dretske, F. 1981. Knowledge and the flow of information. Cambridge, MA: MIT Press. Evans, J. S. B. T., and Frankish, K., eds. 2009. In two minds: dual processes and beyond. New York: Oxford University Press. Evans, J. S. B. T., and Stanovich, K. E. 2013. “Dual-process theories of higher cognition: advancing the debate”. Perspectives on Psychological Science, 8: 223–241. Fodor, J. A. 1968a. Psychological explanation: an introduction to the philosophy of psychology. New York: Random House. Fodor, J. A. 1968b. “The appeal to tacit knowledge in psychological explanation”. The Journal of Philosophy, 65: 627–640. Fodor, J. A. 1975. The language of thought. Cambridge, MA: Harvard University Press.

29

J. Robert Thompson Fodor, J. A. 1981. RePresentations: philosophical essays on the foundations of cognitive science. Cambridge, MA: MIT Press. Gascoigne, N., and Thornton, T. 2013. Tacit knowledge. Durham: Acumen. Gendler, T. S. 2008. “Alief and belief ”. Journal of Philosophy, 105: 634–663. Gleitman, H., Gross, J. J., and Reisberg, D. 2010. Psychology. 8th edn. New York: W. W. Norton  & Company. Haugeland, J. 1998. Having thought: essays in the metaphysics of mind. Cambridge, MA: Harvard University Press. Hendrix, J. S. 2015. Unconscious thought in philosophy and psychoanalysis. New York: Palgrave Macmillan. Hirstein, W. 2005. Brain fiction: self-deception and the riddle of confabulation. Cambridge, MA: MIT Press. Lycan, W. 1991. “Homuncular functionalism meets PDP”. In W. Ramsey, S. P. Stich, and D. Rumelhart, eds., Philosophy and connectionist theory. Hillsdale, NJ: Laurence Erlbaum Associates: 259–286. Mandelbaum, E. 2015. “The automatic and the ballistic: modularity beyond perceptual processes”. Philosophical Psychology, 28: 1147–1156. Miller, A. 1997. “Tacit knowledge”. In B. Hale and C. Wright, eds., A companion to the philosophy of language. Oxford: Blackwell. Millikan, R. 1984. Language, thought, and other biological categories: New foundations for realism. Cambridge, MA: MIT Press. Moors, A., Spruyt, A., and De Houwer, J. 2010. “In search of a measure that qualifies as implicit: recommendations based on a decompositional view of automaticity”. In B. Gawronski and B. K. Payne, eds., Handbook of implicit social cognition: measurement, theory, and applications. New York: Guilford Press: 19–37. Newell, A., and Simon, H. A. 1976. “Completer science as empirical inquiry: symbols and search”. Communications of the association for computing machinery, 19: 113–126. Nisbett, R. E., and Wilson, T. D. 1977. “Telling more than we can know: verbal reports on mental processes”. Psychological Review, 84: 231–259. Nosek, B. A., Hawkins, C. B., and Frazier, R. S. 2011. “Implicit social cognition: From measures to mechanisms”. Trends in Cognitive Sciences, 15: 152–159. Payne, B. K. 2012. “Control, awareness, and other things we might learn to live without”. In S. T. Fiske and C. N. Macrae, eds., The SAGE handbook of social cognition. Thousand Oaks, CA: Sage: 12–30. Polanyi, M. 1966/2009. The tacit dimension. Chicago: University of Chicago Press. Reber, A. S. 1993. Implicit learning and tacit knowledge: An essay on the cognitive unconscious. New York: Oxford University Press. Robbins, J., and Rumsey, A. 2008. “Introduction: cultural and linguistic anthropology and the opacity of other minds”. Anthropological Quarterly, 81: 407–420. Ryle, G. 1945. “Knowing how and knowing that”. Proceedings of the Aristotelian Society, 46: 1–16. Ryle, G. 1949. The concept of mind. Chicago: University of Chicago Press. Stich, S. P. 1983. From folk psychology to cognitive science: The case against belief. Cambridge, MA: MIT Press. Tanney, J. 2022. “Gilbert Ryle”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Summer 2022 edn. https://plato.stanford.edu/archives/sum2022/entries/ryle/. Wiers, R. W., and Stacy, A. W., eds. 2005. Handbook of implicit cognition and addiction. Thousand Oaks, CA: Sage. Wright, C. 1986. “Theories of meaning and speakers’ knowledge”. In Realism, meaning, and truth. Oxford: Blackwell.

30

PART 1

Defining Features? Identifying Implicitness Among Cognate Notions

1 IMPLICIT MENTAL REPRESENTATION William Ramsey

1. Introduction One of the central divisions within both commonsense psychology and scientific theories of cognition is the distinction between explicit and implicit (tacit) mental representation.1 Unfortunately, it is also a distinction that lends itself to a variety of different interpretations. In philosophy and cognitive science, there is currently no real consensus about where or even how one should draw the line between explicit and implicit representation, as different investigators promote alternative demarcating criteria. Moreover, for some it is thought to be a distinction between different types of representation, while for others it is a distinction demarcating different types of employment of roughly the same representational structures. Adding to the confusion is the fact that for some, but not all, the distinction is closely linked to older distinctions between things like explicit and implicit knowledge or belief. A survey of the literature suggests it would be more accurate to refer not to the explicit-implicit representation distinction but instead to a range of different distinctions arrayed along a spectrum of explicitness or implicitness. Despite all this ambiguity, appealing to implicit representations can nevertheless be useful when this criterial diversity is taken into account. My goal in this chapter is to present some of the more common ways in which people think about implicit representation and talk about how these notions are related. I also want to consider the potential explanatory value of abandoning some of these notions of implicit representation in order to reduce the conceptual clutter. I’ll begin by laying out a relatively uncontroversial, paradigmatic case of explicit representation. Then I’ll consider some of the popular ways in which implicit representation is thought to diverge from this, including a brief survey of four prominent ways that explicit and implicit mental representation are demarcated. Finally, I’ll recommend dropping a couple of these notions of implicit representation for the sake of conceptual simplicity and theoretical coherency.

2.  Common Ways of Demarcating Implicit Representation Since the explicit-implicit representation distinction can be located at various points along a spectrum, let’s start by looking at one of the endpoints, the explicit end. A non-contentious version2 of an explicit mental representation would be an occurrent, consciously attended DOI: 10.4324/9781003014584-3 33

William Ramsey

propositional attitude, like my currently entertained belief that my middle name is Max. Let’s assume that this belief is instantiated in the brain by some sort of distinct neural state or structure that, at some level of analysis, functions to represent the proposition that my middle name is Max. I am consciously aware of this state, and it actively functions in an array of different causal relations, including the production of linguistic and other types of behavior. For instance, if asked, this belief plays a crucial role in causing me to announce out loud that my middle name is Max. I have some control over the belief in that I am deliberately considering it and can choose to stop attending to it. As a paradigm of explicitness, we should assume that an investigator could, in principle, isolate this representational state/structure somewhere within the neural or computational architecture of my brain. If we were to make a list of the sort of “explicit-making” features suggested by this example it would look something like this: • • • • • • • •

has specific propositional content is conscious is attended to is active can be expressed via linguistic pronouncements is deliberately entertained and is under the agent’s control is physically or computationally identifiable, concrete and discrete plays a specific causal role in cognitive processing

If representations with these features are paradigmatically explicit mental representations, then we can start to understand implicit representations as representations that lack one or more of these properties. However, while all of these features have appeared in the literature defining explicitness, some have traditionally been much more central to the explicit/implicit distinction than others. For instance, as we’ll see in more detail later, it is quite common to treat conscious, introspective access as a defining characteristic of explicit mental representation. For many the distinction between conscious and unconscious mental representation defines the distinction between explicit and implicit mental representation. Yet at the same time, other features on the list have been less relevant, put forward by only a few outliers. For example, Payne and Cameron (2013), borrowing from Bargh (1994), suggest that implicit representations are representations over which we lack control. However, this factor is a much less popular demarcating criterion for obvious reasons. We often have very little control over explicit representational states, such as visual representations, as Locke famously emphasized. It is also not uncommon for individuals to have explicit beliefs or memories that they wish they could somehow avoid but cannot. It seems voluntary control is a poor candidate for a demarcating criterion for the implicit/explicit distinction. Moreover, some of these features are arguably better understood as effects or consequences of explicitness and not as defining characteristics. For example, some have suggested that linguistic expressibility is an essential dimension of explicit representation, especially with regard to explicit knowledge, and thus implicit representation would be the representation of information that is not expressible (see for example, Dummett 1991; Davies 2001). Yet most people in the cognitive science community would insist that pre-linguistic children have explicit mental representations, despite no expressive capacity. We also know that people with certain types of aphasia, such as extreme forms of Broca’s aphasia, have a very limited ability to linguistically express the content of their occurrent, very explicit thoughts. The breakdown here is not in terms of explicitness of the mental representations but instead in the causal link between the representation and its linguistic expression. Linguistic expression thereby seems to be more of 34

Implicit Mental Representation

a reliable indicator or product of true explicitness, not a defining characteristic.3 It is a consequence of other, more basic dimensions of explicitness, and these more basic dimensions should be our primary focus. So what are these more basic, fundamental criteria that have been more central to our understanding of implicit versus explicit representation? Generally speaking, there are roughly four common ways in which people have defined implicit representation and distinguished it from explicit representation: 1) Explicit = conscious

Implicit = unconscious (or subconscious)

2) Explicit = active

Implicit = dormant

3) Explicit = existing representational vehicle

Implicit = implied by existing ­ representational vehicles

4) Explicit = instantiated by a distinct state

Implicit = instantiated by broader ­ functional architecture

Before discussing each of these demarcation criteria, it is worth considering some interesting dimensions of this collection. For example, some of these distinctions are potentially overlapping or at least mutually consistent, whereas others appear to be incompatible. It is possible for a representation to be implicit in the sense of being both unconscious and dormant. Indeed, presumably most of our beliefs currently have both features, and it is likely that the two are closely related. But some of these criteria are not compatible. A mental representation that is implicit in the sense of being merely entailed by other things explicitly represented, but has never been tokened, cannot be the same as a previously instantiated state that is currently unconscious or dormant. It is also worth noting the different ways that these sets of demarcators present different answers to the question, do explicit and implicit representations share the same representational vehicle in different modes, or are the vehicles different? With the first demarcation criterion – conscious versus unconscious, it seems either answer is possible. That is, the difference between a conscious and an unconscious representation could either be due to the same representational vehicle in different modes (i.e., a vehicle in either conscious mode or unconscious mode) or due to different sorts of representational vehicles (i.e., a conscious representational structure/state with content P is a different sort of structure/state than the unconscious representational structure/ state with content P). However, in the case of active versus dormant representations, it seems that the distinction must involve the same representational vehicles in different conditions – after all, the difference between active and dormant status just is a difference in modes. By contrast, the latter two sets of demarcators require different types of representational vehicles undergirding the difference between explicit and implicit representation. That is, an explicitly tokened representation of P is necessarily a different sort of state than a representation of P in the sense of P being implied by other explicit representations. And a representational vehicle that is a distinct, discrete structure or state is necessarily different from a representational vehicle that is encoded in the overall cognitive functional architecture. Let’s consider each of these types of demarcators in more detail.

2.1  Conscious Versus Unconscious Representation This first distinction captures a common way in which both philosophers and psychologists frame the explicit/implicit distinction, as (roughly) corresponding to the conscious/unconscious mental state distinction (see Armstrong 1968; Carruthers 2018; Frankish 2004; Schacter 1987, 1992; Reber 1967, 1993; and Augusto 2010 for a good overview). Although anathema 35

William Ramsey

to Descartes, the idea that there can be unconscious states like desires and beliefs has become increasingly commonplace, in part because of the work of Freud.4 In philosophy, unconscious mental representations are sometimes described as lacking “state consciousness”, although there are different views of what this entails (Rosenthal 1986). While an unconscious representation might also be dormant, the identification of implicit representation with unconscious representation makes sense only if it is the unconsciousness that really matters. In cognitive science, there is now a huge literature on the acquisition and utilization of unconsciously represented information in such diverse cognitive processes as memory and learning (e.g., Allen and Reber 1980; Schacter 1992), vision (e.g., de Gelder et al. 2002; Cowey 2004), inference (e.g., Gigerenzer and Gaissmaier 2011; Kahneman 2013), language processing (Chomsky 1965; Reber 1967) and moral and social reasoning (Sinnot-Armstrong et al. 2010; Payne and Cameron 2013). To just take one famous example, Chomsky’s well-known work in linguistics invokes an internal grammar that represents syntactic rules that serve to govern language processing (Chomsky 1965). This grammar is famously not accessible to introspection, although its nature can be explored by probing various linguistic intuitive judgments. In more recent years, the notion of unconscious information processing has served as the cornerstone of dual-processing accounts, where mental representations are invoked to explain Type 1 processing that is not accessible to consciousness (Evans and Frankish 2009). One important area where the conscious-unconscious distinction looms large and that merits special attention is the growing body of work in sociology and social, developmental, cognitive and moral psychology focused upon implicit bias as a source of prejudice against various groups. Here there are different perspectives on the sort of implicit representations that give rise to prejudicial behavior without the agent’s awareness of their underlying role. For example, some accounts treat implicit representations as attitudes (Greenwald and Banaji 1995), others as beliefs (Egan 2008), others invoke Gendler’s (2008) notion of aliefs and still others treat these as forms of mental imagery (Nanay 2021). It is likely that a very significant proportion of future research on implicit representation will be in this area (see Brownstein 2019, for an excellent overview). It should be noted that a small minority of philosophers deny the possibility of entirely unconscious mental representations, as such, because they believe that intentional content is at least partly grounded in phenomenal consciousness. For example, Searle (1992) argues that there are specific ways in which things are represented – the “aspectual shape” of representations – and that this requires consciousness. For Searle, talk about unconscious mental representations should be treated as talk about dispositions for having conscious mental representations.

2.2  Active Versus Dormant Representation A different way to demarcate explicit and implicit mental representations is to ground it in the distinction between active and dormant mental representations. Even if the dormant representation is also unconscious, here it is the dormancy that matters. Moreover, with this distinction dormant mental states need to be actual states or entities, and not merely some sort of dispositional condition; otherwise, this distinction threatens to collapse into other dispositional accounts, like the functional architecture account described later. Traditional computational accounts of cognition that suggest mental representations are symbol-like states that can be, in some computational sense, “stored” in retrievable formats, provide one appealing model of dormancy (Fodor 1975; Pylyshyn 1984). If our brains really do possess computational symbols that can be put away in some sort of memory cache, only to be reactivated later, this provides a nice model for understanding the difference between a dormant representation and its explicit re-engagement (see also Tulving and Craik 2000). 36

Implicit Mental Representation

Apart from classical computational notions of stored information, there are other, perhaps more biologically motivated, conceptions of dormant representations. On some accounts of cognitive representation, specific neurons, or perhaps clusters of neurons, serve as indicators of some condition, becoming highly active (more or less) when and only when that condition obtains (e.g., Dretske 1987; Lettvin et al. 1959). When the neurons are dormant, researchers would typically characterize them as still having the functional status of being neural representations of whatever they indicate when active. In a sense, they are like a smoke alarm when no smoke is present and are thereby inert. Along similar lines, this notion of dormant versus active representations is employed by certain localist theories of connectionist representation, where the activation of individual nodes or units is thought to involve the activation of a representational entity with specific content, but that was previously inert (McClelland 1981).

2.3  Existing Representations Versus Potentially Inferred Representations A third way to regard implicitness assumes the nonexistence of any actual representational vehicle with the relevant representational content. Instead, the represented content is entailed or implied by other already-existing representations. Unlike the sort of dormant representation just discussed, which is made explicit via a process of recall (or reactivation), the cognitive process that converts this sort of implicit representation into an explicit representation would be something more like inference (Field 1978; Audi 1994). As Dennett puts it, “let us have it that for information to be represented implicitly, we shall mean that it is implied logically by something that is stored explicitly” (1987: 216). Before reading this sentence, there is a sense in which you held the belief that planets are larger than your laptop, even though you had never actually considered that proposition. The sense in which you believed it is just the sense in which that fact is obviously entailed by a variety of other beliefs you possessed (in some way) about planets and laptops. For the most part, the belief is dispositional because, if prompted, you would immediately make the relevant judgments that generate that specific belief. As Dennett (1987) and Lycan (1986) note, however, one set of propositions (A) entailing another proposition (B) is not the same as having the psychological disposition to infer B on the basis of A, even when A is explicitly believed. After all, whether or not someone actually would make the relevant inference depends upon a variety of factors like the intelligence and rationality of the agent. Given the foibles of human reasoning, not only can there be a failure to see the entailment, but sometimes mental representations are generated by existing ones when they shouldn’t be, when new representations are not properly entailed or implied by existing ones. Hence, we can make a logical/psychological (or normative/descriptive, competence/performance) distinction with regard to this category, and focus either on actual logical entailments or upon the psychological dispositions to make certain judgments. In practice, when most people invoke this version of implicitly representing P, they tend to adopt a charitable level of rationality in the subject. If P is a fairly obvious implication or consequence of other things represented, then an implicit commitment to P is ascribed.

2.4  Distinct Representations Versus Functional Architectural Representation The fourth and final way in which philosophers and cognitive researchers have thought about the implicit/explicit representation distinction involves the difference between having a discrete, identifiable structure or state that functions as a representation versus having the overall 37

William Ramsey

functional architecture of the system encode information through its dispositional nature. As Clapin puts it, “functional architectures represent. They represent tacitly, and this tacit content is distinct from whatever is represented by the symbols” (2002: 299). Apart from Clapin, the philosopher who has probably done the most to promote this way of thinking about mental representation is Dennett, whose account of intentional state ascription famously points out that such ascriptions need not involve any “separately specifiable state of the mechanical elements for each of the myriad intentional ascriptions” (1978: 21). Dennett is plausibly interpreted as promoting the view that most ordinary ascriptions of mental representations depend exclusively on the overall dispositional nature of the underlying physical system. In cognitive science, this notion of implicit representation has become more prominent with advent of connectionist models, where it is assumed that after training the connection weights represent the network’s long-term knowledge base, also characterized as “superpositional” representation. Here, no individual unit or weight is thought to represent any particular, specifiable content. Instead, the smallest unit of semantic evaluability is the entire weight matrix, representing the entire background knowledge base. As Rumelhart and McClelland state, almost all knowledge is implicit in the structure of the device that carries out the task rather than explicit in the states of the units themselves. . .. Each connection strength is involved in storing many patterns, so it is impossible to point to a particular place where the memory for a particular item is stored. (1986: 75, 80) While connectionist accounts have bolstered this notion of implicit representation, it also has a long history in more conventional computational cognitive modeling. It was exactly this notion of implicit representation that Dennett highlighted when relaying a programmer’s description of a classical computational chess playing system that “thinks it should get its queen out early” (1978: 107). Dennett’s point was that there was no such “thought” explicitly represented anywhere in the system; the strategy was represented implicitly in the behavioral disposition of the computer’s functional architecture. Outside of artificial intelligence, this notion of implicit representation has been invoked to account for various cognitive phenomenon, including certain forms of implicit bias discussed earlier. Here, rather than treat the prejudicial responses as the manifestations of a discrete unconscious cognitive state, it is instead regarded as arising from underlying dispositional traits (Machery 2016). These four demarcation strategies are some of the more prominent ways in which people in cognitive science, including philosophers, demarcate explicit and implicit mental representation and thus define implicit representation.5 As I said at the outset, this abundance of ways to draw the implicit/explicit distinction can be navigated as long as writers are clear about which distinction(s) they are invoking in their theorizing and analysis. Nonetheless, it is worth considering if our understanding of cognition and mental representation could be enhanced by rethinking some of these ways in which we conceptualize implicit representation.

3.  Attenuating Implicit Representation In the last section, we examined four different ways in which implicit representations are distinguished from explicit representations, and thus four ways in which they are defined. Even with attention to the details, it is hard not to regard the current state of affairs as one of conceptual clutter, inviting confusion and cross-talk. Consequently, we should explore whether it is possible to streamline things by dropping one or more of these notions. 38

Implicit Mental Representation

The distinction between conscious and unconscious mental representation is, if anything, growing in importance in our theorizing about the mind. With the development of such trends as dual-process accounts of cognitive architecture and various mechanistic, sub-personal conceptions of representational processes, along with the explosion of interest in implicit bias, the notion of implicit mental representation qua-unconscious mental representation is increasingly indispensable (Ramsey 2007; Evans and Frankish 2009; Brownstein 2019). Much the same is true about the distinction between occurrent versus inactive representational states. In fields such as the philosophy of mind, cognitive neuroscience, artificial intelligence, and cognitive psychology and linguistics, it is hard to see how we could get by without sometimes positing implicit mental representations that take the form of dormant, inert structures. While these notions of implicit representation clearly have an essential role to play in our theorizing, I’d like to suggest that the same is not true about the other two notions: implicit representation by virtue of being implied by stored representations, and implicit representation by virtue of the dispositional nature of a cognitive system’s functional architecture. There are good reasons for thinking that we could abandon these ways of conceptualizing implicit representation without much theoretical or explanatory loss. In fact, there is some reason to think that much could be gained, as these notions of implicit representation are arguably based upon a conceptual mistake about the explanatory necessity of representations. The mistake comes from thinking that if we want to preserve a notion of implicit something, we must also invoke a notion of implicit representation. The “something” varies among different accounts, but it often is knowledge, belief, know-how, intelligence, competence, attitudes and so on. For our discussion here, I’ll just use the notion of implicit knowledge6 as a catchall phrase for the pertinent something that is mistakenly assumed to always require implicit representation. My proposal is that some of the ways we think about implicit knowledge should be cashed out not in terms of implicit representation, but instead in terms of the absence of representation. To begin, we should note a certain looseness in the way we talk about knowledge (and belief, intelligence, memory, etc.). We often describe all sorts of things as knowing something where we recognize this means little more than the possession of a certain capacity or competence. Does memory foam really “know” how to adjust itself to the contours of your body? Well, in one sense yes, as it has this simple capacity. Our concept of knowledge is sufficiently permissive to include types of systems that simply perform appropriately, and where we don’t care much about the nature of the internal mechanics and states, including the presence of representations. By contrast, our notion of representation is more closely tied to something playing a specific functional role. Representation is a functional kind – it involves something doing a certain job, like signifying, indicating, encoding and relaying information. Representations must have representational content. Consequently, the notion of representation demands more of a realist interpretation of representational entities or states than, say, our notion of knowledge (Ramsey 2021). The notions of implicit representation that are indispensable involve structures or states that have these ontologically real, functional qualities. An unconscious mental representation engages in many of the same causal, intentional and inferential relations as their conscious counterpart. And dormant cognitive representations are best understood as potentially functionally active structures, albeit it in a resting mode. In these two cases, implicit representations possess the core features required to actually qualify as representations. The same is not true, however, for the implied and architectural/dispositional notions of implicit representation. When someone implicitly knows P only by representing propositions that entail or imply P, that person, by definition, currently lacks any state that counts as the representation of P. And when the functional architecture of a cognitive system is thought to encode a body of knowledge holistically due to its dispositional properties, then here again 39

William Ramsey

there is nothing in that architecture functioning as a representation with the standard qualifying features. Attempts to invoke a notion of implicit representation in these cases must somehow explain how a type of representation can be present when there is nothing present that functions as the relevant representation! Consider another way in which we think about information, albeit along the lines of information transfer as opposed to information storage. Think about the following exchange. Question: “Is John at home?” Answer: “Well, his car is there and it is impossible for him to go out without his car.” The answer conveys the information that John is at home, but it does so implicitly, not explicitly. It explicitly states that John’s car is at his house and that John never goes out without his car. So how do we demarcate this difference between an explicit statement and an implicit statement in this sort of case? By noting the presence or absence of a sentence that unambiguously expresses (or fails to express) the proposition in question. An explicit statement of a proposition involves the tokening of linguistic representations (words) that, given their syntactic arrangement, literally represent that proposition. An implicit statement of a proposition does not involve the tokening of words that straightforwardly express that proposition; instead, the proposition is entailed by words that mean something else. Here, implicitness is demarcated by the absence of certain representational entities – that is, words. Think about another case. Question: “Did John tell you if he got the job?” Answer: “Yes, but not explicitly. It was implied by the way he carried himself, his smile and his presenting a thumbs up.” Here the answer involves a different sort of implicit information conveyance. The information is not conveyed linguistically, but by the physical demeanor and activity of the individual. The demeanor and behavior convey the information that John got the job. The information is implicit because it is conveyed without words. We don’t think the information is transferred by the use of special, implicit words. There are no words. Here again, implicitness is demarcated by the absence of representational entities. I recommend the same sort of demarcation approach when making sense of the two notions of implicit knowledge under consideration. Take the notion of implicitly knowing that P by believing things that strongly imply P. To even make sense of this form of implicit knowing, we need to assume that there is no explicit mental representation with the content of P. But why suppose there is a different sort of mental representation – an implicit one – with that content? Such a supposition would be just as misguided as supposing that when words are uttered that explicitly state a content implying some further proposition P, that the utterance must also involve something like implicit words that express P. For assertions, we can define a notion of implicit assertion by appealing to a lack of words. Similarly, we can define a notion of implicit knowledge by invoking a lack of mental representations that possess the implied content. The same point applies to the idea that the functional architecture of a cognitive system has a sort of implicit knowledge that allows it to perform in certain ways. Remember that with this sort of knowledge or know-how, there are no explicit representational entities or states with the representational content in question. Instead, as in connectionist networks, it is claimed that there is a type of implicit representation, holistically encoded in the overall functional architecture (e.g., in the weight configuration between nodes). In earlier work, I have strongly criticized this notion of implicit representation, arguing that it confuses representational states with dispositional states (Ramsey 2007). I am, however, willing to allow an innocent sense in which we can describe such systems as possessing implicit knowledge or possessing acquired know-how via a learning process. But the proper way to conceptualize such a type of implicit knowledge is as a type of knowledge that involves no representations. It is the same sense in which the automatic transmission in my car knows how to shift gears, or my washing machine knows how to go through various cycles. It is similar to the way in which a person’s appearance 40

Implicit Mental Representation

can communicate something without the use of words. Again, we wouldn’t claim this communication involves the employment of a type of implicit words. Instead we describe it as wordless communication. We should similarly describe this notion of implicit knowledge, based upon the system’s dispositions, as a kind of knowledge that is actually representationless. One might argue that any sort of knowledge requires information storage, and information storage requires representation, so if there is implicit knowledge there must also be implicit representation. But this would be to simply ignore the reality of distinctive senses of knowing that are idiosyncratic precisely because they are not based upon traditional notions of information encoding, but instead upon capacities and potentialities. Rather than retaining a distinction between explicit and implicit representations, and then try to demarcate the two by emphasizing of presence or absence of actual structures playing a representational role, it makes far more sense to invoke a distinction between explicit and implicit knowledge, and then demarcate the two by emphasizing the existence or nonexistence of representations. Insisting that all knowledge must be grounded in representation is as misguided as insisting that all communication must be grounded in words.

4. Conclusion In this chapter I’ve tried to make a little clearer some of the ways in which people have traditionally thought about implicit mental representation and I’ve also recommended some changes that I believe will improve the conceptual landscape. The main change I’ve proposed involves dropping conceptions of implicit representation that involve the absence of any sort of identifiable structure that functions as a representation with specifiable content. As I’ve argued, the key thing to note is that abandoning these notions of implicit representation would not require that we abandon the idea of implicit something that we care about keeping, like, say, implicit knowledge. This approach to making sense of an area of implicit cognition – distinguishing implicit cognition from explicit cognition by emphasizing the absence of certain elements – reflects a common way in which we often think about implicit cognition. For example, blindsight is sometimes described as a type of “implicit processing of visual qualities” (Stoerig and Cowey 2007: 824) in the absence of phenomenal sensory experience. In any event, trimming down the ways we think about implicit representation will in no way diminish the explanatory import of the notions of implicit representation that should be retained. Quite the contrary, the conceptions of implicit representation that are valuable – unconscious and dormant mental representations – will actually become more prominent with the reduction in conceptual excessiveness and greater emphasis upon their role in implicit cognition.7

Related Topics Chapters 3, 8, 14, 16, 18, 20, 24, 25, 27, 29

Notes 1 Some people distinguish between implicit and tacit representation, although there is no uniform way of doing this, and most writers treat these as the same. For our purposes here, ‘implicit’ and ‘tacit’ will be treated as synonymous. 2 By non-contentious, I  don’t mean that there is no contention as to whether or not such a thing exists. Instead, I mean were it to exist, we would all pretty much agree on characterizing it as explicit representation.

41

William Ramsey 3 Of course, the same is true for various forms of nonverbal (and often nonvoluntary) behavior, which are often used by psychologists as indicators of both explicit and implicit mental representations in very young children, children with autism and other conditions, and even ordinary adults. 4 As a reviewer has pointed out, Leibniz also deserves credit for rejecting Descartes’ view that thought must be conscious (see Leibniz 2008). 5 It should be noted that there are still plenty of other ways of thinking about the implicit and explicit representation distinction. For example, it can be done by focusing on the distinction between conceptual and nonconceptual formats of knowledge, the difference between declarative and procedural knowledge, or the difference between knowing something and knowing that one knows something (see, for example, Dienes and Perner 1999). Arguably, some of these further distinctions can often be at least partially reduced to the divisions listed here. For instance, the distinction between conceptual and nonconceptual representation is at least sometimes made in terms of the distinction between conscious and unconscious mental representations (Evans 1982); the distinction between declarative and procedural knowledge is at least sometimes cashed out as the distinction between possessing a distinct representational vehicle as opposed to a capacity embedded in the system’s functional architecture. 6 Here I am thinking of less restrictive attributions of knowledge that don’t require criteria like justification or truthfulness. 7 A version of this chapter was presented to the Washington University Mind and Perception Group on May 5, 2020. Comments and suggestions from this group and an anonymous reviewer were extremely helpful.

References Allen, R., and Reber, A. S. 1980. “Very long term memory for tacit knowledge”. Cognition, 8: 175–185. Armstrong, D. 1968. A materialist theory of mind. London: Routledge. Audi, R. 1994. “Dispositional beliefs and dispositions to believe”. Nous, 28: 419–434. Augusto, L. M. 2010. “Unconscious knowledge: a survey”. Advances in Cognitive Psychology, 6: 116–141. https://doi.org/10.2478/v10053-008-0081-5. Bargh, J. A. 1994. “The four horsemen of automaticity: awareness, intention, efficiency and control in social cognition”. In R. Wyer and T. Srull, eds., Handbook of social cognition. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.: 1–40. Brownstein, M. 2019. “Implicit bias”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Fall 2019 edn. https://plato.stanford.edu/archives/fall2019/entries/implicit-bias/ Carruthers, P. 2018. “Implicit versus explicit attitudes: differing manifestations of the same representational structures?”. Review of Philosophy and Psychology, 9: 51–72. Chomsky, N. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press. Clapin, H. 2002. The philosophy of mental representation. Oxford: Oxford University Press. Cowey, A. 2004. “Fact, artefact, and myth about blindsight”. Quarterly Journal of Experimental Psychology, 57A: 577–609. Davies, M. 2001. “Knowledge (explicit and implicit): philosophical aspects”. In N. Smelser and P. Baltes, eds., International encyclopedia of social & behavioral sciences. Amsterdam: Elsevier: 8126–8132. de Gelder, B., Vroomen, J., and Pourtois, G. 2002. “Covert affective cognition and affective blindsight”. In B. de Gelder, E. de Haan, and C. Heywood, eds., Out of mind: varieties of unconscious processes. Oxford: Oxford University Press: 205–221. Dennett, D. 1978. Brainstorms. Cambridge, MA: MIT Press. Dennett, D. 1987. The intentional stance. Cambridge, MA: MIT Press. Dienes, Z., and Perner, J. 1999. “A theory of implicit and explicit knowledge”. Behavioral and Brain Sciences, 22: 735–808. Dretske, F. 1987. Explaining behavior. Cambridge, MA: MIT/Bradford Press. Dummett, M. 1991. The logical basis of metaphysics. Cambridge, MA: Harvard University Press. Egan, A. 2008. “Seeing and believing: perception, belief formation and the divided mind”. Philosophical Studies, 140: 47–63. Evans, G. 1982. The varieties of reference. Oxford: Oxford University Press. Evans, J. S. B. T., and Frankish, K., eds. 2009. In two minds: dual processes and beyond. Oxford: Oxford University Press. Field, H. 1978. “Mental representations”. Erkenntnis, 13: 9–61.

42

Implicit Mental Representation Fodor, J. 1975. The language of thought. New York: Thomas Y. Crowell. Frankish, K. 2004. Mind and supermind. Cambridge: Cambridge University Press. Gendler, T. 2008. “Alief in action (and reaction)”. Mind and Language, 23: 552–585. Gigerenzer, G., and Gaissmaier, W. 2011. “Heuristic decision making”. Annual Review of Psychology, 62: 451–482. Greenwald, A., and Banaji, M. 1995. “Implicit social cognition: attitudes, self-esteem, and stereotypes”. Psychological Review, 102: 4–27. Kahneman, D. 2013. Thinking fast and slow. New York: Farrar, Straus and Giroux. Leibniz, G. 2008. New essays on human understanding. Trans. P. Remnant and J. Bennett. Cambridge: Cambridge University Press. Lettvin, J., Maturana, H., McCulloch, W., and Pitts, W. 1959. “What the frog’s eye tells the frog’s brain”. Proceedings of the Institute of Radio Engineers, 47: 1940–1951. Lycan, W. 1986. “Tacit belief ”. In R. J. Bogdan, ed., Belief. Oxford: Oxford University Press: 61–82. Machery, E. 2016. “De-Freuding implicit attitudes”. In M. Brownstein and J. Saul, eds., Implicit bias and philosophy. Vol. 1. Oxford: Oxford University Press: 104–129. McClelland, J. 1981. “Retrieving general and specific information from stored knowledge of specifics”. Proceedings of the Third Annual Meeting of the Cognitive Science Society: 170–172. Nanay, B. 2021. “Implicit bias as mental imagery”. Journal of the American Philosophical Association, 7: 329–347. Payne, B. K., and Cameron, C. D. 2013. “Implicit social cognition and mental representation”. In D. Carlston, ed., Oxford handbook of social cognition. Oxford: Oxford University Press: 220–238. Pylyshyn, Z. 1984. Computation and cognition: toward a foundation for cognitive science. Cambridge, MA: MIT Press. Ramsey, W. 2007. Representation reconsidered. Cambridge: Cambridge University Press. Ramsey, W. 2021. “Defending representational realism”. In J. Smortchkova, K. Dolega, and T. Schlicht, eds., What are mental representations? Oxford: Oxford University Press: 54–78. Reber, A. S. 1967. “Implicit learning of artificial grammars”. Journal of Verbal Learning and Verbal Behavior, 77: 317–327. Reber, A. S. 1993. Implicit learning and tacit knowledge: an essay on the cognitive unconscious. Oxford: Oxford University Press. Rosenthal, D. 1986. “Two concepts of consciousness”. Philosophical Studies, 49: 329–359. Rumelhart, D., and McClelland, J. 1986. Parallel distributed processing. Vol. 1. Cambridge, MA: MIT Press. Schacter, D. L. 1987. “Implicit memory: history and current status”. Journal of Experimental Psychology: Learning, Memory, and Cognition: 13: 501–518. Schacter, D. L. 1992. “Implicit knowledge: new perspectives on unconscious processes”. Proceedings of the National Academy of Sciences, 89: 11113–11117. Searle, J. 1992. The rediscovery of the mind. Cambridge, MA: MIT/Bradford Press. Sinnot-Armstrong, W., Young, L., and Cushman, F. 2010. “Moral intuitions”. In J. Doris, ed., The moral psychology handbook. Oxford: Oxford University Press: 246–272. Stoerig, P., and Cowey, A. 2007. “Blindsight”. Current Biology, 19: 822–824. Tulving, E., and Craik, F. I. M. 2000. The Oxford handbook of memory. Oxford: Oxford University Press.

43

2 MEASURING AND MODELING IMPLICIT COGNITION Samuel A.W. Klein and Jeffrey W. Sherman

The widespread use of indirect measures in the psychological sciences, particularly social psychology, was motivated by multiple concerns with direct, self-report measures. First, selfpresentational motives may lead respondents to misrepresent their true attitudes and beliefs, particularly when responding to questions about socially sensitive topics, such as race. Second, people sometimes have limited access to their own attitudes and beliefs (Greenwald and Banaji 1995). As such, people may be willing to respond accurately but are unable to do so. Lastly, even if people are able to accurately introspect about their attitudes and beliefs, they may not be sufficiently motivated to do so (Hahn and Gawronski 2019). Indirect measures were designed to circumvent these potential problems by inferring attitudes and beliefs from behavior, rather than by directly asking people to report them. Toward that end, in some cases, indirect measures conceal the purpose of the measure or possess properties that make them resistant to manipulation. The advent of indirect measures has led to an explosion of empirical research and theoretical advances in the study of people’s cognition and behavior. At the same time, there remain significant conceptual and methodological problems that impede the use and interpretation of indirect measures. In this chapter, we will discuss these problems and a solution via the use of formal models that disentangle cognitive processes underlying implicit cognition data.

Definitional Issues There are now a great variety of indirect measures of implicit cognition, including the Implicit Association Test (IAT; Greenwald et al. 1998), evaluative priming tasks (e.g., Fazio et al. 1995), Go/No Go Association Task (GNAT; Nosek and Banaji 2001), the Weapons Task (Payne 2001), the First Person Shooter Task (Correll et  al. 2002), the Affect Misattribution Procedure (Payne et al. 2005), and the Stereotype Misperception Task (Krieglmeyer and Sherman 2012). These, among other indirect measures, vary on a long list of features, including the time between presentations of various stimuli (stimulus onset asynchrony), the ambiguity of target information, when judgments can be made within each trial, and many more (see Gawronski and Brannon 2018 for a more comprehensive review of indirect measures and their varying features). Perhaps the most basic conceptual problem concerns the name we use to describe these measures. Most commonly, they are referred to as implicit measures. One problem with this label is that the term “implicit” has multiple meanings in experimental psychology. In some 44

DOI: 10.4324/9781003014584-4

Measuring and Modeling Implicit Cognition

traditions, the term refers to automatic processes that occur without awareness or intention, cannot be controlled, and are highly efficient. In other traditions, the term refers to processes that merely operate unconsciously. When people use the term “implicit measure,” it is not clear which of these features are being invoked (for a review, see Payne and Gawronski 2010). The term often suggests features of the measures that they may not possess. For example, although “implicit” often implies lack of awareness of the measure’s purpose, people are well aware of the intent of the IAT after a few trials. Moreover, among the plethora of measures available, there is great variation in the extent to which they possess these various processing attributes. One feature that is common to all measures labeled “implicit” is that they infer attitudes from task performance rather than by directly asking participants to report them. Therefore, we refer to any measure that indirectly assesses psychological attributes via task performance as an “indirect measure” (versus “direct measure”). The same definitional problem applies to the constructs presumed to be revealed by the measures. Most commonly, they are referred to as implicit attitudes or evaluations. However, the same ambiguities surrounding the meaning of “implicit” apply. It is not clear whether the implication of “implicit” is that people are unaware of the evaluations, that they are formed and used without intention, that they cannot be controlled, or that they operate efficiently. In fact, which of these features apply is dependent on both the means by which the evaluation was measured (i.e., which indirect measure is used) and the subject of the evaluation (e.g., race, age, fruit, dogs, etc.). For these reasons, ideally, implicit evaluations would, instead, be referred to as indirect evaluations. However, given the extent to which the term “implicit bias” has saturated academic and popular culture, we reluctantly retain the term “implicit evaluation.” It is simply too impractical to change. Still, note that our use of the term is intended to signify only that the evaluations are implied by performance on an indirect measure rather than explicitly provided on a direct measure. In that sense, the evaluations are, indeed, implicit in the given responses. We prefer the term “evaluation” over “attitude.” Whereas evaluations are assessment outcomes that may be based on a variety of sources, “attitudes” in this context implies the existence of distinct mental representations (implicit and explicit attitudes) that exist in our mind and are uniquely accessed by direct and indirect measures. One problem with viewing indirectly measured outcomes as Things is that it discourages a deeper understanding of the constructive nature of responses on indirect measures. Those evaluations are not mental apples waiting to be picked by an indirect measure; they are constructed from a variety of sources and processes in the act of responding to the task demands. For example, the stimulus must be attended to and interpreted, the correct or intended response must be determined, and the response must be made, which requires mental/physical coordination, self-regulation, motor action, and so on (e.g., Sherman et al. 2008). Thus, implicit evaluation is something we do, not something we have (De Houwer 2019). The view of indirect measure outcomes as Things also implies a level of situational and temporal stability that has not held up to scrutiny (Gawronski et al. 2017). Indeed, implicit evaluations show considerable contextual and temporal variability, and are quite malleable in response to interventions (e.g., Brannon and Gawronski 2017). In contrast, when viewed as measure-induced constructed evaluations, the expectation of stability is significantly diminished.1 A further implication of the Thing view is that the outcomes of different indirect measures should correlate strongly with another. After all, if they are all measuring the same Thing in our heads, then there should be high correspondence among them. However, correlations among indirect measures are modest, at best (Bar-Anan and Nosek 2014). These low correlations are indicative of the fact that the demands imposed by indirect measures and the processes recruited to meet those demands differ among measures in many ways. The outcomes of these measures 45

Samuel A.W. Klein and Jeffrey W. Sherman

reflect the impact of these differing demands and processes in completing the tasks. Thus, it is not a simple matter of measuring the Thing sitting in our heads: Implicit evaluations are constructed in real time.

Conceptual Challenges in Measuring and Interpreting Implicit Cognition In one way or another, the main conceptual issues with indirect measures all can be traced to the framing of indirect (versus direct) measures in terms of dual-process models of human cognition. Dual-process models propose that there are two distinct types of mental processes that characterize human cognition (Sherman et al. 2014). Whereas automatic processes occur without awareness or intention, cannot be controlled, and are highly efficient, controlled processes operate with awareness and intention, can be controlled, and require cognitive resources. Upon the introduction of indirect measures, the distinction between indirect and direct measures was mapped onto dual process models, with indirect measures tapping automatic processes and direct measures tapping controlled processes. With this mapping, indirect measures and their outcomes were endowed with the presumed features of automatic processes. First and foremost, this implied that the properties and outcomes of indirect measures are qualitatively distinct from those of direct measures. Further, it implied that indirect measures and their outcomes would be characterized by the operation of specific types of processes that operated under specific types of conditions and were based on specific types of mental representations. None of these implications were directly tested, and their assumption has greatly impacted how indirect measures are understood and how their outcomes are explained. We have already described how use of the term “implicit” to denote features of automaticity has clouded thinking about implicit measures and evaluations. Here, we will describe four ways in which features of indirect measures and implicit evaluations associated with automaticity have been conflated, leading to conceptual and explanatory uncertainty. First, the processes that underlie implicit cognition tend to be conflated with the conditions under which those processes operate. Second, indirect measures are often assumed to operate under a unique set of conditions. Third, measures (i.e., indirect vs. direct) are often conflated with the constructs they are designed to reveal. Finally, indirect measures have been conflated with the processes that drive their output (see also Sherman and Klein 2021).

Operating Principles and Operating Conditions Whereas operating principles refer to the qualitative nature of a process or representation, describing what it does (e.g., activation of associations; inhibition), operating conditions refer to the conditions under which those processes or representations operate (e.g., Does the process still occur when the person is mentally exhausted?; Sherman et  al. 2014). Operating principles and operating conditions are bidirectionally conflated. Sometimes, assumptions or knowledge about operating conditions influences assumptions about the operating principles. Other times, assumptions or knowledge about operating principles drives assumptions about the operating conditions. In neither case are the assumptions warranted. Knowledge about the conditions under which processes driving implicit cognition occur does not tell us anything about what those processes are (operating principles). For example, knowing that implicit evaluations do not change when participants are under cognitive load does not necessarily mean that the indirect measure is capturing the activation of associations in memory, a central assumption of many theories of implicit cognition (e.g., Gawronski and Bodenhausen 46

Measuring and Modeling Implicit Cognition

2006; Strack and Deutsch 2004). That is, although the process seems “automatic,” the extent of its automaticity does not define what the process does, merely when it may occur. Likewise, knowledge about what the processes are does not tell us anything about when the processes may or may not operate. For example, knowing that a process is inhibitory in nature does not necessarily mean that the process may only operate when people have full processing capacity. Indeed, there is now substantial evidence that a number of processes considered to be controlled in nature nevertheless operate in seemingly highly efficient ways (Calanchini and Sherman 2013). The key point is that conclusions about the processes contributing to implicit evaluations require independent tests of the nature of those processes. Assessing the role of inhibition in implicit evaluations requires research that specifically examines the role of inhibition. It cannot be inferred from knowing the operating conditions. Likewise, conclusions about the conditions under which a process may influence implicit evaluations require research that specifically examines those conditions. Conclusions about the intentionality, awareness, controllability, and efficiency of a process require direct tests of those features. They cannot be inferred from knowledge about the nature of the process in question (e.g., it is an associative or inhibitory process; Sherman et al. 2008).

Measures and Operating Conditions The commonly assumed relationship between implicit evaluations and automatic processes in implicit cognition research introduces yet another conflation. Typically, indirect measures are presumed to reflect processes that are automatic in nature, whereas direct measures are presumed to reflect processes that are considered controlled. At this point, such assumptions are no longer tenable. For example, there is growing evidence that people are aware of their implicit evaluations and how they will influence responses on indirect measures (e.g., Hahn et al. 2014). There also is evidence that people can inhibit implicit bias while completing the measures (e.g., Glaser and Knowles 2008; Krieglmeyer and Sherman 2012; Moskowitz and Li 2011; Sherman et al. 2008) and that responses on the measures are influenced by the availability of processing resources, indicating that performance is not entirely efficient (e.g., Conrey et al. 2005; Correll et al. 2002; Krieglmeyer and Sherman 2012). There may be significant costs in assuming that indirect/direct measures reflect automatic/ controlled processes, particularly given that there are many differences between direct and indirect measures that may not be related to the automatic/controlled distinction. An instructional example can be found in the implicit memory literature. For many years, indirect measures of memory were assumed to reflect the automatic influence of memories whereas direct measures of memory were assumed to reflect the intentional use of memory. After years of research built on this assumption, Roediger and his colleagues (e.g., Roediger 1990) observed that performance on indirect measures of memory depended largely on the encoding and retrieval of perceptual features of stimuli, whereas performance on direct measures of memory largely depended on the encoding and retrieval of conceptual (meaning) features of stimuli. When the type of processing was equated between direct and indirect measures of memory, the dissociations disappeared. Thus, the dissociations were between perceptual and conceptual memory rather than between implicit and explicit memory. In the same way, indirect and direct measures of evaluation differ along many dimensions, such as the use of visual stimuli in indirect but not direct measures. When such differences in the structural properties of the tasks are reduced, the correspondence between implicit and explicit responses rises (Payne et al. 2008). Again, claims about the operating conditions of a measure must be independently established with careful empirical work. They cannot be assumed based on the type of measure. 47

Samuel A.W. Klein and Jeffrey W. Sherman

Measures and Constructs The confounding of measures and operating conditions directly implicates a corresponding confound between measures and constructs. The assumption that indirect/direct measures reflect automatic/controlled processes forms the basis for the claim that implicit and explicit evaluations are, in fact, qualitatively distinct constructs. Again, the central problem is that the measures (and their associated constructs) differ in many ways beyond the extent to which responses are relatively implicit or explicit. Thus, differences between implicit and explicit evaluations may reflect differences in the extent to which those evaluations are implicit/explicit or they may reflect other differences in the procedural demands of indirect and direct measures.

Measures and Operating Principles Measures also are often conflated with operating principles. Whereas indirect measures are assumed to capture mostly associative processing, direct measures are assumed to capture mostly inhibitory or propositional processing (Fazio 1995; Gawronski and Bodenhausen 2006; Strack and Deutsch 2004). However, just as is the case with conflating measures with operating conditions or constructs, the problem is that indirect and direct measures differ in many ways. To the extent that direct and indirect measures differ in their structural features (e.g., the use of visual images) the processes invoked in responding to those measures also will differ. As such, the responses may reflect structural features of the measures that have nothing to do with the presumed operating principles. Another problem with conflating measures with particular operating principles is that no measure is process-pure. That is, no measure, indirect or direct, reflects the operation of a single type of process. Though assumed to primarily reflect associative processes, a plethora of other types of processes have been shown to influence responses on indirect measures, including the inhibition of associations, the accurate identification of stimuli, intentional coding strategies, attributional processes, and response biases (Sherman and Klein 2021). Ultimately, the interaction of many processes determines responses, and measure outcomes cannot reveal, on their own, the nature of the underlying processes that produced the outcomes. Conclusions about operating principles must be established directly through empirical work and cannot be inferred from operating conditions or measures. As a concrete example, consider the finding that aging is associated with increased implicit racial bias. Typically, this effect would be attributed to the fact that older people have more biased associations. However, such differences may also be related to changes in executive function associated with aging. Indeed, the second author’s own research showed that bias associated with aging was related to failures of self-regulation rather than to differences in underlying associations (see Applying the Quad Model section for elaboration; Gonsalkorale, Sherman et al. 2009). Or consider the fact that younger and older people have been observed to demonstrate similar degrees of implicit anti-aging bias. One might conclude that negative attitudes about aging are so pervasive that even older people possess them. However, aging was, in fact, associated with less negative associations with older people. At the same time, aging also was associated with a weaker ability to regulate the expression of said negative associations. In effect, these two processes cancelled each other out. Older and younger people responded similarly because even though older people had less negative associations, they were less able to control them (Gonsalkorale et al. 2014). In the next section, we will describe how one can simultaneously measure associations and the ability to overcome them.

48

Measuring and Modeling Implicit Cognition

Process Modeling of Indirect Measures Following the theoretical and methodological concerns outlined in the first part of this chapter, we turn to formal mathematical modeling as a solution that has seen a steep rise in popularity over the last two decades (Hütter and Klauer 2016; Ratcliff et  al. 2016; Sherman et  al. 2010). The purpose of formal modeling is to identify the processes underlying indirect task performance, measure those processes, and describe the ways in which they interact to produce responses. To do so, formal models propose a set of parameters representing the hypothesized processes (e.g., activation of association; inhibition of associations) and a set of equations that describe the ways in which the processes interact and constrain one another. Solving for the parameters yields estimates of the extent to which they contribute to responses. This technique offers a number of important advantages. First, because formal models input data from a single task, differences in the natures or extents of the processes cannot be due to differences in the features of measures. As described earlier, when different measures are used to assess different operating conditions, constructs, or operating principles, it is always possible that the observed differences are related to differences in the features of the measures that have nothing to do with the proposed constructs, operating conditions, or operating principles. Second, inherent in the use of formal models is the assumption that multiple processes interact to drive outcomes. Thus, the measures are not assumed to reflect only one process. Third, constructing a model requires the use of an explicit theory about which processes contribute to performance and the manner in which those processes interact with one another. Therefore, many of the key assumptions underlying conceptual process models of implicit cognition can be directly tested via formal modeling techniques. Finally, competing models that identify different processes or different relationships among the processes can be compared in terms of their ability to fit the data. In essence, this is a means of comparing the validity of different theories of implicit cognition. Though formal modeling provides a means for proposing and testing the operating principles that determine implicit evaluations, it is important to note that the psychological meanings of the parameters must be independently established with empirical research. If we want to claim that a model parameter represents the inhibition of associations, we need to empirically demonstrate that the parameter responds the way inhibition should. As stated earlier, operating principles cannot be assumed; they must be tested. Likewise, any claims about the conditions under which the parameters operate must be established independently. If we want to claim that a parameter is dependent on the availability of cognitive resources, we need to show that empirically. A wide variety of formal modeling techniques have been used toward these ends, including signal detection, process dissociation, diffusion models, and multinomial tree models. Though a full discussion of these different types of models is beyond the scope of this chapter (for comprehensive reviews, see Hütter and Klauer 2016; Ratcliff et al. 2016; Sherman et al. 2010; Wixted 2020), we will present one example in some detail. Specifically, we present an overview of the development and use of the Quadruple Process model (Quad model; Conrey et al. 2005: Sherman et al. 2008), which was initially advanced to account for performance on the IAT. We show how the Quad model enhances our understanding of the processes that drive performance on indirect measures, the manner in which those processes contribute to individual differences in implicit evaluations, and the meaning of contextual variations and malleability in indirect task performance. We also demonstrate how the model helps to explain relationships between implicit evaluations and behavior.

49

Samuel A.W. Klein and Jeffrey W. Sherman

The Quadruple Process Model The Quad model proposes four distinct processes that interact to produce responses on indirect measures of evaluation. The model proposes parameters for the activation of associations (AC), the detection of correct responses to target stimuli (D), the overcoming of biased associations in favor of correct responses (OB), and a general response bias (G). The structure of the Quad model as applied to the IAT is depicted as a processing tree in Figure 2.1. In the tree, each path represents a likelihood. Processing parameters with lines leading to them are conditional on all preceding parameters. For instance, OB is conditional on both AC and D. The conditional relationships described by the model form a system of equations that predicts the numbers of correct and incorrect responses on different trial types (e.g., compatible and incompatible trials). For example, there are three ways in which an incorrect response can be returned on a trial in which Black and ‘pleasant’ share a response key for a person with pro-White bias. The first is the likelihood that biased associations between ‘Black’ and ‘unpleasant’ are activated (AC), detection succeeds (D), and OB fails (1 – OB), which can be represented by the equation AC × D × (1 – OB). The second is the likelihood that the biased associations are activated (AC) and detection fails (1 – D), which can be represented by the equation AC × (1 – D). The third is the likelihood that biased associations are not activated (1 – AC), detection fails (1 – D), and a bias toward guessing ‘unpleasant’ (1-G) produces an incorrect response, which can be represented by the equation (1 – AC) × (1 – D) × (1 – G). As such, the overall likelihood of producing an incorrect response on such a trial is the sum of these three conditional probabilities: [AC × D × (1 – OB)] + [AC × (1 – D)] + [(1 – AC) × (1 – D) × (1 – G)]. The respective equations for each item category (i.e., White faces, Black faces, pleasant words, and unpleasant words in both trial types) are then used to predict the observed proportions of errors in a given data set. The model’s predictions are compared to the actual data to determine the model’s ability to account for the data. An estimate of statistical fit is computed for the difference between the predicted and observed responses. To best approximate the model to the data, the parameter values are changed through estimation methods (e.g., maximum likelihood) until they produce

Compatible Incompatible OB D

Detection Achieved

Association Activated

Black stimulus

1-OB 1-D

D

Bias Not Overcome

Detection Not Achieved

Detection Achieved

Association Not Activated

G 1-D

Bias Overcome

Detection Not Achieved 1-G

Figure 2.1  The quadruple process model

50

Positive Response Bias Negative Response Bias

Measuring and Modeling Implicit Cognition

a minimum possible value of misfit between the observed and predicted responses. The final parameter values that result from this process are interpreted as relative levels of the processes.

Applying the Quad Model The Quad model has provided consistently good fit across data sets (Conrey et al. 2005; Sherman et al. 2008) and has been applied toward understanding a number of fundamental questions about implicit evaluation (Sherman et al. 2008).

Individual Differences One central question concerns the interpretation of differences in implicit evaluations among respondents. What does it mean when two individuals or two groups of people differ in implicit evaluations? Traditional approaches to understanding implicit evaluation would suggest that such differences must be due to differences in the underlying evaluative associations among individuals. Application of the Quad model has shown that, indeed, sometimes that is the case, as in differences in pro-White bias between White and Black respondents (Gonsalkorale et al. 2010). However, in other cases, modeling has shown group differences to depend not on the strength of an underlying bias, but on the likelihood of overcoming it. For example, the observation that older individuals display more negative implicit evaluations of some lower-status groups has commonly been understood to represent the presence of more negative associations due to prejudicial social norms during their youth. However, application of the Quad model has shown that greater bias among the elderly is based not on the involvement of more biased associations, but on the likelihood that bias is overcome (Gonsalkorale, Sherman et al., 2009, 2014).

Contextual Variation and Malleability of Bias Another central question is how to explain changes in implicit evaluation across contexts or in response to interventions. Traditional approaches would suggest that such changes must be due to differences in the underlying evaluative associations across contexts or interventions. Application of the Quad model has shown that, indeed, sometimes that is the case, as in reductions in implicit evaluative race bias when an IAT includes pictures of positive Black and negative White persons (e.g., Gonsalkorale et  al. 2010), among respondents focused on a common ingroup identity (Scroggins et al. 2016), among respondents with greater intergroup contact (Rae et al. 2020), or among respondents who have suffered a temporary blow to self-esteem (Allen and Sherman 2011). However, in other cases, modeling has shown such variation to depend not on the strength of an underlying bias, but on the likelihood of overcoming it. For example, the observation of greater evaluative race bias among intoxicated respondents corresponds not with alterations in activated associations but with the likelihood of effectively regulating the influence of those associations (Sherman et al. 2008). The likelihood of overcoming bias similarly accounts for reduced implicit race evaluation when an IAT presents Black and White persons in positive and negative contexts, respectively (for a review, see Calanchini et al. 2021).

Predicting Behavior A third fundamental question concerns the extent to which implicit evaluations predict behavior. Specifically, when implicit evaluations predict behavior, which component processes of the evaluation are responsible? Traditional approaches would suggest that variations in underlying 51

Samuel A.W. Klein and Jeffrey W. Sherman

associations direct behavior and are responsible for the evaluation-behavior link. However, it may be more complicated than that. In one study (Gonsalkorale, von Hippel, et al. 2009), after interacting with a Muslim research confederate (i.e., a research assistant acting the part), White participants completed an anti-Muslim GNAT. As well, the confederate rated how much he enjoyed his interaction with each participant. Results showed that the more negative the subjects’ implicit evaluations of Muslims on the GNAT, the less the confederate enjoyed interacting with them. Application of the Quad model to the GNAT data showed that the confederate’s liking of the subjects was not predicted solely by the extent of the subjects’ negative association with Muslims. Rather, the confederate’s liking of the subjects was predicted by an interaction between the subjects’ anti-Muslim associations and the likelihood that they overcame them in performing the GNAT. Specifically, when the subjects’ associations with Muslims were only moderately negative, the confederate’s liking of the subjects did not depend on the likelihood that they overcame their bias on the GNAT. In contrast, when the subjects had strongly negative associations with Muslims, the confederate liked them to the extent that they successfully overcame their bias when performing the GNAT. These findings indicate that the ability to regulate implicit evaluations may play an important role in a person’s direct behaviors with members of a group toward whom they are biased.

Summary In this section, we briefly described the application of one multinomial model, the Quad model, toward understanding key questions in the implicit evaluation literature. Beyond providing a more detailed understanding of the fundamental meaning and basis of implicit responses, modeling deepens understanding of individual differences in implicit evaluations, variability/ malleability of implicit evaluations, and evaluation-behavior links. The common understanding of implicit evaluation explains all of these effects by reference to the activation and application of biased associations stored in memory. Modeling shows that, in many cases, these effects are driven by a variety of processes, and sometimes do not involve associations at all. Other research has shown that some of these processes have nothing to do with a specific attitude object, per se, but, rather, represent domain-general cognitive skills. For example, the extent of Detection and Overcoming Bias in the Quad model in evaluations of one domain correlates robustly with the extent of those processes in other domains (Calanchini and Sherman 2013). This indicates that these processes are not tied to specific domains but rather assess general cognitive abilities that influence responses across domains. These observations made possible by modeling also have important implications for designing interventions to alter implicit evaluations. Traditionally, the assumption has been that such efforts must be targeted toward and are effective to the extent that they change the evaluative associations in people’s heads. However, modeling work shows that interventions that alter general cognitive abilities may also be effective in changing implicit evaluations. For instance, training people to more accurately identify stimuli or more effectively inhibit routinized responses may be effective means of implicit bias reduction. As an example, Calanchini et al. (2013) demonstrated that Detection in the Quad model is responsive to training: Participants who completed a counter-prejudicial training task subsequently had higher levels of Detection. Those participants also demonstrated less IAT bias than control participants, suggesting that increases in Detection may be tied to diminished bias. Given the domain-generality of these cognitive skills, developing these abilities may have relative broad payoffs across attitude domains (Calanchini et al. 2014).

52

Measuring and Modeling Implicit Cognition

Conclusion Many of the initial foundational assumptions of work on implicit cognition have proven to be problematic. Unfounded assertions regarding the nature of indirect measures, the constructs that they assess, the processes that generate responses on the measures, and the conditions under which those processes operate have significantly complicated efforts to measure, characterize, and understand implicit cognition. Those who use indirect measures for research and those who consume that research should be aware of these complications. For researchers, one potential remedy to these problems is the use of mathematical models combined with careful research to validate any claims about mechanisms and the conditions under which they operate. In addressing these issues, we hope that this chapter helps to provide a framework for the future of doing and understanding implicit cognition.

Related Topics Chapters 1, 3, 6, 7, 8, 25, 27, 28, 31

Notes 1 A popular critique of implicit cognition research is that its data often lacks stability. Removing this stability expectation by viewing the data as measure-induced constructed evaluations strongly devalues the weight of that criticism, as it is founded on a view that separate implicit and explicit Things are sitting in the mind. Although outside the scope of this chapter, see Brownstein et al. (2020) for a more in-depth discussion.

References Allen, T. J., and Sherman, J. W. 2011. “Ego threat and outgroup derogation: a test of motivated activation versus self-regulatory accounts”. Psychological Science, 22: 331–333. Bar-Anan, Y., and Nosek, B. A. 2014. “A comparative investigation of seven indirect attitude Measures”. Behavior Research Methods, 46: 668–688. https://doi.org/10.3758/s13428-013-0410-6 Brannon, S. M., and Gawronski, B. 2017. “A second chance for first impressions? Exploring the context-(in) dependent updating of implicit evaluations”. Social Psychological and Personality Science, 8(3): 275–283. Brownstein, M., Madva, A., and Gawronski, B. 2020. “Understanding implicit bias: putting the criticism into perspective”. Pacific Philosophical Quarterly, 101: 276–307. Calanchini, J., Gonsalkorale, K., Sherman, J. W., and Klauer, K. C. 2013. “Counter-prejudicial training reduces activation of biased associations and enhances response monitoring”. European Journal of Social Psychology, 43: 321–325. Calanchini, J., Lai, C. K., and Klauer, K. C. 2021. “Reducing implicit racial preferences: III. A process-level examination of changes in implicit preferences”. Journal of Personality and Social Psychology, 121: 796. Calanchini, J., and Sherman, J. W. 2013. “Implicit attitudes reflect associative, non-associative, and nonattitudinal processes”. Social and Personality Psychology Compass, 7: 654–667. https://doi.org/10.1111/ spc3.12053 Calanchini, J., Sherman, J. W., Klauer, K. C., and Lai, C. K. 2014. “Attitudinal and non-attitudinal components of IAT performance”. Personality and Social Psychology Bulletin, 40: 1285–1296. Conrey, F. R., Sherman, J. W., Gawronski, B., Hugenberg, K., and Groom, C. J. 2005. “Separating multiple processes in implicit social cognition: the Quad model of implicit task performance”. Journal of Personality and Social Psychology, 89: 469–487. Correll, J., Park, B., Judd, C. M., and Wittenbrink, B. 2002. “The police officer’s dilemma: using ethnicity to disambiguate potentially threatening individuals”. Journal of Personality and Social Psychology, 83: 1314–1329. https://doi.org/10.1037/0022-3514.83.6.1314

53

Samuel A.W. Klein and Jeffrey W. Sherman De Houwer, J. 2019. “Implicit bias is behavior: a functional-cognitive perspective on implicit bias”. Perspectives on Psychological Science, 14: 835–840. Fazio, R. H. 1995. “Attitudes as object-evaluation associations: determinants, consequences, and correlates of attitude accessibility”. In R. E. Petty and J. A. Krosnick, eds., Ohio State University series on attitudes and persuasion, vol. 4: attitude strength: antecedents and consequences. Hillsdale: Lawrence Erlbaum Associates: 247–282. Fazio, R. H., Jackson, J. R., Dunton, B. C., and Williams, C. J. 1995. “Variability in automatic activation as an unobtrusive measure of racial attitudes: a bona fide pipeline?”. Journal of Personality and Social Psychology, 69: 1013–1027. Gawronski, B., and Bodenhausen, G. V. 2006. “Associative and propositional processes in evaluation: an integrative review of implicit and explicit attitude change”. Psychological Bulletin, 132: 692–731. https://doi.org/10.1037/0033-2909.132.5.692 Gawronski, B., and Brannon, S. M. 2018. “Attitudes and the implicit-explicit dualism”. In D. Albarracin and B. T. Johnson, eds., The handbook of attitudes. London: Routledge: 158–196. Gawronski, B., and Hahn, A. 2019. “Implicit measures: procedures, use, and interpretation”. In H. Blanton, J. M. LaCroix, and G. D. Webster, eds., Measurement in social psychology. New York, NY: Taylor and Francis: 29–55. Gawronski, B., Morrison, M., Phills, C. E., and Galdi, S. 2017. “Temporal stability of implicit and explicit measures: a longitudinal analysis”. Personality and Social Psychology Bulletin, 43: 300–312. Glaser, J., and Knowles, E. D. 2008. “Implicit motivation to control prejudice”. Journal of Experimental Social Psychology, 44: 164–172. https://doi.org/10.1016/j.jesp.2007.01.002 Gonsalkorale, K., Allen, T. J., Sherman, J. W., and Klauer, K. C. 2010. “Mechanisms of exemplar exposure effects on implicit attitudes”. Social Psychology, 41: 158–168. Gonsalkorale, K., Sherman, J. W., and Klauer, K. C. 2009. “Aging and prejudice: diminished regulation of automatic race bias among older adults”. Journal of Experimental Social Psychology, 45: 410–414. https:// doi.org/10.1016/j.jesp.2008.11.004 Gonsalkorale, K., Sherman, J. W., and Klauer, K. C. 2014. “Measures of implicit attitudes may conceal differences in implicit associations: the case of antiaging bias”. Social Psychological and Personality Science, 5: 271–278. https://doi.org/10.1177/1948550613499239 Gonsalkorale, K., von Hippel, W., Sherman, J. W., and Klauer, K. C. 2009. “Bias and regulation of bias in intergroup interactions: implicit attitudes toward Muslims and interaction quality”. Journal of Experimental Social Psychology, 45: 161–166. https://doi.org/10.1016/j.jesp.2008.07.022 Greenwald, A. G., and Banaji, M. R. 1995. “Implicit social cognition: attitudes, self-esteem, and stereotypes”. Psychological Review, 102: 4–27. https://doi.org/10.1037/0033-295X.102.1.4 Greenwald, A. G., McGhee, D. E., and Schwartz, J. L. K. 1998. “Measuring individual differences in implicit cognition: the implicit association test”. Journal of Personality and Social Psychology, 74: 1464–1480. https://doi.org/10.1037/0022-3514.74.6.1464 Hahn, A., and Gawronski, B. 2019. “Facing one’s implicit biases: from awareness to acknowledgement”. Journal of Personality and Social Psychology, 116: 769–794. https://doi.org/10.1037/pspi0000155 Hahn, A., Judd, C. M., Hirsch, H. K., and Blair, I. V. 2014. “Awareness of implicit attitudes”. Journal of experimental psychology. General, 143: 1369–1392. https://doi.org/10.1037/a0035028 Hütter, M., and Klauer, K. C. 2016. “Applying processing trees in social psychology”. European Review of Social Psychology, 27: 116–159. https://doi.org/10.1080/10463283.2016.1212966 Krieglmeyer, R., and Sherman, J. W. 2012. “Disentangling stereotype activation and stereotype application in the stereotype misperception task”. Journal of Personality and Social Psychology, 103: 205–224. https:// doi.org/10.1037/a0028764 Moskowitz, G. B., and Li, P. 2011. “Egalitarian goals trigger stereotype inhibition: a proactive form of stereotype control”. Journal of Experimental Social Psychology, 47: 103–116. https://doi.org/10.1016/j. jesp.2010.08.014 Nosek, B. A., and Banaji, M. R. 2001. “The Go/No-Go association task”. Social Cognition, 19: 625–666. Payne, B. K. 2001. “Prejudice and perception: the role of automatic and controlled processes in misperceiving a weapon”. Journal of Personality and Social Psychology, 81: 181–192. Payne, B. K., Burkley, M. A., and Stokes, M. B. 2008. “Why do implicit and explicit attitude tests diverge? The role of structural fit”. Journal of Personality and Social Psychology, 94: 16–31. https://doi. org/10.1037/0022-3514.94.1.16 Payne, B. K., Cheng, C. M., Govorun, O., and Stewart, B. D. 2005. “An inkblot for attitudes: affect misattribution as implicit measurement”. Journal of Personality and Social Psychology, 89: 277–293.

54

Measuring and Modeling Implicit Cognition Payne, B. K., and Gawronski, B. 2010. “A history of implicit social cognition: where is it coming from? Where is it now? Where is it going?”. In B. Gawronski and B. K. Payne, eds., Handbook of implicit social cognition: measurement, theory, and applications. New York: The Guilford Press: 1–15. Rae, J. R., Reimer, N. K., Calanchini, J., Lai, C. K., Rivers, A. M., Dasgupta, N., . . . Schmid, K. 2020. “Intergroup contact and implicit racial attitudes: contact is related to less activation of biased evaluations but is unrelated to bias inhibition”. PsyArXiv. https://doi.org/10.31234/osf.io/h4nxd Ratcliff, R., Smith, P. L., Brown, S. D., and McKoon, G. 2016. “Diffusion decision model: current issues and history”. Trends in Cognitive Sciences, 20: 260–281. https://doi.org/10.1016/j.tics.2016.01.007 Roediger, H. L. 1990. “Implicit memory: retention without remembering”. American Psychologist, 45: 1043–1056. https://doi.org/10.1037/0003-066X.45.9.1043 Scroggins, W. A., Mackie, D. M., Allen, T. J., and Sherman, J. W. 2016. “Reducing prejudice with labels: shared group memberships attenuate implicit bias and expand implicit group boundaries”. Personality and Social Psychology Bulletin, 42: 219–229. https://doi.org/10.1177/0146167215621048 Sherman, J. W., Gawronski, B., Gonsalkorale, K., Hugenberg, K., Allen, T. J., and Groom, C. J. 2008. “The self-regulation of automatic associations and behavioral impulses”. Psychological Review, 115: 314–335. https://doi.org/10.1037/0033-295X.115.2.314 Sherman, J. W., Gawronski, B., and Trope, Y., eds. 2014. Dual-process theories of the social mind. New York: The Guilford Press. Sherman, J. W., Klauer, K. C., and Allen, T. 2010. “Mathematical modeling of implicit social cognition: the machine in the ghost”. In B. Gawronski and B. K. Payne, eds., Handbook of implicit social cognition: measurement, theory, and applications. New York: Guilford Press: 156–175. Sherman, J. W., and Klein, S. A. W. (2021). “The four deadly sins of implicit attitude research”. Frontiers in Personality and Social Psychology, 11: 60430. https://doi.org/10.3389/fpsyg.2020.604340 Strack, F., and Deutsch, R. 2004. “Reflective and impulsive determinants of social behavior”. Personality and Social Psychology Review, 8: 220–247. https://doi.org/10.1207/s15327957pspr0803_1 Wixted, J. T. 2020. “The forgotten history of signal detection theory”. Journal of Experimental Psychology: Learning, Memory, and Cognition, 46: 201–233. https://doi.org/10.1037/xlm0000732

55

3 IMPLICIT COGNITION AND UNCONSCIOUS MENTALITY Tim Crane and J. Robert Thompson

1.  The Unconscious It is now a commonplace amongst those who study the mind that it is largely unconscious, with only a small part of it manifesting itself in our conscious lives. Cognitive scientists routinely postulate unconscious states and processes among the central psychological machinery. The idea that the perceptual system, for example, makes ‘unconscious inferences’ has been around since Helmholtz (1867), and is part of the orthodoxy in computational theories of vision (Marr 1982). Cognitive psychologists standardly appeal to unconscious priming effects and subliminal perceptions (Kihlstrom 1987). The puzzling phenomenon of blindsight seems an example of perception that is unconscious in some sense (Weiskrantz 1986), and some theories of perception treat the faculty itself as essentially unconscious (Burge 2010). Unconscious implicit biases and heuristics that affect our behavior and judgment are among the best-known hypotheses of social psychology (Wilson 2002) and behavioral economics (Tversky and Kahneman 1974). In theoretical linguistics, the program of generative grammar explains human linguistic behavior in terms of stored, unconscious knowledge of grammatical rules (Chomsky 1980). In the philosophy of mind, the central mental states such as belief or desire (what Bertrand Russell (1921) called the ‘propositional attitudes’) are generally treated as essentially unconscious states, characterized by their causal or functional role. Perhaps more familiar in popular circles is the unconscious as conceived by psychoanalysis, essentially involving the repression of desires and memories that affect our behavior but are difficult to bring to consciousness (Freud 1915). Many of these psychological phenomena contribute to the popular image of the mind as an iceberg, with only its tip visible in consciousness and all the real action going on underneath. There appears to be a bewildering variety of phenomena that the study of the mind classifies as unconscious, but does anything unite all these phenomena? Does the unconscious have an essence? Can there be a general theoretical account of unconscious mentality? We proceed in this chapter with three aims. The first is to dispute the standard view of the relationship between conscious and unconscious mentality, and with it, the standard view of the relationship between consciousness and intentionality. The second is to lay out several options

56

DOI: 10.4324/9781003014584-5

Implicit Cognition and Unconscious Mentality

for replacing the standard view, ones that allow for substantive differences between conscious and unconscious mentality. The third is to sketch the foundations of a unifying conception of the unconscious across the various disciplines that study the mind. Along the way, we apply these conjectures to examples of implicit cognition.

2.  The Very Idea of Unconscious Mentality To get an adequate theoretical overview of what the unconscious means in these contexts, it helps to have a brief account of how the various conscious/unconscious distinctions in philosophy and psychology arose and developed from the late 19th century to the present day (for a longer account, see Crane (2020)). Without this context, it is difficult to locate where accounts of the unconscious go astray.

2.1  A Short History Psychology as a science began in the late 19th century chiefly as the study of conscious mental phenomena, but the emphasis was shifted to the study of behavior alone during the behaviorist period (roughly from the 1920s until the late 1950s). However, the return of mentalism or cognitivism in psychology after the Second World War did not reinstate the study of conscious phenomena as its chief concern. Instead, the psychology that emerged largely involved the attribution of systems of mental representation to subjects or to their brains to explain subjects’ behavior – systems that are mostly unconscious. This focus on representation appeared in philosophy as well, treating intentionality – what Franz Brentano called ‘the mind’s direction upon its objects’ – rather than consciousness as its central focus. The object of a mental state, or its ‘intentional object’, is what it concerns, or is about, or is otherwise directed on. The intentional content, on this view, is the way the object is represented in the intentional state – the same object can be represented in different ways, and different objects can be represented in the same way. Many philosophers defended the idea that intentionality is what is distinctive of all the things we classify as mental (see Chalmers 2004; Crane 2003, 2009). Though the notion of unconscious mentality had been around in philosophy in some form or another at least since Leibniz (1704), drawing a sharp distinction between intentionality and consciousness allowed 20th-century philosophers a straightforward way of accounting for unconscious mentality – consciousness was not essentially intentional, and intentionality was not essentially conscious. One result of drawing this distinction between consciousness and intentionality is that consciousness came to be conceptualized in predominantly sensory terms. Paradigmatic conscious states were bodily sensations like pains, and visual and other sensory experiences, selected for this role because they appeared to be characterizable in terms unrelated to any intentional or representational content they might possess. So such ‘qualitative’ or ‘phenomenal’ states are thought by many to be, for all intents and purposes, conscious states. A corollary of drawing this distinction so sharply is that because conscious intentional states were hard to incorporate within this framework, intentional states were treated as, for all intents and purposes, unconscious. Drawing this sharp distinction enabled theorists to distinguish between conscious and unconscious mentality. But, such accounts resulted in a notion of conscious mentality that was ready-made for sensory states while remaining largely silent both about forms of nonsensory consciousness as well as what unconscious mentality was like. Even as unconscious mentality was invoked in an expanding number of contexts, it was modeled upon conscious mentality

57

Tim Crane and J. Robert Thompson

and treated as though it was not truly distinctive. As one pioneer in exploring unconscious mentality puts it: This ‘nothing special’ line of argument is a direct entailment of taking the stance that consciousness is primary and that the default position should be that conscious processes lie at the heart of human cognition. From this perspective, unconscious, implicit functions are dealt with derivatively and virtually all interesting cognitive functions are to be seen as dependent on conscious processes. (Reber 1993: 25) This primacy of the conscious did not deny the existence of the unconscious, per se, but it quietly shaped conceptions of the unconscious, treating conscious cognition as the default framework through which all mentality was to be understood (even the unconscious).

2.2 Ramifications Although Reber argues that the unconscious/implicit holds a more legitimate claim to primacy than the conscious/explicit, one need not take a stand on primacy to note that conscious thought appears to be a bit of a latecomer to the cognitive scene. Hence, it is at least worth considering whether unconscious thought is able to achieve what it can because it operates in a manner lacking the particularities imposed by conscious thought. As an evolutionarily older mode of thinking with a wider scope than what thought happens to appear in consciousness, it may share few operational features with the conscious realm. It is important, then, to make explicit the often-unarticulated assumptions involved in the primacy of the conscious. Even if humans are, for obvious reasons, more familiar with the conscious realm, the primacy of the conscious needs to be justified. It is not inevitable to begin with conscious mentality and treat the unconscious states and processes as though they were more of the same, but just lacking some qualitative ‘glow’ or ‘buzz’. In fact, conscious mentality serves as a poor guide to the nature of the unconscious, given the range of paradigm unconscious states and processes. Models of consciousness were developed in a way to account for sensations. So, what kind of guide can they be for characterizing aspects of our mentality that aren’t sensations? If we are right that theorists who believe in the primacy of the conscious tend to draw a sharp distinction between consciousness and intentionality, then they also tend to hold that intentional states are deemed to be primarily unconscious, perhaps essentially unconscious. But, if there is no obvious model of conscious intentional states to shape their characterization of their unconscious nature, how are we supposed to characterize unconscious mentality for this range of nonsensory states? This puzzle about conscious and unconscious mentality stems from attempts to shoehorn intentional states into a perspective that sees the conscious as primary but utilizes a conception of consciousness that is ill-suited for its explanatory aims. Most current attempts to explain the unconscious are a result of a specific and implausible picture of consciousness that arose out of the behaviorist movement in philosophy and psychology (Crane 2020). It is critical to note that this picture is not mandatory. For one thing, these attempts fail to explain conscious intentionality. And without this conscious framework to adjust in order to explain the nature of the intentional, it’s not clear how they can explain unconscious intentionality either. So, we need to rethink conscious intentionality, or we need to rethink unconscious intentionality – or both. Philosophy of mind in recent years has begun

58

Implicit Cognition and Unconscious Mentality

to concentrate on the former task (e.g., Bayne and Montague (2011), Farkas (2008), Kriegel (2013)), but in in this chapter, we concentrate on the latter.

2.3  Basic Elements With these origins in mind, we can place the account of unconscious mentality on a stronger footing. Let’s begin by considering an incorrect, but instructive, view of the relationship between mentality and consciousness. John Searle (1992) has famously argued that consciousness is the only true mark of the mental; there are no unconscious mental states, there are only states of the brain that have the disposition to produce conscious states. Defenders of this position would need to demonstrate that the range of phenomena included in this volume are either not unconscious or not mental, which would be a tall order. But, although Searle’s position is surely incorrect and is rejected by a majority of philosophers and psychologists, it presents a challenge to this majority to specify precisely what alternative they are defending. In particular, the idea of unconscious mentality requires clarification in two dimensions: what it means for an unconscious mental state or event or process to be mental and what it means for a mental state or event or process to be unconscious. ‘Intentionalism’, as we use the term here, is the view that all mental phenomena are intentional. Intentionalism gives a ready answer to the question of what makes unconscious states mental – they have intentional contents, or (in other words) they represent their objects in certain ways. Although we assume this intentionalist view in what follows, the conclusion of the chapter does not depend on it. Someone who rejects intentionalism could nonetheless accept the view of unconscious intentionality proposed here. They would have then to give a separate account of the mentality of the unconscious in non-intentional terms. Intentionalism holds that all mental states and processes have intentionality – both those that are conscious and those that are not. Hence, attempts to explain the conscious/unconscious distinction solely in terms of the absence or presence of intentionality will not be viable. In what follows, we stress that the critical distinction to elucidate is that between conscious and unconscious intentionality.

3.  Unconscious Intentionality Among those who have studied unconscious intentionality, two broad approaches have emerged (cf., Katsafanas 2016). What we shall call the Dominant Approach is uninformative and deflationary: unconscious mentality is just mentality that lacks consciousness. This is clearly in line with assumptions about the primacy of the conscious. The unconscious functions more or less like the conscious mind, simply without the presence of consciousness. The other approach denies this mentality-minus-consciousness claim. The unconscious functions according to its own special governing principles. In this section, we will give reasons in favor of the alternative approach that unseats the primacy of the conscious. According to the Dominant Approach, unconscious representational states share their essential representational nature with their conscious counterparts; they just lack whatever it is that makes conscious states conscious. For philosophers who think of consciousness in terms of ‘qualia’, this will mean that unconscious states have intentionality but lack qualia; for higherorder thought theories, unconscious states will be ones that are not the objects of higher-order thought (Carruthers 2003, 2011; Rosenthal 2005); for theorists who postulate an ambiguity in ‘consciousness’ (Block 1995) states may be ‘phenomenally’ conscious and not ‘access’ conscious, or access conscious and not phenomenally conscious.

59

Tim Crane and J. Robert Thompson

The notion of consciousness presumed here is (for the most part) modeled on sensations, so the Dominant Approach seems best poised to explain what an unconscious sensation would be – it is a sensory state that lacks its conscious-making feature. This is, perhaps, what occurs in cases of blindsight described in §1. The Dominant Approach does little to characterize the nature of these unconscious intentional states, except to say what they lack, but this is still somewhat helpful in dealing with occurrent mental episodes like sensations and perceptual experiences like blindsight. Nevertheless, an account of the unconscious also needs to deal with persisting or ‘standing’ states (like beliefs and intentions, for example). Such standing states cannot be an afterthought for a theory of unconscious intentionality. Yet if we model consciousness on sensory states, this looks inevitable. The Dominant Approach must claim that an unconscious standing state can be understood as a conscious standing state that lacks consciousness. But what is a conscious standing state, that is, in what sense can consciousness be present or absent for such states? Many, if not most, standing states are unconscious in their very nature (Crane 2013), so a proper account of unconscious intentionality needs to explain them. Consider the paradigmatic example of a standing mental state – belief. Although it is sometimes said that people have conscious beliefs, this is a problematic idea. When it is said that a subject has conscious beliefs, what is typically being identified is not a belief or belief state, but an episodic judgment or assertion by the subject. That episode is justified or grounded by the agent’s beliefs or is a report that came about because of her beliefs, but these are not conscious versions of beliefs, or beliefs with some conscious-making feature added. Unlike episodes of sensations, there are no episodes of believing, per se, conscious or not. This critical difference renders the Dominant Approach largely silent about the nature of unconscious beliefs. If there are not conscious beliefs, then one cannot characterize the unconscious versions of these by taking a conscious version and removing its conscious-making feature. It’s the features it possesses as a standing state that make it an unconscious belief, not the absence of some conscious-making features that a singular episodic occurrence of it might possess. To be sure, it is somewhat cumbersome to distinguish between the standing belief states and an episode in which some judgment related to them is delivered, but this critical difference is precisely what is being muddled in the Dominant Approach, and what prevents the Dominant Approach from accounting for unconscious standing states. This stems from a commitment to the primacy of the conscious, where a model of standing belief states is erroneously shaped by a model of episodic judgment. According to the mentality-minus-consciousness approach, unconscious beliefs are explained as something like an episode of conscious judgment that has had its conscious element removed. An episodic sensory model of consciousness is thereby shaping our picture of unconscious belief in problematic ways. Taking conscious mentality as a model for all mentality distorts the phenomena – elements are treated as episodic when they are not, as akin to sensory episodes or by postulating “a psychological structure which corresponds in a more or less direct way with the structure of conscious judgement or assertion” (Crane and Farkas 2022: 36–37). This seems innocent enough at first, in line with the primacy of the conscious, but as useful as these might be in commonplace explanations of behavior involving beliefs and desires, not all behavior can be attributed to episodes that parallel this doxastic/conative episode structure, but at the unconscious level. What’s occurring at the unconscious level may not be episodic, and/or it may involve elements that diverge in significant ways from the doxastic/conative elements that appear in our conscious level explanations. This primacy of the conscious is engrained enough either to escape notice or to feel unproblematic as the lens to view the unconscious, but it has profound impacts on how we view unconscious mentality. 60

Implicit Cognition and Unconscious Mentality

There is a more helpful way to characterize the relationship between the conscious and the unconscious. Standing states are not, by default, conscious, and it is not clear how one would take a conscious standing state, remove some conscious-making feature, and have an unconscious standing state as a result. Nor would it seem that one can take the unconscious standing state and add some conscious-making feature and render it the conscious version of that state. A pair of methodological decisions drives this distortion. The first is the assumption that the conscious and the unconscious realms operate using similar states and processes, that one can use a model of the conscious as a guide to the unconscious. The second specifies which similarities can be anticipated – that the features of conscious occurrent judgment will have corresponding features present among the unconscious intentional states. Yet, there are reasons to think the realms are distinct, and there are reasons to think that the occurrent judgment model is misleading. There should be room for a radically different approach that takes seriously the hypothesis that the unconscious mind differs profoundly from the conscious in the way it represents the world, that unconscious mental representation works in very different ways from representation in consciousness. The unconscious is the basis of our psychological organization, but it may not be organized in the way that our conscious minds are. Our preceding discussion focuses on the need for an account of unconscious intentionality to handle standing states. Such standing states play essential roles in our mental lives and our accounts of behavior, so they demand a proper treatment. But most of the mechanisms and states attributed by cognitive science are unconscious, whether they are occurrent or standing, so a proper account of the reality of these mechanisms and states requires a robust account of unconscious intentionality.

4.  Nondeflationary Accounts In what follows, we outline a range of nondeflationary alternatives that allow for the possibility that the unconscious bears little resemblance to the conscious. Our primary goal is to establish this approach as superior to those approaches closely aligned with the primacy of the conscious. With this alternative approach established, theorists can pursue a number of paths in exploring the different ways that the unconscious might be organized differently from the conscious.

4.1  Option One: A Common Alternative The most common alternative to the Dominant Approach asserts that the unconscious is truly distinctive from the conscious but holds that the unconscious realm possesses states and processes that are characterized in contrast to those that exist in the conscious realms. For example, one can adopt a view of conscious mentality that involves rational, deliberative, and propositional thought while stipulating that the unconscious realm involves nonrational, nondeliberative, nonpropositional thought. The two realms operate according to different principles. To better understand this approach, consider the phenomenon of implicit bias. A widely discussed case is that of implicit racist beliefs. It is often supposed that in such cases, a subject has consciously held egalitarian beliefs, but unconsciously held racist aliefs (Gendler 2008), attitudes, or biases. It stands to reason that the consciously held beliefs involve a psychological structure with a propositional content that drives egalitarian verbal reports, whereas the unconscious associations between racial categories and evaluative terms drive different nonverbal behaviors, like the amount of time it takes to react to a stimulus or sort objects. A proponent of the Dominant Approach might explain the discordant behaviors as stemming from a conflict between two beliefs, an egalitarian conscious belief and a racist belief. The 61

Tim Crane and J. Robert Thompson

racist belief could have been a conscious belief, but it contingently lacks the conscious-making feature. So, on this view, there are not two realms, one rational/evidence-sensitive/propositional and one irrational/evidence-insensitive/associationistic. Since all beliefs operate similarly, any discordant behavior in cases of implicit bias is due to whether the conscious-making feature is present (or not) in one belief or the other. Many writers have indicated the benefits and drawbacks to handling these sorts of cases along the lines of the Dominant Approach – in adapting one’s conception of belief to handle such cases, in developing additional ‘in-between’ approaches (see Schwitzgebel (2010), Brownstein and Saul (2016a, 2016b)) – or along the lines of the alternative described here involving associationistic structures. Given our commitments laid out earlier, the reader should be able to determine that we don’t find explanations that focus on the presence or absence of consciousness for some beliefs as being particularly illuminating, or even coherent. Moreover, we think the alternative just sketched is overly limiting as a general approach to the unconscious. Our alternative approach would start with the point argued earlier: that belief should not be treated as a conscious level phenomenon. If so, then it is preferable treat the egalitarian element as a conscious episode of judgment that is in conflict with racist unconscious associations, rather than a conscious belief that is in conflict with unconscious associations. Secondly, the Dominant Approach tends to assume a sharp delineation between the states and processes that are sensitive to evidence/rational/propositional and those that are not. And this is paired with the expectation that the states and processes that are sensitive to evidence/rational/propositional will be the conscious ones. Current research into implicit bias indicates that the distinctions are far less tidy than this, and that the unconscious states and processes involved in these cases are sensitive to evidence in ways that are not captured by the alternative. For example, at least some of the elements involved in implicit bias appear capable of modulation by rational argumentation and/ or logical interventions (e.g., Sullivan-Bissett, this volume Ch. 8; Mandelbaum 2016). We would like to draw a more general moral here: what happens in cases of discordant behavior like those described as “implicit bias” is not necessarily a struggle between two realms, one conscious, propositional, evidence-sensitive, and one that lacks all of those features. Indeed, it may be that both conflicting elements exist in the absence of conscious awareness. Crane and Farkas (2022) also note that the roots of discordant behavior can be combined in many ways that the alternative seems to overlook. The roots might be not at all doxastic (e.g., emotions, associations), somewhat doxastic (e.g., aliefs), or fully doxastic (e.g., beliefs). The point is that implicit bias and similar discordant phenomena are unlikely to be explained only as the result of two realms operating according to different principles, one that is rational, deliberative, propositional, and conscious and one that is nonrational, nondeliberative, nonpropositional, and unconscious.

4.2  Option Two: Inferential Integration Not every distinctive feature of an unconscious mental state or process can be characterized as resulting from their operating in nonrational, nondeliberative, or nonpropositional ways. This is perhaps the most serious limitation with Option One: some unconscious mental states are distinctively different from consciously available states, but not in a way that precludes them from participating in inferences, being sensitive to rational argumentation, or possessing propositional contents. One way of characterizing this distinctiveness is to say they lack a certain level of inferential integration that limits their functional profiles and thereby limits their availability for conscious access and verbal report. Another way of describing this limitation of integration is in terms of participation: the relevant knowledge or structures don’t appear to participate in many projects 62

Implicit Cognition and Unconscious Mentality

(Miller 1997) or are largely “harnessed to [a] single project” (Wright 1986: 227). In this sense, what is implicit will not be fully cognitively or inferentially integrated, will rarely or never be accessed by some other system for some other purpose, and will rarely (if ever) be subject to verbal report. Many examples of implicit cognition exhibit this sort of unconscious profile. As noted in discussing Option One, some cases of implicit bias, upon examination, may involve unconscious mental states and processes that fail to be nonrational, nondeliberative, or nonpropositional in the requisite ways. As noted in §1, many of the representations postulated by cognitive scientists involve unconscious structures. Option One and the Dominant Approach suggest only two possibilities in such cases: either these structures are nonrational, nondeliberative, and nonpropositional, or they are just like conscious representations, only (contingently) lacking the conscious-making feature. Option Two rejects this dichotomy. The unconscious structures can share features with conscious mentality but differ in terms of their inferential integration – they are limited in terms of their participation but are unlike the unconscious elements proposed in Option One. On this view, there is a range of unconscious mentality that seems to differ from conscious mentality in ways that suggest that the presence of mere associations could not be the full story. At least some of the behaviors are rich enough to posit a range of full-blown mental states to the subject that are unconscious. These do not get reported as frequently (if ever) and have a limited impact on behavior, but there is reason to deem them as propositional. Option Two allows for thoughts with propositional contents and inferential connections, but their limited integration explains many of the features found implicit cognition. For example, cognition that is merely implicit is often used by an agent but not acknowledged by that agent when verbally prompted. And such implicit understanding often appears somewhat sporadically – it is exploited in some contexts but not in others, even where it would have been helpful. But, the fact that some understanding does not arise in conscious or verbally reported forms in every possible circumstance does not require us to conclude that this understanding is flawed, limited, must be insensitive to evidence, or is the result of mere association, and the like. Option Two explains why the presence of cognitive machinery that lacks a certain level of inferential integration would manifest itself in these ways that theorists have identified as implicit. To fully distinguish these sort of cases from those covered by Option One, theorists will need to do the difficult work of showing that the unconscious understanding driving some bit of behavior is too rich and sophisticated in the requisite ways to come from non-evidence sensitive, nonpropositional mental structures. And it needs to explain how it differs from the Dominant Approach by demonstrating how a lack of inferential integration accounts for the phenomena better than mere absence of the conscious-making feature. Option One views the unconscious realm as exclusively irrational and associationistic (both in its processes and their contents), but this is too narrow a conception to account for all cases of unconscious mentality. Option Two shows that while some unconscious mentality may take the form suggested by Option One, there will be some cases where the features suggested by Option One will not suffice. Although we think there are several domains in cognitive science where Option Two is a viable approach to adopt, detailed work needs to be done in such cases to characterize the particular features exhibited by the unconscious intentional states and identify why the alternatives are inadequate (see Thompson 2014 for an example of what this might look like in developmental psychology).

4.3  Option Three: A Unifying Interpretative Difference? The preceding discussion suggests a certain heterogeneity in the unconscious realm that might warrant a pluralistic approach – perhaps the unconscious operates in ways that differ from the 63

Tim Crane and J. Robert Thompson

conscious in some instances, and ways that share some similarities with the conscious in others. In this section, we consider the extent to which there may be a more profound distinction between conscious and unconscious mentality that is reflected in and captured by our interpretational practices that deal with the unconscious realm. This may indicate something important about the states and processes themselves, about the nature of the sorts of representations that are brought to bear in these situations. Implicit cognition is like any theoretical construct in psychology – it is described by a process of interpretation. This need for interpretation applies both to the familiar kinds of unconscious mental states that we attribute to ourselves and to others in our everyday psychological thinking (thoughts, feelings, desires, intentions, etc.), as well as to the unconscious mental states that cognitive science attributes to the brain or the subject. On this approach, although there clearly is unconscious mental representation, it is less determinate and explicit than conscious representation, and requires interpretation in a way that consciousness does not. Drawing from Crane (2017), the alternative sketched here is that unconscious intentional states bear a different relationship to their own interpretation than conscious intentional states do. The central hypothesis is that the interpretation of unconscious mental states – whether by subjects attributing mental states to other human beings and to animals, by subjects themselves in self-attribution of belief, or by scientists attributing representational states to mechanisms in the brain – imposes an order on something that is much less ordered, explicit, and determinate than the representations we find in consciousness. This is the common thread that links all applications of the idea of the unconscious in philosophy and the various branches of psychology – even as philosophy and psychology have failed in their attempts to conceptualize the unconscious. Here we briefly consider both the attribution of intentionality to ourselves, and the attribution of representation in cognitive science. What is it to attribute an intentional state to oneself? And how does one know what one’s own intentional states are? This question has been intensely debated in analytic philosophy under the (not entirely accurate) heading of ‘self-knowledge’ (Cassam 1994, 2014; Macdonald et al. 1998). How can one know what one thinks, or feels, or wants? This question poses a problem because the usual mechanisms of knowledge seem to have no application here – we do not seem to know our mental states by perception, testimony, or inference. Or at least, there is not one of these models that works for all cases of self-knowledge. Saying that we know our mental states by ‘introspection’ seems to name the process rather than describe an actual mechanism. What does the process of the acquisition of knowledge of our unconscious states tell us about the nature of the states themselves? Standard approaches assume that the process of finding out what you think is a matter of finding out what is (so to speak) ‘already there’. There are relatively fixed facts about your dispositions and these facts line up in a straightforward (if complex way) with attributions of intentional states in the language of commonsense psychology. The only question is to figure out what these facts are. But finding out what you think seems to be a different thing from making up your mind. Compare practical reasoning. When you are figuring out what to do, you will often deliberate and weigh up the various options and arrive at a decision or the formation of an intention. Similarly, when you are figuring out what to believe, you weigh up the evidence and come to an opinion. But this is supposed to be a different process from finding out what you already believe, though it will normally draw on things you already believe. The standard approaches assume, in short, that there is a sharp distinction between finding out what you think (and want etc.) and making up your mind. This assumption is questioned by Crane (2017), who argues not that there is no distinction at all between finding out what you 64

Implicit Cognition and Unconscious Mentality

think and making up your mind – that would be absurd – but that it is not a sharp distinction. In other words, there will be cases where the affirmation of something in consciousness could be conceived of as making a judgment about something on which you are genuinely unclear, or it could be a matter of reporting a straightforward belief – and there need be no fact of the matter about which of these is the correct description (Moran 2001). This suggests an account of self-knowledge that allows for self-knowledge to be the product of self-interpretation, where self-interpretation is an essentially creative enterprise. Interpretation shapes one’s vague or inchoate unconscious mental reality into the more determinate, specific form of a conscious judgment. Consciously putting one’s thoughts into words is perhaps the clearest example of this, but there are many others. Recognizing the centrality of interpretation in the understanding of the unconscious also sheds light on the question of how to understand the representational content of the states that cognitive science attributes to the brain or its mechanisms. What does it mean to say that the brain or the visual system represents something in the world outside? A traditional view (defended by Fodor 1975, 1987) is that for psychological theories to be explanatorily useful, they must be literally true, and this implies that there must be distinct representational states within the subject. This theory was opposed by those who thought that explanatory usefulness does not imply that there are representations in this literal, concrete sense (see Dennett 1975). In an older debate, this latter view is sometimes called ‘instrumentalism’, as opposed to Fodor’s ‘realism’. This old dispute between realism and instrumental has stagnated in recent years. Each position seems too extreme. Instrumentalism seems to commit too little on the underlying structure of mental states, while realism over-commits (though see Quilty-Dunn and Mandelbaum 2018). This led some philosophers to create a middle path between the two extremes (for an early attempt, see e.g., Peacocke 1983). Important work on the role of representation in cognitive science has been done more recently (e.g., by Shea 2018 and Rescorla 2020). But nonetheless it is fair to say that there is no general consensus about how to think about ‘realism’ about representation in cognitive science. This suggests that a new approach is needed. Since most of the mechanisms and states attributed by cognitive science are unconscious, a proper account of the reality of these mechanisms and states must draw on a general account of unconscious mentality. Any approach to the role of representation in cognitive science should take on board developments in the philosophy of science. We believe that for this reason we should apply the idea of a model as used in the philosophy of science (Weisberg 2007) to the question of representation. A scientific model is a simplified, idealized description or other object that aims to identify and isolate features of the system under investigation, and to explain the system’s behavior by looking at the behavior of the model. Thus Rutherford’s solar system model of the atom was an idealized representation of the atom’s structure, but one that enabled understanding, explanation, and prediction. Similarly, computational models of the mind or the brain are also simplifying descriptions of the activity of mental faculties, which enable understanding, explanation, and prediction of that activity. The models attribute propositional contents to states of the brain in the way analogous to the way that measurements of physical magnitudes employ numbers (Matthews 2010). This suggests that there will not be a unique content to any given state: the precise content attributed will be relative to the model. What is important is to show that this does not make the intentionality of the brain or cognitive system ‘unreal’ or ‘merely instrumental’. On this picture of the place of representation in cognitive science, abstract propositions do play a role, in modeling the unconscious mental states and processes. But what they are 65

Tim Crane and J. Robert Thompson

modeling is in itself representational – and the models are used to isolate some aspect of that underlying representational structure. The case for saying that this is representational is based on the familiar fact that we cannot identify the ultimate task that the brain or organism is performing without talking in representational terms: for example, the visual system’s ultimate role is to create a representation of the visible world (Marr 1982; Burge 2010). But different theorists will interpret this representational structure in different ways, and use different models, some of which will be better than others in understanding the various subsidiary tasks performed by the system. In its broad outlines, this conception of the role of content in cognitive science owes a lot to Cummins (1989), Dennett (1981), and especially Egan (2012). But these ideas have often been misleadingly associated with ‘instrumentalism’ where that label carries the insinuation that the conception does not treat intentionality as sufficiently real. This is partly because the standard for an intentional state being ‘real’ has been set in an implausibly simplistic way by the Fodor-style realist, or by general metaphysical maxims associating reality with causal efficacy. What has gone wrong here is not the modeling picture but the conception of what realism requires. We need a better picture of what it is for a representational state of the brain to be real, to understand the reality modeled by cognitive science: the representational reality of the mind is made more explicit and determinate by the theorist’s interpretation, which assigns specific contents to specific states in its model. Finally, it is worth pointing out that this picture of the attribution of intentional states to ourselves and to others has a connection with certain central ideas from psychoanalysis. Psychoanalysis remains controversial of course, both as a therapy and as a theory – and we do not endorse any specific psychoanalytic theories. However, our approach shares with psychoanalysis the idea that the unconscious mind operates according to its own principles; mental representation works in a different way in the unconscious from the way it does in the conscious mind (cf., Katsafanas 2016). The relevance of psychoanalysis is that it puts at the heart of its theory and practice the fact that there are reasons why people do things that they do not themselves fully understand; and drawing out what these reasons are may involve imposing an order on something that does not in itself have such an explicit order. This is where conscious interpretation imposes an order on the relatively unformed and chaotic unconscious. Despite the controversies surrounding psychoanalysis, many of the phenomena it attempts to explain are real, and it is an advantage of an account of the unconscious that it can make room for them. Our understanding of the role of interpretation, we claim, has the potential to provide a psychologically realistic account of the relationship between the various manifestations of the unconscious, while also preserving the distinctive unity of the unconscious mind and its distinction from consciousness.

5. Conclusion Cognitive science needs a conception of unconscious mentality that serves its explanatory needs. To achieve this, theorists need to reconsider the primacy of the conscious and its Dominant Approach to unconscious mentality. Although we think there is much to be gained by pursuing the approach sketched in §4.3, once theorists are able to view the unconscious realm in its own right, more adequate attention can be given to the different ways in which intentional states might be unconscious.

Related Topics Due to the centrality of consciousness to the study of implicit cognition, there are no principled grounds for excluding any of the other chapters in the Handbook. 66

Implicit Cognition and Unconscious Mentality

References Bayne, T., and Montague, M., eds. 2011. Cognitive phenomenology. Oxford: Oxford University Press. Block, N. 1995. “On a confusion about a function of consciousness”. Behavioral and Brain Sciences, 18: 227–247. Brownstein, M., and Saul, J., eds. 2016a. Implicit bias and philosophy, vol. 1: metaphysics and epistemology. New York: Oxford University Press. Brownstein, M., and Saul, J., eds. 2016b. Implicit bias and philosophy, vol. 2: moral responsibility, structural injustice, and ethics. New York: Oxford University Press. Burge, T. 2010. Origins of objectivity. Oxford: Oxford University Press. Carruthers, P. 2003. Phenomenal consciousness: a naturalistic theory. Cambridge: Cambridge University Press. Carruthers, P. 2011. The opacity of mind: an integrative theory of self-knowledge. Oxford: Oxford University Press. Cassam, Q., ed. 1994. Self-knowledge. Oxford: Oxford University Press. Cassam, Q. 2014. Self-knowledge for humans. Oxford: Oxford University Press. Chalmers, D. 2004. “The representational character of experience”. In B. Leiter, ed., The future for philosophy. Oxford: Oxford University Press: 153–181. Chomsky, N. 1980. Rules and representations. New York: Columbia University Press. Crane, T. 2003. “The intentional structure of consciousness”. In A. Jokic and Q. Smith, eds., Consciousness: new philosophical perspectives. Oxford: Oxford University Press: 33–56. Crane, T. 2009. “Intentionalism”. In A. Beckermann and B. McLaughlin, eds., The Oxford handbook to the philosophy of mind. Oxford: Oxford University Press: 474–493. Crane, T. 2013. “Unconscious belief and conscious thought”. In U. Kriegel, ed., Phenomenal intentionality: new essays. Oxford: Oxford University Press: 156–173. Crane, T. 2017. “The unity of unconsciousness”. Proceedings of the Aristotelian Society, CXVII: 1–22. Crane, T. 2020. “A  short history of the philosophy of consciousness in the Twentieth Century”. In A. Kind, ed., Philosophy of mind in the Twentieth and Twenty-First Centuries: the history of the philosophy of mind. Vol. 6. London: Routledge. Crane, T., and Farkas, K. 2022. “The limits of the doxastic”. In U. Kriegel, ed., Oxford studies in philosophy of mind. Vol. 2. Oxford: Oxford University Press: 36–57. Cummins, R. 1989. Meaning and mental representation. Cambridge, MA: MIT Press. Dennett, D. C. 1975. “Brain writing and mind reading”. In K. Gunderson, ed., Minnesota Studies in the Philosophy of Science, 7: 403–415. Dennett, D. C. 1981. “A cure for the common code”. In Brainstorms. Hassocks: Harvester: 90–108. Egan, F. 2012. “Metaphysics and computational cognitive science: let’s not let the tail wag the dog”. The Journal of Cognitive Science, 13: 39–49. Farkas, K. 2008. “Phenomenal intentionality without compromise”. Monist, 91: 273–293. Fodor, J. A. 1975. The language of thought. Cambridge, MA: Harvard University Press. Fodor, J. A. 1987. Psychosemantics: the problem of meaning in the philosophy of mind. Cambridge, MA: MIT Press. Freud, S. 1915. The unconscious. London: Hogarth. Gendler, T. S. 2008. “Alief in action (and reaction)”. Mind & Language, 23: 552–585. Helmholtz, H. 1867. Handbuch der physiologischen optik. Vol. 3. Leipzig: Voss. Katsafanas, P. 2016. The Nietzschean self: moral psychology, agency, and the unconscious. Oxford: Oxford University Press. Kihlstrom, J. F. 1987. “The cognitive unconscious”. Science, 237: 1445–1452. Kriegel, U., ed. 2013. Phenomenal intentionality. New York: Oxford University Press. Leibniz, G. W. 1704. New essays on human understanding. Trans. P. Remnant and J. Bennett. 1981. Cambridge: Cambridge University Press. Macdonald, C., Smith, B. C., and Wright, C. J. G., eds. 1998. Knowing our own minds: essays in self-knowledge. Oxford: Oxford University Press. Mandelbaum, E. 2016. “Attitude, inference, association: on the propositional structure of implicit bias”. Noûs: 629–658. Marr, D. 1982. Vision: a computational investigation into the human representation and processing of visual information. San Francisco: W. H. Freeman and Company. Matthews, R. 2010. The measure of mind. Oxford: Oxford University Press. Miller, A. 1997. “Tacit knowledge”. In B. Hale and C. Wright, eds., A companion to the philosophy of language. Oxford: Blackwell. Moran, R. 2001. Authority and estrangement: an essay on self-knowledge. Princeton: Princeton University Press. Peacocke, C. 1983. Sense and content: experience, thought and their relations. Oxford: Oxford University Press.

67

Tim Crane and J. Robert Thompson Quilty-Dunn, J., and Mandelbaum, E. 2018. “Against dispositionalism: belief in cognitive science”. Philosophical Studies, 175: 2353–2372. Reber, A. S. 1993. Implicit learning and tacit knowledge: an essay on the cognitive unconscious. Oxford: Oxford University Press. Rescorla, M. 2020. “Reifying representations”. In J. Smorthchkova, T. Schlicht, and K. Dolega, eds., What are mental representations? Oxford: Oxford University Press: 135–177. Rosenthal, D. M. 2005. Consciousness and mind. Oxford: Oxford University Press. Russell, B. 1921. The analysis of mind. London: George Allen and Unwin. Schwitzgebel, E. 2010. “Acting contrary to our professed beliefs, or the gulf between occurrent judgment and dispositional belief ”. Pacific Philosophical Quarterly, 91: 531–553. Searle, J. 1992. The rediscovery of the mind. Cambridge, MA: MIT Press. Shea, N. 2018. Representation in cognitive science. Oxford: Oxford University Press. Thompson, J. R. 2014. “Signature limits in mindreading systems”. Cognitive Science, 38: 1432–1455. Tversky, A., and Kahneman, D. 1974. “Judgment under uncertainty: heuristics and biases”. Science, 185: 1124–1131. Weisberg, M. 2007. “Who is a modeler?”. British Journal for the Philosophy of Science, 58: 207–233. Weiskrantz, L. 1986. Blindsight: a case study and implications. Oxford: Clarendon Press. Wilson, T. D. 2002. Strangers to ourselves: discovering the adaptive unconscious. Cambridge, MA: Harvard University Press. Wright, C. 1986. “Theories of meaning and speakers’ knowledge”. In Realism, meaning, and truth. Oxford: Blackwell.

68

4 IMPLICIT COGNITION IN RELATION TO THE CONCEPTUAL/ NONCONCEPTUAL DISTINCTION José Luis Bermúdez and Arnon Cahen 1. Introduction This chapter explores implicit cognition in relation to the distinction between conceptual and nonconceptual content, familiar from a range of contexts in philosophy of mind, epistemology, and cognitive science. We focus on two very different explanatory contexts where implicit knowledge is standardly invoked. The first is in explaining linguistic competency – a subject’s ability to form and understand an open-ended range of sentences. Explanations of this capacity standardly appeal to the subject’s possession of some implicit (and specifically tacit) knowledge in the form of propositions encoding the generative grammatical principles, or rules, of the language. A second familiar application of implicit knowledge is in the context of characterizing a subject’s body of procedural knowledge, their knowing how to do something or another. On the face of it, these are two very different cases. In the case of tacit linguistic knowledge, what the subject (implicitly) knows is some set of well specified (by linguists, though not by the subject) rules, which are implicated in the subject’s production of linguistic behavior. In contrast, cases of know-how – say, knowing how to ride a bike – do not seem to involve any such implicit knowledge of rules, nor does appeal to such rules seem necessary to explain the subject’s bike-riding behavior. Yet, though these cases differ significantly, the tendency to assimilate them as instances of implicit knowledge rests on two shared features. First, in both cases it is meaningful to attribute to the subject some knowledge in the service of explaining certain of the subject’s observed behavioral regularities, judgments, and competencies. This is the epistemic condition. Second, in both cases, the subject is not in a position to articulate the epistemic basis that explains those regularities. This is the inarticulability condition. Thus, in the case of linguistic competence, we have reason to attribute to subjects some grammatical knowledge that would explain their competence, yet subjects cannot articulate the grammatical principles or rules that underly their linguistic behaviors. In cases of know-how, a person competently riding a bike clearly knows how to ride a bike yet she is nonetheless not in a position to articulate what it is that she knows that would explain her competency. DOI: 10.4324/9781003014584-6 69

José Luis Bermúdez and Arnon Cahen

In the following, we will argue that such cases of implicit knowledge are more deeply related, in that they involve representations with nonconceptual contents that simultaneously account for both conditions upon implicit knowledge. The involvement of representations with nonconceptual content explains why a purported instance of implicit knowledge counts as knowledge, by virtue of engaging content bearing states. Further, it explains why the subject is not in a position to articulate the contribution that such knowledge makes to the regularities observed within one’s own behaviors, judgments, and competencies, by virtue of that content being nonconceptual. We begin this chapter (Section 2) with an introduction to the distinction between conceptual and nonconceptual content. In Sections 3 and 4, we argue that an appeal to representations with nonconceptual contents can explain the attribution to some subject of implicit knowledge (and implicit cognitive processes operating on such knowledge). In each case, bringing nonconceptual content into the picture deepens our understanding of the phenomenon under discussion. In the first case, linguistic competency, it helps to explain why what is at stake properly counts as knowledge, and to demarcate the scope of that knowledge. In the second case, procedural know-how, it points to a middle ground in ongoing debates between intellectualist and ability-based accounts of that know-how.

2.  Conceptual and Nonconceptual Content The distinction at issue originally emerged from reflecting on the nature of perceptual content and its relation to the contents of (specifically, singular) thought (Evans 1982). Evans, and others after him, provided a wealth of arguments to the effect that, unlike the content of thought, and of other propositional attitudes, one’s capacity to represent the world perceptually is independent of which (if any) concepts one possesses. One can perceptually experience a volcanic eruption just by being awake in its vicinity while looking straight at it, even if one knows nothing about volcanoes and their activities. But one cannot have the thought ‘this volcano is erupting’, without possessing the concepts of a volcano and of eruption and exploiting those concepts in thinking. Thought is concept-dependent: A  specification of how the thinker is epistemically related to the world in thinking about it – that is, a specification of the content of one’s thought – is answerable to concepts exploited (and thereby possessed) by the thinker in having the thought. Not so in the case of perception. A specification of perceptual content, of how the perceiver is epistemically related to the world in perceiving it, need make no appeal to concepts she possesses in perceiving it. For brevity, we will not present the myriad arguments in favor of the concept-independent nature of perception and the debates these have sparked (see Bermúdez and Cahen 2020 for an extensive review). However, it is worth briefly highlighting some of the central considerations in this debate: 1. The fine-grained nature of experience: A canonical specification of one’s perceptual experience – a specification that captures how the world perceptually appears to one and explains the perceptual discriminations one is able to make on the basis of its so appearing – need not be constrained by concepts one possesses. The perceptual discriminations we are able to make are much finer-grained than our conceptual repertoire. Given that our perceptual discriminations are finer-grained than our conceptual repertoire, they cannot be explained by appeal to representations that exploit concepts pertaining to what we discriminate. 2. Ontogenetic and phylogenetic continuity: Perception appears to be something that we more or less share with infants and with others in the animal kingdom who seem not to possess any 70

Implicit Cognition and Conceptual/Nonconceptual Distinction

concepts whatsoever. Non-conceptual creatures perform similar perceptual discriminations as adult human beings, and to the extent that these differ the difference is best explained in terms of physiological differences (e.g., in the nature of their rods and cones) rather than differences in conceptual endowment. 3. Explaining observational concept acquisition and singular thought: One of Evans’s main motivations was to explain how perception can make possible empirical thought. On plausible empiricist assumptions, a central component of such explanation is to show how perception could explain our acquisition of observational concepts. Yet, to avoid vicious circularity, our having a perceptual experience must be independent of those concepts the possession of which the experience is meant to explain (Peacocke 2001). Though the distinction between ways of representing that are concept-dependent and ones that are concept-independent was initially introduced in the context of perception, it has since been applied more broadly. In the next section, we illustrate this broader applicability by arguing that the availability of ways of representing that are independent of the conceptual capacities of the subject can explain our attributing implicit knowledge to that subject.

3.  Implicit Knowledge and Nonconceptual Content It is standard practice in cognitive science to appeal to mental representations to explain certain regularities in a person’s actions, judgments, or competencies. A classic example comes with explanations of linguistic competence, such as grammaticality judgments (e.g., Chomsky 1965, 1980, 1986). Presented with the pair of sentences ‘The cat is on the mat’ and ‘The cat on the mat is’, competent English speakers correctly judge that only the former is grammatical. Given that they do so reliably, it seems plausible to say that they possess some grammatical knowledge that allows them to do so. Nonetheless, subjects are very poor at articulating the basis of their grammatical competencies. That is, although subjects’ grammaticality judgments are made in virtue of certain grammatical principles and rules – it is the best explanation of the regularities we observe in their judgments (a competence) – that information seems not to be available to the subjects themselves. As described, this explanatory project clearly reflects the two conditions identified earlier. The epistemic condition is met because certain behavioral regularities, judgments, and competencies are being explained in terms of the possession of a type of theoretical knowledge, while the inarticulability condition is met because subjects typically cannot explain the epistemic basis for their successful performance. Indeed, there is no reason to think that a subject who was presented (say, in a linguistics textbook) with a correct account of what their implicit knowledge consisted in, would be able to recognize it as such – or, by the same token, to recognize that an incorrect account was incorrect. One way to put this would be to say that the correctness condition of the knowledge attribution is completely relative to a third person observer. Given this, however, it is natural to ask why this counts as a case of knowledge at all. One might ask, for example, whether the proposed body of tacit knowledge is really anything more than a sophisticated description of a complex behavior.1 How might one mark the distinction between a theory-like set of linguistic rules that simply describes (and predicts) patterns of grammaticality judgments and forms of linguistic behavior, on the one hand, and a theory-like set of linguistic rules that articulates a body of knowledge that tacitly regulates a subject’s linguistic behavior and associated judgments? A classic response to this challenge is to stress that the linguistic theories in question are being proposed as causal-explanatory structures. That is, the information corresponding to a particular 71

José Luis Bermúdez and Arnon Cahen

linguistic rule is encoded in a discrete representational item that systematically plays a causal role in generating particular types of linguistic behavior.2 What the third person theoretical articulation captures, therefore, is a structure of causally efficacious representational entities. This causal-explanatory response does indeed seem effective against what might be called the behaviorist objection. It gives a way of thinking about attributions of tacit knowledge so that they plainly come out as doing more than simply describing and codifying linguistic behavior. But it does not answer more general worries about why it is appropriate to talk about knowledge in this context. The problem is that a version of the causal-explanatory response could be given for other systems that it would not be appropriate to describe in knowledge-involving terms. Remaining within the general sphere of language use, perhaps the most basic phenomenon in language comprehension is word segmentation. The stream of speech has to be organized in terms of words before syntactic structure can be determined and semantic processing applied. Bayesian approaches to developmental psycholinguistics have emphasized the importance of transitional probabilities in how young children learn to segment words.3 The transitional probability between two syllables is the probability of the second occurring conditional upon the first. One would expect, for example, that high transitional probabilities occur between two syllables that are part of the same word, whereas low transitional probabilities can mark word boundaries. This area is less developed than syntactic theory. However, there is no discernible reason to think that the same dialectic might not play out here. That is to say, if someone were to raise doubts about rules involving transitional probabilities being no more than redescriptions of word segmentation behavior, then it would be a satisfactory response to say that the relevant rules are encoded in discrete representational items that systematically play a causal role in generating particular types of linguistic behavior, and, moreover, that the structure and relations between the representational items is isomorphic to the theory that we are developing to describe. It does not seem right, however, to say that children and adults have an implicit knowledge of these rules, in the way that they are supposed to have an implicit knowledge of grammar. If this is right, then the causal-explanatory response at best picks up on a necessary condition of knowledge attributions. It is not by itself sufficient. So, what is missing? In the word segmentation case, we do indeed have an attribution of nonconceptual content, in the sense discussed earlier. However, we do not have an attribution at the personal level (on the personal/ subpersonal distinction see Dennett 1969; Drayson 2014). This is because a personal-level content ascription is appropriate only when there is a clearly identifiable activity being exercised at the level of the person, and it does not seem right to think that the person engages in word segmentation. Word segmentation has to occur before the person can engage in linguistic activity, but it is not something that the person herself does. We mark the distinction between subpersonal and personal content ascriptions by considering the user of the representation. When the user is the person, we have a personal-level content-ascription. When the user is a subpersonal module or system, then we have a subpersonal content-ascription. But how can one identify the user of the representation? We propose a two-step functional account. The first step is to identify the cognitive/practical function that the representation serves, and the second step to identify for whom or for what that function is directly performed. In the case we are considering, representations of transitional probabilities have the function of demarcating word boundaries, and this function is performed for the direct benefit of language-processing systems that are further downstream – and, more‑ over, only for the benefit of those systems. There is no activity at the level of the person that corresponds directly to the enterprise of demarcating word boundaries (although there are, of course, many activities that are made possible by that enterprise). One sign of this is that there are no judgments of word boundaries (outside highly specialized contexts). 72

Implicit Cognition and Conceptual/Nonconceptual Distinction

Among those language-processing systems further downstream is the system (or systems) whose function it is to assess the grammaticality of strings of words. Let’s assume for simplicity that this is a single system. Representations within that system serve the function (obviously) of determining grammaticality for other language-processing systems (e.g., those responsible for semantic processing), but they also determine grammaticality for language-users themselves. One index of the difference here is that exploring grammaticality judgments are key to studying grammatical structures and grammatical knowledge, while there is no equivalent for word segmentation, which is usually studied in infants using indirect measures such as the preferential listening paradigm (as pioneered, e.g., in Saffran et al. 1996). In fact, it is not really correct to talk about word segmentation judgments at all. Word segmentation is a quasi-perceptual (or more accurately: pre-perceptual) ability – the ability to parse continuous streams of sound into words, even though there are no auditory markers of the gaps between words. This line of thought is illustrated by, but does not depend on, the specifics of word segmentation and grammatical knowledge. Cognitive science is full of examples where content bearing representational states explain certain behavioral regularities without attributing knowledge to the subject. An example is Marr’s appeal to the representational stage he calls the primal sketch. In contrast with his proposed 3D model, on the basis of which the perceiving subject identifies objects and their allocentric locations, it is implausible to count the primal sketch among the perceiving subject’s epistemic states. This is because though it is plausible to treat the subject as perceptually identifying objects, it is implausible to think that the subject engages in identifying zero-crossings, though something within the subject must perform the latter for the subject to perform the former. Putting this all together yields the following proposal. Subpersonal content-ascriptions are appropriate for representations within systems performing functions that are of direct benefit only to further information-processing systems, and for which there is no directly corresponding activity at the level of the person. In contrast, personal-level content-ascriptions are appropriate for representations within systems for which there is a directly corresponding activity at the level of the person, such as the systems that process information about syntax and grammar. This, then, gives us an answer to the question of how an appeal to nonconceptual content explains why implicit knowledge count as knowledge. Ascriptions of implicit knowledge require the holding of causal-explanatory facts about how representational structures make possible certain types of linguistic behavior. They also require those representational structures to have nonconceptual content. But those two points are not enough, as the example of word segmentation shows. We only have implicit knowledge when what is ascribed to those causal-explanatory representational structures is personal-level nonconceptual content. This, moreover, gives us a principled way of demarcating when attributions of implicit knowledge are appropriate.

4.  Practical Knowledge and Know-How Turning to the second familiar context, the case of knowing how to do something or another, we can again appreciate that the two conditions specified earlier are satisfied. Take as our working example Mary’s knowing how to ride a bike. Mary who has been riding since an early age, rarely falls off her bike, she rides up hills and down valleys, around curves at various speeds, stops for crossing pedestrians, and merges seamlessly into traffic. It seems straightforwardly appropriate to attribute to her a certain kind of knowledge that her successful execution of this wide range of performances manifests. However, it is also a familiar point, that Mary, like most of us, cannot articulate what she knows in a way that accounts for her complex performances.4 As 73

José Luis Bermúdez and Arnon Cahen

such, it appears that both the epistemic condition and the inarticulability conditions hold, so that what Mary knows when she knows how to ride her bike she knows implicitly. Yet, here too deeper worries arise. Starting with the epistemic condition and following Ryle (1946, 1949), who introduced the notion of ‘know-how’ (distinguishing it from the propositional knowledge implicated in ‘know-that’), we might think that for Mary to know how to ride a bike is simply for Mary to have an ability to ride a bike. For Ryle, this ability is to be understood in terms of (complex context-sensitive) corresponding dispositions – for example, a disposition to lean into the direction of a curve, to press the breaks when approaching a crosswalk, etc. Not only is a specification of these dispositions open-ended, as the contexts in which one might (successfully) ride a bike are open ended, but it is clear that Mary, herself, need not be able to specify any of the relevant dispositions that purportedly constitute her knowing how to ride a bike. If so, why might we still think of know-how as a kind of knowledge, rather than ‘a sophisticated description of a complex behavior’? This kind of approach seems to be suggested by anti-intellectualists about skill and practical knowledge (e.g., Noë 2005; Dreyfus 2002, 2005; Glick 2011, 2012; Wiggins 2012). Yet, the anti-intellectualist, ‘ability account’, of know-how has faced multiple objections from intellectualists, who wish to assimilate know-how to the kind of conceptual, propositional, knowledge implicated in knowing that (e.g., Stanley and Williamson 2001; Snowdon 2004; Stanley 2011a, 2011b). They note that knowing how to ride a bike cannot just be having an ability to ride a bike, as the latter is neither necessary nor sufficient for the former. It is unnecessary, as it seems that even if Mary were to become disabled in some way, and therefore lose the ability to ride a bike, it would still be appropriate to characterize her as knowing how to ride a bike (Stanley and Williamson 2001). It is insufficient, as it seems that Mary could have the ability to ride a bike by sheer lucky accident, without thereby knowing how to ride a bike (Carr 1981).5 Reflecting on such purported counterexamples, points to the difficulty at the heart of the ‘ability’ account and urges us closer to the intellectualist camp. The core of the intellectualist complaint is that identifying know-how with a set of dispositions (however complex they may be) is explanatorily vacuous; it is merely a re-description of the phenomenon that an appeal to the subject’s knowledge is supposed to explain. This is but an instance of the familiar, general, complaint against behaviorism.6 In attributing to Mary knowledge how to ride a bike we are not merely characterizing her complex competency but attributing to her some inner state that explains her actual and counterfactual performance – it explains why ‘Disabled Mary’ still knows how to ride a bike and ‘Lucky Mary’ does not. Given that the attribution to Mary of knowledge how to ride a bike is meant to explain her ability to do so, it cannot be identical with that ability. Thus, contrary to the anti-intellectualist ‘ability account’, we have good reason to suppose that when truthfully attributing to Mary knowledge how to ride a bike, we are attributing to her a genuine kind of knowledge on the basis of which (and when other extrinsic conditions prevail – e.g., when she is awake, intact, and in possession of a bike) she proficiently rides her bike. Knowing how to Φ is what the subject has to go on when attempting to Φ. But this raises the further question of what it is that subjects have to go on, that is, what kind of knowledge is this knowledge how to Φ? We suggest that the personal-level knowledge that is available to the subject and guides her skilled performances is available to her nonconceptually. In denying the ability account on the one hand and the conceptual nature of know-how on the other, such a position presents a middle ground between traditional intellectualists and anti-intellectualists.7 Initial motivation for this position is precisely the fact that the subject is not (or need not be) in a position to articulate the epistemic basis of her performances in a way that would show them to be appropriate given what she knows. At best, Mary would be able to say something along the lines of ‘that adjustment simply seemed most appropriate to that curve at that speed’. Yet such a response does nothing 74

Implicit Cognition and Conceptual/Nonconceptual Distinction

to illuminate what it was about her circumstances (e.g., curve and speed) that, given what she knows, favored the particular adjustment she made; that is, it is silent precisely about what her knowing how to ride a bike consists in. The best explanation of this fact is that her knowledge is not available to her in a form that is amenable to such articulation. And this, in turn, is best explained by the fact that such knowledge does not (or at least need not as a condition on playing its epistemic and practical role) employ concepts in her possession.8 Nonetheless, this challenge from inarticulability has been countered by the most influential intellectualist position (Stanley and Williamson 2001; Stanley 2011a, 2011b). According to Stanley (2011b), so long as we allow demonstrative expressions, there is no problem of articulation of what one knows. Responding to an example from Schiffer (2002), he says: “The 8 year old Mozart can assert the proposition that constitutes his knowledge how to compose a symphony; he can just say say [sic.], while composing it, the German translation of ‘this is how I can do it’ ” (2011b: 14, emphasis ours). Thus, though Mary may be unable to verbalize in non-demonstrative terms what she knows in knowing how to ride a bike, she is perfectly able to express the proposition that constitutes her knowing how to ride a bike demonstratively – she can simply get on the bike and say ‘this is how I ride a bike’. However, appeal to demonstrative concepts in articulating what Mary knows is no less problematic than a similar strategy purporting to support a conceptualist account of perceptual content. One influential argument, mentioned earlier, in favor of the nonconceptual content of perception is the claim that our perceptual discriminations far outstrip our conceptual capacities. We can discriminate many more shades than we have concepts for. As a result, how we perceptually represent the world as being – that which accounts for the fine-grained perceptual discriminations we manifest – cannot be constrained by concepts we possess. In response, conceptualists have argued (e.g., McDowell 1994) that though we may not possess fine-grained general concepts for every shade we can perceptually discriminate, when presented with some particular shade, a demonstrative concept such as ‘that shade’ (unsurprisingly) captures the exact fineness of grain with which the shade is perceptually represented. Returning to the practical case, given that on any occasion of Mary’s riding her bike, her way of riding the bike is manifest, the demonstrative ‘that way of riding a bike’ will pick out exactly the way that Mary knows to ride a bike. Yet, as many have argued (e.g., Ayers 2002; Levine 2010; Roskies 2010), the central difficulty with the demonstrative strategy is that such demonstration is only possible on the heels of perceptual discrimination, and therefore cannot account for such discrimination. That is, the employment of the demonstrative concept presupposes our already having perceptually represented and thereby discriminated the shade picked out by the demonstrative. Consequentially, the availability of the demonstrative concept (e.g., ‘this shade’) cannot figure in an explanation of our perceptual capacities – it is the explanandum not the explanans.9 As with the perceptual case so with the practical case. Here, too, the conceptualist’s appeal to demonstrative concepts will not work. Note, first, that what Mary demonstrates at any given moment is a single context-bound manifestation of her practical knowledge. As such, it cannot exhaustively express the body of knowledge that constitutes Mary’s knowing how to ride a bike – indeed, no amount of such demonstrative reference will suffice – as the possible manifestations of her knowledge are open-ended and, as Ryle (1949) notes, ‘indefinitely heterogenous’ (just as Mozart’s knowing how to compose a symphony transcends all of Mozart’s token symphonic compositions).10 More deeply, relying on such demonstrative concepts to articulate what Mary knows has the order of explanation backwards; it is Mary’s knowledge how to ride a bike that, in any given bike-riding situation, makes the demonstrative available. Consequentially, allowing demonstrative concepts in the articulation of what Mary knows is no help to the conceptualist about practical knowledge. 75

José Luis Bermúdez and Arnon Cahen

In summary, the debate between the intellectualist and anti-intellectualist about know-how presents us with a false dichotomy. We need not choose between an ‘ability account’ that is blind to the explanatory significance of knowledge as an internal representation and a conceptual account of the content of that representation. Rather, there is a middle position that is both anti-intellectualist yet firmly representational – namely, recognizing that what one knows when knowing how to Φ is available to one nonconceptually. It is for this reason that one can exhibit what one knows (and make it available for demonstrative reference) even though one need not be able to articulate such knowledge.

5. Summary We have argued that implicit knowledge is attributable to a person when two conditions are met. The first is the epistemic condition, according to which the best explanation of certain personallevel regularities and competencies appeals to content bearing states that make their contents available to the person (rather than to some subpersonal system, as we argued is the case with word segmentation). The second condition is the inarticulability condition, according to which the subject need not be in a position to articulate the contribution that such content bearing states make to their observed regularities and competencies. The fact that implicit knowledge is beholden to this pair of conditions raises a prima facie conflict. The epistemic condition requires that content bearing states make their information available to the subject, yet subjects’ inability to make such information explicit sheds doubt on its availability. Thus, in the linguistic case, subjects’ inability to articulate the grammatical rules that govern their grammaticality judgments and broader linguistic competency suggests that such rules are not available to them, shedding doubt on the claim that such grammatical rules are genuinely known by subjects. Similarly, in the case of know-how, Mary’s failure to articulate the epistemic basis of her bike-riding ability casts doubt on the claim that know-how is a genuine kind of personal-level knowledge attributable to Mary, and may tempt us to an anti-intellectualist account of know-how as a mere ability or complex disposition. In both cases, the inarticulability condition seemingly undercuts the epistemic condition. An appeal to nonconceptual content resolves this apparent tension and thereby makes sense of attributions of implicit knowledge. In both cases, we are ascribing content at the personallevel, that is, content that is utilized by, and thereby available to, the subject in regulating her linguistic, bike-riding, or other behaviors, thus satisfying the epistemic condition and counting as genuine personal-level knowledge. Yet because such knowledge is available only nonconceptually, we have a ready explanation of the inarticulability (by the subject) of the content ascribed.

Related Topics Chapters 1, 3, 5, 13, 18, 19

Notes 1 This is exactly the worry that was first raised against Chomskyan linguistics by Quine (1972). 2 This general strategy was pioneered in Evans (1981) and refined by Davies (1987) and Peacocke (1989). 3 For an overview and references see §10.4 of Bermúdez (2020). 4 As Davies says: “it is manifestly possible to have a skill (such as the ability to tie one’s shoelaces) while being quite unable to give any verbal account of how the performance is achieved” (2001: 8126). Furthermore, as is the case with many such skills, an articulation of the knowledge that accounts for her

76

Implicit Cognition and Conceptual/Nonconceptual Distinction performances is rarely necessary for Mary, herself, to come to know how to ride a bike. Mary, like the rest of us, learned how to ride her bike – acquired the relevant knowledge – simply by riding it. 5 These objections have seen some resistance. As Glick (2012) argues, ‘Disabled’ Mary may have lost the opportunity to ride a bike but not her ‘inner ability’ to do so. Further, sheer lucky performance does not constitute an ability. The counterexamples, thus, fail to tease apart ability from know-how. Nonetheless, as we discuss later, we believe these objections successfully point to a deeper worry underlying the anti-intellectualist account. 6 Thus, by analogy with Putnam’s (1963) Spartans, we might say that Mary’s ability to ride a bike is but a natural sign, a symptom, of her knowing how to ride a bike. One can pretend to know how to ride a bike (by lucky accident) without knowing how, and one can know how to ride a bike without manifesting such knowledge. And, adapting Armstrong’s (1968) objection to the behaviorist, Mary’s knowing how to ride a bike is not merely her being disposed to behave in certain appropriate ways but is the ‘inner’ categorical basis of those dispositions. 7 Toribio (2015), who also defends a nonconceptual account of know-how, aptly presents her position as a non-dispositionalist anti-intellectualism, but it could have just as aptly been described as nonconceptual intellectualism. 8 Jung and Newen (2010) also argue that practical knowledge involves nonconceptual representational. In particular, two forms of nonconceptual content are implicated in know-how: Nonconceptual sensorimotor representations, which underly the abilities in question, and nonconceptual image-like representations, which are implicated in the acquisition and refinement of these same abilities. Our discussion here is broadly consistent with their view, but they may not accept our use of the personal/ subpersonal distinction. 9 Indeed, recall that Evans’s prime motivation for introducing the notion of nonconceptual content was to provide an explanation of the possibility of singular, demonstrative reference. 10 See also Hornsby (2011).

References Armstrong, D. M. 1968. A materialist theory of the mind. London: Routledge & Kegan Paul. Ayers, M. 2002. “Is perceptual content ever conceptual?”. Philosophical Books, 43: 5–17. Bermúdez, J. 2020. Cognitive science: an introduction to the science of the mind. 3rd edn. Cambridge: Cambridge University Press. Bermúdez, J., and Cahen, A. 2020. “Nonconceptual mental content”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Summer 2020 edn. https://plato.stanford.edu/archives/sum2020/entries/ content-nonconceptual/. Carr, D. 1981. “Knowledge in practice”. American Philosophical Quarterly, 18: 53–61. Chomsky, N. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press. Chomsky, N. 1980. “Rules and representations”. Behavioral and Brain Sciences 3: 1–15. Chomsky, N. 1986. Knowledge of language: its nature, origin, and use. New York: Praeger. Davies, M. 1987. “Tacit knowledge and semantic theory: can a five per cent difference matter?”. Mind, 96: 441–462. Davies, M. 2001. “Knowledge (explicit and implicit): Philosophical aspects”. In N. J. Smelser and P. B. Baltes, eds., International Encyclopedia of the Social & Behavioral Sciences. Amsterdam: Elsevier Science: 8126–8132. Dennett, D. C. 1969. Content and consciousness. London: Routledge and Kegan Paul. Drayson, Z. 2014. “The personal/subpersonal distinction”. Philosophy Compass, 9: 338–346. Dreyfus, H. 2002. “Intelligence without representation: the relevance of phenomenology to scientific explanation”. Phenomenology and the Cognitive Sciences, 1: 367–383. Dreyfus, H. 2005. “Overcoming the myth of the mental: how philosophers can profit from the phenomenology of everyday experience”. Proceedings and Addresses of the American Philosophical Association, 79: 43–49. Evans, G. 1981. “Semantic theory and tacit knowledge”. In S. H. Holtzmann and C. M. Leich, eds., Wittgenstein: to follow a rule. London: Routledge and Kegan Paul. Evans, G. 1982. The varieties of reference. New York: Oxford University Press. Glick, E. 2011. “Two methodologies for evaluating intellectualism”.  Philosophy and Phenomenological Research, 83: 398–434.

77

José Luis Bermúdez and Arnon Cahen Glick, E. 2012. “Abilities and know-how attributions”. In J. Brown and M. Gerken, eds., Knowledge ascriptions. Oxford: Oxford University Press: 120–139. Hornsby, J. 2011. “Ryle’s knowing how and knowing how to act”. In J. Bengson and M. A. Moffet, eds., Knowing how: essays on knowledge, mind, and action. New York: Oxford University Press. Jung, E. M., and Newen, A. 2010. “Knowledge and abilities: the need for a new understanding of knowing-how”. Phenomenology and the Cognitive Sciences, 9: 113–131. Levine, J. 2010. “Demonstrative thought”. Mind and Language, 25: 169–195. McDowell, J. 1994. Mind and world. Cambridge, MA: Harvard University Press. Noë, A. 2005. “Against intellectualism”. Analysis, 65: 278–290. Peacocke, C. 1989. “When is a grammar psychologically real?”. In A. George, ed., Reflections on Chomsky. Oxford: Blackwell. Peacocke, C. 2001. “Does perception have a nonconceptual content?”.  The Journal of Philosophy,  98: 239–264. Putnam, H. 1963. “Brains and behavior”. In R. J. Butler, ed., Analytical philosophy. 2nd Series. Oxford: Blackwell. Quine, W. V. 1972. “Methodological reflections on current linguistic theory”. In D. Davidson and G. Harman, eds., Semantics of natural language. Dordrecht: Springer: 442–454. Roskies, A. L. 2010. “ ‘That’ response doesn’t work: against a demonstrative defense of conceptualism”. Noûs, 44: 112–134. Ryle, G. 1946. “Knowing how and knowing that: the presidential address”. Proceedings of the Aristotelian Society, 46: 1–16. Ryle, G. 1949. The concept of mind. Chicago: The University of Chicago Press. Saffran, J. R., Newport, E. L., and Aslin, R. N. 1996. “Word segmentation: the role of distributional cues”. Journal of memory and language, 35: 606–621. Schiffer, S. 2002. “Amazing knowledge”. The Journal of Philosophy, 99: 200–202. Snowdon, P. 2004. “Knowing how and knowing that: a distinction reconsidered”. Proceedings of The Aristotelian Society, 104: 1–29. Stanley, J. 2011a. Know how. Oxford: Oxford University Press. Stanley, J. 2011b. “Knowing (how)”. Noûs, 45: 207–238. Stanley, J., and Williamson, T. 2001. “Knowing how”. The Journal of Philosophy, 98: 411–444. Toribio, J. 2015. “Opacity, know-how states, and their content”. Disputatio, 7: 61–83. Wiggins, D. 2012. “Practical knowledge: knowing how to and knowing that”. Mind, 121: 97–130.

78

5 THE FRAGMENTED MIND Personal and Subpersonal Approaches to Implicit Mental States Zoe Drayson

1.  Implicit Mental States 1.1  Attributions of Implicit Mental States If I ask you what is on your mind, you might answer by telling me that you have noticed a cat on your porch, that you intend to find its owner, and that you want to keep the cat if nobody claims it. Philosophers tend to classify such mental states as explicit, meaning that the person can articulate the intentional contents of their thoughts by means of a sentence, given suitable prompting (see, for example Dummett 1991; Davies 2001). Many of our everyday mental state attributions are of explicit mental states: the person can articulate the ascribed contents when prompted. In some situations, however, we attribute a mental state to a person who cannot suitably articulate the content in question. These are attributions of implicit mental states. Let us consider some examples of situations that might motivate us to attribute implicit mental states to people: A. Skilled behavior. When a person exhibits certain skilled behavior, such as riding a bike or playing the piano, we often say that the person knows how to ride a bike or knows how to play the piano. But the person is often unable to articulate the content of this practical knowledge: they cannot describe what it is they know. If these ascriptions of practical knowledge are genuine attributions of intentional mental states, then they must therefore be attributions of implicit mental states. B. Closure of belief under logical consequence. If beliefs are closed under logical consequence, as models of epistemic logic suggest, then a person believes all the logical consequences of their explicit beliefs. Some of these logical consequences will be the contents of further explicit beliefs: someone who articulates the conjunctive belief that Paris is the capital of France and home to the Louvre gallery will generally also articulate a belief in the conjunct that Paris is home to the Louvre gallery. But some of the logical consequences of our explicit beliefs are not articulable due to real-world constraints on our time, deductive power, and working memory. If we are to be attributed beliefs in these non-articulable contents, such attributions must be of implicit mental states.1 C. Behavior/testimony mismatch. The beliefs we attribute to someone to make sense of what they say are often the same beliefs that would make sense of their behavior: the testimonial DOI: 10.4324/9781003014584-7 79

Zoe Drayson

and predictive-explanatory roles of belief-attribution usually coincide (Schwitzgebel 2021). But there are cases where a person who sincerely voices an opinion also behaves in a way that suggests they endorse a contradictory opinion. Someone might act in a way that would be best explained by attributing a racist belief to them, for example, while articulating and defending a contradictory anti-racist belief. If we are to explain the person’s behavior in terms of their mental states, we might be motivated to attribute the racist belief to them as an implicit mental state. (See Gendler 2008 for more examples and discussions of beliefdiscordant behavior.) In addition to these everyday cases, some pathological cases seem to motivate explanation in terms of implicit mental states: the performance of subjects with blindsight or visual form agnosia seem to be best explained by attributing information to them that is at odds with the beliefs that they articulate. D. Visual perception. Both philosophical and scientific studies of perception have long suggested that people possess more visual information about the world than they can articulate. Evans (1982), for example, proposes that we can perceptually experience more shades of color than we can verbally describe or classify. And in vision science, it is standard practise to attribute the perceiver with information they cannot articulate (e.g. about complex probabilities or retinal disparity) to account for their perceptual capacities (e.g. to discriminate objects from their backgrounds or to make judgments of depth). Both cases involve attributing an implicit mental state to a person: an intentional content that the person cannot articulate.

1.2  Challenges for Implicit Mental State Attribution While it is commonplace to make attributions of implicit mental states in the preceding contexts, this practise faces several philosophical challenges. The very concept of an implicit mental state seems to challenge certain long-held assumptions about the nature of the mind. I’ll consider three such assumptions here: rational conditions on mentality, constraints on concept possession, and the special access that thinkers seem to have their own thoughts. Rationality: Philosophers often emphasize the connection between mentality and rationality. Davidson (1980) famously proposes that attributions of propositional mental states only make sense against a background assumption of rational relations between thought and action: someone who believes that p rationally ought to behave as if p were true, and thus assert that p under the appropriate conditions.2 If such rational conditions are a prerequisite for attributing intentional mental states, then attributions of implicit mental states are never appropriate. Concepts: Assuming we can rationally attribute implicit mental states to people, there is still the question of which implicit mental states to attribute. It is widely held that there is a conceptual constraint on mental state attributions: when we specify the content of a thought, we should only employ concepts possessed by the thinker (Bermúdez and Cahen 2020). And many philosophers (e.g. Evans 1982; Peacocke 1992) propose that concept possession itself faces a further generality constraint: thoughts systematically connect to each other in virtue of their constituent concepts, and so the possessor of a concept must be able to utilize the concept in a variety of different thoughts.3 Attributions of implicit mental states sometimes seem to violate these constraints. A vision scientist might explain a child’s ability to perceive depth in terms of the information possessed about retinal disparity, even where the child is unable to employ the concept of retinal disparity in their thought more generally. Privileged access: We often seem to have a certain special kind of epistemic access to our own mental states. A  strong version of the privileged access claim, on which thinkers are omniscient or infallible with respect to the contents of their mental states, would 80

The Fragmented Mind

presumably rule out implicit mental states completely.4 Weaker forms of privileged access, according to which we merely have some sort of fallible access to certain of our mental states, might be consistent with attributions of implicit mental states, but these raise further questions about why some mental states are introspectable while others are not. One way to respond to these three challenges would be to refrain from attributing implicit mental states. This approach would result in some cases of seemingly rational action being redescribed as reflex-like behavior, and a loss of the ability to distinguish intellectual capacities from bodily abilities. I will set aside such an approach here to focus instead on how philosophers have attempted to retain attributions of implicit mental states in a way that can be reconciled with the challenges outlined earlier. I will propose that there are two distinct ways to attribute implicit mental states which employ different strategies for addressing the challenges of rationality, concept possession, and privileged access.

1.3  Personal and Subpersonal Attributions of Implicit Mental States Traditionally, mental state attributions ascribe intentional content to the person or thinker as a whole. When we say that a person represents that p, calculates that q, or predicts that r, for example, we are describing the person as grasping a propositional thought. These are personal-level attributions of mental states. Since the birth of computational cognitive psychology, however, it has become common to make attributions of intentional content below the level of the person. When we describe a neural structure as representing that p, the visual system as calculating that q, or a Bayesian network as predicting that r, for example, we are attributing an intentional content to some proper part of the thinker, such a functional subsystem or a representational vehicle. These are subpersonal-level attributions of mental states. The distinction here between personal-level and subpersonal-level approaches is first and foremost a distinction between two ways of theorizing about the mind, which need not be understood as competing theories. I will consider questions about the semantic, epistemic, and ontological interpretation of these theories, and the relationship between personal-level and subpersonal-level theories in Section 4. (In this chapter I will be focusing on personal-level and subpersonal-level approaches to attributing implicit mental states, but both approaches can arguably also be used to attribute explicit mental states.5) To get an idea of how personal-level and subpersonal-level approaches can be applied to attributions of implicit mental states, consider a case of skilled behavior such as my ability to play ‘Moon River’ on the piano. Both approaches make attributions of intentional content, but in a different way. For an example of a personal-level approach, consider how Stanley and Williamson (2001) account for skilled behavior in terms of the person standing in a knowledge relation to a propositional content. In this case, I know that w is the way to play ‘Moon River’ on the piano, where w is a proposition that I cannot articulate. For an example of a subpersonal-level approach, on the other hand, consider how Fodor (1968) accounts for skilled behavior as the competence of an information-processing system. He compares the person’s ability (e.g. to play the piano) with a computer’s ability to calculate: intentional contents in the form of “propositions, maxims, or instructions” (Fodor 1968: 638) are attributed to proper parts of the informationprocessing system (vehicles of representation) rather than to the person as a whole. In what follows, I’ll explore the personal-level and subpersonal-level approaches to implicit mental state attribution in more detail and consider further examples. I will show that personallevel and subpersonal-level approaches have very different strategies for addressing the challenges faced by implicit mental state attribution concerning rationality, concept possession, and privileged access. 81

Zoe Drayson

2.  Personal-Level Approaches to Implicit Mental States Personal-level attributions of implicit mental states describe a person as grasping an intentional mental content, and yet not being in a position to assert it. How does the personal-level theorist address the apparent irrationality of such an attribution? Consider, first, the piano-playing example outlined earlier. A personal-level theorist might appeal to different ways of grasping a content, some of which might be rationally compatible with being unable to articulate the content. For Stanley and Williamson (2001), for example, knowing how to φ involves a proposition being presented to the person in a practical rather than theoretical way: it is a case of knowing that w is a contextually relevant way to φ under a practical mode of presentation. This allows the personal-level theorist to reconcile constraints on mental state attribution with the person’s inability to linguistically articulate the proposition involved. This approach, however, doesn’t seem to easily extend to other cases where we want to attribute implicit mental states. What if the implicit beliefs we attribute to the person directly contradict their explicit beliefs? Consider Lewis’s (1982) example of someone who explicitly believes that Nassau Street runs roughly east-west; explicitly believes that the railroad nearby runs roughly north-south; and explicitly believes that the two are roughly parallel. Any two of these propositions entail the negation of the third, so closure under logical consequence requires that we attribute a contradictory (presumably implicit) belief to the person. The fact that the contradictory belief is attributed as an implicit mental state does not make the person any more rational: attributions of contradictory belief violate even the most basic constraints on rationality, leaving no justification for ascribing any intentional mental states at all.6 One way a personal-level theorist might address this problem is to relativize belief attributions to temporal stages (time slices) of the person. Lewis himself takes this approach, suggesting that “the blatantly inconsistent conjunction of the three sentences [. . .] was not true according to my beliefs”: My system of beliefs was broken into (overlapping) fragments. Different fragments came into action in different situations, and the whole system of beliefs never manifested itself all at once. (Lewis 1982: 436) Lewis’s solution is to propose that we understand minds as temporally fragmented: a person’s grasp of a thought is always relativized to a particular time. If the three beliefs are not simultaneously attributable to the person, then the conditions for logical consequence are never met and we do not need to attribute the implicit (contradictory) belief in the first place. Lewis’s strategy of temporally relativized belief attributions might also be applied to behavior/testimony mismatch cases: if we deny any overlap between the person-stage who (explicitly) believes that p and the person stage who (implicitly) believes that not-p, there is no person-stage to whom we must attribute the belief that p and not-p. In some examples of implicit mental state attribution, where we are motivated to attribute the belief that p and the belief that not-p to a person simultaneously, the temporal fragmentation story will not help the personal-level theorist. Consider a visual perception case, for example. To explain why someone is subject to an optical illusion, we might say that the person implicitly believes that light is coming from above while they simultaneously articulate the belief that light is coming from below. In order to preserve the rationality constraint on mental state attributions, the personal-level theorist might relativize such attributions to tasks, contexts, or purposes, as well as times. Varieties of this approach in the literature include Egan’s (2008) suggestion that different beliefs drive different aspects of our behavior in different contexts; and Elga and Rayo’s 82

The Fragmented Mind

(2021) proposal that explaining and predicting behavior requires specifying what information is available to an agent relative to various purposes. In the visual perception case, the personallevel theorist could argue that the person believes-for-a-vision-guiding-purpose that light is coming from above and believes-for-a-more-general-purpose that light is coming from below, where only the latter purpose allows articulation of the belief content.7 Even if we allow that these personal-level approaches to implicit mental state attribution preserve the rationality constraint on mental state attribution, there is still the question of which content to attribute. Some approaches to content interpretation (e.g. Davidson 1982) seem to require the sort of rational holism that is denied by relativizing mental state attributions to temporal or contextual fragments.8 If a personal-level theorist argues that we can relativize concept possession and privileged access to times, contexts, or purposes, this seems to violate the generality constraint: Evans (1982) concludes that someone who can’t see the connection between thoughts involving the same concepts cannot really be said to grasp either of the thoughts in question. Stanley (2011), however, proposes that at least in the skilled behavior case, the inability to linguistically articulate the content of one’s mental state is compatible with being able to conceptualize, introspect, and perhaps even assert the content in a demonstrative form: for example, “I know that this is the way to play Moon River on the piano”.

3.  Subpersonal-Level Approaches to Implicit Mental States Subpersonal-level theories start from the assumption that minds can be understood as information-processing systems, and that the capacities of an information-processing system can be functionally decomposed into informational subsystems and discrete computational states: Sub-personal theories proceed by analyzing a person into an organization of subsystems [. . .] and attempting to explain the behaviour of the whole person as the outcome of the interaction of these subsystems. (Dennett 1978: 154) What makes these subpersonal-level psychological theories, rather than non-psychological descriptions of physical mechanisms, is the intentional interpretation of the subsystems. Subpersonal-level theories describe proper parts of the system (not just whole persons) as representing, evaluating, calculating, expecting, discriminating, and so on. (I’ll address the standard worries about this practise shortly.) Such theories end up attributing contents to subpersonal vehicles of representation: symbols, clusters in state spaces, or attractor basins, depending on the sorts of computational architectures we are dealing with. Importantly, these subpersonal vehicles of representation also have non-semantic properties in virtue of which they can physically implement computational transitions and be assigned contents in a naturalistic way.9 This means that cognitive psychology has at least one way to theorize about representational vehicles that does not characterize them in normatively constrained terms. Considered non-semantically, there is no expectation that these vehicles meet semantic constraints on rationality or concept-possession.10 There is no assumption that subpersonal vehicles are introspectable, and whether we have privileged access to their contents will depend on the nature of the information-processing architecture. The challenges faced by implicit mental state attribution, discussion in Section  1.2, are largely semantic or normative. Where personal-level approaches attempt to show how implicit mental state attributions are compatible with constraints on rationality and concept possession, subpersonal-level approaches suggest that these challenges do not arise in the first place. To see how this works, let us return to the piano-playing example. Cognitive psychology characterizes 83

Zoe Drayson

people’s skilled abilities partly in terms of information-processing performed by their motorcontrol subsystem. Subpersonal-level theories attribute contents (concerning the aims of our movements and how to achieve them, for example) to motor representations, where these representations can be characterized in terms of their non-semantic vehicle properties and thus individuated without appeal to rational norms. If we adopt a naturalistic theory of content determination, we can describe the motor representations as carrying information about the biomechanical constraints and kinematic rules relevant to piano-playing: information about the mathematical relationship between the velocity and amplitude of movement, for example, or laws relating curvature and velocity (see Mylopoulos and Pacherie (2017) for further discussion). Such attributions of content are compatible with the person being unable to reason more generally about these physical and mathematical constraints, and lacking first-person privileged access to these contents. We might account for the lack of articulability of the content in question in terms of the computational architecture of the motor subsystem: perhaps it works independently from other cognitive subsystems that we can introspect, or perhaps motor representations are carried in a different format to those representations that we can articulate. A similar approach can be applied to some of the other examples of implicit mental state attributions outlined in Section 1.1. In the perception case, subpersonal-level theorists can attribute assumptions about retinal disparity to the visual system rather than to the person. One might explain the non-introspectability of early visual processing by positing that low-level perceptual processes use a different representational format from other cognition (e.g. iconic versus discursive), or that they use a different kind of computational processing from other cognition (e.g. connectionist versus classical).11 Notice that subpersonal-level theories tend to be engaged in an explanatory psychological project rather than a justificatory epistemic project: subpersonal theories of visual perception, for example, are not trying to address skeptical worries, but to explain how our perceptual mechanisms operate. Similarly, since subpersonal-level theorists are generally interested in describing our inferential thought processes rather than justifying them, the normative epistemic problems raised by closure under logical consequence are ones they can set aside. Subpersonal-level approaches have faced challenges of their own, however. When cognitive psychologists first started talking about computational states as ‘representing’ states of affairs, some philosophers were skeptical: how can anything other than a person genuinely represent the world as being a certain way?12 There were two prominent criticisms of subpersonal-level approaches. First, does it even make sense to apply mental terminology below the level of the person, or does applying predicates true of the whole person to one of its proper parts commit a ‘mereological fallacy’? And second, even if mental terminology could be applied below the level of the person, wouldn’t any attempt to explain contentful systems in terms of contentful parts lead to some sort of ‘homuncular regress’? But subpersonal-level approaches have been responsible for much of the success of cognitive science, acting as a bridge between traditional personal-level approaches to the mind and lower-level neurophysiological explanation. This has prompted many philosophers to rethink their criticisms and to reject or at least reconsider these challenges: it is not obviously wrong to apply a psychological predicate to a cognitive subsystem, and such attributions may eventually ‘bottom-out’ rather than generate regress worries.13

4.  The Relationship Between Personal-Level and Subpersonal-Level Approaches We have seen that there are two ways to attribute intentional mental content: personal-level approaches and subpersonal-level approaches. It is important to remember, however, that these are not mutually exclusive ways of theorizing about the mind. Many philosophers of cognitive 84

The Fragmented Mind

science propose that personal-level theories can be complemented, expanded upon, and even vindicated by subpersonal-level theories.14 To see how this might work with attributions of implicit mental states, consider the following examples, which aim to combine personal-level and subpersonal-theories theories. Skilled behavior: A personal-level theory of skilled behavior might be combined with a subpersonal-level theory of motor control from cognitive science. Pavese (2017), for example, proposes that by attributing intentional content to representational vehicles in the motor subsystem, we can give a more rigorous characterization of Stanley and Williamson’s (2001) notion of a practical mode of presentation. Testimony/behavior mismatch: The fragmentation approach to personal-level theorizing, which attributes beliefs to a person relative to context or task, might be further cashed out in terms of a subpersonal-level theory of cognitive architecture. Bendaña and Mandelbaum (2021), for example, account for personal-level fragmentation by attributing informational contents to distinct and functionally isolated data structures within the cognitive architecture, instead of assuming one single database for information storage. If these strategies work, then we do not have to choose between personal-level and subpersonallevel theorizing: there are situations in which we can attribute both personal-level and subpersonal-level intentional contents. Whether strategies like these work is a matter for further investigation. If our best subpersonal-level theories of motor representation attribute contents that the person does not conceptually possess and reject a role for contents in reference determination, then motor representations do not seem to be able to play the role attributed by Stanley and Williamson (2001) to practical modes of presentation.15 And what if the personal-level attributions of fragmented beliefs that account for the mismatch between testimony and behavior cross-cut the sorts of informational divisions proposed by our empirically well-founded subpersonal theories of cognitive architecture?16 In these less straightforward cases, combining personal-level and subpersonal-level theories will require making adjustments to one theory or the other. Philosophical opinions, I will suggest, differ widely on whether we should focus on adjusting our person-level theories or our subpersonal-level theories. Some philosophers propose that we ought to prioritize the personal-level approach to intentional content attributions. If the subpersonal level theory doesn’t support the personal level theory, they suggest that we should replace it with a different subpersonal theory, or perhaps even deny the need for subpersonal theorizing. According to this view, we can accept that a person represents that p in a fragmented way, without thinking this requires that some architectural ‘fragment’ of the person represents that p. Schwitzgebel appears to be endorsing this approach to implicit mental states when he suggests that “[w]e can accept disunity [personal-level fragmentation] without embracing the dubious architectural commitments of system [subpersonal-level architectural] fragmentation” (Schwitzgebel 2021: 368). Some philosophers take the opposite approach and argue that we should prioritize the subpersonal-level theory of intentional content attributions. If our best subpersonal-level theory is at odds with our personal-level theory, they propose that our personal-level theory is the one that ought to be reformulated or even rejected. Proponents of this approach to implicit mental states (e.g. Norby 2014) argue that if empirical psychology divides up psychological state types in a way unsuited to a personal-level fragmentation theories, then this is a problem for personal-level theory. In what follows, I will suggest that the two sides in this debate are motivated by very different methodological and metaphilosophical views. 85

Zoe Drayson

Philosophers who prioritize personal-level approaches often propose that such approaches rely on distinctive epistemic methods (e.g. rational reflection, introspection, intuition, conceptual analysis, transcendental reasoning) that they take to be more reliable than the scientific reasoning that results in our subpersonal-level theories. Some (e.g. McDowell 1994) go further, proposing that subpersonal theories are irrelevant to our metaphysical understanding of the mind because they take personal-level theories to be governed by normative principles of rationality that make them completely autonomous from subpersonal-level theories.17 On such views, realism about personal-level theories does not need to be supported or vindicated by subpersonal-level theories: semantic facts don’t need to be explained by non-semantic facts; abstracta don’t need to be explained in terms of concreta. Proponents of prioritizing the personallevel approach often seem to think that only personal-level theorizing can provide genuine metaphysical insight into the fundamental nature of the mind.18 Philosophers who prioritize subpersonal-level approaches, on the other hand, are often motivated by naturalistic concerns about some of the methods associated with personal-level theorizing. They tend to worry about the status of analyticity, the reliability of intuition, and the possibility of a priori knowledge, and instead favor applying scientific reasoning methods to philosophy. Such philosophers (e.g. Churchland 1986) propose that if our best science attributes subpersonal-level content to internal vehicles of information processing, we should take this more seriously than our intuitive folk-psychological frameworks, which attribute content to the person. Prioritizing the subpersonal-level theory in this way might lead us to reject realist interpretations of personal-level theorizing altogether, and even to think that subpersonal-level theorizing can give us a posteriori access to metaphysical truths about the mind.19 Not all philosophers of mind fall into one or other of these camps: we do not have to insist on a realist interpretation of either personal-level or subpersonal-level theories. Perhaps all our attributions of content, whether personal-level or subpersonal-level, are nothing more than heuristic tools. Or perhaps some of our theories are true in a deflationary sense that is not ontologically committing: we can accept them without believing in the entities they posit. (See Drayson (2022) for further discussion of the varieties of anti-realism in philosophy of mind.)

5. Conclusion Some cases of intelligent behavior motivate us to attribute intentional mental states to a person even where they are unable to articulate the intentional content in question. In such cases where we cannot readily attribute explicit mental states, we instead tend to make attributions of implicit mental states. But many of our traditional ideas about mental states – that mental state attributions can only be made against a background assumption of rationality, that the contents of mental states must be specifiable in concepts possessed by the person, and that the person has privileged access to their intentional mental contents – are difficult to reconcile with attributions of implicit mental states. How we respond to this challenge will depend upon whether we take a personal-level approach or a subpersonal-level approach (or both approaches) to attributions of implicit mental states. Personal-level approaches to implicit mental states ascribe intentional content to the person as a whole; while subpersonal-level approaches to implicit mental states ascribe intentional content to cognitive subsystems of the person. The two approaches have different ways of addressing the challenges associated with rationality, concept possession, and privileged access. Personal-level theories find ways to meet the conditions in question by relativizing mental state attributions to persons at a time or a context, for example, or invoking practical modes of 86

The Fragmented Mind

presentation. Subpersonal-level theories tend to relax or reject the conditions in question, engaging in a description of psychological mechanisms with more minimal normative constraints. It is tempting to think that the sort of rational fragmentation proposed by personal-level theorists must map on to the sort of informational fragmentation that we find in different models of subpersonal-level cognitive architecture. While this is one possibility, it is important to remember that the relationship between personal-level theorizing and subpersonal-level theorizing can be characterized in a variety of different ways, depending on one’s background methodological and metaphilosophical views.20

Related Topics Chapters 4, 6, 11, 16

Notes 1 Giordani (2015), for example, acknowledges that the simplest solution to problems of “logical omniscience” involves introducing a distinction between explicit and implicit belief and claiming that only the set of implicit beliefs is closed under logical consequence. 2 Davidson proposes that “if we are intelligibly to attribute attitudes and beliefs, or usefully to describe motions as behavior, then we are committed to finding, in the pattern of behavior, belief and desire, a large degree of rationality and consistency” (Davidson 1980: 237). See also Yalowitz’s claim that “something only counts as being a mind – and thus an appropriate object of psychological attributions – if it meets up to certain rational standards” (Yalowitz 2005). 3 Conceptual constraints are sometimes formulated as linguistic constraints: Davidson (1980) holds that propositional thought contents must be linguistically expressible by the thinker. See also Frege’s claim that we can only grasp the content of a thought when it “clothes itself in the material garment of a sentence” (Frege 1956: 292). 4 Examples of such strong versions of privileged access include claims of self-intimation, self-presentation, or luminosity: being in a mental state suffices for knowing that one is in that mental state. Foundationalist epistemologies often rely on such privileged access claims. 5 Fodor’s (1975) ‘Language of Thought’ hypothesis, for example, is a subpersonal-level theory of information-processing systems on which computational states can correspond to either explicit or implicit mental states. See also Stich’s (1978) claim that explicit beliefs can be understood in terms of cognitive subsystems. 6 Davidson argues that “[n]othing a person could say or do would count as good enough grounds for the attribution of a straightforwardly and obviously contradictory belief ” (Davidson 1985: 138). If propositions are sets of possible worlds, contradictory beliefs would require impossible worlds. 7 For another approach to personal-level theorizing about implicit mental states, not discussed here, see Schwitzgebel’s (2001) account of “in-between believing”. 8 See also Gozzano (1999) for the idea that such mental fragmentation strategies avoid irrationality only by introducing more complex problems. 9 For more on the nature of representational vehicles and the importance of the distinction between their semantic and non-semantic properties, see Drayson (2018). 10 Of course, there may still be syntactic constraints on cognitive subsystems: we might understand the physical system (e.g. the brain) as implementing a formal language. 11 For examples of some of the different computational ways to account for non-introspectable psychological states, see Fodor (1983) on informationally encapsulated modules, Frankish (2010) on type-1 and type-2 processes, and Hohwy (2013) on statistical boundaries in hierarchical Bayesian architectures. 12 See Stich’s acknowledgment, for example, that many philosophers including himself were “skeptical about the idea of invoking internal representations in psychological theories” in the early days of subpersonal-level psychology (Stich 2011: xix). 13 For more on the mereological and homunculus fallacies and their proposed resolution, see Drayson (2012, 2014, 2017). 14 Notable examples include Fodor’s suggestion that “having a particular propositional attitude is being in some computational relation to an internal representation” (Fodor 1975: 198) and Lycan’s proposal

87

Zoe Drayson that the posits of personal-level theories can be identified with “the property of having such-and-such an institutionally characterized state of affairs obtaining in one (or more) of one’s appropriate homunctional departments or subagencies” (Lycan 1988: 41). Davies summarizes such approaches as follows: “we assume that, if a person consciously or occurrently thinks that p, then there is a state that has the representational content that p and is of a type that can figure in subpersonal-level psychological structures and processes” (Davies 2005: 370). 15 For a development of this argument along these lines, see Schwartz and Drayson (2019). 16 Norby (2014) proposes that the sorts of fragmented belief attributions we make in some personal level theories do indeed cut across the division drawn by empirical psychology. Schwitzgebel (2021) concurs that while certain kinds of personal-level fragmentation stories may find empirical support, there are other cases of implicit mental state attribution which don’t match up neatly with the empirically wellfounded varieties of subpersonal-level explanation. 17 This is what Rupert (2018) terms the “Received View” on which personal level facts are known nonscientifically by a priori reasoning, conceptual analysis, and introspection, while cognitive science has a more modest role studying the mere implementation of these facts. 18 In the perception literature, for example, Logue follows McDowell in stating that the fundamental metaphysical structure is “that which provides the ultimate personal-level psychological explanation of the phenomenal, epistemological and behavioural facts” (Logue 2012: 212). Logue contrasts personallevel theories such with scientific theories that appeal to “subpersonal psychological facts (e.g. the perceptual processing in the brain that takes place between stimulation of the sensory organs and experience)” (Logue 2012: 212). For a counterargument, see Drayson (2021). 19 There are different arguments in this vicinity that can lead to eliminativist conclusions. Churchland (1986) claims that none of our neural properties (action potentials, spreading activation, spiking frequencies, etc.) has the appropriate syntactically structured features to vindicate folk-psychological theorizing; while Stich (1983) claims that even if we could make sense of syntactically structured neural states, we couldn’t individuate these structures in the same way that we individuate beliefs in our folk psychological discourse. Rupert (2018) suggests that the success of scientific explanations at the subpersonal level gives us reason not to posit anything essentially normative, rational, and person-level. 20 Earlier versions of this work benefited from discussions at the University of Nevada, Reno (2016) and the University of Arizona (2019). It was also presented at the following conferences: Implicit Attitudes, Explicit Attitudes and Join Action, Ruhr Universitat Bochum (2018); The Self and Its Realization, University of Colorado, Boulder (2018); and the annual meeting of the Southern Society for Philosophy and Psychology, San Antonio (2018). I am grateful to the graduate students of the Philosophy of Mind reading group at the University of California, Davis, for their feedback, and to the volume editor and an anonymous reviewer for helpful comments.

References Bendaña, J., and Mandelbaum, E. 2021. “The fragmentation of belief ”. In C. Borgoni, D. Kindermann, and A. Onofri, eds., The fragmented mind. Oxford: Oxford University Press: 78–107. Bermúdez, J., and Cahen, A. 2020. “Nonconceptual mental content”.  In E. Zalta, ed., The Stanford encyclopedia of philosophy.  Summer 2020 edn. https://plato.stanford.edu/archives/sum2020/entries/ content-nonconceptual/. Churchland, P. 1986. Neurophilosophy: toward a unified science of the mind-brain. Cambridge, MA: MIT Press. Davidson, D. 1980. Essays on actions and events. Oxford: Oxford University Press. Davidson, D. 1982. “Paradoxes of irrationality”. In R. Wollheim and J. Hopkins, eds., Philosophical essays on Freud. Cambridge: Cambridge University Press: 289–305. Davidson, D. 1985. “Deception and division”. In E. Lepore and B. McLaughlin, eds., Actions and events: perspectives on the philosophy of Donald Davidson. Oxford: Blackwell: 138–148. Davies, M. 2001. “Knowledge (explicit and implicit): philosophical aspects”. In Neil J. Smelser and Paul B. Baltes, eds., International Encyclopedia of the Social and Behavioral Sciences. Elsevier: 8126–8132. Davies, M. 2005. “Cognitive science”. In F. Jackson and M. Smith, eds., The Oxford handbook of contemporary philosophy. New York: Oxford University Press. Dennett, D. 1978. Brainstorms. Cambridge, MA: MIT Press. Drayson, Z. 2012. “The uses and abuses of the personal/subpersonal distinction”. Philosophical Perspectives, 26: 1–18. Drayson, Z. 2014. “The personal/subpersonal distinction”. Philosophy Compass, 9: 338–346.

88

The Fragmented Mind Drayson, Z. 2017. “Psychology, personal and subpersonal”. In The Routledge encyclopedia of philosophy. London: Taylor and Francis. www.rep.routledge.com/articles/thematic/psychology-personal-andsubpersonal/v-1. https://doi.org/10.4324/9780415249126-V044-1 Drayson, Z. 2018. “The realizers and vehicles of mental representation”. Studies in History and Philosophy of Science, Part A, 68: 80–87. Drayson, Z. 2021. “Naturalism and the metaphysics of perception”. In H. Logue and L. Richardson, eds., Purpose and procedure in philosophy of perception. Oxford: Oxford University Press: 215–233. Drayson, Z. 2022. “What we talk about when we talk about mental states”. In T. Demeter, T. Parent and A. Toon, eds., Mental fictionalism: philosophical explorations. London: Routledge: 147–159. Dummett, M. 1991. The logical basis of metaphysics. Cambridge, MA: Harvard University Press. Egan, A. 2008. “Seeing and believing: perception, belief formation and the divided mind”. Philosophical Studies, 140: 47–63. Elga, A., and Rayo, A. 2021. “Fragmentation and information access”. In C. Borgoni, D. Kindermann, and A. Onofri, eds., The fragmented mind. Oxford: Oxford University Press: 37–53. Evans, G. 1982. The varieties of reference. Oxford: Clarendon Press. Fodor, J. A. 1968. “The appeal to tacit knowledge in psychological explanation”. Journal of Philosophy, 65: 627–640. Fodor, J. A. 1975. Language of thought. Cambridge, MA: MIT Press. Fodor, J. A. 1983. The modularity of mind: an essay on faculty psychology. Cambridge, MA: MIT Press. Frege, G. 1956. “The thought: a logical inquiry”. Mind, 65: 289–311. Frankish, K. 2010. “Dual-process and dual-system theories of reasoning”. Philosophy Compass, 5: 914–926. Gendler, T. S. 2008. “Alief and belief ”. Journal of Philosophy, 105: 634–663. Giordani, A. 2015. “A suitable semantics for implicit and explicit belief ”. Logique et Analyse, 58: 395–415. Gozzano, S. 1999. “Davidson on rationality and irrationality”. In M. De Caro, ed., Interpretations and causes: new perspectives on Donald Davidson’s philosophy. Dordrecht: Synthese Library, Kluwer. Hohwy, J. 2013. The predictive mind. Oxford: Oxford University Press. Lewis, D. K. 1982. “Logic for equivocators”. Noûs, 16: 431–441. Logue, H. 2012. “Why naive realism?”. Proceedings of the Aristotelian Society, 112: 211–237. Lycan, W. G. 1988. Judgement and justification. Cambridge: Cambridge University Press. McDowell, J. 1994. “The content of perceptual experience”. Philosophical Quarterly, 44: 190–205. Mylopoulos, M., and Pacherie, E. 2017. “Intentions and motor representations: the interface challenge”. Review of Philosophy and Psychology, 8: 317–336. Norby, A. 2014. “Against fragmentation”. Thought: A Journal of Philosophy, 3: 30–38. Pavese, C. 2017. “A theory of practical meaning”. Philosophical Topics, 45: 65–96. Peacocke, C. 1992. A study of concepts. Cambridge, MA: MIT Press. Rupert, R. 2018. “The self in the age of cognitive science: decoupling the self from the personal level”. Philosophic Exchange, 47: 1–36. Schwartz, A., and Drayson, Z. 2019. “Intellectualism and the argument from cognitive science”. Philosophical Psychology, 32: 662–692. Schwitzgebel, E. 2001. “In-between believing”. Philosophical Quarterly, 51: 76–82. Schwitzgebel, E. 2021. “The pragmatic metaphysics of belief ”. In C. Borgoni, D. Kindermann and A. Onofri, eds., The fragmented mind. Oxford: Oxford University Press: 350–376. Stanley, J. 2011. Know how. Oxford: Oxford University Press. Stanley, J., and Williamson, T. 2001. “Knowing how”. Journal of Philosophy, 98: 411–444. Stich, S. P. 1978. “Beliefs and subdoxastic states”. Philosophy of Science, 45: 499–518. Stich, S. P. 1983. From folk psychology to cognitive science: the case against belief. Cambridge, MA: MIT Press. Stich, S. P. 2011. Collected papers, vol. 1: Mind and language, 1972–2010. Oxford: Oxford University Press. Yalowitz, S. 2005. “Anomalous monism”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Winter 2005 edn. https://plato.stanford.edu/archives/win2005/entries/anomalous-monism/

89

6 THE LEVELS METAPHOR AND THE IMPLICIT/EXPLICIT DISTINCTION Judith Carlisle

1. Introduction In cognitive science, implicit and explicit phenomena are often described as operating on distinct “levels” (e.g., see Moore et al. 2012; Vinter and Perruchet 1994; Wittenbrink et al. 1997). For example, Uta and Chris Frith write that, At the lower level there are fast, relatively inflexible routines that are largely automatic and implicit and may occur without awareness. At the higher level there are slow, flexible routines that are explicit and require the expenditure of mental effort. (Frith and Frith 2008: 504) What does it mean to say that explicit cognition, such as perceptual experience or explicit memory, is at a higher level than implicit cognition? What significance is attached to the fact that these cognitive process are assigned to different levels? What is conceptually allowed or prohibited by the fact that two items are at the same or at different levels? In asking these questions, we confront three complicating descriptive facts: First, the term “level” is used to describe many different sorts of relation, and we should be clear in each particular case which we are using (see Churchland and Sejnowski 1988; Craver 2007, 2014). Managers are at a higher level than employees, elephants are at a higher level than slugs, and neurons are at a higher level than ions. Sometimes to be higher-level is to be bigger than, or better at, or more of something. Sometimes higher-level things are served by, coordinate, and organize lower-level things. Sometimes higher-level things are downstream from lower-level things. Sometimes, lower-level things are very literally below or underneath higher-level things. In these different contexts, it means something different to be at a higher or lower level, and different consequences follow from this sorting. Second, the implicit/explicit distinction is often characterized with a loosely connected cluster of opposing properties, where different properties are emphasized in different contexts. Implicit cognition is paradigmatically unconscious (or subconscious), automatic, inflexible, merely associational, and fast. Explicit cognition, in contrast, is effortful or conscious, controlled, flexible, inferential/representational, and comparatively slow (see Ramsey, this volume).

90

DOI: 10.4324/9781003014584-8

The Levels Metaphor and the Implicit/Explicit Distinction

Third, some philosophers talk about a distinction between the personal and the sub-personal in precisely the same terms. It is commonly suggested that both distinctions – namely the implicit/explicit and the personal/subpersonal – involve separate “levels” of cognition (i.e., the personal and the subpersonal are said to be at different levels; see Dennett 1969; Ismael 2014; Lyons 2016; Shea 2013; Rowlands 2010). Sometimes the personal/subpersonal distinction and the implicit/explicit distinction are used almost interchangeably (Frankish 2009; Rupert 2018). For example, Frankish argues that “the distinction between subpersonal and personal reasoning is also associated with a distinction of mental states, similar to that between implicit and explicit” (Frankish 2009: 103). Given the variable use of both the implicit/explicit distinction and the levels metaphor (and their complex relationships to the personal/subpersonal), it is not clear what we mean when we say that explicit cognition is at a “higher level” than implicit cognition. This makes it difficult to assess whether use of the levels metaphor to describe the implicit/explicit distinction is theoretically useful – and whether it genuinely illuminates anything about the distinction. In the following sections, I analyze a variety of different notions of “levels” and attempt to apply those notions to scientific examples of the implicit/explicit distinction from different areas of cognitive science and neuroscience. This will help determine what the levels metaphor is doing in each context and whether it is appropriate. As I progress through the different notions of levels, I make use of a fairly inchoate notion of the implicit/explicit distinction, primarily relying on examples in which researchers use these terms for themselves. However, I conclude by briefly analyzing the presumed notion of implicit/explicit that I argue drives the use of the metaphor in these examples. Finally, my goal is not to dictate some single, appropriate use for the levels metaphor (nor to defend one particular understanding of the implicit/explicit distinction). Again, the term “level” is also used in so many ways (see e.g., Craver 2007: Ch. 5) that it is pointless and unhelpfully stipulative to insist on a single, unequivocal analysis. Different uses of the term “level” serve different purposes – marking size relationships, keeping track of organizational flow in an institution, identifying sociological boundaries in groups, and ordering items by their complexity, their importance, or their fundamentality. My goal is rather to recognize this multiplicity and generate a useful taxonomy of the beneficial (and not so beneficial) ways that the levels metaphor has been applied to the implicit/explicit distinction.

2.  What Are Levels? In this section, I  consider a number of common ways the levels metaphor is used, asking whether any of these apply to empirical examples of the implicit/explicit distinction. My argument within each subsection follows a general pattern: For each notion of level, we find empirical examples where it seems appropriate to describe the implicit and explicit phenomena as being at different levels. However, for each notion of levels, there are also empirical counterexamples, where it seems inappropriate to apply the metaphor of levels to the implicit/explicit divide. Thus, I argue that although there are localized contexts in which it is appropriate to apply each particular version of the metaphor, no version of the metaphor is accurate generally. In other words, there is no general sense in which implicit cognition is at a lower level than explicit cognition.1 In fact, I will ultimately suggest that the implicit/explicit distinction primarily tracks a distinction between states and processes that are available for conscious access and control and those that are not.

91

Judith Carlisle

2.1  Levels of Size, Levels of Science, and Levels of Theory Some common senses of “levels” clearly cannot be applied to the implicit/explicit distinction, and so are unhelpful starting places. Sometimes, for example, scientists use the language of levels to indicate differences in size (Wimsatt 1994; Churchland and Sejnowski 1992; Shepherd 1994: 5). But size is clearly not the relevant factor in this context: Explicit things are not generally bigger than implicit things. Indeed, it’s not clear the distinction has anything to do with space or location at all – the “level” an item is on seems to have no bearing on where it is, what shape it has, or how it is ordered in size with respect to the things at other levels. Likewise, the distinction is not intended to mark a boundary between fields of science, such as economics, psychology, physiology, cell biology, and molecular biology. In general, both implicit and explicit cognition are studied by the same researchers, in the same fields, using many of the same or similar research tools. Finally, implicit and explicit phenomena are not said to be at different levels in virtue of the fact that they are described by different theories – examples of both implicit and explicit cognition are included within single theories and both are often required for successful cognitive and behavioral explanations (Rupert 2018; Perugini 2005). These uses of the level metaphor are clearly inapt in this context. In the following sections, I consider a variety of alternative versions of the levels metaphor that more successfully fit with well-known empirical examples of the distinction. I argue, however, that this fit often turns out to be an accidental feature of the example and not something that generalizes to other examples of implicit or explicit cognition.

2.2  Levels of Composition (Including Levels of Mechanism/Organization) In their most basic form, levels of composition describe the simple part-whole relation of aggregation, in which the property of the whole is a literal sum of the properties of its components. However, in most cases, the properties of wholes are not simply the sums of the properties of their parts – most compositional relations are more complex than simple aggregation. Also relevant are, for example, the sizes, shapes, positions, orientations of the parts, the direction of action of the relevant forces on and among the parts, the location of bonds between parts, and the causal interactions between parts (i.e., who does what to whom at what time and with what intensity, etc.). The parts of many wholes combine in very particular ways, with very particular structures and interactions (for examples of such accounts, see Hardcastle 1996: 38; Churchland and Sejnowski 1992: 18; McClamrock 1995: 185; Wimsatt 1994: 7). In Craver’s levels of mechanisms, for example, “the property or activity at a higher level of mechanisms is the behavior of the mechanism as a whole (the explanandum phenomenon); the parts of the mechanism and their activities are at a lower level” (Craver 2007: 165); things at different levels are related compositionally in that lower-level things are organized into higher level things. Are implicit and explicit phenomena on different compositional levels? In the strictest sense, implicit phenomena are not mechanistic or physical “parts” of explicit phenomena. Levels of composition are often understood to be primarily spatial or anatomical – lower-level phenomena are parts of higher-level phenomena in the same way that wheels are parts of a car or nuclei are parts of cells. On this reading, implicit phenomena are within explicit phenomena. This clearly does not apply to the implicit/explicit distinction generally – implicit phenomena are not generally physiological parts of explicit phenomena, and both implicit and explicit phenomena are made of the same physiological parts (e.g., neurons). 92

The Levels Metaphor and the Implicit/Explicit Distinction

However, in some contexts, there is a sense in which implicit phenomena might be organized together in a way that gives rise to explicit phenomena. For example, according to some interpretations of Chomsky’s account of human language, human conversational competence is made up of functionally characterized, implicit “parts” (see Drayson 2014: 341–343; Lycan 1987: 41; Fodor 1975: 198). According to this picture, our explicit linguistic abilities are made possible by a set of implicit skills that, when organized together, jointly give rise to our linguistic competence – our explicit skills are, in a sense, made up of our implicit skills (Chomsky 1965). The judgment that a sentence is grammatical, for example, is an explicit, conscious act, often performed with some effort. And this explicit capacity, on Chomsky’s model, is constituted by a collection of implicit skills, which may be consciously unavailable. These might include implicitly coded aspects of grammar including, for example, the rules about the proper use of gerunds. Similarly, Marr’s computational theory characterizes vision as a construction process in which internal representations are assembled from basic components that detect subtle changes in shading and, from that, features such as edges and shapes (Marr 1982). Likewise, since Hubel and Wiesel’s work in the 1960s, it’s been common among physiologists to understand vision as a construction process that begins with patterns of light and darkness on the retina and proceeds, via dots, lines, and basic motions (Zeki 1993). The operation of this system as a whole is taken to explain, and to constitute, the explicit capacity of vision. In contexts such as these, when there is a relationship of constitution, or when implicit processes are the realizers of explicit processes, the levels of composition metaphor can be both meaningful and useful. Further, if we seek to understand how consciousness arises from unconscious matter, it will be crucial to separate hypotheses in which consciousness (and so explicit cognition) is a product of antecedent causes, and hypotheses in which the conscious process is constituted by the organization of implicit components, as Chomsky appeared to hold. Although this conception of levels applies to many examples of the explicit/implicit divide, levels of composition do not define the distinction. This is for the simple reason that many implicit and explicit processes cannot be understood compositionally (e.g., Perugini 2005). In learning new skills, for example, we are likely to draw on both explicit representations and implicit learning (see Mathews et  al. 1989). When learning to play a new video game, for example, I  will likely utilize both explicit conceptual representations (e.g., “When I  enter a room that is large and apparently empty, I am likely to get attacked”) and implicit learning (e.g., increased reaction times to familiar/recognizable stimuli). That is, the process of learning to play the new game, which is an apparently explicit process, likely involves both explicit and implicit components interacting with one another. For example, Mathews and colleagues conclude that “knowledge acquired implicitly and explicitly interacts positively in certain situations” (Mathews et al. 1989: 1098). Therefore, the explicit implicit divide is not marked, generally, by part-whole relations. In other words, upon learning that some individual state or process is implicit, we are not then warranted in concluding that it is necessarily “lower level” (in the compositional sense) than all other explicit processes. Such conclusions would be unwarranted given that they ignore the possibility that there may be some explicit states or processes that are not related compositionally, but instead interact with one another (i.e., at the same “compositional level”).

2.3  Levels of Control Levels of control have primarily been understood in terms of regulation: Higher levels of control direct, dominate, or regulate activities at lower levels (Craver 2007: 179–180). The court 93

Judith Carlisle

system is arranged this way, as are some places of employment and military organizations. In the cognitive context, however, the metaphor of levels of control is often applied to information systems and pathways. Are implicit and explicit phenomena on different levels of control? Again, the answer is yes, sometimes. Consider a family of cases in which implicit processes apparently dominate explicit processes: those involving attentional capture. When we hear a sufficiently loud noise, for example, we cannot control the shift in attention that occurs (Fabio and Caprì 2019; Theeuwes 1994, 2010). Here, we seem to have a case in which an implicit (and so, “low-level”) process controls or dominates whichever explicit attentional processes are active. Further, there is a sense in which our implicit perceptual processes “control” our conscious visual experience. In other words, our explicit perceptual experience is directly controlled by the implicit processing that gives rise to that experience – without this implicit processing, we would not have those perceptual experiences. We can, of course, explicitly control whether or not our eyes are open or closed, and thus whether or not we are perceiving. But this is a very limited sense of control. In most cases, once we open our eyes, we cannot control what or how we perceive the world around us. Visual illusions are the most famous example of this perceptual encapsulation – although we can cognitively recognize that we are experiencing an illusion and that reality is being warped by our perception, we cannot stop the distortion from happening through explicit perceptual control. We simply cannot control our perceptual experiences in this way, even though the experience itself is apparently explicit. These sorts of examples show that there are at least some features of our perceptual experience that are not controllable (e.g., because they are encapsulated). However, there are also examples where the domination goes the other direction, where the explicit controls the implicit. Inattentional blindness may be a case in which our implicit attentional capture processes are dominated by our explicit attentional processes (Simons and Chabris 1999; Chabris and Simons 2010). For example, when we are focused on an explicit attentional task, we may ignore fairly large changes in our environment, even though our retina successfully foveates over the changes themselves. In a famous example, people fail to notice a waving gorilla walking through a room of basketball players, if they are already engaged in the explicit task of counting the number of basketball tosses by a particular team in the video (Simons and Chabris 1999; Chabris and Simons 2010). In such cases, it seems apt to describe this as a case where the activity of the explicit processing (i.e., counting basketball tosses) is limiting the influence of whatever implicit attentional processes would usually be responsible for noticing the motion of the large gorilla. In the context of attentional control, then, we see regulation in both directions. The implicit may override the explicit in the presences of loud noises; the explicit may override the implicit when it comes to wandering gorillas.2 These attentional processes show that in some cases implicit processes dominate and capture attentional resources as a result, while in other cases, our explicit processes seem to dominate, and capture attentional resources for themselves. This may explain why some have attempted to understand the implicit/explicit distinction (and personal/subpersonal distinction) in terms of dual process theory, where System 1 processes are associated with a subpersonal and implicit level, and System 2 processes are associated with the personal and explicit level (e.g., Gyurak et  al. 2011; Frankish 2009; Mallon 2016: 149). In these cases, where either the implicit or explicit processes are capable of control and regulation, it seems improper to utilize the levels of control metaphor – it is not the case that implicit processes are generally controlled by explicit processes. It may be more accurate to say that both implicit and explicit processes are controlled by allocation rules, which determine which processes will “receive” the attentional resources they “need.”3 94

The Levels Metaphor and the Implicit/Explicit Distinction

However, there may be other contexts in which the idea of levels of control can usefully be applied to the relationship between implicit and explicit processes. This is easiest to see when we consider research on failures of control. There are a variety of cases in which our explicit processes display a failure to control, direct, dominate, or regulate our implicit processes. Indeed, the implicit bias literature emphasizes a key dissociation between our explicit and implicit attitudes – and suggests an apparent lack of explicit control over these implicit attitudes and the behaviors they produce (see e.g., Follenfant and Ric 2010; Frith and Frith 2008; McClelland et al. 1989). After learning about our own implicit bias, for example, we cannot simply disavow our implicit attitudes and thereby expect that our subsequent behavior will be driven only by our explicit attitudes. Some researchers have suggested that implicit processes may dominate especially in cases when we are tired, distracted, rushed, or otherwise cognitively depleted (e.g., Bartholow et al. 2006). Researchers often describe this as a process by which our usual explicit control mechanisms fail on account of inadequate resources. One study, for example, compared the response of fatigued participants and non-fatigued participants on a weapon identification task, where individuals are asked to identify objects as tools or weapons after being shown either a black face or a white face (Govorun and Payne 2006). Typically, results from this weapons identification task show that participants are more likely to misidentify tools as weapons when the face they had previously seen was black, compared to when the face was white (Correll et al. 2002; Greenwald et al. 2003; Lambert et al. 2003). Govorun and Payne found that fatigued participants showed less explicit control over their implicit attitudes than participants who were less fatigued (Govorun and Payne 2006). This suggests that when we are fatigued we are more likely to rely on our implicit processes and attitudes – and that our behavior is, in these cases, more influenced by implicit processes than explicit processes. The researchers conclude that fatigue “interferes with processes requiring intentional control and cognitive resources” (Govorun and Payne 2006: 128). These failures of control seem to allow for the application of the metaphor of levels to the implicit/explicit distinction. In these cases, researchers believe that the “normal state” is one in which the explicit processes control the implicit processes, and it is only when we are negatively affected by exhaustion, for example, that our explicit processes fail to control our implicit processes. In these contexts, where the “normal state” of the system is one in which explicit phenomena control implicit phenomena, it may be useful to describe this relationship (and failures of this relationship) in terms of levels of control.4 However, as with the other notions of level, it does not seem generally true that explicit processes control, direct, dominate, or regulate implicit processes, or vice versa – there are many cases in which the order is reversed: our implicit processes control or dominate our explicit processes (e.g., attentional capture).

2.4  Levels of Analysis (Including Marr’s Three Levels) Finally, levels of analysis as originally proposed by Marr, are conceptual orientations adopted in the effort to understand a complex system. Bechtel and Shagrir succinctly describe the main features of Marr’s levels: The computational perspective provides an understanding of how a mechanism functions in broader environments that determines the computations it needs to perform (and may fail to perform). The representation and algorithmic perspective offers an understanding of how information about the environment is encoded within the mechanism and what are the patterns of organization that enable the parts of the mechanism 95

Judith Carlisle

to produce the phenomenon. The implementation perspective yields an understanding of the neural details of the mechanism and how they constrain function and algorithms. (Bechtel and Shagrir 2015: 312) Each Marrian level is associated with distinctive research questions: At the computational level, we ask: What is the system trying to achieve? And why? At the algorithmic level, we ask: How does the system achieve that goal? At the implementational level, we ask: What physical structures make it possible for the system to achieve that goal? Are implicit and explicit phenomena on different levels of analysis? One way of answering this question is to ask whether we can give a “computational” analysis of both the implicit and the explicit – or, in other words, whether we can ask Marr’s computational question of both implicit and explicit phenomena. If so, this would suggest that Marr’s levels do not align with the implicit/explicit distinction. This same argumentative strategy was used by Churchland and Sejnowski to show that Marr’s levels of analysis do not align with compositional levels of organization in the nervous system (e.g., molecules, synapses, neurons, networks, layers, maps, and systems). For each of these levels of organization, they argue, we can ask “computational” questions: What does this organization do for the subject? What is its purpose/goal? (Churchland and Sejnowski 1992: 19–21). This suggests that Marr’s framework is abstract enough to apply to research questions throughout the sciences. The term “computational” in Marr’s account is very broad and can be applied to any complex system – brains are complex systems, but so are circuits, pathways, and even cells. That is, Marr’s levels do not pick out a distinctive class of computational or cognitive machines; they are instead ways of looking at a problem to be solved by a complex system and the means by which that problem is solved. It would be a mistake to suppose that Marr’s levels offer a theory of what it is to be a cognitive system in particular, or to assume that his levels might only be fruitfully applied to computational systems. This suggests that each of Marr’s questions might be usefully applied to a variety of different “systems within systems” including the circuits and pathways that underlie both implicit and explicit processes, and to a variety of noncognitive systems as well. Similarly, Marr’s three levels might be used to describe either implicit or explicit cognitive processes: We can ask all three of Marr’s questions (i.e., the computational, algorithmic, and implementational) about both implicit and explicit phenomena. We can ask, for example, about the purpose that implicit biases are meant to serve for the subject (computational), how the system manages to store such attitudes and apply them (algorithmic), and which neurons and parts of the brain are involved (implementational). Given that we can apply the computation-level questions fruitfully to both implicit and explicit phenomena, it is clear that Marr’s levels do not generally align with the implicit/explicit distinction. On the other hand, in some contexts, Marr’s levels have been associated with the compositional sense of levels (e.g., levels of organization or mechanism). In Section 3, I argued that the compositional interpretation of the levels metaphor is not going to be an apt way of describing the implicit/explicit in all (or even most) cases. However, I also allowed that some cases of explicit cognition will occur at a higher compositional level than implicit cognition – or that “levels of composition” could sometimes be fruitfully applied to the implicit/explicit distinction (e.g., to Chomsky’s internal grammar). Given that Marr’s levels are sometimes interpreted compositionally (and that this compositional notion of levels can sometimes be fruitfully applied to the implicit/explicit distinction) we might think that Marr’s levels of analysis could also be fruitfully applied to the implicit/explicit distinction. For example, Chomsky’s internal grammar and Marr’s early visual states could be interpreted in this way. We might say that at the computational level of Chomsky’s grammar, we see 96

The Levels Metaphor and the Implicit/Explicit Distinction

linguistic abilities, such as the ability to determine whether a sentence is grammatically correct or incorrect. At the algorithmic level we see the unconscious cognitive processes that give rise to such explicit linguistic abilities – the implicit systems or processes responsible for checking subject verb agreement, for example, that give rise to the explicit feeling that the sentence is grammatically correct or incorrect (Chomsky 1965). This sort of picture allows for a more successful pairing of Marr’s levels of analysis/explanation with the implicit/explicit distinction. However, it seems somewhat unlikely that all sets of implicit and explicit processes fit this model. Considering implicit bias, it is not likely the case that the explicit attitude (or explicit behavior) is made up of the implicit attitudes – instead, in many cases, it seems that both our explicit and implicit attitudes work together (or against one another) causally, and so at the same compositional level, in giving rise to behavior (Perugini 2005). Further, it seems that we can apply Churchland and Sejnowski’s strategy again, to the compositional interpretation of Chomsky’s grammar. In other words, it seems that we could ask Marr’s “computational” questions about the implicit cognitive processes that give rise to our explicit linguistic abilities (i.e., the phenomena described at the lower compositional level). Initially, I proposed a compositional picture in which the purpose of the system is to determine whether sentences are grammatical or not (i.e., the computational level). These skills are underwritten by a set of unconscious rules that comprise our internal grammar (i.e., the algorithmic level). But even this compositional picture can be reinterpreted and broken down into multiple complex systems. We can, for example, ask about the purpose or goal of the “lower-level” unconscious processes in much the same way we can ask about the purpose or goal of the explicit linguistic skills. In other words, we can attempt to understand these implicit grammar rules from the computational perspective. Such “computational” investigations into the implicit rules can provide novel answers: This implicit system’s purpose is to check subject-verb agreement. Given that we can apply computational analysis to the underlying implicit processes that were previously described at the algorithmic level, it seems that Marr’s levels of analysis are not aligned (in any general sense) with the implicit/explicit distinction.

3.  Conclusion: Levels of Consciousness? Thus far, I have considered levels of composition, control, and analysis, and I have attempted to map those relations onto the implicit/explicit distinction. This survey supports two conclusions about the use of the levels metaphor in conjunction with the implicit/explicit distinction. First, there is no singular sense of the levels metaphor that applies univocally across diverse examples of the implicit/explicit distinction. When we find ourselves tempted to describe implicit and explicit cognition in terms of levels, we should reflect on which notion of level we are drawing on and what work it is intended to do in that context. Further, if we intend to draw a more general conclusion about the “levels” of the implicit/explicit distinction, we must test these intuitions against a variety of empirical examples as I have done in this chapter. Second, the implicit/explicit distinction (when combined with the metaphor of levels) is often used in combination with the conscious/unconscious distinction: implicit phenomena are considered “lower-level” primarily in the sense that they are unconscious and relatively effortless. Of course, as we have seen, the implicit/explicit distinction can sometimes fruitfully be understood in terms of levels of composition, control, and analysis/explanation. However, these applications seem at best accidental associations with what is at the heart of the distinction in each case: conscious availability and conscious control. Consider, for example, Roediger’s classic discussion of implicit and explicit memory. He defines explicit memory in terms of “conscious attempts to retrieve studied events” as opposed 97

Judith Carlisle

to implicit memory, which influences behavior without the awareness of the subject (Roediger and Amir 2005: 121; see also Roediger and Geraci 2005 for more information about implicit memory tests). Roedeiger and Amir open their discussion of the matter with an exceptionally clear formulation: Implicit memory refers to the influence of past events in current behavior when people are not trying to retrieve the past events and when they are usually not even aware of the events’ influence. The contrast is with explicit memory, which refers to conscious attempts to retrieve memories of past events. On implicit memory tests there is no conscious effort to retrieve studied material, whereas on explicit memory tests instructions refer to a specific encoding event. (2005: 121) A similar pattern is evident in descriptions of implicit and explicit systems of emotion regulation: we define explicit emotion regulation as those processes that require conscious effort for initiation and demand some level of monitoring during implementation, and are associated with some level of insight and awareness. Implicit processes are believed to be evoked automatically by the stimulus itself and run to completion without monitoring and can happen without insight and awareness. (Gyurak et al. 2011: 2) Such examples suggest that the main contrast between the implicit/explicit distinction (in the context of the levels metaphor) primarily revolves around two key and defining features of the implicit/explicit distinction – explicit processes are effortful and conscious, while implicit processes are automatic and unavailable to the subject through conscious introspection. Let me briefly clarify this second point. Throughout the chapter, I argue that many of the examples of implicit cognition are considered implicit because they are automatic and unavailable to consciousness. There are, of course, other cases in which “implicit” is used to pick out features that are unrelated (or distantly related) to consciousness, such as intentionality (see Ramsey, this volume). However, when we feel tempted to describe “implicit” phenomena in terms of the levels metaphor, it seems that consciousness is often at the forefront of the ­discussion – implicit memory, perception, and grammar are all implicit and lower-level, primarily because we cannot access them through introspection, nor can we reliably control these states and processes through conscious effort.5 Our ability to flip the examples and identify cases in which the proposed interlevel order (composition, control, etc.) runs in the opposite direction of the implicit/explicit divide suggests that there must be some way of marking the implicit/explicit divide independently of those accidental level-relations. That is to say, it seems like the different notions of levels on the table (composition, control, analysis) are only locally apt, and only appear to be useful ways of making sense of the implicit/explicit divide because we’ve failed to consider a broad enough array of examples. That said, I’m inclined to think there must be some explanation as to why we’re so consistently drawn to characterizing the implicit/explicit distinction in terms of levels. I tentatively suggest that the levels metaphor is consistently trying to track the distinction between cognitive states that are consciously available and controlled versus those that are unconscious and not under volitional control. For Chomsky, implicit grammar is implicit because it is unconscious and because we cannot control its processing. Similarly, when early visual states are described as implicit, this is often because these processes are automatic (i.e., uncontrolled) and unavailable 98

The Levels Metaphor and the Implicit/Explicit Distinction

for conscious reflection. This may lead us to conclude that the relevant notion of levels is “levels of consciousness” – Freud, for example, has been associated with the “iceberg metaphor of mind,” where “only a small portion of the iceberg pokes out above the surface of the water, while the bulk of it is hidden below” (Green 2019: 369). Of course, this is itself a metaphor, and we would need to understand what “levels of consciousness” are – what does it mean to say that conscious phenomena are at a higher level than unconscious phenomena? Perhaps our intuition that explicit processes are at a higher level than implicit processes is just the same as our intuition that explicit processes are conscious and effortful while implicit processes are not. In other words, perhaps we have been using the term “higher-level” as a synonym for “conscious.” Conscious cognitive phenomena may seem higher-level for a variety of reasons: perhaps because they seem superior to unconscious cognitive phenomena, or because non-conscious beings (e.g., sea slugs) do not have them, or because such conscious cognitive phenomena require a certain level of organization in order to occur (Edelman and Tononi 2001). As we have seen, it is not clearly the case that explicit cognitive phenomena are univocally “higher” in terms of levels of composition, control, or analysis/explanation. If I am right in suggesting that our intuitions about consciousness have coopted the language of levels merely to express a difference in their introspective availability and the amount of conscious effort they require, then the metaphor of levels is doing nothing over and above the distinction between conscious/unconscious. Instead, we should simply discuss these processes as differing in their conscious availability or control, and leave the metaphor of levels aside.6

Related Topics Chapters 1, 4, 5

Notes 1 In this chapter, I primarily analyze levels of composition, control, and analysis – these are the three notions of level that I take to be most closely related to the implicit/explicit distinction. But, there are other notions of level that I do not analyze here, including levels of processing (Craver 2007; Churchland and Sejnowski 1992) and levels of complexity (Hardcastle 1996). However, my argument concerning levels of processing and levels of complexity follows the same pattern as stated earlier – for both, there are some contexts in which it is reasonable and useful to discuss implicit states and processes as on a different level of processing and complexity than explicit states and processes. However, there will also be counterexample cases, in which it will not be fruitful to describe implicit states and processes as on a different level of processing and complexity than explicit states. 2 For a helpful review attentional capture and inattentional blindness, see Simons (2000). 3 Clearly, this use of the levels metaphor is distinct from the use of levels of composition in Chomsky’s grammar. The two processes are not related by constitution, but are in competition with one another for attentional resources, and thus, numerically and causally distinct. 4 I am not arguing that this is the most accurate understanding of implicit bias – it could be that these processes are more like the attentional processes described earlier, where their relationship is not aptly described in terms of control. If explicit attitudes do not normally control implicit bias, it would be inappropriate to describe implicit bias in terms of levels of control. 5 Given that the personal/subpersonal distinction and the implicit/explicit distinction are so closely related, I find it telling that Frankish suggests that “in the case of mental processes at least, the distinction between personal and subpersonal corresponds roughly with that between conscious and nonconscious” (Frankish 2009: 91). 6 Many thanks to Carl Craver for his guidance and feedback on this chapter – and also to Rob Rupert, Gualtiero Piccinini, Ron Mallon, Caleb Pickard, and many others for their constructive comments along the way.

99

Judith Carlisle

References Bartholow, B., Dickter, C., and Sestir, M. 2006. “Stereotype activation and control of race bias: cognitive control of inhibition and its impairment by alcohol”. Journal of Personality and Social Psychology, 90: 272–287. Bechtel, W., and Shagrir, O. 2015. “The non-redundant contributions of Marr’s three levels of analysis for explaining information-processing mechanisms”. Topics in Cognitive Science, 7: 312–322. Chabris, C., and Simons, D. 2010. The invisible gorilla: and other ways our intuitions deceive us. New York: Harmony. Chomsky, N. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press. Churchland, P., and Sejnowski, T. 1988. “Perspectives on cognitive neuroscience”. Science, 242: 741–745. Churchland, P., and Sejnowski, T. 1992. The computational brain. Cambridge, MA: MIT Press. Correll, J., Park, B., Judd, C., and Wittenbrink, B. 2002. “The police officer’s dilemma: using ethnicity to disambiguate potentially threatening individuals”. Journal of Personality and Social Psychology, 83: 1314–1329. Craver, C. 2007. Explaining the brain. Oxford: Oxford University Press. Craver, C. 2014. “The ontic account of scientific explanation”. In M. Kaiser, O. Scholz, D. Plenge, and A. Hüttemann, eds., Explanation in the special sciences: the case of biology and history. Dordrecht: Springer Verlag: 27–52. Dennett, D. C. 1969. Content and consciousness. London: Routledge. Drayson, Z. 2014. “The personal/subpersonal distinction”. Philosophy Compass, 9: 338–346. Edelman, G., and Tononi, G. 2001. A universe of consciousness: how matter becomes imagination. New York: Basic Books. Fabio, R., and Caprì, T. 2019. “Automatic and controlled attentional capture by threatening stimuli”. Heliyon, 5: e01752. Fodor, J. 1975. The language of thought. Cambridge, MA: Harvard University Press. Follenfant, A., and Ric, F. 2010. “Behavioral rebound following stereotype suppression”. European Journal of Social Psychology, 40: 774–782. Frankish, K. 2009. “Systems and levels: dual-system theories and the personal-subpersonal distinction”. In J. Evans, and K. Frankish, eds., In two minds: dual processes and beyond. Oxford: Oxford University Press: 89–107. Frith, C., and Frith, U. 2008. “Implicit and explicit processes in social cognition”. Neuron, 60: 503–510. Govorun, O., and Payne, K. 2006. “Ego-depletion and prejudice: separating automatic and controlled components”. Social Cognition, 24: 111–136. Green, C. 2019. “Where did Freud’s iceberg metaphor of mind come from?”. History of Psychology, 22: 369–372. Greenwald, A., Oakes, M., and Hoffman, H. 2003. “Targets of discrimination: effects of race on responses to weapons holders”. Journal of Experimental Social Psychology, 39: 399–405. Gyurak, A., Gross, J., and Etkin, A. 2011. “Explicit and implicit emotion regulation: a dual-process framework”. Cognition & Emotion, 25: 400–412. Hardcastle, V. 1996. How to build a theory in cognitive science. Albany: SUNY Press. Ismael, J. 2014. “On being some-one”. In A. Mele, ed., Surrounding free will: philosophy, psychology, neuroscience. New York: Oxford University Press: 274–297. Lambert, A., Payne, K., Jacoby, L., Shaffer, L., Chasteen, A., and Khan, S. 2003. “Stereotypes as dominant responses: on the ‘social facilitation’ of prejudice in anticipated public contexts”. Journal of Personality and Social Psychology, 84: 277–295. Lycan, W. 1987. Consciousness. Cambridge, MA: MIT Press. Lyons, J. 2016. “Unconscious evidence”. Philosophical Issues, 26: 243–262. Mallon, R. 2016. “Stereotype threat and persons”. In M. Brownstein and J. Saul, eds., Implicit bias and philosophy, vol. 1: Metaphysics and epistemology. Oxford: Oxford University Press: 130–154. Marr, D. 1982. Vision: a computational investigation into the human representation and processing of visual information. San Francisco: Freeman. Mathews, R., Buss, R., Stanley, W., Blanchard-Fields, F., Cho, J., and Druhan, B. 1989. “Role of implicit and explicit processes in learning from examples: a synergistic effect”. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15: 1083–1100. McClamrock, R. 1995. Existential cognition: computational minds in the world. Chicago: University of Chicago Press.

100

The Levels Metaphor and the Implicit/Explicit Distinction McClelland, D., Koestner, R., and Weinberger, J. 1989. “How do self-attributed and implicit motives differ?”. Psychological Review, 96: 690–702. Moore, J., Middleton, D., Haggard, P., and Fletcher, P. 2012. “Exploring implicit and explicit aspects of sense of agency”. Consciousness and Cognition, 21: 1748–1753. Perugini, M. 2005. “Predictive models of implicit and explicit attitudes”. British Journal of Social Psychology, 44: 29–45. Roediger, H., and Amir, N. 2005. “Implicit memory tasks: retention without conscious recollection”. In A. Wenzel and D. Rubin, eds., Cognitive methods and their application to clinical research. Washington, DC: American Psychological Association: 121–127. Roediger, H., and Geraci, L. 2005. “Implicit memory tasks in cognitive research In A. Wenzel and D. Rubin, eds., Cognitive methods and their application to clinical research. Washington, DC: American Psychological Association: 129–151. Rowlands, M. 2010. The new science of the mind: from extended mind to embodied phenomenology. Cambridge, MA: MIT Press. Rupert, R. 2018. “The self in the age of cognitive science: decoupling the self from the personal level”. Philosophic Exchange, 47: 1–36. Shea, N. 2013. “Naturalising representational content”. Philosophy Compass, 8: 496–509. Shepherd, G. 1994. Neurobiology. 3rd edn. New York: Oxford University Press. Simons, D. 2000. “Attentional capture and inattentional blindness”. Trends in Cognitive Sciences, 4: 147–155. Simons, D., and Chabris, C. 1999. “Gorillas in our midst: sustained inattentional blindness for dynamic events”. Perception, 28: 1059–1074. Theeuwes, J. 1994. “Stimulus-driven capture and attentional set: selective search for color and visual abrupt onsets”. Journal of Experimental Psychology: Human Perception and Performance, 20: 799–806. Theeuwes, J. 2010. “Top – down and bottom – up control of visual selection”. Acta Psychologica, 135: 77–99 Vinter, A., and Perruchet, P. 1994. “Is There an implicit level of representation?”. Behavioral and Brain Sciences, 17: 730–731. Wimsatt, W. 1994. “The ontology of complex systems: levels of organization, perspectives, and causal thickets”. Canadian Journal of Philosophy: Supplementary Volume, 20: 207–274. Wittenbrink, B., Judd, C., and Park, B. 1997. “Evidence for racial prejudice at the implicit level and its relationship with questionnaire measures”. Journal of Personality and Social Psychology, 72: 262–274. Zeki, S. 1993. A vision of the brain. Oxford: Wiley-Blackwell.

101

PART 2

The Nature and Limits of Implicit Processing

7 IMPLICIT COGNITION, DUAL PROCESS THEORY, AND MORAL JUDGMENT Charlie Blunden, Paul Rehren, and Hanno Sauer

Introduction Implicit cognition is cognition that happens automatically and (typically) non-consciously. This chapter is about implicit moral cognition. In moral psychology, implicit moral cognition is almost always understood in terms of dual process models of moral judgment. In this overview, we will address the question whether implicit moral judgment is usefully cashed out in terms of automatic (“type 1”) processes, and what the limitations of this approach are. The distinction between type 1 and controlled (“type 2”) cognition is often not as clear-cut as it seems, and the two types of processing appear to be more deeply enmeshed than the popular dichotomy makes it sound. As a result, identifying implicit moral cognition with automatic, unconscious, and inflexible modes of thinking neglects that intuitive moral judgment is often more nuanced and complicated. Our chapter has six sections. In (1), we provide a brief overview of dual process models of domain-general (moral and non-moral) cognition. Section (2) reviews a recent debate regarding the soundness of dual process models to begin with. Section (3) is about dual process accounts of moral judgment specifically, and sections (4), (5), and (6) survey recent attempts to go beyond a simplistic type 1/type 2 distinction in conceptualizing implicit moral cognition. One of the core questions of this overview is whether type 2 processing can successfully penetrate implicit moral cognition. Recently, three main theoretical approaches to this effect have emerged. First, rational learning approaches highlight how episodes of moral learning can feed back into and shape our moral intuitions (section (4)). Second, so-called triple process models of mental processing suggest that automatic cognitive outputs can under certain circumstances be reined in by reflective cognition (section (5)). Finally, recent evidence shows that moral reasoning is not as impotent as earlier developments in empirical moral psychology claimed to demonstrate (section (6)).

1  Dual Process Theory How should we understand the nature of implicit cognition? Over the past four decades, an increasing body of research in cognitive science, philosophy, neuroscience, and psychology has accumulated suggesting that various aspects of human cognition can be fruitfully understood DOI: 10.4324/9781003014584-10 105

Charlie Blunden, Paul Rehren, and Hanno Sauer

using dual process models. This perspective has in particular been applied to higher cognition such as reasoning, decision-making, or social judgment (for reviews, Evans 2008; Kahneman 2011; Evans and Stanovich 2013; Gawronski and Creighton 2013). Dual process theories of higher cognition (DPT) divide cognitive processes into two types which we will here refer to as type 1 and type 2 (Evans and Stanovich 2013). Most DPTs converge on similar characterizations of type 1 and type 2 in terms of certain features. Type 1 is generally described as unconscious, implicit, domain-specific, automatic, fast, and intuitive, whereas type 2 is conscious, domain-general, controlled, deliberative, slow, and effortful (Evans 2008: 256; Kahneman 2011: 20–21; Evans and Stanovich 2013: 223). On this account, implicit cognition can be broadly identified with type 1 processing (Evans 2008). Proponents of DPT cite several different lines of evidence in support of their theory (for reviews see Evans 2006, 2008; Stanovich 2004, 2011; Kahneman 2011). There are many experiments in which subjects are given tasks for which there is a fast, intuitive (type 1) but incorrect answer which people are inclined to give, and a correct response which requires explicit application of (for example) logical or probabilistic reasoning (type 2). People are often unaware of how they come to their intuitive judgment, and so they often don’t detect their error unless prompted to engage in more effortful reasoning. These experiments show the two types of cognition identified by DPT in action.

2  Are Dual Process Theories Theoretically Coherent? If DPT is supposed to be a useful way of understanding implicit cognition, then it must at least be internally coherent. But while very popular, the theory has received considerable criticism on this front. One such objection is the alignment problem (Keren and Schul 2009; Melnikoff and Bargh 2018b): DPT characterizes its two processing types by two distinct sets of features. For example, recall that type 1 processes are often described as automatic, efficient, and non-conscious, while type 2 processes are controlled, inefficient, and conscious. This means that if DPT were true of cognition in general, most cognitive processes should either display type 1 or type 2 features. In contrast, mixtures of the two sets should be rare. In fact, however, such mixtures are observed in many domains of cognition (for examples, Melnikoff and Bargh 2018b). According to the critics, this constitutes decisive evidence against DPT. A second objection, which also focuses on these processing features, states that many of them lack internal consistency (Melnikoff and Bargh 2018b; Moors and De Houwer 2006). To illustrate the point, take controllability. Melnikoff and Bargh (2018b) suggest that processes can be controllable in at least two ways: first, they can be stopped once triggered; second, their output can be modified. If this is correct, then controllability is not a unitary feature, but encompasses multiple distinct types of controllability. The same, argue Melnikoff and Bargh, is true for many of the other processing features mentioned by DPT. But then, using these features to distinguish between type 1 and type 2 processing without specifying for each feature which of its sub-types is at issue makes little theoretical sense. Yet this, so the objection goes, is precisely what DPT does. Proponents of DPT (Pennycook et al. 2018) have responded that both objections are based on a misrepresentation of the theory. They point out that (modern) DPT draws a distinction between the defining features of a processing type, and its typical correlates. Defining features are at the core of DPT’s two-type distinction – they are what makes a process either type 1 or type 2. In contrast, typical correlates are features that, while associated with a given processing type, neither define it nor distinguish it from other processing types. Yet according to

106

Implicit Cognition, Dual Process Theory, and Moral Judgment

Pennycook et  al., both objections exclusively target typical correlates, not defining features. Therefore, both miss their mark. To this, critics have objected that if DPT is understood in this more minimal way, the theory becomes unfalsifiable. The thought here is that “[t]heories must generate predictions, and it is unclear how any predictions can be derived from a ‘defining feature’ that is not correlated with anything” (Melnikoff and Bargh 2018a: 668). This in turn has been challenged by dual process theorists (Pennycook 2017). DPT as a general theory of cognition has its limitations, but this does not imply that dual process models of specific phenomena cannot be useful or valid (cf. Gawronski and Creighton 2013). Rather, such models must be evaluated individually on their own merits. In the next section, we review DPTs of implicit moral cognition. We then spend the rest of the chapter evaluating them on their own merits.

3  Dual Process Theories of Moral Judgment For the past two decades, moral psychology has been dominated by two DPTs of moral cognition. Each paints a different picture of the relationship between type 1 and type 2 processes. The Social Intuitionist Model (SIM; Haidt 2001; Haidt and Björklund 2008; Haidt 2012) puts the focus squarely on type 1 processes. According to the SIM, moral judgments are usually the result of affectively charged intuitions which people feel almost instantly upon encountering morally salient situations. In contrast, moral reasoning (type 2 cognition) generally only kicks in once an intuitive moral judgment has already been formed, and then does little to influence this judgment. Instead, the reasons people provide in defense of their moral judgments are mostly post hoc rationalizations. Post hoc rationalization occurs when people offer factors in their justification of a belief or judgment which did not actually produce that belief or judgment (Haidt 2001: 822–823; Schwitzgebel and Ellis 2017). The factor which actually caused the judgment is often not made explicit because it is an automatic (type 1) process that is not introspectively accessible (Schwitzgebel and Cushman 2012: 149). There is ample evidence of post hoc rationalization in people’s moral cognition. For instance, Cushman et  al. (2006) have provided evidence that people’s responses to moral dilemmas are driven by factors (such as whether harm was intended rather than merely foreseen) that they are not able to articulate in their justifications: instead people offer other factors as explaining their judgments, often citing various (unwarranted) assumptions about the cases in question that would explain their pattern of judgment (Cushman et al. 2006: 1086; see also Hauser et al. 2007; Schwitzgebel and Cushman 2012). That people sometimes engage in post hoc rationalization is perhaps never more apparent than when it breaks down. In a famous but unpublished study, Haidt et al. (2000) asked participants to justify their moral judgments about actions in moral dilemmas. They found that the reasons participants provided often did not apply to the dilemma at hand. For example, participants sometimes mentioned that an action would result in harm, even though the dilemma explicitly stated that no harm would result. Once this was pointed out, participants often admitted that they could not provide good reasons in support of their judgment, yet at the same time did not change their initial moral judgment. Instead, they often ended up saying things like “I know it’s wrong, but I just can’t come up with a reason why” (Haidt et al. 2000: 11). Haidt et al. call these cases of moral dumbfounding and argue that such cases show that the reasons people give in support of their moral judgments or belief are often not the actual reasons for why they hold them. In moral dumbfounding cases, people’s judgments are being driven by intuitions that they

107

Charlie Blunden, Paul Rehren, and Hanno Sauer

do not have access to, and they are unable to construct rationalizations after the fact to justify their judgments (Haidt et al. 2000: 10–11). While much type 2 cognition about moral judgments may be post hoc rationalization, the SIM does allow for some corrective influence of type 2 processes on moral judgment. First, certain individuals are sometimes able to arrive at moral decisions purely by way of moral reasoning; more on this in sections (5) and (6). However, Haidt hypothesizes that this is very rare (Haidt 2001: 819). Second, in social settings, one person’s rationalization can sometimes change another person’s moral judgment, often by drawing their attention to aspects of an issue which trigger new intuitions (Haidt and Björklund 2008: 190–192). According to Greene’s (Greene et al. 2001; Greene 2013, 2014) Dual Process Model, type 1 and type 2 processing often produce conflicting moral judgments. Deontological judgments (judgments more easily supported by appeals to rights and duties) are supported by emotionally charged automatic intuitions (type 1). In contrast, consequentialist judgments (judgments more easily supported by cost-benefit considerations) are supported by conscious controlled reasoning (type 2). Most of the evidence for Greene’s model comes from studies of how subjects respond to sacrificial dilemmas (for a review, see Greene 2014). For example, in the trolley problem, an out-of-control trolley is speeding toward a group of five people. In Switch, the trolley can be diverted onto a separate track by hitting a switch, where it would kill one person instead. In Footbridge, the five people on the tracks could be saved by throwing a sixth person from a footbridge into the path of the trolley (Foot 2002; Thompson 1985). People tend to approve of hitting the switch but not of throwing the person from the footbridge (e.g., Hauser et al. 2007). Greene’s explanation is that the type 1 processes underlying deontological judgments act like an emotional alarm in cases that involve up-close physical harm (Footbridge). In contrast, they are less sensitive to harm as a side-effect (Switch). Therefore, in Footbridge, people’s deontological intuitions tend to dominate the consequentialist reasoning that leads them to approve of the sacrifice in Switch: Greene argues that deontological moral reasoning is merely post hoc rationalization of these emotionally charged intuitions (Greene 2007; Greene 2013: Ch. 9).

4  Beyond Dual Process I: Moral Judgment and Rational Learning Despite their differences, both accounts reviewed in the last section draw a sharp line between implicit moral cognition or moral intuition (type 1) and moral reasoning (type 2). This, in combination with the characterization of type 1 processing as fast, automatic, and (typically) non-conscious, has led several authors to raise doubts about the rationality of our implicit moral cognition. Moral intuitions have been characterized as blunt, inflexible, unresponsive to relevant evidence, and hence frequently unreliable (Greene 2014; Haidt 2012; Singer 2005). More recently, however, this strong identification of implicit moral cognition with type 1 processing has come under fire. Three theoretical approaches in particular are now gaining traction which suggest that type 2 cognition influences and shapes our moral intuitions in various ways. In contrast to standard DPTs of moral cognition, these approaches see implicit moral cognition as deeply intertwined with moral reasoning. First, rational learning approaches highlight how episodes of moral learning can feed back into and shape our moral intuitions. Second, so-called triple process models of higher cognition suggest that type 1 processing can sometimes actively be reined in by reflection. Third, recent evidence shows that type 2 processing is not as impotent as earlier developments in moral psychology claimed to demonstrate. We will take these up in turn, starting with rational learning. The central claim of rational learning approaches is that moral intuitions (along with most other cognitive processes) are shaped by sophisticated learning systems (Cushman et al. 2017). 108

Implicit Cognition, Dual Process Theory, and Moral Judgment

This shaping starts at the initial acquisition of an intuition, but can be ongoing throughout an individual’s life. Even though in the moment, they may be experienced as fast and automatic, moral intuitions are often the result of processes that operate flexibly, integrate information from a variety of sources, and are highly computationally complex (Railton 2017; Woodward and Allman 2007). Thus, rational learning approaches suggest that implicit moral cognition cannot straight-forwardly be identified with type 1 processes, because much of it is continually shaped by processes that display many of the hallmarks of explicit moral reasoning (cf. Railton 2017: 180). One influential rational learning account (Crockett 2013; Cushman 2013) draws on the distinction (originally from computer science) between model-free and model-based reinforcement learning. In model-free learning, actions are evaluated based on their previously learned value – that is, based on whether and to what extent they have been rewarded in the past. In contrast, model-based learning works by building up a causal model of the world inhabited by the agent. Actions are then evaluated by choosing the option that maximizes expected value. Crockett (2013) and Cushman (2013) suggest that classical dual process models of moral cognition should be understood in terms of this distinction. More specifically, on their view, moral intuitions are the output of model-free learning and decision-making. In contrast, model-based systems are associated with moral reasoning and deliberation. While Crockett (2013) and Cushman (2013) update but ultimately retain a dual process model of moral cognition, Railton (2017) goes further. He too makes use of the model-based/ model-free learning distinction. However, Railton disagrees that moral intuitions are exclusively the output of model-free learning systems. Instead, drawing on evidence from a number of disciplines and domains of cognition, he argues that model-based systems frequently play an important role in shaping our moral intuitions as well. Thus, Railton considerably blurs the line between moral intuition and moral reasoning. One set of studies which nicely illustrates the rational moral learning approach is Nichols et al. (2016). To explain the acquisition of deontological rules, Nichols et al. appeal to a Bayesian principle called the size principle. It states that if multiple hypotheses are compatible with the same evidence, then the smaller the hypothesis, the greater the likelihood it receives (Tenenbaum 1999). Nichols et al. (2016) report evidence that the rule learning behavior of adults can be modeled using this principle. Moreover, they suggest that given the kind of feedback children likely receive, and if their learning is governed by the size principle, we would expect them to end up learning deontological rules such as the doing-allowing and the intended-foreseen distinction. Both distinctions are in fact familiar to ordinary moral thought (e.g., Baron and Ritov 2004; Spranca et al. 1991). Thus, Nichols et al. seem to have identified one mechanism of rational moral learning. What does all this mean for the rationality of our moral intuitions? Some think that it provides cause for optimism (Allman and Woodward 2008; Railton 2017; Sauer 2017). In their view, through the lens of rational learning, moral intuitions are revealed to embody the virtues of the learning systems that shape them. Thus, they are much more sophisticated, flexible, responsive to the evidence, and hence ultimately rational than their critics would have us believe. Other authors have been less impressed. One prominent objection, sometimes referred to as “garbage in, garbage out”, is that any learning system, no matter how rational, can still produce systematically misguided intuitions if it does not receive appropriate input (Greene 2017; Gjesdal 2018). For example, applying the size principle in an environment in which racism is prevalent will likely lead children to acquire racist intuitive rules (cf. Greene 2017: 72–73). Thus, so the objection goes, rational learning alone cannot guarantee the rationality of our moral intuitions. Rather, we also need to carefully investigate the conditions under which such learning systems can function properly. 109

Charlie Blunden, Paul Rehren, and Hanno Sauer

5  Beyond Dual Process II: Triple Process Theory and the Reflective Mind Suppose many of our moral intuitions are as non-rational as authors like Greene, Haidt, and Singer claim. Does this mean that all is lost? Clearly not. At least sometimes, we can engage type 2 processing to reason explicitly about an issue and override our implicit moral intuitions. An approach that puts intuition override front and center is the triple process theory of cognition (TPT). TPT maintains the category of type 1 (intuitive processing) from DPT, but divides type 2 processing into two types. First, algorithmic processing (type 2 in the TPT) is the raw capacity of individuals to carry out explicit and often hypothetical thinking. It relies on working memory capacity and roughly corresponds to a person’s general cognitive ability. Second, reflective processing (or type 3) is the ability of an individual to think critically. It encompasses a whole slew of epistemic dispositions, including the disposition to seek various points of view before coming to a conclusion, to seek nuance, to avoid absolutism, to seek consistency, and to enjoy cognitive processing (Stanovich 2009a, 2011). TPT is motivated by evidence showing that many cognitive biases are independent of cognitive ability. For example, in multiple studies, intelligence scores were found to be unrelated to susceptibility to a number of well-known cognitive biases (Stanovich and West 2007, 2008; West et al. 2008). According to TPT, this is because subjects primarily fall prey to such biases not due to a lack of intelligence, but because they either do not detect the need to override their intuition, do not initiate such override, or because their algorithmic processing does not operate properly (Stanovich et al. 2013). These are the main roles of reflective processing: to detect the need for intuition override, to initiate algorithmic processing to carry out this override, and to monitor algorithmic processing to make sure that works. TPT provides further insight into the mechanics of intuition override by introducing the concept of mindware: “The rules, procedures, and strategies that can be retrieved by the analytic system (the algorithmic and reflective minds)” (Stanovich 2009a: 71). Mindware is crucial for intuition override to work effectively. For example, one notable piece of mindware is the procedure of considering evidence that could falsify one’s hypotheses (Stanovich 2009b: 141–144). This piece of mindware is crucial to getting the correct answer in many of the experiments referred to in section (1). However, mindware can be missing. Moreover, it can be contaminated: someone who has acquired incorrect, self-serving, or evaluation-disabling mindware may produce an alternative response to their initial intuition, but this response will also be incorrect (Stanovich 2011: 95–104). Sauer (2018) has recently applied TPT to moral cognition. For example, recall that according to the SIM, reasoning is typically limited to providing post-hoc rationalizations of our intuitive moral judgments. TPT explains this as algorithmic processing being engaged in serial associative mode in the absence of reflective monitoring (Sauer 2018: 75–76). The widespread inability of people to detect inconsistent moral beliefs can be seen as an instance of missing moral mindware (Sauer 2018: 67). Finally, zero-sum bias, in which people perceive the distribution of desirable resources as having zero-sum character when it in fact has a positive-sum character (Meegan 2010), is arguably an instance of contaminated moral mindware (Sauer 2018: 67–68). TPT emphasizes that we can sometimes override our moral intuitions through critical thinking. To the extent that the thinking dispositions and the mindware necessary for this to happen successfully can be taught, this is cause for optimism about the rationality of moral cognition – even if our moral intuitions frequently go astray (Stanovich 2009b: Ch. 10, 2011: 230–244). However, TPT also points out that engaging in intuition override requires considerable effort, leading to it generally being quite rare (Stanovich 2011: 29; Evans and Stanovich 2013: 237; Sauer 2018: 56–57). 110

Implicit Cognition, Dual Process Theory, and Moral Judgment

6  Beyond Dual Process III: Implicit Moral Judgment and Moral Reasoning Some authors working within the dual process paradigm have tried to show that, at the end of the day, all moral cognition is implicit cognition. One way for moral cognition to be implicit is for moral judgments to be embodied. Here, the idea is that it may be possible to manipulate people’s moral judgments by manipulating their somatic states. Early attempts to do so focused on the emotion of disgust, which is typically thought to be an evolved function facilitating the avoidance of toxins and pathogens by triggering a withdrawal response (Kelly 2011). Initially, the evidence seemed to suggest that tampering with subjects’ feelings of disgust changes people’s moral beliefs, often making them more severe (Schnall, Haidt, et al. 2008; Schnall, Benton, et al. 2008). This approach has turned out to be much less promising than originally thought. For one thing, the generalizability of the existing studies can be questioned: most only ever managed to elicit their effects for some vignettes and some subgroups of people, and rarely managed to genuinely change people’s minds (May 2014). Perhaps even more damningly, the “incidental disgust affects moral judgment” paradigm has, together with studies on the effect of behavioral priming more generally, been ground zero of the so-called reproducibility crisis (Johnson et al. 2014). When other researchers tried to obtain similar effects, they often were not able to, and when they were, the strength of their results was typically much smaller. This is reflected in a recent meta-analysis (Landy and Goodwin 2015). It showed that the aggregate effect size for the manipulation of moral beliefs via extraneous disgust is small (d = .11) and, when publication bias was accounted for, the effect essentially disappeared. Some are wondering why, if moral judgment is based on explicit reasoning rather than implicit processes, there should be any effect of emotional variation whatsoever (Prinz 2016)? But this is unsurprising, because it is perfectly possible for affective reactions to bias our responses in all sorts of domains (such as logic, maths, or the assessment of purely descriptive facts) without those domains thereby being grounded in affect all the way down. Another way of demonstrating that moral cognition is implicit is by showing that explicit moral cognition does not influence people’s moral beliefs (see section (3)). This paradigm, too, is increasingly coming under attack. It should be emphasized that upon closer inspection, Haidt et al’s (2000) original (unpublished) dumbfounding study already failed to establish its more radical sounding conclusions regarding the impotence of moral reasoning: in their experiment, around 20% of people did change their mind after being exposed to pertinent reasons. The boundary between the implicit and the explicit is likely more permeable than the dichotomy of the two systems suggests. The automaticity of moral judgment may well reflect the fact that our moral judgments stem from habitualized patterns of moral judgment (see section (4)) rather than rationally impervious subpersonal processing. Some philosophers (Jacobson 2013) have suspected that when subjects refuse to give up their condemnation of allegedly harmless consensual incest or secretive cannibalism, they actually engage in a subtle appreciation of the potential moral risks associated with those actions, the fact that the experimental vignettes stipulate that no harm actually occurred notwithstanding. Recent studies have confirmed this suspicion and found that perceptions of riskiness are the strongest predictors of moral disapproval in dumbfounding cases (Stanley et al. 2019; Guglielmo 2018). Often, subjects simply refuse to buy into many of the far-fetched details stipulated by the outlandish scenarios they are asked to evaluate. When confronted with stories that elicit strong moral objections, subjects engage in “dyadic completion” (Gray et al. 2014), implicitly assuming that when there has been a transgression, there must therefore be a victim that has been 111

Charlie Blunden, Paul Rehren, and Hanno Sauer

harmed. Royzman et al. (2015) showed that participants are unwilling or unable to import the factual assumptions required by various “harmless taboo” vignettes into their moral appraisal. They do not believe that consensual incest won’t have any dire consequences for those involved. Importantly, subjects can nevertheless be deliberately maneuvered into a state of apparent moral dumbfounding – sticking to a recalcitrant moral intuition they are unable to justify. In order to gauge the influence of reasons on people’s moral judgments more precisely, Stanley et al. (2018) tested whether subjects would change their mind about various morally dilemmas familiar from the literature – the crying baby, the organ harvesting case – when exposed to reasons for or against their initial judgment. They found that “reasons probably won’t change your mind”, but the key word here is probably. Their results show that 10–12% of participants did reconsider their initial judgment. These numbers may seem low. Then again, rational agents hold on to their beliefs at least to some extent. Most of our beliefs are acquired more or less reliably, via perception, experience, inference, or social learning, and a rational mind doesn’t carelessly toss away its belongings either. Moreover, people’s judgment revisions display precisely the right kind of pattern for their cognition to qualify as sensitive to reasons: people don’t change their minds when they receive reasons in line with their initial evaluation, reevaluate their beliefs when confronted with opposing reasons, and land somewhere in the middle when they receive both sets of reasons.

Conclusion The current state of the evidence suggests that (a) moral cognition, to the extent that it is implicit, is a reasonably smart and sophisticated subdomain of decision-making and (b) that implicit moral cognition can be successfully penetrated by explicit forms of moral reasoning in all sorts of ways. While dual process theories remain a valuable framework for understanding human cognition, theories of implicit moral cognition could benefit from adopting more sophisticated models of cognitive processing. This chapter has shown that implicit cognitive processes are often shaped by mechanisms of rational learning, that implicit outputs can, under the right conditions, be kept in check by higher-order cognitive control and that explicit (moral) reasoning has a discernible influence on automatic processing.1

Related Topics Chapters 1, 2, 3, 8, 26, 29, 31

Notes 1 All authors contributed to this chapter equally.

References Allman, J., and Woodward, J. 2008. “What are moral intuitions and why should we care about them? A neurobiological perspective”. Philosophical Issues, 18: 164–185. Baron, J., and Ritov, I. 2004. “Omission bias, individual differences, and normality”. Organizational Behavior and Human Decision Processes, 94: 74–85. Crockett, M. J. 2013. “Models of morality”. Trends in Cognitive Sciences, 17: 363–366. Cushman, F. 2013. “Action, outcome, and value: a dual-system framework for morality”. Personality and Social Psychology Review, 17: 273–292.

112

Implicit Cognition, Dual Process Theory, and Moral Judgment Cushman, F., Kumar, V., and Railton, P. 2017. “Moral learning: psychological and philosophical perspectives”. Cognition, 167: 1–10. Cushman, F., Young, L., and Hauser, M. 2006. “The role of conscious reasoning and intuition in moral judgment: testing three principles of harm”. Psychological Science, 17: 1082–1089. Evans, J. 2006. “The heuristic-analytic theory of reasoning: extension and evaluation”. Psychonomic Bulletin & Review, 13: 378–395. Evans, J. 2008. “Dual-processing accounts of reasoning, judgment, and social cognition”. Annual Review of Psychology, 59: 255–278. Evans, J., and Stanovich, K. E. 2013. “Dual-process theories of higher cognition: advancing the debate”. Perspectives on Psychological Science, 8: 223–241. Foot, P. 2002. “The problem of abortion and the doctrine of the double effect”. In P. Foot, ed., Virtues and vices and other essays in moral philosophy. Oxford: Oxford University Press: 19–32. Gawronski, B., and Creighton, L. A. 2013. “Dual process theories”. In D. E. Carlston, ed., The Oxford handbook of social cognition. New York: Oxford University Press: 282–312. Gjesdal, A. 2018. “Moral learning, rationality, and the unreliability of affect”. Australasian Journal of Philosophy, 96: 460–473. Gray, K., Schein, C., and Ward, A. F. 2014. “The myth of harmless wrongs in moral cognition: automatic dyadic completion from sin to suffering”. Journal of Experimental Psychology: General, 143: 1600–1615. Greene, J. D. 2007. “The secret joke of Kant’s soul”. In W. Sinnott-Armstrong, ed., Moral psychology, vol. 3: The neuroscience of morality: emotion, brain disorders, and development. Cambridge, MA: MIT Press: 35–79. Greene, J. D. 2013. Moral tribes: emotion, reason, and the gap between us and them. London: Atlantic Books. Greene, J. D. 2014. “Beyond point-and-shoot morality: why cognitive (neuro)science matters for ethics”. Ethics, 124: 695–726. Greene, J. D. 2017. “The rat-a-gorical imperative: moral intuition and the limits of affective learning”. Cognition, 167: 66–77. Greene, J. D., Sommerville, B. R., Nystrom, L. E., Darley, J. M., Cohen, J. D. 2001. “An FMRI investigation of emotional engagement in moral judgment”. Science, 293: 2105–2108. Guglielmo, S. 2018. “Unfounded dumbfounding: how harm and purity undermine evidence for moral dumbfounding”. Cognition, 170: 334–337. Haidt, J. 2001. “The emotional dog and its rational tail: a social intuitionist approach to moral judgement”. Psychological Review, 108: 814–834. Haidt, J. 2012. The righteous mind: why good people are divided by politics and religion. New York: Vintage. Haidt, J., and Björklund, F. 2008. “Social intuitionists answer six questions about moral psychology”. In W. Sinnott-Armstrong, ed., Moral psychology, vol. 2: The cognitive science of morality: intuition and diversity. Cambridge: MIT Press: 181–217. Haidt, J., Björklund, F., and Murphy, S. 2000. “Moral dumbfounding: when intuition finds no reason”. Unpublished manuscript, University of Virginia. Hauser, M. et al. 2007. “A dissociation between moral judgments and justifications”. Mind and Language, 22: 1–21. Jacobson, D. 2013. “Moral dumbfounding and moral stupefaction”. In M. Timmons, ed., Oxford studies in normative ethics. Vol. 2. Oxford: Oxford University Press: 289–316. Johnson, D. J., Cheung, F., and Donnellan, M. B. 2014. “Does cleanliness influence moral judgments?”. Social Psychology, 45: 209–215. Kahneman, D. 2011. Thinking, fast and slow. New York: Farrar, Straus and Giroux. Kelly, D. 2011. Yuck! The nature and moral significance of disgust. Cambridge: MIT Press. Keren, G., and Schul, Y. 2009. “Two is not always better than one: a critical evaluation of two-system theories”. Perspectives on Psychological Science, 4: 533–550. Landy, J. F., and Goodwin, G. P. 2015. “Does incidental disgust amplify moral judgment? A meta-analytic review of experimental evidence”. Perspectives on Psychological Science, 10: 518–536. May, J. 2014. “Does disgust influence moral judgment?”. Australasian Journal of Philosophy, 92: 125–141. Meegan, D. V. 2010. “Zero-sum bias: perceived competition despite unlimited resources”. Frontiers in Psychology, 1: 1–7. Melnikoff, D. E., and Bargh, J. A. 2018a. “The insidious number two”. Trends in Cognitive Sciences, 22: 668–669. Melnikoff, D. E., and Bargh, J. A. 2018b. “The mythical number two”. Trends in Cognitive Sciences, 22: 280–293.

113

Charlie Blunden, Paul Rehren, and Hanno Sauer Moors, A., and De Houwer, J. 2006. “Automaticity: a theoretical and conceptual analysis”. Psychological Bulletin, 132: 297–326. Nichols, S., Kumar, S., Lopez, T., Ayars, A., and Chan, H-Y 2016. “Rational learners and moral rules: rational learners and moral rules”. Mind & Language, 31: 530–554. Pennycook, G. 2017. “A perspective on the theoretical foundation of dual process models”. In W. De Neys, ed., Dual process theory 2.0. New York: Routledge: 5–27. Pennycook, G., De Neys, W., Evans, J. S. B. T., Stanovich, K. E., and Thompson, V. A. 2018. “The mythical dual-process typology”. Trends in Cognitive Sciences, 22: 667–668. Prinz, J. 2016. “Sentimentalism and the moral brain”. In S. M. Liao, ed., Moral Brains: The neuroscience of morality. New York: Oxford University Press: 45–73. Railton, P. 2017. “Moral Learning: conceptual foundations and normative relevance”. Cognition, 167: 172–190. Royzman, E. B., Kim, K., and Leeman, R. F. 2015. “The curious tale of Julie and Mark: unraveling the moral dumbfounding effect”. Judgment & Decision Making, 10: 296–313. Sauer, H. 2017. Moral judgments as educated intuitions. Cambridge, MA: MIT Press. Sauer, H. 2018. Moral thinking, fast and slow. Abingdon: Routledge. Schnall, S., Benton, J., and Harvey, S. 2008. “With a clean conscience: cleanliness reduces the severity of moral judgments”. Psychological Science, 19: 1219–1222. Schnall, S., Haidt, J., Clore, G. L., and Jordan, A. H. 2008. “Disgust as embodied moral judgment”. Personality and Social Psychology Bulletin, 34: 1096–1109. Schwitzgebel, E., and Cushman, F. 2012. “Expertise in moral reasoning? Order effects on moral judgment in professional philosophers and non-philosophers”. Mind & Language, 27: 135–153. Schwitzgebel, E., and Ellis, J. 2017. “Rationalization in moral and philosophical thought”. In J.-F. Bonnefon and B. Trémolière, eds., Moral inferences. London: Routledge: 170–190. Singer, P. 2005. “Ethics and intuitions”. The Journal of Ethics, 9: 331–352. Spranca, M., Minsk, E., and Baron, J. 1991. “Omission and commission in judgment and choice”. Journal of Experimental Social Psychology, 27: 76–105. Stanley, M. L., Dougherty, A. M., Yang, B. W., Henne, P., De Brigard, F. 2018. “Reasons probably won’t change your mind: the role of reasons in revising moral decisions”. Journal of Experimental Psychology: General, 147: 962–987. Stanley, M. L., Yin, S., and Sinnott-Armstrong, W. 2019. “A reason-based explanation for moral dumbfounding”. Judgment and Decision Making, 14: 120–129. Stanovich, K. E. 2004. The robot’s rebellion. Chicago: University of Chicago Press. Stanovich, K. E. 2009a. “Distinguishing the reflective, algorithmic and autonomous minds: is it time for a tri-process theory?”. In J. Evans and K. Frankish, eds., In two minds: dual processes and beyond. Oxford: Oxford University Press: 55–88. Stanovich, K. E. 2009b. What intelligence tests miss: the psychology of rational thought. London: Yale University Press. Stanovich, K. E. 2011. Rationality and the reflective mind. Oxford: Oxford University Press. Stanovich, K. E., and West, R. F. 2007. “Natural myside bias is independent of cognitive ability”. Thinking and Reasoning 13: 225–247. Stanovich, K. E., and West, R. F. 2008. “On the relative independence of thinking biases and cognitive ability”. Journal of Personality and Social Psychology, 94: 672–695. Stanovich, K. E., West, R. F., and Toplak, M. E. 2013. “Myside bias, rational thinking, and intelligence”. Current Directions in Psychological Science, 22: 259–264. Tenenbaum, J. B. 1999. “Bayesian modeling of human concept learning”. In M. S. Kearns, S. A. Solla and D. A. Cohn, eds., Advances in neural information processing systems, 11. Cambridge, MA: MIT Press: 59–66. Thompson, J. J. 1985. “The trolley problem”. The Yale Law Journal, 94: 1395–1415. West, R. F., Toplak, M. E., and Stanovich, K. E. 2008. “Heuristics and biases as measures of critical thinking: associations with cognitive ability and thinking dispositions”. Journal of Educational Psychology, 100: 930–941. Woodward, J., and Allman, J. 2007. “Moral intuition: its neural substrates and normative significance”. Journal of Physiology-Paris, 101: 179–202.

114

8 IMPLICIT BIAS AND PROCESSING Ema Sullivan-Bissett

This chapter will consider the kinds of processing involved in implicit bias, and how they relate to the debate on what kind of mental construct implicit biases are. I will begin by identifying what is meant by implicit bias and the indirect psychological measurement instruments designed to track it, before overviewing two streams of psychological research on implicit cognition. I will focus on dual process theories in this chapter, which recognize a distinction between the implicit and explicit, both with respect to processing and mental constructs. I will overview some versions of the canonical view in psychology of implicit biases as associations, before moving to recent empirical work, which has motivated an alternative, propositional understanding of implicit biases and the processes in which they partake. Next I motivate the need to recognize wide-ranging heterogeneity in the category of implicit bias, which accommodates various processes and mental constructs. Finally, I overview my preferred model of implicit biases as constituted by unconscious imaginings which is uniquely placed to accommodate this heterogeneity.

1.  Two Streams of Attitude Research The term attitude is used differently in psychology and philosophy. In psychology it is used to pick out likings or dislikings, which are understood to be embodied in associations between objects and evaluations (Brownstein 2018: 264). In philosophy, the term is used to pick out a whole host of mental states, including beliefs, desires, and intentions (Brownstein and Saul 2016: 7). Talk of implicit biases then captures both kinds of ways of conceptualizing attitudes, since, as we will see, implicit biases have been understood both associatively and propositionally. If implicit biases are propositional, then there is a specific relation between their constituents in virtue of this, a relation which is absent if they are associations (Levy 2015: 804). Michael Brownstein identifies two streams of research key to the development of the field of implicit social cognition, which he labels True Attitudes and Driven Underground (2018: 266, see also Payne and Gawronski 2010). The True Attitudes stream came from cognitive theories of learning and selective attention, and recognizes two ways in which information is processed: automatic and controlled. However, this way of conceptualizing the landscape does not recognize two attitudes (one implicit, one explicit) that subjects might have towards a given target (e.g. a particular social group), but rather recognizes two kinds of information processing and

DOI: 10.4324/9781003014584-11 115

Ema Sullivan-Bissett

measurement instruments – direct and indirect – where the latter is understood as helping us to get at a subject’s true attitude. One account of this kind is Russ Fazio’s Motivation and Opportunity as Determinants (MODE) model. The driving thought behind this is that when subjects have motivation and opportunity to deliberate, they are able to respond to direct measures (that is, measures which allow for time and cognitive resources required for deliberation, and are based on self-report). However, when that control is taken away, indirect measures give us access to automatically processed information representative of one’s true attitude, understood as an association in memory between an object and a subject’s evaluation of that object (Fazio 1990: 82). The strength of the association influences its accessibility and the likelihood that it will be automatically activated in response to certain stimuli (Fazio 1990: 92). On the other hand, there is the Driven Underground stream according to which we have dissociated attitudes towards the same target object (e.g. members of a particular social group). Theories under this umbrella recognize the distinction between the implicit and explicit both in terms of processes and mental constructs. This stream of research grew out of work on implicit memory and gives awareness a key role in distinguishing the implicit from the explicit. In what follows I focus on this stream of research, since this is the space in which philosophers interested in the mental constructs and processing underlying implicit bias have operated.

2. Implicit Bias and Indirect Measurement To begin, we can understand implicit biases as ‘the processes or states that have a distorting influence on behavior and judgement and are detected in experimental conditions with implicit measures’ (Holroyd 2016: 154). Very roughly, implicit biases are posited as mental constructs which influence common micro-behaviours1 and discriminations, which cannot be tracked, predicted, or explained by a subject’s explicit attitudes (i.e. those attitudes a subject has introspective access to). For example, I  might believe that men and women are equally adept at producing excellent philosophy, but when marking my students’ work, if it is not anonymized, I tend to give female students less good marks than male students. My explicit attitudes will not explain that. Implicit biases are also thought to be inaccessible to consciousness, automatically activated, and prevalent among even those who identify as egalitarian.2 And, even though for any specific bias we care to name (e.g. concerning women and weakness) it might not be likely that any particular individual would harbour it, they are nevertheless likely to harbour some bias regarding that particular social group (i.e. women) (Holroyd and Sweetman 2016: 83, fn. 4). In the literature on implicit bias, the term implicit is used to refer to at least four things: a psychological construct, a kind of measurement instrument, a set of processes (cognitive and affective), or a kind of evaluative behaviour (Brownstein 2019). For simplicity, and in honour of my particular interest in the topic, I will use the term implicit to characterize a particular kind of mental construct or process, where ‘mental construct’ is intended to be neutral between exactly what mental item implicit biases are (that is, neutral between implicit biases being associations, beliefs, imaginings, etc.). Another point of order is that the term implicit bias is used in at least two ways: to pick out a particular mental construct responsible for biased judgements and behaviours, or to pick out particular judgements and behaviours thought to be the result of certain implicit cognitions. This is a mere terminological issue; both ways of using the term recognize the existence of certain evaluative judgements and behaviours and mental constructs responsible for them. I will use the term implicit bias to pick out the mental construct involved in the generation of biased evaluative judgements, behaviours, and results on indirect psychological measures. 116

Implicit Bias and Processing

I now turn to overviewing a handful of these indirect (or implicit) measures,3 which were designed to circumvent the shortfalls of direct (or explicit) measures when it came to assessing subjects’ attitudes towards certain social categories (such as people not being honest about their attitudes, perhaps due to social desirability concerns). One way of measuring implicit biases is with an Implicit Association Test (IAT). IATs measure the speed at which subjects are able to pair two categories of objects (e.g. pictures of old and young faces) with, for example, pleasant and unpleasant stimuli (e.g. the words wonderful and horrible). The idea is that the speed and accuracy in the categorization performance of combinations of categories can give us insight into which categories a subject associates with one another (De Houwer et al. 2009: 347). A second implicit measure is semantic priming, which is thought to assess a subject’s strength of association between two concepts. One version of the semantic priming procedure presents two words in close succession (prime and target), where participants have to judge the second word (target), and their reaction time in doing so is recorded. One example of this class of measure comes from Mahzarin R. Banaji and Curtis D. Hardin, who were interested in the speed of judgements of gender-consistent or gender-inconsistent targets, and how that was influenced by a prime (Banaji and Hardin 1996: 136). In their experiment, subjects were exposed to an orientation symbol (+) (500ms), followed by a prime word (either male related, female related, gender neutral, or nonword) (200ms), then a blank screen (100ms), and the target pronoun appeared for as long as it took for a participant to enter a response (Banaji and Hardin 1996: 137). They found that when the target gender matched the prime gender, judgement was faster than when the target gender did not match the prime gender. Overall, Banaji and Hardin took their procedure to provide ‘evidence for automatic gender stereotyping’, which occurred ‘regardless of subjects’ awareness of the prime-target relation, and independently of explicit beliefs about gender stereotypes’ (Banaji and Hardin 1996: 139). Another implicit measure thought to measure the strength of associations is the Go/No-Go association task. Association strength is assessed by the participant’s ability to discriminate items belonging to a target category and attribute from distractor items which are not members of the target category or attribute (Nosek and Banaji 2001: 627). For example, in one condition participants might be asked to simultaneously identify stimuli representing the target category fruit and the attribute good, and in a second condition to simultaneously identify stimuli representing the target category fruit and the attribute bad. The idea is that the ease or otherwise of identifying stimuli of the target category and attribute gives us insight into implicit attitudes regarding the target category (Nosek and Banaji 2001: 627). Finally, the Affect Misattribution Procedure has participants make evaluative judgements in an ambiguous context. In one version of the task, participants are exposed to a positively or negatively valenced prime (e.g. President George W. Bush), and are then told to evaluate an ambiguously valenced target (e.g. an abstract symbol), and that they should avoid expressing any influence of the valenced prime in their evaluation of the ambiguously valenced target (Payne et al. 2005: 277). The idea is that this set-up leads to subjects ‘projecting their own psychological state onto an ambiguous source’ when they misattribute their reactions which are caused by the prime (Payne et al. 2005: 277). The misattribution observed is taken to be a reflection of implicit attitudes.

3. Associationism For those who want to recognize distinct mental constructs with talk of implicit and explicit attitudes, a common way of distinguishing these constructs, broadly and not just with respect to theorizing about implicit bias, is along associative-propositional lines. That is, those theories which endorse the implicit–explicit cognition distinction posit distinct implicit and explicit 117

Ema Sullivan-Bissett

processes characteristic of implicit and explicit mental constructs: one automatic, one controlled. The canonical view of implicit biases is that they are associations whose existence can be traced to the learning history of the subject (Levy 2015: 803). Concepts are associated with valences (e.g. women with negative valence) or with other concepts (e.g. women with weakness) in such a way that the activation of one makes more accessible the activation of another. In the presence of appropriate stimuli, stored associations between, say, women and weakness become activated. In psychology, dual process models of bias take associations to be paradigmatic of implicit cognition, whilst explicit cognitions (e.g. beliefs) involve propositional processing. For example, according to the Reflective-Impulsive Model (RIM), two systems (reflective and impulsive) guide human behaviour, and those systems operate according to distinct principles of information processing. The impulsive system, via the activation within associative networks, elicits spontaneous tendencies of approach and avoidance, whilst the reflective system, via reasoned decisions, influences behaviour (Strack and Deutsch 2004). Similarly, the Associative-Propositional Evaluation model (APE), takes it that the outcomes of associative processes are implicit evaluations, whilst the outcomes of propositional processes are explicit evaluations (Gawronski and Bodenhausen 2014: 188). As this chapter has already made clear, thinking of implicit biases in associative terms permeates the literature. Not only has the distinction between associative and propositional processing been thought to map onto implicit and explicit cognition, the indirect measures overviewed earlier are all thought to be tracking implicit associations (although as Levy 2015: 804 points out, such measures might equally be tracking propositional constructs). While the orthodoxy in psychology has been to understand implicit biases associatively, in philosophy there has been a recent move away from this picture, to instead understanding implicit biases as having propositional contents and as partaking in propositional processing.

4. Propositionalism States with propositional contents can have satisfaction conditions, whereas states with only associative contents cannot, instead associative contents are ‘(relations among) mental representations that lack any syntactic structure’ (Mandelbaum 2013: 199, fn. 1). To see the difference between associative and propositional processing, consider the following examples. The first is what you might be primed to think (implicitly or explicitly) when you hear salt (pepper!). This is a merely associative process, the idea being that your concepts of salt and pepper are associatively linked, such that the activation of one makes more accessible the activation of the other, but there’s no relationship thought to hold between these concepts. Now, second, imagine you get engrossed in a long running TV Series about two academics, Dr Salt and Dr Pepper, engaged in a Shakespearean love affair across university faculties traditionally at odds with one another. Now the relationship in your mind between salt and pepper is not a matter of one making more accessible the other (although perhaps that too), but rather that there’s a specific relation which one bears to the other, e.g. Salt loves Pepper. Return to implicit biases. On an associationist understanding of one kind of gender bias, it might be said that my concept of women is associatively linked to another concept, weakness. But this conception of the mental apparatus does not specify a particular relationship which holds between women and weakness, it is rather just a matter of one concept making more accessible the activation of the other. On a propositionalist understanding of this particular bias however, the story might be that I have a propositionally structured mental construct with the content women are weak. In this case we have a single mental item (rather than an association between two), and the specification of a particular relationship between the constituents, that of predication.

118

Implicit Bias and Processing

Some empirical work on the behaviour of implicit biases has motivated a move to propositionalism, the case for which has been most robustly put by Eric Mandelbaum (2016). One study he draws on by Bertram Gawronski and colleagues (2005) had participants introduced to a photo of an unfamiliar person (CS1), which was then paired with either positive or negative statements (in order to set up the association between the CS1 and a particular evaluation). Then participants were introduced to a second photo of an unfamiliar person (CS2) and told either that the CS1 liked the CS2 or that the CS1 did not like the CS2. Subjects then underwent an explicit likeability rating to gauge explicit attitudes towards the CS1 and CS2, and an affective priming task in which they had to identify positive or negative words as such, having been primed with one of the previous images, to gauge implicit attitudes towards the CS1 and CS2 (Gawronski et al. 2005: 621). The results showed that, for example, if the CS1 was paired with negative statements, and participants were told that the CS1 did not like the CS2, then participants liked the CS2. Reflecting on this Mandelbaum suggests that an associative theory would predict the opposite of this result: negative valence + negative valence = negative negative valence (that is, an associative account would predict ‘enhanced negative reactions toward the CS2 because you a) are encountering the CS2 as yoked to negatively valenced CS1 and b) you are activating another negative valence because you are told that the CS1 dislikes the CS2’ (2016: 639)). Mandelbaum sums up the implication of this by noting that ‘if you find two negatives making a positive, what you’ve found is a propositional, and not an associative, process’ (2016: 639). Further support for the claim that associative models are unable to account for many experimental findings is given by discussions of other work (for evidence that implicit biases are sensitive to argument strength, see Brinol et al. 2009; for evidence that implicit biases are adjustable in light of peer judgement, see Sechrist and Stangor 2001). Experiments such as these have led some philosophers to develop propositional models of bias. Before overviewing Mandelbaum’s view, it is worth noting that the argument from the results of these studies to the propositional structure of bias is not without its critics. As Josefa Toribio points out, the move made by Mandelbaum is from biased behaviour (i.e. that measured by implicit measures) being modulated by logical and evidential considerations, to the claim that that which is responsible for the behaviour is propositional (Toribio 2018: 41; see also Brownstein et al. 2019: 5). Mandelbaum moves between these claims via a supposed inference to the best explanation. However, Toribio argues that an associationist view can accommodate the results from the empirical studies Mandelbaum uses in his case for propositionalism (Toribio 2018: 44). Moreover, she argues that ‘[e]ven if implicit attitudes can sometimes be modulated by evidential and rational considerations, this is not their most distinctive characteristic, and any inquiry into their nature that emphasizes this aspect will be off-target as far as best explanations go’ (Toribio 2018: 44). Toribio’s argument is as follows: at the general level, sensitivity to logical or evidential considerations is not necessarily best explained by the state in question being propositionally structured. She uses the example of pain and disgust, both of which can be modulated logically or evidentially, and yet we do not take that fact to be grounds on which to pronounce that they are thus propositionally structured (Toribio 2018: 44–47). In addition, Toribio argues that the move from evidential/logical moderation to the propositional structure of implicit bias would work only if implicit biases were the sole cause of implicitly biased behaviour. However, no one in social psychology would deny that all sort of factors – associative, nonassociative and even non-attitudinal processes – have to be taken into account when

119

Ema Sullivan-Bissett

offering explanations of implicit attitudes’ modulation [. . .] It is perfectly consistent to maintain that implicit attitudes are associations and that factors other than counterconditioning and extinction can modulate implicitly biased behaviour. (Toribio 2018: 50) Notwithstanding criticism of Mandelbaum’s argument for the propositional nature of implicit bias, I  now move on to look at his model, on which implicit biases are unconscious beliefs.4 Philosophical views which characterize the target phenomenon as a belief fall under the heading of doxasticism, and so Mandelbaum’s view is a version of this approach. On its face, that might sound surprising. On a traditional Cartesian way of thinking about the matter, beliefs are largely evidence-responsive, and propositions can be deliberated upon, and then taken up in belief, or not (Gilbert et al. 1993: 221). Beliefs might be thought to be propositional states whose contents we take to be true, whose contents we take ourselves to be committed to, and it is also usually thought that it is not possible to hold conflicting beliefs. If implicit biases are beliefs, these are features we might have to deny when we reflect on cases of implicitly biased egalitarians (that is, those folk for whom their implicit attitudes do not cohere with their egalitarian explicit attitudes). However, Mandelbaum is not working with a traditional Cartesian conception of belief, but rather understands belief in a Spinozan fashion. According to this understanding, we believe any truth-apt proposition that we represent. So there is no gap between representing a truth-apt proposition, and believing it, that is, ‘the act of understanding is the act of believing’ (Gilbert et al. 1993: 222, my emphasis). In light of the putative worry that such an account would attribute conflicting beliefs to single subjects, the Spinozan doxasticist has at her disposal the idea of the mind as fragmented (a la Lewis 1982; Stalnaker 1984; or more recently, Egan 2008). Mandelbaum’s claim that implicit biases are beliefs has been subject to a host of objections.5 However, Mandelbaum claims not to be motivated primarily by capturing implicit biases as a particular kind of mental construct, but rather as capturing them as propositionally structured. He notes that if the term belief offends readers should feel free to understand his hypothesis as one about structured thoughts (Mandelbaum 2016: 636). To this I note that if it is in the spirit of Mandelbaum’s position to understand structured thoughts non-doxastically, then what we have arrived at is only the propositional nature of implicit bias (at best). The question of what kind of mental construct implicit biases are remains open.

5. Heterogeneity Many authors have pointed out the need to recognize serious heterogeneity in our model of bias. That is, although it is common to talk of implicit bias simpliciter (without making any finer grained distinctions within the category), it is almost certainly the case that the category admits of significant heterogeneity. Subsuming all implicit biases under a single mental construct partaking in one kind of process risks making unwarranted generalizations about how implicit biases behave, in particular, how they might influence judgements of, and behaviour towards, members of certain social groups (Holroyd and Sweetman 2016: 84). Firstly, there are differences with respect to implicit biases having particular features. For example, although it is often said that implicit biases are automatic (as opposed to controlled), some have argued that if not the activation of implicit biases (e.g. the activation of a stored association), their expression (i.e. the influence on behaviour) admits of control (see Chehayeb 2020: 123–126). Implicit biases are also often characterized as not being introspectively accessible, or as being unconscious (indeed, it is this feature which is central to dual process theories of 120

Implicit Bias and Processing

implicit cognition).6 Some studies show that under certain conditions, there is more convergence between a subject’s reported explicit attitudes and their implicit attitudes as identified by indirect measures (see e.g. Nier 2005; Hahn et al. 2014). This has persuaded some theorists that implicit biases may well admit of some introspective accessibility (see e.g. Brownstein et al. 2019; Chehayeb 2020: 129–133; Gawronski 2019: 575–578; Levy 2014: 30).7 Secondly, implicit biases differ in their contents, insofar as they are about different social groups. For example, we can have implicit biases about women, Black and ethnic minority people, homosexual people, members of certain religious communities, and so on.8 Thirdly, implicit biases concerning the same social groups can vary with respect to their expression. That is, even though a given individual may score highly on an IAT testing for associations between Black men and stereotypical traits, they may nevertheless not score highly on an IAT testing for negatively valenced associations involving their concept of Black men, and vice versa (see Amodio and Devine 2006). These distinct biases are also predictive of different behaviours (the former influencing judgements of competence, the latter influencing seating distance from a Black confederate). One way of understanding what the two kinds of IAT are tracking here is as two kinds of association which fall under the label of implicit bias: semantic and affective (see Holroyd and Sweetman 2016: 92ff for arguments against the explanatory utility of this distinction). So how do we make good on this heterogeneity? One approach comes from Bryce Huebner, who argues for a variety of ways that implicit associations get internalized. Specifically, he has it that implicit biases reflect the combined influence of three computationally and psychologically distinctive evaluative systems (associative processing, associative Pavlovian systems, and associative model-free systems) (2016: 51). Guillermo Del Pinal and Shannon Spaulding also speak to the heterogeneity of biases, though at the level of their encoding. Usually salient-statistical associations are thought to be the relevant ones for modelling implicit biases. But Del Pinal and Spaulding argue that some biases are encoded ‘in the dependency networks which are part of our representations of social categories’, and not all as salient-statistical associations (2018: 96). This would mean that some biases can be encoded in our concepts in ways that systematically dissociate from salient-statistical properties. Rather, concepts can encode information regarding cue-validity and saliency, but also the degree of centrality of their associated features. We have seen then that the mental constructs falling under the label of implicit bias admit of heterogeneity with respect to key features (control, introspective accessibility, content, and influences on behaviour). We have also seen that those theorists keen to accommodate heterogeneity do so within the confines of associationism, that is, heterogeneity is grounded in different kinds of association (semantic/affective, salient-statistical/dependency networks) or ways in which these associations become a part of our cognitive architecture (Chehayeb 2020 is a notable exception). In the next section I argue that we should take heterogeneity more seriously than this, before briefly defending my preferred view of bias which is able to do so.

6.  Biased by Our Imaginings Although many theorists have wanted to recognize heterogeneity, we have seen that accounts of implicit biases fall squarely into either associationism or propositionalism, and any heterogeneity posited remains within the boundaries of these respective frameworks. On my preferred view9 implicit biases are constituted by unconscious imaginings, and the heterogeneity within the category spans the associative-propositional distinction.10 Before getting to the details, I will say something about how I am understanding unconscious imagination. 121

Ema Sullivan-Bissett

In place of a robust account of imagination, I appeal to three features of it upon which there is ‘wide agreement’ (Kind 2016: 1): it is a primitive mental state (that is, it is irreducible to other mental states, cf. Langland-Hassan 2012), it has representational content (it is about something), and it is not connected to truth11 (Kind 2016: 1–3). It is this kind of state which I think can do good explanatory work when turning to the nature of implicit bias, and is also uniquely placed to take part in both associative and propositional processes. Unconscious imaginings then are simply states with the three features upon which there is wide agreement, and which are tokened in a way as to be inaccessible to introspection.12 In pursuit of recognizing a wide-ranging heterogeneity, I also distinguish two kinds of imagining on grounds of content: propositional imaginings (imagining that your partner’s birthday is next month) and imagistic imaginings (imagining your partner’s face). On my view, both kinds of imaginings can be tokened unconsciously, and both kinds of imaginings have a role to play in our account of implicit bias. If implicit biases are constituted by unconscious imaginings, and unconscious imaginings are candidate mental constructs for partaking in both associative and propositional processes, my view thus recognizes heterogeneity at the level of mental constructs and processing. That is, it can accommodate implicit biases being associatively structured (i.e. associations between multiple imaginings) and it can accommodate them being propositional (i.e. single imaginings with propositional contents apt to e.g. partake in inference). This is a theoretical virtue given the mounting evidence for heterogeneity in this class of attitudes. My account can do two key things. It can say what is common among all implicit biases which justifies grouping all of these things together under a single label, and it can also admit of finer distinctions within the overall category, by appeal to the different ways in which implicit biases can be constituted by unconscious imaginings (e.g. by associatively linked imaginings, or a relationship of identity in cases of single propositional imaginings). I will now run through an example to see the various ways my model allows for implicit biases to be constituted. Our starting point is that to have an implicit bias is to unconsciously imagine certain things in response to stimuli. For biases structured associatively, the constituents of bias are associatively linked and do not stand in determinate syntactic relations. Against such a background, one of three things could be going on in the presence of certain stimuli, say, a woman. The first way of understanding implicit bias on my view is as associatively linked unconscious imagistic imaginings (i.e. of woman and weakness) (as Toribio points out, understanding implicit biases as associations is consistent with thinking of the associated mental constructs as images (2018: 42)). Alternatively implicit bias could be understood as associatively linked propositional imaginings (i.e. there is a woman and there is weakness), or as an unconscious imagistic or propositional imagining associatively linked with a negative valence.13 As we have seen though, there has been a recent move to modelling implicit bias as propositionally structured, and empirical work suggesting that this is required, in at least some cases. If that is right, we should understand the constituents of implicit bias as standing in determinate relations to one another. There are two ways my imagination model can capture what form implicit biases could take against a propositional background when presented with certain stimuli, say, a woman. A subject could have an unconscious imagistic imagining of a weak women, or an unconscious propositional imagining that women are weak. In the first case we have a single imagistic imagining (rather than an association between two such imaginings), and in the second case we have a single propositional imagining (rather than an association between two such imaginings). This last way of understanding the possible structure of implicit bias is where Mandelbaum’s unconscious belief model and my unconscious imagination model look very similar and may share predictions. 122

Implicit Bias and Processing

Before closing, I will say something about the theoretical and predictive benefits that might be gained if we understand implicit biases in the way I suggest. The different features, structures, and behaviours characteristic of implicit biases are theoretically interesting in their own right, but are also likely to be significant when we think of the kinds of normative recommendations regarding mitigation strategies that might be suggested on the basis of the nature of implicit biases. My general point is that my account can admit of more particular carvings of the category of implicit bias along the lines of which kind of imaginings and processes are in play. Different kinds of imaginings may be predictive of different behaviour, and the more we learn about the operations of different kinds of unconscious imaginings, the more predictions we will be able to make about implicit biases understood in such terms. In their argument for associative heterogeneity, Holroyd and Sweetman suggest that the way different associations operate may be explained by differences in content and underpinning processes (Holroyd and Sweetman 2016: 88). The way imagination can be involved in associative processing (where the components are imagistic or propositional) may be one way in which different underpinning processes are involved in associative processing. But, in addition, we should also take seriously the possibility that dissociative scores on indirect measurements may not only be down to different kinds of associations and the processes which underpin them, but may also be the result of measures tapping into different kinds of implicit processes (associative and propositional). My model also allows for variation in the mechanisms responsible for biased behaviours. Where implicit biases are underpinned by associative processing (with unconscious imaginings as the constituents of the association), the explanation for certain judgements or behaviours can be given by appeal to one imagining activating another. But where implicit biases are nonassociative, and instead involve single propositionally structured mental constructs (i.e. single unconscious imaginings), the imagining itself has motivational credentials. For example, the idea would be that when presented with particular stimuli (e.g. a woman’s face), a subject with an unconscious propositional imagining takes there to be a determinate relation between woman and weakness (i.e. the constituents of her attitude). So some implicit biases, in virtue of being propositional, posit determinate relations between the target stimuli and some stereotypical feature or valence. It might be thought that the cost of accommodating significant heterogeneity is giving up on understanding implicit biases as a single kind of mental construct. My account has the twin benefits of accommodating both unification and heterogeneity. It is unifying insofar as it can principally group all implicit biases under a single mental category (as constituted by unconscious imaginings). But it also allows for diverse ways in which implicit biases can be constituted, as well as various processing in which implicit biases might partake.

7. Conclusion In this chapter I  have overviewed some ways of thinking about the processing involved in explicit and implicit cognition, with a focus on implicit bias. With respect to this category, evidence is mounting for the idea that we are not in the domain of a neat and tidy subset of implicit cognition, but rather that an extensionally adequate account of implicit bias and the processing involved needs to recognize significant heterogeneity. Extant accounts are unable to do this insofar as they situate themselves in either an associative or propositional framework. I suggested that my preferred model of implicit bias as constituted by unconscious imaginings is uniquely able to meet the challenge of heterogeneity, whilst also offering a unifying model of the mental constructs and processes in implicit bias.14 123

Ema Sullivan-Bissett

Related topics Chapters 1, 2, 7, 16, 24, 25

Notes 1 Two notes of caution are called for here: some have argued on the basis of a meta-analysis of IAT results, that such results are a poor predictor of behaviour (Oswald and colleagues 2013, see Brownstein et al. 2019 for discussion). Others have argued that procedures designed to change results on implicit measures can do so (in a limited way, and in the short term), but that there is little evidence that such changes are reflected in behaviour (Forscher and colleagues 2019). Since this chapter is primarily about the processing underlying implicit biases, and less about the behavioral effects of them, I say no more about these points. 2 In their review of over 2.5 million results from Implicit Association Tests (described later) across seventeen topics, Brian Nosek and colleagues report that ‘[i]mplicit and explicit comparative preferences and stereotypes were widespread across gender, ethnicity, age, political orientation, and region’ (2007: 40). 3 See Brownstein and colleagues (2019) and Gawronski and De Houwer (2014) for excellent critical overviews of implicit measures. 4 It is beyond the scope of this piece to mention all of the views on the nature of implicit bias put forward by philosophers. Other candidate mental constructs have included: alief (Gendler 2008), character traits (Machery 2016), implicit beliefs (Frankish 2016), in-between beliefs (Schwitzgebel 2010), patchy endorsements (Levy 2015), and mental imagery (Nanay 2021). More radical views include Chehayeb’s (2020) mosaic view and Johnson’s (2020) non-representational account. 5 For example see Holroyd (2016), Levy (2015), Madva (2016). See section 4.1 of my (2019) for an overview of these objections. 6 Gawronski and colleagues (2006) distinguish three types of awareness: source, content, and impact, and they argue that implicit attitudes only differ from explicit attitudes with respect to impact awareness (cf. Gawronski 2019: 577). 7 I think that these studies might in fact be consistent with implicit biases being mental constructs to which we do not have introspective awareness, and indeed, the psychologists running the studies suggest that the results can be interpreted in a way that retains the introspective inaccessibility of implicit biases. In short the idea is that subjects make pretty good predictions about the biases they have by inferring from their affective reactions to certain stimuli to the existence of a bias (as suggested by Hahn and colleagues (2014: 1387)). But that is not to say that they have access to the bias itself, but rather an affective reaction downstream of it. See Berger (2020) for an alternative way of understanding the distinction between implicit and explicit cognitions in terms of awareness. 8 Of course, we also have implicit cognitions which are not about social categories, for example, associations concerning feared objects (Holroyd and Sweetman 2016: 86, fn. 7). 9 A full statement and defence of my view of implicit bias can be found in Sullivan-Bissett (2019), a defence of my view in light of studies on mitigating bias through virtual reality can be found in Sullivan-Bissett (manuscript). 10 For a radically different account of the mental constructs underlying social behaviour, which accommodates more far-reaching heterogeneity than my view, see Chehayeb (2020, Ch. 6). 11 Kind understands this third feature as specifying the absence of a constitutive connection to truth. Elsewhere (Sullivan-Bissett 2017, 2018, 2020), I defend a contingent relationship between belief and truth, and so I have dropped the ‘constitutive’. Whatever one makes of the strength of the relationship, the point is that belief is connected to truth in a way that imagining is not. 12 It might be thought that understanding imaginings as tokened unconsciously departs from a standard view of imagination, and thus that my preferred view has ontic costs to pay. However, the three features upon which there is wide agreement are neutral with respect to whether imaginings can be tokened unconsciously. So, if unconscious imagination does represent a departure from a standard view, that departure is not to be found in these three uncontroversial features (for a more through defence of the claim that allowing for imaginings to be tokened unconsciously is not to endorse a revisionary notion of the imagination see Sullivan-Bissett (2019: 631–635)). 13 There is no reason to rule out at this stage associations between different kinds of mental items (i.e. an imagistic imagining and a propositional imagining), but it is unclear to me what the empirical evidence would have to look like to motivate this possibility.

124

Implicit Bias and Processing 14 I am grateful to Fidaa Chehayeb, Michael Rush, Robert Thompson, and an anonymous reviewer for comments on an earlier draft of this chapter.

References Amodio, D. M., and Devine, P. G. 2006. “Stereotyping and evaluation in implicit race bias: evidence for independent constructs and unique effects on behavior”. Journal of Personality and Social Psychology, 91: 652–661. Banaji, M. R., and Hardin, C. D. 1996. “Automatic stereotyping”. American Psychological Society, 7: 136–141. Berger, J. 2020. “Implicit attitudes and awareness”. Synthese, 197: 1291–1312. Brinol, P., Petty, R., and McCaslin, M. 2009. “Changing attitudes on implicit versus explicit measures: what is the difference?”. In R. Petty, R. Fazio, and P. Brinol, eds., Attitudes: insights from the new explicit measures. New York: Psychology Press: 285–326. Brownstein, M. 2018. “Implicit bias and race”. In P. C. Taylor, L. M. Alcoff, and L. Anderson, eds., The Routledge companion to the philosophy of race. New York: Routledge: 261–276. Brownstein, M. 2019. “Implicit bias”. In E. Zalta, ed., The Stanford encylopedia of philosophy. Fall 2019 edn. https://plato.stanford.edu/archives/fall2019/entries/implicit-bias/ Brownstein, M., Madva, A., and Gawronski, B. 2019. “What do implicit measures measure?”. Wires Cognitive Science, 10: e1501. Brownstein, M., and Saul, J. 2016. “Introduction”. In M. Brownstein and J. Saul, eds., Implicit bias and philosophy, vol. 1: Metaphysics and epistemology. Oxford: Oxford University Press: 1–19. Chehayeb, F. 2020. Contra implicit bias. PhD Thesis. University of Birmingham. De Houwer, J., Teige-Mocigemba, S., Spruyt, A., and Moors, A. 2009. “Implicit measures: a normative analysis and review”. Psychological Bulletin, 135: 347–368. Del Pinal, G., and Spaulding, S. 2018. “Conceptual centrality and implicit bias”. Mind & Language, 33: 95–111. Egan, A. 2008. “Seeing and believing: perception, belief formation, and the divided mind”. Philosophical Studies, 140: 47–63. Fazio, R. 1990. “Multiple processes by which attitudes guide behavior: the MODE model as an integrative framework”. Advances in Experimental Social Psychology, 23: 75–109. Forscher, P. S., Lai, C. K., Axt, J. R., Ebersole, C. R., Herman, M., and Devine, P. G. 2019. “A meta-analysis of procedures to change implicit measures”. Journal of Personality and Social Psychology, 117: 522–559. Frankish, K. 2016. “Playing double: implicit bias, dual levels, and self-control”. In M. Brownstein and J. Saul, eds., Implicit bias and philosophy, vol. 1: Metaphysics and epistemology. Oxford: Oxford University Press: 23–46. Gawronski, B. 2019. “Six lessons for a cogent science of implicit bias and its criticism”. Perspectives on Psychological Science, 14: 574–595. Gawronski, B., and Bodenhausen, G. V. 2014. “The associative-propositional evaluation model: operating principles and operating conditions of evaluation”. In J. W. Sherman, B. Gawronski, and Y. Trope, eds., Dual-process theories of the social mind. New York: Guilford Press: 188–203. Gawronski, B., and De Houwer, J. 2014. “Implicit measures in social and personality psychology”. In H. T. Reis and C. M. Judd, eds., Handbook of research methods in social and personality psychology. New York: Cambridge University Press: 283–310. Gawronski, B., Hofmann, W., and Wilbur, C. J. 2006. “Are ‘implicit’ attitudes unconscious?” Consciousness and Cognition, 15: 485–499. Gawronski, B., Walther, E., and Blank, H. 2005. “Cognitive consistency and the formation of interpersonal attitudes: cognitive balance affects the encoding of social information”. Journal of Experimental Social Psychology, 41: 618–626. Gendler, T. 2008. “Alief and belief ”. Journal of Philosophy, 105: 634–663. Gilbert, D. T., Tafarodi, R. W., and Malone, P. S. 1993. “You can’t not believe everything you read”. Attitudes and Social Cognition, 65: 221–233. Hahn, A., Judd, C. M., Hirsh, H. K., and Blair, I. V. 2014. “Awareness of implicit attitudes”. Journal of Experimental Psychology: General, 143: 1369–1392. Holroyd, J. 2016. “What do we want from a model of implicit cognition?”. Proceedings of the Aristotelian Society, 166: 153–179. Holroyd, J., and Sweetman, J. 2016. “The heterogeneity of implicit bias”. In M. Brownstein and J. Saul, eds., Implicit bias and Philosophy, vol. 1: Metaphysics and epistemology. Oxford: Oxford University Press: 80–103.

125

Ema Sullivan-Bissett Huebner, B. 2016. “Implicit bias, reinforcement learning, and scaffolded moral cognition”. In M. Brownstein and J. Saul, eds., Implicit bias and Philosophy, vol. 1: Metaphysics and epistemology. Oxford: Oxford University Press: 47–79. Johnson, G. M. 2020. “The structure of bias”. Mind, 129: 1193–1236. Kind, A. 2016. “Introduction: exploring imagination”. In A. Kind, ed., The Routledge handbook of philosophy of imagination. London: Routledge: 1–11. Langland-Hassan, P.  2012. “Pretense, imagination, and belief: the single attitude theory”. Philosophical Studies, 159: 155–179. Levy, N. 2014. “Consciousness, implicit attitudes and moral responsibility”. Noûs, 48: 21–40. Levy, N. 2015. “Neither fish nor fowl: implicit attitudes as patchy endorsements”. Noûs, 49: 800–823. Lewis, D. 1982. “Logic for equivocators”. Noûs, 16: 431–441. Machery, E. 2016. “De-Freuding implicit attitudes”. In M. Brownstein and J. Saul, eds., Implicit bias and Philosophy, vol. 1: Metaphysics and epistemology. Oxford: Oxford University Press: 104–129. Madva, A. 2016. “Why implicit attitudes are (probably) not beliefs”. Synthese, 193: 2659–2684. Mandelbaum, E. 2013. “Against Alief ”. Philosophical Studies, 165: 197–211. Mandelbaum, E. 2016. “Attitude, inference, association: on the propositional structure of implicit bias”. Noûs, 50: 629–658. Nanay, B. 2021. “Implicit bias as mental imagery”. Journal of the American Philosophical Association, 7: 329–347. Nier, J. 2005. “How dissociated are implicit and explicit racial attitudes? A  bogus pipeline approach”. Group Process & Intergroup Relations, 8: 39–52. Nosek, B., and Banaji, M. 2001. “The Go/No-go association task”. Social Cognition, 19: 625–664. Nosek, B., Smyth, F. L., Hansen, J. J., Devos, T., Lindner, N. M., Ranganath, K. A., Tucker Smith, C., Olson, K. R., Chugh, D., Greenwald, A. G., and Banaji, M. 2007. “Pervasiveness and correlates of implicit attitudes and stereotypes”. European Review of Social Psychology, 18: 36–88. Oswald, F. L., Mitchell, G., Blanton, H., Jaccard, J., and Tetlock, P. E. 2013. “Predicting ethnic and racial discrimination: a metaanalysis of IAT criterion studies”. Journal of Personality and Social Psychology, 105: 171–192. Payne, K., Cheng, C. M., Govorum, O., Stewart, B. D. 2005. “An inkblot for attitudes: affect misattribution as implicit measurement”. Journal of Personality and Social Psychology, 89: 277–293. Payne, K. B., and Gawronski, B. 2010. “A history of implicit social cognition: where is it coming from? Where is it now? Where is it going?”. In B. Gawronski and K. B. Payne, eds., Handbook of implicit social cognition: measurement, theory, and applications. New York: Guilford Press: 1–15. Schwitzgebel, E. 2010. “Acting contrary to our professed beliefs, or the gulf between occurrent judgment and dispositional belief ”. Pacific Philosophical Quarterly, 91: 531–553. Sechrist, G. B., and Stangor, C. 2001. “Perceived consensus influences intergroup behavior and stereotype accessibility”. Journal of Personality and Social Psychology, 81: 645–654. Stalnaker, R. 1984. Inquiry. Cambridge, MA: MIT Press. Strack, F., and Deutsch, R. 2004. “Reflective and impulsive determinants of social behavior”. Personality and Social Psychology Review, 8: 220–247. Sullivan-Bissett, E. 2017. “Biological function and epistemic normativity”. Philosophical Explorations, 20: 94–110. Sullivan-Bissett, E. 2018. “Explaining doxastic transparency: aim, norm, or function?”. Synthese, 195: 3453–3476. Sullivan-Bissett, E. 2019. “Biased by our imaginings”. Mind & Language, 34: 627–647. Sullivan-Bissett, E. 2020. “We Are Like American Robins”. In S. Stapleford and K. McCain, eds., Epistemic Duties: New Arguments, New Angles. Routledge: 94–110. Sullivan-Bissett, E. manuscript. “Virtually imagining our biases”. Toribio, J. 2018. “Implicit bias: from social structure to representational format”. Theoria, 33: 41–60.

126

9 PREDICTIVE PROCESSING, IMPLICIT AND EXPLICIT Paweł Gładziejewski

1. Introduction The implicit/explicit distinction – as applied to internal representations and to the rules according to which they are processed – has played a significant role in the late 20th century, when the sciences of cognition saw a shift from symbolic to connectionist accounts of information processing underlying cognition. According to the classical symbolic approach, representations are explicitly encoded in strings of symbols and processed according to transition rules that are (at least sometimes) explicitly stored (Clapin 2002; Dennett 1983). Connectionist modeling introduced an altogether different view, where ‘representations’ are not stored in separable, localized vehicles but are rather implicit in the connection weights of neurons comprising the network; and the ‘knowledge’ of processing rules as implicitly embodied in the dispositions of the network to transform input vector of activation into output vectors (Haugeland 1991; Ramsey 2007: Ch. 5). This chapter aims to put the implicit/explicit distinction into service once again to help make sense of the changing theoretical landscape of cognitive science. My focus will be on the predictive processing (henceforth PP) view of cognition that recently stirred some excitement among many philosophers and cognitive (neuro)scientists. According to PP, the brain is in the business of minimizing the prediction error, which measures the mismatch between internally predicted and actual sensory input. This single idea, the story goes, could explain perception, attention, and motor control, as well as potentially scale up to account for more off-line forms of cognition. Two drastically different ways of understanding the commitments of PP have emerged in the literature. The ‘intellectualist’ interpretation views prediction error minimization as a process that rests on using an internal representation to infer the causes of sensory stimulation. Proponents of the ‘radical’ view argue that PP is best viewed as an enactive (non-representational and non-inferential) framework. Here, I will propose a conciliatory reading of PP, where intellectualist and radical interpretations correspond to different ways in which predicting sensory states and computing prediction errors can be realized. In some cases (those in line with ‘intellectualist’ PP), the representational notions at use in PP pick out activities of an internal model that serves as an explicit

DOI: 10.4324/9781003014584-12 127

Paweł Gładziejewski

representation. The processes of updating and revising such a model count as genuinely inferential, although in a fairly minimal sense of implicit inference (that is, although the representations are explicit, the rules of processing are not). In other cases (those in line with ‘radical’ PP), the representational notions in PP pick out states merely implicit in processing biases attuned to statistics of natural scenes. Here, representational talk is arguably best read as a gloss on biased feed-forward processing that implicitly ‘predicts’ certain regularities in sensory input. Rather than inferring the causes of the sensory input, these implicit ‘predictions’ enable efficient encoding of the input. I will argue that this hybrid reading of PP introduces an interesting spin on the philosophical notion that explicit knowing-that presupposes and rests on a background of implicit knowing-how (Clapin 2002; Dennett 1983; Ryle 1949/2000). The discussion is structured as follows. In section 2, I familiarize the reader with basic tenets of PP and summarize what intellectualist and radical readings of the framework amount to. Section 3 is devoted to distinguishing two types of mechanisms that often go under a single umbrella of PP, where mechanisms of one sort employ an explicit representations, while mechanisms of the other sort rely on processing biases that implicitly embody ‘predictions’ about sensory regularities. In section 4, I outline the sense in which the mechanisms of the former, explicit-representation-using type perform genuine, albeit non-explicit (i.e. implicit or tacit) Bayesian causal inference. I end with a brief summary.

2.  ‘Intellectualist’ and ‘Radical’ Predictive Processing 2.1  Core Tenets of Predictive Processing To perceive the world, the cognitive system needs to reduce its uncertainty with respect to distal causes of its sensory states.1 One possible way to achieve this is by employing a detectorbased strategy, where pieces of internal machinery selectively react to relevant features of the environment (say, the presence of edges, flies, or human faces). However, complications associated with interpreting the sensory input in terms of its distal causes suggest a different solution. These complications have to do, first, with the fact that sensory states are inherently corrupted by random noise. And, second, they stem from the fact that the flow of sensory input depends on a whole manifold of interacting causes, and there are multiple possible ways of ‘unmixing’ the input to yield a perception of a scene. Somehow, the cognitive system has to decide on just one interpretation. Consider, then, a solution to the problem of overcoming sensory ambiguity where the causes of sensory states are inferred or predicted, rather than detected (c.f. Bogacz 2017; Clark 2013b, 2016; Friston 2005, 2010; Hohwy 2013; Rao and Ballard 1999). On this approach, the cognitive system produces a set of hypotheses about the way the sensory signals are generated by their worldly causes. The hypotheses regarding the causes of sensory states comprise a generative model. Technically, the generative model is defined in terms of a joint probability distribution P(H,D), where H stands for hypotheses about the distal causes ‘hidden’ behind the sensory states (e.g. edges, flies, or faces) and D stands for ‘observable’ data, namely the sensory input itself (say, the retinal image). This joint distribution is equal to a product of the prior distribution P(H), which specifies how likely a given hypothesis is prior to obtaining sensory data, and the likelihood distribution P(D|H), which expresses the conditional probability of obtaining the data, given the hypothesis.2 The generative model can thus serve as a basis for simulating the sensory input through drawing samples from the model: one takes a hypothesis from the prior distribution, and then, using the likelihood distribution, generates data that this hypothesis deems most likely.

128

Predictive Processing, Implicit and Explicit

To subserve perception (rather than free-flowing imagery), the hypotheses derived from the generative model ought to become answerable to the input itself (i.e. the samples ‘drawn’ from the environment that the model is trying to recapitulate). At this point, the notion of prediction error comes in. The model-based hypotheses are treated as predictions about what streams of sensory input are most likely to be obtained. The hypotheses can then be then adjusted by measuring the mismatch (for example mean squared error) between the sensory states predicted and the states actually sampled. This process enables the system to settle on a hypothesis that is most likely true, given the current sensory input. This is equivalent to performing perceptual inference by inversing the generative model to yield an approximate posterior distribution P(H|D) (later on I will come back to the reservation about the posterior being ‘approximate’). In perceptual learning, the parameters of the generative model are adjusted to more effectively minimize long-term, average prediction error. One can think of the statistical structure of the sensory input in terms of a set of nested patterns. Some of those patterns unfold rapidly (e.g. the arrangement of edges may change with every saccadic eye movement), while others are more invariant (e.g. ones revealing a stable 3D shape of an object), and some may go quite a bit beyond what can be immediately perceived (e.g. slight changes in average illumination due to changing seasons). Reflecting this nested structure, the generative model itself is thought of as hierarchical. Lower levels of the model predict the flow of rapidly changing sensory patterns, while higher levels predict more invariant, ‘abstract’ regularities. In this hierarchical setting, the job of predicting sensory input is parceled into a series of local subproblems, whereby each level L exclusively predicts the activity at the directly adjacent level L-1, and serves as input to be predicted at level L+1. In effect, this means that when brain uses the model at a given level to predict the activity at a level below, it is using a more abstract description of the causes of sensory input to predict causes at a less abstract level of description. The updating of hypotheses is regulated by estimating the relative precisions of prior knowledge and sensory samples. Precision here is understood formally as inversely proportional to the variance of a probability distribution. In the present setting, if the sensory input is estimated to be less precise than prior expectations, then perceptual inference proceeds conservatively, i.e. it is less perturbed by the prediction error (when trying to find your way in a heavy fog, the retinal input may prove close to useless). If the opposite is the case, the inference puts relatively more weight on sensory evidence, i.e. the prediction error (e.g. when trying to make sense of a completely new situation for which one has no prior knowledge). It is precision estimation that accounts, on this story, for attention. Allocation of attention is explained in terms of precisionbased weighting of prediction errors. This theoretical toolkit can also be employed to account for motor control. PP builds on traditional predictive theories of motor control (Grush 2004; Pickering and Clark 2014) by claiming that action just is a way of minimizing the prediction error (Friston 2010). Roughly, movement amounts to intervening on the world to induce sensory data that conform to some prior hypothesis about one’s action (e.g. ‘I am now reaching for a piece of pizza’) or a sequence of actions (also called ‘policy’, e.g. ‘I am faithful to my diet’). If successful, acting results in minimizing the prediction error. Motor control thus construed relies on a kind of self-fulfilling prophecy, where to initiate movement, the brain needs to decrease the estimated precision of current proprioceptive prediction error (regardless of its actual precision) and increase the estimated precision of the prior which is to be actualized by action (and which is false at the onset of movement). Predictive processing (PP), then, is the view that perception, attention and action all result from using a hierarchical generative model to generate predictions and minimize their precision-weighted errors.

129

Paweł Gładziejewski

2.2  Intellectualist PP The preceding description of PP made heavy use of the ‘intellectualist’ vision of cognition as consisting of building a model to perform inferences about the world hidden beyond the veil of sensory states. How literal should we treat this description to be? Is the brain really constructing models to infer the world? Or is this sort of talk merely a heuristically useful (albeit perhaps in many respects misleading) gloss on a theory that, under closer scrutiny, is much more in line with embodied/enactive/embedded views of cognition? This is where the philosophical literature on PP breaks down into two seemingly mutually inconsistent approaches, an ‘intellectualist’ and a ‘radical’ one.3 Space forbids me from discussing in detail the many nuanced ways in which those outlooks differ. I will restrict this general summary to two dimensions along which the distinction can be drawn. These dimensions have to do with PP’s commitment (1) to internal representations and (2) to the idea of perception and action being underpinned by inference. As will transpire later on, these two axes of the debate over PP can be connected to the implicit/explicit distinction, which applies to (explicit vs implicit) representations and to the (explicit vs implicit) rules that guide information-processing, including inferential rules. Regarding the commitment to representation, proponents of the intellectualist PP take the generative model to perform the function of an internal representation (Gładziejewski 2016; Kiefer and Hohwy 2017, 2019; Williams 2017). In particular, the generative model constitutes a representation grounded in exploitable structural similarity (an S-representation), akin to cartographic maps or scale models. This essentially means that: (1) the generative model encodes a hierarchical relational structure of hidden or latent variables (‘hypotheses’); (2) the capacity to minimize the prediction error depends on the degree to which the model’s relational structure (including its dynamics) maps onto the nested causal structure which produces sensory states. For example, the conditional dependencies between hidden variables may map onto the causal relations between respective entities in the world, thus contributing to the successful prediction of the sensory input.4 Like maps, scale models and other paradigmatic cases of representations grounded in structural resemblance, the generative model is explicit in that its constituents (i.e. neurally implemented hidden variables; see section 3.2) act as separable representational vehicles that bear intentional content in virtue of their placement in a larger relational structure.5 As for the commitment to inference, the proponents of intellectualist PP see it as the latest incarnation of a historical lineage of theories that construe perception as unconscious inference (Gregory 1980; Helmholtz 1855). The idea is that the predictive brain updates its representations in a way that conforms to the famous Bayes rule: P(H|D) = P(D|H) P(H)/P(D) Note that this equation partially corresponds to inverting a generative model (notice the numerator on the right-hand side) to compute the posterior P(H|D). However, exact Bayesian inference requires one to calculate the right-hand-side denominator P(D). Because the value of P(D) is conditional on all the hypotheses that could potentially explain data, Bayesian inference is computationally intractable for any hypothesis space of non-trivial size. Thus, the idea is that the brain approximates Bayesian inference. Here is an outline of how this may work. Think first of an exact Bayesian inference as an incremental process where, to use a catchphrase, yesterday’s posteriors become today’s priors. Assume that you start with some prior which is a normal (i.e. Gaussian, bell-shaped) probability distribution. Assume further that the sensory samples actually obtained are best explained by a hypothesis which is also defined as 130

Predictive Processing, Implicit and Explicit

a normal distribution (this is the likelihood). Making the assumption that we are dealing with normal distributions is mathematically crucial here, as distributions of this form can be fully described in terms of their mean and variance, which simplifies calculations. The distance between the means of both distributions (i.e. the prior and the likelihood) is the prediction error. A posterior is formed that updates the prior in light of the prediction error. The posterior turns into a prior in the next iteration of inference, and then a new posterior is formed in light of consecutive prediction error. Over time, this process of incrementally learning and revising your priors should reduce the average prediction error. In PP, the brain is thought to implement an algorithm that indirectly maximizes the posterior probabilities of hypotheses solely by gradually minimizing the prediction error. That is achieved by way of applying a gradient descent on the prediction error, whereby the system ‘tinkers’ with the model to optimize its ability to minimize the error, eventually converging on a true posterior mean. This procedure will tend to produce results that reliably correspond to what exact Bayesian inference would yield. As Hohwy puts it, the crux is “to turn around the observation that inference leads to less prediction error, and say that, if a system is somehow able to continuously minimize prediction error, then the system will approximate Bayesian inference” (Hohwy 2020: 211).

2.3  Radical PP A common thread unifying ‘radical’ approaches to PP consists in the denial that representations or inferences are involved in PP, or at least in proposing a substantially deflated reading of those commitments. However, beyond that, the radical camp is much more internally theoretically diverse. There are two general ways in which the radical view is usually unpacked in the literature. One relies on interpreting the commitments of PP through the lens of Karl Friston’s Free Energy Principle (FEP; for useful discussions of FEP, see Friston 2010; Hipolito 2019; Kirchoff, Parr, et al. 2018). FEP is a mathematically sophisticated way of understanding self-organization in statistical terms and it applies to any system that manages to keep itself in a non-equilibrium steady state (i.e. which avoids dispersal or keeps itself separate from its environment). From this broad and abstract perspective, PP can be seen as a particular realization or ‘implementation’ of the FEP. Some authors have forcefully argued that the FEP-centric way of looking at things largely reconfigures how PP itself should be understood, namely as an enactive, nonrepresentational theory (Allen and Friston 2018; Bruineberg et al. 2016; Ramstead, Kirchoff, et al. 2020; however, see also more representation-friendly recent work on FEP in: Ramstead, Friston, et al. 2020). For the present purposes, however, I want to set the FEP aside. This is partly due to limitations of space, but also because the working assumption of this chapter that PP can stand on its own as an account of cognitive architecture, in a broad sense of being a theory of an overarching information processing strategy that underlies (much of) cognition. PP presumably applies to only some self-organizing systems, namely those equipped with mechanisms whose causal architecture realizes the computations postulated by the theory (Miłkowski 2013; Piccinini 2015). PP thus construed can be held in a way that is agnostic with respect to its exact connection to the FEP (although see Hohwy 2021). It is the second sort of arguments for the radical reading of PP that I do want to focus on here. These arguments purport to establish that the Bayesian, representation-centric rendering of the theory misconstrues the actual workings of PP-based systems. When it comes to the representational commitment, authors that subscribe to the radical reading deny that any of the explanatory posits of PP, including the generative model, 131

Paweł Gładziejewski

play genuinely representational roles in the cognitive system. Kirchhoff and Robertson (2018) argue that, formally, minimizing the average prediction error is equivalent to maximizing mutual information between the internal states of the system and the states of the environment (mutual information is a measure of how well the generative model fares at reducing uncertainty about the states of the world). But this simply means that the internal states come to reliably covary or carry Shannon information regarding distal causes of sensory input. Now, most philosophers agree that carrying Shannon information is not sufficient for being a content-bearing, representational state (although see Isaac 2019). The functional profile of information carriers in cognitive theories often boils down to mere causal mediation (Hutto and Myin 2013; Ramsey 2007). On this view, then, the function of the generative model is too causal-mediator-like to count as representational (see also Hutto 2018; Orlandi 2016, 2018). Furthermore, this stance may be only apparently inconsistent with the fact that scientists and philosophers who subscribe to PP routinely employ content-involving vocabulary invoking ‘hypotheses’ entertained by the brain. Such contentful states may be interpreted in a deflationary way, as purely instrumental or fictional (i.e. not literally true) posits that help researchers make sense of how the internal dynamics of the cognitive system relate to the external environment (Downey 2018). As for the commitment to inference, proponents of radical PP argue that despite Bayesian appearances, the framework does not postulate inferences in the full-bloodied sense of rational transitions between contentful states. This point has been very clearly fleshed out by Nico Orlandi (2016, 2018; see also Anderson 2017). Ever since Helmholtz (1855), the inferentialist view of perception is motivated by the ‘ambiguity of the senses’ argument, whereby the brain supposedly needs to rely on prior knowledge to decide between numerous possible explanations of the sensory input. But perhaps the sensory ambiguity is oversold here. Orlandi (2016, 2018) points to empirical research on natural scene statistics, which aims to discover recurring statistical patterns in images produced by natural environments (Geisler 2008). The proposal is that this research demonstrates that the sensory input is much less underdetermined by environmental causes than is often assumed. Thus, the structure of the input signal contains enough information to significantly reduce the uncertainty about its distal causes. Rather than search through an enormous hypothesis space (even using approximate shortcuts), the brain may simply rely on the sensory signal itself, assuming that the sensory/perceptual mechanisms are appropriately sensitive or attuned to the statistics of natural scenes. This approach does not have to deny that ‘priors’, in some sense of the word, may play a role in disambiguating the input if needed. But priors may be realized as mere processing biases (Orlandi 2016), which nudge the feedforward processing of signals towards ecologically valid interpretations (where ‘interpretation’ is read nonrepresentationally, e.g. as attunement to affordances). It would be superfluous and unjustified to consider such ecologically tuned processing inferential.

3.  Explicit and Implicit Representation in Predictive Processing 3.1  Implicit ‘Representations’ in Predictive Processing Let us now zoom in on the representational commitments of PP. The aim here is to argue that there are two quite different (although related) types of information-processing mechanisms that come under the common umbrella of ‘PP’. One type of mechanism merely implicitly ‘predicts’ sensory patterns and, overall, fits the radical reading of PP. The other type of mechanism relies on an explicit internal model of the environment, and hence fits the reading of PP favored by the ‘intellectualists’. 132

Predictive Processing, Implicit and Explicit

My focus will be on PP as applied to perception. In this domain, PP overlaps with a family of algorithms traditionally dubbed ‘predictive coding’ (see Huang and Rao 2011; Spratling 2017). At heart, predictive coding is a computational strategy in which predictable patterns in some original signal are used to compress this signal by exclusively encoding the prediction errors, that is, those parts of the original signal that do not match the predictable patterns. Consider a seminal work on predictive coding that obtains at the very earliest stages of vision (Srinivasan et al. 1982; see also Huang and Rao 2011; Spratling 2017). Retinal ganglion cells encode a signal that is sent, through the optic nerve, for further processing upstream in the visual cortex. However, this signal itself is an output of information processing that takes place within the retina, starting at photoreceptors and then mediated by layers composed of bipolar, horizontal and amacrine cells. The synaptic connections between those layers are often set up in such a way that a ganglion cell will be triggered if the light intensity at the center of its receptive filed differs from the weighted sum of the intensities of ‘pixels’ surrounding the center. Crucially, this process is ecologically grounded in the statistics of natural scenes. Small patches of retinal images tend to be uniform in light intensity, which means that, usually, the center of the ganglion cell’s receptive field matches its surround. From this perspective, the activation of a ganglion cell signifies a prediction error between an ecologically grounded ‘prediction’ and the image sampled from the environment. Note that this canonical example of predictive coding underlying early vision is in some crucial respects different from how PP is usually portrayed in the literature (including the brief exposition I provided in section 2). The processing is purely feedforward (but also includes lateral inhibition, see Srinivasan et al. 1982) rather than bidirectional. The retina cannot be meaningfully described as estimating the external causes of the pattern to which it is (predictively) sensitive. The prediction errors encoded in activations of ganglion cells are not used to correct the internal hypotheses about the causes of input. What is at stake here is more accurately described in terms of the retina producing a sparse encoding of the raw input image by subtracting spatial (and temporal, see Srinivasan et al. 1982) redundancies (see also Barlow 1961).6 The resulting message is encoded in fewer bits and can be effectively ‘squeezed’ through the informational bottleneck of the optic nerve. Even in this simple case, representational notions may prove useful in making sense of what the retina is doing. One may say that the retinal processing ‘predicts’ that input images are locally uniform in terms of light intensity, or that the ganglion cells encode errors with respect to a ‘hypothesis’ or ‘prediction’ that the center of a receptive field matches the surround. What should we make of such intentional ascriptions? I propose that the best way to read them is by appeal to a notion of ‘implicit’ representation (Clapin 2002; Dennett 1983; Haugeland 1991; Ramsey 2007: Ch. 5; Ramsey: this volume). Although this notion is ambiguous and has historically meant different things for different authors, for the present purposes I interpret it in a way that is nicely illustrated by the famous Dennett’s classic example of a chess-playing computer program to which one attributes the desire to take the queen early (see also Ramsey, this volume). Although there is no localized, separable, causally active internal vehicle that bears this content, the overall dispositional pattern of the program’s behavior embodies the desire implicitly. To generalize, the hallmark of implicit representations in the present sense is the lack internal structures where separable items act as vehicles that encode intentional contents. Instead, on this notion, (implicit-) representational ascriptions are grounded in the fact that a given information-processing system is wired in a way that allows it to embody dispositions to behave or respond ‘as if ’ it (explicitly) represented the world to be certain way. In the context of PP in particular, the notion of implicit representation would cover cases where the cognitive machinery is set up so as to allow the organism to be attuned to or appropriately 133

Paweł Gładziejewski

responsive to relevant statistical regularities, without explicitly representing those regularities or the causal structures that subserve them (see Orlandi 2016). So, to get back to the retina example, the ‘knowledge’ or ‘predictions’ one may want to attribute to the retina are implicitly embodied in the capacity or disposition to turn raw light intensity distributions into sparse messages. This disposition is tuned to natural statistics of retinal images, and hence the retina implicitly embodies ‘knowledge’ or ‘prediction’ of those statistics. Of course, the retina has this disposition in virtue of its feedforward synaptic wiring. But it would be misleading to say that this wiring is explicitly encoding a model or a representation of how the signals are generated.7 A worry may be raised that the retina is a special case and that the example hardly generalizable beyond the very periphery of perceptual processing.8 However, it seems that the case is generalizable to a non-trivial degree, and that it is just one illustration of a larger recurring strategy where, due to the hard-wiring of the nervous system, purely feedforward processing implicitly ‘predicts’ the natural statistics of the input (for an extensive review, see Teufel and Fletcher 2020). Other examples may include the findings that in the primary visual cortex, neurons tuned to the cardinal orientations outnumber those tuned to oblique orientations, echoing natural orientation statistics (Li et al. 2003); or that neurons in the visual cortex tuned to particular object categories tend to have their receptive fields predictively biased towards locations usually occupied by the objects belonging to those categories (Kaiser, Quek, et al. 2019). Akins (1996) provides an intriguing case of more imperative (desire-like) ‘predictions’ of this kind, whereby the variable distribution of thermoreceptors on different parts of the body (e.g. the scalp vs the hands) implicitly embodies preferences regarding how cold/hot the organism can allow those parts to become. I submit that the understanding of representational states as merely implicit in feedforward processing biases fits the radical view of PP much better than the intellectualist one. The line between implicit representationalism and nonrepresentationalism is thin. We may treat intentional talk in such contexts as aiming to highlight how the internal processing relates to the statistics of natural scenes. For example, the center-surround antagonism of the retinal ganglion cells makes sense once you consider what those cells are doing in the context of natural image statistics. However, this sort of connection to the environment is arguably not literally representational (Downey 2018), and may be more parsimoniously cashed out in terms of ecological embeddedness of perceptual/sensory mechanisms (Orlandi 2016, 2018). Relatedly, it has been forcefully argued that the notion of implicit ‘representation’ does not even pick out representations in an explanatorily valuable sense of the term (Ramsey 2007: Ch. 5; see also Ramsey, Chapter 1, this volume), which, in line with the radical view, invites a purely nonrepresentational reading.

3.2  Predictive Processing and Explicit Representations The way PP is usually understood in the literature today originates from a significant extension of vanilla predictive coding, put forward by Rao and Ballard (1999), and subsequently refined by Friston (2005; see also a precursor to this approach in Dayan, Hinton et al. 1995; Hinton 2007). I want to argue that this development constitutes a shift from implicit PP to a version that relies on an explicit model of the environment. Consider an extremely simple, informal illustration of this extended predictive coding, which I  adopt from Bogacz (2017), who in turn draws on the model proposed by Friston (2005). Imagine a simple organism trying to perceptually estimate the value of some hidden environmental variable, V. Assume that V is the size of an item of food, which the organism needs to estimate while only having direct access to a sensory signal, s (corresponding to, say, a 134

Predictive Processing, Implicit and Explicit

state of a light receptor). This signal depends causally on V, such that s is a function of V, g(V). Suppose further that the organism is equipped with a prior expectation Vp about what the most likely size of a food item is before obtaining any input. Given some technical details which I will leave out here, the job faced by the organism is then equivalent to computing P(V|s), that is, a posterior distribution over V, given s. Importantly, in this model, instead of establishing the whole posterior distribution, the organism attempts to estimate a determinate value φ, which encodes the most likely value of V, given the current sensory signal, s (see Figure 9.1). The idea is that the system may start with a random guess, and then engage in a recursive trial-and-error procedure (gradient descent) to gradually bring φ closer to the mean of the posterior distribution. This process relies on minimizing the prediction error. The estimate φ generates an inhibitory prediction signal that is a function of φ, g(φ). This prediction is propagated downwards and compared to s, giving rise to a prediction error εs, which measures the difference between s and the prediction. The error is weighted according to the variance Σs of the sensory signal, and propagated up the hierarchy to allow revision of φ. Simultaneously, the updating of φ is driven by a prior estimate Vp. Again, this is realized by minimizing the prediction error εp, i.e. the difference between φ and Vp, weighted by the variance of the prior, Σp (see Figure 9.1). The upshot is that φ is constrained both by the incoming sensory signals and the prior, weighted by their respective reliabilities. At the implementational level, the variables s, φ, εs and εp correspond to occurently active nodes and may be implemented in the activity of individual neurons or neural populations. The prior Vp acts as a model parameter and is encoded in the strength of a synaptic feedback connection between a ‘tonically active’ neuron and the node encoding φ (Bogacz 2017; see also Friston 2005). The values of Σp, Σs similarly act as parameters and are implemented in selfinhibitory synaptic connections of the error nodes. Now, what this sort of scheme and the retinal predictive coding have in common is that they both rely on the idea that predictions (in a broad sense) are used to compute prediction errors to efficiently encode data.9 However, there are crucial differences as well. The model just discussed involves a bidirectional information flow that includes feedback projections. Under this regime, predictions correspond to descending inhibitory signals. The ascending prediction error signals are not the endpoint of processing, but allow the correction of internal estimates. Most importantly for the present purposes, on the scheme discussed earlier, the descending predictions are based on the estimates of the cause(s) of the input signal. The prior assumptions about the causal structure generating the sensory input are encoded in Vp.10 The ‘best guess’ about the current state of the environment is encoded in φ. The relation between φ and the prediction signal g(φ) encodes the functional relations between the worldly causes and their

Sensory signal S

g( )

Posterior estimate

Prior estimate Vp

s

p

s

Vp

p

Figure 9.1 The bidirectional predictive coding scheme, adapted from Bogacz (2017). Arrows signify excitatory connections, and the lines ending with circles signify inhibitory connections. See main text for details.

135

Paweł Gładziejewski

sensory effects. Because prior assumptions about the most likely causes of the sensory input and about how these causes generate the input correspond to specifiable components of the neural architecture, we may say that the generative model is explicitly encoded. Furthermore, the idea that this model constitutes an explicit representation gets traction once we recognize that, functionally, it fits the ‘job description’ of a representation (Ramsey 2007). Like many things we pretheoretically regard as representations, it relies on structural similarity, allows learning through error correction, guides adaptive action (in realistic models which encompass both perception and action) and can potentially underpin more off-line types of cognition (see Gładziejewski 2016 for an extensive defense of this view).11 As should be clear from the preceding discussion, the idea that PP postulates an explicit internal representation of the environment is very much in line with the intellectualist reading of PP. Crucially, however, the point here is not to argue whether the cognitive system engages in PP by implicitly ‘predicting’ natural scene statistics (rehabilitating the radical view of PP) or by storing an explicit model of the environment (rehabilitating intellectualism). Instead, it is perfectly sound to think that the cognitive system could rely on both strategies. Not only could these strategies peacefully co-exist, but they may functionally complement each other in interesting ways. In fact, some empirical results and theoretical considerations give leverage to the idea that intellectualism and radicalism about PP can be married along those lines. In visual perception, a rapid (150ms) feedforward phase of processing is followed by a second phase, carried out using feedback projections (Kreiman and Serre 2020). From the PP perspective, the initial feedforward sweep could correspond to ‘implicit’ predictions, and the onset of the ‘feedback’ phase would demarcate the point at which explicit generative model(s) kicks in. This would also suggest a functional division of labor, as we know that while the initial phase enables quick object recognition, the feedback phase is required to interpret the meaning of complex visual scenes by flexibly relating them to background knowledge (see Kreiman and Serre 2020 for an extensive discussion).12 Relatedly, it has been argued that the ‘predictions’ implicit in feed-forward processing track context-invariant regularities in the sensory input, and the job of mechanisms corresponding to (what I construe as) explicit PP is to flexibly track context-dependent sensory patterns (Teufel and Fletcher 2020; see also the distinction between ‘structural’ and ‘contextual’ priors in Seriès and Seitz 2013).13 Philosophically, this discussion sheds new light on the old idea – due to Ryle (1949/2000), and raised in the cognitive-scientific context by Dennett (1983) – that explicit knowledge-that requires a background of practical abilities (implicit ‘know-how’). The conciliatory view of PP vindicates this overall approach in the following way. Orlandi (2016) notes that a challenge to intellectualist PP lies in the sheer vastness of the space of all the prior hypotheses that the cognitive system could bring to bear when interpreting the sensory input. How does the system constrain this space to only consider a subset of (relevant) hypotheses? The present view suggests that much of the computational or epistemic load is distributed to implicit machinery. This way, the explicit machinery is freed from the need to learn and represent certain prior assumptions (that small patches of the retinal input tend to be uniform; that vertical and horizontal edges dominate visual scenes; that clouds are usually found in the upper side of the visual field; that it is more dangerous to expose one’s head to cold than it is to expose one’s hands, etc.). Instead, much of what is ‘known’ or ‘predicted’ about the environment is implicit in feedforward information-processing mechanisms. Speculatively, perhaps the abstract priors regarding space, time, or objecthood that structure perception are (at least in part) implicitly embodied in this way. If true, this would dovetail with the Kantian reading of the implicit/explicit distinction in cognitive science (Clapin 2002). 136

Predictive Processing, Implicit and Explicit

4.  Implicit Bayesian Inference in Predictive Processing Are PP-based cognitive systems inferential? What would it mean to say that they are? Here again I want to recommend a conciliatory approach, on which predictive cognitive systems are partially inferential. In particular, they are inferential to the degree to which they use explicit generative models. Intellectualists about PP get this part of the story right. However, the part of the story that encompasses computational strategies where feedforward processing implicitly embodies predictions is non-inferential, in line with claims made by the ‘radicals’. To put this discussion in context, the questions surrounding the status of Bayesian inference in cognitive science reach beyond PP. PP is just one incarnation of a larger trend of employing Bayesian probability theory to model cognitive phenomena (c.f. Griffiths et al. 2008). According to one influential criticism, Bayesian cognitive modeling confounds descriptive use of Bayesian mathematics with an explanatory use (Bowers and Davis 2012; Colombo and Seriès 2012; Jones and Love 2011). Bayesian models typically aim to establish that human performance on some cognitive task overlaps with an optimal Bayesian solution. However, the mere fact that subject’s response patterns can be captured using Bayesian formalism does not any way guarantee that the internal causal mechanisms responsible for those patterns perform or implement Bayesian inference. We may situate this against an even wider philosophical background, as the issue at hand relates to a distinction between behavior that conforms to a rule and behavior that is guided by a rule (Davies 2015). Quine (1970) once used this distinction to argue against Chomskyan linguistics. He pointed out that the fact that linguistic behavior conforms to a set of syntactic rules (to which a subject has no conscious access) does not imply that these rules genuinely guide this behavior. In fact, one might posit a completely different set of unconscious rules which correspond to the very same set of behaviors, and, as long as we rely on behavioral data alone (which was Quine’s assumption), there will be no principled way to decide which rules guide the behavior. The aforementioned criticism of modern Bayesian models in cognitive science boils down to raising a similar point: the fact that task performance conforms to the Bayes rule does not imply that it is guided by it. A natural way out of this problem for anyone subscribing to a Bayesian-inferential view of cognition (including those who buy into intellectualist PP) would consist in trying to spell out conditions under which a given system is guided by the Bayes rule. Arguably, these conditions should have to do with whether the Bayesian inferences postulated by a cognitive model map onto or correspond to causal transactions within a cognitive mechanism responsible for a given explanandum phenomenon (Miłkowski 2013; Piccinini 2015). This, however, is easier said than done, as one needs to steer the account so that it avoids specifying, on the one hand, conditions that are too restrictive or, on the other hand, excessively liberal ones. An exemplary approach belonging to the former category would require the Bayes rule to be explicitly (even if without conscious awareness of the subject) encoded or stored in the mechanism, and have each particular instance of inference ‘consult’ and the follow this rule. Because exact Bayesian inference is intractable, such an account would be utterly naïve. An example of the opposite, too-relaxed sort of approach would consist in simply equating predictive coding, in all its forms (see Spratling 2017), with Bayesian inference. A position of this kind would treat even the predictive coding scheme employed by the retina as an instance of Bayesian inference. This would be gratuitous to say the least, as retinal processing hardly aims to estimate the most likely hypotheses about the causes of sensory stimulation. Its function, to repeat, is to produce a sparse encoding of the raw image registered by photoreceptors. In fact, predictive coding is best seen as an ‘algorithmic motif ’, which may serve several different 137

Paweł Gładziejewski

computational goals, Bayesian inference being just one (Aitchison and Lengyel 2017). To generalize this point, whenever the cognitive system ‘predicts’ states of affairs simply by virtue of the implicit attunement of feedforward processing to natural scene statistics, the usage of hefty cognitivist notions like ‘inference’ will be at least deeply problematic. Proponents of radical PP are right in this respect. How about the cases like the one discussed in section 3.2, where the prediction error minimization relies on an explicit model of the environment? This is where an argument can be made that, in line with intellectualist PP, genuine (approximate) Bayesian inferences sometimes guide the cognitive system. Of course, an explicit inscription of the Bayes rule is nowhere to be found in Figure 9.1. However, the proposal here is that although the representations at play are explicitly encoded, the rule of processing (i.e. the Bayes rule) is itself implicit in the causal transitions between the representational states. In other words, although the updating of representational states merely conforms to Bayes rule (without explicitly following it), the states undergoing the updating are states of components of an internal causal mechanism, and so there is a clear sense in which the rule guides perception, rather than merely describes it. That causal transitions conform to Bayes rule should already be clear from the preceding discussion in sections 2.2 and 3.2. Two considerations are crucial. First, the computational goal of a mechanism like the one present in Figure  9.1 corresponds to the computational objective of Bayesian (perceptual) inference, i.e. it consists in estimating the most likely causes of the sensory signal. Second, as long as the output posterior estimate (φ) relies on prediction error minimization, with the prior estimate (Vp) and the current sensory samples (s) serving as inputs, the result will tend to approximate the true posterior distribution that exact Bayesian inference would yield.14 Before I  conclude, I  want to briefly sketch out another, perhaps more revealing sense in which implicit inference is at play in PP. What I have in mind here is inspired by a solution proposed by Evans (1981) to the aforementioned problem raised by Quine. Evans put forward an account of non-behavioral evidence that could determine whether the behavior of a given system merely conforms to a rule or is guided by it. The account relies on the notion of ‘causal systematicity’. Put simply, according to this view, the rule is guiding the system as long the behavioral manifestations of said rule have a common causal origin in the system. Take an illustrative example (Davies 1995, 2015). Suppose that a person’s behavior conforms to a rule according to which if a word starts with ‘b’, then its pronunciation starts with the sound /B/. We intend to establish whether the person’s behavior is also guided by this rule. To achieve this, the proposal goes, we need to verify whether particular behaviors that conform to this rule are causally mediated by a single state of the subject. That is, under the hypothesis that the /b/-B rule guides the system, the particular instantiations of behavior conforming to this rule count as manifestations of a complex disposition underpinned by a single internal causal state. If, however, we found out that that the person’s brain stores a long look-up table such that each instantiation of the rule-conforming behavior is generated by a different causal state, the claim that the person is guided by the ‘b’-/B/ rule would be falsified. Note now that in (explicitly representational) PP, the error-minimizing brain gradually uncovers the hidden causal structure that produces the sensory input (Clark 2013b; Friston 2005; Kiefer and Hohwy 2017; Hohwy 2013). Over time, through learning (or iterated perceptual inference), the raw stream of sensory stimulation is ‘unmixed’ into a generative model comprising latent variables that capture the complex, nonlinear interactions between the worldly causes of the sensory input. The structure of the model can be expressed as a Bayesian network (see Danks 2014 for an extensive discussion of cognitive representations construed as Bayesian networks). That is, the model can be regarded as a network comprising (1) nodes that encode the values of latent variables,15 and (2) edges which encode dependencies between 138

Predictive Processing, Implicit and Explicit

those variables. On this view, each single node – simply in virtue of its position within a larger relational structure – systematically encodes a whole range of inferential connections16 between itself and other nodes (see also Clark 2013a; Kiefer 2017, 2019; Williams 2018). In addition, the network is productive in that, in principle, the system could endogenously tinker with the values of certain variables to run counterfactual simulations that do not have to correspond to any previous perception (see also Clark 2013a; Kiefer and Hohwy 2017; Williams 2020). What this amounts to is that whenever the cognitive system brings a given latent variable to bear on a cognitive process – be it online in perceptual inference or in counterfactual imagery – this process will be mediated by a single node in the network. This cognitive story becomes causal once we allow the generative model to be explicitly encoded in the system so that the nodes correspond to internal causal states or components of a causal mechanism (although the exact implementational details may turn out quite messy). Hence, we may expect the inferential transitions attributable to the system to map onto causal transitions in the system (see also Kiefer 2017, 2019). Seen this way, the predictive cognitive system – to the extent that it encodes an explicit generative model of the environment – embodies causal systematicity that allows it to count as inferentially guided rather than merely inferentially describable.

5. Conclusions In this chapter, I set out to show that the implicit/explicit distinction in cognitive science – in particular, as regarding the nature of representation and the rules of information processing – can illuminate our understanding of predictive processing. I proposed that depending on the details, PP may come in two flavors: (1) explicitly representational and implicitly inferential, or (2) implicitly representational and non-inferential. This distinction at least partially maps onto the ‘intellectualist’ and ‘radical’ readings of PP, respectively. The upshot is that ‘intellectuals’ and ‘radicals’ may both get part of the story right, in that there is some theoretical and empirical traction to the claim that the brain uses both aforementioned strategies to deal with sensory uncertainty. This position also restores the classical idea that explicit cognition is built on top of implicit know-how. In the present, PP-based guise, this means the computational strain on explicit, Bayesian mechanisms is reduced by the fact that those mechanisms process signals already ‘filtered’ by expectations implicitly embodied by feedforward processing biases.17

Related topics Chapters 1, 10, 11

Notes 1 Readers already familiar with PP may wish to skip this part and proceed straight to section 2.2. 2 For expository purposes, I  am considering a simple, non-hierarchical model with just one level of hypotheses that directly predict the flow of input. Things become more complex in hierarchical models, where predictions that a hypothesis at a given level makes about sensory data are mediated by a set of hypotheses at intermediate levels of the generative model (see main text). 3 This is not to deny that some attempts to peacefully marry the two interpretations have been made (see Allen and Friston 2018; Clark 2016; Constant et al. 2019; Dołęga 2017; Korbak 2021; Rutar et al. 2022). 4 It may be noted that similarity does not have to constitute a strict isomorphism between the two structures. Partial mapping will suffice as long as it is exploitable for the system (see Gładziejewski

139

Paweł Gładziejewski and Miłkowski 2017). Moreover, this view of representation is largely sympathetic to embodied and action-centric theories of cognition, as it essentially treats representations as guides of adaptive action (Gładziejewski 2016; Williams 2017). 5 Note that there is a particular sense in which structural representations may also represent implicitly. Namely, some relations which are not explicitly encoded in the structure of the representational vehicle may be implicit in this structure in that it is possible to infer them from the relations that are explicitly represented (for discussions of ‘implicit’ representation understood in terms of representation that is derivable from an explicit representation, see Dennett 1983; Ramsey 2007, this volume). 6 This is not to suggest that there is a dichotomy between producing a sparse encoding of a signal and estimating or modeling its causes. Explicit generative models of causal structures that generate input (to be discussed in section 3.2) also encode compressed information about sensory data. The point here is that in the case of the retina, creating a sparse code of the input signal is the sole computational goal which is achieved without modeling the causes of the signal. 7 This claim is consistent with the idea that there is also an entirely non-semantic meaning of ‘prediction’ (as a matching between two non-semantic, physical signals) on which the activities of neighboring ‘surround’ bipolar cells in the retina predict the activity of ‘center’ cells, and the activity of the output ganglion cell is determined by the degree to which this prediction is ‘accurate’ (see also Anderson and Chemero 2013; Cao 2020). 8 I am indebted to Alex Kiefer for raising this point. 9 In cases like the one just discussed, the internal model of how the sensory data are generated (see main text) constitutes an informationally parsimonious encoding of the data (Huang and Rao 2011). 10 The example assumes that a single variable is sufficient to describe the environment. In more realistic scenarios, prior knowledge would encompass multiple interrelated (also hierarchically) hidden variables mirroring, to some biologically affordable degree, the complex causal dependencies in the environment. In those more sophisticated cases the idea that (explicit) PP involves structural representations becomes clearer. 11 Note that in our example, φ will covary with or carry Shannon information about the sizes of food items. It would be manifestly wrong, however, to draw from this a conclusion that the whole set-up boils down to using detectors or causal mediators. In explicit PP, the covariance is established through predictive, error-corrected modeling of the environment. More generally, what is decisive in figuring out whether representations are involved in a given story is the mechanics of how the informational relationship is established, for example, through modeling vs detector-like causal dependence (see Kiefer and Hohwy 2017). 12 It needs to be stressed that, for now, this is a speculative interpretation of empirical results that requires further research. It may be the case that the distinction between the feedforward sweep and the feedback phase does not map nicely onto the distinction between implicit predictions/representations (embodied in feedforward processing biases) and the explicit generative model. I am grateful to Alex Kiefer for pointing this out. 13 Note, however, that the invariant-and-inflexible vs contextual-and-flexible division may be oversimplified, as there are demonstrations of flexible adaptation to changing input statistics taking place even at the earliest, retinal stage of visual processing (Hosoya et al. 2005). 14 There are other relevant considerations that I leave out here for the sake of space. These have to do with the fact that representational transitions in mechanisms of this kind should naturally exhibit truthpreservation and tend to maximize coherence among representations, which are traditional hallmarks of genuinely inferential processes (see Kiefer 2017 for an extensive discussion). 15 Importantly, the contents encoded by the nodes do not have to correspond to personal-level conceptual or propositional contents. Because the nodes in Bayesian networks are ‘atomic’ and lack a constituent syntactic structure, doubt has been cast over whether these networks (and PP in general) can capture the ‘rich’ systematicity of human thought (Williams 2018; see Kiefer 2019 for an attempt to answer this challenge). 16 Note, however, that the claim is not that the system performs Bayesian inference simply in virtue of the fact that it encodes a Bayesian network (Bayesian networks do not have to undergo Bayesian updating, see Danks 2014). 17 I thank Alex Kiefer for his insightful critical remarks on the previous version of this chapter. My work on this chapter was kindly supported by the National Science Center in Poland (grant no. 2019/33/B/ HS1/00677).

140

Predictive Processing, Implicit and Explicit

References Allen, M., and Friston, K. J. 2018. “From cognitivism to autopoiesis: towards a computational framework for the embodied mind”. Synthese, 195: 2459–2482. Aitchison, L., and Lengyel, M. 2017. “With or without you: predictive coding and Bayesian inference in the brain”. Current Opinion in Neurobiology, 46: 219–227. Akins, K. 1996. “Of sensory systems and the ‘aboutness’ of mental states”. The Journal of Philosophy, 93: 337–372. Anderson, M. L. 2017. “Of Bayes and bullets”. In T. Metzinger and W. Wiese, eds., Philosophy and predictive processing. MIND Group. https://predictive-mind.net/papers/of-bayes-and-bullets. Anderson, M. L., and Chemero, T. 2013. “The problem with brain GUTs: conflation of different senses of ‘prediction’ threatens metaphysical disaster”. Behavioral and Brain Sciences, 36: 204–205. Barlow, H. B. 1961. “Possible principles underlying the transformations of sensory messages”. In W. A. Rosenblith, ed., Sensory communication. Cambridge: MIT Press: 217–234. Bogacz, R. 2017. “A tutorial on the free-energy framework for modelling perception and learning”. Journal of Mathematical Psychology, 76, Part B: 198–211. Bowers, J. S., and Davis, C. J. 2012. “Bayesian just-so stories in psychology and neuroscience”. Psychological Bulletin, 138: 389–414. Bruineberg, J., Kiverstein, J., and Rietveld, E. 2016. “The anticipating brain is not a scientist: the freeenergy principle from an ecological-enactive perspective”. Synthese, 195: 2417–2444. Cao, R. 2020. “New labels for old ideas: predictive processing and the interpretation of neural signals”. Review of Philosophy and Psychology, 11: 517–546. Clapin, H. 2002. “Tacit representation in functional architecture”. In H. Clapin, ed., The philosophy of mental representation. Oxford: Oxford University Press: 295–312. Clark, A. 2013a. “Expecting the world: perception, prediction and the origins of human knowledge”. The Journal of Philosophy, 110: 469–496. Clark, A. 2013b. “Whatever next? predictive brains, situated agents and the future of cognitive science”. Behavioral and Brain Sciences, 36: 181–204. Clark, A. 2016. Surfing uncertainty: prediction, action, and the embodied mind. Oxford: Oxford University Press. Colombo, M., and Seriès, P. 2012. “Bayes in the brain – On Bayesian modelling in neuroscience”. The British Journal for the Philosophy of Science, 63: 697–723. Constant, A., Clark, A., and Friston, K. J. 2019. Representation wars: enacting an armistice through active inference. Preprint available at: http://philsci-archive.pitt.edu/16125/ Danks, D. 2014. Unifying the mind: cognitive representations as graphical models. Cambridge, MA: The MIT Press. Davies, M. 1995. “Two notions of implicit rules”. Philosophical Perspectives, 9: 153–183. Davies, M. 2015. “Knowledge (explicit, implicit and tacit): philosophical aspects”. In J. D. Wright, ed., International encyclopedia of the social & behavioral sciences. Oxford: Elsevier Ltd: 74–90. Dayan, P., Hinton, G. E., Neal, R. M., and Zemel, R. S. 1995. “The Helmholtz machine”. Neural Computation, 7: 889–904. Dennett, D. 1983. “Styles of mental representation”. Proceedings of the Aristotelian Society, 83: 213–226. Dołęga, K. 2017. “Moderate predictive processing”. In T. Metzinger and W. Wiese, eds., Philosophy and predictive processing. https://predictive-mind.net/papers/moderate-predictive-processing. Downey, A. 2018. “Predictive processing and the representation wars: a victory for the eliminativist (via fictionalism)”. Synthese, 195: 5115–5139. Evans, G. 1981. “Semantic theory and tacit knowledge”. In S. Holtzman and C. Leich, eds., Wittgenstein: to follow a rule. London: Routledge and Kegan Paul: 118–137. Friston, K. J. 2005. “A theory of cortical responses”. Philosophical Transactions of the Royal Society B, 360: 815–836. Friston, K. J. 2010. “The free-energy principle: a unified brain theory?”. Nature Reviews Neuroscience, 11: 127–138. Geisler, W. S. 2008. “Visual perception and the statistical properties of natural scenes”. Annual Review of Psychology, 59: 167–192. Gładziejewski, P. 2016. “Predictive coding and representationalism”. Synthese, 193: 559–582. Gładziejewski, P., and Miłkowski, M. 2017. “Structural representations: causally relevant and different from detectors”. Biology & Philosophy, 32: 337–355.

141

Paweł Gładziejewski Gregory, R. L. 1980. “Perceptions as hypotheses”. Philosophical Transactions of the Royal Society B: Biological Sciences, 290: 181–197. Griffiths, T. L., Kemp, C., and Tenenbaum, J. B. 2008. “Bayesian models of cognition”. In R. Sun, ed.,  Cambridge handbook of computational cognitive modeling.  Cambridge: Cambridge University Press: 59–100. Grush, R. 2004. “The emulation theory of representation: motor control, imagery, and perception”. Behavioral and Brain Sciences, 27: 377–396. Haugeland, J. 1991. “Representational genera”. In W. Ramsey, S. Stich, and D. Rumelhart, eds., Philosophy and connectionist theory. Hillsdale, NJ: Lawrence Erlbaum: 61–89. Helmholtz, H. 1855. “Über das Sehen des Menschen”. In Vorträge und Reden von Hermann Helmholtz. 5th ed., Vol. 1. Braunschweig: F. Vieweg: 85–117. Hinton, G. E. 2007. “Learning multiple layers of representation”. Trends in Cognitive Sciences, 11: 428–434. Hipolito, I. 2019. “A simple theory of every ‘thing’ ”. Physics of Life Review, 31: 79–85. Hohwy, J. 2013. The predictive mind. Oxford: Oxford University Press. Hohwy, J. 2020. “New directions in predictive processing”. Mind and Language, 35: 209–223. Hohwy, J. 2021. “Self-supervision, normativity and the free energy principle”. Synthese, 199: 29–53. https://doi.org/10.1007/s11229-020-02622-2. Hosoya, T., Baccus, S. A., and Meister, M. 2005. “Dynamic predictive coding in the retina”. Nature, 436: 71–77. Huang, Y., and Rao, R. P. N. 2011. “Predictive coding”. Wiley Interdisciplinary Reviews: Cognitive Science, 2: 580–593. Hutto, D. 2018. “Getting into predictive processing’s great guessing game: bootstrap heaven or hell?”. Synthese, 195: 2445–2458. Hutto, D., and Myin, E. 2013. Radicalizing enactivism: basic minds without content. Cambridge, MA: The MIT Press. Isaac, A. M. 2019. “The semantics latent in Shannon information”. The British Journal of Philosophy of Science, 70: 103–125. Jones, M., and Love, B. 2011. “Bayesian fundamentalism or enlightenment? On the explanatory status and theoretical contributions of Bayesian models of cognition”. Behavioral and Brain Sciences, 34: 169–188. Kaiser, D., Quek, G. L., Cichy, R. M., Peelen, M. V. 2019. “Vision in a structured world”. Trends in Cognitive Sciences, 23: 672–685. Kiefer, A. 2017. “Literal perceptual inference”. In T. Metzinger and W. Wiese, eds., Philosophy and predictive processing. MIND Group. https://predictive-mind.net/papers/literal-perceptual-inference. Kiefer, A. 2019. A defense of pure connectionism. Unpublished dissertation. The Graduate Center, City University of New York. https://doi.org/10.13140/RG.2.2.18476.51842. Kiefer, A., and Hohwy, J. 2017. “Content and misrepresentation in hierarchical generative models”. Synthese, 195: 2387–2415. Kiefer, A., and Hohwy, J. 2019. “Representation in the prediction error minimization framework”. In S. Robins, J. Symons, and P. Calvo, eds., The Routledge companion to philosophy of psychology. London: Routledge: 384–409. Kirchoff, M. D., Parr, T., Palacios, E., Friston, K. J., and Kiverstein, J. 2018. “The Markov blankets of life: autonomy, active inference and the free energy principle”. Journal of the Royal Society Interface, 15: 20170792. Kirchhoff, M. D., and Robertson, I. 2018. “Enactivism and predictive processing: a non-representational view”. Philosophical Explorations, 21: 264–281. Korbak, T. 2021. “Computational enactivism under the free energy principle”. Synthese, 198: 2743–276. https://doi.org/10.1007/s11229-019-02243-4. Kreiman, G., and Serre, T. 2020. “Beyond the feedforward sweep: feedback computations in the visual cortex”. Annals of the New York Academy of Sciences, 1464: 222–241. Li, B., Peterson, M. R., and Freeman, R. D. 2003. “Oblique effect: a neural basis in the visual cortex”. Journal of Neurophysiology, 90: 204–217. Miłkowski, M. 2013. Explaining the computational mind. Cambridge, MA: The MIT Press. Orlandi, N. 2016. “Bayesian perception is ecological perception”. Philosophical Topics, 44: 255–278. Orlandi, N. 2018. “Predictive perceptual systems”. Synthese, 195: 2367–2385. Piccinini, G. 2015. Physical computation: a mechanistic account. Oxford: Oxford University Press. Pickering, M. J., and Clark, A. 2014. “Getting ahead: forward models and their role in cognitive architecture”. Trends in Cognitive Sciences, 18: 451–456.

142

Predictive Processing, Implicit and Explicit Quine, W. V. O. 1970. “Methodological reflections on current linguistic theory”. Synthese, 21: 386–398. Ramsey, W. 2007. Representation reconsidered. Cambridge: Cambridge University Press. Ramstead, M. J. D., Friston, K. J., and Hipolito, I. 2020. “Is the free-energy principle a formal theory of semantics? From variational density dynamics to neural and phenotypic representations”. Entropy, 22: 889. Ramstead, M. J. D., Kirchoff, M. D., and Friston, K. J. 2020. “A tale of two densities: active inference is enactive inference”. Adaptive Behavior, 28: 225–239. https://doi.org/10.1177/1059712319862774 Rao, R. P. N., and Ballard, D. H. 1999. “Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptive-field effects”. Nature Neuroscience, 2: 79–87. Rutar, D., Wiese, W., and Kwisthout, J. 2022. “From representations in predictive processing to degrees of representational features”. Minds and Machines, 32: 461–484. Ryle, G. 1949/2000. The concept of mind. Chicago: University of Chicago Press. Seriès, P., and Seitz, A. R. 2013. “Learning what to expect (in visual perception)”. Frontiers in Human Neuroscience, 7. https://doi.org/10.3389/fnhum.2013.00668 Spratling, M. W. 2017. “A review of predictive coding algorithms”. Brain and Cognition, 112: 92–97. Srinivasan, M. V., Laughlin, S. B., and Dubs, A. 1982. “Predictive coding: a fresh view of inhibition in the retina”. Proceedings of the Royal Society of London, B, 216: 427–459. Teufel, C., and Fletcher, P. 2020. “Forms of prediction in the nervous system”. Nature Reviews Neuroscience, 21: 231–242. Williams, D. 2017. “Predictive processing and the representation wars”. Minds and Machines, 28: 141–172. Williams, D. 2018. “Predictive coding and thought”. Synthese, 197: 749–775. Williams, D. 2020. “Imaginative constraints and generative models”. Australasian Journal of Philosophy. https://doi.org/10.1080/00048402.2020.1719523.

143

10 COGNITIVE PENETRATION AND IMPLICIT COGNITION Lucas Battich and Ophelia Deroy

1 Introduction In The Bluest Eye by Nobel Prize laureate Toni Morrison, the young black heroine, Pecola, keeps finding herself and her family physically unattractive, which creates terrible feelings of distress and insecurity. Later, she starts to understand that her impression comes from her having internalised the idea that African Americans could not be as beautiful as Europeans: You looked at them and wondered why they were so ugly; you looked closely and could not find the source. Then you realized that it came from conviction, their conviction. It was as though some mysterious all-knowing master had given each one a cloak of ugliness to wear, and they had each accepted it without question. Pecola’s plight illustrates the pernicious effects that implicit views can have on people. The racist views that the young heroine has come to accept, or at least internalise, shape not only her judgements and decisions, but her perception of herself and others. In Morrison’s novel, we are encouraged to think that if Pecola had not internalised the racist canon of beauty around her, she would be able to see herself in the mirror differently. Whether and how cases like this one occur in the real world remains an animated debate (see, e.g., Burnston 2017; Cermeño-Aínsa 2020). But almost all the cases that are discussed in the philosophical and scientific literature resemble Pecola’s case, in that the cognitive influence bypasses the perceiver’s awareness: people are not conscious that the way they see the world is shaped by how they already expect the world to look (section 2). In some specific cases, people are also not aware of what precisely the content of the influencing cognitive state is (section 3). The distinction between the process and the content of cognitive influence helps us distinguish between two sides of the epistemic risk introduced by cognitive penetration (section 4): one is the epistemic threat, which comes from the perceiver not being capable of detecting that her perception is shaped by prior beliefs, and the other is the epistemic fault, which comes from the perceiver failing to prevent problematic contents to bear on their perception, and subsequent beliefs and decisions. If the perceiver has no awareness of the content of the cognitive state shaping their perception, are they then free of epistemic fault? We conclude by highlighting some issues standing in the way of this epistemic absolution (section 5). 144

DOI: 10.4324/9781003014584-13

Cognitive Penetration and Implicit Cognition

2  Cognitive Influences Are Implicit Establishing whether cognitive states such as beliefs can influence perception or only the judgements based on perception, is a challenging experimental question. Nevertheless, the possibility remains: perception may be cognitively penetrated, if the contents of the higher cognitive states directly and causally affect low-level perceptual processing or perceptual experience, so that, had the higher-level contents been absent or different, the perception of basic aspects of the world would have been different (Siegel 2012; Stokes 2013). To count as cognitive penetration, the effects must concern how perceptual properties and objects are experienced. Canonically, these would be colours, shapes, sizes, slopes, brightness, loudness, weight, sweetness and so on. However, if one embraces a richer view of perceptual content, the affected properties can also be object-kinds, moral or aesthetic properties such as elegance, grace, balance and so on (Bergqvist and Cowan 2018). In Pecola’s case, we may suppose that both could be at stake: her internalised critical views could affect her perception of the shape, size, colour of her face, or it could affect how harmonious it looks to her. The first way in which cognitive penetration is implicit is therefore when the top-down process of influence of cognitive states on experience is involuntary and unconscious: the subject does not have any explicit awareness or understanding that their perception has been influenced by their cognitive states. This is typically assumed to be the case for most candidate instances of cognitive penetration. Awareness of such influence does not necessarily require awareness of all the processing steps through which perception has been influenced, only of that fact that some influence has occurred. This focus arises in part from the stipulation that for cognitive penetrations to occur, the causal connection between the influencing state and the influenced percept must be internal and direct (Stokes 2013). If you have a blurred perception and put on your glasses because you believe you will see better with them, your perception changes as a result of your belief. Here you are certainly aware that your belief that glasses will make you see things better has led you to put them on, and that, as a result, your experience has changed. But this is not a case of cognitive penetration. The direct cause of the change is something external (the glasses), and not your internal mental states. Attention can have the same mediating role to play as the glasses, and act as an intermediate between cognitive states and perception. If you are presented with an ambiguous duck-rabbit drawing and made to realise that the drawing can be seen in two ways, you can voluntarily intend to see the drawing as of a duck or a rabbit, by attending to either duck- or rabbit-like features. Again, you may be aware that your beliefs or intentions have led you to attend to the drawing differently and, as a result, have shaped your perceptual experience. But the cause of the change occurs indirectly, mediated by attention. Classically, if a cognitive state influences attention, and attention, in turn, affects experience, no cognitive penetration has occurred (Pylyshyn 1999). The case of attention, however, is different from cases involving an external mediator. While attention can be voluntarily allocated, it can also be involuntarily affected by desires, intentions, or cognitive expectations. If so, attention itself could be cognitively penetrated. In such cases, the effects of the cognitive states on perception may be indirect, but they raise the same epistemic worries as cases where cognitive penetration occurs directly on perception itself, without an attentional mediation. Whether this counts as a mechanism of cognitive penetration, or constitutes another categorical phenomenon remains debated (see Mole 2015; Marchi 2017, for discussion). The cases discussed so far suggest that cognitive penetration occurs without explicit awareness that some influence has occurred. You could still have reasons to believe that your perception 145

Lucas Battich and Ophelia Deroy

is highly susceptible to certain kinds of top-down influence (Lyons 2011), for instance if someone else warns you that this is the case or because you have the feeling that there is something suspect or impure about your perceptual experience. Because we are not talking about awareness of the precise mechanisms of influence, but only of its occurrence, such indirect warnings or inner suspicions are sufficient to show that cognitive influences are not necessarily implicit. Notably, it is in principle possible for the perceiver to gain access to the fact that their perception is being, or has been, influenced through second-order or metacognitive awareness. Metacognition provides information about how well certain cognitive and perceptual are performing in a given context or how different processes interact (e.g., Deroy et al. 2016), and could indicate whether cognitive penetration has occurred in a given case. In many cases, metacognitive awareness is manifested consciously, including explicit judgements that something is wrong or unreliable, for example, but in others, it may only come as a sense or feeling that things are not going smoothly (Proust 2010). Whether it leads to a judgement or a feeling, metacognitive reflection suggests that people can become reflectively aware of how sub-personal processes are running or interacting, and eventually of their perceptual processes relying on cognitive states. To test this, Travers et al. (2020) examined the race-lightness bias. Levin and Banaji (2006) found that faces with features typical of an African person appear darker than faces with typical features of a European person, even when both faces have equal luminance. Whether all instances of this bias occur because of cognitive penetrability of perception is debated (Firestone and Scholl 2016), but the results from Travers et al. (2020) suggest that theirs were partly under cognitive influence. Their main question was to test whether subjects had metacognitive access to the fact that their responses about luminance are being affected by facial morphology. In other words, would people report or feel less confident when their perception of brightness is strongly biased by the task-irrelevant feature of morphology, compared to cases when they don’t rely on such irrelevant cues? Results show that this was not the case: people were not able to detect that their percepts were biased when reflecting on their confidence. Opacity to metacognition seems to be here the sign that the fact that an influence has occurred (let alone the mechanism behind this influence) is implicit through and through. This metacognitive opacity stands in sharp contrast with decades of experiments in perception showing that people’s confidence ratings can accurately track whether their responses were correct or not when performing classic visual tasks (e.g. Song et al. 2011).

3  Not All Influencers Are Implicit The process of influence in cognitive penetration should be considered as implicit, as it operates without the agent’s will and awareness. But what about the influencing state? There seems to be no restrictions on whether the influencer state should be implicit or explicit, as long as the influence on perceptual content occurs, and qualifies as coming from a cognitive state. Suppose that you are searching for wild strawberries along the forest trail, and are rehearsing in your head how much you desire to find strawberries. Suppose that this desire causes you to see something red where, in fact, there is no more than shadowy green leaves. The influencing state here is explicit in the strongest sense of being (i) consciously manifested; (ii) accessible to introspection, report and rational examination; (iii) generated voluntarily; and (iv) under voluntary control (e.g., by deciding to stop desiring strawberries, because you think you found enough already). Not all candidate cases of cognitive penetration will satisfy these four conditions. Take the case where your belief that most bananas are yellow influences your current colour perception, and makes you see bananas as more yellow than they really are (assuming this is the case; Hansen et al. 2006, but see Deroy 2013, and Valenti and Firestone 2019, for conceptual and empirical 146

Cognitive Penetration and Implicit Cognition

objections, respectively). You are not consciously aware or deliberating about that belief every time you see a banana. You may also not be able to change this belief at will, by stopping to believe that bananas are yellow. Nevertheless, you can report and bring the belief that most bananas are yellow into consciousness, satisfying at least the condition (ii). As pointed out by Dummett (1991), an informational state is explicit when that state allows for the possibility of eliciting a verbal statement about it when prompted. Even when the informational state is not actually verbalised, the mere possibility of being able to do so suffices to call it explicit (see Davies 2015 for an overview). On this dispositional reading, explicit sources at stake in cognitive penetration are those of which the agent can gain direct access to, given her current mental make-up, and without acquiring any new external information. Conversely, implicit influences and sources are such that the agent cannot be aware of, or gain direct access to, the implicit content that influences her perception, given her current state. For instance, in the case of Pecola, the idea that black people cannot be beautiful is not something she (i) is currently conscious of when she looks at herself in the mirror, or (ii) could report on, and examine rationally at this time, unless she acquires new information. Neither is it something she has (iii) chosen or (iv) could change at will. The state that influences Pecola’s perception then fully qualifies as implicit. One question is whether the way this idea has been internalised still counts as cognitive enough to qualify as a cognitive penetration. If the implicit states influencing perception are akin to beliefs (Carruthers 2018), then there is no issue in including them among cases of cognitive penetration: the representation distorting Pecola’s perception could be the generic proposition that “Black isn’t beautiful”. Implicit attitudes and biases could also differ from beliefs, and consist of non-propositional contents or associations. This may not be an issue as long as those states still count as cognitive. However, if the influence on perception comes from a non-conceptual source, such as an affective state or another perceptual state, then the influence would not count as cognitive. Some biases in behaviour may be thus explained if, for example, the mere perception of a Black face triggers an affective response that biases behaviour (Azevedo et al. 2017), or if the visual perception of mouth movements influences which phoneme is heard (McGurk and MacDonald 1976). Determining whether the influencing states or associations are “cognitive enough” to count as cognitive penetration is increasingly at odds with current views where there is no sharp distinction between cognition and perception (Newen and Vetter 2017). Nevertheless, the distinction could be maintained as a matter of degree by determining whether the effect is responsive to other clearly cognitive states, such as explicit beliefs or intentions (Deroy 2019).

4  Implicit Cognitive Influence Pose Epistemic Threats, but Not Equally Epistemic Blame One key philosophical motivation to focus on cases of cognitive penetration where the influence is not accessible to the subject is because they pose specific epistemic challenges (e.g., Lyons 2011). Perceptual experience should provide epistemic justification for our beliefs, intentions and desires. Barring extreme scepticism, what we believe about the external world is grounded on what we see, hear, touch, smell and taste. I perceive the banana in front of me to be yellow, and that perceptual experience makes it reasonable for me to believe that the banana is yellow: It is because perceptual experience has the phenomenal character of confronting one with objects and properties in the world around me that it justifies forming beliefs about those objects and properties. (Smithies 2014: 103) 147

Lucas Battich and Ophelia Deroy

Perceptual experiences thus influence cognitive states such as beliefs, which in turn can influence what we desire, or decide to do. If perceptual experiences are themselves influenced by cognitive states, their justificatory role is in part jeopardised. If my previous belief that most bananas are yellow determines my current perceptual experience of the banana in front of me as being yellow, then it becomes suspect to justify my current belief regarding the colour of the banana on that perceptual experience. Cognitive penetration introduces a circular justificatory structure (Siegel 2017). It does not seem epistemically rational to rely on our perceptual experience to justify or increase the credence in our pre-existing beliefs, if these experiences are already influenced by the very beliefs they are meant to justify. Arguably, not all cases of cognitive penetrability have this circular structure. When my perception is modulated by desire, so that I see something because I want it to be true, then I have a cognitively penetrated experience, albeit no circular one (Lyons 2011). A more general problem is that some forms of cognitive penetration diminish the reliability of perception, although other forms could increase it. A different, though related, epistemic worry is that, if cognitive penetration occurs, perception conforms more to the penetrating cognitive state than to the evidential data provided by the environment (Raftopoulos 2019; but see Burnston 2017). Adopting Smithies’ preceding formulation, a cognitively penetrated perceptual experience no longer confronts me with objects and properties in the world around me as they are, which threatens its justificatory role in forming accurate beliefs about those objects and properties. The implicit or explicit character of both the influencing state and the process of influence allows us to better distinguish between two different epistemic issues. The first one is the general epistemic threat that comes from the fact that the agent is not explicitly aware that her perception is influenced. The second is the epistemic fault of the agent when she becomes aware of the influencing content and/or the process of influence on perception, but fails to counteract this influence. Arguably, the epistemic threat is the same independently of whether the content of the influencing state is explicit (like the belief that bananas are yellow) or implicit (like the representation that Black people cannot be as beautiful as White people). In both cases, we have the epistemic worry that perceptual experience fails to represent the external facts accurately, and the worry of a circular justification of beliefs. The epistemic threat of cognitive penetration occurs, however, when the process of influence on perception is not internally accessible to the perceiver. Even if the influence comes from an accurate cognitive representation, the perceiver does not realise that their perception is shaped by background cognitive expectations, and takes it to represent reality as it is. Not all would agree on this threat. According to a specific internalist theory of perceptual justification, known as accessibilism, only factors that are consciously accessible to the believer can be relevant to epistemic justification (Feldman and Conee 2001). If the influence of cognitive states on perceptual experience is always implicit, it would then be irrelevant to determining the epistemic value of the penetrated experience. However, the intuition that perceptual experience should have more epistemic weight when it accurately represents the world, rather than our background cognitive states, suggest that cognitive penetration on perception should count as relevant to epistemic justification. If this is so, then accessibilism does not have the resources to account for the epistemic threats posed by the implicit cognitive penetration of experience (Siegel 2012; Puddifoot 2016). As we saw earlier, an agent may gain access to the fact that their perception is influenced, though this access needs to come from instructions or metacognitive inference. Once they realise that their perception is epistemically threatened, however, the epistemic threat opens a question of epistemic responsibility. If the agent realises that their perception is influenced, they 148

Cognitive Penetration and Implicit Cognition

seem epistemically responsible for the judgments or beliefs formed on the basis of this percept. If someone has access to the fact that they have a cognitive state that is currently influencing their perceptual experience, and if they realise that this cognitive state is inaccurate or unjustified, then they would have reasons – understood here as epistemic obligations – to stop holding the influencing state, diminish its influence on experience, or/and stop relying on that experience to make judgements or support beliefs. All these options rely on the conscious, deliberative agency of the person. Having reasons to act, however, arguably depends on being aware of such reasons and on whether one can perform the action at all. Determining the epistemic faultiness in cases where an inaccurate or unjustified cognitive representation bears on perception depends on two factors: whether the person can be aware that their perception is being influenced, and whether the person has ways to intervene on this state (Rettler 2018). Voluntary control is here key in determining the epistemic faultiness of cognitive penetration: if people are aware of the influence or its source, but cannot counteract the influence or intervene on the source, then they are also not much at fault. Many beliefs, desires and intentions are often hard to hold back. I may not be able to deliberately change racial biases, for example, even when I consciously repudiate them. In many cases, we don’t have agential control over the influencing cognitive state, nor over the process of influence, even when we are aware of them both. Some philosophers, however, argue that direct voluntary control is not necessary: some beliefs may still be under indirect control (e.g., Rettler 2018); and even when voluntary control is lacking, accountability to other agents may be sufficient to determine epistemic blame (see Brown 2019 for an overview). Even when voluntary control is possible, epistemic blameworthiness will depend on how difficult it is to exercise this control (Nelkin 2016). Once Pecola is made aware of the unjustified view affecting her perceptual experience, she has good reasons to try and stop herself from holding such views, and, at least, to stop these views from shaping her experience. If exercising such control is feasible, then we could find her at fault for continuing to hold these views or let them shape her perception. If, however, at it is often the case, exercising control requires special skills, effort, or repeated attempts, we may find her less epistemically faulty for failing to counteract the cognitive penetration of her experience.

5  Further Complexities in Attributing Blame to Implicit Influencers On this reading, when Pecola believes that she is not beautiful, because her perception of herself is influenced by a representation that she has no awareness or easy control of, she is not under an epistemic fault. Shall we grant the same epistemic absolution to all kinds of cognitive penetration, when the influencing state is implicit? Here we consider in turn three different considerations that may change this verdict: awareness checks, moral obligations and access to social interventions. 1. Awareness checks. An agent is under epistemic blame if they are aware of an inaccurate, or unjustified representation influencing their perception, and have reasons to diminish the influence itself or the consequences of that influence. If a state is implicit, then the first condition cannot be fulfilled. Several lines of research suggest, however, that one of the canonical examples of implicit cognition, such as implicit biases, are at least partly accessible to consciousness (Gawronski et al. 2006). If this is so, implicit influences on perception would always be epistemically blame-worthy to some degree. For example, people can correctly predict their score in an Implicit Association Test presenting Black and White faces, before taking the test (Hahn et al. 2014). People also tend to misreport their biases when explicitly asked, as they may be concerned with appearing prejudiced (Hall and 149

Lucas Battich and Ophelia Deroy

Payne 2010), an interpretation which supposes that they realise that they have certain biases. One issue with conclusions based on verbal self-reports and predictions is that they are not a robust method to assess whether people have introspective or conscious access to the content of the implicit biases, even less if they possibly can. For example, people could merely guess how they would perform, rather than introspectively access the content of their supposed implicit attitudes (Carruthers 2018). 2. Moral obligations are never far away when discussing implicit biases. Pecola’s example is partly chosen to avoid mixing the moral wrongness of implicit representations bearing on perception, from the epistemic risk they introduce. But consider a white person who has internalised the same views as Pecola, and is not aware of them. While they do not explicitly believe that Black people cannot be as beautiful as White people, their perception, like Pecola’s, may still be biased by this implicit representation. Shall they also be absolved epistemically of any fault? Surely, their perception is also threatened by the same epistemic risks as Pecola’s, but the fact that this influence does not hurt them directly, but hurts others, seems to call for more blame. The precise boundaries between epistemic and moral blame are still a matter of debate (Brown 2019). On one approach, even when one is not epistemically responsible for beliefs based on experiences that have been influenced by implicit states, one may still be morally responsible for the outcomes of these beliefs. Attributions of moral responsibility raise the stake for epistemic obligations. If our implicit states put us at a greater risk of a morally reprehensible action, because they influence our perception, we seem to bear a greater responsibility for learning about them, or doing something about them. 3. Role of social feedback. Voluntary control of own’s cognitive states and how they might affect perception can be assumed to rely on an agent’s direct internal access to those states. But this is not necessarily the case. Importantly, there are several means through which perceivers could become aware that their perception is influenced, or even of the specific cognitive contents which are influencing it. These include direct internal means, such as self-reflection and metacognition. But their awareness may also be externally mediated, obtained through other people’s trusted testimony or warning. In this sense, both individual reflective awareness and social influences will affect the epistemic fault arising from cognitive penetration. If such externally mediated access is easy, then, again, the agent may be considered more epistemologically faulty than if it is difficult. Can we then diminish the cognitive influences over our perceptions, if we only have access to this influence through others’ testimony? Travers et  al. (2020) tested for this, by telling participants halfway through the experiment that their responses were biased by either “racial stereotypes” or “facial features”, and encouraged them to do better. They found that when so informed the race-lightness effect was reduced: participants were less influenced by the facial morphologies in their judgements, compared to a control group which had no social feedback. The fact that third-party information can be sufficient to make people realise that their perception is influenced by implicit states, and intervene to diminish this effect, delivers a mixed verdict. On the one hand, people could not, at least under these experimental conditions, fully suppress the effects of morphology on their perception of brightness, suggesting that implicit representations continued to exert some influence (or that all the remaining effects were due to low-level differences, see Firestone and Scholl 2016). On the other hand, social intervention was successful in partly alleviating the influence and making people perform better.

150

Cognitive Penetration and Implicit Cognition

6. Conclusions Cognitive influences on perception operate implicitly and sometimes come from an implicit cognitive state. These two characteristics explain why these influences are epistemic threatening, and help determine how much epistemic fault rests with the agent. Independently of whether the influencing state is explicit or implicit, the fact that perception can be influenced by cognitive states without us directly being aware of this influence raises an epistemic threat. Importantly, if it is possible and easy for perceivers to indirectly realise that their perception is influenced, their epistemic responsibility seems to be engaged. Cognitive penetrability and its epistemic consequences have been studied and debated as a phenomenon affecting the individual. The role of social factors, however, highlights that both awareness and the possibility or difficulty of control need to be assessed at the level of the individual and the level of cultures and groups. Groups may be epistemically appraisable, rather than individuals themselves, for normalising ill-founded beliefs, and failing to investigate and propagate awareness of the implicit influences that distort individuals’ perceptions and viciously shape their beliefs (see Siegel 2017). Even individual metacognitive awareness, which individuals lack when it comes to implicit influences on perception (Travers et al. 2020), may eventually be shaped by social factors (Pescetelli et al. 2016). Though perception is an individual state and process, cognitive penetration makes it socially dependent in two important ways. Social feedback may be needed for us to realise that our perception is influenced, and it may also be needed for us to realise what states exactly influence it.

Related topics Chapters 8, 15, 16, 17

References Azevedo, R. T., Garfinkel, S. N., Critchley, H. D., and Tsakiris, M. 2017. “Cardiac afferent activity modulates the expression of racial stereotypes”. Nature Communications, 8: 13854. Bergqvist, A., and Cowan, R., eds. 2018. Evaluative perception. Oxford: Oxford University Press. Brown, J. 2019. “Epistemically blameworthy belief ”. Philosophical Studies, 177: 3595–3614. Burnston, D. C. 2017. “Cognitive penetration and the cognition – perception interface”. Synthese, 194: 3645–3668. Carruthers, P. 2018. “Implicit versus explicit attitudes: differing manifestations of the same representational structures?”. Review of Philosophy and Psychology, 9: 51–72. Cermeño-Aínsa, S. 2020. “The cognitive penetrability of perception: A blocked debate and a tentative solution”. Consciousness and Cognition, 77: 102838. Davies, M. 2015. “Knowledge (explicit, implicit and tacit): Philosophical aspects”. In J. Wright, ed., International encyclopedia of the social and behavioral sciences. 2nd edn. Oxford: Elsevier. Deroy, O. 2013. “Object-sensitivity versus cognitive penetrability of perception”. Philosophical Studies, 162: 87–107. Deroy, O. 2019. “Predictions do not entail cognitive penetration: ‘Racial’ biases in predictive models of perception”. In C. Limbeck-Lilienau and F. Stadler, eds., The philosophy of perception. Berlin: De Gruyter. Deroy, O., Spence, C., and Noppeney, U. 2016. “Metacognition in multisensory perception”. Trends in Cognitive Sciences, 20: 736–747. Dummett, M. 1991. The logical basis of metaphysics. London: Duckworth. Feldman, R., and Conee, E. 2001. “Internalism defended”. American Philosophical Quarterly, 38: 1–18. Firestone, C., and Scholl, B. J. 2016. “Cognition does not affect perception: Evaluating the evidence for ‘top-down’ effects”. Behavioral and Brain Sciences, 39: e229.

151

Lucas Battich and Ophelia Deroy Gawronski, B., Hofmann, W., and Wilbur, C. J. 2006. “Are ‘implicit’ attitudes unconscious?”. Consciousness and Cognition, 15: 485–499. Hahn, A., Judd, C. M., Hirsh, H. K., and Blair, I. V. 2014. “Awareness of implicit attitudes”. Journal of Experimental Psychology: General, 143: 1369–1392. Hall, D. L., and Payne, B. K. 2010. “Unconscious influences of attitudes and challenges to self-control”. In R. R. Hassin, K. N. Ochsner, and Y. Trope, eds., Self control in society, mind, and brain. Oxford: Oxford University Press. Hansen, T., Olkkonen, M., Walter, S., and Gegenfurtner, K. R. 2006. “Memory modulates color appearance”. Nature Neuroscience, 9: 1367–1368. Levin, D. T., and Banaji, M. R. 2006. “Distortions in the perceived lightness of faces: the role of race categories”. Journal of Experimental Psychology: General, 135: 501–512. Lyons, J. 2011. “Circularity, reliability, and the cognitive penetrability of perception”. Philosophical Issues, 21: 289–311. Marchi, F. 2017. “Attention and cognitive penetrability: the epistemic consequences of attention as a form of metacognitive regulation”. Consciousness and Cognition, 47: 48–62. McGurk, H., and MacDonald, J. 1976. “Hearing lips and seeing voices”. Nature, 264: 746–748. Mole, C. 2015. “Attention and cognitive penetration”. In J. Zeimbekis and A. Raftopoulos, eds., The cognitive penetrability of perception: new philosophical perspectives. Oxford: Oxford University Press. Nelkin, D. K. 2016. “Difficulty and degrees of moral praiseworthiness and blameworthiness”. Noûs, 50: 356–378. Newen, A., and Vetter, P. 2017. “Why cognitive penetration of our perceptual experience is still the most plausible account”. Consciousness and Cognition, 47: 26–37. Pescetelli, N., Rees, G., and Bahrami, B. 2016. “The perceptual and social components of metacognition”. Journal of Experimental Psychology: General, 145: 949–965. Proust, J. 2010. “Metacognition”. Philosophy Compass, 5: 989–998. Puddifoot, K. 2016. “Accessibilism and the challenge from implicit bias”. Pacific Philosophical Quarterly, 97: 421–434. Pylyshyn, Z. 1999. “Is vision continuous with cognition? The case for cognitive impenetrability of visual perception”. Behavioral and Brain Sciences, 22: 341–365. Raftopoulos, A. 2019. Cognitive penetrability and the epistemic role of perception. London: Palgrave Macmillan. Rettler, L. 2018. “In defence of doxastic blame”. Synthese, 195: 2205–2226. Siegel, S. 2012. “Cognitive penetrability and perceptual justification”. Noûs, 46: 201–222. Siegel, S. 2017. The rationality of perception. Oxford: Oxford University Press. Smithies, D. 2014. “The phenomenal basis of epistemic justification”. In M. Sprevak, and J. Kallestrup, eds., New waves in philosophy of mind. London: Palgrave Macmillan. Song, C., Kanai, R., Fleming, S. M., Weil, R. S., Schwarzkopf, D. S., and Rees, G. 2011. “Relating interindividual differences in metacognitive performance on different perceptual tasks”. Consciousness and Cognition, 20: 1787–1792. Stokes, D. 2013. “Cognitive penetrability of perception”. Philosophy Compass, 8: 646–663. Travers, E., Fairhurst, M. T., and Deroy, O. 2020. “Racial bias in face perception is sensitive to instructions but not introspection”. Consciousness and Cognition, 83: 102952. Valenti, J. J., and Firestone, C. 2019. “Finding the ‘odd one out’: memory color effects and the logic of appearance”. Cognition, 191: 103934.

152

PART 3

Ways of Perceiving, Knowing, Believing

11 HELMHOLTZ ON UNCONSCIOUS INFERENCE IN EXPERIENCE Lydia Patton

Hermann von Helmholtz’s career was among those that defined scientific research in 19thcentury Germany. He contributed to physiology, mathematics, optics, acoustics, electromagnetism, fluid mechanics, and related fields. His research into perceptual phenomena, including color perception (for which he invented an opthalmoscope at the same time as Charles Babbage’s), spatial perception, and harmony and tone, led him to a characteristic multilevel account of perception. Helmholtz’s account of the role of ‘unconscious inferences’ in perception makes him a natural precursor of research into implicit cognition in contemporary philosophy of mind and neuroscience. Before diving into that research, it is worth sorting out the foundation of Helmholtz’s multilevel account of perception (section 1). This in turn will allow for a closer understanding of how Helmholtz employs physiology, generic concepts, and group theory in his analysis of representation of the external world (section 2). Section 3 will investigate the recent theory of ‘predictive processing’, a promising contemporary development of a broadly Helmholtzian approach. The chapter will conclude with a brief analysis of the explanatory and perspectival features of Helmholtz’s theory that distinguish it from contemporary approaches.

1.  The Mind in Perception A recent history of ‘scientific psychology’ focusing on the tradition to which Helmholtz belonged (Murray and Link 2021) considers psychophysics and the measurement of sensations (e.g., Weber’s Law, Fechner’s Law) to be part of psychology. This is accurate given the current way that the disciplines are divided. But, in the context in which Helmholtz was working, the measurement of responses to physical stimuli, the mapping of stimulus-response curves, the measurement of the time it takes for a sensation to propagate along a nerve, and the like were all part of physiology.1 In the lab of Johannes Muller in which Helmholtz began his career, and in the work of his colleagues including Emil Du Bois-Reymond, Wilhelm Weber, and Gustav Fechner, physiology was pursued using experimental methods. Experimental setups included mechanisms that duplicated binocular vision (stereoscopes), wooden models that simulated the way the eye’s muscles moved in focusing, and the like. The mechanistic system of the body could be investigated using experimental methods. The methods of much early 19th-century psychology, on DOI: 10.4324/9781003014584-15 155

Lydia Patton

the other hand, were ‘introspective’ or ‘descriptive’. One strain of research, the physiological one, leads from Johann Friedrich Herbart’s work in the early 19th century, to the physiological research of Wundt and Helmholtz. Another strain begins from the hermeneutic or descriptive methods found in German Idealism, and culminates in Wilhelm Dilthey’s psychological foundation of the human sciences (Damböck 2020). As a result, “in the nineteenth century, arguments about the task, method, and relevance of psychology were more heated than ever before” (Reiners 2020: 241). As the debates went on, even the definition of ‘psychology’ itself changed: toward the end of the 19th century, the methods of psychology began to shift, from introspective to experimental. Helmholtz himself was a significant influence on this change: The psychology of 1850 was the fruit of the introspective method of research. The value of this method as applied in the study of the higher intellectual processes can not be gainsaid. Even here, however, it is to be recognized as only one of several methods of extending knowledge of the higher intellectual faculties and processes. The psychology of to-day [1902] is an experimental science, or, if not wholly so, at least the foundations and a large part of the superstructure represent experimental and laboratory research, the introspective method coming in only at the last to complete and round out the superstructure. The metamorphosis in the methods of research in psychology has had the effect of giving it a new and much higher standing in the scientific world than it ever had in the period of a purely introspective method. This renaissance of psychology is due in very large measure to the influence of von Helmholtz. (Hall 1902: 561) Despite the fact that Helmholtz’s contributions to the study of sensation and perception are crucial to the history of psychology when viewed in retrospect, he would have insisted that most were results in physiology. In the 1860s and 70s, Helmholtz believed that, because it employed the introspective methods detailed earlier, psychology was not yet a true science.2 In the latter part of the 19th century, as Hall details, psychology became more experimental, and Helmholtz’s earlier work became assimilated to the new tradition. A curious feature of the relation of Helmholtz to psychology is that those contributions of his which have come to be fundamental in psychology were at the time they were made published as contributions to physiology. In this connection it is significant to remember that the methods which Helmholtz originated in the fifth and sixth decades for the study of the senses and sensation, are the methods of the new psychology of the last three decades of the nineteenth century. (Hall 1902: 561) Helmholtz’s work in psychology had an outsized influence. Despite his insistence on a scientific, ‘physiological’ method, Helmholtz recognized that representing objects as external to the subject, and as standing in causal relationships to each other and to the subject, required an inference from sensation to reality, and an investigation of the human activity of symbolization: the depiction, not just the experience, of an outer world. Helmholtz’s ‘sign’ theory was well known to his contemporaries. One of them, the eminent psychologist Carl Stumpf, sums it up as follows: The Physiologische Optik, as is well known, contains also [Helmholtz’s] philosophical theory of the relation of our consciousness to the outer world, to which Helmholtz 156

Helmholtz on Unconscious Inference in Experience

in his Thatsachen der Wahrnehmung latterly returned, the theory that the a priori selfevident law of causation necessitates the acceptation by us of an outer world; that, however, our knowledge of this outer world must remain essentially a symbolical one, and that our sensations are merely signs of what is real. (Stumpf 1895: 8) The fundamentals of Helmholtz’s mature epistemology of perception can be summarized as follows. “Our sensations are merely signs of what is real”, as Stumpf summarizes.3 These signs must be interpreted, as signals coming from objects causally related to the subject. In order to do that, sensations must be worked up into perceptions that are taken to be representations of external objects. The working up consists of inductive inferences from the causal properties of objects previously experienced, as well as modal judgments about the possible properties of those objects. In what follows, we will investigate how Helmholtz combines this epistemological account with his research into the physiology of perception. The bulk of the discussion in this chapter is drawn from Helmholtz’s Handbook of Physiological Optics. In that work, Helmholtz provides a thorough investigation and presentation of novel research into the physiology of perception. However, as Stumpf notes earlier, in the Handbook Helmholtz also presents his philosophical account of our experience and knowledge of the world. Helmholtz’s analysis of our perception of the world stratifies perception into the following stages or levels (Helmholtz 1867, §26: 427): (1) Sense-Physiology. A physical level of responses to stimuli, investigated by physiology and established independently of the mind’s influence. An account of “which specific features of the physical stimulant, and of the physiological stimulation [of a nerve], give an occasion for the formation of this or that specific representation of a type of perceived external object”. (2) Physiological psychology. An account of “mental operations and their laws, insofar as they come into the consideration of sensible perception”. (3) Pure psychology. An inquiry “whose principal task is to establish the laws and nature of the operations of the mind”. Helmholtz begins his analysis of experience with the results of his empirical, physiological research into a series of physical processes: signals from proximal or distal sources transmitted via waves to our sense organs. Our reception of these signals consists in the reactions of the sensory nerves of the body.4 Helmholtz was well known for re-creating the physical capacities of the mind using instruments (e.g., using objects with resonant cavities to re-create the eardrum’s capacity to resonate at different frequencies). This method reproduces the physiological level of experience. In contrast, both of the other levels involve mental activity. Physiological psychology investigates how the operations of the mind enter into the perception and representation of objects of experience. Sensory effects – for example, the vibration of the eye’s cones, or of the organ of Corti in the ear’s cochlea – are the basis for the construction of sensory perceptions, which are distinct from sensations. These perceptions are interpreted as signs of the presence and features of their objects, whether external or internal.5 Pure psychology, on the other hand, investigates the laws of mental operation in isolation from the mind’s contribution to the perception and representation of observable objects. Helmholtz emphasizes that his own account of experience will focus on objective representation, and that he will leave pure psychology to one side.6 His account of experience, and of the representation of objects, will rely on physiology and on physiological psychology. 157

Lydia Patton

In §26 of the Handbook, Helmholtz makes the strong claim that there is no perception without the involvement of the mind. This claim depends on his distinction between sensations and perceptions. Sensations are signals transmitted along nerves. Sensations are used – in Helmholtz’s words – to form perceptions of objects (1867, §26: 427). But perceiving an object already involves representing it as the source of the perception. Specifically, perceiving an object as such – as opposed to merely receiving sensations from some unidentified source – requires that we think of some object, distinct from the subject, as the cause of the sensations. Helmholtz concludes, Since perceptions [Wahrnehmungen] of external objects thus belong to representations, and representations always are acts of our mental operation, perceptions can come about only in virtue of mental operation, and thus the doctrine of perceptions in fact already belongs to the domain of psychology. (1867, §26: 427) This statement seems to give ‘psychology’ pride of place in Helmholtz’s account of perception. But ‘psychology’ in this sense is ‘mental operation’, which for Helmholtz is grounded in an empirical account of mental activity. Helmholtz sees himself to be giving an account based on natural science, not on pure psychology or on philosophy. Helmholtz intends his method in natural science to be based on “secure facts and a method grounded in general, recognized, and clear principles” (1867, §26: 427; see also 1879/1878, 1894). Psychology, as an account of mental operation, is necessary to explain how we move from sensation to representation of an external world of objects and events. But we need only as much psychology as epistemology requires. Helmholtz allows for a ‘scientific’ account of how we move from sensation, to perception, to knowledge, in a way that is continuous and based on our sensory experience (see the next section).

2.  Helmholtz’s Analysis of Perception We are presented with perceptual phenomena in experience. These phenomena are not unmediated, according to Helmholtz: they are constructed, in a process involving inductive inferences that we use to interpret sensations using memory and reasoning. But that experience, and that interpretation, is embedded in a more fundamental theory regarding what in our experience is certain and secure, and what is derived or even arbitrary (the fundamental question of Helmholtz 1879/1878). Perceptual manifolds – visual images, haptic fields, smells, harmonies, and so on – are the result of inductive inferences, which appeal to (1) the physiology of the subject, (2) the sensations that arise from interaction with proximal or distal causes, and (3) past experience of similar phenomena. To Helmholtz, a visual image is constructed through a series of inferences. Sensory signs are interpreted to be indications of features of our visual experience.7 While ‘mental operations’ must have an experimental basis, they can be extended in various ways. For Helmholtz, our perceptual experience is enriched with counterfactual claims that appeal to inferences: for instance, inferences about what would happen if the subject were to change her position or behavior with respect to an object or event being perceived.8 Those counterfactual claims are grounded in the inductive inferences we make based on experience, as well as on the mathematical and formal conditions of measurement.9 For instance, if I am perceiving a table, I can infer what aspect of a table would present itself to me if I were to walk closer to the table or to approach it from another angle. That inference is 158

Helmholtz on Unconscious Inference in Experience

derived from empirical, inductive knowledge about how similar massive objects have behaved in past interactions with me as perceiving subject, as well as from mathematical knowledge of visual angles. Helmholtz’s theory in this connection is found partly in §26 of the Handbook. Helmholtz argues that manifold aspects of our perceptual experience are inferences, and, in fact, are unconscious inferences (Helmholtz 1867: 430). It is an illusion, Helmholtz argues, that the objects we perceive are extended in space, are colored, and produce sounds (second ed., 1878: 376–377). The spatial, acoustic, and optic manifolds are produced by the mind by combining and synthesizing the sequences of perceptions we do experience. Even more refined aspects of our experience might be products of inferences. For instance, I might have experience that tells me that objects farther away appear smaller. Moreover, that objects farther away in a landscape appear lighter, while closer objects are darker. So, if I see two trees in a painting, one of which is lighter and smaller, and the other of which is darker and larger, I will perceive the darker, larger one as closer, and the lighter, smaller one as farther away. Helmholtz explains such phenomena by appealing to an inference made by the mind in response to inductive generalizations from experience. As De Kock (2014: 727) notes, for Helmholtz .  .  . the idea of external existence was entirely derived a posteriori from the recognition of something like Mill’s “permanent possibilities,” based on the learned association between the coming into being of certain “circles of presentables” and voluntary movement. Once this kind of knowledge is acquired, Helmholtz added, the perceptual process takes the form of an unconscious inductive syllogism, with the current sensation as a minor premise, acquired knowledge of lawlike covariation between movement and sensation as a major premise, and the object as a conclusion. What Helmholtz calls an ‘inference’ or ‘syllogism’ in perception is not a conscious process of bringing to mind the major and minor premises, and drawing a conclusion. Instead, Helmholtz argues that we arrive at inference-enriched perceptual experience through a process of association and projection. First, we link what is present to us now with the learned context of how our voluntary movements are associated with the presence of broader classes of objects or possible ‘presentabilia’ (De Kock 2014: 727–728). Second, our mind projects the nature and behavior of the kind of object that we think is likely to cause our present sensations into our enriched perceptual experience.10 Perceptual signs are not direct evidence of the properties of our perceptual experience them, partly because that experience is enriched via inference in this way. To understand how that works, we must have more information about how a particular kind of object stands in actual and possible relations with the subject’s own body and perceptual apparatus. Neither the eye nor the hand gives a direct depiction [Bild] of a spatial magnitude extended in three dimensions. The representation of bodies arises first through the comparison of the depictions [Bilder] of both eyes, or through the motion of the body, with reference to the hand. [. . .] The representation of a spatially extended body, for instance of a table, contains a host of single observations. Within it lie included the entire range of depictions that can be granted to me of this table, if I were to regard it from different angles and at different distances, further, the entire range of tactile impressions, which I would receive, if I were to lay my hands one after another on the different positions of its surface. (Helmholtz 1867: 447) 159

Lydia Patton

If the representation of a body includes all the ways a subject could interact with it, and could depict it as an object, then that already involves the concept of that object. The concept, in this sense, is a principle that organizes all the infinite ways that an object can be present to us. Such a representation of a single individual body thus in fact is already a concept, which contains an infinite number of single intuitions that follow each other in time, which all can be derived from [the concept], just as the genus concept “table” in turn contains all single tables in itself, and expresses their common features. The representation of a single individual table, which I carry in myself, is correct and exact if I can derive from it correctly and exactly which sensations I would have, it I were to bring my eye and my hand in this or that particular position against the table. I do not understand how there could be any other kind of similarity between such a representation and the body represented by it. The former is the mental sign for the latter. The nature of this sign is not arbitrarily chosen by me, but forced on me by the nature of my sense organs and my mind. (Helmholtz 1867: 447) Note that Helmholtz’s account of concepts is related to the Kantian account: Helmholtz’s concepts contain single intuitions in space and time, for instance.11 But Helmholtz’s account is enriched by his physical and physiological research, which allows him to engage in counterfactual reasoning. In 1944, Ernst Cassirer wrote “The Concept of Group and the Theory of Perception”, in which he explored and extended Helmholtz’s approach.12 Cassirer begins, “Helmholtz avowed that it was his dealing with the fundamental problems of ‘Physiological Optics’ that encouraged and, in a certain sense, even enabled him to undertake this synthesis” of “highly different fields of study”, including geometry and physiology. “From the outset,” Cassirer continues, “his attention was drawn to the question as to whether and to what extent experience contributes toward shaping the notion of space” (1944: 1–2). Helmholtz accepted that space is a ‘transcendental form’ in Kant’s sense, according to Cassirer, but the transcendental form merely designates the general “possibility of coexistence” – as space had been defined by Kant. As soon as we attempt to specify this possibility – and only through such specification can it be made applicable to and fruitful for the problems of physics – we find ourselves faced with a whole new set of questions. We must now introduce a metrical determination. (Cassirer 1944: 2) Doing so requires that we introduce axioms governing the motions of rigid bodies (see Biagioli 2014 for a detailed discussion). But once we do this, a new horizon opens up for the analysis of perception. After all, any body whatsoever can now be viewed as falling under a set of axioms governing its possible motions with respect to the subject. Helmholtz’s account is extended through the application of group theory to his experimental analysis of perception.13 The experimental method investigates the empirical objects we encounter, and experimental investigation may change how we understand empirical concepts. That experimental method is informed by mathematical and dynamical structures, including group-theoretic structures that describe possible motions. The infinite possible positions of the subject with respect to the percept, and the infinite possible paths the subject could take to have access to the object, and vice versa, are described using groups of rigid motions. Helmholtz’s 160

Helmholtz on Unconscious Inference in Experience

group-theoretic or perspectival reading of the spatial properties of our representations allows for the imaginative extension of geometrical figures encountered in experience to a broader context of the space of possible presentabilia: the geometrical objects that could be present to us if we changed our orientation with respect to them, for instance. Elaborating Helmholtz’s perspective with respect to experience allows for the following argument to be made, summarizing points Helmholtz makes in the Handbook and in his essay “The Facts in Perception”. (1) We acquire concepts in experience of objects: that is, concepts that show how the object would appear to a subject in a certain position in space and time. These concepts contain “the entire range of depictions that can be granted to me” of an object of this kind (1867: 447). (2) Those concepts are acquired through experience, but are also products of mental operation (for instance, the implicit recognition of the symmetries of certain rigid motions). (3) Once we have acquired these concepts, we can extend them through reasoning about the spatial axioms and group symmetries. (4) We can also use models – perhaps even physical models – to extend our knowledge of the modal physical properties of object and subject. We can show how a subject might be related to an object by building a model of their relationship.14 For instance, we might show how a subject can perceive an object that is quite far away, by showing that the subject had access to a telescope.

3.  Helmholtz’s Causal Perspectivism and Contemporary Perceptual Theory Helmholtz developed a causal, perspectival analysis of perception that distinguishes his theory from contemporary perceptual theories, both in its motivating questions and in its basic approach. Helmholtz’s theory is intended to grasp “the facts in perception”:15 the knowledge we may have about perception given the limitations of our epistemic position. We do not have secure scientific knowledge about the inner workings of the mind via pure psychology. Neither does perception afford direct access to the properties of external objects. Objective knowledge must be constituted by analyzing the perceptual signs encountered in experience. It is possible to have access, Helmholtz argues, to the causal laws that govern the interactions between perceiving subjects and the external world, as it is presented in perceptual signs.16 Moreover, we do have access to information about the perspective of the perceiving subject. This information includes physiological facts about the subject’s response to stimuli, and modal facts about the possible ways objects with certain qualities could be present to the subject in perception, determined through reasoning about geometric perspective, group theory, and the like. The facts Helmholtz cites are the basis of a causal, perspectival account intended to support broad explanatory claims about the knowledge we can establish through an analysis of perception. This causal account rests on a scientific analysis of the facts in perception, which are extended to an explanatory framework employing causal laws and counterfactual reasoning. While the properties and behavior of the perceiving subject are a key pillar of that account, Helmholtz does not intend primarily to explain how perceptual mental states arise in the subject, or to analyze how these states play a role in broader ascriptions of beliefs and desires to a particular subject. Thus, his account differs significantly from contemporary approaches, as we will see later. The discussion of Helmholtz’s approach in §2 identifies three distinctive features of his view.17 First, Helmholtz’s account of actual perceptual experience situates the perceiving subject’s perspective within a larger context of possible experience.18 Second, a perspectival element 161

Lydia Patton

is operative even in actual experience, in the (at times implicit) expectations that are projected onto perceptual experience based on inductive inferences from previous experience. Finally, formal, explicit reasoning about the perspectives operative in experience allows for experimental and formal investigation of perception. With these features of Helmholtz’s view on board, we can evaluate the relationship between his approach and more contemporary versions of inferentialist accounts of perception, especially those that involve attributing mental states to the perceiving subject. Modern approaches attribute mental states to the perceiving subject to explain, predict, and understand the behavior of that subject. There are two general types of theories, distinguished by the type of attribution involved: the theory theory and the simulation theory.19 The theory theory argues that the subject is working from an ‘information base’ and a set of inferential principles in order to ‘read the mind’ of another subject. ‘Mind-reading’ involves thinking of other subjects as rational agents who have capacities similar to one’s own (e.g., Gopnik’s and Meltzoff’s “child scientists”, 1997). For theory theorists, those capacities are best described as rational inferences from an inductive basis.20 The simulation theory, by contrast, works from a more practical and less theoretical understanding of mental capacities.21 Heal (1996: 46–48) uses the example of a model airplane, used to simulate, and thus to understand, the flight of real airplanes. Heal compares this process to Ryle’s ‘knowing how’, as opposed to ‘knowing that’ (1996: 48). One does not need to have a full theory of mental processes and inference to be able to grasp ways in which others’ mental capacities resemble one’s own. If a person has learned successfully how to infer from others’ facial expressions what their mood might be, for instance, then – according to the simulation theory – that person doesn’t need to know how the expression-mood relationship works overall to read someone else’s mind on their face. They only need to be able to understand, in a coarsegrained way, how others’ mental processes resemble their own.22 It may seem at first glance as if Helmholtz’s account is a theory theory. Helmholtz frequently uses the word “inference” in this theory of perception, which implies that conclusions are being drawn from evidence. Moreover, the pedigree of Helmholtz’s theory includes John Stuart Mill’s theory of induction via syllogistic inference (De Kock 2014, §3: 725–728). However, many of Helmholtz’s experiments in physiology of perception involve trying to put a subject into a particular situation, perhaps trying to induce a certain perceptual experience using physical, physiological, or mental cues. Moreover, Helmholtz denies that we can explain perception a priori, arguing that experimental investigation is necessary to discover the contributions of the perceiving subject to experience. So, while Helmholtz’s approach could be retrospectively compared to a theory theorist in his explanations of the physico-physiological basis of perceptual experience, he seems to more closely resemble a simulation theorist in his methods for scientifically investigating the phenomena of perceptual experience.23 But in fact, Helmholtz is neither a theory theorist nor a simulation theorist. The fundamental motivation for Helmholtz’s account is quite different from the project undertaken by these contemporary approaches. Helmholtz’s view was not developed in an attempt to explain mental processes, much less ‘beliefs’ and ‘desires’ along the lines of contemporary philosophy of mind. As explained earlier, Helmholtz deliberately avoided giving psychological explanations for perceptual experience. Instead, he built his theory to provide an explanation and epistemological account of the phenomena of sensory experience on the basis of an experimental investigation of the physiology of perception, and the modal causal reasoning that could be justified on the basis of that experimental information. To be sure, Helmholtz thought that ordinary sensory images (visual, haptic, auditory, and so on), as well as observable events, required inference to explain. But an account of the mental 162

Helmholtz on Unconscious Inference in Experience

processes of a single subject would never meet Helmholtz’s criteria for an epistemology adequate to explain perceptual experience or the role of inference in it. In Helmholtz’s analysis of perception, explaining a single subject’s experience of objects involves appeal to concepts which have two features (see §2): 1. The concepts are acquired inductively, through experience. 2. The concepts, when put into practice using groups of rigid motions, contain an infinite number of possible experiences of (or perspectives on) the objects of experience. Explaining the perceptual experience of a single subject requires inferential knowledge about how possible subjects could have experience of the same objects (or types of object), just from a different spatial perspective, or under different physical circumstances. Helmholtz’s analysis of perception of external objects relies on thick concepts that involve a rich characterization of those objects: not only the actual features of the objects as experienced, but counterfactual information about how they would be experienced by different possible subjects in different possible situations. Restricting this account to an analysis or simulation of the inferences or experience of a single subject would cut off its central epistemological foundation. Helmholtz’s explanation of how we can have knowledge of the possible experience of other people is not equivalent to the ‘mind-reading’ explored by contemporary theory theorists. Instead, Helmholtz’s view invites the conclusion that our own experience is one node of a network of possible experiences. We must involve mental and physical reasoning in order to represent objects in the first place. In doing so, we acquire thick concepts of the kind described earlier. Those concepts, when employed in experience, show us how to characterize a single subject’s experience as only one among an infinite number of possible perspectives on the external world.24 Thus, the theory theory and the simulation theory are not well suited to the Helmholtzian approach. This is not a negative assessment of these theories on their own, but rather a recognition that they are motivated by different questions and follow an entirely different methodology. A contemporary theory involving Helmholtzian ideas is found outside philosophy, in the field of computational neuroscience. Recently, Helmholtz’s synthesis of Kantian and inductive methods has been identified (Swanson 2016) as one of the ‘roots’ of a new research field: predictive processing (PP). Like Helmholtz’s causal perspectivism, predictive processing analyzes perception as an interaction between distal signals, physiological responses to stimuli, and mental operations that become part of the perceptual process: ‘generative models’ for PP, ‘inductive inferences’ for Helmholtz. The approach is distinctive in that it takes mental activity as part of the causal process giving rise to perceptual experience. In other words, instead of seeing perception as a physical or physiological event that is then interpreted by the brain, Helmholtz, and predictive processing theorists, analyze perception as a causal process from the beginning. A scientific, causal explanation of perception is necessary to explain perceptual experience, in their view. An exclusively philosophical account of the mental states of a subject does not go far enough to achieve such an explanation. Predictive processing is an explanation of perception that involves inference, but where the activity of the mind is analyzed as one of the processes that generate perceptual experience. This account does not depict perception as built from detections of signals that are then processed by modules in the brain that merely interpret signals received in a distinct perceptual process. Instead, detections are worked up by ‘generative models’ in the brain, based on hypotheses about the current input signals using past experience as an inductive basis. These generative models are operative in perception. Perception in predictive processing thus becomes a kind of testing procedure, 163

Lydia Patton

in which the nervous system constantly generates perceptual experience using, not only the detection of perceptual signals, but also causal and statistical reasoning.25 This framework has had some successes in explaining difficult phenomena including binocular rivalry.26 Several features of predictive processing align it with Helmholtz’s approach. Predictive processing requires determining a broader set of possible ways of generating experience from signals first, before focusing on the mental states of a single agent as tokens of those broader types. Moreover, predictive processing is ‘top down’, invoking causal reasoning from the beginning: PP begins with causal laws or ‘generative models’ and derives general explanations of possible perceptual experience on that basis.27 For Helmholtz (and for PP), mere detection of distal signals is not enough to have rich perceptual experience of external objects and processes. Further, having knowledge of that rich experience requires reasoning about how mental activity factors into its causal structure. Recreating causal structure in perceptual experience requires inductive reasoning that is constantly being updated.28 Of the contemporary versions of cognition that invoke inference, predictive processing is perhaps the closest to Helmholtz’s methods and approach. Comparing Helmholtz’s theory to contemporary theories allows us to identify a central feature of Helmholtz’s view: its causal perspectival nature (Patton 2019). Helmholtz’s analysis of perception requires a causal account of how a single subject’s experience is generated from distal signals. The processing used to generate an actual perceptual experience is represented, in his analysis, as one possible outcome among a larger set of possible outcomes. That account identifies a subject’s actual experience as one node in a network of possible experiences, and thus as defining one perspective within a larger set of perspectives. The causal perspectival approach to perception is not aimed at an analysis of a single perceptual experience, then. It is geared toward providing an account of the origins and generation of actual perceptual experience in causal and inferential processes that can be shown to be the possible source of a much broader set of possible experiences. Helmholtz argues that what I’ve called the causal perspectival approach is the real source of our knowledge about perception.29

Related Topics Chapters 9, 10, 16, 21, 29, 31

Notes 1 The Weber-Fechner law governing stimulus-response curves, established by Ernst Weber and Gustav Fechner, were seen as part of physiology, not psychology, at the time Helmholtz was working. 2 There were significant debates at the time concerning the scientific status of ‘descriptive’ versus ‘experimental’ methods in psychology. See Damböck (2020); Reiners (2020). 3 On the “sign theory” see Hatfield (1990, Ch. 5: 171ff). 4 Each nerve has a “specific sense energy” or “specific nerve energy”: a characteristic way that a particular sense organ or nerve reacts to stimuli (Helmholtz 1867, §17). These sense energies determine effects in the body from physical causes. 5 Helmholtz’s own multilevel theory came over 100 years before David Marr’s (1982), though of course the two are quite different. 6 Helmholtz argues that the pure part of psychology is not based on “secure facts and a method grounded in general, recognized, and clear principles”. In 1867, Helmholtz’s view of ‘pure psychology’ as a pure (non-empirical) investigation of mental operations could have been influenced by a number of movements in psychology, including nativism, and in philosophy, including speculative idealism (De Kock 2014) and neo-Kantianism. Hermann Cohen, founder of the neo-Kantian Marburg School, faulted Helmholtz for connecting the conditions for objective representation with psychology and with the physiology of perception. Cohen distinguishes between the formal, a priori conditions for representing objects, involving the logical, necessary conditions for the unity of consciousness, and Helmholtz’s

164

Helmholtz on Unconscious Inference in Experience contingent requirements, conceived in terms of the “psycho-physical organization” of the knowing subject, for forming coherent representations of objects and events. For Cohen, concepts are “pure”, meaning that they are not derived from any empirical representation or content. Such concepts are derived from the a priori conditions for the unity of consciousness. See Pecere (2021), §2, for detailed discussion. Helmholtz was a keen opponent of “nativist” theories along the lines of Ewald Hering, according to which perceived relationships or phenomena are explained by innate mechanisms or modules in the brain. Ewald Hering appealed to innate mental mechanisms to explain phenomena Helmholtz considered the result of inference from sensation. The most famous debates they engaged in concerned phenomena like simultaneous contrast or binocular vision. In both cases, Hering argued for an innate mental module or organ that would explain the phenomenon. Helmholtz argued that simultaneous contrast, for instance, could be an epiphenomenon resulting from an artefact of our perceptual system. The eye accommodates itself to perceiving colors at the same time, and interprets one as white and the other as a different color (Helmholtz 1867: 396–397). For more detailed discussion of Helmholtz and his nativist nemesis Ewald Hering, see Patton forthcoming; Turner 1994: Ch. 9. Still, it’s not quite right to say that Helmholtz rejected all explanations appealing to innate structures. After all, Helmholtz himself appealed to the cones of the eye as structures explaining why we pick up on visual signals of certain wavelengths and not others. (See Meulders 2010: 184 for a discussion of how organic structures of the ear contribute to the perception of combination tones, for Helmholtz.) But that explanation is at the level of sensation. In appealing to cones to explain why we perceive light at a certain wavelength and not at others, we are making a physical hypothesis that can be confirmed or falsified using what Helmholtz considered to be secure epistemological methods. 7 Sometimes Helmholtz went further, arguing that the mind must ‘project’ some features of images: three dimensions, for instance. See Turner 1994: 14–20 and passim. 8 PO 26: 447. 9 See Biagioli 2014 for analysis of how a priori spatial conditions of measurement are active in perception, for Helmholtz. 10 Helmholtz’s view that perceptual experience is enriched with ‘unconscious inferences’ was subject to considerable criticism. 11 See Pecere 2021 and De Kock 2014 for a more detailed account of the relation between Helmholtz and Kant. 12 Cassirer 1944. See Biagioli 2016. 13 Helmholtz’s account of group theory in perception draws on manifold theory, Felix Klein’s Erlangen Programm, and measurement theory. See, e.g., Biagioli 2014. 14 Wittgenstein later picked up on this reasoning in Helmholtz and his student Heinrich Hertz. See Sterrett 2005: Chs. 7–8. 15 The title of a well known address Helmholtz gave in 1878 (Helmholtz 1879/1878). 16 This argument can be found in Helmholtz as early as “On the Conservation of Force” (1847). See Hatfield 1990 for discussion. 17 I am grateful to Robert Thompson for clarification and even formulation of the central themes of this section. 18 For more on the perspectival elements of Helmholtz’s view, see Patton 2019. 19 For a recent classification, see Goldman (2006: 4). I will not discuss his category of ‘rationality theory’ here. 20 The theory theory is often attributed to Wilfrid Sellars (1956). Contemporary defenders include Nichols and Stich (2003) and Gopnik and Meltzoff (1997). 21 Simulation theorists include Jane Heal (1996) and Robert Gordon (1986). 22 In the 1990s, there was an interesting debate in which Jane Heal (e.g., 1996) defended the simulation theory against criticisms from Shaun Nichols and Stephen Stich (e.g., 1992), who favored the theory theory. 23 In this sense, Helmholtz could be compared to ‘hybrid’ theorists, like Alvin Goldman (2006), who combined aspects of both. 24 To an extent, of course, this is a matter of emphasis: theory theorists might respond that their view can also enable explanations of a Helmholtzian kind. But their approach is not motivated by the same questions. 25 “Kant and PP both define the primary function of cognition and perception as the ability to track causal structure without direct access to real-world causes. . . . The PP paradigm thus aims to provide a neurally plausible set of mechanisms by which brains accomplish causal inference and overcome the

165

Lydia Patton challenges of induction. . . . The PP paradigm has been framed as an answer to Hume’s challenge in that it aims to offer an account for how causal structure is extracted from statistical regularities that occur in sensory stimulation. . . . PP’s answer to Humean problems of induction rests on proposed neural computations based on Bayesian principles” (Swanson 2016: 3). 26 Hohwy et al. 2008; see Walsh et al. 2020 for a review of evidence for and against PP. 27 “PP urges that psychology and neuroscience would make better progress on the problems of perception if they would . . . assume that brains actively generate percepts in a top-down manner, not by accumulating and combining input signals, but rather, by issuing predictions or accounts of the current state of the input signals based on hierarchical generative models that rely on prior probabilities and likelihood estimates” (Swanson 2016: 4). 28 See Fazelpour and Thompson 2015; Swanson 2016. This is the sense in which the PP ‘paradigm’ is Kantian. However, it is worth noting that Kant would not have argued for an inductive account of the concepts and causal reasoning that enter into perceptual experience and objective knowledge. Helmholtz, however, would have, and so his framework is closer to that of PP. 29 Helmholtz never uses this exact terminology, of course.

References Biagioli, F. 2014. “What does it mean that ‘space can be transcendental without the axioms being so’? Helmholtz’s claim in context”. Journal for General Philosophy of Science, 45: 1–21. Biagioli, F. 2016. Space, number, and geometry from Helmholtz to Cassirer. Dordrecht: Springer. Cassirer, E. 1944. “The concept of group and the theory of perception”. Philosophy and Phenomenological Research, 5: 1–36. Damböck, C. 2020. “What is descriptive psychology? Ebbinghaus’s 1896 criticism of Dilthey revisited”. HOPOS: The Journal of the International Society for the History of Philosophy of Science, 10: 274–289. De Kock, L. 2014. “Hermann von Helmholtz’s empirico-transcendentalism reconsidered”. Science in Context 27 (4): 709–744. Fazelpour, S., and Thompson, E. 2015. “The Kantian brain: brain dynamics from a neurophenomenological perspective”. Current Opinion in Neurobiology, 31: 223–229. http://dx.doi.org/10.1016/j. conb.2014.12.006 Goldman, A. I. 2006. Simulating minds: the philosophy, psychology, and neuroscience of mindreading. Oxford: Oxford University Press. Gopnik, A., and Meltzoff, A. 1997. Words, thoughts, and theories. Cambridge, MA: MIT Press. Gordon, R. M. 1986. “Folk psychology as simulation”. Mind and Language, 1: 158–171. Hall, W. S. 1902. “The contributions of Helmholtz to physiology and psychology”. Journal of the American Medical Association, 38: 558–561. https://doi.org/10.1001/jama.1902.62480090010001c Hatfield, G. 1990. The natural and the normative. Cambridge, MA: MIT Press. Heal, J. 1996. “Simulation and cognitive penetrability”. Mind and Language, 11: 44–67. Helmholtz, Hermann von. All translations mine: Helmholtz, Hermann von. 1847. “On the conservation of force”. In J. Tyndall, ed., Scientific memoirs. London: Taylor and Francis. Helmholtz, Hermann von. 1867. Handbuch der physiologischen Optik, Leipzig: Leopold Voss. Published in parts from 1856 to 1866, then published in toto in 1867 as Volume Nine of the Allgemeinen Encyclopädie der Physik, ed. Gustav Karsten. Helmholtz, Hermann von. 1879/1878. “Die Thatsachen in der Wahrnehmung”. Berlin: Hirschwald. Helmholtz, Hermann von. 1894. “Über den Ursprung der richtigen Deutung unserer Sinneseindrücke”. Zeitschrift für Psychologie und Physiologie der Sinnesorgane. Band VII, S. 81–96. Hohwy, J., Roepstorff, A., and Friston, K. 2008. “Predictive coding explains binocular rivalry: an epistemological review”. Cognition, 108: 687–701. http://dx.doi.org/10.1016/j.cognition.2008.05.010 Marr, D. 1982. Vision: a computational investigation into the human representation and processing of visual information. San Francisco: W. H. Freeman and Company. Meulders, M. 2010. Helmholtz: from enlightenment to neuroscience. Cambridge, MA: MIT Press. Murray, D., and Link, S. 2021. The creation of scientific psychology. New York: Routledge. Nichols, S., and Stich, S. P. 1992. “Folk psychology: simulation or tacit theory?”. Mind and Language, 7: 35–71.

166

Helmholtz on Unconscious Inference in Experience Nichols, S., and Stich, S. P. 2003. Mindreading: an integrated account of pretence, self-awareness, and understanding other minds. Oxford: Oxford University Press. Patton, L. 2019. “Perspectivalism in the development of scientific observer-relativity”. In M. Kusch, K. Kinzel, J. Steizinger, and N. Wildschut, eds., The emergence of relativism. New York: Routledge: 63–78. Patton, L. Forthcoming. “The Helmholtz-Hering debate: the perils of perigenesis”. In C. Wolfe, P. Pecere, and A. Clericuzio, eds., Mechanism, life and mind in the ‘modern’ era. Dordrecht: Springer. Pecere, P. 2021. “ ‘Physiological Kantianism’ and the ‘organization of the mind’: a reconsideration”. Intellectual History Review, 30: 693–714. Reiners, S. 2020. “ ‘Our science must establish itself ’: on the scientific status of Lazarus and Steinthal’s Völkerpsychologie”. HOPOS: The Journal of the International Society for the History of Philosophy of Science, 10: 234–253. Sellars, W. 1956. “Empiricism and the philosophy of mind”. Minnesota Studies in the Philosophy of Science, 1: 253–329. Sterrett, S. 2005. Wittgenstein flies a kite. New York: Pi Press. Stumpf, C. 1895. “Hermann Von Helmholtz and the new psychology”. Psychological Review, 2: 1–12. Swanson, L. 2016. “The predictive processing paradigm has roots in Kant”. Frontiers in Systems Neuroscience, 10: 79. www.frontiersin.org/articles/10.3389/fnsys.2016.00079/full Turner, S. 1994. In the eye’s mind: vision and the Helmholtz-Hering controversy. Princeton: Princeton University Press. Walsh, K., McGovern, D., Clark, A., and O’Connell, R. 2020. “Evaluating the neurophysiological evidence for predictive processing as a model of perception”. Annals of the New York Academy of Sciences, 1464: 242–268. https://nyaspubs.onlinelibrary.wiley.com/doi/10.1111/nyas.14321

167

12 HUSSERL ON HABIT, HORIZONS, AND BACKGROUND Dermot Moran

In this chapter, I  outline the main features of the phenomenological approach to implicit knowing, focusing on embodied cognition, pre-predicative knowledge, habits, and horizonconsciousness. Generally speaking, twentieth-century analytic philosophy approached implicit cognition either under the category of ‘knowing how’, construed as an ability or complex of dispositions (Gilbert Ryle 1949; but see Stanley and Williamson 2001), or as nonverbal, ‘tacit knowledge’ (“we can know more than we can tell,” Polanyi 1966: 4; Fodor 1968). The European phenomenological tradition (especially Husserl, Heidegger, Gurwitsch, Merleau-Ponty, Sartre, Schutz, see Moran 2000), on the other hand, has a longer and more complex tradition of analyses of intuitive, tacit, ‘pre-predicative’ knowledge, centered on embodiment, that developed prior to and independently of recent analytic discussions, although there have been recent attempts to mediate between these traditions (see Dreyfus 2002a, 2002b, 2005, 2007; Dreyfus and Taylor 2015). British philosophy did have some mid twentieth-century connections with phenomenology, largely through Michael Polanyi and Gilbert Ryle, who offered discussions of tacit, skillful, habitual knowledge, but besides these figures, but mainstream analytic philosophy did not have engagement with the phenomenological tradition until recently largely due to a revival of interest in consciousness (Moran 2011).1 Phenomenology focuses especially on intuitively apprehended, embodied, skillful behavior. Husserl’s mature phenomenology, greatly elaborated on by the French phenomenologist Maurice Merleau-Ponty (who himself was trained in empirical and Gestalt psychology), specifically focuses on this pre-reflective, pre-predicative level of human experience. Philosophy of mind tended to ignore embodiment completely and now that has changed there is increasing interest in the phenomenological contribution.

The Phenomenological Approach Phenomenology as a methodology was announced by Edmund Husserl in his Logical Investigations (1901). He went on to develop his specific account of the phenomenological reduction in Ideas I (Husserl 2014) that brackets what is accidental in the experience in order to arrive at the essence. The key aim is the careful, unprejudiced description of conscious, lived experiences, precisely according to the manner in which they are experienced, without the imposition of

168

DOI: 10.4324/9781003014584-16

Husserl on Habit, Horizons, and Background

external explanatory frameworks, whether from the natural or social sciences, from religion or philosophy, or even from common sense or ordinary language use. One attends to the phenomena of experience as given to the subject. Phenomenology continues in the European philosophy of cognition (e.g., Descartes, Kant) that begins from the centrality of the egoic subject. Scientific objective knowledge usually excludes or brackets the input of the knowing subject. Phenomenology, on the other hand, seeks to restore the knowing subject; the world is what it is for a knowing subject. Phenomenology thus opposes objectivism and naturalism and seeks to relate all knowing back to subjectivity (and the cooperation between subjects known as intersubjectivity). Indeed, phenomenology was especially critical of the emerging empirical psychology due to its inherent objectivistic naturalism. The contribution of the acting, knowing subject must be documented. Knowledge is viewed as an essential, a priori correlation between subject and object. Traditionally, philosophy focused on one or other side of this correlation (i.e., objectivist or subjectivist) and has rarely (prior to Kant) foregrounded the essential subject-object relation. To unpack the subject-object correlation, phenomenology approaches all experience as intentional, that is, object-directed. The human subject is embodied, embedded, and enactive in a world that it endows (‘constitutes’) with sense. Human conscious experience, then, is a ‘sense-giving’ (Sinnbegung) or meaning-constituting enterprise. As the phenomenological psychologist Aron Gurwitsch put it (Gurwitsch 2009: 155), to experience a conscious act is to actualize a sense. Explicit propositional judgments and thoughts express articulated subject-predicate meanings, but perceivings, rememberings, imaginings are also objectually shaped (one imagines something), as are hopes, fears, feelings of sadness or elation, and so on. The objects intended and their ‘modes of givenness’ can be quite diverse and appear in different ways to different mental attitudes (so an artwork can captivate the senses but can also be an object of rational scrutiny or economic appraisal). Phenomenology rejects all accounts of knowledge that posited the existence of meaningless ‘sense data’ that have to be processed by the mind. Rather sense perceptions, and even feelings, emotions and moods, all present themselves with intrinsic meaningfulness that conveys the ‘world’ to us in a special way. The world has multiple modes of senseful ‘givenness’ or ‘phenomenality’ to human conscious subjectivity. This pre-propositional, intuitive ‘givenness’ (Gegebenheit) of the world as meaningful is, phenomenology maintains, the foundational basis for propositional knowledge. Accordingly, phenomenologists pay more attention to the original connection between subjects and their world, namely the embodied, embedded, and enactive presence that makes the world phenomenally present to the subject. Bodily capacities and movements disclose the world in certain ways and these are largely intuitive and unconscious. I instinctively have access to the range of bodily movements that produces a particular action in the world, for example, I can freely vary my touching of the button with my finger or my elbow. I can lean closer to see something or step closer. I am in possession of a range of bodily capacities that I activate freely to apprehend the world.

Describing the Life of Consciousness Husserlian phenomenology (Husserl in part was inspired by William James) showed a specific interest in the dynamics of consciousness, at the very time when experimental psychology largely embraced behaviorism as a methodology with an explicit exclusion of all references to consciousness. For phenomenology, consciousness is a multilayered, dynamic, flowing whole, a unified ‘complex’ or ‘nexus’ (Zusammenhang), a ‘field’ (Feld, champ, in Aron Gurwitsch’s terms,

169

Dermot Moran

following Gestalt psychology, Moran 2019).2 Already in the Logical Investigations Husserl sees a ‘field of meaning’ as an interconnected whole: As regards the field of meaning, the briefest consideration will show up our unfreedom in binding meanings to meanings, so that we cannot juggle at will with the elements of a significantly given, connected unity. (Husserl 2001, vol. 2: 62) Consciousness is a seamless, unified, living stream, a temporally unfolding, interacting web of interrelated emotional and affective states, desires, feelings, moods, and so on. Although this unity is experienced by the subject as a seamless, flowing whole, it is composed of complex, a priori, eidetic structures that can be mapped through a reflective practice. Perceptions are modified into memories, all the while the sense of time continues in the background. Phenomenology thus seeks to uncover the alphabet or “ABC of consciousness” (in Husserl’s phrase), its grammar and syntax; but also how the interlocking experiences “interpenetrate or intersaturate” (Husserl 1997: 62). Husserlian phenomenology, moreover, maintains there are many layers or strata to consciousness; knowing or cognition takes place on multiple levels. Experiences, acts and attitudes (Husserl’s ‘position-takings’, Stellungnahmen, explicit stances taken whether in the form of judgments or points of view) are founded on one another, interpenetrate, and modify one another. Furthermore, much of our experience (not just as children but as adults) is pre-linguistic and not formulated in explicit propositional thoughts. We simply see, touch, and feel things in our environment. Explicit perceptual judgments are founded on sensuous perceptions that already contain a degree of complexity. I do not just see patches of color and shapes but specific states of affairs, processes, and events. I see the bird-flying in a single, yet complexly stratified visual seeing, and then I may judge (perhaps pre-verbally) that the bird is flying. This complex pre-linguistic perception can become the evidential basis for any number of linguistic statements about the situation, for example, ‘I saw a blackbird’, ‘That bird flew in a startled manner’, and so on. For Husserl, the sensuous seeing is saturated with significance and always contains more than the articulated perceptual judgments that are founded on it. In this sense, our perceptual connection with the world has an intrinsic richness that is never captured completely in our explicit judgments. Husserl (followed by Merleau-Ponty) also emphasize that our responses to experience are not governed by strict causality but in terms of what he and others called motivation (Husserl 1989: 231–259). There is no strict cause-effect relation in human action. Rather there is an inherent ‘ambiguity’ (Merleau-Ponty) or freedom in conscious responses. There are certain pathways available that allow one action to proceed to the next. A single thought makes sense within a complex of motivated intentions and fulfilments. I can feel the need to open the window because the room is stuffy, but I decide not to, because of the noise in the street outside. Motivation is closely connected with habit and association. In a response, I can follow blindly or I can also take a decisive stand. I may have a desire to smoke but I resist because I consider that smoking is harmful to my health. I come to see my desire as unwelcome but perhaps I still give in to the desire to smoke while taking a negative stance toward it. All this can take place at an intuitive, pre-verbal level. There is a motivated chain of experiences and attitudes that evolve in a specific way because of my character, my mood, my subjective states. In the phenomenological tradition, cognition is founded not just on embodied perception but also on feelings, moods, the general sphere of affectivity (also embodied), that are deeply involved in shaping our experience as meaningful. Pervasive moods, such as anxiety, as Heidegger explicates in Being and Time (Heidegger 1927: 172–188) provide an overall affective framework 170

Husserl on Habit, Horizons, and Background

through which our sense of world is primordially ‘disclosed’. One’s overall mood determines how emotions can display themselves, their temporal trajectory, and so on. It is within overall moods that particular emotions, such as anger or joy, get their peculiar configuration. Propositional thought, too, is cradled in a more pervasive state of mind or mood, such that our thoughts might be contemplative or agitated, imbued with urgency, or ‘wanting to get things’ done. Moods simply befall us. One is always in a mood. Even casual everyday normality is a mood although it is usually noticed only when it is disturbed by another ‘counter-mood’ (Heidegger 1927: 175). In this sense, moods belong to the domain of implicit cognition; through moods we make sense of our experience.

Tacit and Reflective Self-Awareness Husserl thought of phenomenological analysis as proceeding in self-conscious reflection. However, he also maintained that self-reflection presupposes an underlying, more basic level of ‘unreflected’ or ‘pre-reflective’ consciousness (Husserl 1989: 259). The self or ego is not necessarily fully present in a self-aware manner in our direct object-oriented experiences. We do not have complete possession of our egoic experiences. Normally, we are focused outside ourselves. Jean-Paul Sartre famously discusses this ‘first-order’ ‘unreflective’ consciousness in The Transcendence of the Ego (Sartre 1937) in an example repeated by Dreyfus 2005; Kelly 2000). In immediate, direct consciousness one focuses on the object and one is not aware of oneself except in a ‘non-positional’ way (Sartre 1937: 7). There is only the ‘street-car-to-be-caught’. Sartre writes: When I run after a tram, when I look at the time, when I become absorbed in the contemplation of a portrait, there is no I. There is a consciousness of the tram-needingto-be-caught, etc., and a non-positional consciousness of consciousness. In fact, I am then plunged into the world of objects, it is they which constitute the unity of my consciousnesses, which present themselves with values, attractive and repulsive values, but as for me, I have disappeared, I have annihilated myself. There is no place for me at this level, and this is not the result of some chance, some momentary failure of attention: it stems from the very structure of consciousness. (Sartre 1937: 8) Self-awareness, on this account comes as a new and secondary experience, building on a pre-reflective experience, as when one experiences the gaze of others, promoting shame, for instance (when I see myself as the other sees me, I experience shame). Merleau-Ponty similarly rejects a purely intellectualist or cognitive reading of the Cartesian cogito. Beneath the reflexive cogito, is a ‘pre-reflective’ or ‘tacit’ cogito, that is, a sense of the ‘I am’ as given in my immediately bodily self-presence as ‘me-here-now’ which is not yet articulated into the reflective reasoning form, ‘I think therefore I am’, that is mediated through language (Merleau-Ponty 1945: 422). The active ‘performance’ of the cogito is merely a reflective highlighting of what is already tacitly present. Merleau-Ponty further maintains that I encounter my thoughts in ways I had not consciously expected. He writes in his late The Visible and the Invisible: Genuine conversation gives me access to thoughts that I did not know myself capable of, that I was not capable of, and sometimes I feel myself followed in a route unknown to myself which my words, cast back by the other, are in the process of tracing out for me. (Merleau-Ponty 1968: 13) 171

Dermot Moran

There is a vast domain of pre-reflective conscious experience, not just at the perceptual but at the cognitive level. I am not always aware of what I am going to say next. Sometimes, something just ‘slips out’ and I see my own understanding in a new way.

The Phenomenology of Embodied Perception Merleau-Ponty, especially in his Phenomenology of Perception (1945), greatly elaborated on Husserl’s accounts of this pre-theoretical, preconscious experience that is closely connected to the motor significance of the body. For phenomenologists, the ‘perceptual body’ or ‘phenomenal body’ (Gurwitsch) is the opposite of the body as described by science. Gurwitch writes (in his review of Merleau-Ponty’s Phenomenology of Perception) of Merleau-Ponty’s conception of the body: The body in question is not the organism in the sense of biological science but, on the contrary, the phenomenal body, the body with which I live, which I experience as mine, which defines my situation within and my point of view upon the world. (Gurwitsch 2010b: 493) The body implicitly possesses an orientation, and is responsible for giving a sense of what is up, down, near, far, reachable or not reachable. As Husserl puts it, the body is the ‘zeropoint’ (Nullpunkt) or orientation. Husserl and Merleau-Ponty both begin from the complexity of multimodal sensuous perception as the anchor-point for our embodied and embedded being-in-the-world (Wheeler 2005). Perception has more thickness and complexity that classical empiricism appreciated. Perception is a whole-body activity involving the ‘interlacing’ and ‘overlapping’ (Merleau-Ponty’s terms) of the senses (seeing, hearing, tasting touching, smelling, feeling), combined with proprioceptive experiences of bodily movement (Merleau-Ponty’s ‘motility’), balance, orientation. The body also is experienced within the fields of vision and touch. There is also an overlapping and intertwining between the different sense modalities. Perception is multimodal; ‘the senses communicate among themselves’, as Merleau-Ponty puts it – I  grasp by different sensory pathways that a surface is rough (sight, touch). In paintings, Merleau-Ponty, writes I  see the raised, embossed pattern on the cloak in the painting even though I cannot touch it. Indeed, Merleau-Ponty thought there was a degree of synaesthesis in all perception. I see the surface as rough and scratchy on my skin even without touching it. Pre-linguistic bodily perception is our core being-in-the-world. Husserl speaks of perception as providing an ‘primordial belief ’ (Urdoxa, Husserl 1973a: 59), which Merleau-Ponty interprets as a ‘perceptual faith’ or belief in the world that is presupposed by science (MerleauPonty 1968: 14). Merleau-Ponty writes: The perceiving person is not spread out before himself in the manner that a consciousness must be; he has an historical thickness, he takes up a perceptual tradition, and he is confronted with a present. In perception we do not think the object and we do not think the thinking, we are directed toward the object and we merge with this body that knows more than we do about the world, about motives, and about the means available for accomplishing the synthesis. (Merleau-Ponty 1945: 247–248) The body has an inbuilt, antecedent knowledge; the body literally incorporates knowledge. Polanyi agrees: “Our body is the ultimate instrument of all our external knowledge, whether 172

Husserl on Habit, Horizons, and Background

intellectual or practical” (Polanyi 1966: 15). Bodily systems cooperate with each other to render our experience of the objective world. As Gurwitsch puts it in his review of Merleau-Ponty’s Phenomenology of Perception, the perceptual body is an interlacing system of competencies – a ‘synergetic system’: Since the body as a whole forms a synergetic system, the qualities mediated by the several sense organs (i.e., visual, auditory, tactile, and other) imply, symbolize, and modify each other. (Gurwitsch 2010b: 489) There are bodily self-movements integrated into all perceptual activities. For Husserl, visual perception involves a specific series of eye movements, tiltings of the neck, turning of the head or body, in the experience of looking at something (and Husserl noticed that one chain of movements can stand in for another, for example, I can turn just my head or rotate my body). The body instinctively substitutes one chain of movements with another to gain the same equivalent result. Similarly, touch requires movement (not just staccato tapping); one has to move one’s fingers over a surface to apprehend smoothness (a single touch will not yield smoothness, although it might yield hardness or softness). As Merleau-Ponty points out (Polanyi makes a similar point), in seeing, one is not aware of the accompanying blinking of the eyes, but only the vista to be seen. The whole of bodily perceptual experience involves implicit knowledge of this embodied but consciously felt kind. The hands ‘instinctively’ reach out to feel the surface of the desk or to steady oneself. This is perceptual knowledge even before it is articulated in explicit thoughts or words. Skilled performance takes this knowledge to a higher more expert level, but it is still primarily intuitive and inarticulate. A dancer can make minor adjustments to a pose without following an explicit map of rules (although beginners often use these rules to guide their practice). The primary experience of embodiment is sensuous, ‘instinctual’ and habitual. I  simply inhabit my body in an animated (Husserl 1989: 252, speaks of ‘ensouled’) manner. There is a receptivity in sensibility that is already an openness to the flow of the specific sensory fields. Furthermore, each individual has a unique embodiment, which Husserl calls ‘style’ or ‘habitus’ (Husserl 1989: 290), involving typical gestures, peculiar mannerisms, facial expressions, tone of voice, accent, walking gait, stance, pattern of thought or speech. Memories, skills, practical abilities are likewise literally incorporated into the body in an individual manner, in the way we hold ourselves, move our bodies, walk, sit, eat, look weary, adopt a defeated air, and so on (Young 1990; Sheets-Johnstone 2003). All these idiosyncrasies shape, inform, and characterize an individual’s style in a uniquely identifiable way. Moreover, the subject experiences her agency in and through this embodied style. Knowledge involves an integration of parts into a whole. We recognize a familiar face even if much is occluded. I recognize your face even if I do not consciously know the color of your eyes (or whether you have changed your hair). Indeed, facial recognition is a paradigm for implicit knowledge (cf. Polanyi 1966: 4). Similarly, one has an implicit bodily sense of how far one can reach or whether one can jump over a certain gap. Of course, this bodily intentionality can be ‘inhibited’ (Young 1990) for all kinds of reasons, including assumptions about age or gender, confidence, or ability, but that is not to deny a certain intrinsic bodily intentionality operating at an implicit level and providing a platform for conscious thought and action. Although Husserl was primarily a logician and epistemologist interested in systematic propositional knowledge connected through logical consequence, he also recognized that there is a deep, pre-propositional, presupposed, embodied knowledge, which yields vistas that are 173

Dermot Moran

passively ‘synthesized’ and combined into unified forms. In his Logical Investigations (Husserl 2001), Husserl describes how we not only perceive patterns in the carpet or on the tiled floor, but we also expect those patterns to continue under the table or desk to the hidden parts of the carpeted floor. There is a built-in, implicit anticipatory intention in our perceptual apprehension; our eyes move towards to limit or horizon of our viewing, but it is not at the level of an expectation. Husserl writes penetratingly: Intention is not expectancy, it is not of its essence to be directed to future appearances. If I see an incomplete pattern, e.g. in this carpet partially covered over by furniture, the piece I see seems clothed with intentions pointing to further completions – we feel as if the lines and coloured shapes go on ‘in the sense’ of what we see – but we expect nothing. It would be possible for us to expect something, if movement promised us further views. But possible expectations, or occasions for possible expectations, are not themselves expectations. (Husserl 2001, vol. 2: 211) From the outset of his mathematical and logical investigations, Husserl was fascinated by the way one can apprehend intuitively not just individual entities (grasped as ‘unities’) but also collectivities, groups or sets of things – flocks of birds, herds of cattle, fields of grass, a mountain range. Husserl thought of these groups as grasped by a particular intuitive act that he called ‘collective combination’ (Husserl 2003). For him it was the pre-theoretical basis for acts of counting – first I  have to apprehend a group of items, before I  traverse them or colligate them in an act of explicit counting. So, right from the beginning, Husserl was concerned especially with implicit perceptual knowledge. He starts from what he calls generally ‘intuition’ (Anschauung), the direct sensuously founded perception or insight not yet mediated by concepts. Husserl begins from the initial passivity or receptivity of experience: noises, colors, tickles, itchings, scratchings, impose themselves on consciousness. They have a salience over the receding or sunken background. A sudden noise wakes me from my reverie and I turn my attention towards it. A pre-linguistic child can see and react to the pattern on the carpet, is drawn towards the flickering light coming through the window. Furthermore, our perceptual awareness is always editing out or selecting what we choose to perceive or not (Polanyi speaks about ‘attending from’, Polanyi 1966: 10). Thus, I look at the surface of the table in sunlight and I literally see the table as flat, smooth and uniformly colored, even though there are clearly patches of sunlight and shadow crisscrossing the table that make the surface appear mottled, and, if I look at it again, I can pay attention to this explicit patterning of light and shade on what I had initially apprehended as a unified colored surface. A painter will need to attend to that mottled surface to render it on the canvas, whereas someone buying the table as furniture attends only to its uniformly colored surface and excludes the mottled patterns of light. There are different forms of ‘givenness’ of the object depending on the intuitive intentional approach.

Passive Synthesis The later Husserl, in Experience and Judgment (Husserl 1973a), introduces a new and paradoxical notion of ‘passive synthesis’ that is supposed to capture the genesis of categoriality in sensibility, offering a critique of Kant (for whom all syntheses are active). Husserl speaks of a “passive agreement of intentionalities in a synthetic unity” (Husserl 1973a, 62) and of “the world as the universal ground of belief ‘pregiven’ (vorgegeben) for every experience of individual objects” (Husserl 1973a: 28). For Husserl, pre-theoretical, ‘prepredicative’ judgments, 174

Husserl on Habit, Horizons, and Background

for example, perceptions form the basis for explicit predicative judgments (Husserl 1973a: 61). In this work, Husserl claims that explicit logical acts, such as negation, have their foundation in the pre-predicative sphere. Thus, he writes: It thus appears that negation is not first the business of the act of predicative judgment but that in its original form it already appears in the prepredicative sphere of receptive experience. (Husserl 1973a: 90) I have to have an intuitive sense of something being rejected before I can make an explicit judgmental act of denial.3 Similarly, mental experiences such as doubt are already anchored in bodily perception, when the continuous flow of harmonious confirmations is disrupted, for example, in looking at a man turns out to be disrupted when the man turns out to be a mannequin. Again Husserl writes: One and the same complex of sense data is the common foundation of two apprehensions superimposed on each other. Neither of the two is canceled out during the period of the doubt. They stand in mutual conflict; each one has in a certain way its own force, each is motivated, almost summoned, by the preceding perceptual situation and its intentional content. But demand is opposed to demand; one challenges the other, and vice versa. In doubt, there remains an undecided conflict. (Husserl 1973a: 92) One further feature of this passivity is our experience of the temporal flow of experience itself. All experience is temporally unified and cumulatively compounded in an organic way. As Merleau-Ponty puts it: “In every movement of focusing, my body tied a present, a past and a future together. It secretes time” (Merleau-Ponty 1945: 249). For Husserl, time is the deepest, most fundamental organizing feature of consciousness. It is a kind of passive synthesis, joining one moment to the next to yield the sense of harmonious experience. Perception is dynamic, self-organizing, and constitutes a harmonious continuum that is glued together without my specific active synthetic intervention. The inner coherence or organization of the field of consciousness occurs on both what Husserl terms the noetic (or experiential act) and the noematic (or objectual content) sides. I have a sense of continuity of myself (I wake up as the same person each morning) and of my surrounding world (my room). The data in a field are unified by their relevance with each other, as Gurwitsch elaborates. The field of consciousness is always structured with a ‘focus’ or ‘theme’ (of varying width) of attention surrounded by a structured periphery of inattentional contents. I am only vaguely aware of the world outside my window, of the weather, of the clouds. Gurwitsch and Merleau-Ponty both describe in detail the inner organization of our perceptual fields. The field of perception consists not just of the actual contents present (whether sharply given or vaguely apprehended) but also of a background of potentialities and “inactualities that constitute a ‘field of freedom’ ” (Gurwitsch 2010a: 197). This field contains affordances and solicitations, to use the language of J.J. Gibson (1979), which he encountered in Husserlian phenomenology through his teacher David Katz, a student of Husserl, see Moran 2015). A climber will see a rockface as offering footholds and handholds that simply stand out. Thus, Gurwitsch states: “All perceptual consciousness is supported and pervaded by an inexplicit, unformulated, and silent reliance on the familiarity of the world” (Gurwitsch 2010b: 488). Furthermore, as Husserl puts it, “everything unfamiliar is the horizon of something familiar” (Husserl 2014: 116). Experience 175

Dermot Moran

has a gradient from familiar to unfamiliar. I can only grasp what is unfamiliar within an overall horizon of familiarity. I reach for the hotel doorhandle even if it is not shaped like my doorhandle at home. There is a passive knowledge guiding my hand to grip the handle.

The Habitual Body Phenomenologists emphasize how this deep, instinctual, inner embodied awareness becomes ‘sedimented’ (Husserl’s term) or bedded down in habit and shaped by wider cultural tradition. Habit is associated with disposition, possession, skill, performance of routines, the very embodiment of activities (the pianist’s fingers reach for the appropriate keys). According to Husserl, moreover, habit (Habitus, Habitualität, Hexis, Habe) belongs with association, memory, temporal synthesis, to the essence of consciousness (Husserl 1989: 143). Habit, for Husserl, plays an essential role in the constitution of meaningfulness (Sinnhaftigkeit) at all levels: in perceptual experience, the developmental formation of the self, to the development of wider social bonds, establishing history and tradition. Habit is the glue to binds together the harmonious course of life in the world.4 Each individual possesses a corporeal or bodily habitus (Husserl uses Habitus, Husserl 1973b, 76). In medicine, the term ‘habitus’ refers to someone’s overall ‘bearing’, ‘form’, how they present themselves (Husserl speaks of a ‘habitual style’, Husserl 1989: 260n. 1). There is, furthermore, a degree of contingency or ‘facticity’ in bodily givenness. Some people simply have better ‘innate’ or ‘natural’ balance, a pregiven ability to navigate water with less effort when swimming, a capacity to retain musical sequences. One may take joy in hearing specific sounds, another not (Husserl speaks of this as belonging to sheer facticity, Husserl 1989: 288), and so on. Husserl and Stein recognize that embodiment gives one particular traits and disposition that belong to ‘nature’ (Husserl 1989: 289). One person has longer limbs than another. Such natural capabilities (perceptive, proprioceptive, perceptive, cognitive), of course, may be isolated, strengthened and indeed fine-tuned by training (balance can be tuned by visualization techniques for instance). As Edith Stein puts it, capacities can be strengthened through ‘habituation’ (Stein 1989: 51). Intellectual and explicit thematic structures can be imposed on this bodily pre-givenness, but only within certain limits. One cannot become anything one likes. Habits, furthermore, can be passive or active. As Husserl writes in Ideas II: Habits are necessarily formed, just as much with regard to originally instinctive behavior . . . as with regard to free behavior. To yield to a drive establishes the drive to yield: habitually. Likewise, to let oneself be determined by a value-motive and to resist a drive establishes a tendency (a “drive”) to let oneself be determined once again by such a value-motive . . . and to resist these drives. (Husserl 1989: 267, translation modified) Habits and skills are in a continuum between actions have a degree of purposive intentionality, and more ‘automatic’ or ‘mechanical’ forms of behavior. Of course, the margins between conscious and preconscious experience are very difficult to delimit precisely. According to Merleau-Ponty, habitus has to do primarily with our ambiguous bodily insertion in the world. Merleau-Ponty in particular characterizes different kinds or levels of knowing (savoir) that he uses to explain the case of the phantom limb and similar embodied experiences: the ambiguity of knowledge amounts to this: it is as though my body comprises two distinct layers (couches) that of the habitual body (corps habituel) and that of the actual

176

Husserl on Habit, Horizons, and Background

body (corps actuel). Gestures of manipulation that appear in the first have disappeared in the second, and the problem of how I can feel endowed with a limb I no longer have comes down to knowing how the habitual body can act as a guarantee for the actual body. (Merleau-Ponty 1945: 84) Merleau-Ponty thinks of habit as also providing a kind of anticipatory knowledge, for example, one walks through a low doorway without first measuring it or one knowing intuitively that my car will fit through the gap when passing (Merleau-Ponty 1945: 144). We essentially incorporate ourselves in the car, into the door, which are not encountered as external objects but rather instruments expressing my abilities. As Merleau-Ponty summarizes: Habit expresses our power of dilating our being in the world, or of altering our existence through incorporating new instruments. (Merleau-Ponty 1945: 145)

Skillful Coping In a recent discussion, the Berkeley philosopher Hubert L. Dreyfus has combined MerleauPonty’s phenomenology of motor intentionality with Heidegger’s concept of everyday beingin-the-world to offer his own interpretation of everyday expertise, or ‘skillful coping’, which prioritizes bodily response and claims to avoid a Cartesian intellectualist (rule-following) and representationalist construal (Dreyfus 2002a, 2002b, 2005). Dreyfus argues that, for an expert practitioner, the action must be a form of absorbed coping, where no degree of self-aware ego is prominent and there is no implicit or explicit conceptualization or rule representing. Countering Dreyfus, John McDowell insists on practice being permeated by a degree of self-awareness. The Dreyfus/McDowell debate highlights the issue of whether habitual action requires conscious deliberation or is illuminated by recourse to reasons or even some kind of self-awareness. Phenomenologists tend to emphasize latent, habitual, embodied knowledge that is not conceptually shaped as the starting point. Indeed, higher level intellectual reasoning is also shaped by habit and make follow largely instinctual pathways.5 Polanyi claimed that skillful practices follow rules of which the follower is not aware (Polanyi 1958: 51), but Dreyfus thinks there is no rule following involved at all. Polanyi also points out that too much self-consciousness can intrude – he mentions the phenomenon of ‘stage fright’ (Polanyi 1958: 58), where an actor becomes overly self-conscious. In this regard Dreyfus and Polanyi agree that intrusion of the self into the process can disrupt the absorbed coping. But the phenomenological dimensions of this ‘coping’ need to be much more clearly articulated. Someone skilled in mental arithmetic will be able to ‘see’ certain conclusions of calculations without explicitly running through them. A writer may habitually invoke certain metaphors and never use others even when consciously composing a text. There is much debate about the kinds of ‘knowledge’ or ‘cognition’ incorporated in the body. The main phenomenological features are that this implicit cognition is experienced directly in a felt first – person way, can be passive or active, but it is always temporally unfolding, dynamic, and elaborated in phases. For Husserl, higher cognitive processes such negation or doubt are anchored in perceptual experiences of cancellation or conflict. For Husserl as well as for James, the seamless continuity and flow of conscious experience itself involves a series of harmonious syntheses.

177

Dermot Moran

The Concept of Horizon Increasingly, Husserl and the phenomenological tradition recognized the importance of consciousness of horizon, or what Husserl called ‘horizon-intentionality’. Husserl introduced the concept of ‘horizon’ as “ ‘what is co-given’ but not genuinely” in Ideas I (Husserl 2014: 77), and it is given explicit treatment in Experience and Judgment § 33 where he speaks of “the horizon of typical pre-acquaintance in which every object in pregiven” (Husserl 1973a: 150). Husserl distinguishes between the object apprehended and the ‘potential perceptual field’ (Husserl 2014: 162) surrounding it. This constitutes a ‘background’ of actualizations and also of ‘inactualities’ that constitute a ‘field of freedom’ (Gurwitsch 2010a: 197). I see an object from one side but I have the implicit awareness that it has other sides and I can traverse those sides by walking around it or moving the object around. All experience possesses a focal point and a wider field or context (Husserl borrows the terms ‘fringe’ and ‘horizon’ from William James). Both focal point and horizon are apprehended intuitively and without language. For example, I pick up my pen to write, but maybe I also use it absent-mindedly as to lever a paperclip. The object has an intrinsic openness to become thematic in new, unforeseen contexts (the screwdriver used to open a paint tin). The theme furthermore can also be ‘released’ back into the thematic field (Gurwitsch 2010a: 254). There is an implicit accompanying consciousness of the horizon when one is perceiving or thinking about an intentional object. Horizon-consciousness, for Husserl is indefinite and empty but it has a particular character relative to the theme. There is always what is relevant or irrelevant, interesting or uninteresting, wrapped up in the experience. Every experience has specifically and lawfully determined but also essentially unlimited horizons of intentional implication, including not just what is actually given but also available potentialities and possibilities in which such intentional objects are apprehended and made meaningful. As Gurwitsch elaborates, the proposition (statement or state of affairs) being focusing on is the theme but it always indicates another such that a chain of implications becomes visible. The theme unfolds in an overall ‘meaning-field’ (Gurwitsch 2009: 317). To invoke Gurwitsch’s example, each mathematical problem has its theme, field, and horizons. These horizons are only vaguely apprehended – they present as possibilities that can be actively traversed. Some mathematicians will just ‘see’ or ‘intuit’ these horizons better than others. Nevertheless, these possibilities are, for phenomenology, predelineated according to the particular essence of the tool in question, depending on the horizonal foresight of the user. In Husserl’s terms, we have tacit knowledge of the overall horizon or context of a problem and motivation to find a solution and this motivation may not need to be articulated (I simply want to discover the truth). The manner we experience the wider context of our intersubjective, cultural lives (our implicit knowledge of our language or religious tradition) is also manifest in this horizonal way. An individual has a broad, vague, passively available access to the known vocabulary in her or her own language or to the rituals in her religious tradition. Sometimes, specific words may have to be a actively searched in memory, but mostly, reasonably appropriate words present themselves effortlessly to the speaker. Each speaker depends upon the wider context of the language as known to that speaker and in dialogue one’s horizons are further expanded. Indeed, Hans-Georg Gadamer, drawing on phenomenology speaks of mutual understanding as involving ‘fusion of horizons’ (Horizont-Verschmelzung, Gadamer 1960), seeing beyond what is immediately present. Similarly, one has an implicit, unarticulated sense of one’s family, friends, familiars, customs, habits, traditions, and so on. One knows one’s place in the social order. This is a vast domain of mostly passive, unarticulated knowledge that makes possible other kinds of articulated acting 178

Husserl on Habit, Horizons, and Background

and knowing (such as deciding to visit my cousin, or choosing to stay in Anglophone countries). Phenomenologists have explored how such traditional knowledge becomes embedded or sedimented, how it is transmitted to others, but also how it can be distorted or worn down to become an impediment or prejudice. The wine-taster or tea-taster (what Polanyi calls the ‘connoisseur’) has knowledge that comes from deep experience and copying the mastery of others. This habitual knowledge is passed on through apprenticeship and training and similarly skills can be lost.6 Heidegger, for instance, discusses how a work of art belongs to a world and if that world becomes lost (e.g. the world that believed in the Greek gods), then the art-work can no longer exhibit that world, but perhaps enters another context, for example, as a museum piece representing ancient Greek sculpting technique (but no longer radiating the presence of a god). Art objects and indeed all cultural objects and practices depend heavily on the contexts and horizons within which they are disclosed and which shape their meaningfulness to the subject.

Conclusion Overall, the phenomenological tradition with its detailed analyses of the perceptual process as a bodily incorporated ‘knowledge’ that involves a range of bodily movements coordinated with disclosure of properties of the object, habitual skillful action, intuitive awareness of the focal point (‘theme’) of a particular act of concentration, and also of the horizons of our objects and actions, as well as the overall implicit background awareness of culture and tradition, offer extremely rich discussions of implicit cognition that deserve much closer scrutiny by the cognitive sciences and philosophies of mind and action. Classical phenomenology (Husserl and his followers) have a deep analysis of knowledge and agency as based on and enabled by implicit bodily awareness and situated overall in broader horizons of significance and possibilities. This broader understanding of implicit cognition deserves much deeper scrutiny.

Related Topics Chapters 13, 14, 15, 23, 24

Notes 1 Both Ryle and Polanyi had exposure to European philosophy. Ryle was familiar with phenomenology (Ryle 1932), especially Husserl, Heidegger (Ryle reviewed Being and Time, Ryle 1928), and MerleauPonty (O’Connor 2012), whereas Polanyi drew on Gestalt psychology (Polanyi 1958; Fuchs 2001), and was familiar with Ryle’s concept of ‘knowing how’ (Polanyi 1966: 7). Polanyi was not favorably disposed to the analytic currents of his day (logical positivism, behaviorism, ordinary language philosophy, as he encountered them in Manchester and Oxford). In fact, Polanyi’s notion of ‘indwelling’, is his adaption of ‘empathy’ (Einfühlung), or sympathetic understanding, found in Theodor Lipps and Wilhelm Dilthey (Polanyi 1966: 16–17). For Polanyi, indwelling does not distinguish the human sciences from natural sciences, rather it is involved in all knowing. It is by dwelling in things that we learn to see them (Polanyi 1966: 18). Although not an existentialist (he rejected Sartre’s idea that we can choose our values), Polanyi saw knowledge as a committed, existential act guided by values that are tacitly grasped. Indeed, Polanyi asserts that “into every act of knowing there enters a passionate contribution of the person knowing what is being known, and that this coefficient is no mere imperfection but a vital component of his knowledge” (Polanyi 1958: v), and he maintains that it is “our personal participation that governs the richness of concrete experience to which our speech can refer” (Polanyi 1958: 90). 2 Having studied initially with Husserl, Gurwitsch went on to study at the University of Frankfurt with the renowned Gestalt psychologists Adhemar Gelb and Kurt Goldstein. Gestalt psychology approaches conscious experiences in a holistic way as a field made up of interconnected moments rather than distinct parts) organized in a coherent structure. Husserl too speaks regularly of the ‘field of intuition’ (Husserl

179

Dermot Moran 2014: 56), ‘sensory fields’ (Sinnesfelder, Husserl 2014: 73), ‘fields of perception’ (Wahrnehmungsfeld, Husserl 2014: 51), ‘fields of memory’ (Erinnerungsfelder, Husserl 2014: 51), ‘field of view’ (Blickfeld), and so on. 3 This phenomenon is important for the phenomenology of disability and informed consent. A person can refuse food by keeping their mouth closed, for example, even if they cannot articulate that refusal. Bodily resistance can be understood as itself an act of refusal or negation. 4 Husserl’s analyses of habit (Husserl 1973a, 1989) deeply influenced the French sociologist Pierre Bourdieu in his own influential discussions (Bourdieu 1977, 1985, 1990). 5 Polanyi, who spoke several languages, gives the example of reading his morning correspondence that arrives in several languages, but he is so intent on the meaning of the letters that he does not notice explicitly or remember in what language the letter is written (Polanyi 1958: 59). 6 As Polanyi writes: “It follows that an art which has fallen into disuse for the period of a generation is altogether lost. There are hundreds of examples of this to which the process of mechanization is continuously adding new ones. These losses are usually irretrievable. It is pathetic to watch the endless efforts – equipped with microscopy and chemistry, with mathematics and electronics – to reproduce a single violin of the kind the half-literate Stradivarius turned out as a matter of routine more than 200 years ago” (Polanyi 1958: 55).

References Bourdieu, P. 1977. Outline of a theory of practice. Cambridge: Cambridge University Press. Bourdieu, P. 1985. “The genesis of the concepts of habitus and field”. Sociocriticism, 2: 11–24. Bourdieu, P. 1990. “Structures, habitus, practices”. In P. Bourdieu, ed., The logic of practice. Stanford: Stanford University Press: 52–65. Dreyfus, H. L. 2002a. “Intelligence without representation: the relevance of phenomenology to scientific explanation”. Phenomenology and the Cognitive Sciences, 1: 367–383. Dreyfus, H. L. 2002b. “Refocusing the question: can there be skillful coping without propositional representations or brain representations?”. Phenomenology and the Cognitive Sciences, 1: 413–425. Dreyfus, H. L. 2005. “Overcoming the myth of the mental: how philosophers can profit from the phenomenology of everyday expertise”. Proceedings and Addresses of the American Philosophical Association, 79: 47–65. Dreyfus, H. L. 2007. “Detachment, involvement, and rationality: are we essentially rational animals?”. Human Affairs, 17: 101–109. Dreyfus, H. L., and Taylor, C. 2015. Retrieving realism. Cambridge, MA: Harvard University Press. Fodor, J. 1968. “The appeal to tacit knowledge in psychological explanation”. Journal of Philosophy, 65: 627–640. Fuchs, T. 2001. “The tacit dimension”. Philosophy, Psychiatry & Psychology, 8: 323–326. Gadamer, H-G. 1960. Truth and method. Trans. J. Weinsheimer and D. G. Marshall. 1989. New York: Crossroad. Gibson, J. J. 1979. The ecological approach to visual perception. Boston: Houghton Mifflin. Gurwitsch, A. 2009. The collected works of Aron Gurwitsch (1901–1973), vol. II: Studies in phenomenology and psychology. Ed. F. Kersten. Dordrecht: Springer. Gurwitsch, A. 2010a. The collected works of Aron Gurwitsch (1901–1973), vol. III: The field of consciousness: phenomenology of theme, thematic field, and marginal consciousness. Eds. P. Zaner and L. Embree. Dordrecht: Springer. Gurwitsch, A. 2010b. The collected works of Aron Gurwitsch (1901–1973), vol. I: Constitutive phenomenology in historical perspective. J. García-Gómez, ed. Dordrecht: Springer. Heidegger, M. 1927. Being and time. Trans. J. Macquarrie and E. Robinson. 1962. New York: Harper & Row. Husserl, E. 1973a. Experience and judgment: investigations in a genealogy of logic. Trans. J. S. Churchill and K. Ameriks. Ed. L. Landgrebe. Evanston, IL: Northwestern University Press. Husserl, E. 1973b. Zur Phänomenologie der Intersubjektivität. Texte Aus Dem Nachlass. Erster Teil. 1905–1920. In I. Kern, ed., Husserliana XIII. The Hague: Nijhoff. Husserl, E. 1989. Ideas pertaining to a pure phenomenology and to a phenomenological philosophy, second book. Trans. R. Rojcewicz and A. Schuwer. Dordrecht: Kluwer. Husserl, E. 1997. Thing and space: lectures of 1907. In Edmund Husserl – collected works, vol. vii. Trans. R. Rojcewicz. Dordrecht: Springer.

180

Husserl on Habit, Horizons, and Background Husserl, E. 2001. Logical investigations. 2 volumes. Trans. J. N. Findlay. Ed. D. Moran. London and New York: Routledge. Husserl, E. 2003. Philosophy of arithmetic: psychological and logical investigations with supplementary texts from 1887–1901. Trans. D. Willard. Netherlands: Springer. Husserl, E. 2014. Ideas for a pure phenomenology and phenomenological philosophy. First book: general introduction to pure phenomenology. Trans. D. Dahlstrom. Indianapolis: Hackett Publishing Company. Kelly, S. D. 2000. “Grasping at straws: motor intentionality and the cognitive science of skillful action”. In J. Malpas and M. Wrathall, eds., Heidegger, coping, & cognitive science. Cambridge, MA: MIT Press: 161–177. Merleau-Ponty, M. 1945. Phenomenology of perception. Trans. D. Landes. 2012. London: Routledge. Merleau-Ponty, M. 1968. The visible and the invisible, followed by working notes. Trans. A. Lingis. Evanston: Northwestern University Press. Moran, D. 2000. Introduction to phenomenology. London and New York: Routledge. Moran, D. 2011. “Edmund Husserl’s phenomenology of habituality and habitus”. Journal of the British Society for Phenomenology, 42: 53–77. Moran, D. 2015. “Phenomenologies of vision and touch: between Husserl and Merleau-Ponty”. In R. Kearney and B. Treanor, eds., Carnal hermeneutics: perspectives in continental philosophy. New York: Fordham University Press: 214–234. Moran, D. 2019. “Husserl and Gurwitsch on horizonal intentionality. The Gurwitch Memorial Lecture 2018”. Journal of Phenomenological Psychology, 50: 1–41. O’Connor, J. K. 2012. “Category mistakes and logical grammar: Ryle’s Husserlian tutelage”. Symposium: Canadian journal of continental philosophy/Revue canadienne de philosophie continentale, 16: 235–250. Polanyi, M. 1958. Personal knowledge: towards a post-critical philosophy. Corrected Edn. 2005. London: Routledge. Polanyi, M. 1966. The tacit dimension. 2009. Chicago: University of Chicago Press. Ryle, G. 1928. “Heidegger’s Sein und Zeit”. Mind, XXXVIII, reprinted in Ryle, Collected papers. Vol. 1: critical essays. 1971. London: Hutchinson & Co: 197–214. Ryle, G. 1932. “Phenomenology”. Proceedings of the Aristotelian Society, XI, reprinted in Ryle, Collected papers. Vol. 1: critical essays. 1971. London: Hutchinson & Co: 167–178. Ryle, G. 1949. The concept of mind. London: Hutchison. Sartre, J-P. 1937. The transcendence of the ego. Trans. A. Brown. 2004. London and New York: Routledge. Sheets-Johnstone, M. 2003. “Kinesthetic memory”. Theoria et Historia Scientiarum, 7: 69–92. Stanley, J., and Williamson, T. 2001. “Knowing how”. Journal of Philosophy, 98: 411–444. Stein, E. 1989. On the problem of empathy. Trans. W. Stein. Washington, DC: ICS Publications. Wheeler, M. 2005. Reconstructing the cognitive world: the next step. Cambridge, MA: MIT Press. Young, I. M. 1990. Throwing like a girl. Bloomington: Indiana University Press.

181

13 POLANYI AND TACIT KNOWLEDGE Stephen Turner

Michael Polanyi is responsible for the popularity of the term “tacit knowledge,” and for developing the concept in its multiple aspects by generalizing from the case he makes for its role in science to all forms of intelligent engagement with the world. By identifying it with science, and arguing that tacit knowledge was indispensable for science, he assured its respectability, but at the same time assured that it would be controversial, and it has been. This entry will provide a brief account of the concept, the interpretive controversies surrounding it, and the criticisms of it. The core of the idea of tacit knowledge is contained in the slogan, “we know more than we can tell” (Polanyi 1966: 4). Tacit knowledge is knowledge that we cannot articulate. But beyond this slogan, and even in the interpretation of it (including the meaning of the “can” in the slogan), there is a great deal of controversy, some of which Polanyi himself is responsible for. He changed his usages over time, and reinterpreted his own prior statements about the term and the terms he associated with it, and used a variety of terms normally associated with the tacit, such as “premises,” in ways that seemingly refer to tacit epistemes and presuppositions, despite being opposed to and rejecting the philosophical theories that the use of these terms was associated with. To understand the Polanyian conception of tacit knowledge requires us to cut through these confusions, many of which are generated by his own (often shifting) usages. At the center of his own distinctive conception there are two ways of speaking about articulating the tacit. The positive conception is concerned with the process of knowing, and the dependence of this process on elements of the knowing process that cannot be articulated, the paradigm of which is the process of scientific discovery. The tacit in these cases he assimilates to skills and comments that it is “continuous in its inarticulateness with the knowledge possessed by animals and infants, who .  .  . also possess the capacity for reorganizing their inarticulate knowledge and using it as an interpretative framework” ([1958] 1962: 90). The other conception is concerned with unaccountable elements, particularly in verbalization, which pose limits to such things as the formalization of scientific knowledge, but also to any attempt to make tacit knowledge fully explicit. In discussing “easily intelligible speech,” where “the tacit is coextensive with the text of which it carries the meaning,” he comments that “the tacit component is the information conveyed,” ([1958] 1962: 87). The “information conveyed,” the meaning, is different from the text, despite being co-extensive with it; similarly, “hearing a message and knowing what it conveys to us” are two different things. 182

DOI: 10.4324/9781003014584-17

Polanyi and Tacit Knowledge

In neither the verbal nor the nonverbal case can the knowledge be adequately articulated. In an important passage in the Introduction to Science, Faith and Society, he makes a point similar to the famous rule-following argument in Wittgenstein’s Philosophical Investigations, when he observes that “There are an infinite number of mathematical formulae that will cover any series of numerical observations. Any additional future observations can still be accounted for by an infinite number of formulae” ([1946] 1964: 9). What is common to both ways of talking about the tacit is that there are “limitations to articulation” ([1958] 1962: 90). The two ways of thinking about the tacit come together in the thought that “To assert that I have knowledge which is ineffable is not to deny that I can speak of it, but only that I can speak of it adequately” ([1958] 1962: 91): that there is always an unaccountable residue beyond the possibility of articulation. Moreover, this residue is “personal” in that the process of integration of elements of thought that lead to discovery or understanding, and then, sometimes, to articulation, is individual and based on the skills and elements of thought of a specific person. in Science, Faith and Society he analogizes this to the personal coefficient of the observational astronomer recording positions of stars from a telescope that is on a moving earth. But this is a problematic analogy: the point of this personal coefficient is that it can be made explicit; the process of coming to understanding contains elements that cannot be. What is the relation between this thesis and traditional ways of talking about the tacit intellectual content of activities such as scientific discovery, for instance, “assumptions,” “presuppositions,” and “premises”? Premises are, normally, semantic objects: things that can be articulated, or made explicit, without residue. The principles of causation indeed would seem to be a paradigm case of the kind of thing that can be articulated, and indeed would not be “principles” or “premises” if they could not be. This is exactly what Polanyi denies: “The premises underlying a major intellectual process are never formulated and transmitted in the form of definite precepts. When children learn to think naturalistically they do not acquire any explicit knowledge of the principles of causation” (Polanyi [1946] 1964: 42). But there is still ambiguity here. If Polanyi is saying that they are never formulated or transmitted in the form of definite precepts, we must ask whether this is simply an observation about how this knowledge is acquired, or a claim that such “premises” cannot be understood – at least fully – as explicit “premises,” principles, or any other semantic object. In his major work, Personal Knowledge, where he works the concept out in greatest detail, he is attentive to what he describes as the “participation of the tacit in the process of articulation” ([1958] 1962: 87), which suggests that articulation is not only possible, but a goal of science. Indeed, the process of scientific discovery itself may be understood as a process in which our tacit knowledge, for example the unarticulated intuitions that lead to a discovery, is made explicit. This is not an inconsistency. The core thought here is that there are some tacit things that could be, and are, articulated, but these are never adequately articulated, and the personal element is never eliminated. But the role and nature of “premises” and similar concepts for Polanyi is still obscure. When he discusses the changes in the premises of science that led to Einstein and to Heisenberg’s revision of Bohr, for example, he attributes them to Mach’s philosophy of science and the idea that “essentially unverifiable” implications of theories should be eliminated from scientific propositions ([1958] 1962: 87). But he denies that “analytical operations” of this kind produced any great discoveries: “What happened was that scientific intuition made use of the positivist critique for reshaping its creative assumptions concerning the nature of things,” and comments that what the discoveries illustrated refuted the positivist conception of science by showing that it depended on a “faculty of speculative discovery” ([1958] 1962: 88). In the same passages, when he speaks of the premises of science, he says that “no exhaustive statement of the 183

Stephen Turner

premises of science can possibly exist” ([1958] 1962: 90). But he then adds that “The common ground of science is, however, accessible to all scientists and is accepted by them as they become apprenticed to the traditional practice of science” ([1958] 1962: 90). Sorting such claims out, and understanding how the “personal” relates to them is the central puzzle in accounting for Polanyi’s view of the tacit.

The Background The term “tacit” as applied in the conceptual domain of reasoning and cognition appears as early as Herbert Spencer (Spencer 1887: Ch. VIII, §302: 87; §305: 99). But the dominance of neo-Kantian philosophy in the later nineteenth century led to the incorporation of the general thought behind “tacitness” into the notion of presupposition, understood in the sense of Kantian categories of thought. The original Kantian concept involved the general categories of thought required to apprehend and organize experience, and were understood to be not psychological but “logical,” or conceptual preconditions of experience that were built into experience but not in the empirical world itself; in the hands of the neo-Kantians these preconditions were understood differently. The key idea was the “faktum der Wissenschaft”: the fact that there was a conceptually organized domain of thought and experience that presupposed certain concepts that validated these conceptual presuppositions. Thus the validity of the presuppositions of physics was warranted by the existence of physics as a science. These presuppositions were tacit in the sense that they were normally not articulated. It was the task of philosophy to identify and reveal them. It became a commonplace to speak of the presuppositions of science, especially in contrast to those of theology, and to regard each domain as in some sense dependent on an act of faith in the presuppositions and therefore equal. Writers like Pierre Duhem applied this reasoning to the history of science itself, and especially to the change from Aristotelian to Newtonian science, which could then be understood as a succession of coherent bodies of thought, each with its own distinctive conceptual presuppositions. An alternative to this way of thinking both preceded and paralleled Kantianism and neoKantianism. Kant was responding to Hume’s argument that our grasp of causality was based on habit or custom: Kant thought this was insufficient, as it could not justify our beliefs. The pragmatists broadened the idea of habit to deal more generally with the tacit, with such notions as habits of mind, and suggested that conscious reasoning arose from failures of habitual responses to the situations faced by people in the course of their interaction with the world. This implied that notions of belief and truth applied to the results of this kind of situation, and that their basis was itself pragmatic – that truth was equivalent to the best beliefs, and the best were those that were best for us to believe, and that the quest for certainty was based on an illusion. A variant of this idea was that truth in the full sense meant the opinion on which investigators ultimately converged. This image of knowledge suggested that knowledge rested on a set of habits which had not yet and might never become problematic through failure. One might identify yet a third tradition, less well-defined, which Polanyi explicitly identifies with: the tradition of phenomenology and existentialism (for him represented by Dilthey), and its product, Gestalt psychology. He does give some credit to the neo-Kantians, notably Windelband, whom he praises for the idea that ideographic and nomothetic knowledge are “logically distinct parts of all knowledge” (1959: 100), meaning that despite being distinct, all knowledge contained both inseparably. This was analogous to Polanyi’s own view that explicit knowledge could nor exhaust knowledge, and that the personal and tacit were an ineliminable part of all knowledge. The neo-Kantians thought the objectively experienced world was constituted by categories and concepts, but that these were not psychological facts, but rather “logical” ones built into the 184

Polanyi and Tacit Knowledge

experience of the world which it was the task of philosophy to elucidate, and make explicit. So although they were, in Polanyian terms, tacit, they were the sorts of things we could come to tell. From the neo-Kantian point of view, “tacit knowledge,” is an oxymoron: if knowledge is, classically, “justified true belief,” it appears that tacitness, by definition, precludes justification, and cannot be understood as “belief ” and therefore not as knowledge. This is the problem the neo-Kantians “solved” by appealing to the notion of presuppositions, which were understood as something that could be made explicit rather than tacit. Polanyi’s relation to these ideas is somewhat ambiguous. On the one hand he uses the notion of interpretive frames, a concept very similar to Kuhn’s “paradigms” and the neo-Kantian notion of an intellectual domain – physics was for them, as for Kant, a favorite example, constituted by concepts. But Polanyi denied that one was trapped into a single interpretive frame. They were for him, like theories, a kind of resource which could be brought to bear on problems. Moreover, the rejected the basic Kantian critique of Hume, when he distinguished “demonstration” and knowledge and said “I deny that truth is demonstrable, [but] I assert that it is knowable” ([1958] 1962: 82). To understand these claims, one must understand his basic model of tacit knowing.

Polanyi’s Model Polanyi’s own model of tacit knowing involved a triangular relation between two elements of knowledge, focal and subsidiary, and the knower, who is an active agent. The key to this relation was the indeterminacy of the links between them. He built the model from the paradigm case of perception, in which he rejected contemporary positivist notions of sense-data, and identified tacit knowing with what he called “from-to knowing.” He explains this in terms of a simpler example: the integration of images that occurs when we view a stereoscopic image and experience it phenomenally as an image of one thing. The grounds of all tacit knowing are items – or particulars – like the stereo pictures of which we are aware in the act of focusing our attention on something else, away from them. This I call the functional relation of subsidiaries to the focal target, and we may also call it a from-to relation. Moreover, I can say that this relation establishes a fromto knowledge of the subsidiaries, as linked to their focus. Tacit knowing is a from-to knowing. Sometimes it will also be called a from-at knowing, but this variation will be only a matter of convenience. (1968: 29) The point of the analogy with the stereoscopic image is that what we see depends on what we focus on. To focus on is to focus away, or “from,” something we are not focused on which our focus is nevertheless dependent on. The thing we focus away from is what he calls subsidiaries; the focus “to” is on particulars. The relation between the “from” or subsidiary part and the focal part is indeterminate, and personal, because the content of the subsidiary part is personal: it is what we personally bring to the process of understanding or discovery. These features carry over to all forms of knowing. What does Polanyi have in mind here? The key idea is that from-to knowing is a skilled performance, even in the act of perception, though in the most elementary forms of perception the element of skill is attenuated and less notable. He distinguishes three “aspects” of tacit knowing in which this relation operates. The functional relation is contained in the act of focusing which establishes the relation between the object of focal attention and the subsidiaries. The phenomenal aspect is the creation of a novel sensory experience – the stereoscopically integrated image – which 185

Stephen Turner

has “a new sensory quality not possessed by the sense perceptions from which it was tacitly created” (1974: 35). What we see depends on the subsidiaries we possess and which are employed in the course of our meaning-making activities. The semantic relation is the meaning of the image, and it is also the joint product of the subsidiaries and the focal object. The “familiar use of a word, which is our subsidiary awareness of it, renders it in a way bodiless” (1974: 35). This relation of dependence, the relation between subsidiary and focal, is neither a logical nor a causal, mechanical one. It is the product of our attention and activity. But Polanyi goes beyond this language to describe the incorporation of the objects of thought into our world in a different way, reflecting the nature of our engagement with the world, in terms of what he calls indwelling. The example he uses to explain this concept is the use of a stick to feel one’s way in the dark. Such exploration is a from-to knowing for we attend subsidiarily to the feeling of holding the probe in our hand, while the focus of our attention is fixed on the far end of the probe, where it touches an obstacle in its path. (1974: 35) This generalizes to all perception. As with the probe, “all sensation is assisted by some (however slight) skillful performance, the motions of which are performed with our attention focused on the intended action so that our awareness of the motion is subsidiary to the performance” (1974: 36). So “the structure of tacit knowing . . . includes a joint pair of constituents. Subsidiaries exist as such by bearing on the focus to which we are attending from them” (1974: 37–38). A skilled act requires an agent, who possesses particular properties. Integration is neither logically determinate nor mechanically causal, because it depends on personal effort using personal skills. The result can be valid or invalid, true or mistaken, and our sense of coherence, the coherence we produce through integration, may be correct or illusory (1974: 39). But once achieved, an integration “can be damaged by contrary facts only if these items are absorbed in an alternative integration which disrupts the one previously established” (1974: 42). Only an “alternative that appears to us to be more meaningful and so true will produce a new perception that will correct our errors” (1974: 42). But this alternative will be produced in the same way: through personal effort assisted by personal skills. Over time Polanyi changed some of his terminology, and some of his emphases, but the core from-to idea remained, along with the idea that the integration achieved was a personal result. The focal-subsidiary distinction was replaced by proximal and distal, based on the idea of indwelling, which will be discussed later. The core of the later thought was still the use of a stick as a probe: the cavity at the end of the stick was “distal” and object-like, while the stick and the body itself were proximal and instrument-like, and dropped out of attention. But he extended this reasoning to “meaning,” as when he says that “all meaning tends to be displaced away from ourselves, and that is in fact my justification for using the terms ‘proximal’ and ‘distal’ to describe the first and second terms of tacit knowing” (1966: 13). The way he elucidated this process, and also the way he understood his relation to Kant, was through a psychological account. Kant had had said that “The schematicism by which our understanding deals with the phenomenal world . . . is a skill so deeply hidden in the human soul that we shall hardly guess the secret trick that Nature here employs” (Kant [1781/1787] 1998: A.141; quoted in Polanyi 1962: 1, 1969: 105). Polanyi provided a solution to the mystery: the psychological operations that produce not only perception but advanced scientific thinking are hidden and dependent on peripheral impressions that are integrated according to Gestalt like operations to form our perceptions, rather than anything resembling rules or logical 186

Polanyi and Tacit Knowledge

derivations. It is here, he thought that “We are at last facing here fully that secret power of our nature that Kant despaired of knowing.” The power is embodied: “The main clues on which perception relies are deeply hidden inside the body and cannot be experienced in themselves by the perceiver” ([1958] 1962: 9, 1969: 115). “[T]hey are seen only incidentally and their effect on the appearance of the object perceived can be revealed only by elaborate experimental investigations” (1969: 115). Yet all these hidden evaluations of not identifiable clues are conducted, . . . by our full powers of our intelligence, relying on operations which . . . must be acquired by efforts which in certain cases may prove so strenuous that their persistent pursuit leads to mental breakdown. ([1958] 1962: 10, 1969: 115) The power is that of integrating clues in our thought, what he calls “peripheral impressions” ([1958] 1962: 9, 1969: 114) in the course of “skillful integration.” This is a vivid picture of the hidden character of skillful integration in science, and of the difficulties in acquiring the skills, their hidden character, and of the hidden character of the processes of integration themselves. And it is, in its purely psychological character, a picture radically opposed to the neo-Kantian one. But it has implications for any version of the Kantian approach to the tacit, and to the idea of presuppositions in general. But he makes an important concession to the idea that we can make these processes explicit: although “integration cannot be replaced by any explicit mechanical procedure . . . one can paraphrase the cognitive content of an integration,” though “the sensory quality which conveys this content cannot be made explicit” (1968: 32).

The Relation to Empathy Polanyi also said a few other things that located him in relation to previous philosophy, though he was careful to explain how he had radicalized these earlier thoughts. He said little about empiricism, which reasoned something like this: we acquire knowledge empirically, in accordance with habits that we also acquire empirically. His emphasis on agency separated him from this tradition, which he thought was too passive in its account of the acquisition of knowledge as a matter of inputs. He appealed to Brentano ([1874] 1942) [who] “taught that consciousness necessarily attends to an object, and that only a conscious mental act can attend to an object.” Polanyi added that My analysis of tacit knowing has amplified this view of consciousness. It tells us not only that consciousness is intentional, but also that it always has roots from which it attends to its object. It includes a tacit awareness of its subsidiaries. (1968: 32) Brentano was a thinker outside of, and opposed to, the neo-Kantian tradition, as was Dilthey, in a somewhat more ambiguous way. But one of Polanyi’s clearest statements of his views dealt with the issue of “indwelling” by reference to the idea of empathy, Einfühlung (“in-feeling”), developed by Theodor Lipps. Polanyi accepts the lessons taught by Lipps and Dilthey. He says that “German thinkers postulated that indwelling, or empathy, is the proper way of knowing man and the humanities” (1966: 16). But Polanyi radicalizes these lessons, by saying that empathy is just a special case of “indwelling,” a term that he regards as not only more comprehensive in that it applies to all knowing, and more precise as a concept. 187

Stephen Turner

He applies this term to the tacit inferences that allow us to make discoveries and form an awareness of objects, the core of his basic account of the role of the tacit in knowing, but also to the process by which we assimilate theories. This is a crucial step: understanding, interpretation, and discovery now appear to be products of the same process of knowing, with the same basic structure. Polanyi describes “another indication of the wide functions of indwelling when we find acceptance of moral teachings described as their interiorization” (1966: 17). He explains that interiorization makes the teachings “function as the proximal term of a tacit moral knowledge, as applied to practice” (1966: 17). In other words, they are the things we focus on, but can only focus on by virtue of knowledge we are not focused on. When he says that “mathematical theory can be learned only by practicing its application: its true knowledge lies in our ability to use it” (1966: 17), he is making a related point. In both cases, using our knowledge, as well as acquiring it, depends on much that operates out of the range of direct attention.

The Reception The reception and use of Polanyi’s ideas about tacit knowledge have been divided. Although philosophers of science have typically acknowledged the role of tacit knowledge, they have regarded it as a psychological fact, and taken Polanyi’s insistence on the tacit element and its personal character to be an embrace of subjectivism. This would appear to be the implication of such statements as these: We have yet to recognize an important element of all personal judgements affecting scientific statements. Viewed from outside as we described him the scientist may appear as a mere truth-finding machine steered by intuitive sensitivity. But this view takes no account of the curious fact that he is himself the ultimate judge of what he accepts as true. His brain labors to satisfy its own demands according to criteria applied by its own judgement. It is like a game of patience in which the player has discretion to apply the rules to each run as he thinks fit. Or, to vary the simile, the scientist appears acting here as detective, policeman, judge, and jury all rolled into one. He apprehends certain clues as suspect; formulates the charge and examines the evidence both for and against it, admitting or rejecting such parts of it as he thinks fit, and finally pronounces judgement. While all the time, far from being neutral at heart, he is, himself passionately interested in the outcome of the procedure. (Polanyi [1946] 1964: 38) An important dissertation by Alan Musgrave (1968), a follower of Karl Popper, outlines the charge of subjectivism in detail. The large set of issues raised by this criticism, and Polanyi’s general approach to objectivity, lay largely outside of his approach to tacitness, but his solution to the problem of objective knowledge, and his response to the charge of subjectivism, reflects his account of skillful performance. Relating to a comprehensive entity, i.e., something real, is a skilled performance. As commentators have noted, this is personal, but far from subjective: it is about engagement with the world, which produces a common, but not a “constructed” response. As Polanyi puts it, the characteristic features of the situation are seen more clearly when we consider the way one man comes to understand the skillful performance of another man. He must try to combine mentally the movements which the performer combines practically and he must combine them in a pattern similar to the performer’s pattern of 188

Polanyi and Tacit Knowledge

movements. . . . He dwells in these moves by interiorizing them. By such exploratory indwelling the pupil gets the feel of the master’s skill and may learn to rival him. (1966: 29–30) This placed Polanyi outside of the common view that arose in science studies, which nevertheless took over the term tacit knowledge, and outside the debates that dominated philosophy of science, each of which took over the idea of science creating its objects by shared conceptual schemes. Outside of this domain, however, the idea of tacit knowledge, in its negative form of the denial of the reducibility of thought to rules, found considerable support. The significance and attractiveness of the idea of tacit knowledge is largely unconnected to the model Polanyi himself constructs. The key attraction, the claim of the existence and indispensability of tacit knowledge, is taken as evidence of the limits of attempts to reduce science, or any complex intellectual activity, or even simple manual labor, to a rule-driven computer program. It is in this form that the concept has been applied to multiple topics.1

Related Topics Chapters 3, 4, 14, 15, 31

Notes 1 There is a large secondary literature on Polanyi, only some of which deals with the concept of tacit knowledge. Notable explications of Polanyi’s views include Jha (1997, 2002), Blum (2010), Jacobs (2001), Scott and Moleski (2005), Zmyślony (2010), Drew Leder (1990), Jerry H. Gill (2000), and Harry Prosch (1986). Special mention should be made of the relationship between Polanyi and Marjorie Grene, who contributed greatly to his philosophical efforts, and provided an important explication of his thought (1977), in addition to editing a collection of his essays, Knowing and Being (1969). A larger collection is Walter Gulick, Recovering Truths: A Comprehensive Anthology of Michael Polanyi’s Writings (www.polanyisociety.org/). For a critical approach which attempts to reconcile tacit knowledge with recent analytic epistemology, see Gascoigne and Thornton (2013). For a response to the Popperian critique, see Andy Sanders (1988). For a more general account of his thought and science, see Mary Jo Nye (2011). The journal Tradition and Discovery, devoted to Polanyi’s thought, and available on the web, includes many articles on tacit knowing.

References Blum, P. R. 2010. “Michael Polanyi: the anthropology of intellectual history”. Studies Eastern European Thought, 62: 197–216. Brentano, F. [1874] 1942. Psychologie vom empirischem Standpunkt. Leipzig: Oskar Kraus. Gascoigne, N., and Thornton, T. 2013. Tacit knowledge. Durham, UK and Bristol, CT: Acumen. Gill, J. H. 2000. The tacit mode: Michael Polanyi’s postmodern philosophy. Albany: SUNY Press. Grene, M. 1977. “Tacit knowing: grounds for a revolution in philosophy”. Journal of the British Society for Phenomenology, 8: 164–171. Jacobs, S. 2001. “Michael Polanyi, tacit cognitive relativist”. The Heythrop Journal, 42: 463–479. Jha, S. R. 1997. “A new interpretation of Michael Polanyi’s theory of tacit knowing: integrative philosophy with ‘intellectual passions”. Studies in the History of Philosophy of Science, 28: 611–631. Jha, S. R. 2002. Reconsidering Michael Polanyi’s philosophy. Pittsburgh: University of Pittsburgh Press. Kant, I. [1781/1787] 1998. Critique of Pure Reason. Trans. and ed. P. Guyer and A. W. Wood. Cambridge: Cambridge University Press. Leder, D. 1990. The absent body. Chicago: University of Chicago Press. Musgrave, A. 1968. “Personal knowledge: a criticism of subjectivism in epistemology”. Dissertation, University of London, London School of Economics and Political Science, December.

189

Stephen Turner Nye, M. J. 2011. Michael Polanyi and his generation: origins of the social construction of science. Chicago: The University of Chicago Press. Polanyi, M. [1946] 1964. Science, faith, and society. Chicago: The University of Chicago Press. Polanyi, M. [1958] 1962. Personal knowledge: towards a post-critical philosophy. Chicago: The University of Chicago Press. Polanyi, M. 1959. The study of man. London: Routledge. Polanyi, M. 1962. “The unaccountable element in science”. Philosophy, 37: 1–14. Polanyi, M. 1966. The tacit dimension. Chicago: The University of Chicago Press. Polanyi, M. 1968. “Logic and psychology”. The American Psychologist, XXII: 27–43. Polanyi, M. 1969. Knowing and being. Ed. M. Grene. London: Routledge. Polanyi, M. 1974. Meaning (with H. Prosch). Chicago: The University of Chicago Press. Prosch, H. 1986. Michael Polanyi: a critical exposition. Albany: SUNY Press. Sanders, A. 1988. Michael Polanyi’s post-critical epistemology: a reconstruction of some aspects of ‘tacit knowing’. Amsterdam: Rodopi. Scott, W. T., and Moleski, M. X. 2005. Michael Polanyi, scientist and philosopher. Oxford: Oxford University Press. Spencer, H. 1887. The principles of psychology. Vol. ii. New York: D. Appleton and Company. Zmyślony, I. 2010. “Various ideas of tacit knowledge – is there a basic one?”. In T. Margitay, ed., Knowing and being: perspectives on the philosophy of Michael Polanyi. Cambridge: Cambridge University Press.

190

14 TACIT KNOWLEDGE Tim Thornton

1. Introduction The main promoter of the idea of tacit knowledge (or ‘tacit knowing’, as he preferred) in twentieth century philosophy was the chemist turned philosopher of science Michael Polanyi. His book The Tacit Dimension starts with the manifesto: I shall reconsider human knowledge by starting from the fact that we can know more than we can tell. (Polanyi 1967b: 4) But, as he is quick to acknowledge, it is not easy to say exactly what this means. It does, however, suggest a via negativa. What is tacit is what is not ‘tellable’ or explicit under a suitable understanding of that. The manifesto accords with some of the phenomena that are sometimes called ‘tacit knowledge’ in the popular adoption of the label: recognising someone’s face, or a few hastily drawn lines as a face; throwing and catching a ball; riding a bicycle; touch typing; skilfully playing the piano; reading a book or map; immediately diagnosing a patient; navigating the shoals of interpersonal relationships; understanding a language. These phenomena seem to involve normative, intentionally directed activities that express some form of knowledge but also something that cannot be fully, at least, put into words. Two less-everyday examples are worth setting out because they are often used as paradigmatic examples in more academic discussions of tacit knowledge. The first example is that of skilled Polynesian navigators who were found to be able to navigate small out-rigger canoes ‘across two or three hundred miles of open sea; and do so in almost any weather, and even when less than fully sober. How is it done?’ (Gellatly 1986: 5). Investigation suggested that the skill took years to master and was context specific: they were only able to navigate the seas and in the wind conditions of their familiar part of the world and that they were unable to put this skill into words on dry land. The second relates to the need in poultry farming to determine the gender of chicks as soon as possible after they hatch. In the 1920s, Japanese scientists came upon a Chinese method by which this could be done based on subtle perceptual cues with a suitably held one day old chick (Martin DOI: 10.4324/9781003014584-18 191

Tim Thornton

1994: 3). It was, nevertheless, a method that required a great deal of skill developed through practice. After four to six weeks of instruction and practice, a newly qualified chick-sexer might be able to determine the sex of 200 chicks in 25 minutes with an accuracy of 95% rising with years of practice to 1,000–1,400 chicks per hour with an accuracy of 98% (Gellatly 1986: 4). Although the theory of chick sexing by genital differences was simple, the nature of the recognitional skill involved was elusive. According to R.D. Martin’s history, having read about the theory, Mr Kojima, a poultry farmer from Nagoya, reversed engineered the skill using first 60 day, then 30 day, then younger and younger chicks (Martin 1994: 4). Dreyfus and Dreyfus’ 1986 account of this case seems to imply that not even the chick-sexers could explain how they did it (Dreyfus and Dreyfus 1986: 197–198). Their own knowledgeable ability was opaque even to them and hence doubly ‘untellable’. But this interpretation risks distorting the fact that it was necessary to make an immediate perceptual judgement as chick-manipulation changed the key visual cue. Chick-sexers also often learnt their judgements aided by diagrams later ‘outgrown’ once the ability was second nature (Martin 1994). The very idea of tacit knowledge presents a challenge, however. Any account of it must show that it is both tacit and knowledge. But it is not easy to meet both conditions. Emphasising the tacit status, threatens the idea that there is anything known. Articulating a knowable content, that which is known by the possessor of tacit knowledge, risks making it explicit. What limits can be placed on what can be said while still leaving something that can be known? I will suggest that an account can be developed that meets this dual condition influenced by Polanyi, Ryle and Wittgenstein. One consequence is that not all potential instances of tacit knowledge in philosophy – such as of the axioms of a theory of meaning for a language (see more on this later) – will fit.

2.  Polanyi, Tacit Knowledge and the Distinction Between Focal Versus Subsidiary Awareness Given, however, that the notion of tacit knowledge is only partly embedded in ordinary language and is specifically associated with Polanyi, why not simply adopt his own account? There are two reasons. First, Polanyi is promiscuous with his use of the idea and so it is difficult to derive a clear account. As one Polanyian scholar concedes: Now we are in a position better to understand why Polanyi’s notion of the tacit is unclear. Since he uses the term in different contexts to embrace all in human cognition that is not explicit, it is a comprehensive term that includes processes and contents from many ontological levels and different disciplines. Tacit processes and components include: (1) all the autonomic physiological processes that support at lower levels an act of knowing at any given moment; (2) plus the embodied skills, learned and perhaps innate, that contribute to one’s intentional actions (including cognitive actions); (3) plus all the psychological dynamics (integrations, judgments) involved in this knowing; (4) plus all the background presuppositions and beliefs that shape the prevailing personal framework of interpretation; (5) plus the immediately preceding thoughts and actions still active in working memory; 192

Tacit Knowledge

(6) plus the network of sensitivities, moods, and feelings salient at the moment of analysis; (7) plus the influx of sensations, all vying for attention, relayed by the central and peripheral nervous systems, including sensory data from the body’s receptors; (8) plus the interests, goals, and expectations pushing thought and action toward certain satisfactions; (9) plus the words and grammar evoked as the transparent means to express and evaluate the meaning of this concatenation of influences. (Guilick 2016: 303) The second reason is that Polanyi’s own account emphasises a phenomenological distinction between subsidiary and focal awareness, which applies either to too narrow a range of cases or is merely implausibly generalised. To distinguish between focal and subsidiary awareness, he uses the example of pointing to something with a finger. There is a fundamental difference between the way we attend to the pointing finger and its object. We attend to the finger by following its direction in order to look at the object. The object is then at the focus of our attention, whereas the finger is not seen focally, but as a pointer to the object. This directive, or vectorial way of attending to the pointing finger, I shall call our subsidiary awareness of the finger. (Polanyi 1967a: 301) In attending from the finger to the object, the object is the focus of attention whilst the finger, though seen, is not attended to. Applying this notion more broadly, he gives the following example. I may ride a bicycle and say nothing, or pick out my macintosh among twenty others and say nothing. Though I cannot say clearly how I ride a bicycle nor how I recognise my macintosh (for I don’t know it clearly), yet this will not prevent me from saying that I know how to ride a bicycle and how to recognise my macintosh. For I know that I know how to do such things, though I know the particulars of what I know only in an instrumental manner and am focally quite ignorant of them. (Polanyi 1962: 91) This passage suggests that the skills involved in recognising a macintosh and in bicycle riding are similar (an important clue, to which I  will return). In both cases, the ‘knowledge-how’ depends on something which is not explicit and hence tacit: the details of the act of bike riding or raincoat recognition. But while one can recognise one’s own macintosh, in this example, one is ignorant, in some sense, of how. This connects to the idea that focal awareness that it is one’s own depends on merely subsidiary awareness of macintosh details. If this argument were successful it would be of general significance because it would also apply to all linguistic labelling and Polanyi explicitly makes this connection. [I]n all applications of a formalism to experience there is an indeterminacy involved, which must be resolved by the observer on the ground of unspecified criteria. Now we may say further that the process of applying language to things is also necessarily unformalized: that it is inarticulate. Denotation, then, is an art, and whatever we say about things assumes our endorsement of our own skill in practising this art. (Polanyi 1962: 84) 193

Tim Thornton

The wider argument for denotation being tacit seems to rest on an appeal to the clearer case of the recognition of particulars – such as a particular macintosh – which, Polanyi argues, depends on features of which one is focally ignorant. (The example concerns recognition of a particular macintosh as one’s own rather than merely that an object is a macintosh but nothing turns on this difference.) However, to justify the wider argument, he needs to defend the general claim that explicit recognition of something as an instance of a type is always based on the implicit recognition of subsidiary properties of which one is focally ignorant. To recognise a feature (F, say) one must a) always recognise it in virtue of something else (subsidiary features G, H and I, e.g.) of which b) one is focally ignorant. But it is not clear that either part of this claim is true. Polanyi seems to assume that the question of how one recognises something to be of an instance of a particular kind – for example, a patch of colour on a summer’s afternoon – always has an informative answer. To cover cases where it is not obvious what this is, he claims – implausibly – that one can have subsidiary awareness of one’s own ‘neural traces in the brain’ (Polanyi 1969: 147). But although the question may sometimes have an informative answer, there is no reason to think that it always does at the level of the person. An excursion into optics and neurology may explain how some perceptual discrimination is possible without trading in (deploying rather than explaining) person level knowledge. Even in cases where one recognises a particular as an F in virtue of its subsidiary properties G, H, I, and cannot give an independent account of those properties, it is not clear that one need be focally ignorant of them. It may be, instead, that the awareness one has of G, H, I is manifested in the recognition of something as an F or a particular F. One might say, I recognise that this is a macintosh (or my macintosh) because of how it looks here with the interplay of sleeve, shoulder and colour even if one could not recognise a separated sleeve, shoulder or paint colour sample as of the same type. Although it seems plausible that one might not be able to say in context-independent terms just what it is about the sleeve that distinguishes a macintosh from any other kind of raincoat (lacking the vocabulary of tailoring) that need not imply that one is focally ignorant of, or not attending to, just those features that make a difference. Recognition may depend on context-dependent or demonstrative elements, such as recognising shapes or colours for which one has no prior name. That suggests one has to be focally aware of them. (This notion of context-dependent or demonstrative elements will play a role shortly.) For these reasons and contrary to Polanyian scholarship, it is better to sever an account of tacit knowledge from Polanyi’s preferred distinction between focal versus subsidiary awareness (contra Grene 1977).

3.  A Polanyi-, Ryle- and Wittgenstein-Inspired Account of Tacit Knowledge I suggested that the very idea of tacit knowledge presents a challenge. Tacit knowledge is both tacit and knowledge and it is hard to meet both conditions. There is a second strand in Polanyi’s work, which helps address this, though abstracting away from his own account. At the start of his book Personal Knowledge he says: I regard knowing as an active comprehension of things known, an action that requires skill. (Polanyi 1962: vii)

194

Tacit Knowledge

In combination with the first slogan – ‘I shall reconsider human knowledge by starting from the fact that we can know more than we can tell’ – and the idea of context-dependent demonstratives (preceding), this suggests a clue to tacit knowledge. Taking tacit knowledge to be practical suggests one way in which it is ‘untellable’. It cannot be articulated except by deploying some context-dependent practical demonstrations and, further, only to those with ‘eyes to see’. It cannot be made explicit in words alone. Yet, at the same time, it has content: practical knowledge of how to do something. In Tacit Knowledge, Neil Gascoigne and I set out an account of tacit knowledge as conceptually structured, context dependent, practical knowledge (Gascoigne and Thornton 2013). We locate tacit knowledge at the level of the person, rather than the sub-personal, and within the ‘space of reasons’, the space of justifying and being able to justify what one says. (McDowell 1994: xiv; Sellars 1997: 76). We motivate the account partly in response to ‘regress arguments’ offered by Gilbert Ryle and Ludwig Wittgenstein and defend the former against new intellectualist criticism (e.g. Stanley and Williamson 2001). But the support the regress arguments offer to a conception of tacit knowledge is nuanced. Ryle’s regress argument is summarised pithily thus: If a deed, to be intelligent, has to be guided by the consideration of a regulative proposition, the gap between that consideration and the practical application of the regulation has to be bridged by some go-between process which cannot by the presupposed definition itself be an exercise of intelligence and cannot, by definition, be the resultant deed. This go-between application process has somehow to marry observance of a contemplated maxim with the enforcement of behaviour. So it has to unite in itself the allegedly incompatible properties of being kith to theory and kin to practice, else it could not be the applying of the one in the other. For, unlike theory, it must be able to influence action, and, unlike impulses, it must be amenable to regulative propositions. Consistency requires, therefore, that this schizophrenic broker must again be subdivided into one bit which contemplates but does not execute, one which executes but does not contemplate and a third which reconciles these irreconcilables. And so on for ever. (Ryle 1945: 2–3) There has been a recent flurry of literature on the precise nature of this argument and thus whether it is successful (e.g. Stanley and Williamson 2001; Noë 2005; for detailed assessment see Gascoigne and Thornton 2013: 51–79). But it involves something like the following regress. Suppose all know-how can be articulated in words as a piece of knowledge-that: grasping some proposition that p. Grasping the proposition that p is itself something one can do successfully or unsuccessfully, so it is also a piece of know-how. So, on the theory in question, it will involve grasping another proposition, call this q. But grasping the proposition that q is itself something one can do successfully or unsuccessfully, so it is also a piece of know-how. So, on the theory in question, it will involve grasping another proposition, call this r. . . . etc. If the first step of the reductio is designed to ‘articulate’ or represent a piece of otherwise merely tacit knowledge at the heart of recognition, it will lead to a regress. Ryle himself suggests that it can be used to undermine what he calls an ‘intellectualist legend’ which attempts to explain practical knowing-how through a deeper, theoretical form of knowing-that based on grasping a proposition. His counter argument is that, since grasping a proposition can be done well or badly, the only way to avoid the vicious regress is to grant that intelligence can accrue

195

Tim Thornton

to, and be manifested in, practical knowledge directly, without further theoretical explanation or underpinning. Knowing-how is more basic than knowing-that. What is the relationship between Ryle’s argument and the clues extracted (preceding) from Polanyi to the nature of tacit knowledge? On the one hand, it offers some nuanced support. Polanyi’s claim that we know more than we can tell is one clue. Given that Ryle argues that knowing-how cannot be explained through knowing-that or grasp of a proposition – both paradigmatic of what can be put into words – and if, following Polanyi’s second clue, tacit knowledge is equated with practical knowledge, then Ryle’s argument suggests limits to the way of putting it into words. It cannot be explained, at least, in knowing-that terms. But, on the other hand, the idea that practical knowledge can express intelligence directly – without needing to inherit it from grasp of a proposition – suggests the following thought which runs counter to the claim that we know more than we can tell. (And thus it puts Polanyi’s two clues in tension.) Consider the following piece of practical knowledge: the ability to recognise a raincoat as a macintosh and thus denote it ‘macintosh’. Why is the denoting of raincoats as ‘macintoshes’ not what articulating or expressing this piece of recognitional knowledge amounts to, thus discharging any tacit element? In this particular example, because it involves linguistic denotation, why is that not putting all the relevant knowledge in play into words (calling this coat a ‘macintosh’, e.g.)? Why assume, as Polanyi does, that there is always a further, though tacit, answer as to how one recognises that something is an instance of a general kind? Another regress argument, this time from Wittgenstein, suggests an answer. Wittgenstein considers teaching a pupil to continue a mathematical series correctly who ‘judged by the usual criteria . . . has mastered the series of natural numbers’ and is then taught to continue series of the form 0, n, 2n, 3n, etc. for arbitrary values of n. Then we get the pupil to continue one series (say “+ 2”) beyond 1000 – and he writes 1000, 1004, 1008, 1012. We say to him, “Look what you’re doing!” – He doesn’t understand. We say, “You should have added two: look how you began the series!” – He answers, “Yes, isn’t it right? I thought that was how I had to do it.” – – Or suppose he pointed to the series and said, “But I did go on in the same way”. – It would now be no use to say, “But can’t you see . . . ?” – and go over the old explanations and examples for him again. In such a case, we might perhaps say: this person finds it natural, once given our explanations, to understand our order as we would understand the order “Add 2 up to 1000, 4 up to 2000, 6 up to 3000, and so on”. This case would have similarities to that in which it comes naturally to a person to react to the gesture of pointing with the hand by looking in the direction from fingertip to wrist, rather than from wrist to fingertip. (Wittgenstein 2009: §185) The pupil reacts in a divergent way (from us) to explanations of how to continue. He appears to have acted under a divergent interpretation of the expression of the rule. But if, to be successful, an explanation of a rule has to be correctly interpreted, then any expression of the interpretation might be subject to divergent understandings, whether explicitly stated in a symbolism or by examples. “But how can a rule teach me what I have to do at this point? After all, whatever I do can, on some interpretation, be made compatible with the rule.” – No, that’s not what one should say. Rather, this: every interpretation hangs in the air together with what 196

Tacit Knowledge

it interprets, and cannot give it any support. Interpretations by themselves do not determine meaning. (Wittgenstein 2009: §198) Since any interpretation has to be specified in some way, if understanding depends on an interpretation, no specification would be sufficient in itself. Hence the label ‘regress of interpretations’. In an early work – Changing Order – on the role of tacit knowledge in science, the sociologist Harry Collins assumes that the regress argument shows that the correct way to continue, the correct notion of sameness, cannot be specified or said (Collins 1985). [S]ince in spite of this we all know the correct way to go on, there must be something more to a rule than its specifiability. (Collins 1985: 14) Thus there is a genuine gap between what is said in an explanation of a rule and what is understood when all goes well and something is needed to plug the gap so as to explain the otherwise mysterious ability to go on correctly. The extra element is social entrenchment such that the sense of sameness in continuing in the same way is relative to a community. It underpins the mysterious abilities that enable us to know when to continue ‘2,4,6,8’ with ‘10,12,14,16’ and when with ‘who do we appreciate?’ (Collins 1985: 22, italics added) Collins calls this ‘tacit knowledge’. [T]he member of a social group who has the ability to continue the sequence ‘2,4,6,8’ with ‘10,12,14,16’ as a matter of course, without even thinking about it, also possesses something that the stranger to our culture and the newborn do not. This is sometimes referred to as ‘social skill’ but we can call it tacit knowledge without doing too much violence to the term. (Collins 1985: 56, italics added) According to Collins, tacit knowledge cannot be put into words, is invisible in its transmission and capricious (Collins 1985: 129). This suggests a picture of rule following in which explanations are always inadequate but are patched by invisible and capricious tacit knowledge: the picture that Wittgenstein puts into the mouth of his interlocutor in order to reject it. “But do you really explain to the other person what you yourself understand? Don’t you leave it to him to guess the essential thing? You give him examples – but he has to guess their drift, to guess your intention.” – Every explanation which I can give myself I give to him too. “He guesses what I mean” would amount to: “various interpretations of my explanation come to his mind, and he picks one of them”. So in this case he could ask; and I could and would answer him. (Wittgenstein 2009: §210) There is an honourable tradition of Wittgenstein scholarship that takes the question of how something can be grasped, from an explanation and that determines subsequent correct 197

Tim Thornton

applications of a rule, seriously and attempts to construct substantial philosophico-theoretical answers. Crispin Wright, for example, uses the Wittgenstein-inspired phrase ‘rails to infinity’ to label a view of rules he rejects and then attempts to construct a notion of rule following as depending on the ongoing piecemeal judgements subjects make as to what accords with the rule (Wright 2001). The challenge for any such account is to avoid the objection that by making the context of a rule plastic to ongoing judgement that it undermines the very idea of being governed by rules (Thornton 1997). There is, however, a more promising way to respond to Wittgenstein’s regress argument which also connects it to tacit knowledge. The moral of the regress of interpretations according to John McDowell is that: We learn that it is disastrous to suppose there is always a conceptual gap between an expression of a rule and performances that are up for assessment according to whether or not they conform to the rule, a gap that is made vivid by saying the expression of the rule stands there. . . . We must not acquiesce in the idea that an expression of a rule, considered in itself, does not sort behaviour into performances that follow the rule and performances that do not. (McDowell 2009: 100–101) This, according to McDowell, is the moral of §201 of the Philosophical Investigations: ‘there is a way of grasping a rule which is not an interpretation, but which, from case to case of application, is exhibited in what we call “following the rule” and “going against it” ’ (Wittgenstein 2009: §201). For those who are party to the relevant practices, there is no gap between the expression of the rule and being told which way to go. A signpost, for example, not merely a signpost under an interpretation, points the way. On this account, the regress of interpretations is stopped before it can start. Hence, the expression of a rule is explicit in its directions. But it is only explicit for those with the right eyes to see, or ears to hear, an explanation: for those, in Stanley Cavell’s phrase, who share the ‘whirl or organism’ of our form of life (Cavell 1969: 52). This in turn grounds out in the applications of rules, which are context-dependent, conceptually structured and practical. The regress of interpretations argument shows that any informative explanation of what following a rule correctly, going on in the same way, comprises soon gives out. Any general account of the relation of rules or conceptual generalities to particular judgements falls prey to the regress of interpretations. What can be made explicit via the art of denotation – in Polanyi’s phrase – rests on the tacit, not because of a distinction between focal and subsidiary awareness, but because going on in each specific way simply is accord with a rule or concept. It is also context dependent and, since it links conceptual generalities to particular cases in judgements, it is practical. Because the relation of the tacit and explicit is subtle, it is worth giving a worked example. It is possible to make a béchamel sauce by following a recipe: to measure ingredients and to cook at a specific temperature for a predetermined time. But it is also possible to make one ‘bi t’rack o’ t’ee’ (by the rack of the eye) as it is said in Yorkshire dialect. One judges a rough equivalence of butter and flour, sautés the resultant roux until it looks right (so as to cook the flour) and then adds milk gradually and intermittently while stirring continually, watching the consistency fall to almost that of the added milk and then stiffen again. Gradually, the rate of returning to stiffness slows. Were one to add too much milk, it would never thicken again. Thus, the end point is fixed by a judgement that the rate of stiffening has slowed sufficiently that a final addition of a particular amount of milk will, combined with the residual heat energy of the sauce, yield the right consistency by the time it is served. Cooking thus is an expression of a conceptual 198

Tacit Knowledge

understanding of the effects of dilution, heating, stirring, thickening against the jeopardy of adding too much milk. But the expression of the concepts comprises demonstratives of the form ‘that degree of stiffness’ and ‘that quantity of additional milk’. It involves the right kind of manipulation of the sauce mixture with a spoon and the right degree of heat applied so that it bubbles thus. Although it is possible to make a béchamel sauce following an explicit recipe, the cook who makes it ‘by eye’ does so employing a form of conceptually structured, context dependent, practical knowledge, or tacit knowledge, on my account. Further, its expression and articulation in a practical demonstration depends for its success on the ability of the audience to see in the particularity of a few cases the generality of the rules involved. They must have ‘eyes to see’. They must share the whirl of organism. The dual challenge for an account of tacit knowledge can be met in this way as follows. It is tacit because it cannot be codified in context-independent and general terms – it is in that sense ‘untellable’ – while at the same time it is knowledge because it is articulated in the application of general concepts to the particular case, grounded in practice, which is a form of know-how. One consequence of this account is that explicit knowledge always rests on a body of tacit knowledge. Against that background – the art of denotation – some things can be made explicit. I advertised earlier that this account would not meet all the circumstances in which philosophers call on tacit knowledge. One such concerns compositional semantics. Gareth Evans argued that while the ‘theorems’ of a theory of meaning that state the meaning (typically via truth conditions) of sentences of an object language can be objects of knowledge, their axioms cannot be because they do not have general application except to yield meaning theorems (Evans 1981). They thus fail to express conceptual understanding because they fail the Generality Constraint (Evans 1982). Despite that, he argued that it is helpful to characterise the linguistic dispositions charted by compositional semantic theories as expressions of ‘tacit knowledge’ and his account has been further developed, against criticisms put forward by Wright, by Martin Davies (Davies 1987; Wright 1986; see also Miller 1997). Whatever the merits of calling a subject’s relation to the compositional semantic theories of the languages they possess, this use does not fit the Polanyi-inspired account developed earlier because the former is couched using concepts the subject need not possess and is postulated at a merely sub-personal, sub-doxastic level. It is thus a much more radical extension from the root notion of knowledge. There is, of course, some freedom to use words in whatever way we wish but not complete freedom. As Lewis Carroll put it: ‘But “glory” doesn’t mean “a nice knock-down argument,” ’ Alice objected. ‘When I use a word,’ Humpty Dumpty said, in rather a scornful tone, ‘it means just what I choose it to mean – neither more nor less.’ ‘The question is,’ said Alice, ‘whether you can make words mean so many different things.’ ‘The question is,’ said Humpty Dumpty, ‘which is to be master – that’s all.’ (Carroll 1875: 124)

4.  An Ontology Rather Than Epistemology of Tacit Knowledge? Taking tacit knowledge to be person-level and conceptual (as well as practical and context specific) is not the only possible approach, however. In his Tacit and Explicit Knowledge, Collins also contrasts tacit knowledge with what is explicit (Collins 2010). But the characterisation of the explicit is not what can be expressed linguistically – as Polanyi and I assume – but, rather, with what Collins calls ‘strings’ (Collins 2010: 57). Strings are ‘bits of stuff inscribed 199

Tim Thornton

with patterns: they might be bits of air with patterns of sound waves, or bits of paper with writing, or bits of the seashore with marks made by waves, or patterns of mould, or almost anything’ (Collins 2010: 9). The motivation for this is to avoid the ‘freight of inherent meaning that makes the notions of signs, symbols and icons so complicated’ (Collins 2010: 9). One repeated worry is that strings (and hence signs, as kinds of strings) do not have an essential meaning: ‘strings are without meaning . . . a book is a physical thing, not a meaningful thing’ (Collins 2010: 34). So rather than presupposing meaning or mental content, Collins stresses the physical nature of strings: ‘a string is just a physical object and it is immediately clear that whether it has any effect and what kind of effect this might be is entirely a matter of what happens to it’ (Collins 2010: 9). But since strings are just ‘bits of stuff inscribed with patterns’ their elaboration, transformation, mechanisation and explanation are all counted by Collins as instances of what is explicit. Hence the focus of the book turns out not to be the nature of the knowledge that rational subjects possess but the worldly objects of knowledge: the processes or tasks or worldly patterns themselves. One example of this is the way that Collins, as I have, attempts to demystify tacit knowledge by connecting it to what is practical: We are just like complicated cats, dogs, trees, and sieves . . . Sometimes we can do things better than cats, dogs, trees and sieves can do them, and sometimes worse. A sieve is generally better at sorting stones than a human (as a fridge is better at chilling water), a tree is certainly better at growing leaves, dogs are better at being affected by strings of smells, and cats are better at hunting small animals . . . if we were to stop talking and just get on with things – that is, if the tacit was not made mysterious by its tension with the explicit – there would be no puzzle at all about the body, per se. That teaching humans to accomplish even mimeomorphic actions is a complicated business, involving personal contact, says nothing about the nature of the knowledge, per se. (Collins 2010: 104–105) The final claim reiterates the idea that what matters to Collins – ‘knowledge, per se’ – does not depend on how humans think of or conceptualise tasks, something in the realm of sense, but rather on the mechanical process, in the realm of reference. In my preceding example, Collins would not distinguish between the béchamel sauce maker who follows a recipe and the one who uses uncodified judgement if, by chance, they undertake the same process. The process, not the knowledge, that is, the way things are understood by a subject, matters. For Collins, the tacit status of the knowledge of a subject can be undermined through a kind of action at a distance if the process they follow is codified elsewhere (Thornton 2013). The passage quoted does not explicitly equate tacit knowledge with the behaviour of cats, dogs, trees and sieves but it does suggest that thinking of sieves sorting stones is a good way to demystify human tacit knowledge. Whatever the independent merits of an ontology of strings, however, it is far from an account of knowledge itself. It is pitched at the level of reference rather than sense and ontology rather than epistemology. Polanyi was right to stress the importance of a form of knowledge at the level of the person, expressive of human minds, tied to context and practical and presupposed by, but exceeding, linguistic codification. The challenge is to articulate it. Construing tacit knowledge as conceptually structured, context dependent, practical knowledge is one way to meet that challenge.

200

Tacit Knowledge

Related topics Chapters 1, 4, 13, 15, 16, 18

References Carroll, L. 1875. Alice through the looking glass. New York: Macmillan. Cavell, S. 1969. Must we mean what we say? Cambridge: Cambridge University Press. Collins, H. 1985. Changing order: replication and induction in scientific practice. London: Sage. Collins, H. 2010. Tacit and explicit knowledge. Chicago: University of Chicago Press. Davies, M. 1987. “Tacit knowledge and semantic theory: can a five percent difference matter?”. Mind, 96: 441–462. Dreyfus, H. L., and Dreyfus, S. E. 1986. Mind over machine. New York: Macmillan. Evans, G. 1981. “Semantic theory and tacit knowledge”. In S. H. Holtzman and C. M. Leitch, eds., Wittgenstein: to follow a rule. London: Routledge and Kegan Paul. Evans, G. 1982. The varieties of reference. Oxford: Clarendon Press. Gascoigne, N., and Thornton, T. 2013. Tacit knowledge. Durham: Acumen. Gellatly, A., ed. 1986. The skillful mind. Milton Keynes: Open University Press. Grene, M. 1977. “Tacit knowing: grounds for a revolution in philosophy”. Journal of the British Society for Phenomenology, 8: 164–171. Guilick, W. 2016. “Relating Polanyi’s tacit dimension to social epistemology: three recent interpretations”. Social Epistemology, 30: 297–325. Martin, R. D. 1994. The specialist chick sexer. Melbourne: Bernal. McDowell, J. 1994. Mind and world. Cambridge, MA: Harvard University Press. McDowell, J. 2009. The engaged intellect. Cambridge, MA: Harvard University Press. Miller, A. 1997. “Tacit knowledge”. In B. Hale and C. Wright, eds., A companion to the philosophy of language. Oxford: Blackwell. Noë, A. 2005. “Against intellectualism”. Analysis, 65: 278–290. Polanyi, M. 1962. Personal knowledge. Chicago: University of Chicago Press. Polanyi, M. 1967a. “Sense-giving and sense-reading”. Philosophy, 42: 301–325. Polanyi, M. 1967b. The tacit dimension. Chicago: University of Chicago Press. Polanyi, M. 1969. Knowing and being. Chicago: University of Chicago Press. Ryle, G. 1945. “Knowing how and knowing that”. Proceedings of the Aristotelian Society, 46: 1–16. Sellars, W. 1997. Empiricism and the philosophy of mind. Cambridge, MA: Harvard University Press. Stanley, J., and Williamson, T. 2001. “Knowing how”. The Journal of Philosophy 97: 411–444. Thornton, T. 1997. “Intention, rule following and the strategic role of Wright’s order of determination test”. Philosophical Investigations, 20: 136–151. Thornton, T. 2013. “Tacit knowledge and its antonyms” Philosophia Scientiae 17: 93–106 Wittgenstein, L. 2009. Philosophical investigations. Oxford: Blackwell. Wright, C. 1986. “Theory of meaning and speaker’s knowledge”. In Realism, meaning and truth. Oxford: Blackwell. Wright, C. 2001. Rails to infinity. Cambridge, MA: Harvard University Press.

201

15 COLLECTIVE AND DISTRIBUTED KNOWLEDGE Studies of Expertise and Experience Harry Collins It is easy to confuse knowledge with information and practical understanding with practical ability. If these things were the same, expertise and skill would just be a matter of study and practice by individuals. It’s tempting to see them this way because that’s how we experience them as we learn to live and succeed in society – the more we study and the harder we practice the more our abilities are recognised with better exam results, more success in competitions, more respect, more standing in society and better jobs. What’s missing from this view is that much of our knowledge and skill diffuses in from our surroundings tacitly and without any conscious effort; grow up in one place and one set of abilities will come naturally, grow up in another place and it will be another set of abilities. The very possibility of knowing something and acquiring a practical skill depends on the embedding society in two respects: the substance depends on the society and the chance of our acquiring that substance depends on how thoroughly we are immersed in it. This is the perspective of an approach known as ‘Studies of Expertise and Experience’, with the acronym ‘SEE’ (e.g., Collins and Evans 2007). A crucial feature of SEE is that the same considerations apply whether what is being acquired is a matter of childhood socialisation or specialist adult abilities. The exposition is divided into three sections. Section 1 relates SEE to the Wittgensteinian idea of a ‘form of life’ and looks at how one acquires a form of life. Science is an especially good field site for illustrating the approach because many of its activities unfold within well-defined boundaries but the results can be generalised. The similarities and differences with Polanyi’s tacit knowledge and Reber’s implicit learning are mentioned. A theory of the tacit is set out. Section 2 describes newer developments.1 Wittgenstein does not consider change, but science, the favoured field site, is always changing. The integral relationship of trust and the transfer and development of new knowledge is explained with real examples. The respective contributions of a shared language to the transfer of understanding of practical matters, on the one hand, and of participation in physical practices on the other, are pulled apart by considering the division of labour; the idea of ‘interactional expertise’ is explained along with the use of the imitation game as an empirical test of the claims. The role of the body is given less importance than in other treatments and this helps to explain how it is that the physically challenged can be linguistically fluent. The new model of expertise as socialisation into specialist groups resolves the philosophical problems of standard approaches which find it hard to cope with the changing definition of ‘expert’ over time and across locations, and of disagreement among co-located 202

DOI: 10.4324/9781003014584-19

Collective and Distributed Knowledge

experts. It allows that an immensely difficult expertise such as natural language speaking can be a property of whole populations irrespective of the distribution of self-conscious effort. A threedimensional model of expertise is described and the ‘fractal model’ of society is also set out. Section 3 looks very briefly at some philosophical, social and political consequences of what has been described in the first two sections. The ‘experimenter’s regress’ leads to new models of science and demands new kinds of justifications for scientific expertise. Socialisation depends on trust and the building of secure knowledge depends of face-to-face communication which cannot be replaced by remote communication over the internet. In principle, human societies of any kind can be understood by other humans if immersion in the linguistic culture is long enough and deep enough so a member of any group could, in principle, understand and represent any other group. Language-based procedures, such as the Turing Test, are sufficient for identifying human-like intelligence in machines but to pass a demanding test the computer would have to be capable of being socialised in human societies and this is the major obstacle to the prospect of AI.2

1.  The Philosophical and Empirical Starting Point Forms of Life The idea that much knowledge is located in society rather than individuals starts with the sociology of knowledge. The notion of ‘society’ employed here is Wittgenstein’s idea of a ‘form of life’ but there are undoubtedly many other philosophical routes to a similar position: ‘knowledge’ comprises the mixture of concepts and practices that make up our lives. An introduction to the idea of a form of life can be found in the late Peter Winch’s 1958 book, The Idea of a Social Science, where he explains it with the example of the idea of ‘germ’. We understand the concept of germ because we see surgeons ritualistically scrubbing and robing before an operation and we understand the scrubbing and robing because we have the concept of germ. Take away the concept of germ and the scrubbing and robing would make no sense; replace the scrubbing and robing with the wearing of filthy, blood-stained, garments and the idea of germ would make no sense; the actions and the ideas combined are what make meaning and meaning is the sum of such actions in society – varying from society-to-society and from epoch-to-epoch within a single society. Society is the locus of knowledge and most of it seeps in ‘tacitly’ without our being aware that we are acquiring knowledge.3 A useful analogy for this process is the way language changes and is learned: no individual is in control of language and when an individual learns language, they learn it by being immersed in language-speaking society and what they learn is a matter of where and when they learn. As we will see, language is more than an analogy because it also has a central role in the acquisition of knowledge, though in deeply tacit ways as well as explicit instruction.

Tacit Knowledge A good way to study these processes is to look at scientific change. In an early study, the author showed that, the inventor aside, builders of a new kind of laser, the ‘Transversely Excited, Atmospheric Pressure, CO2 laser (TEA-laser)’, though they had access to many detailed published descriptions of the device, succeeded in making it work only if they had social interaction in the laboratory of someone who already had a working device: learning to build a TEA-laser was like acquiring fluency in a new language – social immersion among the equivalent of native speakers was necessary (Collins 1974). In Wittgensteinian terms, immersion was the only way to learn the unspoken and, ultimately unspeakable, rules for applying the spoken or speakable rules, the 203

Harry Collins

totality of which cannot be explicated on pain of regress. Some rules can be explained a result of careful attention: for instance, it later turned out that attaching the top electrode to the capacitor had to be done, awkwardly, with a very short lead. In the early days, however, laser builders did not know this. One consequence was that their laser would work only if they mimicked the physical lay-out of a successful device but they did not know why this was important so physical lay-out was not mentioned in published papers; working from a circuit diagram would lead to failure. But that is only a single instance of a rule for the application of a rule becoming revealed; there will always be more that is transmitted only in the course of socialisation.

Usefulness of Science as a Field Site and Brief Comparison With Other Approaches to Tacit Knowledge Though the approach covers all knowledge, science is a useful field site because scientists are continually changing what they count as knowledge and the locations of the processes by which scientific knowledge changes are relatively easier to identify and access than in the case of, say, fashion or politics.4 We can feel relatively confident that we can see most of what is going on when, say, one set of scientific actions and concepts – say those associated with the discovery of gravitational waves – become normal and another set – those associated with failure to discover gravitational waves – disappear from social life. Though the starting point for this approach was the sociology of knowledge and Wittgenstein’s later philosophy, there are resonances with Polanyi’s notion of tacit knowledge (1958, 1966) and with Reber’s work on implicit learning (1967, 1993). Polanyi, though also working out of the sciences, and as the title of his flagship work, Personal Knowledge, indicates, focusses on individual scientist’s ability to intuit fruitful directions for research rather than the social location of knowledge. Reber’s experiments on the learning of implicit grammars, is immediately relevant; Reber shows that people can learn the grammar of language without being aware that they are learning it.5 Both approaches are, however, opposed to idealised models of science which take it to be a kind of fully explicable logical process supported by the unambiguous results of experiments. Thus Polanyi discounts the model in favour of the centrality of personal judgement organised by a republic of science while Reber’s implicit learning began with opposition to the notion, put forward by Bruner in the 1950s (Bruner et al. 1956), that learning is self-conscious, hypotheticodeductive reasoning (which sociological research has also now shown to be unworkable).

Three Kinds of Tacit Knowledge and the Difficult Idea of Explication A series of studies of the ‘transmission of tacit knowledge’, described in the next section, indicated the need for a more general theory of the meaning of ‘tacit’. In Tacit and Explicit Knowledge (2010), tacit knowledge is classified into three kinds depending on the obstacles in the way of making it explicit. The first half of the book is dedicated to trying resolve the far more difficult problem of the meaning of ‘explicable’. The explicit is defined as that which is transmitted between person and person by means of information contained in ‘strings’ – a generalised term for anything that has a pattern, the term ‘symbol’ being rejected since it appears to answer the question of how information is transmitted before it has been asked. Four meanings for ‘explicable’ are extracted from the discussion:6 1. Explicable by elaboration:  A longer string affords meaning when a short one does not. Typically, the domain of teachers and explainers 204

Collective and Distributed Knowledge

2. Explicable by transformation:  Physical transformation of strings enhances their causal effect and affordance. The domain of, for instance, spy-catchers, code-crackers and translators. 3. Explicable as mechanization:  A string is transformed into mechanical causes and effects that mimic human action. Typically, the domain of engineers 4. Explicable as explanation:  Mechanical causes and effects are transformed into strings called scientific explanations. The domain of scientists. The three kinds of tacit knowledge which relate to these four meanings of explicit are termed ‘relational’, ‘somatic’ and ‘collective’. ‘Relational tacit knowledge’ (RTK) is tacit rather than explicit for various logistic and contingent reasons. An example of RTK is the relationship between the top electrode and the capacitor of a TEA-laser, something that is unknown at one time such that it can’t be explicated and has to be transmitted through social interaction but becomes explicable (in meaning 4) at a later time. Crucially, since our understanding of the world is continually developing, while any element of RTK is potentially explicable, not all of it is explicable; one can see how this fits with Wittgenstein’s account of the rules needed to explain the rules for applying rules – each is potentially explicable but to try to explain them all leads to a regress.7 ‘Somatic tacit knowledge’ (STK) is associated with the material affordances of the human body and brain. Once more, one can imagine each of these eventually being explicated in terms of meaning ‘3’ of explication and that is where the program of robotics and artificial intelligence is trying to go. ‘Collective tacit knowledge’ (CTK) is the only category which seems profoundly resistant to explication because it is embedded in social practices such as the rules of language, or of driving in traffic in different countries. Some the rules for these things seem to be explicable but the rules for the application of the rules change as social life changes in ways that are impossible to predict. The way language is used is a ready example of CTK, with the vernacular meaning of words changing rapidly in ways that sometimes make language hard to understand (a common problem for the conversational interaction of adults and teenagers in Western societies). Furthermore, native speakers can accept and ‘repair’ all kinds of creative linguistic rule-breaking behaviour while rejecting other kinds but without being able to describe how and why they do it. We will return to this topic when we get to Artificial Intelligence in Section 3.

2.  New Developments Change Wittgenstein was concerned with the very meaning of knowledge whereas science lends itself to the study of change. In the case of gravitational waves, the change from disbelief that they had been detected or could be detected, to belief that they had been detected, took about 50 years from the start of attempts to detect them.8 The author hung around with the community for all but the first five or so of those 50 years. About half-a-dozen claims to have detected the waves were made during those years but, the final claim aside, the normal response was that they were not valid; many scientists thought that the waves were undetectable, either in theory or in practice. Since 2016 the normal has changed so that nowadays it is almost uniformly established that gravitational waves are being regularly detected. A change in what is normal affects the way people act. Claims before 2015 were disbelieved, now they will affect astrophysical calculations; affect recruitment, promotion and career-choices in physics departments; affect the way astronomers point their telescopes if the claim is of the right sort to emit electromagnetic radiation; affect the spending of research grant money in many countries; and so on. All these 205

Harry Collins

are the equivalent of the surgeons discarding their bloody waistcoats and starting to scrub their hands. The 2015 discovery will affect the meaning of words in our society in the same way as the ‘discovery’ of germs affected it. The process of change is worth studying because scientific knowledge can no longer be seen as simply a matter of testing by experiments and observations confirmed by replication. Because the acquisition of experimental skills involves socialisation, the success of which can be monitored only by experimental outcomes, in areas where results are disputed there is ample room to dispute whether an experiment or observation has been completed successfully; this gives rise to the ‘experimenter’s regress’.9 Because of the experimenter’s regress, scientific ‘discovery’ cannot be done by formula but is a matter of shifts of credibility within the scientific community – a cultural shift in the assessment of the value of observations and calculations; scientific change can no longer be described as arising quasi-automatically from observations and experiment.10 Because scientific change is a cultural shift, it is no surprise that there are still small groups of scientists who won’t accept the change and class the trustworthy and untrustworthy in a different way when it comes to the detection of gravitational waves.11

Tacit Knowledge Transfer, Change and Trust Collins (2001) shows the intimate relationship between the transfer of tacit knowledge and trust. It involves the transfer, from Russia to the UK, of the ability to make a very delicate but crucial measurement on the properties of sapphire, a measurement that, at the time, was believed to be vital for the construction of new generations of gravitational wave detectors. The measurement reported by the Russians was so out of line with other measurements that it would not be believed by Western scientists unless they could repeat it to their satisfaction. A Russian visitor who had pioneered the measurement spent two weeks in a Scottish laboratory, bringing his own samples of sapphire, but without success; he left a sample of sapphire with the Scots for them to continue the work but when they still failed, he informed them that the sample was faulty. All this would normally have confirmed the scepticism felt at the outset but the way the Russian visitor behaved was enough to persuade the UK scientists that he was credible and to persevere well beyond the normal limits of time and energy that would have been expended under these circumstances. Only as a result of this kind of perseverance, well beyond the normal, did the Scots scientists manage to repeat the measurement. Trust is at the heart of scientific change.

Language Practice and Interactional Expertise Another new departure is to examine, much more closely, the contributions of language and physical actions to the process of socialisation, with language being given a central role. This involves marked disagreement with philosophers such as Hubert Dreyfus.12 Dreyfus, following Heidegger, stressed the importance of practices and the role of the body in understanding. The central term is ‘interactional expertise’; this is fluency in the spoken language of a practical domain. A ‘special’ interactional expert is one who possesses interactional expertise but no practical competence in any of the practices of the domain and it is argued that such an expert can possess as much understanding of the domain as one who is practically proficient when tested in their use of language – that is, the practical judgments they will make will be just as good as those made by someone who is physically proficient. Part of the long resistance to the idea that fluency in language alone can give rise to sound practical judgement seems to have to do with the associating language with explicit knowledge 206

Collective and Distributed Knowledge

and practice with tacit abilities, whereas fluency in language is itself a tacit-knowledge laden skilful practice acquired through social interaction. The stress on the importance of language was developed in the course of debate, starting in the early 1990s. Dreyfus’s well-known position was that to understand a practice you had to engage in it. ‘Understanding’, in the interactional expertise sense, does not mean being able to execute the practices that are understood but being able to make practice-relevant judgements that are indistinguishable from those which would be made by those who actually practice; Dreyfus claimed this was impossible.

Interactional Expertise and the Division of Labour There are various ways of mounting the argument about interactional expertise but a simple one is to consider the division of labour.13 Consider a specialist domain such as gravitational wave detection physics (GW physics). Figure 15.1 is meant to show the division of labour in a specialist domain with a number of sub-specialists – numbered 1-n – within it. The hammers and anvils represent the exercise of different specialist practical expertises within GW physics while the bundles of waves represent the common ‘practice language’ – the medium of the interactional expertise – which makes a productive division of labour possible. The possession of interactional expertise in GW physics makes it possible for, say, specialist 2 to co-ordinate his or her work with specialist 4 even though 2 cannot do 4’s work; the same applies across the domain. If the language spoken by the specialist who practices specialty 2 and speciality 4 were not largely the same, when it comes to understanding practice, they would not be able to co-ordinate their work and the same applies across the domain. The domain language – the GW physics ‘practice language’ – must contain practice insofar as it enables all the parties to make similar practical judgements about how best to work together; they have to be able to discuss these things in a mutually comprehensible way. The smiling figure without a hammer and anvil is a ‘special’ interactional expert who has acquired fluency in the language but cannot practice any of the specialist practices – someone

1

…

3 5 2 4

Figure 15.1  A practice domain (originally Figure 15.2 in Collins (2011))

207

n

Harry Collins

like Collins, the sociologist immersed in the domain, or the technical manager of the domain, whose one-time practices pertained to high-energy physics not gravitational wave physics but makes judgements within gravitational wave physics without engaging in any of its many specialist practices. One can see that the only difference between the smiling figure and the numbered specialists is the possession of one or two practical expertises! The difference cannot lie in the ability to practice ‘the practice of the domain’ since no-one practices such a thing: the specialists practice only one or two specialties each; there is no universal ‘domain practice’. This makes it easier to understand how someone who does not practice can come to understand the domain practices through acquisition of fluency in the language, because that is what the ‘contributory experts’ – those who do practice at least one of the specialties, 1-n, have to accomplish too if they are to coordinate their work with one another, and that is why the contributory experts must also be interactional experts – that’s why they spend so much of their time going to conferences and talking to each other. The point is nicely illustrated by Gary Sanders, the one-time project manager of LIGO, who agreed with the idea that he possessed interactional expertise though not actually practising any of the specialties in the domain and who, when interviewed at a later time, when he had become the manager of the rather different 30-meter telescope project, explained his role. Here the subject is the narrowly esoteric topic of ‘adaptive optics’ (which take the twinkle out of the stars in modern terrestrial telescopes): I can sit down in a group of adaptive optics experts who will come to me and say, “ ‘Gary, you’re wrong, multi-object adaptive optics will be ready at first light and it will give the following advantages” – and others will say, “ ‘No, it’s multi-conjugative adaptive optics,” ’ and I can give them four reasons why we should go with multiconjugative adaptive optics based on the kind of science we want to do, the readiness of the technical components, when we need them, and so on. And I will see as I am talking about it that the room is looking back at me and they’re saying, “ ‘He does have a line, he’s thought it through, and yes.’ ” But if someone said to me, “ ‘OK Sanders, we agree with you, now go and design a multi-conjugative adaptive optics system,’ ” I couldn’t do it. I couldn’t sit down and write out the equations – But I can draw a diagram of what each part does, where the technological readiness of each one is, what the hard parts are – I know the language. I actually feel qualified to make the decisions. (Collins and Sanders 2007: 629)

Language Is the Property of the Collective Not the Individual The other thing that the figure makes clear is that the specialist ‘practice language’ found in the domain – the language acquired by the smiling figure and represented by the bundles of waves – would not exist in the absence of the figures 1-n actively carrying out their practices. The practice language could not be developed by a collection of smiling figures doing nothing but conversing because there must be practices going on if there is to be a practice language in the first place. The domain’s practices, as engaged in by the contributory experts, 1-n, are gathered together and captured in the domain practice language and can then be used to transmit an understanding of those domain practices to the smiling figure and each other. One sees clearly, then, that the language is a property of the group which inhabits the domain, not any individual: the language absorbs and is formed by all the practical understandings of the practicing

208

Collective and Distributed Knowledge

individuals, 1-n, and makes all those practices available to every individual in the domain including those, like the smiling figure, who do not practice at all. This is one of the reasons it makes sense to say that ‘language contains practice’ and is another illustration of the way this kind of knowledge is the property of the collective not the individual.

The Role of the Body Because individuals can make practical judgements without practicing so long as they are immersed in the discourse, even though actual practice is essential when we consider the collective level, the role of the body is minimised in the acquisition of the understanding of the practical world. We can, then, understand why excellent sports coaches and commentators do not have to have been top performers in a sport and we can understand how it is that the physically challenged, who are unable to practice most of the practices pertaining to their native society, do not show any defect in their linguistic fluency.14 I believe it was Dreyfus and his colleagues’ failure to make this distinction between the individual and the collective that led them to paint themselves into a corner, proclaiming, following Heidegger, that one could not understand a practice without practising it. It followed that no sports coach could give useful advice unless they had played the game at the highest level and that if Collins could pass as a gravitational wave physicist there must be something ‘funny’ about gravitational wave physics.15 Both of these claims seem incorrect. On the other hand, it would be very difficult to understand the working of complex division of labour, including technical management, peer review, the working of science-grant distribution agencies, and, indeed, the cooperation between different sectors of society such as men and women, if it was impossible to understand the practices of persons in whose practices one could not participate.

Imitation Games as a Test of the Idea of Interactional Expertise To some extent these claims can be tested in ‘Imitation Games’, which test naturalistic phenomena with experimental investigations, so lean a little way toward psychological methodology. The Turing Test is an imitation game with a computer trying to imitate a human and is based on imitation games where men and women try to imitate each other. We capitalise ‘Imitation Game’ to indicate a method with a carefully worked out protocol, where individuals or groups try to pretend to be members of other groups. For example, Collins has used two different methodologies to test his ability to pass as a gravitational wave physicist after many years in the role of Figure 15.1’s ‘smiling figure’, immersed in the culture of GW physics but without participating in its practices. Collins did better than chance in 2006, when he and another GW physicist independently answered a series of technical questions set by a third GW physicist and the answers were compared by a series of other GW physicists. About ten years later, the result was supported in a test with a more elaborate protocol. Imitation Games have also been used to test the theory of interactional expertise in comparative tests of samples of the colour-blind, the blind, and persons with perfect pitch, and used to test wider sociological theses such as the understanding of one social group for another, for instance, Scots and English and vice versa; homosexuals and heterosexuals in various countries; religious and non-religious groups in various countries, and so on. These are ways of measuring and comparing the varying levels of understanding of one group’s implicit (and explicit) knowledge of another group and how this understanding varies from country to country.16

209

Harry Collins

Esotericity

Individual or group accomplishment

Exposure to tacit knowledge of domain

Figure 15.2  Expertise space (originally Figure 15.2 in Collins (2013: 257))

Expertise Is Socialisation The SEE approach also leads to a new model of expertise that differs from standard philosophical and psychological models. The acquisition of expertise is seen as socialisation into an expert group while individual competence at acquiring socialisation, which is the usual focus of philosophical and psychological models is only one dimension (the Z-axis going into the page) of a three-dimensional ‘expertise space’ in which the X-axis is the opportunity and actuality of exposure to the domain and the Y-axis is the extent to which the domain is esoteric or ubiquitous. This is shown in Figure 15.2. This model resolves certain problems such as natural language fluency not counting as an ‘expertise’ in the standard model since expertises are always taken to be the possession of elite groups (even though acquisition of natural language fluency is the topic of the Turing Test for human-like abilities intelligence because it is so difficult). It also resolves the problem of changes in what counts as expertise over historical time (e.g. ordinary car-driving, alchemy) and the, typical, violent disagreement among experts as to what counts as expertise in a domain (such as GW physics). The approach also give rise to ‘the fractal model of society’, as a cascade of expertise domains, with ubiquitous expertises at the top, embedded within one another and with multi- dimensional overlapping.17

3.  Some Social and Political Consequences Science as ‘Craftwork With Integrity’ Is a Check and Balance in Pluralist Democracies What has been described in Sections  1 and 2 has various larger consequences and presents new problems. The experimenter’s regress and the social nature of science in general means 210

Collective and Distributed Knowledge

that science needs to be justified in ways other than its automatic and incontestable ability to uncover the truth. Instead science must be justified as an exemplary model of the search for truth, irrespective of its ability to resolve every issue. Seen as ‘craftwork with integrity’ science can provide leadership in the execution of democratic values without sacrificing its claim to expertise in its domain, and can act as a check and balance in pluralist societies. An understanding of the role of trust in the transfer and development of scientific knowledge also indicates that it is vital to preserve and understand the importance of face-to-face communication in small groups in the age of the internet and the spread of disinformation.18 Public education is needed to explain and reclaim the vital role of expertise in pluralistic democracies and the way citizens should understand and rank the contributions different specialist domains found in (the fractal model of) society when faced with a disputed technical problem.

All Humans Can Potentially Acquire All Practice Languages Assuming all groups of humans share a genetic heritage there is no in-principle obstacle to any human acquiring any initially foreign group culture to the point of their being able to represent it with authenticity. The practical difficulties in the way of this are often severe and may be insuperable (Collins 2020).

The Role of Language Is Such That the Turing Test Is an Adequate Test for Human Intelligence The importance of language in human society and the way language embeds practice elevates the power of the Turing Test as test of human abilities without the computer needing to demonstrate robotic abilities or having a body to exhibit an understanding of practical life. Winograd schemas can already illustrate this point. For example, to translate the alternative form of the following sentence into a gendered language requires an understanding of certain practices that would be ubiquitous in a society but aren’t resolvable by the best deep-learning techniques unless the example can be found in the existing world of accessible text: ‘The trophy did not fit the suitcase because it was too big/small.’ If the final word is big the ‘it’ must refer to the trophy, if ‘small’ it must refer to the valise even though in logic a small trophy does not fit a very big suitcase! As things stand, computers are quite unable to pass Turing Tests that incorporate Winograd schemas (Levesque et al. 2012). This example, and many others, can be related to the fact that expertise, particularly collective tacit knowledge, is acquired only through socialisation. Fluency in language is a matter of collective tacit knowledge and this means that for computers to be intelligent, as defined by a demanding, but purely language based, Turing Test, requires that they be capable of being socialised in the way that human beings are socialised.19

Related Topics Chapters 1, 12, 13, 14, 15

Notes 1 The dividing line between the first and second section is somewhat arbitrary: for instance, is the theory of the tacit new or immanent? 2 This chapter has an unusually high proportion of citations to its author, who initiated many of the developments described here. But the work was done in collaboration with other co-experimenters

211

Harry Collins and co-authors, especially Robert Evans, whose work gave rise to the idea of the large-scale Imitation Game field studies. As will be explained later, some of the developments described here were driven along by a long-running debate with the late Bert Dreyfus. Many other authors are independently exploring aspects of the program. 3 This is Wittgenstein’s ‘later philosophy’ as represented by his, 1953 book, Philosophical Investigations. Winch describes his thesis as an attack on sociology which, he says is ‘misbegotten epistemology’ since social change is ‘really’ conceptual change. But his thesis that the social and the conceptual are two sides of the same coin could equally be interpreted as meaning that epistemology is misbegotten sociology. The crucial thing is the intimate relationship of concepts of actions. Bloor, in his 1983 book, Wittgenstein: A Social Theory of Knowledge, is explicit that the later Wittgenstein can be seen as a sociologist. 4 We now understand that the scientific knowledge is more affected by the wider society than we once thought, hence the ‘relatively’. 5 Though the author did not realise the link with Polanyi until his submitted paper on the TEA-laser was refereed and not until much later in the case of Reber. Major points of similarity and difference between Reber and Collins are worked out in Collins and Reber (2013). Some notable differences between the approaches are the naturalistic setting of early SEE research and the laboratory settings of implicit learning research and SEE’s interest in the nature of knowledge and Reber’s in learning. Thus the act of driving a car to work is interestingly different for Reber, depending on whether the driver is doing it self-consciously or as a background task whereas for SEE it is the same act with the same tacit and explicit components irrespective of how it is executed moment-to-moment. 6 Originally set out in Collins (2010: 81, Table 4). The italicised comments are new to this chapter. 7 Explained (in the sense of meaning 1) with the example of the game ‘Awkward Student’, in Collins (1992). 8 This was preceded by about 50 years of debate about whether Einstein’s theoretical ‘discovery’ was valid (he disbelieved it himself for a while), and that if they did exist they were detectable even in principle (see Kennefick 2007). Collins’s gravitational wave fieldwork was supported by a series of grants: 1975 SSRC £893 ‘Further Exploration of the Sociology of Scientific Phenomena’; 1975–1977 SSRC £8,250 ‘Cognitive Dislocation in Science’; 1976 Nuffield £154 ‘Replication of Experiments in Physics’; 1995–1996 ESRC (R000235603) £39,927 ‘The Life After Death of Scientific Ideas: Gravity Waves and Networks’; 1996–2001 ESRC (R000236826) £140,000 ‘Physics in Transition’; 2002– 2006 ESRC (R000239414) £177,718 ‘Founding a New Astronomy’; 2007–2009 ESRC (RES-000– 22–2384) £48,698 ‘The Sociology of Discovery’; 2010–2019 funded through US National Science Foundation grant PHY-0854812 to Syracuse University ‘Toward Detection of Gravitational Waves with Enhanced LIGO and Advanced LIGO’, P.I.: Peter Saulson. Open-ended. ‘To complete the sociological history of gravitational wave detection’. 9 For a complete analysis see Collins (1992). 10 Collins account of the process of this change in GW physics can be found in three books; it is now less surprising than it might be they total 1654 printed pages: Collins 2004a, 2011, 2017a. 11 Thus the cover of the science magazine, New Scientist, of 3 November, 2018 bears the headline, ‘Did we really discover gravitational waves? Breakthrough physics result questioned’, and the story within sets out the view of some of the (well-qualified and well-established) scientists who are still trying to reverse the trend of normalcy in this regard and re-establish disbelief in the discovery claims on the grounds that the statistical calculations have been done incorrectly. 12 Bert Dreyfus died in 2017. I stress the importance of this friendly tension in my remembrance of him – Collins (2019b). 13 Collins and Evans (2015) is an extensive analysis of the various conceptual sources and consequences of the idea of interactional expertise. 14 Deafness is a much more difficult physical challenge. This theme began from Dreyfus’s famous, 1967, paper entitled ‘Why Computers Must Have Bodies in Order to be Intelligent’ followed up by his heroic, 1972, book, What Computers Can’t Do. More on Dreyfus’s view of the importance of practice and the body can be found in his 2017 book titled Skillful performance: enacting expertise, competence, and capabilities in organizations. Perspectives on process organization studies (P-PROS). For the dispute with Dreyfus and Doug Lenat over the body in the context of the argument over artificial intelligence, see Collins 2004b, 2004c. 15 The, always friendly, debate was carried through in person – usually at conferences – as well as in print. This exchange was at a conference in Austria in 2009. 16 The gravitational wave Imitation Games can be found reported in Giles; 2006 and described in detail in Collins (2019a: Ch. 7) while the second GW test is reported in Collins (2017: Ch. 14). Description of other uses of the methodology can be found in Collins and Evans (2014;) Collins et al. (2006, 2017,

212

Collective and Distributed Knowledge 2019, under contract); Evans et al. (2019). Many of the international tests were supported by 2011– 2016 European Research Council Advanced Grant (269463 IMGAME) €2,260,083 ‘A new method for cross-cultural and cross-temporal comparison of societies’ and 2012–2013 European Research Council Proof of Concept Grant (297467 IMCOM) €150,000 ‘IMGAME Commercial’. 17 See, for example, Collins et al. 2019. 18 These are the themes of a series of books with the titles Are We All Scientific Experts Now? (Collins 2014), Why Democracies Need Science (Collins and Evans 2017), Experts and the Will of the People (Collins 2019) and Face-to-Face: Communication and the liquidity of knowledge (Collins, Evans, Innes, Kennedy, Mason-Wilkes and McLevey; under submission). 19 See Collins (1990); Collins and Kusch (1998); Collins (2018). Collins (2018) suggests that deep learning computers have been so successful because they come closer to being socialisable than previous generations of, so-called intelligent machines, but that they’re failings are still failings of socialisation. Collins (2018: Ch. 10) discusses Winograd schemas and other, simple, but still harder tests of linguistic fluency that depend on readily understood – by fluent humans – creative rule-breaking.

References Bloor, D. 1983. Wittgenstein: a social theory of knowledge. London: Macmillan. Bruner, J. S., Goodnow, J. J., and Austin, G. A. 1956. A study of thinking. New York: John Wiley and Sons. Collins, H. 1974. “The TEA set: tacit knowledge and scientific networks”. Science Studies, 4: 165–186. Collins, H. 1990. Artificial experts: social knowledge and intelligent machines. Cambridge, MA: MIT Press. Collins, H. 1992. Changing order: replication and induction in scientific practice. 2nd edn. Chicago: University of Chicago Press. Collins, H. 2001. “Tacit knowledge, trust, and the q of sapphire”. Social Studies of Science, 31: 71–85 Collins, H. 2004a. Gravity’s shadow: the search for gravitational waves. Chicago: University of Chicago Press. Collins, H. 2004b. “Interactional expertise as a third kind of knowledge”. Phenomenology and the Cognitive Sciences, 3: 125–143 Collins, H. 2004c. “The trouble with Madeleine”. Phenomenology and the Cognitive Sciences, 3: 165–170 Collins, H. 2010. Tacit and explicit knowledge. Chicago: University of Chicago Press Collins, H. 2011. “Language and practice”. Social studies of science, 41: 271–300. https://doi. org/10.1177/0306312711399665 Collins, H. 2013. “Three dimensions of expertise”. Phenomenology and the Cognitive Sciences, 12: 253–273. https://doi.org/10.1007/s11097-011-9203-5. Collins, H. 2014. Are we all scientific experts now? Cambridge: Polity Press. Collins, H. 2017a. “Interactional expertise and embodiment”. In J. Sandberg, L. Rouleau, A. Langley, and H. Tsoukas, eds., Skillful performance: enacting expertise, competence, and capabilities in organizations. Perspectives on process organization studies (P-PROS). Vol. 7. Oxford: Oxford University Press: 125–146. Collins, H. 2017b. Gravity’s kiss: the detection of gravitational waves. Cambridge, MA: MIT Press. Collins, H. 2018. Artifictional intelligence: against humanity’s surrender to computers. Cambridge: Polity Press. Collins, H. 2019a. Forms of life: The method and meaning of sociology, Cambridge, MA: MIT Press. Collins, H. 2019b. “Remembering Bert Dreyfus”. AI and Society, 32(2): 373–376 Collins, H. 2020. “Interactional Imogen: language, practice and the body”. Phenomenology and the Cognitive Sciences, 19: 933–960. https://doi.org/10.1007/s11097-020-09679-x Collins, H., and Evans, R. 2007. Rethinking expertise. Chicago: University of Chicago Press. Collins, H., and Evans, R. 2014. “Quantifying the tacit: the imitation game and social fluency”. Sociology, 48: 3–19. https://doi.org/10.1177/0038038512455735 Collins, H., and Evans, R. 2015. “Expertise revisited I – interactional expertise”. Studies in History and Philosophy of Science, 54: 113–123. Collins, H., and Evans, R. 2017. Why democracies needs science. Cambridge: Polity Press Collins, H., Evans, R., Durant, D., and Weinel, M. 2019. Experts and the will of the people: society, populism and science. Basildon: Palgrave. Collins, H., Evans, R., Ribeiro, R., and Hall, M. 2006. “Experiments with interactional expertise”. Studies in History and Philosophy of Science, 37: 656–674. Collins, H., Evans, R., Weinel, M., Lyttleton-Smith, J., Bartlett, A., and Hall, M. 2017. “The imitation game and the nature of mixed methods”. Journal of Mixed Methods Research, 11: 510–527. https://doi. org/10.1177/1558689815619824

213

Harry Collins Collins, H., Hall, M. Evans, R., Weinel, M., and O’Mahoney, H. (under contract). Imitation games: A new method for investigating societies, Cambridge, MA: MIT Press Collins, H., and Kusch, M. 1998. The shape of actions: what humans and machines can do. Cambridge, MA: MIT Press. Collins, H., and Reber, A. 2013. “Ships that pass in the night”. Philosophia Scientiae, 17: 135–154. Collins, H., and Sanders, G. 2007. “They give you the keys and say ‘drive it’: managers, referred expertise, and other expertises”. Studies in History and Philosophy of Science, 38: 621–641. Dreyfus, H. 1967. “Why computers must have bodies in order to be intelligent”. Review of Metaphysics, 21: 13–32. Dreyfus, H. 1972. What computers can’t do. Cambridge, MA: MIT Press. Dreyfus, H. 2017. “Embodied expertise according to Martin Heidegger, Maurice Merleau-Ponty, and Samuel Todes”. In J. Sandberg, L. Rouleau, A. Langley, and H. Tsoukas, eds., Skillful performance: enacting expertise, competence, and capabilities in organizations. Perspectives on process organization studies (P-PROS). Vol. 7. Oxford: Oxford University Press: 147–159. Evans, R., Collins, H., Weinel, M., Lyttleton‐Smith, J., O’Mahoney, H., and Leonard‐Clarke, W. 2019. “Groups and individuals: conformity and diversity in the performance of gendered identities”. British Journal of Sociology, 70: 1561–1581. Giles, J. 2006. “Sociologist fools physics judges”. Nature, 442: 8. Kennefick, D. 2007. Traveling at the speed of thought: Einstein and the quest for gravitational waves. Princeton: Princeton University Press Levesque, H., Davis, E., and Morgenstern, L. 2012. “The Winograd schema challenge”. In Proceedings of the thirteenth international conference on principles of knowledge representation and reasoning. Rome: AAAI Press: 552–561. Polanyi, M. 1958. Personal knowledge. London: Routledge and Kegan Paul. Polanyi, M. 1966. The tacit dimension. Chicago: University of Chicago Press. Reber, A. S. 1967. “Implicit learning of artificial grammars”. Journal of Verbal Learning and Verbal Behavior, 6: 855–863. Reber, A. S. 1993. Implicit learning and tacit knowledge: an essay on the cognitive unconscious. Oxford: Oxford University Press. Winch, P. G. 1958. The idea of a social science. London: Routledge and Kegan Paul. Wittgenstein, L. 1953. Philosophical investigations. Oxford: Blackwell.

214

16 IMPLICIT BELIEFS Joseph Bendaña

1. Introduction Implicit beliefs can be preliminarily characterized by contrast with explicit beliefs (Lycan 1988; Dummett 1991; Crimmins 1992).1 The nature of explicit beliefs is itself notoriously controversial (Field 1978; Churchland 1981; Dennett 1989; Fodor 1987), but they are often characterized as beliefs that an agent in normal circumstances is aware of and could articulate (Quilty-Dunn and Mandelbaum 2018; Schwitzgebel 2021a). But whatever criterion of explicitness you use, implicit beliefs are, minimally, beliefs that aren’t explicit. However, characterizing implicit beliefs ex negativo raises more questions than it answers. What makes a belief implicit? What makes them beliefs? When is it correct to attribute them to an agent? And perhaps most importantly, why think there are any at all? Making these questions even more difficult is the fact that the term “implicit beliefs” is not used consistently in philosophy or psychology. Its primary usage in philosophy has been in attempts to bolster representationalist theories of belief against dispositionalist, instrumentalist, and eliminativists attacks. Meanwhile, in attitude psychology and the philosophy thereof, when the term is used it is typically applied to attitudes that are measured without verbal reports. Following is an introduction to these characterizations of implicit belief and some of the core philosophical debates that accompany them.

2. Representationalism Representationalism about belief is minimally the view that to have a belief is to stand in a particular relation to a mental representation. Proponents of the view usually also hold that these representations must have semantic content and be able to enter into cognitive functions characteristic of states that capture a world-to-mind relationship, but the view can be fleshed out in a myriad of ways. Details aside, as Quilty-Dunn and Mandelbaum (2018) point out, such an account is about as close as it gets to orthodoxy in cognitive science and the philosophy thereof.

DOI: 10.4324/9781003014584-20 215

Joseph Bendaña

3.  Problems for Representationalism Representationalism seems to have problems accommodating common intuitions about the truth of belief reports. There are a variety of cases where certain belief attributions seem true, but it seems implausible that an agent has a mental representation with the content of the belief that is supposedly correctly attributed. The first kind of case alleges that representationalism demands implausible attributions of representations. For instance, it might seem natural to say that you believe that no T-Rex has ever played an electric guitar. This seems to be the case even if you have never thought about reasons why a T-Rex couldn’t play “Reign in Blood” (Lycan 1988: 54–55 offers additional, less Jurassic examples). However, if you’ve simply never thought about whether or not a T-Rex could play a guitar, it’s hard to see how or why, even according to representationalism, you would have a mental representation with the content “no T-Rex has ever played an electric guitar”. Second, representationalism seems to require too many representations. If you believe there are 8 planets in our solar system, it seems correct to attribute to you the belief that there are fewer than 9 planets in our solar system and the belief that there are fewer than 10 planets . . . etc. up to N planets, for any number N above 8 (Schwitzgebel 2021b). This is supposed to create a problem particularly for naïve forms of representationalism, which hold that each belief is a separate representation, since it would require more representations than could possibly be attributed to a limited, finite human. Third and finally, representations can seem irrelevant to the truth of a belief attribution. Consider Dennett’s (1989) notorious case of a chess playing agent that routinely gets the queen out early in every match without (by hypothesis) having any explicit instruction with the content “get the queen out early”. In this case, according to Dennett, it seems correct and useful to ascribe to the agent a belief that it should get the queen out early, even though the software powering the agent does not explicitly represent that content anywhere.2 Although the preceding cases are all subtly different, they are instances of the same core problem: our practice of everyday belief attribution does not seem to mesh well with representationalism. Representationalism seems to maintain that each true ascription of the belief that P to an agent requires the agent to have a representation that P. However, at least at first glance, it appears that intuitively seems that the truth of belief attributions to an agent is often independent of what, if any, mental representations that agent has.

4.  Implicit Belief Representationalists invoked the notion of implicit belief in direct response to the preceding challenges (Field 1978; Lycan 1988; Crimmins 1992). They maintained that our practices of belief attribution are not limited to explicit beliefs. In the kinds of trouble cases detailed earlier, critics are correct in alleging that the agent in question doesn’t have an explicit belief, but representationalism requires only that all beliefs be represented, not that they all be represented in precisely the same manner as explicit beliefs. Representationalists were painfully aware of the need to (and difficulty of) saying more about what implicit beliefs are for this answer to be at all compelling. Recall that according to representationalists, explicit beliefs are mental representations that fill the functional role of belief, whatever that role is. If implicit beliefs are minimally all the beliefs a subject has that aren’t explicit, then they must not be mental representations with semantic contents that play the typical cognition- and behavior-guiding role of belief. So either they have a different functional role than beliefs or they are not mental representations with semantic content or both. 216

Implicit Beliefs

So at least at first glance, representationalism still appears to be in serious trouble (Lycan 1988; Crimmins 1992).3 There are currently two classes of strategies for dealing with these problems. The first denies that implicit beliefs need to actually be explicit mental representations. Dispositionalism and simple consequence accounts are in this class. The second instead characterizes the implicitness of beliefs as a matter of the functional role of mental representations. The Map view of belief and Belief Fragmentation both fall into this camp. Dispositional theories (Lycan 1981), maintain that to implicitly believe a proposition is just to be disposed to have an explicit belief in the right circumstances. They typically construe having an explicit belief as having an explicit mental representation that plays the belief role in cognition. So on at least some versions of this view implicit beliefs need not be explicitly represented. Simple-consequence accounts (Field 1978), similarly, maintain that an agent implicitly believes a proposition just in case it is a simple or obvious consequence of an explicit belief. So called “virtual” accounts follow suit and hold that an agent implicitly believes P just in case it is as if an agent had an explicit belief that P (Crimmins 1992). Again on both of these views, if explicit belief is a matter of explicit representation, implicit beliefs need not be explicitly represented. However, both views suffer from serious problems. Dispositional theories are notoriously susceptible to counter-examples. Here are just a couple that Crimmins (1992: 246) points out. 1. Obvious discoveries: When talking loudly you might not believe you are talking loudly, yet you likely would explicitly believe that you are after having it pointed out to you (see Audi 1994 for additional variations of this case, which he uses to argue that people don’t actually believe all the things commonly attributed to them as tacit beliefs). 2. Opinionated people: Imagine a person who forms beliefs about a proposition immediately on the basis of irrelevant features of their cognitive state (see Mandelbaum 2016 for evidence that is depressingly all too common). For every P this person is disposed to have a belief that P. But surely this person doesn’t tacitly believe every proposition or its negation (see Lycan 1986 for the original versions and additional variations of this case). Emendations to dispositionalist theories of implicit belief may be worth pursuing, but as it stands it’s not clear that any of the proposed modifications are not susceptible to slightly modified variants of aforementioned counter-examples (see Lycan 1986 for a classic critical overview). Simple consequence theories similarly appear to generate the wrong predictions across a range of cases in which you believe that P, but P is not a logical consequence of your beliefs. For instance, according to Crimmins (1992: 247), it is likely that you believe that you have never eaten a bicycle, but this belief is unlikely to be a logical consequence of your explicit beliefs. Additionally, simple consequence theories can over-generate beliefs depending on how you cash out the notion of “simple”. By some accounts the Pythagorean theorem is just a simple consequence of simple geometric beliefs you acquired as a child, yet it also seems odd to say you believed it before going through the proof in geometry class (see Lycan 1986 and Crimmins 1992 for additional criticisms). Virtual theories, like the one proposed by Crimmins, don’t seem as susceptible to obvious counter-examples, but that is in large part because they are underdeveloped. What is it for a situation to be as if an agent had an explicit belief that P? Crimmins (1992: 249) says it is for an agent’s cognitive dispositions to be relevantly as if the agent had an explicit belief that P. However, if what it is for an agent’s cognitive dispositions to be relevantly as if an agent has an explicit belief that P is for an agent to act and reason as if P, then the analysis seems eliminative 217

Joseph Bendaña

or circular. If on the other hand we constrain the notion of relevant similarity too tightly, the dispositions could just be those sufficient for believing that P. In that case, implicit belief would appear to just boil down to explicit belief. If we leave the notion totally unconstrained, say perhaps by leaving the determination of relevance up to the belief attributor, then it looks very easy to both under- and over-generate true belief reports. In light of the kinds of objections detailed earlier, some philosophers have adopted instrumentalism (Dennett 1989) or outright eliminativism (Churchland 1981) about belief as a whole, while others have attempted to characterize implicit belief more in terms of its functional role instead of in terms of a tenuous relation to explicit belief. Among functional role theorists, some rely on particular theories of the format of the mental representations that constitute beliefs (Braddon-Mitchell and Jackson 1996).4 Others invoke format agnostic theories of cognitive architecture to attempt to draw the distinction, but it should be noted that these theorists are less concerned with accommodating intuitions about belief attributions than with explaining data about implicit attitudes from psychology and cognitive science (Mandelbaum 2016; Porot and Mandelbaum 2021; Bendaña and Mandelbaum 2021). Theories of belief according to which belief has a map-like format (as opposed to a language-like format) seem to provide an attractive solution to the problems implicit belief raises for representationalists (Braddon-Mitchell and Jackson 1996: 202).5 The basic idea is that in the same way an actual map can guide your action through the world, a map-like mental representation can guide action and reasoning in the ways that belief must: you can navigate from desk to your coffee pot because of the mental map you have of your apartment. One might be skeptical that map-like representations can play the role of belief because it is difficult to cash out the notions of similarity that they depend on to represent information or because it’s hard to see how maps can represent very abstract content. But it is helpful to keep in mind that unlike pictures, maps can combine representation via convention with representation via resemblance (see Camp 2007 for more details on how thinking with maps might work). On this view you implicitly believe everything that’s true according to the map-like representations that constitute your beliefs, while your explicit beliefs are just those that you can report or more easily read off the map. All beliefs, implicit or explicit are construed as mental representations that play a particular action and reason guiding functional role in your cognitive architecture. It is merely that some information encoded in the representation is harder to utilize for certain purposes than others. Maintaining that beliefs are map-like thus appears to handle the worries that implicit beliefs drain representationalism of its explanatory power, since the causal power, and status as beliefs of implicit beliefs is fundamentally no different in kind than that of explicit beliefs. As Schwitzgebel (2021b) notes, map theories also appear to more readily handle some of the belief attribution cases detailed earlier. For instance, if you had a map-like representation that represents New York City as 2,446.32 miles away from Los Angeles, then it also represents it as less than N miles away from Los Angeles for every number greater than 2,446.32. As long as that representation plays the right functional role you would have an explicit belief that New York City is 2,446.32 miles away from Los Angeles and an implicit belief that New York City is less than a billion miles away from Los Angeles. The ability of maps to represent infinite propositions in a compact way seems to offer a unique advantage in accounting for implicit belief. Map theories thus also seem to have an advantage enjoyed by dispositional accounts of the distinction, in that they make the distinction between implicit and explicit beliefs one of degree. The more processing it takes to utilize the information encoded in a belief representation, the more implicit it is. Whether this is enough to allow the view to accommodate intuitions about borderline cases where it’s not clear whether we want to say a subject believes that P or not is an open question (Schwitzgebel 2021b). 218

Implicit Beliefs

Map views thus seem to be one of the more attractive representationalist views of implicit belief on offer, but they also have significant problems. At least naïve versions of the view are committed to the logical closure of belief. A subject implicitly believes all the entailments of their beliefs, simply due to fact that map-like representations can arguably represent all the entailments of propositions they represent. However, there are many well-known problems with assuming the logical closure of belief and it’s not clear that maintaining that implicit beliefs are logically closed is much better (Stalnaker 1984, 1991; Yalcin 2016). Moreover, it is hard to see how map views of belief can account for the simple discoveries cases that also caused trouble for dispositional accounts of implicit belief. Whether these problems are as bad as they seem and how best to handle them is a matter of ongoing research (see Braddon-Mitchell and Jackson 1996: 185–218 for a helpful overview that favors the map view; see Camp 2007 for a deeper argument that map-like representations can fill the belief role; and Blumson 2012 and Johnson 2015 for a discussion of whether map representations are truly distinct from language-like representations). In any case, for representationalists who like functional characterizations of implicit beliefs, but do not wish to commit to the view that beliefs have a map-like format, there is an alternative. Belief Fragmentation (“Fragmentation”) is a nascent framework for theorizing about belief that provides several of the advantages that the map view has characterizing implicit belief, without committing to any particular representational format. According to the framework, an agent’s set of beliefs is broken up into subsets, which can be selectively accessed for updating, action guidance and reasoning. Metaphorically, instead of having one unified web of belief, agents have many disparate storehouses of belief and which one is accessed at any given time for any given purpose can depend on a wide variety of contextual factors. This framework can be filled in in many ways (Lewis 1982; Stalnaker 1984; Egan 2008; Elga and Rayo 2021; Bendaña and Mandelbaum 2021), but for representationalists it offers a format-agnostic way of drawing a distinction between implicit and explicit belief. Explicit beliefs are just the representations that are in the fragments that are accessible for whatever cognitive functions are constitutive of the role of explicit belief. Implicit beliefs are just the representations that play a significant enough role guiding action or reasoning such that they count as beliefs, but are also unavailable for in certain contexts or for whichever cognitive functions are deemed necessary for explicitness, for example, the processes that make the representation available for verbal reports. Fragmentation thus allows representationalists to maintain the core of their view, as both implicit and explicit beliefs still just boil down to mental representations, whatever their format, playing slightly different, but still belief-like functional roles. Like the map-view, it also offers a way of starting to explain why some beliefs are implicit in terms of their functional role. The organization of information encoded in human agents makes it such that some represented information is harder to access for some purposes than for others. The view is also capable of accommodating the idea that the distinction between implicit and explicit belief is one of degree. The harder it is to access mental representations for the functions that are constitutive of explicitness, the more implicit those representations will be. Fragmentation just maintains that difficulty will be a function of information organization instead of representational format. It offers comparable, albeit more schematic, solutions in cases where it seems correct to attribute a belief that P to an agent even though the agent arguably doesn’t have an explicit representation that P. The agent can still be characterized as having an implicit representation with the functional role of belief, where an implicit representation can be characterized as a representation to which cognitive processes have limited access.6 Fragmentation thus promises a novel and flexible way of characterizing and explaining implicit beliefs.7 However, the use of Fragmentation to give an account of implicit belief still faces challenges when dealing with some of cases that were supposed to cause problems for representationalism 219

Joseph Bendaña

as a whole. It does not obviously help us accommodate the problem of too many representations described earlier. If it is correct to attribute an infinite number of beliefs to an agent and it is correct that each belief must be represented, having finer grained accessibility conditions will not help with the worry that finite creatures can’t represent infinite propositions. That worry would have to be addressed by explaining how mental representations could compactly represent infinite propositions. It is also not clear if Fragmentation helps with the problem of irrelevant representations as raised by Dennett’s chess playing robot. If the case is supposed to be one in which the agent simply does not have a representation with the content P, but it is correct to say that the agent believes that P, then accessibility will not help. However, if the case is supposed to be one in which the content P is in fact implicitly represented perhaps then Fragmentation’s construal of implicitness in terms of access might be able to do more work. Finally, without further fleshing out, the view threatens to render agents logically oblivious, since the framework alone puts no boundaries on how finely belief fragments can be individuated. This allows for the possibility of arbitrary failures to put the pieces together. For instance, for all the framework has to say, an agent could implicitly believe that P and implicitly believe that Q, but not implicitly believe that P & Q for any content P and any content Q. So it remains to be seen whether Fragmentation can live up to its apparent promise as tool for characterizing implicit belief in a way that shores up representationalism against its problems with belief attribution.

5.  Implicit Attitudes Psychology is rife with discussions of implicit attitudes, but the debates over their nature are largely if not entirely orthogonal to the debates about implicit belief discussed earlier (Greenwald et al. 2020). However, there is some debate over what makes implicit attitudes implicit and also whether or not implicit attitudes count as beliefs. So following is a brief review how psychologists use the term “implicit attitudes” and a sketch of the more philosophical debates over whether implicit attitudes are beliefs. When psychologists discuss implicit attitudes they are typically discussing attitudes measured without utilizing explicit verbal attitude reports, for example, by tools such as the evaluative priming task (EPT) (Fazio et al. 1986) or the Implicit Association Test (IAT) (Greenwald et al. 1998). The IAT, for instance, is a measure of reaction times. During a typical IAT people try to sort words into disjunctive categories as fast and as accurately as possible using computer keys (e.g., a word like “murder” would need to be sorted into the category “good” OR “black” or the category “white” OR “bad”). The results of such tests reveal that most people sort words more accurately when the disjunctive categories are consistent with common stereotypes (e.g., subjects would be faster and less error prone if the categories were “good” OR “white” and “black” OR “bad”). Researchers have long taken this and similar patterns of behavior to indicate implicit stereotypes and preferences.8 Since the attitudes measured by these tools often conflict with explicitly reported attitudes and appear to generally be outside of awareness this is evidence that they are in some sense implicit. Nevertheless, the precise nature of the implicit/explicit distinction in psychology is just as controversial as in philosophy. Initially researchers characterized implicit attitudes simply as attitudes that could not be easily controlled or intentionally altered (Fazio et al. 1995) and that were unconscious (Greenwald and Banaji 1995). Recently both of these characterizations have faced empirical challenges (Gawronski et al. 2020). In part due to these challenges and in part to not pre-judge important theoretical questions about the nature of implicit attitudes, many psychologists researching implicit attitudes have started adopting more operational stances to

220

Implicit Beliefs

the distinction, according to which implicit attitudes just are attitudes measured with indirect measures (e.g., the IAT) (Kurdi and Dunham 2020). Research and theorizing in psychology that attempts to determine what indirect measures actually measure has focused on the nature of the processes and representations underlying the responses captured by such measures. Historically, the core theoretical debate has been between associative models (McConnell and Rydell 2014), which held that the mental representations ungirding implicit attitudes don’t directly represent particular relations between things and propositional accounts (De Houwer 2014) that held the opposite. For instance, in contrast to a propositionally structured representation, an associatively structured representation supposedly can’t represent that a banana is yellow. Rather it would just represent a banana and the color yellow, since associative link is nothing but co-occurrence (see Mandelbaum 2020 for an overview of the difference between associative structures and associative processes and Buckner 2019 for one of associative explanation). The basic idea was that if implicit attitudes are mere associations, then they should not be sensitive to relational information, whereas if they are propositional they should be. However, in light of puzzling patterns of evidence that show implicit attitudes exhibit sensitivity to relational information, but also inconsistency with explicit attitudes and deep recalcitrance to revision, many psychologists have adopted more flexible models that incorporate aspects of both the propositional and associative theories (Gawronski and Bodenhausen 2006; Sherman and Klein 2021). Given the rapid proliferation of hybrid models and large bodies of evidence that don’t seem to neatly fit any single theory, determining the nature of implicit attitudes, as well as the core empirical and theoretical commitments of the various models remains an important open question (Gawronski et al. 2020; Greenwald et al. 2020; Kurdi and Dunham 2020). Philosophers have also been interested in implicit attitudes in the context of their interaction with implicit bias (Brownstein and Saul 2016a, 2016b), but the focus of those concerned with the nature of implicit attitudes has often landed on the question of whether or not such attitudes count as beliefs. Gendler (2008) and Levy (2015) for instance, maintain that implicit attitudes are not beliefs because such attitudes seem to violate various norms of rationality which states must comply with to count as beliefs. Similarly, Madva (2016), argues that implicit attitudes fail to be sensitive to logical structure in the way that belief states must be. Mandelbaum (2016), on the other hand argues that because recent psychological evidence supports propositional models of implicit attitudes and beliefs are propositional mental representations, implicit attitudes are likely beliefs, denying that belief has a special normative profile. Bendaña (2021), argues that those who want to deny that implicit attitudes count as beliefs face a dilemma: evidence from psychology seems to illustrate that the functional role of implicit attitudes and explicit attitudes is similar enough that if either of them count as beliefs, both do, yet as it stands little in this debate is settled. There is little agreement on what the functional role of belief should be even among functionalists about belief. It also remains to be seen whether future empirical research finds functional characteristics that clearly distinguish between implicit and explicit attitudes. Why does anyone care about whether or not implicit attitudes are beliefs? One might be interested in the question simply due to an interest in cognitive architecture. No taxonomy of the mind will be complete without an account of the nature and functional role of implicit attitudes. If we can fit our understanding of implicit attitudes together with more folk-psychological categories like belief all the better. Much of the interest so far though has been driven by the thought that implicit attitudes undergird implicit biases, implicit biases lead to normatively significant behaviors, and that determining whether or not implicit attitudes are beliefs will help us determine how to alter implicit biases. The basic thought was thought that if implicit

221

Joseph Bendaña

attitudes are beliefs, then they should be amenable to rational interventions, while if they are some other kind of state change would call for different methods of intervention (Madva 2016; Mandelbaum 2016). However, it’s far from clear that determining whether or not implicit attitudes play the functional role of belief is really going to provide much if any purchase on how to change implicit biases. It’s not as though there are clear recipes for belief change of any kind and more significantly the connections between beliefs, implicit attitudes, bias, and behavior remain deeply controversial (Oswald et al. 2013; Brownstein and Saul 2016b; Greenwald and Lai 2020). Though the nature of implicit attitudes remains controversial, if implicit attitudes are in fact beliefs, then it would seem natural to categorize them as implicit beliefs, yet it seems unlikely that they would be implicit in the same sense as implicit beliefs typically invoked to defend representationalism. Recall that the notion of implicitness at play in classical defenses of representationalism was intended primarily to deal with our intuitions around belief attributions. In philosophical debates about whether implicit attitudes are beliefs the notion of implicit at play is more often intended to capture features of how an agent’s implicit attitudes interface or fail to interface with attitudes expressed by verbal reports. However, Fragmentation might explain some of the odder features of the psychological evidence for implicit attitudes, such as why implicit attitudes appear to be poorly integrated with explicit beliefs, but still occasionally responsive to rational interventions (Bendaña 2021; though see Karlan 2021 for arguments that Fragmentation isn’t up to the job). If that’s right, then perhaps Fragmentation can offer a general account of implicit beliefs that works both to help meet challenges to representationalism and to characterize implicit attitudes.

6.  Varieties of Implicitness The notions of implicit belief and associated debates sketched earlier illustrate that there are a wide variety of conceptions of what makes a belief explicit or implicit, and accordingly of explicitness and implicitness. One might take explicitness and implicitness to be a function of representation and lack thereof, such that if an agent has a mental representation that P with the right functional role, they have an explicit belief that P. Whereas if the content that P implies Q, but there is no mental representation that Q, then they have an implicit belief that Q. Dispositionalist, simple consequence, and virtual theories of implicit beliefs all offer this kind of notion of implicitness, but for all this notion’s utility and flexibility accommodating intuitions about belief attributions, it seems to not mesh particularly well with the core of representationalism about belief. Alternatively, one might characterize the explicitness and implicitness of belief in terms of its functional role, such that the more accessible represented information is, the more explicit it is, and conversely, the less accessible it is, the more implicit it is. Representationalists that endorse the view that the format of belief is map-like and those that endorse Fragmentation seem to adopt this kind of notion of implicitness.9 Although this functional notion of explicitness and implicitness seems to mesh well with representationalism, it remains to be seen whether it can actually accommodate the intuitions around belief attributions that threatened representationalism in the first place. Finally, taking a page from psychology, one might simply adopt an operational notion of implicitness, according to which the implicitness of a mental state or process is just a function of the measure – typically measures that are in some sense indirect relative to verbal reports. On this kind of view, a mental state such as a belief or mental process counting as implicit need not have any implication about its nature, but the terminology might nevertheless be helpful for distinguishing between different kinds of measures. If there is a general conclusion to draw from the variety of notions of explicitness and implicitness applied to belief sketched earlier, it seems to be just that they serve different theoretical and 222

Implicit Beliefs

practical purposes. Theorists might want to invoke the implicitness of beliefs to defend a particular theory of belief, to accommodate or explain psychological data about implicit attitudes, or to simply help mark practical distinctions between different kind of psychological measures. However, as detailed throughout this chapter, there remain many open questions about the utility of each of the various conceptions of implicitness sketched earlier and also about whether there is yet some more general construal of implicitness to be applied to belief that has greater theoretical or practical utility than the current options.

Related Topics Chapters: 1–10, 14, 17, 21, 24–27

Notes 1 Throughout this chapter, except where explicitly noted, I  will use the term “implicit beliefs” interchangeably with “tacit beliefs”. Although one might want to draw a distinction between tacit and implicit representations (Dennett 1982) the terms “implicit beliefs” and “tacit beliefs” have been used relatively interchangeably in the literature. Note however, that “implicit beliefs” is also sometimes used as a term of art for all the implications of the content of an agent’s belief state (Levesque 1984). This chapter is not just about implicit beliefs in this narrow sense, but whether such notions can help provide a solution to the problems of logical omniscience is an in interesting question (Stalnaker 1991; Yalcin 2016). 2 If you don’t like talk of attributing beliefs to robots or computers, you can run an analogous case with more humanoid creatures. Just imagine some action, like getting the queen out early, emerging reliably out of information processing over a set of representations that doesn’t include one with the relevant content. 3 It is important to separate the question of what it is for a belief to be implicit, from the question of what it is for a mental representation to be implicit. If an explicit belief is just a mental representation that plays a particular functional role, then it is possible that under some notion of implicit representation (see Dennett 1982 for a classic characterization and Davies 2015 for overview) an explicit belief could be an implicit representation. Conversely, under some notions of the functional role that a state must play to be an explicit belief, an explicit representation could count as an implicit belief. 4 The format of a representation can be thought of roughly as its syntactic or more generally structural properties (Quilty-Dunn 2016; Beck 2015). How to make this notion precise is controversial, but the to get a rough feel for the notion of a map-like mental representation it helps to contrast it with a language-like representation. A sentence has a canonical decomposition (e.g. you can separate it into words), syntactic structure, and nothing about it needs to resemble what it represents (Fodor 1975). A map, on the other hand, arguably has a different structure. It doesn’t have to have discrete compositional pieces and, putting aside keys and legends, it typically represents information via some isomorphism to what it represents. 5 What exactly makes a representation map-like is a matter of some debate (Braddon-Mitchell and Jackson 1996; Camp 2018; Rescorla 2009), but they are typically supposed to be to importantly distinct from language-like representations in at least two ways. They resemble what they represent and they arguably can compactly represent infinite propositions. 6 Note that this characterization of implicit representation diverges from more traditional accounts (e.g., Dennett 1982) according to which implicitly represented information is just information implied by explicitly represented information. As Davies (2015) notes, for any notion of explicit representation it is possible to define a host of notions of implicit representation that differ in resources required to extract the information. Moreover, there is no clearly correct notion of what exactly explicit representation is. So if characterizing explicit and implicit representation in terms of accessibility is useful it’s not clear why one shouldn’t. 7 See also Yalcin (2016) for a more formal extension of a Fragmentation that characterizes implicit beliefs as beliefs that are backgrounded by the resolution of logical space to which an agent’s belief state is dialed into, arguably providing even more flexibility than the naïve form of the framework. 8 Although the most notorious of the results of such measures have been about bias, implicit attitudes go well beyond biases (de Bruijn et  al. 2012; Crisinel and Spence 2010; Maison et  al. 2004). The relationship between implicit attitudes, whatever they are, and notions of bias, though fascinating, is

223

Joseph Bendaña very controversial and thankfully not the topic of this chapter. For overviews see Brownstein and Saul (2016a, 2016b); Brownstein et al. (2019); Johnson (2020). 9 See Kirsh (2003) for a similar view of implicitness applied to mental representations more broadly.

References Audi, R. 1994. “Dispositional beliefs and dispositions to believe”. Noûs, 28: 419–434. Beck, J. 2015. “Analogue magnitude representations: a philosophical introduction”. British Journal for the Philosophy of Science, 66: 829–855. Bendaña, J. 2021. “Implicit attitudes are (probably) beliefs”. In C. Borgoni, D. Kindermann, and A. Onofri, eds., The fragmented mind. Oxford: Oxford University Press: 281–302. Bendaña, J., and Mandelbaum, E. 2021. “The fragmentation of belief ”. In C. Borgoni, D. Kindermann, and A. Onofri, eds., The fragmented mind. Oxford: Oxford University Press: 78–107. Blumson, B. 2012. “Mental maps”. Philosophy and Phenomenological Research, 85: 413–434. Braddon-Mitchell, D., and Jackson, F. 1996. Philosophy of mind and cognition. Oxford: Blackwell. Brownstein, M., Madva, A., and Gawronski, B. 2019. “What do implicit measures measure?”. Wiley Interdisciplinary Reviews: Cognitive Science, 10: e1501. Brownstein, M., and Saul, J., eds. 2016a. Implicit bias and philosophy, vol. 1: Metaphysics and epistemology. Oxford: Oxford University Press. Brownstein, M., and Saul, J., eds. 2016b. Implicit bias and philosophy, vol. 2: Moral responsibility, structural injustice, and ethics. Oxford: Oxford University Press. Buckner, C. 2019. “Understanding associative and cognitive explanations in comparative psychology”. In K. Andrews and J. Beck, eds., The Routledge handbook of philosophy of animal minds. London: Routledge: 409–419. Camp, E. 2007. “Thinking with maps”. Philosophical Perspectives, 21: 145–182. Camp, E. 2018. “Why maps are not propositional”. In A. Grzankowski and M. Montague, eds., Nonpropositional intentionality. Oxford: Oxford University Press: 19–45. Churchland, P. M. 1981. “Eliminative materialism and the propositional attitudes”. The Journal of Philosophy, 78: 67–90. Crimmins, M. 1992. “Tacitness and virtual beliefs”. Mind & Language, 7: 240–263. Crisinel, A. S., and Spence, C. 2010. “A sweet sound? Food names reveal implicit associations between taste and pitch”. Perception, 39: 417–425. Davies, M. 2015. “Knowledge (explicit, implicit and tacit): philosophical aspects”. In J. D. Wright, ed., International encyclopedia of social & behavioral sciences. 2nd edn. Amsterdam: Elsevier: 74–90. de Bruijn, G. J., Keer, M., Conner, M., and Rhodes, R. E. 2012. “Using implicit associations towards fruit consumption to understand fruit consumption behaviour and habit strength relationships”. Journal of Health Psychology, 17: 479–489. De Houwer, J. 2014. “A  propositional model of implicit evaluation”.  Social and Personality Psychology Compass, 8: 342–353. Dennett, D. C. 1982. “Styles of mental representation”. Proceedings of the Aristotelian Society, 83: 213–226. Dennett, D. C. 1989. The intentional stance. Cambridge, MA: MIT Press. Dummett, M. 1991. The logical basis of metaphysics. Cambridge, MA: Harvard University Press. Egan, A. 2008. “Seeing and believing: perception, belief formation and the divided mind”. Philosophical Studies, 140: 47–63. Elga, A., and Rayo, A. 2021. “Fragmentation and information access”. In C. Borgoni, D. Kindermann, and A. Onofri, eds., The fragmented mind. Oxford: Oxford University Press: 37–53. Fazio, R. H., Jackson, J. R., Dunton, B. C., and Williams, C. J. 1995. “Variability in automatic activation as an unobtrusive measure of racial attitudes: a bona fide pipeline?”. Journal of Personality and Social Psychology, 69: 1013–1027. Fazio, R. H., Sanbonmatsu, D. M., Powell, M. C., and Kardes, F. R. 1986. “On the automatic activation of attitudes”. Journal of Personality and Social Psychology, 50: 229–238. Field, H. 1978. “Mental representation”. Erkenntnis, 13: 9–61. Fodor, J. A. 1975. The language of thought. Cambridge, MA: Harvard University Press. Fodor, J. A. 1987. Psychosemantics: the problem of meaning in the philosophy of mind. Cambridge, MA: MIT Press. Gawronski, B., and Bodenhausen, G. V. 2006. “Associative and propositional processes in evaluation: an integrative review of implicit and explicit attitude change”. Psychological Bulletin, 132: 692–731. Gawronski, B., De Houwer, J., and Sherman, J. W. 2020. “Twenty-five years of research using implicit measures”. Social Cognition, 38(Supplement): s1–s25.

224

Implicit Beliefs Gendler, T. S. 2008. “Alief and belief ”. The Journal of Philosophy, 105: 634–663. Greenwald, A. G., and Banaji, M. R. 1995. “Implicit social cognition: attitudes, self-esteem, and stereotypes”. Psychological Review, 102: 4–27. Greenwald, A. G., Brendl, M., Cai, H., Cvencek, D., Dovidio, J. F., Friese, M., Hahn, A., Hehman, E., Hofmann, W., Hughes, S., and Hussey, I. 2020. “The Implicit Association Test at age 20: what is known and what is not known about implicit bias”. https://doi.org/10.31234/osf.io/bf97c Greenwald, A. G., and Lai, C. K. 2020. “Implicit social cognition”.  Annual Review of Psychology,  71: 419–445. Greenwald, A. G., McGhee, D. E., and Schwartz, J. L. 1998. “Measuring individual differences in implicit cognition: the implicit association test”. Journal of Personality and Social Psychology, 74: 1464–1480. Johnson, G. M. 2020. “The structure of bias”. Mind, 129: 1193–1236. Johnson, K. 2015. “Maps, languages, and manguages: rival cognitive architectures?”. Philosophical Psychology, 28: 815–836. Karlan, B. 2021. “The rational dynamics of implicit thought”. Australasian Journal of Philosophy. https:// doi.org/10.1080/00048402.2021.1936581 Kirsh, D. 2003. “Implicit and explicit representation”. In L. Nadel, ed., Encyclopedia of cognitive science. Vol. 2. London: Nature Publishing Group: 478–481. Kurdi, B., and Dunham, Y. 2020. “Propositional accounts of implicit evaluation: taking stock and looking ahead”. Social Cognition, 38(Supplement): s42–s67. Lewis, D. 1982. “Logic for equivocators”. Noûs, 16: 431–441. Levesque, H. J. 1984. “A logic of implicit and explicit belief ”. In Proceedings of the fourth national conference on artificial intelligence (AAAI-84). Austin, TX: 198–202. Levy, N. 2015. “Neither fish nor fowl: implicit attitudes as patchy endorsements”. Noûs, 49: 800–823. Lycan, W. G. 1981. “Toward a homuncular theory of believing”. Cognition and Brain Theory, 4: 139–159. Lycan, W. G. 1986. “Tacit belief ”. In R. J. Bogdan, ed., Belief: form, content, and function. New York: Oxford University Press: 61–82. Lycan, W. G. 1988. Judgement and justification. Cambridge: Cambridge University Press. Madva, A. 2016. “Why implicit attitudes are (probably) not beliefs”. Synthese, 193: 2659–2684. Maison, D., Greenwald, A. G., and Bruin, R. H. 2004. “Predictive validity of the Implicit Association Test in studies of brands, consumer attitudes, and behavior”. Journal of Consumer Psychology, 14: 405–415. Mandelbaum, E. 2016. “Attitude, inference, association: on the propositional structure of implicit bias”. Noûs, 50: 629–658. Mandelbaum, E. 2020. “Associationist theories of thought”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Fall 2020 edn. https://plato.stanford.edu/archives/fall2020/entries/associationist-thought/. McConnell, A. R., and Rydell, R. J. 2014. “The systems of evaluation model”. In J. Sherman, B. Gawronski, and Y. Trope, eds., Dual-process theories of the social mind. New York: Guilford: 204–217. Oswald, F. L., Mitchell, G., Blanton, H., Jaccard, J., and Tetlock, P. E. 2013. “Predicting ethnic and racial discrimination: a meta-analysis of IAT criterion studies”. Journal of Personality and Social Psychology, 105: 171–192. Porot, N., and Mandelbaum, E. 2021. “The science of belief: a progress report”. Wiley Interdisciplinary Reviews: Cognitive Science, 12: e1539. Quilty-Dunn, J. 2016. “Iconicity and the format of perception”. Journal of Consciousness Studies, 23: 255–263. Quilty-Dunn, J., and Mandelbaum, E. 2018. “Against dispositionalism: belief in cognitive science”. Philosophical Studies, 175: 2353–2372. Rescorla, M. 2009. “Cognitive maps and the language of thought”. The British Journal for the Philosophy of Science, 60: 377–407. Schwitzgebel, E. 2021a. “The pragmatic metaphysics of belief ”. In C. Borgoni, D. Kindermann, and A. Onofri, eds., The fragmented mind. Oxford: Oxford University Press: 350–376. Schwitzgebel, E. 2021b. “Belief ”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Winter 2021 edn. https://plato.stanford.edu/archives/win2021/entries/belief/. Sherman, J. W., and Klein, S. A. 2021. “The four deadly sins of implicit attitude research”. Frontiers in Psychology, 10.3389/fpsyg.2020.604340 Stalnaker, R. 1984. Inquiry. Cambridge, MA: MIT Press. Stalnaker, R. 1991. “The problem of logical omniscience, I”. Synthese, 89: 425–440. Yalcin, S. 2016. “Belief as question-sensitive”. Philosophy and Phenomenological Research, 97: 23–47.

225

17 IMPLICIT SELF-KNOWLEDGE Kristina Musholt

This chapter aims to give an overview of different types of knowledge that can reasonably be considered forms of implicit self-knowledge in contrast to explicit self-knowledge. In order to do so, it is important to first clarify the notion of self-knowledge. Second, we will need to define the difference between implicit and explicit knowledge. This will allow us to come to an understanding of what implicit self-knowledge might be. In the philosophical literature on self-knowledge, much of the focus is on knowledge about oneself that is paradigmatically expressed by means of the first-person pronoun and that possesses certain epistemic features, in particular the feature of being “immune to error through misidentification relative to the first-person pronouns” (Shoemaker 1968: 556). The interest in this kind of self-knowledge can be traced back to its specific epistemic status and to the fact that it is often thought to be based on a kind of privileged access to one’s own states (Gertler 2020). Importantly, not all self-knowledge is of this kind, and some have argued that it is, indeed, rather trivial compared to the more substantial self-knowledge that can only be reached by less privileged and more laborious routes of investigation (Cassam 2014). Moreover, some authors deny that self-knowledge possesses an epistemically privileged status (e.g. Carruthers 2011). We will return to the issues of substantial self-knowledge and self-deception toward the end, but to begin with this entry will focus on accounts that discuss self-knowledge in the epistemically privileged sense.

Self-Knowledge and Immunity to Error Through Misidentification Self-consciousness can be defined as the ability to think ‘I’-thoughts, that is, thoughts that non-accidentally refer to the subject of the thought. Such thoughts are meant to rely on ways of gaining information about oneself that are privileged in the sense that they are “immune to error through misidentification relative to the first-person pronouns” (Shoemaker 1968; also see Wittgenstein 1958). Accordingly, while the subject of such a thought might be wrong with regard to the predicate they are ascribing to themselves, they cannot be mistaken with respect to the subject (i.e. themselves) they are referring to. For instance, if I experience a sharp pang of pain in my stomach and on this basis form the thought ‘I am hungry’ I cannot be mistaken with regard to the question of who is hungry. Thus, the self-knowledge that is expressed in such thoughts is said to possess a special epistemic status. Much thought has been put into the 226

DOI: 10.4324/9781003014584-21

Implicit Self-Knowledge

question as to how best to explain this epistemic status, and some theories have appealed to implicit elements in self-conscious experience in order to explain the phenomenon of immunity to error through misidentification (IEM). Moreover, it has been argued that although not all self-knowledge is of this specific kind, it provides the basis for the ability to gain any selfknowledge at all.

Nonconceptual Self-Awareness and the Essential Implicitness of Self-Related Information Some authors have argued that the kind of self-knowledge just described is based on nonconceptual forms of self-representation, such as perception and proprioception (e.g. Bermúdez 1998; Hurley 1998; Peacocke 2014). For instance, Bermúdez (1998) argues that while perception provides the subject ostensibly with information about their environment, is also always necessarily contains information about the self, such as the subject’s distance and orientation from the objects it perceives, or the affordances provided by the objects in question. Thus, according to this view, the content of perception always contains a de se component. However, as we share our perceptual capacities with nonconceptual creatures such as animals, this is thought to be at the level of nonconceptual rather than conceptual content. Insofar as the notion of nonconceptual content is often related to the notion of implicit knowledge, another way of expressing this is by saying that certain forms of experience, such as the experience involved in perceiving one’s environment, constitute forms of implicit self-knowledge in virtue of possessing nonconceptual de se content. While the theory of nonconceptual self-consciousness claims that the self is part of the representational content of experience, others have tried to provide impersonal accounts of the content of experience. For instance, Musholt (2015) argues that perception and proprioception do not represent the self but rather contain implicitly self-related information. For example, although the information that is gained through visual experience is necessarily self-related or self-concerning, because visual experience always provides information from the perspective of the perceiving subject, this fact itself need not be represented. Rather, it can be treated as an ‘unarticulated constituent’ of the experience in question (Perry 1986), meaning that it forms part of the context for determining the truth-conditions of judgments regarding the subject’s spatial location. Thus, the explicit content of the visual experience of, for example, seeing a tree should be specified as ‘tree in front’ rather than ‘tree in front of me’. As Campbell (1994) puts it, the egocentric frame of vision operates with ‘monadic’ rather than relational notions. However, the implicit self-relatedness of the information in question can be made explicit – provided that the subject possesses the relevant concepts to do so – when the subject forms an ‘I’-thought (such as ‘I see a tree in front of me’) on the basis of their experience.

Implicit Versus Explicit Knowledge This raises the question as to what it means for a piece of information to be represented implicitly rather than explicitly. Davies (2001) defines implicit knowledge simply in opposition to explicit knowledge, where the latter is defined as something that can be verbally stated by a subject upon suitable enquiry or prompting (Dummett 1991). Thus, in this view, implicit knowledge is knowledge that cannot be verbally stated by the knower, mirroring Polanyi’s (1967) notion of ‘tacit knowledge’. In contrast to this, on the view proposed here, implicit knowledge can sometimes be made explicit by the subject. The proposal presented here draws on Dienes and Perner’s claim that a “fact is explicitly represented if there is an expression 227

Kristina Musholt

(mental or otherwise) whose meaning is just that fact; in other words, there is an internal state whose function is to indicate that fact” (Dienes and Perner 1999: 736). In this view, a fact or a piece of information is represented explicitly when the representational state in question contains a component (an expression) that directly stands for this piece of information. In contrast, a piece of information is represented implicitly when the representational state in question does not contain a component that directly refers to the information, but when the information is still conveyed (and thus can be inferred), for instance as part of the context of the representational state. For example, the functional role of bodily experience is to convey information about the bodily states of the subject undergoing the experience. However, the essential self-relatedness of the content of bodily experience is not explicitly represented; rather it is left implicit (i.e., unarticulated), as part of the mode or functional context in which bodily experience operates (which is determined by facts about the cognitive architecture of the subject). When I have a bodily experience of legs being crossed, it is necessarily my legs that I experience to be crossed. This is simply how bodily awareness works – it delivers information about the experiencing subject’s limbs and bodily states and nobody else’s. Yet this fact need not itself be explicitly represented precisely because the function of bodily awareness is such that it necessarily delivers information about one’s own body. Rather, the content of the bodily experience consists in the property alone, without containing a component that refers to the subject-role (Recanati 2007, 2012). Coming from the perspective of cognitive science, one can also ask how information has to be encoded in order to qualify as implicit or explicit. Karmiloff-Smith (1996) proposes that information that is encoded in procedural form is implicit in the sense of being unavailable to other operations within the cognitive system. She suggests that in a subsequent reiterative process of representational redescription, this information can be transformed into increasingly abstract and less specialized, but more cognitively flexible formats of representation. These latter representations can then be used for cognitive operations that require explicit knowledge. In particular, Karmiloff-Smith distinguishes between three different levels of explicitness: at the first level (E1), information is available as data to the system, in the form of ‘theories-inaction’, although not necessarily to conscious access and verbal report. At the second level (E2), information becomes available to conscious access, but not yet to verbal report. And finally, at the third level (E3), information is re-represented into a cross-system code, which allows it to become verbally expressed (Karmiloff-Smith 1996: 23). Thus, the process of representational redescription involves the recoding of information that is stored in one representational format into another, such that each redescription is a more condensed version of the previous level. As a result, the information becomes increasingly explicit and can be used increasingly flexibly. The advantage of explicit representations consists in the fact that they allow for more cognitive flexibility and domain-generality. In contrast, implicit or procedural representations enable fluid and automatic interactions with the environment, but remain domain-specific. Notice that this model implies that the representational system is much more complex than the dichotomy between nonconceptual and conceptual content would suggest. It also suggests that the implicitly self-related information that is part of the content of perception or proprioception is coded in procedural form, meaning that it is not available to other parts of the cognitive system. That is, it cannot be directly expressed in thought or language, rather it is present in terms of guiding the subject’s interactions with the world. Accordingly, we might think of implicit self-knowledge in this sense as a kind of knowledge of affordances, or of knowledge-how; a point we will return to later. However, it can be made available for explicit representation through a gradual process of representational redescription (see Musholt 2015: Ch. 5 for a detailed discussion). 228

Implicit Self-Knowledge

Immunity to Error Through Misidentification Reconsidered Now, returning to the issue of immunity to error through misidentification, we can see how explicit first-personal judgments that are grounded in impersonal experiential states can be IEM. For a first-person judgment that is based on perceptual or bodily experience simply exploits the fact that the relevant mode of experience is self-specific by making this fact – which is implicit in the experience – explicit. Importantly, such a judgment does not rely on any sort of identity judgment – it is identification-free in Evans’ (1982) sense. Insofar as such a judgment does not rely on a representation of the self, but rather simply makes explicit what is already implicitly contained in the mode of presentation – or in the affordances that are present to the subject –, it cannot possibly misrepresent the self (Recanati 2012). That is, because such judgments do not rely on self-identifying information, they don’t leave any room for error based on misidentification. In contrast, on a view in which the self is explicitly represented in experience, it is harder to explain why judgments that are made on the basis of experience should be IEM. After all, if the self is represented explicitly in experience, there should, at least in principle, be the possibility of misrepresentation. Interestingly, the claim that there are forms of nonconceptual – implicit – self-knowledge has parallels in phenomenological accounts. According to some phenomenological analyses, any state of consciousness is necessarily also a state of self-consciousness. This is because conscious experience is always experience from a particular perspective – it is always experience for a subject. The fact that experience is always perspectival in this sense provides it with a feeling of ‘mineness’ or ‘ipseity’ (Zahavi 2005, 2014). Importantly, though, this does not imply that pre-reflective experience contains a de se component. As Sartre states “there is no place for me at this level” (1937: 13). Rather, the self appears to be a feature of the structure of experience. That is, the feeling of mineness is not to be understood in terms of self-representation (Zahavi and Kriegel 2017). Accordingly, such a phenomenological reading is quite compatible with the impersonal view proposed earlier, namely that it is the implicit self-relatedness of our modes of conscious experience, rather than the de se content of experience, that can explain both the subjectivity of experience and the fact that experience can ground first-personal judgments that are IEM. (Also see Boyle (forthcoming) for an explication of a Sartrean conception of self-awareness in analytical terms that is compatible with this view.) It is noteworthy that the importance of the distinction between pre-reflective and reflective self-consciousness, and hence between explicit self-representation and implicitly self-specifying experiences, has also recently been emphasized with respect to cognitive neuroscience and psychiatry (e.g. Christoff et al. 2011; Sass et al. 2011; Musholt 2013).

From Implicit to Explicit Self-Knowledge However, the difficulty that is raised by theories of implicit self-knowledge lies in explaining how the subject moves from having experiential states with implicitly self-related information to forming explicit first-personal judgments. There are two ways of reading this question: On the one hand, one can ask what entitles the subject to form such judgments. With respect to this question one might appeal to Peacocke’s (1999) notion of ‘representationally free’ uses of the first-person pronoun. According to Peacocke, ‘I’-thoughts (i.e., explicit self-ascriptions) that are made on the basis of conscious states or activities are ‘representationally free’ uses of ‘I’, which renders these self-ascriptions immune to error through misidentification. Such selfascriptions are characterized by the fact that “the thinker’s reason for self-applying the predicate F is not that one of his conscious states has the content ‘I am F’. His reason is not given by the 229

Kristina Musholt

‘representation distinguishing a particular object’ as F. It is rather the occurrence of a certain kind of conscious state itself which is his reason for making the judgment” (Peacocke 1999: 285, emphasis original). According to Peacocke, it is because a conscious state necessarily requires a subject, and because a subject can only ever experience their own mental state and not that of another, that a thinker who possesses the first-person concept is entitled to self-ascribe it. Possession of the self-concept then expresses rational sensitivity to this fact of ownership. On the other hand, one can ask how a subject acquires the first-person concept. That is, one can ask how subjects acquire the rational sensitivity just mentioned. More broadly, one can ask how concept acquisition in general is to be explained. It seems natural to suggest that language acquisition plays an important role in this. However, there is much debate regarding the formation of rationality via language acquisition. For one thing, there is a dispute between proponents of so-called transformative views, who claim that humans acquire rationality and self-knowledge through the socially mediated process of being initiated into language (e.g. McDowell 1994; Bakhurst 2011), and those who argue that humans must always already be rational and self-conscious creatures to begin with and thus cannot be transformed through education (e.g. Rödl 2016). Relatedly, there is a debate regarding the process of language acquisition. On some views, which are sometimes associated with Wittgenstein, language – and hence rationality – is acquired through a process of drill or training. However, many have objected that a process of training is inadequate to explain the transition from the non- or prerational to the rational (e.g. Bakhurst 2011). Put differently, many philosophers hold that one cannot reduce facts about rationality, meaning, or normativity to natural facts. Nonetheless, in order to do justice to the developmental dimension of the acquisition of language and rationality, one might try to explain the emergence of rationality by reference to something more primitive. Along these lines, Ginsborg (2011) has proposed that in order to explain the development of language, as well as other rule-governed activities, in children, we have to assume that the child who is being initiated into such activities already possesses a kind of ‘primitive normativity’. By this she means that a child might have a sense of it being appropriate that, for example, differently colored blocks are sorted in a specific way, or that ‘10’ is the proper followup to the sequence ‘2,4,6,8’, without this requiring the child’s understanding of an antecedently applicable rule. Now, insofar as the ability to let one’s activities be guided by one’s sensitivity to norms – albeit in a primitive (implicit) sense – implies, on the Kantian view advocated by Ginsborg, a form of self-knowledge (i.e. an awareness that what one is doing is appropriate), this provides us with another argument for the claim that there are forms of preconceptual, or implicit, self-knowledge. Moreover, there are also researchers who claim that explicit self-knowledge need not be tied to the linguistic use of the first-person pronoun, and hence to language acquisition at all. For instance, some researchers claim that the ability to recognize oneself in the mirror displayed by some nonhuman animals is indicative of their ability to think about themselves, even in the absence of linguistic abilities (Anderson and Gallup 1999). However, there is much debate regarding the question as to what cognitive abilities really underwrite the ability to recognize oneself in the mirror and in particular regarding the question as to whether this ability warrants the ascription of psychological self-knowledge (e.g. Boyle 2018). Further, some animals have been shown to display so-called metacognitive abilities, that is, they seem to know what they do and do not know. For example, monkeys were shown to be able to master a task in which they were presented with a stimulus that they would later, after a delay, have to recognize among an array of other items (Hampton 2001). Crucially, the monkeys had to decide whether to take the test before being presented with the test stimuli. Alternatively, they could opt for an easier task. As Hampton (2005) argues, in order to succeed in this task, monkeys must rely on a prospective 230

Implicit Self-Knowledge

judgment of their memory, that is on a ‘feeling of knowing’ (or feeling of uncertainty), and thus on a meta-representational mental state. However, others have argued that the results obtained in such studies can alternatively be explained by first-order rather than higher-order cognition (Carruthers 2008). Moreover, it has been argued that even if the behavior in question constitutes a case of metacognition, in the sense of the monkeys’ sense of (un-)certainty referring to an underlying mental state, this need not be interpreted in terms of the animal being aware of their feeling as giving them information about another mental state of theirs. That is, metacognition is not necessarily equal to explicit meta-representation and hence explicit self-knowledge (Proust 2006). Finally, some authors argue that explicit self-representation is primarily a matter of developing a theory of mind (e.g. Carruthers 2011; Baker 2013; Musholt 2015). For instance, Musholt (2015) argues that it is only when a subject comes to realize that there exist other subjects with different perspectives on the world that they gain awareness of their own perspective as their own. Accordingly, in this view, the transition from implicitly self-related information to explicit self-representation involves a gradual process of self-other-differentiation. Thus, in this view self-knowledge and knowledge of others go hand in hand.

Implicit Self-Knowledge and Knowledge-How There is a legitimate question as to whether something that cannot be verbally articulated by the subject, or that isn’t reflectively available to them, can constitute knowledge. One possible answer to this objection is that although implicit self-knowledge is not currently reflectively available, it can in principle be made explicit, provided the subject possesses the relevant concepts, in particular the first-person concept. However, this implies that we should strictly speaking not ascribe implicit self-knowledge to beings who do not possess the relevant conceptual abilities to make this knowledge explicit, such as animals and infants. Some theorists might welcome this implication, while others might consider this to be a problem. A second way of answering this question, which was already hinted at earlier, is by characterizing implicit self-knowledge in terms of knowledge-how, as opposed to (explicit) knowledgethat. That is, we can think of implicit information as being stored in the cognitive system as procedural knowledge that provides the organism with knowledge-how to interact with the world. There are many questions regarding the notion of knowledge-how and its relation to knowledge-that, as well as to notions such as practical knowledge, competence, procedural knowledge, nonconceptual content, automatic vs voluntary control, or skill. For the purposes of this entry, I will simply assume that knowledge-how denotes a distinct type of intelligent, intentional interaction with the world in Ryle’s (1949) sense. Further, on the view proposed here, we can think of the content of experience as being determined precisely by the subject’s implicit knowledge of how to pursue and accomplish their goals and intentions with regard to the objects in their environment (Ward et al. 2011). This knowledge is implicitly self-specifying in the sense that the affordances available to the subject will depend on their specific abilities, their position in relation to the objects surrounding them, their current bodily and mental states, and so on. That is to say, knowledge-how is always knowledge for the specific subject and refers to their specific agentive capacities. In line with our preceding discussion, O’Brien (2007) argues that it is precisely such agent’s awareness that can ground our ability to explicitly self-refer in thought. Insofar as an agent is acting on the basis of assessing possibilities for action, these are necessarily possibilities for them. According to O’Brien, this is because such actions are under the agent’s control – they are something the agent must actively produce. Accordingly, they immediately warrant self-ascription, without the need for observation, reflection, or self-representation. 231

Kristina Musholt

Self-Knowledge and Implicit Bias Finally, as promised in the introduction, let us return to the issue of substantial self-knowledge. As indicated, Cassam (2014) argues that substantial self-knowledge – in contrast to the rather trivial self-knowledge as to whether I am currently experiencing a tree or whether my legs are crossed – is much harder to come by. Substantial self-knowledge refers to knowledge regarding my character, my non-trivial beliefs, my deepest desires or my true ambitions. And the claim is that these are often hidden to ourselves. For instance, I might be convinced that I am someone who values and respects men and women equally. Yet in my daily interactions with male and female colleagues it is apparent to others, if not myself, that I  systematically treat men with more respect. Thus, arguably, I do not truly value men and women equally. At the very least my behavior portrays an implicit bias against women, though there is some debate as to whether this bias qualifies as genuine belief or some other kind of mental state. If it does qualify as a belief, then my explicitly self-ascribed conviction in this case – that is, my explicit knowledge of what I believe – seems to come apart from what I appear to believe implicitly. Hence, we can think of my implicit beliefs as something that cannot be directly expressed, but that can be inferred – both by others and by myself – via an observation of my behavior. If so, this might constitute another form of implicit self-knowledge, albeit one that often stands in tension with my explicit self-ascriptions. This raises important questions as to how such tensions should be resolved and which of my beliefs constitute my true, authentic self. On one view, in cases such as the one just described, I simply suffer from self-deception regarding my true beliefs and values, and thus I suffer from a lack self-knowledge regarding my true self (Cassam 2014; Schwitzgebel 2012; also see Carruthers 2011 for a discussion of various ways in which I  might be mistaken about my mental states). I might be convinced that I am not a sexist, but really I am. On another view, it is only those beliefs and values that I reflectively endorse or that I identify with after critical reflection that form part of my authentic, autonomous self (e.g. Frankfurt 1971; Watson 1975). According to this latter view, in order to figure out what I truly believe, I need to critically reflect on my values, desires and beliefs so as to decide which of them I can identify with. However, this raises the problem of socialization regress, as the norms and values I endorse will inevitably be shaped by my social upbringing (Friedman 1986; also see Haslanger 2019). For instance, an agent who is socialized in an oppressive environment might explicitly endorse and identify with oppressive beliefs, even upon reflection, because her skills for critical reflection are themselves compromised by her socialization. Indeed, she might even develop a strong resistance to critical perspectives that threaten to call into question certain aspects of her sense of self (Benson 1991; Mackenzie 2002). As a result, in such cases it might even be the kinds of motivations, desires, behaviors, and emotional reactions that the agent would disown upon reflection – what we might call implicit self-knowledge – that provide a better indication of her true, authentic self (Mackenzie 2002). So here, again, the crucial question with respect to the relation between implicit and explicit self-knowledge raises its head, albeit in a different form. If our skills of critical reflection are just as much the result of socialization as the oppressive norms and stereotypes we are prone to internalize, how can we ensure that critical reflection brings to the surface real knowledge about our true self? Put differently, what must a kind of socialization look like that will enable the agent to come to an explicit knowledge of their authentic self? A promising direction to take in searching for an answer to this question is outlined by McGeer’s exploration of the relation between the regulative nature of folk psychology, self-knowledge, responsibility and social (in-)justice (e.g. McGeer 2015, 2019). 232

Implicit Self-Knowledge

Related topics Chapters 1, 4, 8, 16, 23, 24, 25, 26, 31

References Anderson, J. R., and Gallup, G. G., Jr. 1999. “Self-recognition in nonhuman primates: past and future challenges”. In M. Haug and R. E. Whalen, eds., Animal models of human emotion and cognition. Washington, DC: American Psychological Association: 175–194. Baker, L. R. 2013. Naturalism and the first-person perspective. Oxford: Oxford University Press Bakhurst, D. 2011. The formation of reason. Hoboken, NJ: John Wiley & Sons. Benson, P. 1991. “Autonomy and oppressive socialization”. Social Theory and Practice, 17: 385–408. Bermúdez, J.-L. 1998. The paradox of self-consciousness. Cambridge: MA: MIT Press. Boyle, A. 2018. “Mirror self-recognition & self-identification”. Philosophy & Phenomenological Research, 97: 284–303. Boyle, M. forthcoming. Transparency and reflection: a study of self-knowledge and the nature of mind. New York: Oxford University Press. Campbell, J. 1994. Past, space and self. Cambridge, MA: MIT Press. Carruthers, P. 2008. “Meta-cognition in animals: a skeptical look”. Mind & Language, 23: 58–89. Carruthers, P. 2011. The opacity of mind. Oxford: Oxford University Press. Cassam, Q. 2014. Self-knowledge for humans. Oxford: Oxford University Press Christoff, K., Cosmelli, D., Legrand, D., and Thompson, E. 2011. “Specifying the self for cognitive neuroscience”. Trends in Cognitive Sciences, 15: 104–112. Davies, M. 2001. “Explicit and implicit knowledge: philosophical aspects”. In N. J. Smelser and P. B. Baltes, eds., International Encyclopedia of the Social and Behavioral Sciences. Amsterdam: Elsevier Science: 8126–8132. Dienes, Z., and Perner, J. 1999. “A theory of implicit and explicit knowledge”. Behavioral and Brain Sciences, 22: 735–808. Dummett, M. 1991. The logical basis of metaphysics. Cambridge, MA: Harvard University Press. Evans, G. 1982. The varieties of reference. Oxford: Oxford University Press. Frankfurt, H. 1971. “Freedom of the will and the concept of a person”. Journal of Philosophy, 68: 5–20. Friedman, M. 1986. “Autonomy and the split-level self ”. Southern Journal of Philosophy, 24: 19–35. Gertler, B. 2020. “Self-knowledge”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Spring 2020 edn. https://plato.stanford.edu/archives/spr2020/entries/self-knowledge/. Ginsborg, H. 2011. “Primitive normativity and skepticism about rules”. The Journal of Philosophy, 108: 227–254. Hampton, R. 2001. “Rhesus monkeys know when they remember”. Proceedings of the National Academy of Sciences, 98: 5359–5362. Hampton, R. 2005. “Can Rhesus monkeys discriminate between remembering and forgetting?”. In H. Terrace and J. Metcalfe, eds., The missing link in cognition: origins of self-reflective consciousness. Oxford: Oxford University Press. Haslanger, S. 2019. “Cognition as a social skill”. Australasian Philosophical Review, 3: 5–25. Hurley, S. 1998. “Nonconceptual self-consciousness and agency: perspective and access”. Communication and Cognition: An Interdisciplinary Quarterly Journal, 30: 207–247. Karmiloff-Smith, A. 1996. Beyond modularity. Cambridge, MA: MIT Press. Mackenzie, C. 2002. “Critical reflection, self-knowledge, and the emotions”. Philosophical Explorations, 5: 186–206 McDowell, J. 1994. Mind and world. Cambridge, MA: Harvard University Press. McGeer, V. 2015. “Mind-making practices: the social infrastructure of self-knowing agency and responsibility”. Philosophical Explorations, 18: 259–281. McGeer, V. 2019. “Mindshaping is inescapable, social injustice is not: reflections on Haslanger’s critical social theory”. Australasian Philosophical Review, 3: 48–59 Musholt, K. 2013. “A philosophical perspective on the relation between cortical midline structures and the self ”. Frontiers in Human Neuroscience, 7: 536. Musholt, K. 2015. Thinking about oneself: from nonconceptual content to the concept of a self. Cambridge, MA: MIT Press. O’Brien, L. 2007. Self-knowing agents. Oxford: Oxford University Press.

233

Kristina Musholt Peacocke, C. 1999. Being known. Oxford: Oxford University Press. Peacocke, C. 2014. The mirror of the world: subjects, consciousness, & self-consciousness. Oxford: Oxford University Press. Perry, J. 1986. “Thought without representation”. Proceedings of the Aristotelian Society, Supplementary Volumes, 60: 137–166 Polanyi, M. 1967. The tacit dimension. London: Routledge and Kegan Paul. Proust, J. 2006. “Rationality and meta-cognition in non-human animals”. In S. Hurley and M. Nudds, eds., Rational animals? Oxford: Oxford University Press. Recanati, F. 2007. Perspectival thought: a plea for (moderate) relativism. Oxford: Oxford University Press. Recanati, F. 2012. “Immunity to error through misidentification: what it is and where it comes from”. In S. Prosser and F. Recanati, eds., Immunity to error through misidentification. Cambridge: Cambridge University Press. Rödl, S. 2016. “Education and autonomy”. Journal of Philosophy of Education, 50: 84–97. Ryle, G. 1949. The concept of mind. London: Hutchinson. Sartre, J. P. 1937. The transcendence of the ego. Trans. A. Brown. 2004. London: Routledge. Sass, L. A., Parnas, J., and Zahavi, D. 2011. “Phenomenological psychopathology and schizophrenia: contemporary approaches and misunderstandings”. Philosophy, Psychiatry, Psychology, 18: 1–23. Schwitzgebel, E. 2012. “Self-ignorance”. In J. L. Liu and J. Perry, eds., Consciousness and the self: new essays. Cambridge: Cambridge University Press: 184–197. Shoemaker, S. 1968. “Self-reference and self-awareness”. The Journal of Philosophy, 65: 555–567. Ward, D., Roberts, T., and Clark, A. 2011. “Knowing what we can do: actions, intentions, and the construction of phenomenal experience”. Synthese, 181: 375–394. Watson, G. 1975. “Free agency”. Journal of Philosophy, 72: 205–220. Wittgenstein, L. 1958. The blue and brown books. Oxford: Blackwell. Zahavi, D. 2005. Subjectivity and selfhood: investigating the first-person perspective. Cambridge, MA: MIT Press. Zahavi, D. 2014. Self and other: exploring subjectivity, empathy, and shame. Oxford: Oxford University Press. Zahavi, D., and Kriegel, U. 2017. “For-me-ness: what it is and what it is not”. In D. O. Dahlstrom, A. Elpidorou, and W. Hopp, eds., Philosophy of mind and phenomenology. London: Routledge: 48–66.

234

PART 4

Language

18 CHOMSKY, COGNIZING, AND TACIT KNOWLEDGE John Collins

1. Introduction The chapter presents a historical survey of Noam Chomsky’s conception of a speaker-hearer’s cognitive relation to language. In broadest terms, the topic is the subject having knowledge of, or, to use Chomsky’s neologism, a cognizing relation to, their language. It will be shown that the operative epistemology has always been somewhat weak and unconstrained by ordinary or more philosophically sophisticated conceptions of knowledge. This becomes clear in Chomsky’s in more recent considerations, but the history is important, for it brings into relief how the epistemic terminology of ‘knowledge of language’ has often been used to demarcate a domain of inquiry rather than specify a definite underlying architecture for linguistic cognition.

2.  What Generative Linguistics Is About For Chomsky (1957: 15), the goal of linguistic theory is to ‘offer[. . .] an explanation for this fundamental aspect of linguistic behaviour [i.e., the cognitive projection from finite input to unbounded competence]’ and the explanatory burden of the theory falls squarely in that province. The larger and older, albeit unpublished at the time, The Logical Structure of Linguistic Theory (1955–56/75) takes linguistic theory to be an account of ‘a large store of knowledge . . . and a mass of feelings and understandings’ that ‘develops’ within each speaker-hearer and the projection from a finite exposure to unbounded competence (essentially, acquisition) (ibid.: 62–63). A grammar, then, is intended to tell us what a competent speaker-hearer knows in knowing a language, and the grammar must be such as to meet whatever conditions hold for grammars in general such that each is a member of a class of grammars that depict the knowledge competent speaker-hearers may acquire. This conception of the goals of a theory of language was quite radical at the time of the behaviourist hegemony of the 1950s. According to the proposal, language is not to be thought of as a set of verbal dispositions to respond to stimuli, but as a body of knowledge, which, qua acquirable, presupposes an account of the child not as a perfectly general unformed system prior to a regime of ‘training’, but as a system capable of representing and processing linguistic information in an abstract fashion detached from behaviour. Here is

DOI: 10.4324/9781003014584-23 237

John Collins

Chomsky on the matter in the famous review of B.F. Skinner, which contributed to the demise of behaviourism as a general theory of cognition: The child who learns a language has in some sense constructed the grammar for himself on the basis of observation of sentences and non-sentences (i.e. corrections by the verbal community). Study of the actual observed ability of a speaker . . . apparently forces us to the conclusion . . . that the young child has succeeded in carrying out . . . a remarkable type of theory construction. .  .  . The fact [of remarkable rapidity] suggests that human beings are somehow specially designed to do this, with datahandling or ‘hypothesis-formulating’ ability of unknown character and complexity. [. . .] In principle it may be possible to study the problem of determining what the built-in structure of an information-processing (hypothesis forming) system must be to enable it to arrive at the grammar of a language from the available data in the available time. (Chomsky 1959: 577–578) The suggestion here is, in effect, to turn the methodological and formal innovation of the earlier work into a full-blown psychological hypothesis. The grammar a theory specifies is what a child acquires or comes to know, and they are able to acquire such knowledge because the knowledge meets certain conditions that the child does not learn at all, but has encoded in the form of a ‘language acquisition device’ or universal grammar. The theory of the theorist thus becomes a hypothesis concerning the ‘theory’ the child acquires. Thus: What [the child] accomplishes can fairly be described as theory construction of a nontrivial kind . . . a theory that predicts the grammatical structure of each of an infinite class of potential physical events. (Chomsky 1962: 528) The view is further specified in the opening methodological chapter of Aspects: It seems plain that language acquisition is based on the child’s discovery of what from a formal point of view is a deep and abstract theory – a generative grammar for his language – many of the concepts and principles of which are only remotely related to experience by long and intricate chains of unconscious quasi-inferential steps. (Chomsky 1965: 58) Chomsky’s model of acquisition mirrors, in effect, Hempel’s (1966) hypothetico-deductive model of explanation, where hypotheses are confirmed by their observational consequences, and compared to alternative hypotheses. It is important to note, though, that Chomsky is not here claiming that the child is as an evidence-hungry creative scientist, who consciously constructs a theory to cover the observed phenomena. To the contrary, the actual acquisition process is characterised by a ‘poverty of stimulus’, with the child being able to acquire a language based on partial and highly misleading data relative to what is acquired, hence the need for ‘intricate chains of unconscious quasi-inferential steps’, whose character is not determined by experience. The puzzle this poses is resolved by the hypothesis that the acquisition process is mostly shaped and enabled by the fixed conditions of an initial state (universal grammar) invariant over evidential variation. Still, the end-state of acquisition (an internalised grammar) is understood to be theory-like in possessing a deductive structure, featuring unobservable posits, 238

Chomsky, Cognizing, and Tacit Knowledge

and having a predictive content in terms of being able to assign structures to observed utterances. Thus: The structure of particular languages may very well be largely determined by factors over which the individual has no conscious control. . . . [A] child cannot help constructing a particular sort of transformational grammar to account for the data presented to him, any more than he can control his perception of solid objects or his attention to line and angle. (Chomsky 1965: 59) In short, ‘theory construction’ describes the acquisition process and its end-states without the presumption that a theory of the language is in any sense an object of knowledge for the speaker-hearer in the way a scientific theory is for a scientist.

3.  Competence/Performance and Tacit Knowledge In Aspects, Chomsky further clarifies what kind of knowledge is at issue with regard to language and just what the core phenomena are. The famous distinction drawn in Aspects is between competence and performance. They are specified as follows: Competence: ‘knowledge of the language that provides the basis for actual use of language’ (Chomsky 1965: 9). Performance: the use of the language affected by extra-linguistic factors such as ‘memory limitations, distractions, shifts of attention and interest, and errors (random or characteristic)’ (ibid.: 3), and ‘intonational and stylistic factors, “iconic” elements of discourse, and so on’ (ibid.: 10–11). Four features bear emphasis. First, the distinction is not one between the interesting and the uninteresting, as it were, but is best viewed as a minimal fractionation of linguistic phenomena in a broad sense in order to gain some theoretical traction. So, if one set out to offer a theory of linguistic performance, a distinction must still be drawn between those factors that operate generally (memory, attention) or have no specific structural basis (many discourse effects, metaphors, etc.) and those factors peculiar to the nature of language itself being a specific kind of rule- and principle-based system. Similarly, if one targeted competence alone, one would not factor in perfectly general considerations such as the bounds on working memory (cp., ibid.: 15). Secondly, in line with the distinction, Chomsky draws a further distinction between acceptability and grammaticality (ibid.: 11). The former refers to what speaker-hearers find readily interpretable or not otherwise untoward. The latter refers to what a specific theory predicts to be interpretable or well-formed by the rules of the system. The two can come apart, especially in cases where rules are iterated in ways that pose difficulties for parsing. For example, (1) will strike English speakers as unacceptable: (1) The rat the cat the dog chased ate died (Miller and Chomsky 1963: 286) The sentence is formed, however, simply by iterated relative clause modification of the rat and then the cat, and so should be grammatical. This suffices to suggest that performance factors can interfere with competence in the sense that the deliverances of competence can fail to meet performance conditions. It is crucial to note that if one were to view (1) as somehow 239

John Collins

ungrammatical, rather than merely unparsable, and so eschew a distinction between grammaticality and acceptability, an onerous explanatory cost would be incurred. One way of denying the distinction would be to stipulate against cases like (1) by specifying relative clause formation to be sensitive to whether the nominal being modified by a relative clause is a constituent of a relative clause or not. This manoeuvre, however, complicates the rule of relative clause formation without explaining anything beyond the unacceptability of cases such as (1). The move, therefore, appears to be an ad hoc stipulation. As a matter of fact, every account of relative clauses within any grammatical theory I am aware of admits centre-embedding to any finite degree, precisely because any bound would amount to a stipulation that offers no explanatory benefit. Similar reasoning holds for myriad other cases. The simplest and most explanatorily efficient rules allow for unacceptable structures, which itself militates for the competence/performance distinction by indicating that, performance-wise, a free generative system functions under external (extragrammatical) conditions whose ensemble issues in the structures speaker-hearers find acceptable. Thirdly, the knowledge that is understood to constitute competence is tacit in ways that we shall further interrogate, but this is how Chomsky (ibid.: 8) introduces the claim: [A speaker-hearer having knowledge does not mean] that he is aware of the rules of the grammar or even that he can become aware of them, or that his statements about his intuitive knowledge of the language are necessarily accurate. . . . [A] generative grammar seeks to specify what the speaker actually knows, not what he may report about his knowledge. The knowledge in play, therefore, is not necessarily conscious, or even potentially conscious, and what a speaker (sincerely) reports may be quite misaligned with the knowledge. This latter point reiterates the preceding second point: speaker-hearers can be inaccurate about what structures are sanctioned by their own knowledge because performance factors interdict against well-formed (licitly generated) structures. Fourthly, from what has just been said, ‘[p]erformance provides evidence for competence’ (Chomsky 1966a: 10) in the sense that the speaker-hearer can only provide acceptability data, not direct grammatically intuitions, for her intuitive reports on her language are not guaranteed to be accurate, are subject to interfering factors, and are not even potentially exhaustive of the underlying knowledge that is the target of inquiry. Pace some recent critics (e.g., Devitt 2006), a speaker-hearer is not treated as a ‘voice of competence’. The theorist of competence, therefore, uses performance/acceptability data to inquire into the interaction of underlying systems, particularly the hypothesised core knowledge system, that issue in the acceptability judgements. Prior to seeing how these ideas developed in subsequent years, we need to diverge from the historical trajectory somewhat in order to say something about knowledge in general and what the content of this knowledge was supposed to be roughly between the late 1950s and the mid-1970s.

4.1 Knowledge From the early 1960s onwards, Chomsky (1964, 1965, 1966b) made appeal to classical rationalist thought as a significant antecedent to his own position. The tradition stretches back to Plato, inclusive of, in the modern era, Descartes, the Port Royal grammar, and some later Romantic thinkers. To render a complex history simple, a keystone of rationalism was the identification of the innate with a priori necessary truth. In this context, it is all too easy to take Chomsky’s appeal to knowledge of language as akin to Plato’s or Descartes’ claims about our knowledge 240

Chomsky, Cognizing, and Tacit Knowledge

of geometry, say. As we have seen already, however, a striking difference that Chomsky makes explicit is that knowledge of language is not essentially open to reflection, no matter how attentive a speaker-hearer might be. We shall see further differences too. In essence, while the historical antecedents Chomsky cites are real and significant, they are liable to mislead unless it is properly borne in mind that Chomsky’s endeavour, we might say, is to render a rationalist position as an empirical claim, not to support an unmodified classical rationalist position by way of empirical considerations. What is rationalist about Chomsky’s claims is that knowledge is innate in the sense that it develops and can be exercised under a poverty of stimulus, which contradicts classical empiricist thought. There is no conflation of the innate with a priori necessary truths. In philosophical discussions, knowledge is typically treated as coming in three apparent flavours corresponding to the kind of complement the verb know may take. Propositional knowledge: Billy knows that Jane went to the party Practical knowledge: Billy knows how to ride a bike Awareness/acquaintance: Billy knows Jane For present purposes, let us ignore the third category, for if the object of knowledge is something that could be rendered propositional, then the third category will reduce to the first. The default view of propositional knowledge is that it involves a true justified belief. So, Billy can’t know that Jane went to the party unless she in fact did go, and this is something Billy must believe (if he doubted it, he would hardly know it), and have some reason or justification for the belief; a lucky guess, or an insistence in the face of contrary evidence appears to undermine knowledge, even if accurate. Suffice it to say, whether true justified belief amounts to a definition of knowledge (singularly necessary and jointly sufficient conditions) is highly moot. Still, the features at least reflect aspects of our common notion, and facticity appears to be an essential feature, even if justification is a more slippery condition. At any rate, we can ask if the knowledge of language Chomsky hypothesised amounts to true justified belief. Practical knowledge (know-how) is traditionally distinguished from propositional knowledge, but with much attendant controversy. According to such a view, for Billy to know how to ride a bike, say, is not for Billy to believe some true propositions about bike riding, not even tacitly, but for Billy to be able to ride a bike in a suitably robust sense (it ought not to be, say, that a fortunate wind always keeps him upright). Of course, one may always ask after the basis of such an ability or disposition and posit tacit knowledge as an answer, but such knowledge is not constitutive of the know-how in question, at least not in the traditional picture I have in mind. Again, then, we may ask if knowledge of language of the kind Chomsky hypothesises amounts to a certain kind of complex ability, with linguistics being the inquiry into the genesis and nature of this ability. We shall return to these issues in §5. We first need to be clear about what is putatively known by the speaker-hearer.

4.2  Grammatical Knowledge Minimally, a grammar in Chomsky’s sense supports a pairing of signal (sound, hand gestures, etc.) with meaning. A direct mapping from concrete symbol types in their surface organisation to a meaning or interpretation, without recourse to specific grammatical rules or principles, is at least conceivable. Natural language, though, is not so simple. One component of a grammar is taken to be a procedure or function that generates a set of structures that can be incarnated by words that reflect properties of meaning. In Aspects up to work from the 1990s, the structures of this kind were called D(eep)-Structure. This amounts to a phrase structure grammar that defines constituency and allows for embedding in 241

John Collins

a manner that reflects semantic relations between the composed elements. A further component is transformational, which is a set of rules or operations that map from D(eep)-Structure to S(urface)-Structure, where the latter encodes linearity and any other properties supported by how a sentence appears in terms of the ordering of the constituents. So, the two levels, linked by transformations, constitute the desired pairing of signal with meaning, insofar as both levels encode properties uniquely relevant to each component of the paring. For the sake of clarity and space, we must restrict ourselves to a single striking example from Chomsky (1964). Consider: (2) a b

Mary is eager to please Mary is easy to please

The cases are the same, save for the difference of adjective, but since both adjectives can take infinitive complements, it seems that there is no grammatical difference, and so D-Structure should be the same as S-Structure, rendering the distinction otiose. Now compare (2) with (3): (3) a b

It is eager to please Mary It is easy to please Mary

(3a) is not a paraphrase of (2a), whereas (3b) does paraphrase (2b). Reflect on the pronoun it, which is lexically ambiguous between a pleonastic construal (cf., It’s raining!) and an indefinite referential construal (cf., It’s alive!). In (3a), it can only be read referentially, as might be used to refer to a particular dog’s eagerness to please Mary; it can’t be pleonastic, as if nothing whatsoever were eager. Thus, the sentence doesn’t mean what (2a) means, which is about Mary’s eagerness, not that of some other object, such as a dog. Conversely, the it of (3b) can only be read pleonastically, where no-one is being easy to please Mary (whatever that might mean), but that Mary’s being pleased is easy to bring about by someone or other. Since (3b) does paraphrase (2b), this shows us that the respective subject positions of the cases in (2) are differentially construed. Since D-Structure is designed to encode such differences, the conclusion is that whereas the cases in (2) appear the same, they have distinct D-Structures approximate to (4): (4) a b

Mary is eager Mary to please e Δ is easy e to please Mary

Transformational rules apply, deleting, moving, and filling in the structures to arrive at S-Structures approximate to the ordering in (2). The details need not detain us, and have been under constant revision, anyway. The bottom line for the moment is that if a speaker-hearer knows a grammar, then she knows a set of rules that produce two kinds of structures derivationally linked, and so she could be said to know the rules and the properties of the structures as they apply to particular cases of her language.

5.  Stripping Knowledge Back As we saw in §3, Chomsky’s conception of knowledge as a relation a subject has to their language is unconscious at the relevant level of analysis. It remains to ask whether it is propositional or practical, and even whether the question of truth arises, let alone justification. First let us clear away the know-how question. 242

Chomsky, Cognizing, and Tacit Knowledge

Chomsky (1968/72: 191) remarked that ‘In general, it does not seem to me true that the concepts “knowing how” and “knowing that” constitute exhaustive categories for the analysis of knowledge’, and the categories fail, in particular, to track the sense of knowledge apt for linguistic theory. Chomsky does not have know-how in mind in the sense of an ability to do various things with language. Of course, knowing a language allows us to communicate and reflect and organise our thoughts, etc., but such abilities neither constitute the knowledge in question nor are even necessary for it. Chomsky (1980: 51, 2000: 50–51) raises a simple objection to the knowhow view. Our ability to do things with language is subject to various constraints and conditions our knowledge is not. A speaker-hearer, for example, might temporarily lose the ability to use language, but they have not, therefore, lost the knowledge, which might be deployed after recovery, say. Tricky issues arise with our ordinary attributions of know-how, which we can’t here explore. Little is to be gained, however, by tarrying over the colloquial locutions; it suffices to note, and here we simply reiterate the moral of the competence/performance distinction, that if linguistic knowledge is a species of ability, then it will fractionate into the activity of distinct systems, and so we may preserve ‘knowledge’ in the intended sense for the core linguistic system that is invariant over changes or damage to otherwise identifiable systems and sensory modalities that support linguistic abilities. The issue becomes merely semantic. Notwithstanding Chomsky’s quoted distancing from ‘knowledge that’ as the operative notion, it is common to think of knowledge of language as propositional knowledge concerning truths about language (cf., Fodor 1968, 1983; Graves et al. 1973; George 1989; Knowles 2000). Fodor (1983: 4–5), for example, avers that ‘what is innately represented should constitute a bona fide object of propositional attitudes; what’s innate must be the sort of thing that can be the value of a propositional variable in such schemas as “x knows (/believes/cognizes) that P” ’ (cf., Fodor 2000: 10–11). The irony here is that Chomsky (1975: 164–165; cf. 1980: 69–70) coined cognize as a new technical term to replace ‘knowledge’ precisely in order to forestall constraints on the object of inquiry from the associated properties of the colloquial idiom. Thus, Chomsky (op cit.) is happy to think of the speaker-hearer as cognizing everything from humdrum facts about ‘dog’ being an English word, to particular transformational rules, to general principles of universal grammar, which are not particular to any language, and, of course, the judgements of acceptability that the grammar makes possible. The point of the locution is just to isolate humans as possessors of distinctive cognitive states specifiable independent of their interaction with other systems and the behaviours they result in. Of course, so much is consistent with all such states only being specifiable via propositional truths to be known or cognized. This claim, however, is implausible in itself, and not what Chomsky thinks, either. Chomsky (1981a: 9) writes: I shall refer to this relation as ‘(tacit) knowledge’, thus saying that a person knows his grammar and knows the rules and principles of his grammar, which of course does not imply that he has propositional knowledge that these are the rules and principles of his grammar. The linguist may develop such propositional knowledge, but that is quite a different matter. There are two basic thoughts behind this claim, which Chomsky appears to treat as obvious. Firstly, for the speaker-hearer, no questions arise as to the truth of the principles or rules at issue, for there is no independent object for them to be true of or apply to. For example, while a speaker-hearer will acquire a language that is (more or less) the same as the speaker-hearers with whom she grows up, the acquired ‘knowledge’ is not about these other agents or their language. 243

John Collins

This latter move became clear in Chomsky’s (1986) eschewal of the notion of language as such for I-language as a specified state of the speaker-hearer rather than an external thing to which the speaker-hearer might be related, but the basic idea goes back to the 1960s: Since the language has no objective existence apart from its mental representation, we need not distinguish between ‘systems of belief ’ and ‘knowledge’. (1968/72: 169, n. 3) The point here being that no question of truth arises beyond that the subject represents certain rules and principles, which is a matter for theoretical determination. The truth of such rules and principles only pertains to their role in fixing what the speaker-hearer understands, not what she is related to, which might be nothing at all, in the relevant sense, i.e., nothing external to the speaker-hearer makes true or false what she ‘knows’. This renders the knowledge quite unlike the supposed a priori truths conjured by classic rationalist thought. The matter is clarified by dropping ‘language’ altogether: The basic cognitive notion is “knowledge of grammar,” not “knowledge of language”. . . . Knowledge of language, so understood, involves or perhaps entails particular instances of knowledge-that and belief-that. .  .  . But knowledge is not constituted of such elements and no atomistic, element-by-element account of the character of growth of knowledge of this sort is possible. A system of knowledge in this sense is an integrated complex represented in the mind. . . . [T]here being no external standard, the domains of potential and actual fact are identical; X’s grammar is what X’s mind constructs. (1981b: 6; cf., Chomsky 1975: 164–166) In other words, ‘knowledge’ covers the interaction of a complex suite of systems that only as an ensemble give rise to propositionally articulate knowledge, but there is nothing for the grammar to be true of beyond what the system represents, and the knowledge in a colloquial sense does not constitute the system. Chomsky’s second reason for resisting propositional attribution is more basic, and sheds light on the previous points. Much of a grammar, especially if our focus is on universal grammar, is simply not propositional in relation to the system as a whole. This is a central feature of the developments in linguistics from the 1970s onwards. As explained, early generative grammars consisted of a set of rules that defined certain levels of representation. It is not implausible to think of such components as propositions a speaker-hearer knows, and so she might be said to know a priori truths. For example, a simple rewrite rule like ‘VP → V NP’ can be read as a derivational rule a device follows, or as stating categorical information that a verb followed by a noun phrase is a verb phrase. Increasingly, however, a grammar was conceived as a set of open parameters, fixed by experience, and a minimal number of principles that constrain certain uniform structure building and movement operations, so that under a so-called ‘minimalist’ orientation (Chomsky 1995), there could just be a single operation Merge, which recursively combines two elements together in a way that meets certain general conditions. The crucial point here is that a speaker-hearer does not, in any sense, represent such conditions or principles as propositions, for the principles describe constraints on the system, not what the system itself might be said to represent. For example, the projection principle (Chomsky 1981a, 1981c) in effect says that the derivational character of a grammar is monotonic, that is, it does not lose information (for example, an adjunct wh-item such as how that moves to the front of a sentence in question formation is still interpreted as modifying the item lower in the structure). This 244

Chomsky, Cognizing, and Tacit Knowledge

is not a statement of a grammar, but a statement about a grammar, a condition every grammar meets. So, one might think of a speaker-hearer as representing words and representing their combination into phrases, perhaps as entries in memory or a work space, but the agent doesn’t represent constraints on such combination: only the theorist does that in an effort to understand the combinatorial principles at issue. Similarly, that all grammatical relations are binary is a consequence of Merge, not a principle a grammar itself represents. It will be noted that Chomsky uses ‘represent’ and ‘cognize’ indifferently between such cases, which might make for some confusion, but does indicate that his concern is not for information that could be rendered as propositional or truth-apt for the competent speaker-hearer. In sum, we might think of knowledge of a grammar, as opposed to language, as an internal relation constituted by a speaker-hearer realising a certain abstract-computational structure, rather than an external relation to something otherwise identifiable: [F]or H to know L is for H to have a certain I-language. The statements of a grammar are statements of the theory of mind about the I-language, hence statements about structures of the brain formulated at a certain level of abstraction from mechanisms. (Chomsky 1986: 23) Here, the apparent external relation ‘K(H, L)’ is analysed as H being in certain brain states abstractly characterised in terms of an I-language, where a grammar or theory is about such states so characterised, that is, at a level of abstraction from mechanisms. So, an I-language is not the object of knowledge or even a product of the mind, but the state of the mind/brain that is picked out by the informal locution ‘H knows L’. An informal gloss is harmless and, indeed, serves to pick out the phenomena at issue, so long as it is not freighted with a philosophical content the informal term appears not to carry.

6. Conclusion The path described is one of a gradual shedding of externalist and epistemic connotations of the epistemic locution know a language. I have endeavoured to show, however, that this shedding did not mark so much a fundamental philosophical change of mind, but more reflected the developments in the theory of grammar itself. It became increasingly odd to attribute to the speaker-hearer just what the theorist is seeking to develop, as if a speaker-hearer’s grammar and the theorist’s theory of a grammar are one and same. A grammar, for the speaker-hearer, is not a propositional body of knowledge, although that is what the theorist aims for: not a theory of language as an external entity, but as an explicit specification of an internal system that a speaker realises. It seems to me that this conception is one that could have been attributed to Chomsky at least at the time of Aspects (1965), but there would have been a certain incongruity between such an understanding and the formal character of the grammars at that time. An improvement in theory led to an improvement in the meta-theory.

Related Topics Chapters 1, 4, 6, 14, 18, 19, 29, 31

References Chomsky, N. 1955–56/75. The logical structure of linguistic theory. Chicago: Chicago University Press.

245

John Collins Chomsky, N. 1957. Syntactic structures. The Hague: Mouton. Chomsky, N. 1959. “Review of B. F. Skinner’s Verbal Behaviour”. Language, 35: 26–58. References to the reprint in J. Fodor and J. Katz, eds. 1964. The structure of language: readings in the philosophy of language. Englewood Cliffs: Prentice-Hall: 547–578 Chomsky, N. 1962. “Explanatory models in linguistics”. In E. Nagel, P. Suppes, and A. Tarski, eds., Logic, methodology and philosophy of science. Stanford: Stanford University Press: 528–550. Chomsky, N. 1964. Current issues in linguistic theory. The Hague: Mouton. Chomsky, N. 1965. Aspects of the theory of syntax. Cambridge, MA: MIT Press. Chomsky, N. 1966a. Topics in the theory of generative grammar. The Hague: Mouton. Chomsky, N. 1966b. Cartesian linguistics: a chapter in the history of rationalist thought. New York: Harper & Row. Chomsky, N. 1968/72. Language and mind. Extended edn. New York: Harcourt Brace Jovanovich. Chomsky, N. 1975. Reflections on language. London: Fontana. Chomsky, N. 1980. Rules and representations. New York: Columbia University Press. Chomsky, N. 1981a. “On the representation of form and function”. Linguistic Review, 1: 3–40. Chomsky, N. 1981b. “Knowledge of language: its elements and origins”. Philosophical Transactions of the Royal Society, B295: 223–234. Chomsky, N. 1981c. Lectures on government and binding: the Pisa lectures. Dordrecht: Foris. Chomsky, N. 1986. Knowledge of language: its nature, origin and use. Westport: Praeger. Chomsky, N. 1995. The minimalist program. Cambridge, MA: MIT Press. Chomsky, N. 2000. New horizons in the study of language and mind. Cambridge: Cambridge University Press. Devitt, M. 2006. Ignorance of language. Oxford: Oxford University Press. Fodor, J. 1968. “The appeal to tacit knowledge in psychological explanation”. Journal of Philosophy, 65: 627–640. Fodor, J. 1983. The modularity of mind. Cambridge, MA: MIT Press. Fodor, J. 2000. The mind doesn’t work that way: the scope and limits of computational psychology. Cambridge, MA: MIT Press. George, A. 1989. “How not to become confused about linguistics”. In A. George, ed., Reflections on Chomsky. Oxford: Basil Blackwell: 90–110. Graves, C., Katz, J., Hishiyama, Y., Soames, S. Stecker, R., and Tovey, P. 1973. “Tacit knowledge”. Journal of Philosophy, 70: 318–330. Hempel, C. G. 1966. Philosophy of natural science. Englewood Cliffs: Prentice Hall. Knowles, J. 2000. “Knowledge of grammar as a propositional attitude”. Philosophical Psychology, 13: 325–353. Miller, G., and Chomsky, N. 1963. “Finitary models of language users”. In R. Luce, R. Bush, and E. Galanter, eds., Handbook of mathematical psychology. Vol. II. New York: John Wiley and Sons, Inc.: 419–491.

246

19 LANGUAGE PROCESSING Making It Implicit? David Pereplyotchik

1. Introduction Discussions of implicit and explicit mental functioning proceed best when they meet two criteria. First, they must, from the very outset, be firmly grounded in empirical results. Second, they should take particular care with the conceptual categories they employ. In this chapter, I aim to meet the first criterion by focusing attention on just one area of cognition – adult, real-time language processing – which has been heavily studied by psycholinguists and neurolinguists in recent decades. The goal is to then to discern, from this narrow case study, which aspects of the mind can be said to be, in one sense or another, implicit. Meeting the second constraint – that is, conceptual clarity about various attendant notions – will occupy much of the remaining discussion.

2.  Language-Processing: A General Schema Understanding a sentence in one’s native language requires an array of information-processing systems. After registering the sound waves from a nearby utterance, the language-processing mechanism in the human mind/brain first categorizes the sounds into phonemes – fundamental linguistic units – and then uses these categorizations to recognize incoming lexemes in a learned mental lexicon. Having done so, it goes on to assemble representations of them into representations of larger structured phrases, and eventually of whole sentences. This allows the language processor to compute a representation of (something like) the logical form of the sentence, to assign literal meanings to it (if such there be), and to fill in information about the referents of any pronouns or indexicals in the utterance – for example, she/he/they, this/that/those, now/then. All of this is in the service of higher cognitive systems, whose role is to determine the utterance’s relevance to the overall discourse, to recognize potential implicatures or indirect speech acts (irony, metaphor, insult, etc.), and to update the hearer’s beliefs and plans. It is a wonder that all of this happens automatically, and in mere milliseconds. As with all overlearned, routinized, and quasi-reflexive processes, language comprehension – particularly in the early processing stages – occurs largely without the benefit of consciousness. We thus don’t have anything like reliable introspective access to how these processes work. Fortunately, we do

DOI: 10.4324/9781003014584-24 247

David Pereplyotchik

know a great deal about them from decades of sustained psycholinguistic research. Figure 19.1 provides a general schema that, while hotly controversial in its details, is sufficiently general for our purposes, conveying a sense of how human language comprehension takes place in real time (for a standard view, see Fernández and Cairns 2011). Psycholinguistics traffics heavily in mental representations. But what meaning does that vexed notion have in the present context? To answer, consider the role of such representations in two examples of psycholinguistic explanation. The primary case explored in this chapter involves mental phrase markers (MPMs). An MPM is an occurrent mental representation of the syntactic structure of incoming linguistic stimuli. Such representations are generated by the subcomponent of the language faculty known as the parser (Figure 19.1). As I will show, the success of various psycholinguistic explanations rests on the posit of MPMs. A second case, far more controversial, involves representations of the grammar – the rules and principles of syntax – that the parser uses. The success of various psycholinguistic explanations does not rest to the same extent on positing such grammatical-rule representations, though this is a more complicated matter. The discussion that follows is aimed at discerning the differences between these two cases and articulating the best account of the relevant psycholinguistic processes. To establish the need for MPMs, consider first the striking phenomenon of garden-path sentences. To appreciate it firsthand, take a moment to read sentences (1)–(4). But, be forewarned:

PRAGMATIC ENRICHMENT, PRONOUN RESOLUTION, IMPLICATURE RECOGNITION, DISCOURSE PROCESSING, ETC.

SEMANTIC ANALYSIS

BACKGROUND BELIEFS DESIRES GOALS EMOTIONS MEMORIES ETC. ???

GRAMMAR SYNTACTIC PARSING

LEXICAL SELECTION LEXICON PHONOLOGICAL PROCESSING

DATA STRUCTURE PROCESSING MODULE OTHER

THE LANGUAGE FACULTY RAW LINGUISTIC INPUT

Figure 19.1  Mechanisms of language comprehension

248

Language Processing

even experienced readers often find these sentences difficult to parse – despite their being, on fuller inspection, perfectly grammatical and sensible. (1) (2) (3) (4)

The dealer sold the forgeries complained. Aron gave the man who was eating the popcorn. The cotton clothing is made of grows in Arkansas. The daughter of the king’s son admired himself.

When sentences with these structures, and many others like them, are presented to experimental subjects, they reliably elicit increased reading and reaction times, lower comprehension and recall rates, and longer gaze-fixations on crucial bits of text, as measured by eye-trackers (Fernández and Cairns 2011). More recently, Electroencephalographic (EEG) studies have identified patterns of neural activity that underpin error-detection and repair processes (Friederici 2017). What makes garden-path sentences so unusual? Standard psycholinguistic accounts point to how the parser represents the syntax of the incoming material – that is, the mental phrase markers (MPMs) that it constructs. Let’s delve deeper into that process. Operating in accordance with an internalized grammar, the parser outputs MPMs that respect relatively rigid phonological, morpho-syntactic, and semantic constraints (Fernández and Cairns 2011). Critically, it does not wait for large chunks of material before initiating its processing routines. Rather, it immediately extracts and incorporates the linguistic information provided by each lexeme, without waiting for further input. Indeed, neurocognitive studies often rely on the parser to take note of one morpheme just milliseconds before another (Friederici 2017). Moreover, parsing is predictive, delivering representations of the structure of probable subsequent input (Fodor and Ferreira 1998). Incremental predictive processing presents a major problem for the parser. This is because real-time linguistic input is rife with syntactic ambiguity. At each step in the process, multiple hypotheses are available regarding how to parse incoming material, and what future material is likely to be. The parser must either take a gamble on one of these, and revise as necessary, or entertain several at once, along with a probabilistic ranking. In revising either an MPM or the ranking of candidate MPMs, the parser deploys intelligent, incremental, and predictive errorcorrection mechanisms; when it encounters a problem, it is able to reanalyze the input on the fly without having to re-scan it (Fodor and Ferreira 1998). In explaining why the parser makes various types of incorrect predictions in a circumscribed class of cases (well-defined for each language), psycholinguists appeal to various heuristics governing ambiguity resolution. These heuristics are, by hypothesis, not themselves mentally represented; rather, they are seen as well-confirmed generalizations about subjects’ processing tendencies. Crucially for our purposes, they make ineliminable reference to the construction and manipulation of MPMs. One style of ambiguity resolution relies on information about the frequencies of phonemes, words, and phrase-structures, to which the parser is demonstrably sensitive (Rumelhart and McClelland 1986). In some simple cases, information about the frequency of a piece of language determines the parser’s behavior. For instance, in deciding whether to take the subject of sentence (3) to be ‘the cotton’ (correct) or ‘the cotton clothing’ (incorrect), the parser chooses the latter, largely because ‘cotton clothing’ is a very common bigram. In more interesting cases, the parser’s decision turns on the syntactic complexity of the competing hypotheses. This is what we see in sentence (4), which typically requires explicit guidance to interpret properly. (Hint: The leader of the gang’s bomb exploded itself.) The parser’s natural tendency is to close off the noun phrase ‘the daughter’ rather than waiting to pack in more material (e.g., ‘of the king’), which 249

David Pereplyotchik

would make for a more complex possessive noun phrase. Such least-effort heuristics usually work, but they fail in cases like (4), where the subsequent input – specifically, ‘himself ’ – renders the sentence semantically incoherent. The parser then has to engage in a kind of reanalysis – that is, to select another structure, or to re-rank the previously considered options – in order to arrive at the more complex structure, ‘[the daughter of the king]’s son.’ The evidence currently available for the psychological reality of MPMs goes well beyond studies of ambiguity resolution. To ascertain the representational contents of MPMs, psycholinguists take advantage of a phenomenon known as structural priming. In general, priming amounts to triggering a mental representation and measuring its observable downstream effects on memory, reaction time, and other cognitive processes. The term ‘structural priming’ refers to those cases where tokening a perceptual representation of a word, a syntactic category, or a whole phrasal structure significantly modulates linguistic processing for up to a few seconds. To illustrate, it has been found that processing sentence (5) “primes” listeners to then describe a generic event using sentence (6), rather than the equally appropriate (7). (5) The soldier moved a tank for the police. (6) The man gave a flower to the woman. (7) The man gave the woman a flower. Here, an active representation of the syntactic structure Noun-Verb-Preposition-Noun, as in (5), primes the choice for that same structure in producing (6) rather than (7). This is, by modern syntacticians’ standards, a relatively coarse-grained representation, as it employs linguistic categories (e.g., Verb) that have been much refined over the past few decades. Nevertheless, by experimenting extensively with other priming relations, psycholinguists can construct an increasingly detailed picture of the parser’s representational scheme, with all the refinements that formal linguistics demands (Branigan and Pickering 2017). Having sketched some aspects of the rationale for positing MPMs, what can we say about the grammar that the parser uses in constructing its proprietary representations? The status of grammatical rules/principles is hotly contested (Devitt 2006; Pereplyotchik 2017). Unlike the case with MPMs, there are currently no decisive grounds for positing representations of syntactic rules/principles of the kind that one finds in linguistics texts. We cannot, at present, directly test the empirical hypothesis that a grammar is represented in the mind/brain in the manner of a data structure in a computer. To better appreciate the difference between MPMs and grammatical representations, we must first clarify several related notions.

3.  Conceptual Distinctions In what sense is parsing an implicit psychological process? The aim of the present section is to clarify various conceptual distinctions that sometimes run together in the literature on this topic. These include the distinctions between (i) occurrent and dispositional states, (ii) procedural and declarative knowledge, (iii) personal and subpersonal processes, and (iv) conscious vs. nonconscious mental functioning. I will argue that careful attention to the differences among these distinctions reveals that they are sometimes orthogonal to one another, and hence must be operationalized differently in the context of empirical research. This, in turn, allows for the formulation and defense of the following Hypothesis: MPMs are subpersonal analogues of personal-level declarative thoughts, whereas the rules/principles of an internalized grammar are a kind of procedural knowledge – 250

Language Processing

a set of dispositions to move from one occurrent subpersonal representation to another. Only once these distinctions are clarified can the reasons supporting Hypothesis be properly understood.

3.1  Conscious and Nonconscious States It’s plain that the implicit/explicit distinction has something to do with consciousness. Unfortunately, the topic of consciousness isn’t exactly a model of clarity. The term ‘conscious’ is used in a bewildering number of ways, both in ordinary language and in theory construction. Let’s note some key distinctions, drawn from Rosenthal (2002). Various creatures are conscious, for stretches of time, but asleep, comatose, or otherwise incapacitated at other times. In this sense, ‘consciousness’ means roughly the same as ‘wakefulness’. Let us call this creature consciousness. Adaptive engagement with an environment also requires a creature to be conscious of various things. Here, the use of ‘conscious’ is grammatically transitive, so we can call it transitive consciousness. In this sense of the term, ‘conscious of ’ is roughly equivalent to ‘has an occurrent mental representation of ’. The representations that most clearly implement this kind of awareness are sensory or perceptual. Transitive consciousness of a stimulus, or of some feature of the environment, can occur consciously, but it can also occur nonconsciously, as in cases of subliminal priming, blindsight, and hemineglect (Dehaene 2014). Here, the use of the term ‘conscious’ pertains to one or another mental state. It is in this sense that we speak of conscious and nonconscious desires, worries, fears, and the like. Let us use the term state consciousness to distinguish this notion from the others surveyed earlier. Any account of implicit processing that makes use of the notion of consciousness must respect the distinctions between state-consciousness, creature-consciousness and transitive-consciousness. Applying this to the case of MPMs, we can say that these are states of transitive consciousness – that is, representations of a linguistic stimulus – and that they occur in conscious creatures (e.g., awake, vigilant adults), but that they are not themselves conscious states. Accordingly, the processes in which they participate are, likewise, nonconscious.

3.2  Personal and Subpersonal Levels of Description Folk psychology describes and explains behavior in terms of states that are ascribed, in the paradigmatic case, to an entire person. By contrast, cognitive science frequently offers subpersonal explanations, which ascribe states to the information-processing mechanisms that jointly constitute a person. When modeling such a mechanism, there are often compelling grounds for thinking of some of its states as representations, as we have seen in the case of the language faculty (§2). However, it would be a mistake to uncritically assimilate subpersonal representations to personal-level beliefs. Let’s take stock of the relevant differences. First, a different flavor of normativity is involved in the ascription of folk-psychological states and processes (e.g., intentions and planning) than in the in the ascription of computational states and operations (e.g., data structures and algorithms). Dennett (1987) captures this idea with his distinction between three stances that one might adopt in describing, explaining, or interacting with natural systems. Adopting what he calls the physical stance, one takes as one’s target a material object, with mass, velocity, temperature, and the like. Presumably, no normative assumptions are presupposed in the ascription of such attributes. By contrast, in taking the 251

David Pereplyotchik

design stance, one characterizes the target as well-functioning designed system. Taking this stance toward a creature involves thinking of it as an aggregate of interacting purposeful mechanisms, each of which has the function of performing a specialized task. It’s from the design stance that we attribute subpersonal states, such as MPMs, to the human sentence processing mechanism – a gadget whose ‘job’ (when functioning properly) is to classify incoming linguistic input. The normative assumptions that are at play here are broadly teleological, grounded in a combination of learning and innate design by natural selection. Finally, one can take on even stronger normative commitments and ascend to the intentional stance, treating one’s target as a rational agent, who perceives and remembers what she ought to, in order to make reasonable inferences that lead to intelligent action. The difference between the teleological normativity of the subpersonal design level and the normativity involved in a personal-level intentional ascription is reflected in the kinds of criticisms that we offer in different circumstances. No one can be thought foolish for falling prey to the Müller-Lyer illusion, or for failing to comprehend sentences (1)–(4) on a first pass. Whatever cognitive failures are involved in such cases accrue to the visual system and the language processor, respectively, not to the person as such. Another mark of subpersonal states is their inexpressibility in speech. Consider again the case of MPMs. Although these undoubtedly play a causal role in language production, their representational contents, which pertain to grammatical categories and relations, do not coincide with the meanings of the speech acts whose production they facilitate. The meaning of ‘Dogs bark?’ is not [NP VP]. Nor does the illocutionary force of those speech acts in any way reflect the mental attitudes of the subpersonal states that facilitate them. The representational content [NP VP] is, if anything, assertoric, whereas the corresponding speech act is interrogative. What is expressed in speech is a personal-level thought, never an MPM. A related feature of subpersonal states is their permanent inaccessibility to consciousness. Notably, the usual grounds for positing nonconscious states seem not to apply to states like MPMs. Psychological defense mechanisms, such as repression, are thought keep troublesome desires and feelings out of consciousness. But these would have no cause to suppress emotionally neutral MPMs. Nor are MPMs dormant, like long-term memories; they are occurrent states, accessed and used in real-time information processing. Finally, while brief or degraded presentation of stimuli can render a perceptual judgment nonconscious, this is implausible as an explanation of why MPMs never occur consciously. In standard cases, they are activated by ordinary linguistic inputs, and participate in a processing stream that ultimately gives rise to a conscious experience of comprehension. Many hold that personal-level states – paradigmatically, propositional attitudes – come to be what they are in part by being caught up in a larger inferential network. Their intentional contents are often described as conceptual, on account of the indefinite range of inferential transitions that constitute concept-driven reasoning (e.g., generalization). By contrast, as Stich (1978) points out, subpersonal states are not inferentially integrated within the network of our propositional attitudes. That is, we cannot arbitrarily draw inferences whose premises or conclusions are MPMs. Very little ‘follows’ from the representational content [NP VP]. Still, just as it would be a mistake to hastily assimilate MPMs to personal-level states, it would be equally wrong to see them as mere causes with no psychological properties. For, although their contact with personal-level states is limited to facilitating linguistic comprehension, they are nevertheless enmeshed in their own subpersonal system of quasi-rational moves – the parsing steps sketched earlier – which has been successfully modeled by computational linguists. Summing up, subpersonal states have only some of the features characteristic of personallevel beliefs and desires. They bear systematic relations to the environment, to behavior, and 252

Language Processing

to one another. This makes it both reasonable and useful to think of them as representations, whose interactions resemble (in relevant respects) the inferential transitions between personallevel propositional attitudes. Doing so allows us to abstract away from largely unknown neural mechanisms and to see subpersonal faculties as engaged in purposeful activities that involve reasonable steps toward a cognitive goal.

3.3  Declarative Representations and Procedural Dispositions Though philosophers often use the term ‘belief ’ to pick out virtually any assertoric propositional attitude, this usage elides an important distinction between having an occurrent assertoric attitude and the disposition to have such an attitude. On a more careful usage – common in ordinary language and folk-psychological explanation – a thought is a state that one comes to be in occurrently, at a certain time and place, whereas beliefs are dispositions to have such thoughts, when the occasion arises. This folk-psychological distinction is analogous to one that is often drawn in cognitive science, between declarative and procedural knowledge. (I ignore here the epistemic baggage of the term ‘knowledge’.) With this distinction in mind, I can articulate the aspect of Hypothesis that deals with rules and principles: the rules/principles of an internalized grammar are a kind of procedural knowledge – a set of dispositions to move from one occurrent subpersonal representation to another. To get a handle on what this means, let’s look at two parallel types of psychological explanation – one belonging to personal-level folk psychology, and the other to subpersonal psycholinguistics. At the personal level, we tell stories such as the following: ‘Kurt believes that spiders live in large numbers. So, when he just now saw several spiders crawl out of a box in his attic, he reasoned that there are probably many more spiders in that box.’ In this story, an occurrent thought regarding the presence of spiders interacts with a standing belief to yield another occurrent thought as a conclusion. The process of syntactic parsing, though it takes place subpersonally, has roughly the same structure. Consider, for instance, what happens when the parser encounters an ambiguous sentence-opening. (8) Have the soldiers . . . [possible continuations: . . . eaten? (question) . . . fed! (command)] The parser houses two standing dispositional states that embody the following grammatical principles: (i) a sentence-initial auxiliary verb (e.g., ‘Have’) serves to introduce a question; (ii) a subsequent noun phrase (e.g., ‘the soldiers’) will be the subject of the clause. As a result of registering the presence of the word ‘Have’ in the input, the parser constructs an MPM that represents ‘Have’ as an inverted auxiliary verb, in accordance with principle (i). From this occurrent MPM and the standing principle (ii), the parser ‘concludes’ – sometimes incorrectly – that the overall sentence will be a question, and hence that the subsequent noun phrase will be the subject of the clause.

4.  What Is It for Language Processing to Be Implicit? I have now laid the groundwork for a careful dissection of the central questions that animate this discussion: First, in what sense are the early stages of language comprehension implicit? Second, what is the relationship between tacit knowledge of grammar and implicit representations? I address these questions in §4.1 and §4.2, respectively. 253

David Pereplyotchik

4.1  Conceptual clarity A useful place to begin is the classic discussion of implicit and explicit rules of grammar (including phonology) in Davies (1995). Pessimistically, he points out: [T]he distinction in experimental psychology between explicit and implicit memory tests does not help us. In an implicit memory test “memory for a recent experience is inferred from facilitations of performance, generally known as repetition or direct priming effects, that need not and frequently do not involve any conscious recollection of the prior experience” [Schacter 1989: 695]. In contrast, explicit memory tests “make explicit reference to and demand conscious recollection of a specific previous experience.” (ibid.) Given that we are considering human information processing much of which is unconscious, and that we also want our notions of knowledge of rules to be applicable to systems – such as small connectionist networks – for which the question of conscious recollection does not even arise, this . . . usage of “explicit” is of no help to us. (Davies 1995: 158–159) As Davies hints, there is a persistent habit in the literature of conflating or ignoring the distinctions discussed in §3. Thus, progress on the psychological reality of grammars requires being as clear as possible about the logical relationships between five different distinctions: conscious/ nonconscious, personal/subpersonal, implicit/explicit, declarative/procedural, and occurrent/ dispositional. Let us now attempt to tease out these relationships. Focusing for the moment on the declarative/procedural distinction, we can compare two ­examples of declarative memory representations: (1) the factual knowledge that dogs are mammals – clearly a personal-level state, which is sometimes conscious (especially when expressed in speech) – and (2) lexical memory, for example, information about the argument places of verbs – a subpersonal state and hence never conscious. It appears that some declarative knowledge is accessible to consciousness, while some isn’t. Next, consider whether the occurrent/dispositional distinction lines up with personal/subpersonal distinction. We find that it doesn’t. The folk-psychological practices of predicting and explaining behavior rest on the ascription of both thoughts (occurrent states) and beliefs (dispositional states) to whole persons, not to subpersonal mechanisms. So personal-level states and processes can have either property. The same is true at the subpersonal level, where we’ve seen both occurrent states, such as MPMs, and procedural dispositions to transition between them – for example, the principles of grammar, if Hypothesis is correct. Lastly, we might ask whether the conscious/nonconscious distinction lines up with personal/ subpersonal distinction. Again, it does not. Personal-level states, including intentions, thoughts, expectations, hopes, worries, fears, desires, volitions, and decisions have all been shown in experimental settings to occur nonconsciously (Berger 2014). Some of these are occurrent states, while others are dispositional. We are not consciously aware, for instance, of all our beliefs and memories at any given time, though we are able to answer questions like ‘Have you ever ridden an antelope?’ without skipping a beat (Dennett 1978). When the question is presented, a standing belief gives rise to an occurrent thought, which is then expressed in speech. If the question is presented in a clever masked-priming study or in a variety of other conditions, the occurrent thought may still be formed, though nonconsciously. Being an occurrent state, then, is neither a necessary nor a sufficient condition for being a conscious state. And being a dispositional state is neither necessary nor sufficient for being a nonconscious state. 254

Language Processing

Here is a summary of the conclusions reached thus far. First, at the personal level of description, some states/processes are conscious, while others aren’t. Likewise, some are occurrent, while others are dispositional; some are declarative, and some procedural. At the subpersonal level, all states/processes are nonconscious; nevertheless, some are occurrent, while others are dispositional, and some are declarative, while others are procedural. With these conclusions in mind, we arrive at several plausible options for defining ‘implicit states/processes’.  (i) Nonconscious (ii) Procedural/dispositional (iii) Subpersonal   (iv) Subpersonal and procedural/dispositional For ‘explicit’, we have the following options:  (i) Conscious (ii) Occurrent/declarative (iii) Personal-level   (iv) Personal-level and conscious Plainly, some of these proposals are broader than others. For instance, if we equate implicit with nonconscious, as in (i), then we get the result that much of language processing is implicit, regardless of whether some of the representations involved are occurrent (and often called ‘explicit’), while others are hardwired procedural dispositions. Similarly, if we equate implicit with subpersonal, as in (iii), then we get the result that nearly all perceptual processing is implicit. This seems too broad. I leave it to the reader to decide which of these is most relevant to their theoretical pursuits. However, given the more fine-grained distinctions now at our disposal, one might argue that it is profitable to be more careful than the implicit/explicit dichotomy allows. Spelling it out, we should say that parsing is a nonconscious, subpersonal process, relying on occurrent representations of incoming material (MPMs) and, if Hypothesis is correct, procedural dispositions to move from one MPM to another.

4.2  Are the Principles of Grammar Explicitly Represented? Some systems – for example, the modern-day personal computer – compute their target functions by explicitly representing, as a separate data structure in their memory banks, the very programs that they are running. Such systems have control states that explicitly represent the instructions of a program, which are causally involved in the inner workings of the machine. Each such instruction encodes one of the basic functions that, together with other basic functions, conspire to produce the output of the overall target function. However, not all computers are stored-program systems of this kind. Some are hardwired circuits, in which the transitions between occurrent states are not the result of interactions between two types of data structure – program and input. Rather, a program that is directly hardwired into the circuit acts, in the manner of a procedural disposition, to facilitate transitions from one occurrent representation to another. Horgan and Tienson (1999) recommend that we think of some types of connectionist networks as hardwired circuits in this sense. In connectionist models, rules for processing representations are not explicitly represented. It is sometimes assumed that classicism is committed to explicitly represented 255

David Pereplyotchik

rules, and that lack of such rules therefore constitutes an essential difference between classicism and connectionism (e.g., Hatfield 1991). But although programs are explicitly represented as stored “data structures” in the ubiquitous general-purpose computer, stored programs are not an essential feature of the classical point of view. In some computational devices – including, for example, many hand-held calculators – the rules are all hardwired into the system and are not explicitly represented. (725) Thus, the subpersonal principles of a grammar, if considered as a kind of procedural knowledge, can accomplish their task either by being explicitly represented as a data structure or by being ‘hardwired’ into the brain. For instance, consider the language-processing system ACT-R, in which [a]ll procedural knowledge is represented as production rules – asymmetric associations specifying conditions and actions. Conditions are patterns to match against buffer contents, and actions are taken on buffer contents. All behavior arises from production rule firing; the order of behavior is not fixed in advance but emerges in response to the dynamically changing contents of the buffers. (Vasishth and Lewis 2006: 414) The authors cast language comprehension as kind of procedural knowledge, but they treat the rules of grammar as explicit representations, on a par with the occurrent representational states that I’ve been calling MPMs. However, Devitt (2006) points out that this is simply an artifact of the researchers’ running their models on conventional stored-program computers. In a different kind of system, he argues – and perhaps, indeed, in the human brain – procedural knowledge might not consist in representational states. ACT-R theory .  .  . “accounts for a wide range of cognitive phenomena, including perceptual-motor tasks, memory, and problem-solving” (Johnson et al. 2003: 32). The key question for us is whether this achievement requires that the production rules . . . be represented. Descriptions of ACT are often vague on this matter but the received view is that the rules are represented: “in ACT both declarative and procedural knowledge are represented in an explicit, symbolic form (i.e. semantic networks plus productions)” (Sun et al. 2001: 235; see also Masson 1990: 223). Yet . . . since the production rules constitute procedural not declarative knowledge there seems to be no immediate and pressing need to take them as represented. Because ACT theories are based on general-purpose computer models (Anderson 1983: 2) it is perhaps not surprising that the cognitive architecture they propose involves the representation of production rules. Still, we wonder whether we should take this aspect of the model seriously if we are looking for a simulation of skills that exist in real organisms. Is there any reason to think that the IF – THEN rules that become embodied in an organism as a result of practice are represented rather than merely embodied? Perhaps we can suppose that the organism has simply learnt to respond to the working memory representation of a certain condition with the appropriate action. Is there any explanatory gain in supposing further that it does this by representing the rule governing this response and applying it? (Devitt 2006: 215)

256

Language Processing

How might we use future empirical studies to address the question of whether the principles of a grammar are ‘hardwired’ procedural dispositions or explicitly represented? An important insight comes from Davies (1995) who explicates the notion of implicit knowledge thus: A device that effects transitions between a set of inputs and their respective outputs can be said to have implicit knowledge of a rule just in case the device contains a state or a mechanism that serves as a common causal factor in all of those transitions. In a series of clever examples, Davies demonstrates that there are ways of satisfying this definition which are logically weaker than explicit representation, but, at the same time, stronger than mere conformity with a rule. His examples contrast different connectionist network architectures, all of which are capable of conforming with a set of basic phonological rules, but only some of which facilitate this conformity by mechanisms that serve as ‘common causal factors’ in the application of only one rule. If Davies is right, then our verdict concerning the leading question of this section will depend partly on the anatomical and neurocognitive facts – specifically whether there are neural mechanisms that act as ‘common causal factors’ in facilitating the application of only one grammatical rule/principle of a grammar, or whether the mechanisms involved are routinely recruited by a variety of other seemingly disparate tasks, or function in even less direct ways. These are empirical issues whose resolution awaits further neurocognitive research. Fortunately, the cutting-edge field of neurolinguistics is now delivering exciting results that may well bear on this question in the years to come. Let us briefly survey some key findings. It has been well known for some time now that the brain reacts strongly to unexpected stimuli. EEG studies of language comprehension take advantage of this fact, in what is known as the violation paradigm. Here, participants encounter linguistic material that contains violations of one or another linguistic constraint. The resulting neural signals, event related potentials (ERPs), are time-locked to crucial stimulus regions which are underlined in the following examples. Syntactic violation: *I admired Eugene’s of sketch the landscape. Semantic violation: *I admired Eugene’s headache of the landscape. Different types of violation have been found to elicit distinct ERPs, providing important clues about their functional interpretation (Friederici 2017). Magnetoencephalographic (MEG) studies have likewise confirmed that the parser tracks phrase-level groupings. For instance, Ding et al. (2016) showed that “although linguistic structure building can clearly benefit from prosodic or statistical cues, it can also be achieved purely on the basis of the listeners’ grammatical knowledge” (158). These researchers found, moreover, that statistical cues regarding the frequency of certain structures “are not always available, and even when they are available, they are generally not sufficient” (163). This conclusion speaks to the inadequacy of computational models of human comprehension that seek to do away with MPMs entirely, and to replace them with sophisticated statistical analyses of the linear order of incoming words. Ding et al. take their results to “underscore the undeniable existence of [not linear, but] hierarchical structure-building operations in language comprehension”, which, they add, “relies on a listeners’ tacit syntactic knowledge” (2016: 162). As such experimental work yields increasing refinements in our understanding of the representational contents of MPMs, it will likely also shed light on whether the mechanisms that produce them are clustered into what Davies calls “common causal factors”. If so, then there is a clear sense in which the rules/principles of an internalized grammar are only implicitly represented, whereas occurrent states like MPMs are arguably explicit, occurrent representations, albeit subpersonal.

257

David Pereplyotchik

5. Conclusion Throughout this discussion, my aim has been to clarify the distinctions that bear on the representations involved in language processing. Having done so, I argued that MPMs are subpersonal analogues of personal-level occurrent, declarative thoughts. By contrast, the rules or principles of an internalized grammar may well be a kind of procedural knowledge – a set of hardwired dispositions to move from one occurrent subpersonal representation to another. This distinction informs debate about whether, and in what sense, either of these theoretical constructs count as implicit. My hope is that the resultant conceptual clarity enhances future empirical work in psycholinguistics, and furthers the philosophical debate about which theoretical posits in successful psycholinguistic theories hold a claim to ‘psychological reality’.

Related Topics Chapters 1, 3-6, 13, 14, 18, 20, 29-31

References Berger, J. 2014. “Mental states, conscious and nonconscious”. Philosophy Compass, 9: 392–401. Branigan, H., and Pickering, M. 2017. “An experimental approach to linguistic representation”. Behavioral and Brain Sciences, 40: 1–17. Davies, M. 1995. “Two notions of implicit rules”. In J. E. Tomberlin, ed., Philosophical perspectives, 9: AI, connectionism, and philosophical psychology. Cambridge, MA: Blackwell. Dehaene, S. 2014. Consciousness and the brain: deciphering how the brain codes our thoughts. New York: Viking Press. Dennett, D. C. 1978. Brainstorms: philosophical essays on mind and psychology. Cambridge, MA: MIT Press. Dennett, D. C. 1987. The intentional stance. Cambridge, MA: MIT Press. Devitt, M. 2006. Ignorance of language. Oxford: Oxford University Press. Ding, N., Melloni, L., Zhang, H., Tian, X., and Poeppel, D. 2016. “Cortical tracking of hierarchical linguistic structures in connected speech”. Nature Neuroscience, 19: 158–164. Fernández, E. M., and Cairns, H. S., eds. 2011. Fundamentals of psycholinguistics. Malden: Wiley-Blackwell. Fodor, J. D., and Ferreira, F., eds. 1998. Reanalysis in sentence processing. New York: Kluwer Academic Publishers. Friederici, A. D. 2017. Language in the brain: the origins of a unique human capacity. Cambridge, MA: MIT Press. Horgan, T., and Tienson, J. 1999. “Rules and representations”. In R. A. Wilson and F. C. Keil, eds., The MIT encyclopedia of the cognitive sciences. Cambridge, MA: MIT Press: 724–726. Pereplyotchik, D. 2017. Psychosyntax: the nature of grammar and its place in the mind. Cham, Switzerland: Springer. Rosenthal, D. 2002. “Explaining consciousness”. In D. Chalmers, ed., Philosophy of mind: classical and contemporary readings. New York: Oxford University Press: 406–421. Rumelhart, D., and McClelland, J., eds. 1986. Parallel distributed processing. Cambridge, MA: MIT Press. Schacter, D. L. 1989. “Memory”. In M. I. Posner, ed., Foundations of cognitive science. Cambridge, MA: MIT Press: 683–725. Stich, S. 1978. “Beliefs and subdoxastic states”. Philosophy of Science, 45: 499–518. Vasishth, S., and Lewis, R. 2006. “Symbolic models of human sentence processing”. In K. Brown, ed., Encyclopedia of language and linguistics. 2nd edn, Vol. 5. Amsterdam: Elsevier: 410–419.

258

20 IMPLICIT KNOWLEDGE IN PRAGMATIC INFERENCE Chris Cummins and Albertyna Paciorek

Introduction Much of linguistic pragmatics deals with the transmission of meanings that go beyond what is explicitly conveyed by the speaker. In some cases, the notion of these meanings as ‘implicit’ is suggested by the terminology (Grice’s implicatures, Bach’s implicitures) and in some cases it is defined more clearly. For instance, Reboul (2017: 91) discusses “implicit communication” and intends this term this “to be understood strictly as referring to presuppositions and, among implicatures, to conversational implicatures”. She contrasts the meanings conveyed in these ways with those meanings that are communicated explicitly. As a simple example, consider the exchange in (1). (1) A: Will Anne’s parents be at her wedding next month? B: Her mother will be there. B’s utterance in (1) explicitly conveys that Anne’s mother will be at her wedding, and implicitly conveys (by means of a conversational implicature) that Anne’s father won’t be at her wedding (or potentially the slightly weaker meaning that B is not sure that Anne’s father will be at her wedding). It also presupposes than Anne is getting married next month: an overhearer who was not aware that this was the case would be able to discern it from A’s utterance, even though A does not explicitly say such a thing. Hence, this could also be a case of implicit communication in Reboul’s terms. The existence of implicit communication, in this sense, raises a number of questions that have been fundamental to research in pragmatics and its interface with semantics. One is the question of how this works at all: how are hearers able to recover implicit meanings? How did this ability emerge, and what does its existence disclose about the nature of human communication and the cognitive abilities that underpin it? Another question is why we do this: what does it benefit us to use implicit rather than purely explicit communication? Is this fundamentally a matter of efficiency, and a way of addressing the lack of bandwidth of the communicative channel (as argued by Levinson 2000), or is it rather that we achieve specific effects through presenting information implicitly versus explicitly (the point discussed by Reboul 2017)? And how are implicit and explicit content demarcated: for instance, to what extent does B in (1) really explicitly say “Anne’s mother will be at the wedding” rather than just “Her mother will be there”? That is, should DOI: 10.4324/9781003014584-25 259

Chris Cummins and Albertyna Paciorek

such things as the resolution of deixis be understood as part of the process of determining the explicit content, or are these also to be understood as an aspect of implicit content? Strikingly, however, these debates appear largely orthogonal to the notion of implicit cognition targeted by this volume. Crucially, what is meant by implicit communication in the preceding sense is merely that it is does not derive directly from the explicit content of the speech stream. Material that is communicated “implicitly” in this way may nevertheless be explicit in the minds of the discourse participants: indeed, from a Gricean standpoint, conversational implicatures are argued to be wholly explicit in their cognitive character. Grice (1989: 31) includes the following among the necessary criteria for the utterance of p to be said to conversationally implicate q: “the supposition that [the speaker] is aware that, or thinks that, q is required in order to make his saying . . . p . . . consistent with [the presumption that he is obeying the Cooperative Principle]” and “the speaker thinks . . . that it is within the competence of the hearer to work out, or grasp intuitively, that the [previous supposition] is required”. Thus, Grice’s definition of conversational implicature requires that the speaker is explicitly aware of q and the hearer (at least from the speaker’s point of view) is also expected to be explicitly aware of q and able to reason about its status as a candidate explanation of how the utterance p is cooperative. This is compatible with the idea of conversational implicature as a species of non-natural meaning in the sense of Grice (1957), and with our intuitions about B’s behaviour in (1): it seems that B wishes to convey that Anne’s father won’t be at the wedding, just as they wish to convey that Anne’s mother will be, and they have chosen their utterance in the expectation that the hearer A will be able to reconstruct both parts of this communicative intention. We would not expect A to have any difficulty in precisely stating this implicated meaning, or in affirming that B “meant” to convey it; and we would not expect B to have any difficulty in articulating the same implicated content in the form of an assertion if they were obliged to do so. In short, much of what is said in pragmatics about ‘explicit’ or ‘implicit’ communication is not directly relevant to the scope of this volume. So rather than attempting to summarise pragmatic research on implicit communication in that sense, in this chapter we will look briefly at some areas of pragmatics in which notions of implicit cognition – typically couched in other terms – are potentially germane: specifically, metaphor interpretation, speech act recognition, implicature in a Rational Speech Act context, and typicality effects.

Metaphor The question of whether communicative meaning can be implicit, in the deeper sense of the word, distinguishes at least two of the major approaches to metaphor: one that considers figurative language to reside on a continuum with ostensibly literal language use, and one which treats it as a distinctly different phenomenon. Consider (2). (2) Juliet is the sun. (Romeo and Juliet, Act 2 Scene 2) Grice (1967) proposed that Romeo, in uttering (2), has made a trivially false statement, and took this as an instance of a speaker flouting the Quality maxim in order to convey a related implicature, although this account leaves open the question of precisely what the implicature is supposed to be. Glucksberg and Keysar (1990) argued instead that metaphors such as (2) were indeed what they superficially appear to be – statements of class inclusion – but with the caveat that the meaning of the metaphor vehicle (here, the sun) is taken to refer to a different and broader class of entities than it would normally denote. They discuss the example (3), noting that jail belongs to various superordinate categories, including punishments, human-made 260

Implicit Knowledge in Pragmatic Inference

structures, etc., and arguing that the speaker of (3) is conveying that their job belongs to one of these categories (ibid., 7). (3) My job is a jail. An alternative view, explored from a Relevance Theory perspective in much recent work, is that the pragmatic modulation involved in metaphor interpretation is of a piece with the interpretation of much of what we would normally think of as ‘literal’ language (as well as other types of ‘figurative’ language such as hyperbole). As Carston (2002) describes it, we should read (2) as affirming that Juliet is the sun* and (3) as affirming that the speaker’s job is a jail*, where these ‘starred concepts’ are pragmatically modulated versions of the concepts lexically encoded by the words sun and jail respectively. As in other cases of pragmatic modulation, the task of the hearer is to reconstruct the concept intended by the speaker. As discussed by Wilson and Kolaiti (2017), we can see this view as standing in opposition to the position articulated by Davidson (1978: 46), who argued that “the thesis that associated with a metaphor is a cognitive content that its author wishes to convey and that the interpreter must grasp if he is to get the message . . . is false”. On Davidson’s view, termed the “Brute Force” account by Hills (2017), the effect of metaphor is to cause us to consider the primary subject of the metaphor in a new light, causing us to notice apparent truths about it that would otherwise have escaped our attention. This position is adopted and developed by, among others, Lepore and Stone (2010: 165), who propose that “though metaphors can issue in distinctive cognitive and discourse effects, they do so without issuing in metaphorical meaning and truth, and so without metaphorical communication.” From this perspective, talking about “metaphorical meaning” is at best a convenient fiction, as such meanings are epiphenomenal – they are consequences of the reconsideration of the primary subject by the hearer. In paraphrasing metaphorical meaning, we are attempting to recapture some of the conclusions that the hearer might arrive at as a result of being induced to think in this way. And empirical observations about the activation of apparent aspects of metaphorical meaning – often conducted within an RT framework – can be explained away in a similar fashion. For instance, Rubio-Fernández (2007) uses a lexical priming task to demonstrate that metaphor vehicles (e.g. cheetah) facilitate the recognition of literally related words (e.g. cat) presented at offsets of 0ms and 400ms, but not at offsets of 1000ms, and facilitate the recognition of figuratively related words (e.g. fast) at all three offsets. From an RT viewpoint, this demonstrates that literal meanings of metaphorical expressions are rapidly suppressed while the figurative meanings persist; from a Davidsonian viewpoint, this demonstrates that hearers given a metaphorical prompt initially entertain inferences involving all aspects of its meaning but rapidly focus their attention on those which repay more detailed consideration. In the following example, we can see how these two kinds of account posit different relations between the metaphorical utterance (4) and its candidate (partial) interpretation (5). (4) My lawyer is a shark. (5) My lawyer is ruthless and ferocious. On an RT account, a speaker who believes (5) may intend to convey it, and in order to do so they may choose to utter (4) in the expectation that the hearer will be able to interpret that utterance as intending to convey (5). On a Davidsonian account, a speaker cannot utter (4) with the intention of conveying (5): rather, they utter (4) in the belief that considering their lawyer in the aspect of a shark will be a helpful way for the hearer to understand that individual, and 261

Chris Cummins and Albertyna Paciorek

with the expectation of instituting a process of reasoning in the mind of the hearer that will result in an indeterminate set of conclusions being drawn. The knowledge that motivates the speaker to utter (4) might be implicit, on this view: it need not be expressible in propositional form, and indeed its inaccessibility in words might be part of the motivation for the use of metaphor. Similarly, the interpretation induced in the hearer may have implicit content. As Hills (2017) points out, “[t]he interpretation of a metaphor often turns not on properties the secondary subject [vehicle] actually has or even on ones it is believed to have but instead on ones we habitually pretend it to have”. We could perhaps think of the effect of (4), on a Davidsonian view, as encouraging the hearer to think of the speaker’s lawyer in a way that they think about metaphorical sharks, based on their experience of how this metaphor is used in language, rather than basing their reasoning on any of the properties that sharks are literally agreed to have. In the context of an RT account, just as for accounts more directly founded on Gricean principles, one of the main challenges is to explain how the hearer could possibly arrive at the precise meaning intended by the speaker through a series of defensible rational reasoning steps. One way of addressing this challenge is to posit a greater role for implicit knowledge, which might be accommodated within a Davidsonian account. However, the corresponding challenge for such an account is perhaps to explain how the speaker knows that a metaphor will be effective or communicatively appropriate if they do not have a clear sense of the specific effects that it is likely to bring about.

Speech Act Recognition An issue identified in the early days of Speech Act Theory is how we recognise illocutionary force when this is not signalled by sentence type: for instance, how we know that an utterance such as (6) is likely to be a request whereas (7) is not. (6) Could you pass the salt? (7) Could you read when you first went to school? Searle (1969, 1975) noted that utterances such as (6), even though they resemble informationseeking questions like (7) in their sentence type (interrogative), appear to work as requests by questioning the truth of a necessary preparatory condition for the corresponding request to be satisfied. It is necessary that the answer to (6) should be affirmative, on its question interpretation, for the hearer to be able to satisfy a request actually to pass the salt. Consequently, the hearer is entitled to infer that the likely goal underlying the speaker’s utterance of (6) is that the speaker wishes the hearer to pass the salt, and that what they have produced is a pre-request; and the cooperative hearer can short-circuit the process of the hearer actually producing a request and simply fulfil it. This account of the reasoning being undertaken is premised on the understanding of (6) being literal-first, in some sense: the hearer understands the interrogative (6) as a question and then tries to make sense of why the speaker would ask such a question, thus deriving an indirect interpretation under which the speaker is in fact issuing a request. However, there are good reasons to think that this literal-first interpretative step might not be appropriate. First, as pointed out by Gazdar (1981) and Levinson (1983), many speech acts are indirect in this sense, and in some cases it is not obvious what kind of ‘literal’ interpretation they should be considered to have (for instance, is ‘How do you do’ literally a question, or a greeting?). Moreover, in some cases the so-called indirect illocutionary force has implications for the surface form of the utterance: for instance, it is possible to insert the word please into (6) but not into (7). It is also not 262

Implicit Knowledge in Pragmatic Inference

clear that this kind of derivation of an indirect interpretation fits into the timeline of utterance processing. The typical gap between utterances in fluent conversation is too short to accommodate reasoning of this kind if it must be done after first completing a literal interpretation of the utterance (Stivers et al. 2009). Furthermore, it would be reasonable to expect an utterance that begins Could you . . . to have the force of a request, and that expectation could not be underpinned by reasoning about the utterance’s literal force. In computational work on speech acts, as discussed by Jurafsky (2004), such observations as these motivated the emergence of cue-based accounts of speech act recognition, as an alternative to plan-based approaches which reconstruct the type of reasoning proposed by Searle (1975). On a cue-based approach, the hearer (or the system) attempting to recognise an utterance as instantiating a particular kind of speech act does so by exploiting the probabilistic cues provided by the utterance (and its context). These might include syntactic cues, such as the association between imperative sentences and the speech act of requesting; lexical cues, such as the association between the word sorry and the speech act of apology; prosodic cues, such as the association between utterance-final pitch rises and the speech act of questioning (Sag and Liberman 1975; Sadock and Zwicky 1985); and cues based on dialogue structure, such as the association between one utterance being an invitation and the following one being a response to that invitation. Importantly, as we have already seen for the syntactic cue provided by sentence type, these cues are all probabilistic and uncertain in character: the use of an interrogative sentence type in (6) might incline us to think that it is a question, but other factors militate against that interpretation, eventually leading us to the conclusion that it is not. Similarly, the words Could you in (7) might incline us to think that it is a request until we discern otherwise. Although the initial interest in cue-based approaches was principally computational in character, and directed towards the development of better artificial dialogue systems, the idea that human speech act recognition might rely on similar representations has gained traction (as foreseen by Jurafsky 2004). Such an account would help to resolve the problem of turn-taking, as discussed earlier: observationally, consecutive conversational turns are causally related to one another at the level of speech acts, questions being followed by answers, invitations by responses, and so forth (Schegloff and Sacks 1973), yet they follow one another in quick succession with gaps that are often too short to permit speech planning (Stivers et al. 2009). This suggests that dialogue participants must exploit cues in order to form an early expectation about the likely purpose of an interlocutor’s utterance, in order to commence planning a suitable response to it. Such a capability would also explain our sense that encountering a strong cue like Could you or I’m sorry early in an utterance induces us to expect a particular kind of speech act, even before we have heard the rest of the utterance’s content, let alone commenced to reason about its speaker’s deeper agenda. In this sense, we could argue that the shift from a purely plan-based approach to speech act recognition to one that also appeals to cues is one that introduces an appeal to implicit knowledge. While it would be possible to possess explicit knowledge about the availability of certain cues as signposts of particular speech act – and indeed to try to impart that knowledge, as when a parent reminds a child to say please or say sorry – the cue-based account supposes a fineness of detail about the strength of cues that is not accessible to introspection. Knowledge about the relevant cues, and their respective strengths, would generally be acquired by establishing statistical generalisations based on experience, in common with other types of implicit learning (Perruchet and Pacton 2006). On this account, speech act recognition typically proceeds through integrating the information about the likely utterance purpose provided by multiple (sometimes conflicting) cues, below the level of conscious awareness. 263

Chris Cummins and Albertyna Paciorek

Implicature and Rational Speech Acts Recall that the Gricean notion of quantity implicature is one in which the speaker is intentionally conveying the falsity of a stronger proposition by means of the assertion of a weaker one. For instance, the speaker of (8) is understood to convey the same meaning as (9), the difference being that the “but not all” meaning is asserted by (9) but only implicated by (8). We assume here that some has purely existential semantic meaning (that is, the meaning “not all” is not part of the semantics of some).   (8)  I read some of the papers.   (9)  I read some, but not all, of the papers. On a traditional Gricean account, the hearer would be able to recover this implicature by noting that the speaker of (8) could have uttered (10), which would have been more informative, and concluding that the speaker’s refusal to utter (10) is because doing so would violate the maxim of Quality (“Do not say what you believe to be false. Do not say that for which you lack adequate evidence”; Grice 1989: 27), which would represent a more serious violation of cooperativity than merely being underinformative. (10)  I read all of the papers. Under the additional assumption that the speaker is knowledgeable about the truth or falsity of (10), the hearer can infer that the relevant submaxim of Quality is the first one, and that the speaker who utters (8) does so because they believe (10) to be false. Hence, a hearer is entitled to interpret (8) as conveying (9) – and indeed, on Grice’s view, the speaker intends them to arrive at just this conclusion. Accepting that a hearer could reason in this way, the question of whether they actually do so has been a much-discussed one in recent pragmatic research, with several competing processing accounts under discussion. One recent approach to pragmatic interpretation, which has proved influential in the field at large, is the Rational Speech Act (RSA) model (Frank and Goodman 2012), which aims to explain hearer inference using Bayesian cognitive modelling. In an RSA approach, a hearer is modelled as reasoning about a speaker’s choice of utterance in order to reconstruct the goals that led that speaker to produce it. This retains Grice’s insight that language is a type of rational action while dispensing with the idea that its use is best described by a series of maxims. Within this type of model, the quantity implicature from (8) is obtained as follows. In a circumstance in which (10) is true, the speaker is more likely to utter (10) than (8), on the basis that this is a more effective way to guide a (literal) hearer to a correct understanding of the world state. Conversely, if (8) is true but (10) is false, the speaker is much more likely to utter (8) than (10) – in fact, they cannot utter (10) except in error. On encountering the utterance (8), the hearer performs Bayesian inference to establish the likely world state, and arrives at the conclusion that it is more likely to be one in which (10) is false than one in which (10) is true. Note that when we say the speaker is more likely to choose the more effective than the less effective communicative option, the RSA model does not exclude the possibility that the speaker might opt for the less effective communicative option – for instance, saying (8) even though (10) is true. RSA typically encodes this as a soft-max process governed by a rationality parameter.

264

Implicit Knowledge in Pragmatic Inference

Correspondingly, the hearer is not entitled to be certain about the falsity of (10), given the utterance of (8) – they are merely expected to increase the probability that they assign to it. Thus, the analysis of quantity inference via RSA is another case in which the idea of an inferential process that is largely deterministic in character and relies on explicit knowledge – for instance, one that is guided by appeal to Grice’s maxims – is supplanted by one in which the hearer is performing probabilistic inference in order to derive an understanding of the worldstate given the speaker’s utterance. It is debatable whether the RSA-based inference is in fact a quantity implicature, in the traditional sense of the term. If the speaker of (8) intends to convey the falsity of (10), then they fail to do so, according to RSA – they merely invite the hearer to decrease the subjective probability they attach to the truth of (10). On the other hand, if the speaker’s communicative intention in uttering (8) is indeed simply to cause the hearer to doubt (10), we could argue that they are fully successful on an RSA model. In either case, the RSA model does not presume that the (literal) hearer is necessarily reasoning about the speaker’s intention as such, and hence it could be argued that this is not a species of intentional communication, but rather a way of transmitting probabilistic information about the likely world state. Although some attempt can be made to elicit the probability that a hearer assigns to each of the competing possibilities (see Goodman and Stuhlmüller 2013), in principle this information could be implicit for both the speaker and hearer and accessible only in an impressionistic form. In short, the RSA model appears to offer an account of (at least some cases of ) quantity inference that is consistent with a greater reliance on implicit knowledge than would be the case for more traditional accounts that assume hearers to be reasoning about propositions with known truth-values.

Typicality Effects A further form of implicature discussed in the pragmatics literature is the so-called markedness implicature, in which an utterance such as (11) is interpreted as conveying an additional meaning beyond that of (12). (11)  Mary caused the car to stop. (12)  Mary stopped the car. One way of capturing this intuition is to note that the choice to say (11) rather than (12) runs the risk of violating the Gricean maxim of manner: it is needlessly verbose. Hence, the speaker who utters (11) invites the hearer to infer that something additional is meant, for instance that Mary stopped the car in a non-standard way. Correspondingly, a speaker who utters (12) might be presumed to suggest that Mary stopped the car in the standard way, i.e. by pressing the brake pedal while driving it. As Levinson (2000: 31–33) puts it in his formulation of a set of proposed heuristics governing pragmatic interpretation, “What is simply described is stereotypically exemplified. . . . What’s said in an abnormal way isn’t normal”. Typicality effects are potentially quite widespread in communication, because surprisingly many concepts appear to exhibit typicality structure. This is not just the case for classes that are putatively organised around prototypes (Rosch 1975), but also extends to those with crisp membership criteria: for instance, Armstrong et al. (1983) provide evidence that 2 is considered the “most typical” even number. Geurts and Van Tiel (2013) argue that the quantifier some exhibits typicality effects too: in a partitive context, it is judged most appropriate for values in the mid-range, and appreciably

265

Chris Cummins and Albertyna Paciorek

less satisfactory as a description of a situation in which eight out of 10 objects possess the relevant property than one in which five out of 10 do. A possible explanation of this is that rational hearers have experienced the fact that some is less frequently used for proportions in that higher range, where alternatives such as most are available, even though they have not engaged in explicit reasoning about the availability of possible implicatures (some is not generally judged to implicate the negation of most as strongly as it implicates the negation of all). That is, they are using apparently implicit knowledge about the use of these quantifiers in order to constrain their likely interpretation. The acquisition of these distributional preferences might proceed in much the same way as the acquisition of subtle collocational preference that govern, for instance, the use of big versus large (see Paciorek and Williams 2015 for a review). In a similar spirit, Cummins et al. (2012) discuss how expressions such as more than 100 are taken pragmatically to refer to particular ranges of values, with hearers inferring upper bounds on the intended values that are not given semantically. A possible explanation of this is quantity implicature, with hearers inferring the falsity of specific stronger alternatives such as more than 200 or more than 150: however, Hesse and Benz (2020) find evidence that hearers prefer to infer that the true value lies within a certain proportional range of the numeral. They argue that the existence of this pattern and the specific size of the range are due to the way humans represent approximate quantities and the acuity with which they do so. In common with other psychophysical properties of human cognition, this is not accessible to introspection, and could reasonably be considered implicit. This invites the conclusion that either speakers are using implicit knowledge to constrain their production of numerical expressions of quantity, or they are being systematically misunderstood by hearers.

Conclusion Linguistic pragmatics is sometimes considered to be the study of implicit communication, but the relevant sense of ‘implicit’ is typically that information is not directly discernible in the speech stream. In the preceding sections, we briefly discuss four topics in pragmatics for which ongoing theoretical debates can be construed as questions about the status of pragmatic knowledge as implicit or explicit in the sense of this volume. In these four cases – metaphor interpretation, speech act recognition, quantity implicature, and the interpretation of quantity information more generally – this perspective helps us identify a tension between two theoretical approaches. In each case, we discuss one account which holds that speakers convey determinate, introspectively accessible content to hearers, who reconstruct it via a discrete series of reasoning stages; and an alternative that argues instead that the content communicated by the speaker need not be determinate and that the process of meaning (re)construction in the hearer makes use of implicit knowledge, which could be acquired through the extraction of statistical regularities from the hearer’s linguistic experience. Although the details of the alternatives discussed here vary in many important respects across these four topics, the resulting dialectics are perhaps surprisingly similar: the accounts relying on explicit knowledge encounter difficulties in explaining how the precise meanings intended by the speaker could possibly be reconstructed under the conditions of normal interaction, while the accounts relying on implicit knowledge grapple with the challenge of explaining the subjective certainty and introspective accessibility of many of the inferences that originally motivated these strands of pragmatic enquiry.

266

Implicit Knowledge in Pragmatic Inference

Funding Acknowledgement Albertyna Paciorek gratefully acknowledges the funding from the National Science Centre in Poland, grant nr: 2016/21/D/HS2/02493.

Related Topics Chapters 3, 6, 14, 15, 18, 19

References Armstrong, S. L., Gleitman, L. R., and Gleitman, H. 1983. “What some concepts might not be”. Cognition, 13: 263–308. Carston, R. 2002. Thoughts and utterances: the pragmatics of explicit communication. Oxford: Blackwell. Cummins, C., Sauerland, U., and Solt, S. 2012. “Granularity and scalar implicature in numerical expressions”. Linguistics and Philosophy, 35: 135–169. Davidson, D. 1978. “What metaphors mean”. Critical Inquiry, 5: 31–47. Frank, M. C., and Goodman, N. D. 2012. “Predicting pragmatic reasoning in language games”. Science, 336: 998. Gazdar, G. 1981. “Speech act assignment”. In A. K. Joshi, B. Webber, and I. Sag, eds., Elements of discourse understanding. Cambridge: Cambridge University Press: 64–83. Geurts, B., and Van Tiel, B. 2013. “Embedded scalars”. Semantics and Pragmatics, 6: 1–37. Glucksberg, S., and Keysar, B. 1990. “Understanding metaphorical comparisons: beyond similarity”. Psychological Review, 97: 3–18. Goodman, N. D., and Stuhlmüller, A. 2013. “Knowledge and implicature: modeling language understanding as social cognition”. Topics in Cognitive Science, 5: 173–184. Grice, H. P. 1957. “Meaning”. Philosophical Review, 66: 377–388. Grice, H. P. 1967. “Logic and conversation: The William James Lectures, Harvard University”. Published as Part 1 of Grice (1989). Grice, H. P. 1989. Studies in the way of words. Cambridge, MA: Harvard University Press. Hesse, C., and Benz, A. 2020. “Scalar bounds and expected values of comparatively modified numerals”. Journal of Memory and Language, 111: 104068. Hills, D. 2017. “Metaphor”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Fall 2017 edn. https:// plato.stanford.edu/archives/fall2017/entries/metaphor/ Jurafsky, D. 2004. “Pragmatics and computational linguistics”. In L. R. Horn and G. Ward, eds., Handbook of pragmatics. Oxford: Blackwell: 578–604. Lepore, E., and Stone, M. 2010. “Against metaphorical meaning”. Topoi, 29: 165–180. Levinson, S. C. 1983. Pragmatics. Cambridge: Cambridge University Press. Levinson, S. C. 2000. Presumptive meanings: the theory of generalized conversational implicature. Cambridge, MA: MIT Press. Paciorek, A., and Williams, J. N. 2015. “Semantic implicit learning”. In P. Rebuschat, ed., Implicit and explicit learning of languages. Amsterdam: John Benjamins: 69–90. Perruchet, P., and Pacton, S. 2006. “Implicit learning and statistical learning: one phenomenon, two approaches”. Trends in Cognitive Sciences, 10: 233–238. Reboul, A. 2017. “Is implicit communication a way to escape epistemic vigilance?”. In S. Assimakopoulos, ed., Pragmatics at its interfaces. Berlin: De Gruyter: 91–112. Rosch, E. 1975. “Cognitive representations of semantic categories”. Journal of Experimental Psychology: General, 104: 192–233. Rubio-Fernández, P. 2007. “Suppression in metaphor interpretation: differences between meaning selection and meaning construction”. Journal of Semantics, 24: 345–371. Sadock, J. M., and Zwicky, A. M. 1985. “Speech acts distinctions in syntax”. In T. Shopen, ed., Language typology and syntactic description. Cambridge: Cambridge University Press: 155–196. Sag, I. A., and Liberman, M. 1975. “The intonational disambiguation of indirect speech acts”. In R. E. Grossman, L. J. San, and T. J. Vance, eds., Papers from the eleventh regional meeting Chicago Linguistic Society (CLS-75). Chicago: University of Chicago: 487–498.

267

Chris Cummins and Albertyna Paciorek Schegloff, E. A., and Sacks, H. 1973. “Opening up closings”. Semiotica, VIII: 289–327. Searle, J. R. 1969. Speech acts: an essay in the philosophy of language. Cambridge: Cambridge University Press. Searle, J. R. 1975. “Indirect speech acts”. In P. Cole and J. L. Morgan, eds., Syntax and semantics 3: Speech acts. New York: Academic Press: 59–82. Stivers, T., Enfield, N., Brown, P., Englert, C., Hayashi, M., Heinemann, T., Hoymann, G., Rossano, F., De Ruiter, J. P., Yoon, K-E, and Levinson, S. C. 2009. “Universals and cultural variation in turntaking in conversation”. Proceedings of the National Academy of Sciences of the United States of America, 106: 10587–10592. Wilson, D., and Kolaiti, P. 2017. “Lexical pragmatics and implicit communication”. In P. Cap and M. Dynel, eds., Implicitness: from lexis to discourse. Amsterdam: John Benjamins: 147–175.

268

PART 5

Agency and Control

21 IMPLICIT MECHANISMS IN ACTION AND IN THE EXPERIENCE OF AGENCY Sofia Bonicalzi

1. Introduction In the past decades, research on implicit mechanisms has attracted attention both in the philosophy and the cognitive science of action. However, systematic attempts to clarify the nature of the action-relevant implicit mechanisms and the contribution they provide to specific actiontypes are still missing. The chapter aims to provide an overview of selected action-related research topics, notably goal-directedness (section  2), action correction (section  3), and the sense of agency (section 4), where implicit mechanisms have proven productive. The focus will remain on individual actions, although much interest in implicit processes has been sparked by their contribution to social perception and action (Frith and Frith 2008) via automatic imitation (Chartrand and Bargh 1999) and simulation (Gallese 2001). Overall, implicit mechanisms have been referred to in terms of a range of automatic or unconscious processes that run independently of volitional, top-down control and are inaccessible to introspective awareness (Evans 2008; Payne and Gawronski 2010). Whereas those are features of how the information is computed in the brain, implicitness can also characterise the kind of the (stored) information, that is, as non-conscious or non-amenable to verbal reports but still able to influence behaviours (Davies 2001). These features are commonly clustered together, although some researchers have focused more on the automaticity-related (Shiffrin and Schneider 1977) or the consciousness-related aspects (Schacter 1992), or even expressed doubts about whether these aspects consistently co-exist. De Houwer and Moors (2012) treat implicit as a synonym of automatic, while acknowledging that automatic is also a multi-faceted notion, so that implicit processes may be automatic in some respects without being automatic in others. In turn, discussing implicit learning, Cleeremans and Jiménez (2002) introduce implicit as primarily meaning unconscious and defend a multi-dimensional understanding of consciousness. In this respect, a most credited framework distinguishes between phenomenal and access consciousness. Phenomenal consciousness refers to the qualitative aspect of our experience, including the experience of being an agent. Access consciousness corresponds to the ability to use the content of mental states for reasoning, verbal report, and deliberate control of behaviour (Block 1995; Overgaard 2018). Finally, besides signalling the properties of a phenomenon, the notion of implicit is also associated with specific measurement tools, that is, readouts of objective performance parameters, DOI: 10.4324/9781003014584-27 271

Sofia Bonicalzi

used to assess people’s attitudes. Conversely, explicit measures are based on self-reports of past performances (Gawronski and Brannon 2017). Implicit measures are seen as more immune from person-level cognitive biases and can capture attitudes that are inaccessible to introspection or that the subject is unwilling to report (Nisbett and Wilson 1977). As the provided examples will prove, these different senses of implicitness come together insofar as implicit measures (e.g., of speed or reaction time) are used to evaluate whether an action-related process runs implicitly or whether a piece of action-relevant information is implicitly stored (Ramsey 2018). Whenever one embarks in discussions about implicitness, it is therefore vital to specify which of the senses of this multi-faceted notion is at stake. To facilitate discussion, it is also important to remove an ambiguity hanging over actionrelated interdisciplinary debates. Philosophers working on the topic are traditionally mostly interested in (a) the features of intentional actions, often conceived of as a subset of the wider set of voluntary actions, which in turn includes behaviours we can interrupt at will but that are not straightforwardly intention-driven, such as habitual actions (Levy 2013). By contrast, cognitive scientists mainly distinguish (b) between internally generated or goal-directed and externally or sensory-triggered actions (Passingham et  al. 2010). As a result, the very same action, for example, consider “pressing the accelerator when the traffic light is green”, can paradoxically count as intentional under (a) and as externally triggered under (b). To overcome the issue, a non-dichotomous, continuous view (Bonicalzi and Haggard 2019) of action-types – ranging from automatic reflexes to progressively more complex, intelligent, and flexible behaviours – is adopted throughout the chapter.

2. Goal-Directedness One of the key features of intentional and voluntary actions is their being goal-directed. In classic action theory, the goal of an intentional action is the explicit content of the conscious intention or plan to act (Bratman 2007). Although the intention or plan itself might be more or less deliberately or consciously formed (Holton 2009), the subject is expected to be able to gain awareness of its content and motivating factors (Mele 2010). The action-guiding role of the intention then secures control, which is achieved when the action matches the representational content of the intention (Shepherd 2014). Given this quintessentially explicit view of goal-directedness, in what sense might implicit mechanisms, as types of processes or knowledge, play a role? Building on De Houwer and Moors’ (2012) implicitness as automaticity, if a behaviour is automatic, then controllability might be lost, in the sense that the behaviour lacks purposefulness and occurs (a) in the absence of a goal, or (b) in contrast with a goal, or (c) independently of a goal. Tics and reflexes, which can or cannot be inhibited (Ganos 2016), are examples of (a) and (b). Deviant causal chains, identifying situations where a subject accidentally achieves the very same goal that they were planning to achieve (Schlosser 2010), work as an example of (c). Alternatively, understanding implicit as unconscious, behaviours can remain purposeful without their goals’ being amenable to introspective awareness. Key examples of actions that subserve goals without being explicitly goal-directed are habitual and skilled actions. In cognitive science, habitual and goal-directed actions are often contrasted based on their behavioural outputs (Dickinson 1985), learning processes (Keramati et al. 2011), neural realisers (Dolan and Dayan 2013), and computational mechanisms (Keramati et al. 2016). The ability to engage in habitual behaviour, computationally less demanding but also less flexible than goal-directed behaviour (Adams and Dickinson 1981), is recognised as fundamental to free up useful cognitive resources (Daw 2015). 272

Implicit Mechanisms in Action and in the Experience of Agency

A fertile field of research investigates how the goal-directed and the habitual-system cooperate to originate behaviours, how the smooth shift between the two takes place within the same action sequence, and how goal-directed behaviour becomes routinised over time (Dezfouli and Balleine 2013). Indeed, research has shown that habitual actions can be eventually triggered by environmental cues (Shiffrin and Schneider 1977), even contrary to conscious motivation (Neal et al. 2011), but can also be activated, within the same causal sequence, by conscious processes (Bargh 1994). For example, producing conscious implementation intentions earlier on increases the chance of subsequently engaging in the corresponding automatic behaviour, in response to situational cues (Gollwitzer 2003). Skilled actions, particularly in their motor aspects, can be thought of as types of habitual behaviours that become perfected via repetition and are maintained through exercise. An expert player will be aware of the general goals of the action, while possibly lacking both conscious awareness of the sub-goals enabling action performance and discursive knowledge of the implementation details. In discussing skilled actions, one must avoid caricature in introducing too simplistic dichotomies between flexible and intelligent cognition (knowing that) as opposed to fixed and bottom-up motor execution (knowing how) (Fridland 2016), but the highly automatised nature of skilled actions is what often permits fast and efficient execution (Shepherd 2017). In particular, the high speed achieved during some sporting skills might be incompatible with top-down, conscious control (Papineau 2013). Behavioural evidence in this sense is provided by findings indicating disruptions in speed, smoothness, and accuracy when expert performers are required to make conscious changes to habitual behavioural patterns (Beilock and Carr 2001; Logan and Crump 2010). Martens and Roelofs (2019) apply the notion of implicit intentions to habitual and skilled actions. Initially explicitly intentional, once learned these actions become weakly intentional: they have identifiable aims and incorporate a representation of reality, although they progressively operate through implicit intentions, having an impact on behaviour without being available for reflection and integration with conscious mental states. Implicit intentions remain, however, accessible to the extent that one is usually able to bring them back to consciousness when needed. Correspondingly, one of the challenges in the study of attentional control is to explain how consciousness may resume a guiding role, should circumstances so require (Norman and Shallice 1986). Even when actions are goal-directed, not all their aspects and sub-goals must be guided by explicit intentions. Within the same causal sequence, types of intentions can be classified based on their temporal, content-based, or functional features. For example, Gollwitzer (2003) distinguishes between a deliberative and an implemental mindset and between more abstract goal intentions and more practical implementation intentions, including situational details; Searle (1983) between planned and unplanned actions based on whether they are guided also by prior intentions (future-oriented plans) or only by intentions in action (the intentional content of the action itself). Examples of the second type are spontaneous behaviours (e.g., suddenly hitting somebody) and subsidiary actions that are part of more complex chains of behaviours (e.g., all the steps one must get through to get home). Different cognitive models of action control have attempted to explain how more abstract and more context-dependent aspects of action planning and implementation come together, generally accepting the basic distinction between abstract and sensorimotor intentions (Pockett 2006), with the former originating in the dorsolateral prefrontal cortex (Jahanshahi et al. 2001) and the pre-supplementary motor area (Lau et al. 2004), and the latter in the posterior parietal cortex (Andersen and Buneo 2002). In particular, cascade models propose that action control is made possible by a cascade of top-down control from rostral to caudal LPFC and premotor 273

Sofia Bonicalzi

regions, with specialised areas devoted to more abstract or more context-dependent aspects of the action (Koechlin and Summerfield 2007). These more fine-grained, context-dependent aspects may rely on motor intentions or representations, that is, non-necessarily conscious or propositional states representing the outcome of the action and the kinematic properties of the movement (Pacherie 2008). Evidence that covert aspects of the goal, including the details of the movement, may remain outside conscious awareness is provided by tasks in which participants have to reach and grasp visual targets under time pressure. Related findings suggest that subjects become consciously aware of the target appearance after the movement has already started (Castiello et al. 1991). Longer reaction times, when no time pressure is applied, may have the function to keep the motor action synchronised with the subjective experience of the target (Jeannerod 2006). The key theoretical question thus concerns how abstract intentions, as propositional states with semantic content, may interlock with sensorimotor intentions (Butterfill and Sinigaglia 2014; Mylopoulos and Pacherie 2017). What has been said so far is meant to account for how purposeful action sequences unfold via the integration of implicit and explicit processes and knowledge. By contrast, more radical views erode the notion of goal-directedness, by claiming that people’s apparently rational behaviours are governed by implicit, automatic processes above which they have no or little conscious control (Doris 2015). On the one hand, a range of now classic experiments shows that people’s behaviours can be easily manipulated via automatic environment-perception-behaviour links that bypass awareness and are not amenable to introspective reflection (Bargh and Chartland 1999). On the other hand, this very same literature, jointly with findings on the unconscious neural antecedents of conscious will (Libet et al. 1983) and on illusions of control (Wegner 2002) has been used to support epiphenomenalism about the causal role of conscious mental states. Overall, epiphenomenalism, as a metaphysical thesis, is the view that conscious mental states, and their neural realisers, cannot be among the causes of physical states, including bodily actions (Bonicalzi 2020; Bonicalzi and De Caro 2022; Nahmias 2014). At the empirical level, illusions of control highlight that people’s reports about their having consciously authored given actions are often untrustworthy. In typical experiments, participants are tricked into believing that they have authored an action or produced an effect, which was in fact the work of a confederate. Illusions of control are generated inferentially, in virtue of a mistakenly perceived association between a prior mental state, the action, and the effect (Wegner 2002). The action-effect repeated bidirectional pairing is central also to classic ideomotor theories of actions, arguing that, once this pairing has been established, the anticipation of future consequences can be sufficient to directly trigger the action (Shin et al. 2010). However, epiphenomenalists leverage on people’s practice of making a posteriori inferences about the causes of behaviours to argue that most or all actions are driven by implicit processes, not available to introspection. Conceptually problematic assumptions, based on the questionable claim that all actions must in principle fall under the same category, concern the generalisability of this conclusion to all types and instances of behaviours (Mele 2018). Acknowledging that people fall prey to recurring illusions of control, however, neither justifies the metaphysical claim that conscious mental states are never among the causes of behaviours nor the empirical claim that all actions are exclusively caused by implicit processes.

3.  Error Detection, Adjustments, and Corrections Goals may run through actions in a more or less explicit manner. A different question concerns the role of implicit mechanisms in motor action correction and error detection, which requires incorporating new information and feedback from proprioception and the sensory systems, and making predictions about the consequences of motor commands (Archambault et al. 2015). 274

Implicit Mechanisms in Action and in the Experience of Agency

Existing evidence suggests that people are scarcely aware of the mechanisms underlying action readjustments, often occurring without diverting conscious attention and in the absence of top-down control. Evidence from reaching and grasping tasks shows that, especially under time pressure, people become aware of changes in the target after the correction has already been initiated and without them being aware of the time discrepancy (Castiello et al. 1991). When adjustments are small, they may even occur without participants ever becoming aware of the correction, as shown by their implicit and explicit reactions to environmental perturbations. For example, in a study by Fourneret and Jeannerod (1998), participants had to draw a line moving a stylus on a graphic tablet. Participants could follow the stylus trajectory by monitoring a cursor on the screen, without directly seeing their arm movements. Unbeknownst to them, the trajectory of the cursor was sometimes slightly distorted, forcing them to deviate their arm movements 10° to the left to maintain the cursor on a straight line. The corrections were effectively made without participants verbally reporting the deviations and the corresponding adjustments. Research on patients with blindsight – whose case is famously discussed by Block to tease phenomenal and access consciousness apart (1995) – has been extensively used to study how implicit knowledge can affect overt behaviour. In particular, people with blindsight, who are cortically blind, are usually able to make correct perceptual judgments and visuomotor actions in response to cues presented in the blind visual field (Prentiss et al. 2018). Several studies have highlighted that these patients are also able to correct ongoing movements in response to changes in the target while remaining unable to consciously discriminate it (Perenin and Rossetti 1996). Once again, a key issue consists in understanding how, in healthy individuals, conscious awareness can be brought back when needed, as it happens, for example, when the magnitude of the distortion crosses a threshold. Troubleshooting is indeed indicated by Norman and Shallice as one of the tasks where automatic action schema might be insufficient and lead to mistakes, thus requiring the intervention of deliberate attentional resources (1986). The modulation of automatic processes with conscious attention thus suggests that the contours of automaticity cannot be so easily demarcated (Tzelgov et al. 1997). Errors might depend on unexpected changes in the target but also on poor performance. Especially under time pressure, automatic responses to errors remain difficult to suppress even when participants are explicitly instructed to avoid correction (Rabbitt 1966). Key open issues concern whether errors need to be consciously detected to prompt corrections (Hommel et al. 2016) and what brain mechanisms underlie conscious and unconscious error detection. The error-related negativity (ERN) – an event-related brain potential reflecting activity in the anterior cingulate cortex (ACC) and whose amplitude depends on whether the action counted as a success or a failure (Gehring et al. 1993), and possibly on mismatches between intended and actual actions (Steinhauser and Yeung 2010) – is a widely studied implicit measure of sensitivity to errors. The ERN is one of the most investigated ERP components, but its connection with conscious awareness of mistakes remains debated (Wessel 2012), with some studies providing evidence that it is associated with both conscious and unconscious error detection (Nieuwenhuis et al. 2001), and others suggesting that it is exclusively linked with conscious error detection (Scheffers and Coles 2000). More generally, existing research suggests that distinct performance monitoring processes underlie conscious and unconscious error detection so that the impairment of one system does not necessarily have an impact on the other. One example is provided by schizophrenic patients presenting deficits in conscious error detection, that is, low ERN amplitude in response to mistakes, without this having an impact on unconscious error detection (Charles et al. 2017). 275

Sofia Bonicalzi

4.  The Sense of Agency Besides considering human action from a third-party perspective, due attention in the literature has been paid to agency from the first-person point of view. The phenomenological experience of being an agent is related to perceiving oneself as the source of the action, and as deliberately, purposefully, and freely acting (Gallagher 2000). The scientific literature emphasises, in particular, the control component of the sense of agency, identifying the sense of agency as the experience of being in control of our voluntary actions and, through them, of their consequences in the external world (Haggard 2017). The sense of agency is often further segmented into a non-conceptual, pre-reflective feeling and a conceptual, explicit judgment of agency, with the two being mediated by different processing systems (Synofzik et al. 2008) No conclusive explanation of the underlying mechanisms of the sense of agency has been provided (Grünbaum 2015), but existing research suggests that it derives from a combination of prospective/pre-motor and retrospective signals (see Pacherie 2008; Bayne 2011; Shepherd 2016; Mylopoulos and Shepherd 2020). Whereas the latter are linked to post hoc reconstructions (Wegner 2002), the former might crucially involve a sub-personal comparison between an internal forward model, based on the efference copy of the motor command, and the actual action effect. Classic comparator models suggest that the feeling of agency arises when the implicit prediction of the sensory consequences of the action matches (Blakemore et al. 1998), or does not deviate too much from (Farrer et al. 2003), the actual sensations (but see Mylopoulos (2015) for a criticism of the sense of agency as sensory awareness, and Mylopoulos (2017) for an alternative account). Specific pathologies, including anarchic hand syndrome or schizophrenia (Garbarini et al. 2016), may lead to disruptions in the sense of agency via deficits in such comparator mechanisms (Shergill et al. 2005). The sense of agency can be quantified by means of explicit and implicit measures. Explicit measures rely on the explicit recollection of the judgment of agency associated with past performances (Sirigu et al. 1999). Implicit measures are readouts of behavioural and neurophysiological parameters interpreted as proxies of the feeling of agency (for a criticism of the connection between given implicit measures and the sense of agency, see Mylopoulos (2012)). Most used laboratory-based implicit measures include the intentional binding effect (Haggard et al. 2002) and sensory attenuation in self-touch (Pyasik et al. 2019), as well as electrophysiological measures indicating modulations in the ERP components associated with action monitoring and control (Beyer et al. 2017). In particular, the intentional binding effect consists in the shortening, in subjective time perception, of the temporal interval between voluntary actions (e.g., a keypress) and their effects (e.g., a tone), compared to analogous time intervals following passive movements. Shorter time intervals are interpreted as indicating a stronger sense of agency. The binding relies on explicit reports of temporal intervals, but the measure per se is implicit since naive participants are expected to be unaware of what is truly measured, of the relation between the sense of agency and the temporal dimension, and of the systematic shortening of the perceived time interval co-occurring with voluntary actions. As in the case of the sense of agency, a clear understanding of the mechanisms underlying the binding is missing (David et al. 2008), and the effect elicits several questions about the relationship between implicit and explicit measures, and between implicit measures and their targets (Moore et al. 2012). Indeed, the simple co-occurrence of the binding with voluntary actions does not allow one to infer that, when the binding occurs, participants also have explicit feelings or beliefs concerning their agency. On the one hand, the linkage can be indirectly reconstructed by observing that factors heightening the explicit sense of agency (Chambon et al. 2013) – e.g., choosing 276

Implicit Mechanisms in Action and in the Experience of Agency

between multiple and motivationally significant options (Barlas et al. 2018) – have an analogous boosting effect on the implicit sense of agency. On the other hand, similar binding patterns have been observed when physical causality, but not intentionality, is involved (Suzuki et al. 2019). Implicit knowledge can be also deployed to effectively modulate the sense of agency. An example of implicit modulation of explicit judgments of agency is provided by an experiment by Wenke and colleagues (2010). Using a visual reaction-time task, the experimenters subliminally primed participants with quickly presented arrows pointing towards the right or left. Following the priming, participants responded to an overt cue by selecting between a right/ left keypress (cued trials) or freely decided between the right/left keypress (free trials), in both cases causing the appearance of one colour out of a six-colour palette. The displayed colour was independent of the prime and the keypress but reflected the compatibility/incompatibility between them. Participants had then to evaluate their explicit sense of control relative to each colour. In both cued and free trials, participants were faster in making prime-compatible keypresses, feeling also more in control over the corresponding colour compared to colours associated with prime-incompatible keypresses. This suggests that prime-compatible, conflictfree choices incorporating facilitatory implicit knowledge are conducive to an increased explicit sense of control. In healthy subjects, aspects of the feeling of agency are modulated by the specific actiontypes, for example the feeling of doing something purposive is more or less salient depending on whether the action is preceded by deliberation rather than being routinely performed. Horgan and colleagues suggest that the action may be perceived as purposive, while the specific goal or purpose is not explicitly conscious (e.g., in habitual actions), or is not even accessible to consciousness (e.g., in fast-paced skilled actions) (2003). By contrast, the phenomenology of agency becomes vivid when the sense of control is suddenly lost, due to unexpected mismatches between prior expectations and outcomes (Haggard 2017). Implicit learning can nonetheless be used to modify the sense of agency associated with a given action-type. For example, Khalighinejad and Haggard (2015) show the effect of associative learning in extending the sense of agency typical of voluntary actions to involuntary ones. In their experiment, participants made repeated voluntary keypresses followed by a tone with one hand, while a robotic arm induced kinematically analogous movements from their other hand. Following this repeated and systematic pairing, the intentional binding effect usually associated with voluntary actions was observed also in association with involuntary movements.

5. Conclusions Without any pretension of being exhaustive, this chapter aimed to provide an overview of how and why implicit processes matter to individual action and agency, a fertile research area at the intersection between the philosophy and the cognitive science of action. Overall, the possibility to outsource tasks to implicit mechanisms is recognised as a fundamental cog in human action. As discussed, however, the notion of implicitness remains unsatisfactorily vague if it is not adequately conjugated with a detailed analysis of the sense in which a given mechanism is meant to be implicit. The chapter discusses implicitness with reference to information processing, information kind, and measures of people’s performances as applied to three action-related domains, notably goal-directedness, action correction, and the sense of agency.1

Related Topics Chapters: 6, 14, 16, 17, 23, 24, 277

Sofia Bonicalzi

Notes 1 S.B. benefitted from the PRIN grant 20175YZ855 from the Italian government and from a fellowship at the Paris Institute for Advanced Study (France), with the financial support of the French State, programme “Investissements d’avenir” managed by the Agence Nationale de la Recherche (ANR-11LABX-0027–01 Labex RFIEA+).

References Adams, C. D., and Dickinson, A. 1981. “Instrumental responding following reinforcer devaluation”. Quarterly Journal of Experimental Psychology [B], 33: 109–121. Andersen, R. A., and Buneo, C. A. 2002. “Intentional maps in posterior parietal cortex”. Annual Review of Neuroscience, 25: 189–220. Archambault, P. S., Ferrari-Toniolo, S., Caminiti, R., and Battaglia-Mayer, A. 2015. “Visually-guided correction of hand reaching movements: the neurophysiological bases in the cerebral cortex”. Vision Research, 110: 244–256. Bargh, J. A. 1994. “The four horsemen of automaticity: awareness, intention, efficiency, and control in social cognition”. In R. S. Wyer and T. K. Srull, eds., Handbook of social cognition, Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.: 1–40. Bargh, J. A., and Chartland, T. L. 1999. “The unbearable automaticity of being”. American Psychologist, 54: 462–479. Barlas, Z., Hockley, W. E., and Obhi, S. S. 2018. “Effects of free choice and outcome valence on the sense of agency: evidence from measures of intentional binding and feelings of control”. Experimental Brain Research, 236: 129–139. Bayne, T. 2011. “The sense of agency”. In F. Macpherson, ed., The senses. Oxford: Oxford University Press: 490–524. Beilock, S., and Carr, T. 2001. “On the fragility of skilled performance: what governs choking under pressure”. Journal of Experimental Psychology, 130: 701–725. Beyer, F., Sidarus, N., Bonicalzi, S., and Haggard, P. 2017. “Beyond self-serving bias: diffusion of responsibility reduces sense of agency and outcome monitoring”. Social Cognitive and Affective Neuroscience, 11: 138–145. Blakemore, S. J., Wolpert, D. M., and Frith, C. D. 1998. “Central cancellation of self produced tickle sensation”. Nature Neuroscience, 1: 635–640. Block, N. 1995. “On a confusion about the function of consciousness”. Behavioral and Brain Sciences, 18: 227–247. Bonicalzi, S. 2020. “Actions and intentions”. In B. Feltz, M. Missal, and A. C. Sims, eds., Free will, causality and neuroscience. Leiden: Brill – Rodopi: 120–142. Bonicalzi, S., and De Caro, M. 2022. “How the Libet tradition can contribute to understanding human action rather than free will”. In C. Austin, A. Marmodoro, and A. Roselli, eds., Free will and the laws of nature, Synthese Library 451, Cham: Springer: 199–225. Bonicalzi, S., and Haggard, P. 2019. “From freedom from to freedom to: new perspectives on intentional action”. Frontiers in Psychology, 10: 1193. Bratman, M. E. 2007. Structures of agency: essays. Oxford: Oxford University Press. Butterfill, S. A., and Sinigaglia, C. 2014. “Intention and motor representation in purposive action”. Philosophy and Phenomenological Research, 88: 119–145. Castiello, U., Paulignan, Y., and Jeannerod, M. 1991. “Temporal dissociation of motor responses and subjective awareness: a study in normal subjects”. Brain, 114: 2639–2655. Chambon, V., Wenke, D., Fleming, S. M., Prinz, W., and Haggard, P. 2013. “An online neural substrate for sense of agency”. Cerebral Cortex, 23: 1031–1037. Charles, L., Gaillard, R., Amado, I., Krebs, M., and Dehaene, S. 2017. “Conscious and unconscious performance monitoring: evidence from patients with schizophrenia”. Neuroimage, 144: 153–166. Chartrand, T. L., and Bargh, J. A. 1999. “The chameleon effect: the perception-behavior link and social interaction”. Journal of Personality and Social Psychology, 76: 893–910. Cleeremans, A., and Jiménez, L. 2002. “Implicit learning and consciousness: a graded, dynamic Perspective”. In A. Cleeremans and R. French, eds., Implicit learning and consciousness: an empirical, philosophical and computational consensus in the making. New York: Routledge: 1–40. David, N., Newen, A., and Vogeley, K. 2008. “The ‘sense of agency’and its underlying cognitive and neural mechanisms”. Consciousness and Cognition, 17: 523–534.

278

Implicit Mechanisms in Action and in the Experience of Agency Davies, M. 2001. “Knowledge (explicit and implicit): philosophical aspects”. In N. Smelser and P. Baltes, eds., International encyclopedia of social and behavioral sciences. Amsterdam: Elsevier: 8126–8132. Daw, N. D. 2015. “Of goals and habits”. PNAS, 112: 13749. De Houwer, J., and Moors, A. 2012. “How to define and examine implicit processes?”. In R. W. Proctor and E. J. Capaldi, eds., Psychology of science: implicit and explicit processes. New York: Oxford University Press: 183–198. Dezfouli, A., and Balleine, B. W. 2013. “Actions, action sequences and habits: evidence that goal-directed and habitual action control are hierarchically organized”. PloS Computational Biology, 9: e1003364. Dickinson, A. 1985. “Actions and habits: the development of behavioural autonomy”. Philosophical Transactions of the Royal Society B, 308: 67–78. Dolan, R. J., and Dayan, P. 2013. “Goals and habits in the brain”. Neuron, 80: 312–325. Doris, J. M. 2015. Talking to our selves: reflection, ignorance, and agency. New York: Oxford University Press. Evans, J. S. 2008. “Dual-processing accounts of reasoning, judgment, and social cognition”. Annual Review of Psychology, 59: 255–278. Farrer, C., Franck, N., Georgieff, N., Frith, C. D., Decety, J., and Jeannerod, M. 2003. “Modulating the experience of agency: a positron emission tomography study”. Neuroimage, 18: 324–333. Fourneret, P., and Jeannerod, M. 1998. “Limited conscious monitoring of motor performance in normal subjects”. Neuropsychologia, 36: 1133–1140. Fridland, E. 2016. “Skill and motor control: intelligence all the way down”. Philosophical Studies, 174: 1–22. Frith, C. D., and Frith, U. 2008. “Implicit and explicit processes in social cognition”. Neuron, 60: 503–510. Gallagher, S. 2000. “Philosophical conceptions of the self: implications for cognitive science”. Trends in Cognitive Sciences, 4: 14–21. Gallese, V. 2001. “The ‘shared manifold’ hypothesis: from mirror neurons to empathy”. In E. Thompson, ed., Between ourselves: second-person issues in the study of consciousness. Thorverton: Imprint Academic: 33–50. Ganos, C. 2016. “Tics and Tourette’s: update on pathophysiology and tic control”. Current Opinion in Neurology, 29: 513–518. Garbarini, F., Mastropasqua, A., Sigaudo, M., Rabuffetti, M., Piedimonte, A., Pia, L., and Rocca, P. 2016. “Abnormal sense of agency in patients with schizophrenia: evidence from bimanual coupling Paradigm”. Frontiers in Behavioural Neuroscience, 10: 1–10. Gawronski, B., and Brannon, S. 2017. “Attitudes and the implicit-explicit dualism”. In D. Albarracín, and B. T. Johnson, eds., The handbook of attitudes, vol. 1: Basic principles. 2nd ed. New York: Routledge: 158–196. Gehring, W. J., Goss, B., Coles, M. G., Meyer, D. E., and Donchin, E. 1993. “A neural system for errordetection and compensation”. Psychological Science, 4: 385–390. Gollwitzer, P. 2003. “Why we thought the action mindsets affect illusions of control”. Psychological Inquiry, 14: 259–267. Grünbaum, T. 2015. “The feeling of agency hypothesis: a critique”. Synthese, 192: 3313–3337. Haggard, P. 2017. “Sense of agency in the human brain”. Nature Reviews Neuroscience, 18: 196-207. Haggard, P., Clark, S., and Kalogeras, J. 2002. “Voluntary action and conscious awareness”. Nature Neuroscience, 5: 382–385. Holton, R. 2009. Willing, wanting, waiting. New York: Oxford University Press. Hommel, B., Brown, S. B., and Nattkemper, D. 2016. Human action control: from intentions to movements. Cham, Switzerland: Springer International Publishing. Horgan, T., Tienson, J., and Graham, G. 2003. “The phenomenology of first-person agency”. In S. Walter and H. D. Heckmann, eds., Physicalism and mental causation: the metaphysics of mind and action. Exeter: Imprint Academic: 323–340. Jahanshahi, M., Dirnberger, G., Liasis, A., Towell, A., and Boyd, S. 2001. “Does the pre-frontal cortex contribute to movement-related potentials? Recordings from subdural electrodes”. Neurocase, 7: 495–501. Jeannerod, M. 2006. “Consciousness of action as an embodied consciousness”. In S. Pockett, W. P. Banks, and S. Gallagher, eds., Does consciousness cause behavior? Cambridge, MA: MIT Press: 25–38. Keramati, M., Dezfouli, A., and Piray, P. 2011. “Speed/accuracy trade-off between the habitual and the goal-directed processes”. PLoS Computational Biology, 7: e1002055. Keramati, M., Smittenaar, P., Dolan, R. J., and Dayan, P. 2016. “Adaptive integration of habits into depthlimited planning defines a habitual-goal – directed spectrum”. PNAS, 113: 12868–12873. Khalighinejad, N., and Haggard, P.  2015. “Extending experiences of voluntary action by association”. PNAS, 113: 8867–8872. Koechlin, E., and Summerfield, C. 2007. “An information theoretical approach to prefrontal executive function”. Trends in Cognitive Sciences, 11: 229–235.

279

Sofia Bonicalzi Lau, H. C., Rogers, R. D., Haggard, P., and Passingham, R. E. 2004. “Attention to intention”. Science, 303: 1208–1210. Levy, Y. 2013. “Intentional action first”. The Australasian Journal of Philosophy, 91: 705–718. Libet, B., Gleason, C. A., Wright, E. W., and Pearl, D. K. 1983. “Time of conscious intention to act in relation to onset of cerebral activity (readiness-potential): the unconscious initiation of a freely voluntary act”. Brain, 106: 623–642. Logan, G. D., and Crump, M. J. 2010. “Cognitive illusions of authorship reveal hierarchical error detection in skilled typists”. Science, 330: 683–686. Martens, J., and Roelofs, L. 2019. “Implicit coordination: acting quasi-jointly on implicit shared intentions”. Journal of Social Ontology, 4: 93–120. Mele, A. R. 2010. Effective intentions: the power of conscious will. New York: Oxford University Press. Mele, A. R. 2018. “Free will, moral responsibility, and scientific epiphenomenalism”. Frontiers in Psychology, 9: 2536. Moore, J. W., Middleton, D., Haggard, P., and Fletcher, P. C. 2012. “Exploring implicit and explicit aspects of sense of agency”. Consciousness and Cognition, 21: 1748–1753. Mylopoulos, M. 2012. “Evaluating the case for the low-level approach to agentive awareness”. Philosophical Topics, 40: 103–127. Mylopoulos, M. 2015. “Agentive awareness is not sensory awareness”. Philosophical Studies, 172: 761–780. Mylopoulos, M. 2017. “A cognitive account of agentive awareness”. Mind and Language, 32: 545-563. Mylopoulos, M., and Pacherie, E. 2017. “Intentions and motor representations: the interface challenge”. Review of Philosophy and Psychology, 8: 317–336. Mylopoulos, M., and Shepherd, J. 2020. “The experience of agency”. In U. Kriegel, ed., The Oxford handbook of the philosophy of consciousness. Oxford: Oxford University Press. Nahmias, E. 2014. “Is free will an illusion? Confronting challenges from the modern mind sciences”. In W. Sinnott-Armstrong, ed., Moral psychology, vol. 4: Freedom and responsibility. Cambridge, MA: MIT Press: 1–25. Neal, D. T., Wood, W., Wu, M., and Kurlander, D. 2011. “The pull of the past: when do habits persist despite conflict with motives?”. Personality and Social Psychology Bulletin, 37: 1428–1437. Nieuwenhuis, S., Ridderinkhof, K. R., Blom, J., Band, G. P., and Kok, A. 2001. “Error-related brain potentials are differentially related to awareness of response errors: evidence from an antisaccade task”. Psychophysiology, 38: 752–760. Nisbett, R. E., and Wilson, T. E. 1977. “Telling more than we can know: verbal reports on mental processes”. Psychological Review, 84: 231–259. Norman, D. A., and Shallice, T. 1986. “Attention to action: willed and automatic control of behavior”. In R. J. Davidson, G. E. Schwartz, and D. Shapiro, eds., Consciousness and self regulation: advances in research. Vol. IV. New York: Plenum Press: 1–18. Overgaard, M. 2018. “Phenomenal consciousness and cognitive access”. Philosophical Transactions of the Royal Society B, 373: 1–6. Pacherie, E. 2008. “The phenomenology of action: a conceptual framework”. Cognition, 107: 179–217. Papineau, D. 2013. “In the zone”. Royal Institute of Philosophy Supplement, 73: 175–196. Passingham, R. E., Bengtsson, S. L., and Lau, H. C. 2010. “Medial frontal cortex: from self-generated action to reflection on one’s own performance”. Trends in Cognitive Sciences, 14: 16–21. Payne, B., and Gawronski, B. 2010. “A  history of implicit social cognition: where is it coming from? Where is it now? Where is it going?”. In B. Gawronski and B. Payne, eds., Handbook of implicit social cognition: measurement, theory, and applications. New York: Guilford Press: 1–17. Perenin, M. T., and Rossetti, Y. 1996. “Grasping without form discrimination in a hemianopic Field”. Neuroreport, 7: 793–797. Pockett, S. 2006. “The neuroscience of movement”. In S. Pockett, W. P. Banks, and S. Gallagher, eds., Does consciousness cause behavior?, Cambridge, MA: MIT Press: 9–24. Prentiss, E. K., Schneider, C. L., Williams, Z. R., Sahin, B., and Mahon, B. Z. 2018. “Spontaneous Inflight accommodation of hand orientation to unseen grasp targets: a case of action blindsight”. Cognitive neuropsychology, 35: 343–351. Pyasik, M., Salatino, A., Burin, D., Berti, A., Ricci, R., and Pia, L. 2019. “Shared neurocognitive mechanisms of attenuating self-touch and illusory self-touch”. Social Cognitive and Affective Neuroscience, 1: 1–9. Rabbitt, P. 1966. “Errors and error correction in choice-response tasks”. Journal of Experimental Psychology, 71: 264–272.

280

Implicit Mechanisms in Action and in the Experience of Agency Ramsey, R. 2018. “What are reaction time indices of automatic imitation measuring?”. Consciousness and Cognition, 65: 240–254. Schacter, D. L. 1992. “Implicit knowledge: new perspectives on unconscious processes”. PNAS, 89: 11113–11117. Scheffers, M. K., and Coles, M. G. 2000. “Performance monitoring in a confusing world: error-related brain activity, judgments of response accuracy, and types of errors”. Journal of Experimental Psychology. Human perception and performance, 26: 141–151. Schlosser, M. E. 2010. “Bending it like beckham: movement, control and deviant causal chains”. Analysis, 70: 299–303. Searle, J. R. 1983. Intentionality: an essay in the philosophy of mind. Cambridge: Cambridge University Press. Shepherd, J. 2014. “The contours of control”. Philosophical Studies, 170: 395–411. Shepherd, J. 2016. “Conscious action/zombie action”. Noûs, 50: 419–444. Shepherd, J. 2017. “Skilled action and the double life of intention”. Philosophy and Phenomenological Research, 170: 395–411. Shergill, S. S., Samson, G., Bays, P. M., Frith, C. D., and Wolpert, D. M. 2005. “Evidence for sensory prediction deficits in schizophrenia”. American Journal of Psychiatry, 162: 2384–2386. Shiffrin, R. M., and Schneider, W. 1977. “Controlled and automatic human information processing: II. Perceptual learning, automatic attending and a general theory”. Psychological Review, 84: 127-190. Shin, Y. K., Proctor, R. W., and Capaldi, E. J. 2010. “A review of contemporary ideomotor theory”. Psychological Bulletin, 136: 943–974. Sirigu, A., Daprati, E., Pradat-Diehl, P., Franck, N., and Jeannerod, M. 1999. “Perception of self-generated movement following left parietal lesion”. Brain, 122: 1867–1874. Steinhauser, M., and Yeung, N. 2010. “Decision processes in human performance monitoring”. Journal of Neuroscience, 30: 15643–15653. Suzuki, K., Lush, P., Seth, A. K., and Roseboom, W. 2019. “Intentional binding without intentional action”. Psychological Science, 30: 842–853. Synofzik, M., Vosgerau, G., and Newen, A. 2008. “Beyond the comparator model: a multifactorial twostep account of agency”. Consciousness and Cognition, 17: 219–239. Tzelgov, J., Henik, A., and Porat, Z. 1997. “Automaticity and consciousness: is perceiving the word necessary for reading it?”. American Journal of Psychology, 3: 429–448. Wegner, D. M. 2002. The illusion of conscious will. Cambridge, MA: MIT Press. Wenke, D., Fleming, S. M., and Haggard, P.  2010. “Subliminal priming of actions influences sense of control over effects of action”. Cognition, 115: 26–38. Wessel, J. R. 2012. “Error awareness and the error-related negativity: evaluating the first decade of evidence”. Frontiers in Human Neuroscience, 6: 88.

281

22 IMPLICIT COGNITION AND ADDICTION Selected Recent Findings and Theory Reinout W. Wiers and Alan W. Stacy 1. Introduction Fifteen years ago, we finalized the handbook on Implicit Cognition and Addiction (Wiers and Stacy 2006). In the same period, one of the authors received a letter from a US prisoner, asking him to provide an expert testimony for court to indicate that he had committed his crime without free will, as he lost his free will through addiction. Perhaps more than reviews and questions of colleagues, this letter raised a series of important questions: first, was this the implication of our research and the volume we put together, that people lose their free will when they get addicted? Second, in cognitive science and neuroscience the concept of “free will” was widely questioned and called an illusion (“the mind’s best trick”, Wegner 2003). However, if that was true, the next question was whether people really lost only an illusion as they became addicted; or: what is the difference between an addicted and a non-addicted person? (see Wiers, Field et al. 2014) In this chapter, we will first discuss theory: what is implicit cognition, and how does it relate to the phenomena of addiction, including loss of voluntary control. Second, we will review and update the assessment of implicit cognition in relation to addiction. Third, we will briefly review the role of implicit cognition in the treatment of addiction, specifically attempts to directly influence implicit cognitive processes to ameliorate the recovery process (varieties of cognitive bias modification or CBM), which took off after publication of the handbook.

2.  Implicit Cognition and Addiction: Theory The term “implicit” in implicit cognition or implicit memory is used in different ways, as outlined by Jan De Houwer (2006) in his excellent chapter in the handbook. First, and without controversy, it can refer to an indirect assessment method, through which a cognitive motivational variable is derived indirectly, which can be contrasted with a direct measure, which comes down to asking somebody the reasons for doing something. For example, if we want to know why somebody smokes, despite knowing that smoking severely decreases life expectation, we can ask the expected benefits, either in a (clinical) interview or with a questionnaire. That would be the direct assessment of cognitive motivational constructs to smoke (e.g., expectancies, motives, etc.). The alternative would be to assess the same constructs indirectly, for example through a reaction-time test or through assessing first memory associations (described later). 282

DOI: 10.4324/9781003014584-28

Implicit Cognition and Addiction

Hence, implicit in the sense of indirect measurement is uncontroversial: motivation is inferred indirectly from responses. The second meaning concerns the outcome of the measurement procedure, the implicit cognitions or memories. According to dual process perspectives, these should be seen as a different entity than their explicit counterparts, hence, it would be possible that one is explicitly negative about smoking but implicitly positive (or the other way around). However, dual process models have been criticized (Hommel and Wiers 2017; Keren and Schul 2009; Melnikoff and Bargh 2018). Moreover, as De Houwer (2006) already indicated, it is important to characterize and assess the functional properties of implicit cognitions: what exactly do we mean, when we use the term. Are they unconscious, assessed outside awareness of the person being assessed, automatic, effortless, uncontrolled or even uncontrollable? In line with De Houwer (2006) and Fazio (Fazio 1990; Fazio and Olson 2014), we would propose that implicit cognition and memory concerns an index of relatively automatic or spontaneous processes that play a role in the (addictive) behavior of interest, which can be initiated outside conscious awareness, but does not have to. The assessment can be seen as a mini-experiment in which participants’ spontaneous reactions to addiction-related stimuli are assessed and compared with reactions to other stimuli (De Houwer 2006; Wiers, Houben et al. 2010). The resulting indirect measure can be compared regarding its predictive validity of the behavior with more direct measures (e.g., questionnaires), as has been done in many studies and two summarizing meta-analyses (Rooke et al. 2008; Reich et al. 2010), which both concluded that implicit measures predict unique variance after controlling for explicit measures. However, it should be noted that this conclusion has been contested with reference to differences in the measures employed (Blanton et  al. 2016). There is also evidence that implicit measures predict other aspects of behavior, more spontaneous nonverbal behavior, while explicit measures better predict deliberate aspects of behavior (Gawronski 2019). One way in which implicit cognition has been viewed is as automatic in the sense of overlearned or habitual (Tiffany 1990). Habits develop with repetition, which frees up attentional control: someone learning to drive cannot talk and drive well, but an experienced driver can. Similarly, a smoker may simply light a cigarette when waiting outside, without elaborate decision making. The crucial question is what happens when the normal routine is blocked. If addictive behaviors are habitual and automatic in the strong sense, as has been proposed as an important mechanism in addiction (Everitt and Robbins 2005, 2016), the stimulus (cue) will elicit the response in a compulsive way, even when it is known that the result of the action is negative (e.g., a foot-shock in animal models). However, voluntary actions in humans and many other animals are typically goal-directed, and when the routine is blocked for some reason (e.g., the driver finds the normal road blocked), attention is switched back to the problem-solving mode, and the drive may stop talking (Kruglanski and Szumowska 2020). Hence, the critical question for the (strong) habit account is what happens in addicted people, in case their routine to obtain the drug reward is blocked? Typically they do not persevere in the old routine, but try to find new ways to fulfill their goal (Robinson and Berridge 2008). In a recent review of both the human and the animal literature, Hogarth (2020) concluded that the empirical evidence for the strong (compulsive) habit account for addiction is weak at best. Of course, addictions are often habitual in the weaker sense of frequent, overlearned and resistant to change, but these characteristics do not make them immune to consequences and not goal-directed (Hogarth 2020; cf., Hommel 2019; Kruglanski and Szumowska 2020; Wiers et al. 2020; but see Wood et al. 2022, for counterarguments). A different theoretical perspective currently gaining popularity, is that addictive behaviors are goal-directed, even in the problematic later phases of addiction, while recognizing that people 283

Reinout W. Wiers and Alan W. Stacy

have multiple goals and that active goals can automatically trigger means to achieve the goal (Köpetz et al. 2013; Kruglanski and Szumowska 2020; Pickard 2020). Human (or more general vertebrate) brains have many different subsystems, functionally defined by their concerns and different subsystems may yield their own suggested actions, which may conflict (e.g., one system suggesting approach of an attractive object, while another system signals danger), and conscious awareness may have evolved to align these inputs and to reach a decision (Morsella 2005; Morsella et al. 2015). According to this view, the primary function of the integrated mind/brain is to predict what is about to happen and how to optimally react (Clark 2013; Verschure 2016; Pezzulo et al. 2018). Subsystems can signal needs (e.g., thirst) that which will activate a goal (drink), and learned ways to achieve the goal (e.g., fetch drink from the fridge). From this perspective, addictions result in strong desires, but these remain goal-directed in an integrated overall system that integrates reactions to different goals (Field et al. 2020; Hogarth 2020; Wiers, Van Gaal et al. 2020). The question then becomes whether the addictive behavior can be irrational, in the sense of an example of weakness of will or akrasia in classical Greek philosophy (see Heather and Segal 2013; Heather 2017). In philosophy, akrasia means that somebody does “A”, while knowing that “B” is the better choice, all things considered. For example, an ex-smoker knows smoking kills and severely reduces life-expectancy, and yet (s)he resumes smoking. The tricky part appears to be the “all things considered”: while in a philosophical armchair it is easy to state that all things can (theoretically) be considered, in real life this is typically not the case. Decisions have to be made, and there is no time to consider all possible relevant consequences. Therefore, it is possible that a decision is made which serves one salient goal at the moment of decision making (e.g., need to bond with an old friend, who offers the cigarette), while the ex-smoker is aware that accepting the cigarette goes against other goals (health, life-expectancy, self-image as stronger than the addiction, etc.); which, upon careful consideration, are more important, “all things considered”. Accepting the cigarette is irrational then, from the overall perspective, but “rational” regarding the dominant goal at that moment (Field et al. 2020; Gladwin et al. 2011). Returning to the initial questions relating to the (ir)rationality of the addicted prisoner, the current perspective would not lead to the conclusion that the addicted prisoner had lost free will, but rather that his drug use before the incident for which he was convicted, was biased by his history of addiction.

3.  Implicit Cognition and Addiction: Assessment It is one thing to theorize about implicit or spontaneously activated cognitive processes in addiction, it is another thing to assess them in a reliable and valid way. Progress in assessment of implicit cognition in addiction has continued since the last major reviews of this research area (Stacy and Wiers 2010; Wiers, Gladwin, Hofmann, et al. 2013). This progress is underscored by further investigation of measures that have been used for one or more decades as well as through use of new promising measures and more rigorous psychometric documentation.

3.1  Assessment of Addiction-Relevant Memory Associations As outlined earlier, all assessments of implicit cognition use some form of indirect assessment, in which the key association or relationship presumably underlying an implicit process is assessed without direct questioning about that relationship. We primarily discuss progress in the assessment of memory associations here, and briefly assessment of biases in attention and action tendencies later. 284

Implicit Cognition and Addiction

3.1.1  Assessment of Addiction-Relevant Memory Associations Using Reaction Time (RT) Measures The most popular assessment instrument in addiction research and social psychology is the Implicit Association Test (Greenwald et al. 1998; Greenwald et al. 2009; Greenwald et al. 2021). The IAT assesses through a series of reaction time (RT) trials whether people more strongly associate a target behavior with one attribute versus another attribute. Attributes can be almost anything, but have usually been positive and negative affect categories as reflected in words presented on the computer screen (e.g., good, pleasant, bad, unpleasant). In this case, a series of RT trials provides the critical comparisons that yield an indirect measure of attitude toward the target behavior, commonly inferred as an implicit attitude. IATs have also been used to assess associations of a behavior with other attributes such as arousal or excitement (Wiers et al. 2002, 2005) and motivational orientation (approach vs. avoidance, Palfai and Ostafin 2003; Ostafin and Palfai 2006). Some longitudinal studies have revealed that the IAT is predictive of later substance use, such as use of tobacco (e.g., Chassin et al. 2010), and alcohol (e.g., Lindgren et al. 2016; Thush and Wiers 2007). An intriguing recent version of the IAT focuses on self-identity as the attribute, rather than affect or arousal. In the self-identity IAT, the comparisons of focus address the target behavior in relation to the attributes “me” compared to “not me”. Lindgren and her colleagues (2016) recently conducted probably the most rigorous longitudinal study ever addressing the IAT in any area of research. This study followed college students 21 months over eight waves of assessment and reported that the drinking-identity IAT was a significant and reliable predictor of alcohol use and related problems. Prediction remained significant even after statistically controlling for prediction by other versions of the IAT, explicit cognition variables, and previous alcohol consumption. Although several other IATs (excitement, approach-avoid) were evaluated as predictors as well, the identity IAT was the only variation of the IAT that remained predictive after 21 months. Innovations in further development of the IAT for addiction research continue, for example, with a promising version with nonverbal stimuli (Palfai et al. 2016). A recent new variety of the IAT attempts to assess implicit “wanting” (like the earlier excitement-IAT, Wiers et al. 2002, and motivational IAT, Palfai and Ostafin 2003), in accord with incentive sensitization theory (Robinson and Berridge 1993; Berridge and Robinson 2016), which differentiates the neural processes underlying wanting and liking, and showed that addiction is characterized by excessive wanting, in the absence of liking. In this new variety (Koranyi et al. 2017; Grigutsch et al. 2019), a truly motivational wanting quality was added to the measurement, by first making participants thirsty before the assessment. During assessment, participants obtained water as a consequence of the response categorized as “I want”. Adding these action-effects may improve measurement properties, going back to the idea that implicit assessments can be seen as mini-experiments (De Houwer 2006; Wiers, Rinck et al. 2010). Another novel way to investigate the IAT is in terms of neural correlates during assessment. Ames and her colleagues have found that engagement in substance-related IATs is correlated with activation in certain brain regions hypothesized to support implicit processes (Ames et al. 2013, 2014; cf., Ernst et al. 2014; C.E. Wiers, Stelzel et al. 2014).

3.1.2  Assessment of Addiction-Relevant Memory Associations Using Non-RT Measures One of the earliest and most impactful experimental literatures on implicit processes addressed what became known as an implicit associative response (IAR, Bousfield et al. 1958; Deese 1962; 285

Reinout W. Wiers and Alan W. Stacy

Underwood 1965). This compelling early work, surprisingly unrecognized in some research domains, experimentally manipulated IAR by varying the associative strength or structure of words presented as cues or in word-lists. The assessments employed to define associative strength or structure were simple tests of word association. In addiction research, word association assessments use simple top-of-mind instructions, which ask participants to write the first word or behavior that comes to mind in response to cue words, phrases, or pictures. Responses to these types of tests have been found to prospectively predict use of alcohol (Ames et al. 2007, 2017; Van Der Vorst et al. 2013; Salemink and Wiers 2014) and marijuana (Ames et al. 2017; Shono et al. 2018) in a variety of populations. In a study of over 1,000 Canadian adolescents, associations in memory assessed with word association were found to predict the onset of alcohol consumption (Van Der Vorst et al. 2013). When the target behavior (e.g., alcohol consumption) is not mentioned in the instructions or cue, word association tests are quite indirect (see for a more elaborate discussion, Stacy et al. 2006). The relevance to implicit processes is inferred not only on the basis of the indirect structure of word association tests but also on the basis of several decades of basic research, spanning research on IAR, multiple memory systems, and research showing that word association yields normative parameters strongly predictive of responses in independent paradigms thought to involve implicit or automatic processes, such as semantic priming, extralist cued-recall, and illusory memory (for a recent review, see Stacy et al. 2020). Payne and his colleagues have provided strong support for yet another distinct indirect test of association: the affect misattribution procedure (AMP, Payne and Lundberg 2014), which has been adapted to wide variety of topics, including the prediction of onset of alcohol use (Payne et al. 2016). When the AMP is used to study alcohol use, the computerized procedure presents an image of an alcoholic or soft drink. Almost immediately afterward, a Chinese pictograph is presented. Participants are asked to decide whether the pictograph is more or less pleasant than the average pictograph shown and to press one of two response keys indicating their preference. The idea is that the image of the drink should rapidly prime or activate related affect, such as a positive or negative feeling toward alcohol (similar to affective priming, a technique widely used in social psychology, but hardly in addiction research). Within a short time-interval, this is hypothesized to carry over or prime pleasantness judgments of the neutral pictograph, which does not normally engender an affective response (among those who cannot read Chinese). In a unique analysis of potentially different prospective effects of implicit processes in initial non-drinkers versus drinkers, Payne and his colleagues followed 868 adolescents and found that scores on the AMP predicted alcohol consumption one year later regardless of initial drinker status. This is important because it is well established that heavy drinking (or drug use) at an early age is a risk-factor for later addiction (Kuntsche et al. 2013; de Goede et al. 2021). Importantly, the AMP procedure includes instructions that warn participants to avoid letting the preceding image (prime) influence responses on the pleasantness task. However, it should be noted that responses are not immune to faking (Teige-Mocigemba et  al. 2016). Further, creative investigators have transformed the AMP into an assessment of semantic associations, without requiring an evaluative judgment of pleasantness (Imhoff et al. 2011), which could also be interesting to use in addiction research, but has not been done, to the best of our knowledge. A number of other indirect tests of association and implicit or automatic processes have been used in addiction research that should be briefly mentioned. These include, for example, semantic priming in lexical decision (e.g., Zack et al. 1999), semantic and mediated priming in word naming (e.g., Sayette et al. 2001), affective priming (e.g., Glock et al. 2015), illusory memory (e.g., Reich et al. 2004), the Extrinsic Affective Simon Test (e.g., de Jong et al. 2007) and process dissociation procedures (e.g., Fillmore et al. 1999). Most likely additional indirect test variations could be “mined” or translated from research on memory, cognitive neuroscience, 286

Implicit Cognition and Addiction

social psychology or decision research for potential use in the addiction area. A few examples are the relational responding task (De Houwer et al. 2015) and the implicit relational assessment procedure (IRAP, Power et al. 2009). However, many of these varieties have shown poorer reliability compared to the IAT (e.g., De Houwer and De Bruycker 2007), related to measurement properties of the tasks (explained in the next paragraph).

3.2  Assessment of Addiction-Relevant Attentional Bias and Approach Bias While all measures described so far can be conceptualized as ways to assess aspects of (affective or semantic) memory, two other broad categories of assessment tasks should be briefly mentioned (we do not have room for a full review): those aiming to assess attentional biases and those aiming to assess biases in action tendencies, as these are important targets in cognitive bias modification (discussed in the next section).

3.2.1  Assessment of Attentional Bias in Addiction The most used tasks to assess attentional bias (AtB) in relation to addiction are the modified Stroop task (Cox et al. 2006) and the visual-probe task (Field et al. 2006; Field and Cox 2008). Importantly, both these tasks have been criticized regarding their psychometric properties, especially the visual probe task, which has a very low reliability (Ataya et al. 2012). Physiological measures such as eye-movements show better psychometric properties (Field et al. 2009). In addition, recently, new “dual probe” varieties have been developed that show much better reliability (MacLeod et al. 2019), that are currently adapted for use in the addiction domain (Cahill et al. 2021; Wiechert et al. 2021).

3.2.2  Assessment of Approach Bias in Addiction Regarding biased action tendencies (in addiction the tendency to approach a conditioned stimulus relating to the addictive behavior, ApB or Approach bias) has been assessed with varieties of the approach-avoidance task (AAT, Wiers et al. 2009). The measure includes an outcome effect (after pulling or pushing a joystick, the stimulus increases or decreases in size, respectively), and has shown to be related to problematic use of alcohol, cigarette smoking, cannabis and gambling (Boffo et al. 2017; Cousijn et al. 2011; C.E. Wiers, Kühn et al. 2013; Wiers et al. 2009). A second often used task involves a symbolic approach-avoidance movement of a manikin approaching or avoiding a specific category of stimuli (Mogg et  al. 2003; Field et  al. 2006, 2008). This task is sometimes confusingly called the Stimulus Response Compatibility (SRC), which is confusing, because variability in stimulus response compatibility underlies RT-measures of implicit processes in general. The latter task is typically assessed in a relevant-feature format (De Houwer 2003): participants are requested to make a manlike figure approach the substance stimuli in one block (and avoid the other category, e.g., stationary) and to avoid the substance stimuli in another block (and approach the other category). The ApB is calculated as the difference in RT between the approach substance block and the avoid substance-stimuli block. As participants react to a relevant feature of the stimulus (approach/avoid substance stimuli), this is a relevant feature task, like the IAT. These tasks typically have good reliabilities, but have rather explicit instructions related to what is measured (approach/avoid addiction-related stimuli). In irrelevant feature tasks, in contrast, participants react to a feature unrelated to the contents of the stimuli (e.g. the format or tilt of the picture). ApB is then calculated as the difference in 287

Reinout W. Wiers and Alan W. Stacy

reaction times to approach/avoid the stimuli of a certain type, sometimes corrected for the bias for the alternative category (Boffo et al. 2017; Cousijn et al. 2011). Direct comparisons have demonstrated a much lower reliability for irrelevant feature tasks compared with relevant feature tasks (De Houwer and De Bruycker 2007; Field et al. 2011). The advantage of irrelevant feature tasks, however, is that content can be modified without changing any instructions, which is useful when the task is transformed into a training instrument, as will become clear in the next section. But for assessment purposes, it is clear that for RT-tasks, relevant feature versions are preferable over irrelevant feature versions of tasks, although this may be difficult to combine with training when one wants to use indirect instruction in the training (Wiers, Gladwin, and Rinck 2013). Note that the aforementioned non-RT measures of word association (typically not covered in general overviews of implicit assessment from a social cognition perspective) also have good reliabilities (Shono et al. 2016). Some final remarks about assessment (see further for general lessons regarding RT-measures of implicit cognition, Gawronski 2019). First, implicit measures are not process-pure measures of implicit processes. For example, in relevant-feature tasks such as the IAT, there is a switch in response-assignments between the blocks and people differ in how easily they adapt, related to executive functions (Klauer et al. 2010). This is important because a number of studies have demonstrated that implicit measures like the IAT better predict (addictive) behaviors in participants with relatively weak executive control (Hofmann et al. 2008; Thush et al. 2008; Houben and Wiers 2009). Importantly, the same pattern has also been observed for measures of word association, which do not have an obvious executive control component in the measurement procedure (Grenard et al. 2008), and with an irrelevant feature AAT (Peeters et al. 2012), which also does not require switching. Further, mathematical procedures have been developed to estimate the contribution of associative and control processes in outcomes of indirect measures such as the IAT, such as the QUAD model (Conrey et al. 2005) and related procedures, which have also been applied to the analysis of the changes caused by cognitive bias modification (Gladwin et al. 2015), discussed later. Finally, as measurement procedures move from the computer to the smartphone, with some issues concerning measurement (Pronk et al. 2020), more intense measurement is possible in relevant situations, which opens up a myriad of new possibilities to assess contextual effects on implicit cognition (e.g., Marhe et al. 2013).

4.  Implicit Cognition and Addiction: Recovery This first topic under this heading concerns whether implicit measures can teach us something about recovery from addiction, for example, regarding prediction of relapse. The second topic concerns the question if implicit (in the sense of automatically triggered, relatively spontaneous) processes can be influenced and whether that has (or adds to) treatment effects.

4.1  Implicit Measures Predicting Treatment Outcomes Some studies have reported that implicit measures predict relapse (Cox et al. 2002; Marhe et al. 2013; Spruyt et al. 2015), but results have been far from conclusive (Snelleman et al. 2015; Field et al. 2016), probably related to the poor psychometric quality of most currently used RT-based implicit measures, which is not the case for relevant-feature RT-based tasks such as the IAT (Greenwald et al. 2009), and non-RT based measures such as word-association (Shono et al. 2016; Shono et al. 2018). Promisingly, new tests are also developed to assess attentional bias in a more reliable way (MacLeod et al. 2019; Wiechert et al. 2021), but these have not been used yet to predict relapse. 288

Implicit Cognition and Addiction

4.2  Targeting Implicit Cognitive Processes: Cognitive Bias Modification (CBM) We briefly review CBM here, because other recent reviews are available (Boffo et  al. 2019; Wiers et al. 2018). First, it is important to recognize two different subclasses of CBM-research: proof-of-principle (PoP) studies, aimed at testing the hypothesis that the cognitive bias has a causal role in determining (symptoms of) the addictive behavior (or other form of mental health problem), and randomized controlled trials (RCTs) in clinical populations, aimed at establishing therapeutic effects, two different phases of the experimental medicine approach to intervention development (Sheeran et  al. 2017). Following other areas of experimental psychopathology, especially anxiety (MacLeod et al. 2002), initial PoP studies attempted to manipulate cognitive biases in healthy volunteers, to investigate if this had effects on a proxy of addictive behavior (e.g., how much alcohol was consumed in a taste-test, e.g., Field and Eastwood 2005). Note that in order to test causality, the bias can be manipulated in both directions (toward alcohol or away from alcohol), but that in clinical applications or in studies with problem drinkers, typically a neutral control condition is used for ethical reasons to contrast effects of training away from the substance (see Wiers et  al. 2008). Different implicit processes have been addressed in these studies: attentional bias (e.g., Field and Eastwood 2005; Schoenmakers et al. 2007), approach bias (e.g., Wiers, Rinck, et al. 2010) and memory associations with a variety of different procedures, including selective inhibition training (e.g., Houben et al. 2011) and evaluative conditioning (e.g., Houben et al. 2010). The pattern of results over these PoP studies was rather consistent: if the bias was successfully manipulated, a short-lived effect on behavior was typically observed (review: Wiers et al. 2018), such as an effect on drinking in a taste-test or on craving (Field and Eastwood 2005; Wiers, Houben et al. 2013), or an effect on drinking behavior in the week following the experiment (Houben et al. 2011, 2012). Note, however, that assessing the effect of the manipulation on the bias is difficult, given the weak psychometric properties of most measures used (typically irrelevant feature RT-measures). In the second class of CBM studies, effects on clinical outcomes are assessed. These studies include either patients who are treated for an addiction and motivated to change or volunteers suffering from addictive behaviors who seek help (typically online), which is a very different group from the healthy volunteers (typically students) in PoP studies, who participate for course credit, money or free beer (Wiers et al. 2018). A number of studies have added CBM to regular treatment of alcohol use disorders (AUDs) and all found that adding CBM to treatment reduced relapse (Eberl et al. 2013; Manning et al. 2016, 2021; Rinck et al. 2018; Salemink et al. 2021; Schoenmakers et al. 2010; Wiers et al. 2011). Effects on one-year follow-up are modest (around 10% reduction in relapse), but clinically important and similar in size to the effects of current medications in alcohol use disorders (Jonas et al. 2014). These consistent positive add-on effects have led to the recommendation to use CBM as an add-on to the treatment of AUD in Germany and Australia, where most of the RCTs took place. Regarding mechanisms, approach bias modification has been shown to reduce neural cue-reactivity in a circuit including the amygdala (C.E. Wiers, Stelzel et al. 2015) and medial prefrontal activity underlying approach-avoidance tendencies (C.E. Wiers, Ludwig et al. 2015). In addition, a decomposition analysis of the IAT demonstrated that the effects of approach-bias modification in increasing alcohol-avoidance associations predicted lower relapse (Gladwin et al. 2015). While the effects of adding CBM to the treatment of AUD have consistently been positive (less relapse compared to sham-training or no training), effects of online CBM in volunteers have been less promising, with participants in all conditions reducing their alcohol use (van Deursen et al. 2016; Jones et al. 2018; R.W. Wiers, Houben et al. 2015). There are at least two 289

Reinout W. Wiers and Alan W. Stacy

possible explanations for this difference: it could be related to the lack of supporting therapy or to the different treatment goal; in regular treatment abstinence and in online training participants typically want to reduce their drinking (which they do, no matter which condition they are in). One study in the domain of smoking did find effects of online CBM versus control training in heavy smokers, which may have been related to the abstinence goal, which was verified prior to training (Elfeddali et al. 2016), supporting the idea that the abstinence goal might be crucial for differential effects of CBM compared with sham-training. Note that an early meta-analysis on CBM in addiction (Cristea et al. 2016) concluded that there was some evidence for CBM having effects on cognitive biases but not for clinical effectiveness. However, this analysis combined (many) PoP-studies in volunteers not motivated to change with the few published clinical randomized controlled trials (RCTs), as noted two different phases of intervention development (Sheeran et al. 2017; Wiers et al. 2018), which invalidates the conclusions drawn. A more recent meta-analysis included only studies with a clinical purpose (behavior change), both online and clinical RCTs (Boffo et al. 2019), and confirmed significant (small) effects of CBM on cognitive bias and on relapse, but not on reduced use. In conclusion, there is evidence that CBM has a small but clinically relevant add-on effect in the abstinence-oriented treatment of AUD, and less is known as-yet about effects in other addictions, although one positive study in heroin addiction shows promise regarding other addictions (Ziaee et al. 2016). While CBM was developed from a dual-process perspective, from the idea that it would uniquely change implicit cognitive processes, not influenced by regular therapy, that would address primarily explicit cognitions (Wiers et al. 2008), recent evidence has questioned this interpretation. For example, training effects have been found to be moderated by beliefs about the consequences of the trained actions (Van Dessel et al. 2018), by instructions (Van Dessel et al. 2015) and by subjective awareness of stimulus-action relations (Van Dessel et al. 2016). These findings appear to be more in line with a theoretical framework of automatic inferences (Van Dessel et al. 2019), which suggests that the human mind can be seen as an integrated “prediction machine” building on hierarchical belief modules (Clark 2013; Verschure 2016; Moors et al. 2017; Wiers and Verschure 2021). From this new perspective, approach bias modification does not work by changing specific automatic associations, but by changing propositional beliefs that give rise to automatic inferences underlying (maladaptive) behavior (Wiers, Van Dessel, et al. 2020). This new perspective has led to a new variety of CBM, ABC-training, in which personally relevant behavioral alternatives (Bs) are trained in personally relevant risk situations (Antecedent contexts, As), where all behavioral choices have (personally relevant) consequences (Cs) (Wiers, Van Dessel, et al. 2020). Personalizing alternative activities appears to be especially important in other addictions than AUD, where participants are trained away from alcohol toward a generally relevant alternative (non-alcoholic drinks), which does not exist for other addictions. In smoking, personalizing alternative actions has already shown promise (Köpetz et al. 2017). Note that this approach is conceptually related to implementation intentions (Gollwitzer and Sheeran 2006), which have been shown to be effective in changing health behaviors, but does not entail a training component. ABC-training adds such a personalized training component, which is likely to benefit patients (cf., Eberl et al. 2014).

5. Conclusion As could have been expected, a lot has happened in the field of implicit cognition and addiction, since we published the handbook on the topic 15 years ago. Regarding theory, dual process models have been criticized, the idea that addictive behaviors can become fully automatic 290

Implicit Cognition and Addiction

or habitual in the strong sense (no longer influenced by intentional processes and insensitive to punishment) appears to be too strong (Field et al. 2020; Hogarth 2020; Wiers, Van Gaal, et al. 2020). At the same time, it is clear that decision making is biased in addiction, and that many different neural circuits underlie the associated biases in attention, memory and action tendencies. However, although decision making is biased in addiction, it still typically involves intentional decision making, and perhaps only in very extreme cases this can no longer be called intentional (e.g., AUD with severe Korsakoff syndrome, Fenton and Wiers 2017). This is promising from a treatment perspective, especially given that at least some of the neural abnormalities in addiction appear to recover after prolonged abstinence (Schulte et al. 2014; Lees et al. 2019), and this may also reduce cognitive biases (C. E. Wiers, Kühn et al. 2013). While the dominant perspective in biomedical research has been that addiction concerns a chronic brain disease (Leshner 1997; Volkow et al. 2016), the new perspective yields a more optimistic perspective on recovery after addiction (see Wiers and Verschure 2021). Regarding assessment, a lot of work has been done, but it has been difficult to develop reliable and valid measures of implicit cognition, which may be related to the context-sensitive fluctuating nature of the processes targeted, which also applies to the relatively spontaneous behaviors it should primarily predict (Gawronski 2019). This is not unique to implicit measures, the same has been argued for measures of impulsivity (White et al. 1994). Importantly, there are new promising developments concerning assessment, which is also important for the prediction of treatment effects and the assessments of effects of CBM. Regarding the latter, there is convincing evidence that current CBM has a small but clinically relevant add-on effect to the treatment of AUD (Boffo et al. 2019; Wiers et al. 2018), with as-yet, less clear effects on other addictions. A new variety of training might do better there (ABC-training) as it includes personalized alternatives and consequences, but the future will tell if this is confirmed or not.

Related Topics Chapters 2, 3, 21, 23, 31

References Ames, S. L., Grenard, J. L., Stacy, A. W., Xiao, L., He, Q., Wong, S. W., Xue, G., Wiers, R. W., Bechara, A., W. Stacy, A., Xiao, L., He, Q., Wong, S. W., Xue, G., W. Wiers, R., and Bechara, A. 2013. “Functional imaging of implicit marijuana associations during performance on an Implicit Association Test (IAT)”. Behavioural Brain Research, 256: 494–502. https://doi.org/10.1016/j.bbr.2013.09.013 Ames, S. L., Grenard, J. L., Thush, C., Sussman, S., Wiers, R. W., and Stacy, A. W. 2007. “Comparison of indirect assessments of association as predictors of marijuana use among at-risk adolescents”. Experimental and Clinical Psychopharmacology, 15: 204–218. https://doi.org/10.1037/1064-1297.15.2.218 Ames, S. L., Wong, S. W., Bechara, A., Cappelli, C., Dust, M., Grenard, J. L., and Stacy, A. W. 2014. “Neural correlates of a Go/NoGo task with alcohol stimuli in light and heavy young drinkers”. Behavioural Brain Research, 274: 382–389. Ames, S. L., Xie, B., Shono, Y., and Stacy, A. W. 2017. “Adolescents at risk for drug abuse: a 3-year dualprocess analysis”. Addiction, 112: 852–863. https://doi.org/10.1111/add.13742 Ataya, A. F., Adams, S., Mullings, E., Cooper, R. M., Attwood, A. S., and Munafò, M. R. 2012. “Internal reliability of measures of substance-related cognitive bias”. Drug and Alcohol Dependence, 121: 148–151. https://doi.org/10.1016/j.drugalcdep.2011.08.023 Berridge, K. C., and Robinson, T. E. 2016. “Liking, wanting, and the incentive-sensitization theory of addiction”. American Psychologist, 71: 670–679. https://doi.org/10.1037/amp0000059 Blanton, H., Burrows, C. N., and Jaccard, J. 2016. “To accurately estimate implicit influences on health behavior, accurately estimate explicit influences”. Health Psychology: Official Journal of the Division of Health Psychology, American Psychological Association, 35: 856–860. https://doi.org/10.1037/hea0000348

291

Reinout W. Wiers and Alan W. Stacy Boffo, M., Smits, R., Salmon, J. P., Cowie, M. E., de Jong, D. T. H. A., Salemink, E., Collins, P., Stewart, S. H., and Wiers, R. W. 2017. “Luck, come here! Automatic approach tendencies toward gambling cues in moderate-to-high risk gamblers”. Addiction, 113: 289–298. https://doi.org/10.1111/add.14071 Boffo, M., Zerhouni, O., Gronau, Q. F., van Beek, R. J. J., Nikolaou, K., Marsman, M., and Wiers, R. W. 2019. “Cognitive bias modification for behavior change in alcohol and smoking addiction: Bayesian meta-analysis of individual participant data”. Neuropsychology Review, 29: 52–78. https://doi. org/10.1007/s11065-018-9386-4 Bousfield, W. A., Whitmarsh, G. A., and Danick, J. J. 1958. “Partial response identities in verbal generalization”. Psychological Reports, 4: 703–713. Cahill, C., White, M., MacLeod, C., Grafton, B. 2021. “Craving mediates the association between attentional bias to alcohol and in vivo alcoholic beverage consumption in young social drinkers”. Psychology of Addictive Behaviors, 35: 895–900. Chassin, L., Presson, C. C., Sherman, S. J., Seo, D.-C., and Macy, J. T. 2010. “Implicit and explicit attitudes predict smoking cessation: moderating effects of experienced failure to control smoking and plans to quit”. Psychology of Addictive Behaviors, 24: 670–679. Clark, A. 2013. “Whatever next? Predictive brains, situated agents, and the future of cognitive science”. Behavioral and Brain Sciences, 36: 181–204. https://doi.org/10.1017/S0140525X12000477 Conrey, F. R., Gawronski, B., Sherman, J. W., Hugenberg, K., and Groom, C. J. 2005. “Separating multiple processes in implicit social cognition: the quad model of implicit task performance”. Journal of Personality and Social Psychology, 89: 469–487. https://doi.org/10.1037/0022-3514.89.4.469 Cousijn, J., Goudriaan, A. E., and Wiers, R. W. 2011. “Reaching out towards cannabis: approach-bias in heavy cannabis users predicts changes in cannabis use”. Addiction, 106: 1667–1674. https://doi. org/10.1111/j.1360-0443.2011.03475.x Cox, W. M., Fadardi, J. S., and Pothos, E. M. 2006. “The addiction-stroop test: theoretical considerations and procedural recommendations”. Psychological Bulletin, 132: 443–476. https://doi. org/10.1037/0033-2909.132.3.443 Cox, W. M., Hogan, L. M., Kristian, M. R., and Race, J. H. 2002. “Alcohol attentional bias as a predictor of alcohol abusers’ treatment outcome”. Drug and Alcohol Dependence, 68: 237–243. https://doi. org/10.1016/S0376-8716(02)00219-3 Cristea, I. A., Kok, R. N., and Cuijpers, P. 2016. “The effectiveness of cognitive bias modification interventions for substance addictions: a meta-analysis”. PlosOne, 11: e0162226. https://doi.org/10.1371/ journal.pone.0162226 Deese, J. 1962. “Form class and the determinants of association”. Journal of Verbal Learning and Verbal Behavior, 1: 79–84. de Goede, J., van der Mark-Reeuwijk, K. G., Braun, K. P., le Cessie, S., Durston, S., Engels, R. C. M. E., Goudriaan, A. E., Moons, K. G. M., Vollebergh, W. A. M., de Vries, T. J., Wiers, R. W., and Oosterlaan, J. 2021. “Alcohol and brain development in adolescents and young adults: a systematic review of the literature and advisory report of the Health Council of the Netherlands”. Advances in Nutrition, 20: 1–32. https://doi.org/10.1093/advances/nmaa170 De Houwer, J. 2003. “A structural analysis of indirect measures of attitudes”. In J. Musch and K. C. Klauer, eds., The psychology of evaluation: affective processes in cognition and emotion. Hillsdale: Erlbaum: 219–244 De Houwer, J. 2006. “What are implicit measures and why are we using them?”. In R. W. Wiers and A. W. Stacy, eds., Handbook of implicit cognition and addiction. Thousand Oaks, CA: SAGE: 11–28. https://dx.doi.org/10.4135/9781412976237.n2 De Houwer, J., and De Bruycker, E. 2007. “The Implicit Association Test outperforms the Extrinsic Affective Simon Task as an implicit measure of inter-individual differences in attitudes”. British Journal of Social Psychology, 46: 401–421. https://doi.org/10.1348/014466606X130346 De Houwer, J., Heider, N., Spruyt, A., Roets, A., and Hughes, S. 2015. “The relational responding task: toward a new implicit measure of beliefs”. Frontiers in Psychology, 6: 319. https://doi.org/10.3389/ fpsyg.2015.00319 de Jong, P. J., Wiers, R. W., Van de Braak, M., and Huijding, J. 2007. “Using the extrinsic affective simon test as a measure of implicit attitudes towards alcohol: relationship with drinking behavior and alcohol problems”. Addictive Behaviors, 32: 881–887. Eberl, C., Wiers, R. W., Pawelczack, S., Rinck, M., Becker, E. S., and Lindenmeyer, J. 2013. “Approach bias modification in alcohol dependence: do clinical effects replicate and for whom does it work best?”. Developmental Cognitive Neuroscience, 4: 38–51. https://doi.org/10.1016/j.dcn.2012.11.002

292

Implicit Cognition and Addiction Eberl, C., Wiers, R. W., Pawelczack, S., Rinck, M., Becker, E. S., and Lindenmeyer, J. 2014. “Implementation of approach bias re-training in alcoholism-how many sessions are needed?”. Alcoholism, Clinical and Experimental Research, 38: 587–594. https://doi.org/10.1111/acer.12281 Elfeddali, I., de Vries, H., Bolman, C., Pronk, T., and Wiers, R. W. 2016. “A randomized controlled trial of web-based attentional bias modification to help smokers quit”. Health Psychology, 35: 870–880. https://doi.org/10.1037/hea0000346 Ernst, L. H., Plichta, M. M., Dresler, T., Zesewitz, A. K., Tupak, S. V., Haeussinger, F. B., Fischer, M., Polak, T., Fallgatter, A. J., and Ehlis, A. C. 2014. “Prefrontal correlates of approach preferences for alcohol stimuli in alcohol dependence”. Addiction Biology, 19: 497–508. https://doi.org/10.1111/adb.12005 Everitt, B. J., and Robbins, T. W. 2005. “Neural systems of reinforcement for drug addiction: from actions to habits to compulsion”. Nature Neuroscience, 8: 1481–1489. Everitt, B. J., and Robbins, T. W. 2016. “Drug addiction: updating actions to habits to compulsions ten years on”. Annual Review of Psychology, 67: 23–50. https://doi.org/10.1146/annurev-psych-122414-033457 Fazio, R. H. 1990. “Multiple processes by which attitudes guide behavior: The MODE model as an integrative framework”. In M. P. Zanna, ed., Advances in experimental social psychology. Vol. 23. New York: Academic Press: 75–109 Fazio, R. H., and Olson, M. A. 2014. “The MoDe model”. In J. W. Sherman, B. Gawronski, and Y. Trope, eds., Dual-process theories of the social mind. New York: Guilford Press: 155–171. Fenton, T., and Wiers, R. W. 2017. “Free will, black swans and addiction”. Neuroethics, 10: 157–165. https://doi.org/10.1007/s12152-016-9290-7 Field, M., Caren, R., Fernie, G., and De Houwer, J. 2011. “Alcohol approach tendencies in heavy drinkers: comparison of effects in a relevant stimulus-response compatibility task and an approach/avoidance Simon task”. Psychology of Addictive Behaviors, 25: 697–701. https://doi.org/10.1037/a0023285 Field, M., and Cox, W. M. 2008. “Attentional bias in addictive behaviors: a review of its development, causes, and consequences”. Drug and Alcohol Dependence, 97: 1–20. https://doi.org/10.1016/j. drugalcdep.2008.03.030 Field, M., and Eastwood, B. 2005. “Experimental manipulation of attentional bias increases the motivation to drink alcohol”. Psychopharmacology, 183: 350–357. https://doi.org/10.1007/s00213-005-0202-5 Field, M., Eastwood, B., Bradley, B. P., and Mogg, K. 2006. “Selective processing of cannabis cues in regular cannabis users”. Drug and Alcohol Dependence, 85: 75–82. https://doi.org/10.1016/j. drugalcdep.2006.03.018 Field, M., Heather, N., Murphy, J. G., Stafford, T., Tucker, J. A., and Witkiewitz, K. 2020. “Recovery from addiction: behavioral economics and value-based decision making”. Psychology of Addictive Behaviors, 34: 182–193. https://doi.org/10.1037/adb0000518 Field, M., Kiernan, A., Eastwood, B., and Child, R. 2008. “Rapid approach responses to alcohol cues in heavy drinkers”. Journal of Behavior Therapy and Experimental Psychiatry, 39: 209–218. https://doi. org/10.1016/j.jbtep.2007.06.001 Field, M., Munafò, M. R., and Franken, I. H. A. 2009. “A meta-analytic investigation of the relationship between attentional bias and subjective craving in substance abuse”. Psychological Bulletin, 135: 589–607. https://doi.org/10.1037/a0015843 Field, M., Werthmann, J., Franken, I., Hofmann, W., Hogarth, L., and Roefs, A. 2016. “The role of attentional bias in obesity and addiction”. Health Psychology, 35: 767–780. https://doi.org/10.1037/ hea0000405 Fillmore, M. T., Vogel-Sprott, M., and Gavrilescu, D. 1999. “Alcohol effects on intentional behavior: dissociating controlled and automatic influences”. Experimental and Clinical Psychopharmacology, 7: 372–378. Gawronski, B. 2019. “Six lessons for a cogent science of implicit bias and its criticism”. Perspectives on Psychological Science, 14: 574–595. https://doi.org/10.1177/1745691619826015 Gladwin, T.E, Figner, B., Crone, E. A., and Wiers, R. W. 2011. “Addiction, adolescence, and the integration of control and motivation”. Developmental Cognitive Neuroscience, 1: 364–376. https://doi. org/10.1016/j.dcn.2011.06.008 Gladwin, T. E., Rinck, M., Eberl, C., Becker, E. S., Lindenmeyer, J., and Wiers, R. W. 2015. “Mediation of cognitive bias modification for alcohol addiction via stimulus-specific alcohol avoidance association”. Alcoholism: Clinical and Experimental Research, 39: 101–107. https://doi.org/10.1111/acer.12602 Glock, S., Klapproth, F., and Müller, B. C. N. 2015. “Promoting responsible drinking? A mass media campaign affects implicit but not explicit alcohol-related cognitions and attitudes”. British Journal of Health Psychology, 20: 482–497. https://doi.org/10.1111/bjhp.12130

293

Reinout W. Wiers and Alan W. Stacy Gollwitzer, P. M., and Sheeran, P.  2006. “Implementation intentions and goal achievement: a metaanalysis of effects and processes”. Advances in Experimental Social Psychology, 38: 69–119. https://doi. org/10.1016/S0065-2601(06)38002-1 Greenwald, A. G., Brendl, M., Cai, H., Cvencek, D., Dovidio, J. F., Friese, M., Hahn, A., Hehman, E., Hofmann, W., Hughes, S., Hussey, I., Jordan, C., Kirby, T. A., Lai, C. K., Lang, J. W. B., Lindgren, K. P., Maison, D., Ostafin, B. D., Rae, J. R., Ratliff, K. A., Spruyt, A., and Wiers, R. W. 2021. “Best research practices for using the Implicit Association Test”. Behavior Research Methods. https://doi. org/10.3758/s13428-021-01624-3. Greenwald, A. G., Mcghee, D. E., and Schwartz, J. L. K. 1998. “Measuring individual differences in implicit cognition: the Implicit Association Test”. Journal of Personality and Social Psychology, 74: 1464–1480. https://doi.org/10.1037/0022-3514.74.6.1464 Greenwald, A. G., Poehlman, T. A., Uhlmann, E. L., and Banaji, M. R. 2009. “Understanding and using the Implicit Association Test: III. Meta-analysis of predictive validity”. Journal of Personality and Social Psychology, 97: 17–41. https://doi.org/10.1037/a0015575 Grenard, J. L., Ames, S. L., Wiers, R. W., Thush, C., Sussman, S., and Stacy, A. W. 2008. “Working memory capacity moderates the predictive effects of drug-related associations on substance use”. Psychology of Addictive Behaviors : Journal of the Society of Psychologists in Addictive Behaviors, 22: 426–432. https:// doi.org/10.1037/0893-164X.22.3.426 Grigutsch, L. A., Lewe, G., Rothermund, K., and Koranyi, N. 2019. “Implicit ‘wanting’ without implicit ‘liking’: a test of Incentive-Sensitization-Theory in the context of smoking addiction using the Wanting-Implicit-Association-Test (W-IAT)”. Journal of Behavior Therapy and Experimental Psychiatry, 64: 9–14. https://doi.org/10.1016/j.jbtep.2019.01.002 Heather, N. 2017. “Is the concept of compulsion useful in the explanation or description of addictive behaviour and experience?”. Addictive Behaviors Reports, 6: 15–38. https://doi.org/10.1016/j. abrep.2017.05.002 Heather, N., and Segal, G. 2013. “Understanding addiction: Donald Davidson and the problem of akrasia”. Addiction Research and Theory, 21: 445–452. https://doi.org/10.3109/16066359.2012.754014 Hofmann, W., Gschwendner, T., Friese, M., Wiers, R. W., and Schmitt, M. 2008. “Working memory capacity and self-regulatory behavior: toward an individual differences perspective on behavior determination by automatic versus controlled processes”. Journal of Personality and Social Psychology, 95: 962–977. https://doi.org/10.1037/a0012705 Hogarth, L. 2020. “Addiction is driven by excessive goal-directed drug choice under negative affect: translational critique of habit and compulsion theory”. Neuropsychopharmacology, 45: 720–735. https:// doi.org/10.1038/s41386-020-0600-8 Hommel, B. 2019. “Binary theorizing does not account for action control”. Frontiers in Psychology, 10. https://doi.org/10.3389/fpsyg.2019.02542 Hommel, B., and Wiers, R. W. 2017. “Towards a unitary approach to human action control”. Trends in Cognitive Sciences, 21: 940–949. https://doi.org/10.1016/j.tics.2017.09.009 Houben, K., Havermans, R. C., Nederkoorn, C., and Jansen, A. 2012. “Beer à no-go: learning to stop responding to alcohol cues reduces alcohol intake via reduced affective associations rather than increased response inhibition”. Addiction, 107: 1280–1287. https://doi.org/10.1111/j.1360-0443.2012.03827.x Houben, K., Nederkoorn, C., Wiers, R. W., and Jansen, A. 2011. “Resisting temptation: decreasing alcohol-related affect and drinking behavior by training response inhibition”. Drug and Alcohol Dependence, 116: 132–136. https://doi.org/10.1016/j.drugalcdep.2010.12.011 Houben, K., Schoenmakers, T. M., and Wiers, R. W. 2010. “I didn’t feel like drinking but I don’t know why: the effects of evaluative conditioning on alcohol-related attitudes, craving and behavior”. Addictive Behaviors, 35: 1161–1163. https://doi.org/10.1016/j.addbeh.2010.08.012 Houben, K., and Wiers, R. W. 2009. “Response inhibition moderates the relationship between implicit associations and drinking behavior”. Alcoholism, Clinical and Experimental Research, 33: 626–633. https://doi.org/10.1111/j.1530-0277.2008.00877.x Imhoff, R., Schmidt, A. F., Bernhardt, J., Dierksmeier, A., and Banse, R. 2011. “An inkblot for sexual preference: a semantic variant of the Affect Misattribution procedure”. Cognition and Emotion, 25: 676–690. https://doi.org/10.1080/02699931.2010.508260 Jonas, D. E., Amick, H. R., Feltner, C., Bobashev, G., Thomas, K., Wines, R., Kim, M. M., Shanahan, E., Gass, C. E., Rowe, C. J., and Garbutt, J. C. 2014. “Pharmacotherapy for adults with alcohol use disorders in outpatient settings: a systematic review and meta-analysis”. JAMA – Journal of the American Medical Association, 311: 1889–1900. https://doi.org/10.1001/jama.2014.3628

294

Implicit Cognition and Addiction Jones, A., McGrath, E., Robinson, E., Houben, K., Nederkoorn, C., and Field, M. 2018. “A randomised controlled trial of Inhibitory Control Training for the reduction of alcohol consumption in problem drinkers”. Journal of Consulting and Clinical Psychology, 86: 991–1004. Keren, G., and Schul, Y. 2009. “Two is not always better than one: a critical evaluation of two-system theories”. Perspectives on Psychological Science, 4: 533–550. https://doi.org/10.1111/j.1745-6924.2009.01164.x Klauer, K. C., Schmitz, F., Teige-Mocigemba, S., and Voss, A. 2010. “Understanding the role of executive control in the Implicit Association Test: why flexible people have small IAT effects”. Quarterly Journal of Experimental Psychology, 63: 595–619. Köpetz, C. E., Lejuez, C. W., Wiers, R. W., and Kruglanski, A. W. 2013. “Motivation and self-regulation in addiction: a call for convergence”. Perspectives on Psychological Science, 8: 3–24. https://doi. org/10.1177/1745691612457575 Köpetz, C. E., MacPherson, L., Mitchell, A. D., Houston-Ludlam, A. N., and Wiers, R. W. 2017. “A novel training approach to activate alternative behaviors for smoking in depressed smokers”. Experimental and Clinical Psychopharmacology, 25: 50–60. https://doi.org/10.1037/pha0000108 Koranyi, N., Grigutsch, L. A., Algermissen, J., and Rothermund, K. 2017. “Dissociating implicit wanting from implicit liking: development and validation of the Wanting Implicit Association Test (W-IAT)”. Journal of Behavior Therapy and Experimental Psychiatry, 54: 165–169. https://doi.org/10.1016/j. jbtep.2016.08.008 Kruglanski, A. W., and Szumowska, E. 2020. “Habitual behavior is goal driven”. Perspectives on Psychological Science, 15: 1256–1271. Kuntsche, E., Rossow, I., Simons-Morton, B., Bogt, T. Ter, Kokkevi, A., and Godeau, E. 2013. “Not early drinking but early drunkenness is a risk factor for problem behaviors among adolescents from 38 European and North American countries”. Alcoholism: Clinical and Experimental Research, 37: 308–314. https://doi.org/10.1111/j.1530-0277.2012.01895.x Lees, B., Mewton, L., Stapinski, L. A., Squeglia, L. M., Rae, C. D., and Teesson, M. 2019. “Neurobiological and cognitive profile of young binge drinkers: a systematic review and meta-analysis”. Neuropsychology Review, 29: 357–385. https://doi.org/10.1007/s11065-019-09411-w Leshner, A. I. 1997. “Addiction is a brain disease, and it matters”. Science, 278: 45–47. https://doi. org/10.1126/science.278.5335.45 Lindgren, K. P., Neighbors, C., Teachman, B. A., Baldwin, S. A., Norris, J., Kaysen, D., Gasser, M. L., and Wiers, R. W. 2016. “Implicit alcohol associations, especially drinking identity, predict drinking over time”. Health Psychology, 35: 908–918. https://doi.org/10.1037/hea0000396 MacLeod, C., Grafton, B., and Notebaert, L. 2019. “Anxiety-linked attentional bias: is it reliable?”. Annual Review of Clinical Psychology, 15: 529–554. https://doi.org/10.1146/annurev-clinpsy-050718-095505 MacLeod, C., Rutherford, E., Campbell, L., Ebsworthy, G., and Holker, L. 2002. “Selective attention and emotional vulnerability: assessing the causal basis of their association through the experimental manipulation of attentional bias”. Journal of Abnormal Psychology, 111: 107–123. https://doi. org/10.1037/0021-843X.111.1.107 Manning, V., Garfield, J. B. B., Staiger, P. K., Lubman, D. I., Lum, J. A. G., Reynolds, J., Hall, K., Bonomo, Y., Lloyd-Jones, M., Wiers, R. W., Piercy, H., Jacka, D., and Verdejo-Garcia, A. 2021. “Effect of cognitive bias modification on early relapse among adults undergoing inpatient alcohol withdrawal treatment: a randomized clinical trial”. JAMA Psychiatry, 78: 133–140. https://doi. org/10.1001/jamapsychiatry.2020.3446 Manning, V., Staiger, P. K., Hall, K., Garfield, J. B. B., Flaks, G., Leung, D., Hughes, L. K., Lum, J. A. G., Lubman, D. I., and Verdejo-Garcia, A. 2016. “Cognitive bias modification training during inpatient alcohol detoxification reduces early relapse: a  randomized controlled trial”. Alcoholism: Clinical and Experimental Research, 40: 2011–2019. https://doi.org/10.1111/acer.13163 Marhe, R., Waters, A. J., van de Wetering, B. J. M., and Franken, I. H. A. 2013. “Implicit and explicit drug-related cognitions during detoxification treatment are associated with drug relapse: an ecological momentary assessment study”. Journal of Consulting and Clinical Psychology, 81: 1–12. Melnikoff, D. E., and Bargh, J. A. 2018. “The mythical number two”. Trends in Cognitive Sciences, 22: 280–293. https://doi.org/10.1016/j.tics.2018.02.001 Mogg, K., Bradley, B. P., Field, M., and De Houwer, J. 2003. “Eye movements to smoking-related pictures in smokers: relationship between attentional biases and implicit and explicit measures of stimulus valence”. Addiction, 98: 825–836. https://doi.org/10.1046/j.1360-0443.2003.00392.x Moors, A., Boddez, Y., and De Houwer, J. 2017. “The power of goal-directed processes in the causation of emotional and other actions”. Emotion Review, 9: 310–318. https://doi.org/10.1177/1754073916669595

295

Reinout W. Wiers and Alan W. Stacy Morsella, E. 2005. “The function of phenomenal states: supramodular interaction theory”. Psychological Review, 112: 1000–1021. https://doi.org/10.1037/0033-295X.112.4.1000 Morsella, E., Godwin, C. A., Jantz, T. K., Krieger, S. C., and Gazzaley, A. 2015. “Homing in on consciousness in the nervous system: an action-based synthesis”. Behavioral and Brain Sciences, 39: E168. https://doi.org/10.1017/S0140525X15000643 Ostafin, B. D., and Palfai, T. P.  2006. “Compelled to consume: the implicit association test and automatic alcohol motivation”. Psychology of Addictive Behaviors, 20: 322–327. https://doi. org/10.1037/0893-164X.20.3.322 Palfai, T. P., Kantner, C. K., and Tahaney, K. D. 2016. “The image-based alcohol-action implicit association test”. Journal of Behavior Therapy and Experimental Psychiatry, 50: 135–138. https://doi.org/10.1016/j. jbtep.2015.07.002 Palfai, T. P., and Ostafin, B. D. 2003. “Alcohol-related motivational tendencies in hazardous drinkers: assessing implicit response tendencies using the modified-IAT”. Behaviour Research and Therapy, 41: 1149–1162. https://doi.org/10.1016/S0005-7967(03)00018-4 Payne, B. K., Lee, K. M., Giletta, M., and Prinstein, M. J. 2016. “Implicit attitudes predict drinking onset in adolescents: shaping by social norms”. Health Psychology, 35: 829–836. https://doi.org/10.1037/ hea0000353 Payne, K., and Lundberg, K. 2014. “The Affect Misattribution Procedure: ten years of evidence on reliability, validity, and mechanisms”. Social and Personality Psychology Compass, 8: 672–686. https://doi. org/10.1111/spc3.12148 Peeters, M., Wiers, R. W., Monshouwer, K., van de Schoot, R., Janssen, T., and Vollebergh, W. A. M. 2012. “Automatic processes in at-risk adolescents: the role of alcohol-approach tendencies and response inhibition in drinking behavior”. Addiction, 107: 1939–1946. https://doi.org/10.1111/j.13600443.2012.03948.x Pezzulo, G., Rigoli, F., and Friston, K. J. 2018. “Hierarchical active inference: a theory of motivated control”. Trends in Cognitive Sciences, 22: 294–306. https://doi.org/10.1016/j.tics.2018.01.009 Pickard, H. 2020. “What we’re not talking about when we talk about addiction”. Hastings Center Report, 50: 37–46. https://doi.org/10.1002/hast.1172. Power, P., Barnes-Holmes, D., Barnes-Holmes, Y., and Stewart, I. 2009. “The Implicit Relational Assessment Procedure (IRAP) as a measure of implicit relative preferences: a first study”. The Psychological Record, 59: 621–640. Pronk, T., Wiers, R. W., Molenkamp, B., and Murre, J. 2020. “Mental chronometry in the pocket? Timing accuracy of web applications on touchscreen and keyboard devices”. Behavior Research Methods, 52: 1371–1382. https://doi.org/10.3758/s13428-019-01321-2 Reich, R. R., Below, M. C., and Goldman, M. S. 2010. “Explicit and implicit measures of expectancy and related alcohol cognitions: a meta-analytic comparison”. Psychology of Addictive Behaviors, 24: 13–25. https://doi.org/10.1037/a0016556 Reich, R. R., Goldman, M. S., and Noll, J. A. 2004. “Using the false memory paradigm to test two key elements of alcohol expectancy theory”. Experimental and Clinical Psychopharmacology, 12: 102–110. https://doi.org/10.1037/1064-1297.12.2.102 Rinck, M., Wiers, R. W., Becker, E. S., and Lindenmeyer, J. 2018. “Relapse prevention in abstinent alcoholics by Cognitive Bias Modification: clinical effects of combining Approach Bias Modification and Attention Bias Modification”. Journal of Consulting and Clinical Psychology, 86: 1005–1016. Robinson, T. E., and Berridge, K. C. 1993. “The neural basis of drug craving: an incentive-sensitization theory of addiction”. Brain Research Reviews, 18: 247–291. https://doi.org/10.1016/0165-0173(93)90013-P Robinson, T. E., and Berridge, K. C. 2008. “The incentive sensitization theory of addiction: some current issues”. Philosophical Transactions of the Royal Society B: Biological Sciences, 363: 3137–3146. https://doi. org/10.1098/rstb.2008.0093 Rooke, S. E., Hine, D. W., and Thorsteinsson, E. B. 2008. “Implicit cognition and substance use: a metaanalysis”. Addictive Behaviors, 33: 1314–1328. https://doi.org/10.1016/j.addbeh.2008.06.009 Salemink, E., Rinck, M., Becker, E., Wiers, R. W., and Lindenmeyer, J. 2021. “Does comorbid anxiety or depression moderate effects of approach bias modification in the treatment of alcohol use disorders?”. Psychology of Addictive Behaviors. https://doi.org/10.1037/adb0000642 Salemink, E., and Wiers, R. W. 2014. “Alcohol-related memory associations in positive and negative affect situations: drinking motives, working memory capacity, and prospective drinking”. Psychology of Addictive Behaviors, 28: 105–113. https://doi.org/10.1037/a0032806

296

Implicit Cognition and Addiction Sayette, M. A., Martin, C. S., Perrott, M. A., and Wertz, J. M. 2001. “Parental alcoholism and the effects of alcohol on mediated semantic priming”. Experimental and Clinical Psychopharmacology, 9: 409–417. Schoenmakers, T. M., de Bruin, M., Lux, I. F. M., Goertz, A. G., Van Kerkhof, D. H. A. T., and Wiers, R. W. 2010. “Clinical effectiveness of attentional bias modification training in abstinent alcoholic patients”. Drug and Alcohol Dependence, 109: 30–36. https://doi.org/10.1016/j.drugalcdep.2009.11.022 Schoenmakers, T. M., Wiers, R. W., Jones, B. T., Bruce, G., and Jansen, A. T. M. 2007. “Attentional retraining decreases attentional bias in heavy drinkers without generalization”. Addiction, 102: 399–405. https://doi.org/10.1111/j.1360-0443.2006.01718.x Schulte, M. H. J., Cousijn, J., den Uyl, T. E., Goudriaan, A. E., van den Brink, W., Veltman, D. J., Schilt, T., and Wiers, R. W. 2014. “Recovery of neurocognitive functions following sustained abstinence after substance dependence and implications for treatment”. Clinical Psychology Review, 34: 531–550. https://doi.org/10.1016/j.cpr.2014.08.002 Sheeran, P., Klein, W. M. P. P., and Rothman, A. J. 2017. “Health behavior change: moving from observation to intervention”. Annual Review of Psychology, 68: 573–600. https://doi.org/10.1146/ annurev-psych-010416-044007 Shono, Y., Ames, S. L., and Stacy, A. W. 2016. “Evaluation of internal validity using modern test theory: application to word association”. Psychological Assessment, 28: 194–204. https://doi.org/10.1037/pas0000175 Shono, Y., Edwards, M. C., Ames, S. L., and Stacy, A. W. 2018. “Trajectories of cannabis-related associative memory among vulnerable adolescents: psychometric and longitudinal evaluations”. Developmental Psychology, 54: 1148–1158. https://doi.org/10.1037/dev0000510 Snelleman, M., Schoenmakers, T. M., and van de Mheen, D. 2015. “Attentional bias and approach/avoidance tendencies do not predict relapse or time to relapse in alcohol dependency”. Alcoholism: Clinical and Experimental Research, 39: 1734–1739. https://doi.org/10.1111/acer.12817 Spruyt, A., Lemaigre, V., Salhi, B., Van Gucht, D., Tibboel, H., Van Bockstaele, B., De Houwer, J., Van Meerbeeck, J., and Nackaerts, K. 2015. “Implicit attitudes towards smoking predict longterm relapse in abstinent smokers”. Psychopharmacology, 232: 2551–2561. https://doi.org/10.1007/ s00213-015-3893-2 Stacy, A. W., Ames, S. L., and Grenard, J. L. 2006. “Word association tests of associative memory and implicit processes: theoretical and assessment issues”. In R. W. Wiers and A. W. Stacy, eds., Handbook of implicit cognition and addiction. Thousand Oaks, CA: Sage: 75–90. Stacy, A. W., Pike, J. R., and Lee, A. Y. 2020. “Multiple memory systems, addiction, and health habits: new routes for translational science”. In S. Sussman, ed., The Cambridge handbook of substance and behavioral addictions. Cambridge: Cambridge University Press: 152–170. Stacy, A. W., and Wiers, R. W. 2010. “Implicit cognition and addiction: a tool for explaining paradoxical behavior”. Annual Review of Clinical Psychology, 6: 551–575. https://doi.org/10.1146/annurev. clinpsy.121208.131444 Teige-Mocigemba, S., Penzl, B., Becker, M., Henn, L., and Klauer, K. C. 2016. “Controlling the ‘uncontrollable’: faking effects on the affect misattribution procedure”. Cognition and Emotion, 30: 1470–1484. https://doi.org/10.1080/02699931.2015.1070793 Thush, C., and Wiers, R. W. 2007. “Explicit and implicit alcohol-related cognitions and the prediction of future drinking in adolescents”. Addictive Behaviors, 32: 1367–1383. https://doi.org/10.1016/j. addbeh.2006.09.011 Thush, C., Wiers, R. W., Ames, S. L., Grenard, J. L., Sussman, S., and Stacy, A. W. 2008. “Interactions between implicit and explicit cognition and working memory capacity in the prediction of alcohol use in at-risk adolescents”. Drug and Alcohol Dependence, 94: 116–124. https://doi.org/10.1016/j. drugalcdep.2007.10.019 Tiffany, S. T. 1990. “A cognitive model of drug urges and drug-use behavior: role of automatic and nonautomatic processes”. Psychological Review, 97: 147–168. Underwood, B. J. 1965. “False recognition produced by implicit verbal responses”. Journal of Experimental Psychology, 70: 122–129. Van Der Vorst, H., Krank, M., Engels, R. C. M. E., Pieters, S., Burk, W. J., and Mares, S. H. W. 2013. “The mediating role of alcohol-related memory associations on the relation between perceived parental drinking and the onset of adolescents’ alcohol use”. Addiction, 108: 526–533. https://doi.org/10.1111/add.12042 Van Dessel, P., De Houwer, J., Gast, A., and Smith, C. T. 2015. “Instruction-based approach-avoidance effects: changing stimulus evaluation via the mere instruction to approach or avoid stimuli”. Experimental Psychology, 62: 161–169. https://doi.org/10.1027/1618-3169/a000282

297

Reinout W. Wiers and Alan W. Stacy Van Dessel, P., De Houwer, J., Roets, A., and Gast, A. 2016. “Failures to change stimulus evaluations by means of subliminal approach and avoidance training”. Journal of Personality and Social Psychology, 110: e1–e15. https://doi.org/10.1037/pspa0000039 Van Dessel, P., Hughes, S., and De Houwer, J. 2018. “Consequence-based approach-avoidance training: a new and improved method for changing behavior”. Psychological Science, 29: 1899–1910. https://doi. org/10.1177/0956797618796478 Van Dessel, P., Hughes, S., and De Houwer, J. 2019. “How do actions influence attitudes? An inferential account of the impact of action performance on stimulus evaluation”. Personality and Social Psychology Review, 23: 267–284. https://doi.org/10.1177/1088868318795730 van Deursen, D. S., Salemink, E., and Wiers, R. W. 2016. “Effectiveness of online cognitive bias modification for problem drinkers: preliminary results”. In Oral communication at the 8th world congress of behavioural and cognitive therapies. Melbourne, Australia. Verschure, P. F. M. J. 2016. “Synthetic consciousness: the distributed adaptive control perspective”. Philosophical Transactions of the Royal Society B: Biological Sciences, 371: 20150448. https://doi.org/10.1098/ rstb.2015.0448 Volkow, N. D., Koob, G. F., and McLellan, A. T. 2016. “Neurobiologic advances from the brain disease model of addiction”. New England Journal of Medicine, 374: 363–371. https://doi.org/10.1056/ NEJMra1511480 Wegner, D. M. 2003. “The mind’s best trick: how we experience free will”. Trends in Cognitive Sciences, 7: 65–69. White, J. L., Moffitt, T. E., Caspi, A., Bartusch, D. J., Needles, D. J., and Stouthamer-Loeber, M. 1994. “Measuring impulsivity and examining its relationship to delinquency”. Journal of Abnormal Psychology, 103: 192–205. Wiechert, S., Grafton, B., MacLeod, C., and Wiers, R. W. 2021. “When alcohol-adverts catch the eye: a psychometrically reliable dual-probe measure of attentional bias”. International Journal of Environmental Research and Public Health, 18: 13263. Wiers, C. E., Kühn, S., Javadi, A. H., Korucuoglu, O., Wiers, R. W., Walter, H., Gallinat, J., and Bermpohl, F. 2013. “Automatic approach bias towards smoking cues is present in smokers but not in exsmokers”. Psychopharmacology, 229: 187–197. https://doi.org/10.1007/s00213-013-3098-5 Wiers, C. E., Ludwig, V. U., Gladwin, T. E., Park, S. Q., Heinz, A., Wiers, R. W., Rinck, M., Lindenmeyer, J., Walter, H., and Bermpohl, F. 2015. “Effects of cognitive bias modification training on neural signatures of alcohol approach tendencies in male alcohol-dependent patients”. Addiction Biology, 20: 990–999. https://doi.org/10.1111/adb.12221 Wiers, C. E., Stelzel, C., Gladwin, T. E., Park, S. Q., Pawelczack, S., Gawron, C. K., Stuke, H., Heinz, A., Wiers, R. W., Rinck, M., Lindenmeyer, J., Walter, H., and Bermpohl, F. 2015. “Effects of cognitive bias modification training on neural alcohol cue reactivity in alcohol dependence”. American Journal of Psychiatry, 172: 335–343. https://doi.org/10.1176/appi.ajp.2014.13111495 Wiers, C. E., Stelzel, C., Park, S. Q., Gawron, C. K., Ludwig, V. U., Gutwinski, S., Heinz, A., Lindenmeyer, J., Wiers, R. W., Walter, H., and Bermpohl, F. 2014. “Neural correlates of alcohol-approach bias in alcohol addiction: the spirit is willing but the flesh is weak for spirits”. Neuropsychopharmacology, 39: 688–697. https://doi.org/10.1038/npp.2013.252 Wiers, R. W., Boffo, M., and Field, M. 2018. “What’s in a trial? On the importance of distinguishing between experimental lab studies and randomized controlled trials: the case of cognitive bias modification and alcohol use disorders”. Journal of Studies on Alcohol and Drugs, 79: 333–343. https://doi. org/10.15288/jsad.2018.79.333 Wiers, R. W., Eberl, C., Rinck, M., Becker, E. S., and Lindenmeyer, J. 2011. “Retraining automatic action tendencies changes alcoholic patients’ approach bias for alcohol and improves treatment outcome”. Psychological Science, 22: 490–497. https://doi.org/10.1177/0956797611400615 Wiers, R. W., Field, M., and Stacy, A. W. 2014. “Passion’s slave? Cognitive processes in alcohol and drug abuse”. In K. J. Sher, ed., The Oxford handbook of substance use and substance use disorders. Oxford: Oxford University Press: 311–350. https://doi.org/10.1093/oxfordhb/9780199381678.013.009 Wiers, R. W., Gladwin, T. E., and Rinck, M. 2013. “Should we train alcohol-dependent patients to avoid alcohol?”. Frontiers in Psychiatry, 4: 33. https://doi.org/10.3389/fpsyt.2013.00033 Wiers, R. W., Gladwin, T. E., Hofmann, W., Salemink, E., and Ridderinkhof, K. R. 2013. “Cognitive bias modification and cognitive control training in addiction and related psychopathology: mechanisms, clinical perspectives, and ways forward”. Clinical Psychological Science, 1: 192–212. https://doi. org/10.1177/2167702612466547

298

Implicit Cognition and Addiction Wiers, R. W., Houben, K., Fadardi, J. S., van Beek, P., Rhemtulla, M., and Cox, W. M. 2015. “Alcohol cognitive bias modification training for problem drinkers over the web”. Addictive Behaviors, 40: 21–26. https://doi.org/10.1016/j.addbeh.2014.08.010 Wiers, R. W., Houben, K., Roefs, A., Jong, P. De, Hofmann, W., and Stacy, A. 2010. “Why common sense goes out the window”. In B. Gawronski and K. Payne, eds., Handbook of implicit social cognition: measurement, theory and applications. New York: Guilford: 463–488. Wiers, R. W., Rinck, M., Dictus, M., and Van Den Wildenberg, E. 2009. “Relatively strong automatic appetitive action-tendencies in male carriers of the OPRM1 G-allele”. Genes, Brain and Behavior, 8: 101–106. https://doi.org/10.1111/j.1601-183X.2008.00454.x Wiers, R. W., Rinck, M., Kordts, R., Houben, K., and Strack, F. 2010. “Retraining automatic actiontendencies to approach alcohol in hazardous drinkers”. Addiction, 105: 279–287. https://doi. org/10.1111/j.1360-0443.2009.02775.x Wiers, R. W., Schoenmakers, T., Houben, K., Thush, C., Fadardi, J. S., and Cox, W. M. 2008. “Can problematic alcohol use be trained away?”. In C. R. Martin, ed., Identification and treatment of alcohol dependency. Keswick, UK: M&K Update, Ltd: 187–208. Wiers, R. W., and Stacy, A. W., eds. 2006. Handbook of implicit cognition and addiction. Thousand Oaks, CA: SAGE. https://doi.org/10.1111/j.1467-8721.2006.00455.x Wiers, R. W., van de Luitgaarden, J., van den Wildenberg, E., and Smulders, F. T. Y. 2005. “Challenging implicit and explicit alcohol-related cognitions in young heavy drinkers”. Addiction, 100: 806–819. https://doi.org/10.1111/j.1360-0443.2005.01064.x Wiers, R. W., Van Dessel, P., and Köpetz, C. 2020. “ABC training: a new theory-based cognitive bias modification to foster automatization of alternative choices in the treatment of addiction and related disorders”. Current Directions in Psychological Science, 29: 499–505. Wiers, R. W., Van Gaal, S., and Le Pelley, M. E. 2020. “Akrasia and addiction: neurophilosophy and psychological mechanisms”. In J. Harbecke and C. Herrmann-Pillath, eds., Social neuroeconomics: integrating the neurosciences and the social sciences. London: Routledge. Wiers, R. W., Van Woerden, N., Smulders, F. T. Y., and De Jong, P. J. 2002. “Implicit and explicit alcoholrelated cognitions in heavy and light drinkers”. Journal of Abnormal Psychology, 111: 648–658. https:// doi.org/10.1037/0021-843X.111.4.648. Wiers, R. W., and Verschure, P. F. M. J. 2021. “Curing the broken brain model of addiction: neurorehabilitation from a systems perspective”. Addictive Behaviors, 112: 106602. https://doi.org/10.1016/j. addbeh.2020.106602 Wood, W., Mazar, A., and Neal, D. T. 2022. “Habits and goals in human behavior: separate but interacting systems”.  Perspectives on Psychological Science,  17: 590–605. https://journals.sagepub.com/doi/ full/10.1177/1745691621994226 Zack, M., Toneatto, T., and MacLeod, C. M. 1999. “Implicit activation of alcohol concepts by negative affective cues distinguishes between problem drinkers with high and low psychiatric distress”. Journal of Abnormal Psychology, 108: 518–531. Ziaee, S. S., Fadardi, J. S., Cox, W. M., and Yazdi, S. A. A. 2016. “Effects of attention control training on drug abusers’ attentional bias and treatment outcome”. Journal of Consulting and Clinical Psychology, 84: 861–873.

299

23 PHENOMENOLOGY, PSYCHOPATHOLOGY, AND PRE-REFLECTIVE EXPERIENCE Anthony Vincent Fernandez

Introduction Phenomenological psychopathology is the study of the experience of mental disorders. But this field of research doesn’t simply provide descriptions of what it’s like to live with mental illness, like those found in memoirs, literary depictions, or even some qualitative studies. Rather, phenomenological psychopathologists are typically interested in how the structure of experience, including the “tacit” or “implicit” features of experience, may alter in cases of mental illness. In this chapter, I introduce phenomenology and phenomenological psychopathology by clarifying the kind of implicit experiences that phenomenologists are concerned with. Such an introduction should facilitate critical engagement and collaboration between phenomenologists and researchers working across a variety of disciplines, including psychology, psychiatry, the cognitive sciences, and analytic philosophy of mind. In section 1, I introduce the phenomenological concept of pre-reflective experience, focusing especially on its relation to the concept of implicit experience. In section 2, I introduce the structure of pre-reflective self-consciousness, which has been studied extensively by both classical phenomenologists and contemporary phenomenological psychopathologists. In section 3, I show how phenomenological psychopathologists rely on an account of pre-reflective self-consciousness to better understand the experience of schizophrenia and I outline some of the methodological challenges that arise in this field of research.

1.  Phenomenology and the Pre-Reflective If one reads contemporary studies in phenomenology, it’s not difficult to find references to implicit experience. Thomas Fuchs, for example, has popularized the distinction between implicit and explicit temporality (Fuchs 2013). And one can easily find discussions of implicit senses of selfhood, implicit feelings or affects, and implicit experiences of embodiment. However, despite widespread use of the term “implicit,” this is not a technical term in classical or contemporary phenomenology. Rather, when phenomenologists describe experience as “implicit,” it is usually more accurate to use the term “pre-reflective.” But why should we concern ourselves with this terminological difference? Isn’t a pre-reflective experience, by definition, an implicit experience? It is, after all, an experience that hasn’t arisen to reflective 300

DOI: 10.4324/9781003014584-29

Phenomenology, Psychopathology, and Pre-Reflective Experience

awareness. However, the term “pre-reflective” has additional connotations that are important for understanding the kind of experiences that phenomenologists study. To start, phenomenologists distinguish “pre-reflective” from terms such as “non-conscious” or “unconscious,” which may seem to refer to similar mental phenomena. Referring to something as unconscious suggests that it is not yet part of conscious awareness but may arise to the level of consciousness – a psychoanalyst, for example, might argue that we have unconscious desires that we can only become aware of through various kinds of self-reflective acts, such as those practiced in psychoanalytic therapies. Referring to something as non-conscious, in contrast, means that it cannot, in principle, arise to the level of consciousness – a neuroscientist, for example, might investigate subpersonal neural processes that are required for conscious experience, but can never be directly experienced by a conscious subject. Neither of these concepts, however, adequately capture what phenomenologists mean by the pre-reflective. That’s because both the unconscious and the non-conscious refer to something outside of consciousness. The phenomenological concept of the pre-reflective, in contrast, refers to something that is very much part of conscious experience, yet is not an object of consciousness. How can something be part of conscious experience without being an object of consciousness? Depending on one’s philosophical commitments about the nature of consciousness, this may seem like a counterintuitive notion. However, phenomenologists operate with a deep or rich notion of consciousness in which much of our conscious life is experienced pre-­reflectively, rather than as a reflective object. In the case of pre-reflective self-consciousness, which we’ll cover in more detail in the following section, phenomenologists argue that we have a tacit or implicit experience of ourselves, or of our own experiencing. As Shaun Gallagher and Dan Zahavi explain, The phenomenologists explicitly deny that the self-consciousness that is present the moment I consciously experience something is to be understood in terms of some kind of higher-order monitoring. It does not involve an additional mental state, but is rather to be understood as an intrinsic feature of the primary experience. (Gallagher and Zahavi 2019) In other words, I don’t become aware of myself or my experience only in those moments where I  self-reflect. Rather, these reflective moments can only occur because I’m already aware of myself in a tacit or implicit way. How, then, do phenomenologists investigate this relationship between pre-reflective and reflective experience? Phenomenology is typically characterized as a self-reflective method. But phenomenologists don’t think of self-reflection as a way of bringing what was previously unconscious to the level of conscious awareness. Rather, self-reflection allows us to become explicitly aware of those features of experience that were already part of consciousness in an implicit or tacit manner. According to Zahavi, Edmund Husserl “spoke of reflection as a process that discloses, disentangles, explicates, and articulates all those components and structures that were implicitly contained in the pre-reflective experience” (Zahavi 2006: 88). The act of reflection, therefore, accentuates features of experience that typically operate in the background, allowing us to grasp them in a more concrete or immediate way (Zahavi 2006: 88–89). With all of this in mind, we still need to clarify what kind of pre-reflective experience phenomenologists are interested in. When phenomenologists study the pre-reflective, they’re not concerned with just any aspect of experience that we haven’t reflected upon. They aim to identify and describe those fundamental features that constitute the form or structure of experience, rather than the particular content of one’s experience. Martin Heidegger, for 301

Anthony Vincent Fernandez

example, clarifies the subject matter of his investigation in Being and Time when he says, “there are certain structures which we shall exhibit – not just any accidental structures, but essential ones,” which are “determinative” for the character of human existence (1962: 38). Heidegger, and many other phenomenologists, have described a variety of these structures in their work, including temporality, affectivity, and embodiment. In this chapter, we’ll focus on pre-reflective self-consciousness because this feature of experience has played an especially important role in the phenomenological study of mental disorders.

2.  Pre-Reflective Self-Consciousness Self-consciousness, in both everyday and philosophical discourse, is typically understood as a reflective act. To become self-conscious is to engage in an act of self-reflection in which one’s own consciousness or experience is taken as an object. In this moment, one becomes conscious of oneself, rather than simply being conscious of other objects in the lived world. This kind of self-consciousness is certainly a core feature of human subjectivity. Such acts of reflection allow us to develop and formulate a social identity and to think more carefully about who we are and want to be. Phenomenologists, however, argue that this is not the only kind of self-consciousness. And, importantly, this reflective act is derivative of a deeper, pre-reflective self-consciousness. But what does it mean to be self-conscious in a pre-reflective way? It means that, at a basic level, experience is always first-personal. Whenever I have an experience, I have the tacit sense that it is my experience or that I am the one undergoing the experience. This basic, fundamental sense that the experiences I undergo are mine, that they belong to or happen to me, is not something that typically arises to the level of reflective awareness. As Zahavi puts it, conscious experiences have “the quality of mineness, the fact that the experiences are characterized by first-personal givenness. That is, the experience is given (at least tacitly) as my experience, as an experience I am undergoing or living through” (2006: 16). Therefore, phenomenal consciousness necessarily entails what Zahavi calls a “minimal or thin form of self-awareness” (2006: 16). There are a variety of terms that philosophers and psychologists use to refer to this aspect of experiential life. However, for the sake of simplicity, I will follow Zahavi in referring to it as “mineness” or, more recently, “for-me-ness” (Zahavi 2018). This phenomenological view of self-consciousness is similar to views that have been advanced by some analytic philosophers, such as the views of Owen Flanagan (1992) and Uriah Kriegel (2003, 2004). But the phenomenological view should be sharply distinguished from higher-order theories of consciousness (Carruthers 2017; Rosenthal 2004). On the higherorder account, a first-order mental state becomes conscious by being the object of a higherorder mental state, which may be perception- or thought-like (depending on which theory one subscribes to). While higher-order theorists do agree that self-consciousness is a key feature of conscious experience, they disagree with phenomenologists over how, exactly, a subject becomes self-conscious. On the phenomenological view, pre-reflective self-consciousness is understood as primitive or irreducible, in the sense that it does not depend upon some more basic structure of experience. At this point, one may be concerned that the phenomenological account doesn’t actually have much to say about pre-reflective self-consciousness beyond the claim that we are, in fact, self-conscious in this way. However, one may characterize pre-reflective self-consciousness as primitive or irreducible while also acknowledging that this basic feature of subjectivity is amenable to further analysis and articulation (Gallagher and Zahavi 2012: 63). As we’ll see in the following section, it’s precisely phenomenology’s careful and systematic articulation of basic, 302

Phenomenology, Psychopathology, and Pre-Reflective Experience

fundamental structures of experience that makes it so valuable to the study of mental disorders. Phenomenological studies of schizophrenia, for example, penetrate beyond symptoms like hallucinations and delusions to inquire into how the basic structures of consciousness must have altered for these symptoms to arise in the first place.

3.  The Pre-Reflective and Psychopathology Phenomenology provides conceptual and theoretical foundations used across the human sciences, including in nursing, psychology, sociology, anthropology, and the cognitive sciences. But the longest running and most established application of phenomenology is in the field of psychiatry, producing the interdisciplinary field of phenomenological psychopathology. This field owes its existence, in large part, to Karl Jaspers, who argued that our understanding of mental disorders must be underpinned by the kind of conceptual clarity produced by phenomenological research (Jaspers 1968, 1997). He pointed out that, despite the widespread use of concepts such as “delusion” or “hallucination,” we have a fairly impoverished understanding of what we actually mean when we use these terms. Jaspers believed that phenomenology could provide nuanced descriptions of signs and symptoms, thereby mapping the terrain for psychiatric research and clinical practice. However, when he originally proposed the field of phenomenological psychopathology in 1912, phenomenology itself was still in its infancy. Jaspers’ vision of a careful and systematic description of disordered experience, while certainly valuable, therefore did not incorporate phenomenology’s deeper concern with the fundamental pre-reflective structures of experience. Those psychiatrists and clinical psychologists who came after Jaspers – such as Ludwig Binswanger, Wolfgang Blankenburg, Medard Boss, Kimura Bin, Frantz Fanon, Eugene Minkowski, Erwin Straus, and Hubertus Tellenbach, among many others – did the difficult work of integrating philosophical phenomenology with psychopathological research. Drawing on the work of figures such as Heidegger, Jean-Paul Sartre, and Maurice Merleau-Ponty, in addition to Husserl, they provided not only nuanced descriptions of disordered experience, but also attempted to understand and explain such experiences by appealing to alterations at the implicit, pre-reflective level of consciousness. This thoroughly integrated approach to phenomenological psychopathology still characterizes the field today, with some of the best work being produced by interdisciplinary teams of psychiatrists, clinical psychologists, and philosophers. In this section, I  illustrate how phenomenological psychopathologists study alterations in pre-reflective experience to better understand psychiatric symptoms that may at first seem anomalous or even incomprehensible. Phenomenological psychopathologists have investigated a wide range of disorders, including depressive disorders, personality disorders, and substance misuse disorders. But psychotic disorders, especially schizophrenia spectrum disorders, have been studied in the most detail. For this reason, I’ll confine the following discussion to recent phenomenological studies of schizophrenia. To appreciate just what phenomenology offers the field of psychiatry, we should start with a brief overview of how contemporary psychiatry conceptualizes mental disorders. Since the 1980s, the Diagnostic and Statistical Manual of Mental Disorders (DSM), which is the dominant classificatory manual in North America, has used an operational, polythetic system of classification and diagnosis. This means that each category of disorder includes a list of symptoms, some subset of which the patient needs to exhibit for a specified duration. The other dominant classificatory manual, the International Classification of Diseases (ICD), does not use operational criteria, but instead provides a brief description of the disorder that includes a list of symptoms. Despite this difference, the latest versions of both manuals (DSM-5 and ICD-11) typically present the 303

Anthony Vincent Fernandez

same basic symptoms for each disorder. Phenomenologists are critical of these approaches to conceptualizing disorders for at least two reasons: First, the experiential symptoms are typically described in a superficial manner; second, there is no attempt to find an organizing principle among the apparently diverse symptoms (see, e.g., Fernandez 2016; Parnas and Bovet 2015; Parnas and Zahavi 2002). Phenomenologists argue that without careful, nuanced descriptions of experiential symptoms, psychiatrists risk conflating what are, in fact, distinct psychopathological symptoms or conditions. And, without identifying an underlying organizing principle, there’s no way to understand how or why this cluster of symptoms tend to co-occur. It’s important to emphasize that phenomenologists are not alone in critiquing the DSM and ICD for failing to identify any kind of meaningful relationship among signs and symptoms, or to inquire into potential causes. Other philosophers and cognitive scientists have also sought to remedy this shortcoming (see Murphy 2008, 2010 for an overview of some of these approaches). Even within psychiatry, we find new approaches that prioritize explanatory accounts. Perhaps the prime example is the US National Institute of Mental Health’s Research Domain Criteria initiative, which prioritizes psychiatric research into brain circuitry, but also aims to investigate disorders across a broad set of domains, including behavior and genetics (Cuthbert 2014). These other explanatory approaches proposed by philosophers, cognitive scientists, and psychiatrists tend to differ from the phenomenological approach in at least one key respect: They typically prioritize mechanistic explanations at biological, neurological, or cognitive levels. Phenomenological accounts, by contrast, prioritize the experiential level. We might assume that these are therefore competing approaches, and they are sometimes presented as such. However, phenomenologists don’t discount the importance of neurological and biological explanations. Rather, they typically argue that, even if our goal is to explain disorders at a neurological or biological level, we still need a clear formulation of the explanandum. Some psychiatrists have even referred to phenomenological psychopathology as “the basic science of psychiatry,” since it maps out the phenomena that other researchers aim to explain (Stanghellini and Broome 2014). While there’s certainly more work to do to clarify how, exactly, phenomenological accounts can guide various kinds of mechanistic explanation, there are at least some recent attempts to characterize this relationship (Fernandez 2019; Pokropski 2019, 2021). To return to the classificatory frameworks of the DSM and ICD, we can look at how schizophrenia is classified in the DSM-5. According to this manual, schizophrenia should be diagnosed only if the patient exhibits two or more of the following five symptoms over the same one-month period, where at least one of these symptoms must be items (1), (2), or (3): 1. Delusions 2. Hallucinations 3. Disorganized speech (e.g., frequent derailment or incoherence) 4. Grossly disorganized or catatonic behavior 5. Negative symptoms (e.g., diminished emotional expression or avolition) (American Psychiatric Association 2013: 99) Moreover, while the DSM-5 does provide definitions of these symptoms, they remain brief and cursory. Consider the DSM-5 definitions of delusions and hallucinations: “Delusions are fixed beliefs that are not amenable to change in light of conflicting evidence” and “Hallucinations are perception-like experiences that occur without an external stimulus. They are vivid and clear, with the full force and impact of normal perceptions, and not under voluntary control” (American Psychiatric Association 2013: 87). These definitions focus primarily on the content of experience (i.e., the delusional belief or the hallucinated object) and fail to characterize the 304

Phenomenology, Psychopathology, and Pre-Reflective Experience

kind of experiential background that makes the delusion or hallucination possible in the first place. By contrast, rather than ask only “What is a delusion?” or “What is a hallucination?”, phenomenologists also inquire into how one’s overall experiential structure must shift or alter such that delusions or hallucinations can occur in the first place. What have phenomenologists concluded by investigating schizophrenia in this way? The dominant phenomenological theory of schizophrenia characterizes it as a self-disorder or an ipseity disturbance – that is, the primary alterations that characterize schizophrenic experience occur in the structure of selfhood, including in pre-reflective self-consciousness. Louis Sass and Josef Parnas highlight two characteristic ways in which one’s sense of self is altered in schizophrenic experience, which they call “hyper-reflexivity” and “diminished self-affection”: this ipseity disturbance has two fundamental and complementary aspects or components. The first is hyperreflexivity, which refers to forms of exaggerated selfconsciousness in which a subject or agent experiences itself, or what would normally be inhabited as an aspect or feature of itself, as a kind of external object. The second is a diminishment of self-affection or autoaffection – that is, of the sense of basic selfpresence, the implicit sense of existing as a vital and self-possessed subject of awareness. (Sass and Parnas 2003: 428) However, they don’t simply describe these features of schizophrenia in detail. They also reflect on why these particular experiences tend to co-occur: In our view, these two features are best conceptualized not as separate processes but as mutually implicative aspects or facets of the intentional activity of awareness. Thus, whereas the notion of hyperreflexivity emphasizes the way in which something normally tacit becomes focal and explicit, the notion of diminished self-affection emphasizes a complementary aspect of this very same process – the fact that what once was tacit is no longer being inhabited as a medium of taken-for-granted selfhood. (Sass and Parnas 2003: 430) One way to think about the relationship between these two disturbances is that when our implicit sense of self is diminished or reduced, we compensate by engaging in acts that make our sense of self explicit, attempting to hold onto an aspect of ourselves that seems to be fading away. However, whether this is best conceptualized as a compensatory relationship is still debated. Sass and colleagues, for instance, suggest that these might be “complementary facets or tightly interacting processes” – but also admit that both conceptions might be required for adequately understanding the relationship (Sass et al. 2018: Supplemental Material Note II). This kind of analysis demonstrates that phenomenology isn’t a purely descriptive enterprise. It can also be explanatory, insofar as it identifies various kinds of motivational relations among different experiential alterations (see, e.g., Parnas and Sass 2008; Sass 2010, 2014). When phenomenological psychopathologists seek out these kinds of motivational relations, they often attempt to identify what Minkowski (1948) calls the trouble générateur or “generating disorder” and Parnas calls the “core gestalt” (2012). The aim is to identify the core disturbance in a particular category of disorder – but also to explain how this core disturbance helps us make sense of the broad range of signs and symptoms that are characteristic of the condition in question, thereby clarifying the overall gestalt structure of the condition. This addresses phenomenology’s second major criticism of psychiatry: that it presents disorders as clusters of apparently unrelated symptoms, with little attempt to understand why these symptoms co-occur. 305

Anthony Vincent Fernandez

As we see from Sass and Parnas’ descriptions of diminished self-affection, it’s precisely the implicit sense of for-me-ness – the sense that I am the subject of my own experience – that is in some respect compromised in schizophrenia. And they argue that this disturbance of prereflective self-consciousness, once properly understood, can also help us understand some of the symptoms of schizophrenia that may at first appear anomalous or incomprehensible. Consider, for instance, the experience of thought insertion. In the DSM-5, thought insertion is defined as “A delusion that certain of one’s thoughts are not one’s own, but rather are inserted into one’s mind” (American Psychiatric Association 2013: 820). A phenomenological psychopathologist will approach this in two ways: First, they’ll ask if this definition accurately characterizes the experience in question; second, they’ll ask how one’s pre-reflective experience must alter in order for such an experience to occur – or, if there’s already some consensus about the pre-reflective alteration that characterizes the relevant diagnosis, then they may ask if this particular symptom can also be understood as a product of this pre-reflective alteration. In the case of phenomenological research on thought insertion, the standard view is that the root of schizophrenic experience is an alteration in pre-reflective self-consciousness. Since thought insertion is a symptom associated with schizophrenia, one should therefore ask if this experience can also be attributed to this alteration – and, if so, how exactly this structure would need to alter for the experience of thought insertion to occur. This is not, however, the kind of insight that a patient can simply report or describe. Rather, identifying which experiential structure has altered and how it has altered can require a complex and sometimes contentious interpretive process. For example, a number of philosophers have argued that thought insertion involves an experience in which one has neither agency for nor ownership of the thought in question, and, therefore, the basic sense of for-me-ness is entirely absent from the experience (see, e.g., Lane 2012, 2015; López-Silva 2018, 2019; Metzinger 2004). Mads Gram Henriksen, Parnas, and Zahavi, by contrast, argue against this interpretation. They agree that pre-reflective self-consciousness is altered and that this can be understood in terms of a distinct change in the sense of for-me-ness that’s typically an integral part of experience (Henriksen et al. 2019). But they argue that this alteration in the sense of for-me-ness should not be understood as the complete absence of this aspect of pre-reflective experience. How do they develop and defend this argument? Henriksen, Parnas, and Zahavi do not defend their position by simply holding to the phenomenological orthodoxy that pre-reflective self-consciousness, including the sense of for-me-ness, is a necessary and universal feature of experience – and, thus, could not be absent from any experience at all. Rather, they argue that many accounts of the experience of thought-insertion are based on misinterpretations of the empirical evidence. They point out that many philosophers who write on the experience of thought insertion developed their accounts based on limited and, in some respects, non-representative examples. Moreover, they argue that many of these philosophers adopt a literal approach to interpreting reports by people who experience thought insertion. As they write, “In an effort to take patient’s descriptions of their own experiences seriously – ‘patient phenomenology’ as it sometimes is termed – philosophers sometimes slip and mistakenly take this to involve taking patients’ descriptions literally” (Henriksen et al. 2019: 4). Even making sense of one’s own experiences from a first-person perspective still requires conceptualization and interpretation. And this means that we’re also capable of misunderstanding, misinterpreting, and ultimately misdescribing our own experiences. In the case of thought insertion, for instance, what one means by “hearing voices” may differ from person to person. Henriksen, Parnas, and Zahavi give examples of patients who deny “hearing voices” because they attribute the voices they hear to real people, and therefore fail to acknowledge that they are experiencing auditory verbal hallucinations (2019: 5). This is just one example of why taking first-person 306

Phenomenology, Psychopathology, and Pre-Reflective Experience

reports seriously does not necessarily require that we take them literally. Rather, to genuinely understand these kinds of experiences, one will typically require empirical methods, such as well-formulated, semi-structured interviews, that can be used to inquire into the experience in question from a variety of angles. By drawing on a broader range of examples than those used by other philosophers engaged in the study of thought insertion, Henriksen, Parnas, and Zahavi argue that thought insertion does not involve a complete loss of for-me-ness and, therefore, conclude that for-me-ness is altered but not absent in the case of thought insertion. As we see in discussions of the relationship between diminished self-affection and hyperreflexivity as well as the debate over how the sense of for-me-ness alters in thought insertion, developing the best phenomenological interpretation of a pre-reflective experience is a challenging task. But phenomenological psychopathologists also face an additional challenge in the collection of experiential data. For the most part, phenomenological psychopathologists assume that they can obtain the data they require through first-person reports, which might be facilitated through various second-person practices, such as interviewing (Henriksen et al. 2022). However, some phenomenological psychopathologists, such as Sass and Elizabeth Pienkos, acknowledge that this approach has a built in selection bias, since it is limited to studying the experiences of those who can provide such reports through verbal or written expression (Sass and Pienkos 2013: 108). But what about those who cannot accurately describe their own experiences, either because they lack adequate capacities for verbal or written expression, or because they do not understand their own experiences in the first place – for instance, in cases of confabulation? Despite its importance to the discipline, this question has received surprisingly little attention in phenomenological psychopathology. The continued success of this field will depend not only on the soundness of its interpretive arguments, but also on its ability to obtain novel experiential data for further interpretation and analysis. Phenomenologists must think carefully, for instance, about how best to formulate a qualitative interview that will help the interviewee accurately report their pre-reflective experiences (Høffding and Martiny 2016; Køster and Fernandez 2021). And it may even be worth rethinking phenomenology’s traditional privileging of firstperson, self-reflective evidence. Classical phenomenologists, such as Husserl and MerleauPonty, sometimes made claims about experience that could not be supported by first-person, self-reflective reports – such as claims about experience in infancy. Considering this, it may be possible for phenomenologists to use behavioral evidence as a replacement for, or in conjunction with, first-person reports (Klinke and Fernandez 2022). Such an approach may open the door for phenomenologists to investigate a broader range of conditions, and to avoid the kind of selection bias that’s built into approaches that rely strictly on interviewing.

Conclusion In this chapter, I’ve introduced the field of phenomenological psychopathology by clarifying the concept of pre-reflective experience. Using the example of pre-reflective self-consciousness, I’ve shown how phenomenologists approach the study of psychopathological conditions. In contrast with psychiatry’s traditional approaches to characterizing mental disorders, which rely on clusters of superficially defined signs and symptoms, phenomenologists develop rich descriptions of psychopathological experience and inquire into how one’s pre-reflective structures must have altered for such an experience to come about. By providing a better understanding of the overall organization of the disorder, phenomenology helps us to make sense of signs and symptoms that may at first appear anomalous or incomprehensible, ultimately illuminating the experience of mental illness in ways that extend far beyond what contemporary psychiatry 307

Anthony Vincent Fernandez

offers. However, while phenomenology may be key to a proper understanding of mental disorders, it should also be seen as just one among many approaches that are critical of mainstream psychiatry and aim to provide more comprehensive accounts.1

Related Topics Chapters 12, 17, 21, 22, 24

Notes 1 I’m thankful to Rosa Ritunnano, Dan Zahavi, and J. Robert Thompson for their comments on an earlier draft of this chapter.

References American Psychiatric Association. 2013. Diagnostic and statistical manual of mental disorders, 5th edition: DSM5. Washington, DC: American Psychiatric Publishing. Carruthers, P. 2017. “Higher-order theories of consciousness”. In S. Schneider and M. Velmans, eds., The Blackwell companion to consciousness. 2nd edn. Oxford: Wiley-Blackwell: 288–297. Cuthbert, B. N. 2014. “The RDoC framework: facilitating transition from ICD/DSM to dimensional approaches that integrate neuroscience and psychopathology”. World Psychiatry, 13: 28–35. Fernandez, A. V. 2016. “Phenomenology, typification, and ideal types in psychiatric diagnosis and classification”. In R. Bluhm, ed., Knowing and acting in medicine. Lanham, MD: Rowman and Littlefield International: 39–58. Fernandez, A. V. 2019. “Phenomenology and dimensional approaches to psychiatric research and classification”. Philosophy, Psychiatry, & Psychology, 26: 65–75. Flanagan, O. J. 1992. Consciousness reconsidered. Cambridge, MA: MIT Press. Fuchs, T. 2013. “Temporality and psychopathology”. Phenomenology and the Cognitive Sciences, 12: 75–104. Gallagher, S., and Zahavi, D. 2012. The phenomenological mind. 2nd edn. London: Routledge. Gallagher, S., and Zahavi, D. 2019. “Phenomenological approaches to self-consciousness”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Summer 2019 edn. plato.stanford.edu/archives/sum2019/ entries/self-consciousness-phenomenological/ Heidegger, M. 1962. Being and time. Trans. J. Macquarrie and E. Robinson. New York: Harper Perennial Modern Classics. Henriksen, M. G., Englander, M., and Nordgaard, J. 2022. “Methods of data collection in psychopathology: the role of semi-structured, phenomenological interviews”. Phenomenology and the Cognitive Sciences, 21: 9–30. Henriksen, M. G., Parnas, J., and Zahavi, D. 2019. “Thought insertion and disturbed for-me-ness (minimal selfhood) in schizophrenia”. Consciousness and Cognition, 74: 102770. Høffding, S., and Martiny, K. 2016. “Framing a phenomenological interview: what, why and how”. Phenomenology and the Cognitive Sciences, 15: 539–564. Jaspers, K. 1968. “The phenomenological approach in psychopathology”. Trans. J. N. Curran. British Journal of Psychiatry, 114: 1313–1223. Jaspers, K. 1997. General psychopathology. Trans. J. Hoenig and M. W. Hamilton. Baltimore, MD: Johns Hopkins University Press. Klinke, M. E., and Fernandez, A. V. 2022. “Taking phenomenology beyond the first-person perspective: conceptual grounding in the collection and analysis of observational evidence”. Phenomenology and the Cognitive Sciences: 1–21. https://doi.org/10.1007/s11097-021-09796-1 Køster, A., and Fernandez, A. V. 2021. “Investigating modes of being in the world: an introduction to phenomenologically grounded qualitative research”. Phenomenology and the Cognitive Sciences: 1–21. https://doi.org/10.1007/s11097-020-09723-w Kriegel, U. 2003. “Consciousness as intransitive self-consciousness: two views and an argument”. Canadian Journal of Philosophy, 33: 103–132. Kriegel, U. 2004. “Consciousness and self-consciousness”. The Monist, 87: 182–205.

308

Phenomenology, Psychopathology, and Pre-Reflective Experience Lane, T. 2012. “Toward an explanatory framework for mental ownership”. Phenomenology and the Cognitive Sciences, 11: 251–286. Lane, T. 2015. “Self, belonging, and conscious experience: a critique of subjectivity theories of consciousness”. In R. Gennaro, ed., Disturbed consciousness: new essays on psychopathology and theories of consciousness. Cambridge, MA: MIT Press: 103–140. López-Silva, P. 2018. “Mapping the psychotic mind: a review on the subjective structure of thought insertion”. Psychiatric Quarterly, 89: 957–968. López-Silva, P. 2019. “Me and I are not friends, just aquaintances: on thought insertion and self-awareness”. Review of Philosophy and Psychology, 10: 319–335. Metzinger, T. 2004. Being no one: the self-model theory of subjectivity. Cambridge, MA: MIT Press. Minkowski, E. 1948. “Phénoménologie et analyse existentielle en psychopathologie”. L’évolution Psychiatrique, 11: 137–185. Murphy, D. 2008. “Levels of explanation in psychiatry”. In K. Kendler and J. Parnas, eds., Philosophical issues in psychiatry: explanation, phenomenology, and nosology. Baltimore: Johns Hopkins University Press: 99–124. Murphy, D. 2010. “Explanation in psychiatry”. Philosophy Compass, 5: 602–610. Parnas, J. 2012. “The core gestalt of schizophrenia”. World Psychiatry, 11: 67–69. Parnas, J., and Bovet, P. 2015. “Psychiatry made easy: operation(al)ism and some of its consequences”. In K. Kendler and J. Parnas, eds., Philosophical issues in psychiatry III: the nature and sources of historical change. Oxford: Oxford University Press: 190–212. Parnas, J., and Sass, L. 2008. “Varieties of ‘phenomenology’ ”. In K. Kendler and J. Parnas, eds., Philosophical issues in psychiatry: explanation, phenomenology, and nosology. Baltimore: Johns Hopkins University Press: 239–278. Parnas, J., and Zahavi, D. 2002. “The role of phenomenology in psychiatric diagnosis and classification”. In M. Maj, W. Gaebel, J. J. López-Ibor, and N. Sartorius, eds., Psychiatric diagnosis and classification. New York: John Wiley & Sons: 137–162. Pokropski, M. 2019. “Phenomenology and mechanisms of consciousness: considering the theoretical integration of phenomenology with a mechanistic framework”. Theory & Psychology, 29: 601–619. Pokropski, M. 2021. Mechanisms and consciousness: integrating phenomenology with cognitive science. London: Routledge. Rosenthal, D. M. 2004. “‪Varieties of higher-order theory”. In R. J. Gennaro, ed., Higher-order theories of consciousness: an anthology. Amsterdam and Philadelphia: John Benjamins Publishing Company: 17–44. Sass, L. 2010. “Phenomenology as description and as explanation: the case of schizophrenia”. In D. Schmicking and S. Gallagher, eds., Handbook of phenomenology and cognitive science. Dordrecht: Springer Netherlands: 635–654. Sass, L. 2014. “Explanation and description in phenomenological psychopathology”. Journal of Psychopathology, 20: 366–376. Sass, L. Borda, J. P., Madeira, L., Pienkos, E., and Nelson, B. 2018. “Varieties of self disorder: a bio-pheno-social model of schizophrenia”. Schizophrenia Bulletin, 44: 720–727. https://doi.org/10.1093/schbul/sby001. Sass, L., and Pienkos, E. 2013. “Varieties of self-experience: a comparative phenomenology of melancholia, mania, and schizophrenia, part I”. Journal of Consciousness Studies, 20: 103–130. Sass, L. A., and Parnas, J. 2003. “Schizophrenia, consciousness, and the self ”. Schizophrenia Bulletin, 29: 427–444. Stanghellini, G., and Broome, M. R. 2014. “Psychopathology as the basic science of psychiatry”. The British Journal of Psychiatry, 205: 169–170. Zahavi, D. 2006. Subjectivity and selfhood: investigating the first-person perspective. Cambridge, MA: MIT Press. Zahavi, D. 2018. “Consciousness, self-consciousness, selfhood: a reply to some critics”. Review of Philosophy and Psychology, 9: 703–718.

309

PART 6

Social Cognition

24 RACE AND THE IMPLICIT ASPECTS OF EMBODIED SOCIAL INTERACTION Jasper St. Bernard and Shaun Gallagher

Frantz Fanon was riding the train on a cold day. He was noticed. “Look a Negro!” declared a fellow passenger. “Look, a Negro! Maman, a Negro!” the young white boy notified his mother. Both the boy’s and Fanon’s bodies were trembling. “[T]he Negro is trembling with cold, the cold that chills the bones, the lovely little boy is trembling because he thinks the Negro is trembling with rage” (Fanon 2008: 93). In response to this perceived rage the young child runs into his mother’s arms. He is afraid the Negro would eat him. What exactly is happening here? Much scholarship has been devoted to the fifth chapter of Black Skin, White Masks (‘The lived experience of the black man’). Most of it attends to Fanon’s reflection on the effect that moments like these have on Fanon (and thus, other black people in similar situations). In this chapter we ask how these moments affect the young white boy (and by extension those who are like him). We’ll attempt to discover what may have motivated this kind of engagement. What caused the boy to perceive Fanon as angry enough to eat him?

Implicit Bias and Racist Attitudes One concept that claims to explain moments like this is the notion of implicit bias. Were implicit biases the motivating force behind this boy’s evaluation of Fanon that day? The first part of this chapter will look at how implicit biases may be at work here (and what it would mean if this were the case). We’ll argue, however that there are gaps in the literature and that an overemphasis on mental states like beliefs makes it difficult to understand moments like this – especially when the actor in question is a child. In contrast, we will shift attention to an account that focuses on the embodied mind (especially in relation to the formation of such things as attitudes). Bodily processes, many of which are implicit, have a critical impact on the formation of race-related attitudes, manifest in moments like this one. The implicit bias literature finds one of its primary roots in the field of social cognition. Mahzarin Banaji and Anthony Greenwald – two of the primary architects of a popular tool used to measure implicit bias – recount how an understanding of implicit bias emerged during a revolutionary time in social psychology. This revolution “introduced new ways of understanding how much of human judgment and behavior are guided by processes that operate outside conscious awareness and conscious control” (Banaji and Greenwald 2013: 17; also see Devine 1989; Kang et al. 2012). They believe this shift to unconscious functioning has an important DOI: 10.4324/9781003014584-31 313

Jasper St. Bernard and Shaun Gallagher

impact on how we can understand human behavior and mental processes (memory, perception, etc.). The ‘signature’ of this new approach is the idea that unconscious thought, including unconscious attitudes, has a definite bearing on action, even if these thoughts are not readily retrievable. On this view, attitudes hold things together in the mind and are involved in the management of preferences, which come in two different forms: reflective (conscious) preferences and automatic preferences. The latter are unconscious, and typically unfamiliar to their owner and difficult to explain (Banaji and Greenwald 2013: 88). They include implicit biases/attitudes (Brownstein and Saul 2016: 8). These attitudes influence or modulate feelings, thoughts, judgments, and/or actions toward objects in the world (including people). In other words, these attitudes play an integral and pervasive (albeit largely unnoticed) role in the background of experience. Many social and cognitive scientists in this field argue that these implicit attitudes “shape all aspects of social life” (Beeghly and Madva 2020: 1). There are many variations of the Implicit Association Test (IAT) (Greenwald et al. 1998; Banaji and Greenwald 2013), designed to measure implicit attitudes in different categories (e.g., consumer products, political values, etc.). We’ll focus on the ‘Race IAT’ and its measurement of implicit attitudes toward racial groups. Early results from this test suggested a general pattern of the pairing of pleasant words with white faces happening at a faster rate than the alternative. The conclusion was drawn that there was generally a more implicit attitudinal preference for white people compared to black people (Greenwald and Krieger 2006: 953). These interpreters found the rate of preference “a surprisingly high figure” (Banaji and Greenwald 2013: 77). It seemed to suggest that many Americans, including those who professed more egalitarian beliefs, suffered from an aversive type of racism. Many theorists have used this understanding of implicit attitudes to make sense of such things as police violence against people of color and discrimination in the justice system (e.g., Correll et al. 2014; Fridell 2016; Fridell and Lim 2016). The IAT, and the implicit bias literature in general, is not without its detractors. One of the primary criticisms interrogates the legitimacy of the science that undergirds the test and its ability to measure implicit attitudes. Questions have been raised about whether implicit biases are a meaningful predictor of behavior (for a good review, see, e.g., Brownstein et al. 2020; Gawronski 2019). We will focus on the emphasis placed on individual psychology in this literature. One concern that emerges is that this emphasis takes attention away from the “more fundamental causes of injustice” (Beeghly and Madva 2020: 1). We want to interrogate the assumptions surrounding the dynamics of how these biases originate. The IAT suggests that the biases begin in the mind and work their way out in bodily behavior. It seemingly assumes there is a strong distinction between the mind (where the attitudes reside) and the body (where the influence of these attitudes may display itself). What follows from this assumption can confound genuine issues related to racist beliefs and their influence. For instance, what if the influence flows in multiple directions? What can a consideration of such dynamics tell us about the relationship between embodied actions, social contexts, and these implicit attitudes?

Embodied and Socially Contextualized Interactions The dominant view in the implicit bias literature “creates the impression that bias exists exclusively ‘in the head’ of individuals” (Beeghly and Madva 2020: 6), which is not necessarily remarkable in light of the cognitive orientation of those who initially spearheaded implicit bias research. However, this stress on both the individual agent and their (conscious or unconscious) mental states can distort the understanding of how implicit bias actually functions. Embodied cognition raises important questions about the problems that may arise when cognitivist, 314

Race and the Implicit Aspects of Embodied Social Interaction

internalist conceptions of mind are (over-)emphasized. We think that embodied cognition can help shed light on claims about the automaticity and inaccessibility of implicit biases. The framework we’ll employ involves seeing embodied processes and social interactions as primary (Gallagher 2020). On this view, embodied interactions have a critical influence on how we perceive others. On embodied cognition approaches the ‘body’ is understood to be a lived or animate body, “the medium through which we engage with the world (ourselves, and others) but also the condition of our having a world” (Ngo 2017: xiv). Maurice Merleau-Ponty’s (2012) enactive (i.e., action-oriented) phenomenology of the body’s role in perception, and the critical analysis of his work is a good starting point. MerleauPonty combines the phenomenological concept of the lived body (Leib) with psychological studies of the body schema (found, for example, in Head 1920; Lhermitte 1939; Schilder 1935) to show that motor control processes, that remain implicit and mainly unconscious, shape the way that we perceive the world (Gallagher 2005). The body schema is a system of sensory-motor processes responsible for the regulation of bodily posture and movement, which generally functions without the conscious awareness of the individual. For Merleau-Ponty the body-in-action tends to efface itself in most of its purposive activities and allows the perceiving agent to move with ease through the world. This ease of movement is what renders conscious monitoring of the body unnecessary. Fanon takes Merleau-Ponty to be defining the body schema as the “slow construction of myself as a body in a spatial and temporal world. . . . It is not imposed on me; it is rather a definitive structuring of myself and the world” (2008: 91). The body schema, on this view, facilitates a dialogue between the body and the world. It does this by conditioning/enabling an agent’s dispositions toward the world, and these dispositions, in turn, have an influence on the individual’s conscious self-image, including body-image. While the body schema functions in a generally non-conscious way as we move through the world, there are times when perception of one’s body comes to the surface, for example, in “limit situations” that involve a forced reflection brought on by pain, discomfort, pleasure, fatigue, some social circumstances, and so forth. Moments like these reveal complex and reciprocal relationships between the body schema, self-awareness, in the form of body image, and perception of the world. These are dynamical relations that are both impacted by, and help shape, interactions with the surrounding environment. Importantly, this includes social interactions. Embodied, enactive approaches to social cognition emphasize that our relations with others are not based on observational mindreading, but are primarily interactive in ways that involve grasping intentions and emotions in the faces, postures, gestures and actions of others. Enactive phenomenology understands the world in terms of pragmatic and social affordances, to use Gibson’s (1979) term. We perceive others in terms of how we can interact with them. The enactive phenomenology that follows Merleau-Ponty (starting, for example in Varela et al. 1991) provides much more detail, incorporating affordance-based ecological psychology, sensory-motor contingency accounts of perception (e.g., Noë 2004), interactionist accounts of social cognition (e.g., De Jaegher et al. 2010; Gallagher 2020; Ratcliffe 2007), and developmental accounts of intersubjectivity (e.g., Trevarthen 1979; Reddy 2008). That such actionoriented explanations are relevant to the topic of implicit racial bias can be seen in Fanon’s initial critique and the subsequent development of critical phenomenology. Fanon argues that there is actually something more basic than the body schema as MerleauPonty characterizes it. He calls it the historical-racial schema which is intersubjectively composed of “a thousand details, anecdotes, and stories.” Below the corporeal schema I had sketched a historico-racial schema. The elements that I  used had been provided for me not by “residual sensations and perceptions 315

Jasper St. Bernard and Shaun Gallagher

primarily of a tactile, vestibular, kinesthetic, and visual character,” but by the other, the white man, who had woven me out of a thousand details, anecdotes, stories. I thought that what I had in hand was to construct a physiological self, to balance space, to localize sensations, and here I was called on for more. (Fanon 2008: 91) The body schema, which typically functions non-consciously or implicitly, crumbles under the gaze of the white man (or even the white boy), according to Fanon, and reveals in the black man a ‘racial epidermal schema’. The latter, which can also be characterized as a conscious body image, can disrupt the work of the body schema leading to its ‘collapse’. Even though Fanon takes some credit for stitching this ‘historical-racial schema’ together that day on the train, he notes that he received the thread from ‘the Other’ (the white man). It was in this moment, brought on by the declaration, ‘Maman, look, a Negro,’ that Fanon became consciously aware of his body in a specific way. The black man’s movements are under observation – by others and by himself. The racial-epidermal schema brings the body forcefully to consciousness, the focus of attention, which changes how one moves in the world. This is, for Fanon, a moment of involuntary reflection forced on him.

Cultural Permeation and Context The idea that the thread of this historico-racial-epidermal schema is provided by the other, suggests that this situation is primarily an inter-relational one, or what Merleau-Ponty calls an intercorporeal relation. That is, it will not be sufficient to understand this situation simply in terms of implicit biases or mental attitudes in the head of the white man, or in this case, the white boy. To see this, consider a recent experiment that explicates some important features of what Merleau-Ponty calls intercorporeity. In one sense, the experiment, conducted by Soliman and Glenberg (2014), has a relatively limited scope in that it focuses on a specific case of synchronic joint action. Despite this, they manage to explore some much larger implications. With respect to the limited scope, Soliman and Glenberg claim that when two people engage in a joint action that requires synchronic coordinated movement (cooperatively moving a wire back and forth to cut through a candle, in the experiment) a “joint body schema” is formed. The experiment is behaviorally simple, but neuroscientifically complex. We can summarize this complexity by saying that activation in specific areas of the brain that register peripersonal (i.e., reachable) space indicate an expansion of that space during the task, similar to the way that during tool use the body schema extends or, correlatively, peripersonal space expands. In effect the individual’s sense of peripersonal space expands to incorporate the other agent. On one reading, it may be simply that processes in each individual agent change – individual body schemas expand, altering subpersonal processes that generate an individual sense of joint agency and a feeling of being in sync with the other. On a more enactivist reading, the two bodies may form a larger dynamical intercorporeal action system, so that the joint body schema belongs only to this larger system. This latter interpretation is supported in studies of entrainment or sensorimotor synchronization (Glowinski et al. 2013; Repp and Su 2013). This interesting finding is the basis for a much larger point that is relevant to the issue at hand. Specifically, Soliman and Glenberg go on to show that these effects (joint body schemas and expanded peripersonal space) are culturally relative. When they conduct the experiment with subjects from cultures characterized by social independence or individuality (e.g., North American) and then with subjects from interdependent cultures (e.g., Asian), they find that the 316

Race and the Implicit Aspects of Embodied Social Interaction

neuronal and behavioral details are different – the effects are stronger in subjects from interdependent cultures. This suggests, as Soliman and Glenberg note, that culture should not be viewed simply as a top-down effect on behavior, but as something that permeates embodied existence – and specifically reaches into body schematic processes. [C]ulture enters the scene not as a self-contained layer on top of behavior, but as the sum of sensorimotor knowledge brought about by a bodily agent interacting in a social and physical context. As such, culture diffuses the web of sensorimotor knowledge, and can only be arbitrarily circumscribed from other knowledge. (Soliman and Glenberg 2014: 209) This idea of cultural permeation1 clearly supports Fanon’s concept of a historico-racial schema, “woven out of a thousand details, anecdotes, stories” including the encounter with the young boy. Here we note that it is not just the boy’s gaze that decomposes Fanon’s body schema; there are words uttered; there are attitudes involved; and there are also witnesses, so there is a complicated context or social situation involved. In this respect we want to emphasize the importance of context. Indeed, Bertram Gawronski (2019) argues that contextual factors are important for any understanding of implicit bias. In fact, the available evidence suggests that contextual factors determine virtually every finding with implicit measures, including (a) their overall scores, (b) their temporal stability, (c) the prediction of future behavior, and (d) the effectiveness of interventions. Although the significance of contextual factors has been identified in the early years of research with implicit measures . . . contextual thinking has still not penetrated the mainstream narrative about implicit bias. (2019: 584–585) In Gawronski’s review, even as he stays relatively close to the mainstream cognitivist account involving mental states, traits, stored information, internal representations, and conceptual categories, he identifies a number of embodied and contextual factors that have been shown to affect the measure of racial bias, including the emotional states of the perceiver, the environment in which a given target person is encountered, and the social role of the perceiver. If in fact, to understand implicit bias we need to consider not just the specifics of dynamical and embodied social interaction, but also context, and cultural permeation, then a focus on “in-the-head” mental states or mental attitudes is not enough. Practically speaking, whether it’s about adjusting our individual behaviors, or developing training policies for the police force, it is not just about changing minds – as if individual cognitive therapy might be the appropriate way to address the issue.2 Rather, it also has to be about changing culture.

The Boy But what about the boy? Was his encounter with Fanon a limit-situation for him as well? Is there a historical-racial schema underneath his body schema? If so, does it collapse as well? Can we say that the boy’s implicit bias and his declaration are racist? If so, in what sense? Here, there is almost too much to take into account to explain the boy’s actions. We will focus on two interconnected factors: affectivity and the issue of development. For both Fanon and the boy, affect, in the broadest sense, is playing a major role in the encounter. Fanon is cold and this manifests in his body – he is shivering. “[T]he Negro is trembling with 317

Jasper St. Bernard and Shaun Gallagher

cold . . . the lovely little boy is trembling because he thinks the Negro is trembling with rage” (Fanon 2008: 93). The boy is shaking with fear, and it seems clear that there are resonance and looping effects that push along a set of responses and behaviors. The role of affect in perception and social cognition, often ignored in traditional cognitivist mindreading accounts, should not be underestimated (Colombetti 2014; Gallagher and Bower 2014; Ratcliffe 2008). To say the boy has a certain mental attitude, even an implicit one, is to tell only a small part of the story. The boy is reacting in an affective-embodied way to his encounter with a black man. Indeed, one might suggest that the trembling is part of, and not a small part, of his attitude. Affect shapes our ability to cope with the surrounding world. In the broadest sense it includes emotion processes, but also more general and basic bodily, autonomic states such as hunger, fatigue, pain. Affect is a form of world-involving intentionality that can modulate bodily behavior without necessarily possessing informational-representational value of any kind. Affect works differently in different contexts. Some important differences may have to do with the way that affective factors are integrated with motoric/agentive factors – the kinetic and kinaesthetic feelings associated with body-schematic processes. In other contexts, even for highly intelligent adults, affect may interfere with our ability to make formal judgments. Consider a study by Danziger et al. (2011) which shows that hunger can bias cognitive processes. The study shows that the application of legal reasoning is not sufficient to explain a legal judgment. Whether the judge is hungry or satiated may play an important role. In one particular courtroom, The percentage of favorable rulings drops gradually from ≈65% to nearly zero within each decision session [e.g., between breakfast and lunch] and returns abruptly to ≈65% after a [food] break. Our findings suggest that judicial rulings can be swayed by extraneous variables that should have no bearing on legal decisions. (Danziger et al. 2011: 1) Affective factors may be “extraneous” to the formal aspects of legal reasoning but are central to perception and behavior. More directly related to the boy’s encounter with Fanon, there is good reason to believe that affect has direct relevance to how we perceive others. Think of the variety of affective possibilities when others are present: we may simply watch passers-by, we may peek into a room of familiar faces, we may be listening attentively, we may be in a heated quarrel, or accidentally meeting someone with whom we were recently quarreling, etc. Each instance comes along with a certain embodied affect, for example, tensing or loosening of posture or facial expression, folding one’s arms, gesturing with one’s hands, or trembling, etc. Bodily affects are also mediated by acquired habits of social behavior; they have a complexity and history that shape behavior. A habit or routine is “wedded to the lived body” (Ngo 2017: 1), and can become part of one’s affective repertoire. Moreover, agents “tend to embody the habits of the social groups to which [they] belong” (Lebouf 2020: 49). This idea motivates Beeghly and Madva to suggest that implicit biases “consist in bodily habits, rather than mental activity per se” (2020: 6). The child begins life under the influence of others, and in a broad culture where he picks up affections and attitudes from other bodies, and forms habits through active participation with these other bodies. The child thus takes up a history that has been heavily influenced by the habits of his social group. These considerations already bring us to issues that concern development. One might ask how a child so young could possibly be racially biased. There is little consensus in the implicit bias literature. On the one hand, it is acknowledged that children as young as 6 years manifest implicit biases indistinguishable from those of adults (e.g., Banse et al. 2010; Gawronski 2019). 318

Race and the Implicit Aspects of Embodied Social Interaction

On the other hand, there is some indication that children can, as early as 6 years, openly report racial preferences. “These patterns suggest that children form explicit biases early on, but gradually learn that these biases are wrong, and not OK to say out loud” (Beeghly and Madva 2020: 4). This seems to suggest that the explicit ideas/attitudes of childhood become the implicit ideas/attitudes of adulthood, as adults become aware of the social pressures exerted on the overt expression of these ideas. This raises an interesting tension, however, since these same individuals are “immersed in a broadly prejudiced society” (Beeghly and Madva 2020: 4). It seems worth asking if something is missing in such accounts; namely, an explanation of how a broadly prejudiced society both promotes and forbids explicit racial bias. The explanation is not simply that concepts are accessible in the environment (Payne et al. 2017) and “children and adults tend to live in the same cultural environments” (Gawronski 2019: 584), although this goes in the right direction of cultural permeation. Greenwald and Krieger (2006) suggest that there are several important early influences on attitudes (including affective experiences). They argue that these influences may have an important impact on the implicit attitudes of individuals, and that this explains “why implicit attitudes generally reveal more bias [than explicit attitudes]” (Greenwald and Krieger 2006: 959). At least part of the explanation is that bodily affect is present in social perception from early infancy, as demonstrated in studies of infant cognition and imitation. As part of primary intersubjective interactions during the first year, affective tuning occurs as early as five to seven months (Gopnik and Meltzoff 1997; Trevarthen 1979). This type of communicative process can happen non-consciously; children can easily pick up attitudes expressed in their care-givers’ postures, facial expressions, gestures, and vocal intonations. It’s not by explicit instruction, but by communication of specific attitudes in non-conscious bodily expressions (Darwall 1998: 265), which define social affordances (what I can do with others) and disaffordances (what I cannot do with others). The developmental processes underpinning our understanding of and attitudes toward others are extensive. Primary intersubjectivity involves early developing sensory-motor capacities for face-to-face interactions that start in the first year of life and typically involve embodied affective processes of give and take between infant and caregiver. These capacities allow us to engage with others by perceptually attuning to their bodily postures, movements, gestures, facial expressions, gaze direction, vocal intonation, etc. We are able to pick up on the other’s intention and emotional expression, and we can respond with our own bodily movements and actions (Trevarthen 1979). Secondary intersubjectivity, starting with possibilities of joint attention around 9 months of age, involves interactions and joint actions in social and pragmatic contexts (Trevarthen and Hubley 1978; see Gallagher 2020). Both primary and secondary intersubjectivity persist throughout the life span. It is through these capabilities established in primary and secondary intersubjectivities that children begin to learn from others what is ‘reasonably’ expected of them in social situations. Further enculturation comes through narrative practices. Children learn many narratives in interaction with others. At 2–3 years children appropriate the narratives of others for their own (Bruner 1996; Nelson 2003a, 2003b; Trevarthen 2013), so that their own narratives are shaped by the attitudes and actions of others, and by the broader narratives in their culture and society. Children learn from narratives (if defined broadly, this would include not just bedtime stories, but television, movies and other media); they learn not only what actions are suited to particular situations but also what reasons for acting are or are not acceptable (Gallagher and Hutto 2008). Through an education in narratives (sometimes enacted in play) children absorb the values and attitudes and learn how to judge an action’s appropriateness. They learn what others ought to do, think and feel, often indexed to the sort of people they are and what social roles they play. 319

Jasper St. Bernard and Shaun Gallagher

That is, they learn the norms associated with the social roles that pervade our everyday environments, and that are continuous with primary and secondary intersubjective interactions (Guajardo and Watson 2002: 307). This is the child’s historical-racial schema, described by Fanon as intersubjectively composed of “a thousand details, anecdotes, and stories.” In effect children stitch their world together into a habitus that they live through their bodies. Considering processes of embodied intersubjectivity, contextualized interaction, habit formation, and narrative practices, suggests that biases are more than simply internal mental states. In the first instance the body is already in a formative relationship with these biases well before these moments of demonstration. Lebouf argues, On an embodied view of implicit bias, to harbor an implicit bias simply means to ‘use the body’ in a biased way. This does not mean that we actively or consciously choose to use our bodies in biased ways. . . . On an embodied conception . . . what it means to be implicitly biased is to interact with the world – whether directly with other persons or with the objects associated with them – according to patterns that are barely in the background of our awareness. (Lebouf 2020: 48) This is also the child’s body schema, developed in the “slow construction of . . . a body in a spatial and temporal world” that structures self and world (Fanon 2008: 91).

Embodied Racism Where exactly does racism reside? The implicit bias literature argues that explicit racism can be seen in overt actions of individuals (e.g. shouting out a racial slur), whereas the implicit kind of racism consists of a set of unconscious ideas that hide in the head of the individual. An account based on embodied cognition endorses a different view. Although racism is neither exclusively conscious, nor entirely non-conscious, and, as Ngo argues, it “sit[s] in the grey region of acquired orientation” (2017: 26), on an embodied cognition account it is not an abstract thing, but is inscribed and circumscribed in bodies, habits, affective attitudes, interactions, narratives, and more generally in cultural practices. Racial attitudes are among those that are learned through bodily and cultural practices. As a result, racism will be “deeply embedded in our bodily habits of movement, gesture, perception, and orientation” (Ngo 2017: 1). In agreement with Fanon, Ngo argues that “the experience of racism and racialization intrudes into this [body-schematic] coordination straining the fluidity of the experience of the body” (2017: 66). Racism’s effect on the motor capacities of the body may not be the same for every body, however. After all, “[i]n a world where racism exists, racialized bodies come predetermined . . . with coded meanings” (Ngo 2017: 16), while a racist body may feel quite at home. As the (black) body is observed walking down the street, watching him suspiciously is not merely a cognitive response to the social milieu, but the very watching enacts the suspicion and reinforces that social milieu. According to Ngo: Discursive representations come into being through their enactment and embodiment. . .. The ease with which such gestures are enacted in response to the racialized ‘other’ – that is to say, the extent to which they are not anomalous or exceptional in the history of one’s body schema, but rather coherent and consistent with it – supports the ascription of habit. (2017: 17, 24) 320

Race and the Implicit Aspects of Embodied Social Interaction

Such habits find their support in the historical-racial schema that holds the world together. Implicit attitudes, because they resonate with the habitual body, do not appear as aberrant, and because they are drawn from the social environment, they do not appear aberrant in society: “Where norms and expectations about different kinds of people are communicated to us in subtle and not so subtle ways. Biases thus reflect inequalities and norms in society at large” (Beeghly and Madva 2020: 7). Accordingly, instead of hiding in the head, racial biases emerge out of a complex relationship between the living body and the world, and may already be on display in one’s bodily habits as one moves through the world. “Look, a Negro! Maman, a Negro!” The declaration was revelatory, for both Fanon and the young boy. Fanon’s articulation of this moment problematizes any conclusion that places implicit (racial) biases exclusively “in the head.” It frustrates the strict distinction between ideas (in the head) and the expression of those ideas (through the body). Such a distinction leads to some misleading conclusions about racism/racist actions. Color-coded bodies play a critical role in the development and experience of racist actions. They show up so often in the world and often without great fanfare (like this moment on the train) because “the racializing schema is already present – and indeed already operative – on a pre-conscious, pre-reflective level, in situations where race is not already explicitly thematized” (Ngo 2017: 69). It is worth noting how early this schema was available to the boy, already settling (settled) into his body. His body was primed to respond to the encounter with a black body in this way. If the encounter disrupted Fanon’s body schema, did it have the same effect on the little boy? There seems to be a related counterpart to Fanon’s weaving of a ‘thousand details, anecdotes, and stories’ being weaved into the little boy. When the boy announces Fanon’s presence, the boy seems to know what it means to have spotted a ‘Negro.’ At least, he knows he is not one. The boy responds to Fanon’s bodily movement as the world has taught him to. His response – running into his mother’s arms – is a reasonable expectation in light of the culture that has permeated his development. Fanon, however, is just trembling because he is cold. The boy’s response resonates this bodily affect, but he trembles because he is afraid Fanon will eat him. The intersubjective resonance that is always potentially empathic is broken or disrupted. The historical-racial schema has facilitated this fracture; the narrative has intervened.3 The young boy’s mother dismisses the moment as innocent. She reassures Fanon that he too is civilized (i.e., not a cannibal) (Fanon 2008: 85). Does she mean that the boy’s behavior was just an aberration rather than the habitual response reflective of the broader civilized world?

Related Topics Chapters 8, 12, 16, 21, 23

Notes 1 The concept of cultural permeation is contrasted with cognitive penetration. See Hutto et al. (2020). For a good discussion of cognitive penetration in the context of implicit bias, see Siegel (2020). 2 Consider the study by Lai et al. (2014), which examined the effects of 18 different interventions, almost all of them purely cognitive (e.g., having white subjects consider different perspectives via imagining thoughts, feelings, and actions of Black individuals). Half of the interventions were ineffective. In a follow-up study (Lai et al. 2016) compared nine of the most effective interventions to reveal that none of them resulted in stable reductions over time (see Gawronski 2019 who relates this to ignoring context). 3 The question of sincerity may not be a helpful one in interpreting this moment. Ngo argues that fear does not exclude the possibility of sincerity and/or vice versa (Ngo 2017: 17). The question of whether he really thought he was going to be eaten is not critical here. That is, sincerity does not diminish the

321

Jasper St. Bernard and Shaun Gallagher impact of the racist action. The historical-racial schema is a problem whether or not someone genuinely ascribes to it.

References Banaji, M., and Greenwald, A. 2013. Blindspot: the hidden biases of good people. New York: Delacorte Press. Banse, R., Gawronski, B., Rebetez, C., Gutt, H., and Morton, J. B. 2010. “The development of spontaneous gender stereotyping in childhood: relations to stereotype knowledge and stereotype flexibility”. Developmental Science, 13: 298–306. Beeghly, E., and Madva, A. 2020. “Introducing implicit bias”. In E. Beeghly and A. Madva, eds., An introduction to implicit bias: knowledge, justice, and the social mind. London: Routledge: 1–19. Brownstein, M., Madva, A., and Gawronski, B. 2020. “Understanding implicit bias: putting the criticism into perspective”. Pacific Philosophical Quarterly, 101: 276–307. https://doi.org/10.1111/papq.12302 Brownstein, M., and Saul, J. 2016. “Introduction”. In M. Brownstein and J. Saul, eds., Implicit bias and philosophy, vol. 1: Metaphysics and epistemology. Oxford: Oxford University Press: 1–19. Bruner, J. S. 1996. The culture of education. Cambridge, MA: Harvard University Press. Colombetti, G. 2014. The feeling body: affective science meets the enactive mind. Cambridge: MIT Press. Correll, J., Hudson, S. M., Guillermo, S., and Ma, D. S. 2014. “The police officer’s dilemma: a decade of research on racial bias in the decision to shoot”. Social and Personality Psychology Compass, 8: 201–213. Danziger, S., Levav, J., Avnaim-Pesso, L. 2011. “Extraneous factors in judicial decisions”. PNAS, 108: 6889–6892. First published online: www.pnas.org/cgi/doi/10.1073/pnas. Darwall, S. 1998. “Empathy, sympathy, care”. Philosophical Studies, 89: 261–282. De Jaegher, H., Di Paolo, E., and Gallagher, S. 2010. “Does social interaction constitute social cognition?”. Trends in Cognitive Sciences, 14: 441–447. Devine, P. G. 1989. “Stereotypes and prejudice: their automatic and controlled components”. Journal of Personality and Social Psychology, 56: 5–18. Fanon, F. 2008. Black skin, white masks. Trans. R. Philcox. New York: Grove Press. Fridell, L. A. 2016. “Racial aspects of police shootings: reducing both bias and counter bias”. Criminology and Public Policy, 15: 481–490. Fridell, L. A., and Lim, H. 2016. “Assessing the racial aspects of police force using the implicit-and counter-bias perspectives”. Journal of Criminal Justice, 44: 36–48. Gallagher, S. 2005. How the body shapes the mind. Oxford: Oxford University Press. Gallagher, S. 2020. Action and interaction. Oxford: Oxford University Press. Gallagher, S., and Bower, M. 2014. “Making enactivism even more embodied”. AVANT/Trends in Interdisciplinary Studies (Poland), 5: 232–247. Gallagher, S., and Hutto, D. 2008. “Understanding others through primary interaction and narrative practice”. In J. Zlatev, T. Racine, C. Sinha, and E. Itkonen, eds., The shared mind: perspectives on intersubjectivity. Amsterdam: John Benjamins: 17–38. Gawronski, B. 2019. “Six lessons for a cogent science of implicit bias and its criticism”. Perspectives on Psychological Science, 14: 574–595. Gibson, J. J. 1979. The ecological approach to visual perception. Boston: Houghton-Mifflin. Glowinski, D., Mancini, M., Cowie, R., Camurri, A., Chiorri, C., and Doherty, C. 2013. “The movements made by performers in a skilled quartet: a distinctive pattern, and the function that it serves”. Frontiers in Psychology, 4: 841. https://doi.org/10.3389/fpsyg.2013.00841 Gopnik, A., and Meltzoff, A. N. 1997. Words, thoughts and theories. Cambridge, MA: MIT Press. Greenwald, A. G., and Krieger, L. H. 2006. “Implicit bias: scientific foundations”. California Law Review, 94: 945–967. Greenwald, A. G., McGhee, D. E., and Schwartz, J. L. K. 1998. “Measuring individual differences in implicit cognition: the Implicit Association Test”. Journal of Personality and Social Psychology, 74: 1464–1480, Guajardo, N. R., and Watson, A. 2002. “Narrative discourse and theory of mind development”. The Journal of Genetic Psychology, 163: 305–325. Head, H. 1920. Studies in neurology. Vol. 2. London: Oxford University Press. Hutto, D., Gallagher, S., Ilundáin-Agurruza, J., and Hipólito, I. 2020. “Culture in mind – an enactivist account: not cognitive penetration but cultural permeation”. In L. J. Kirmayer, S. Kitayama, C. M.

322

Race and the Implicit Aspects of Embodied Social Interaction Worthman, R. Lemelson, and C. A. Cummings, eds., Culture, mind, and brain: emerging concepts, models, applications. New York: Cambridge University Press. Kang, J., Bennett, M., Carbado, D., Casey, P., Dasgupta, N., Faigman, D. et al. 2012. “Implicit bias in the courtroom”. UCLA Law Review, 59: 1124–1186. Lai, C. K., Marini, M., Lehr, S. A., Cerruti, C., Shin, J. L., Joy-Gaba, J. A., Nosek, B. A., et al. 2014. “A comparative investigation of 17 interventions to reduce implicit racial preferences”. Journal of Experimental Psychology: General, 143: 1765–1785. Lai, C. K., Skinner, A. L., Cooley, E., Murrar, S., Brauer, M., Devos, T., Simon, S., et al. 2016. “Reducing implicit racial preferences: II. Intervention effectiveness across time”. Journal of Experimental Psychology: General, 145: 1001. Lebouf, C. 2020. “The embodied biased mind”. In E. Beeghly and A. Madva, eds., An introduction to implicit bias: knowledge, justice, and the social mind. London: Routledge: 41–56. Lhermitte, J. 1939. L’Image de notre corps. Paris: Nouvelle Revue critique. Merleau-Ponty, M. 2012. Phenomenology of perception. Trans. D. A. Landes. London: Routledge. Nelson, K. 2003a. “Narrative and the emergence of a consciousness of self ”. In G. D. Fireman, T. E. J. McVay, and O. Flanagan, eds., Narrative and consciousness. Oxford: Oxford University Press: 17–36. Nelson, K. 2003b. “Self and social functions: individual autobiographical memory and collective narrative”. Memory, 11: 125–136. Ngo, H. 2017. The habits of racism: a phenomenology of racism and racialized embodiment. Lanham: Lexington Books. Noë, A. 2004. Action in perception. Cambridge, MA: MIT Press. Payne, B. K., Vuletich, H. A., and Lundberg, K. B. 2017. “The bias of crowds: how implicit bias bridges personal and systemic prejudice”. Psychological Inquiry, 28: 233–248. Ratcliffe, M. 2007. Rethinking commonsense psychology: a critique of folk psychology, theory of mind and simulation. Basingstoke, UK: Palgrave Macmillan. Ratcliffe, M. 2008. Feelings of being: psychiatry and the sense of reality. Oxford: Oxford University Press. Reddy, V. 2008. How infants know minds. Cambridge, MA: Harvard University Press. Repp, B. H., and Su, Y. H. 2013. “Sensorimotor synchronization: a review of recent research (2006–2012)”. Psychonomic Bulletin & Review, 20: 403–452. Schilder, P. 1935. The image and appearance of the human body. London: Kegan, Paul, Trench, Trubner and Co.; 1950. New York: International University Press. Siegel, S. 2020. “Bias and perception”. In E. Beeghly and A. Madva, eds., An introduction to implicit bias: knowledge, justice, and the social mind. London: Routledge: 99–115. Soliman, T. M., and Glenberg, A. M. 2014. “The embodiment of culture”. In L. Shapiro, ed., The Routledge handbook of embodied cognition. London: Routledge: 207–220. Trevarthen, C. 1979. “Communication and cooperation in early infancy: a description of primary intersubjectivity”. In M. Bullowa, ed., Before speech. Cambridge: Cambridge University Press. Trevarthen, C. 2013. “Born for art, and the joyful companionship of fiction”. In D. Narvaez, J. Panksepp, A. Schore, and T. Gleason, eds., Evolution, early experience and human development: from research to practice and policy. New York: Oxford University Press: 202–218. Trevarthen, C., and Hubley, P.  1978. “Secondary intersubjectivity: confidence, confiding and acts of meaning in the first year”. In A. Lock, ed., Action, gesture and symbol: the emergence of language. London: Academic Press: 183–229. Varela, F. J., Thompson, E., and Rosch, E. 1991. The embodied mind: cognitive science and human experience. Cambridge, MA: MIT Press.

323

25 IMPLICIT SOCIAL COGNITION Shannon Spaulding

1. Introduction Social cognition refers to the various cognitive processes that are involved in social interactions. These include attention, memory, and meta-cognitive processes implicated in interpreting and interacting with other people. Of particular interest here is mentalizing: inferring others’ mental states in order to interpret or anticipate their behavior. Many theorists propose that mentalizing can occur implicitly (Blackburn 1992: 192; Davies and Stone 1995b: 2; Thompson 2012; Spaulding 2015). Typically, implicit mentalizing means that the cognitive processing involved in mentalizing can occur non-consciously, in the absence of voluntary control, and it may be difficult to verbally articulate how the mental process works or even the outcome of that mental process. This idea plays a central role in defenses of the Theory Theory (Nichols and Stich 2003) and the Simulation Theory, especially “low-level” mental simulation (Goldman 2006). Implicit social cognition also comes up in the developmental psychology literature on infants’ folk psychological abilities (Baillargeon et al. 2010; Onishi and Baillargeon 2005) and in the literature on implicit bias.1 In each of these domains, theorists propose that the mental processes underlying social interactions can be more or less automatic, unarticulated, and non-conscious. Positing implicit social cognitive processes is common. However, there is little effort to articulate what counts as implicit social cognition across all these cases. As a result, theorizing about implicit social cognition is extremely disparate across each of these sub-domains. I will not attempt a systematic review of each claim of implicit social cognition. That would take much more space than I have here. Instead, I will present an account of implicit cognition that promises to be a fruitful, unifying account of implicit cognition in general. In the next section, I will present Michael Brownstein’s account of implicit cognition. Though it is a completely general account, I will argue in section 3 that it is well suited to explain various claims of about implicit social cognition. In the final section, I  will discuss some open questions and future directions for research on implicit social cognition.

2.  Implicit Cognition In a recent book, Michael Brownstein (2018) offers a nuanced, empirically well-grounded account of implicit cognition. He argues that implicit cognition involves a cluster of co-activating 324

DOI: 10.4324/9781003014584-32

Implicit Social Cognition

cognitive components: noticing a salient feature, experiencing a low-level tension – a perceptual unquiet – and acting to alleviate that tension. The feature, tension, behavior, and alleviation (FTBA) cluster together and form habitual responses to aspects of our environments. Let’s analyze these FTBA components in more detail. Not every feature of a situation, behavior, or person is important to us. The features that stand out to us depend on our background beliefs and motivations. Noticing a salient feature is a relatively automatic process. However, on this view, when we notice a salient feature, it is not like noticing a neutral, arbitrary feature of the environment. These salient features have rich perceptual content and an imperatival quality to them, a bit like J.J. Gibson’s notion of affordance (1986). These salient features present to us as to-be-acted-upon in a certain way, a way that is determined by our goals, cares, and the situational context. To take a non-social example, consider how elite athletes perform in their sport. A basketball player weaving her way down the court notices an opponent’s unprotected dribble or an opening lane toward the basket. She does not choose to notice these features. Rather, given her experience and education, these features stand out to her in a way that they may not to a less skilled player. Moreover, the unprotected dribble and open lane are not meaningless features; they represent opportunities for action for her. Noticing the feature automatically triggers a kind of tension, a feeling of needing to do something. The tension is low-level, sometimes just an inarticulate sense of “perceptual unquiet,” but it is always geared toward a behavioral response. In the basketball player’s case, noticing the feature may trigger a feeling of needing to reach in for the ball, alert a teammate who is closer to the ball, or initiate a drive to the basket. The tension will vary in terms of strength and valence, depending on the circumstances and one’s goals. For instance, the fatigue the basketball player feels and the relative importance of the game will mediate the strength of the felt tension. The tension motivates a particular spontaneous behavioral response aimed at alleviating the tension. The behavioral response may simply be a reflexive behavior – stepping back or reaching in – or it may be more sophisticated behavior – alerting a teammate or calling a play. The behavioral response one has to the tension is a function of one’s goals, skills, and the context. Successful responses are ones that alleviate the tension and unsuccessful responses are ones that fail to alleviate the tension. Brownstein characterizes the behavioral response mostly in terms of physical dispositions. However, each of the FTBA components are meant to be cognitive components, and the behavioral component includes psychological dispositions, such as focusing your attention, categorizing, and mental rehearsal. In fact, Brownstein discusses psychological dispositions in explaining implicit empathic responses (2018: Ch. 4) and implicit learning involved in acquiring a new skill (2018: Ch. 3). Thus, although many examples of implicit cognition reference physical dispositions, it is appropriate to include psychological dispositions as part of the behavioral responses to felt tension. This aspect of the account is key to making it an account of implicit cognition. The feature, tension, and behavior components tend to cluster together as we get more skilled in a certain domain, and, as a result, we develop habitual responses to particular kinds of features. The tension and behavior form a dynamic feedback system that can be trained and improved over time through a process of trial and error. Our spontaneous inclinations can adapt in response to rewards, like reduced tension, enabling an agent to notice new features and initiate new responses. Our spontaneous inclinations can also adapt in response to punishments, like persisting tension and inability to move on to the next task. For example, a developing basketball player learns that some moves are more successful in certain circumstances and against certain opponents. She’ll learn how to respond to these situations in ways that avoid obvious fouls, or elicit fouls from her opponent, and achieve her immediate and overarching goals. Her physical athletic skills co-evolve with her cognitive athletic skills. The feature, tension, behavior 325

Shannon Spaulding

components dynamically interact and evolve over time as she learns what works to alleviate the tension and what does not work, until her skills reach a plateau. It is important to notice that the basketball player may not be able to articulate, consciously reflect on, or directly control the salience of a particular feature, the tension this generates, or the behavioral response aimed at alleviating the tension. That is not to say it is impossible, but attempting to articulate or reflect on these components in real time may disrupt the athletic performance. It may be possible to articulate or reflect on these elements of performance after the fact, perhaps in learning to hone one’s skill or in teaching the skill to others. However, many elite athletes confess they don’t know how they perform like they do; they just do it. And this is an important feature of Brownstein’s account of implicit cognition. I have used the highly tuned implicit cognition of elite athletes as example to illustrate how implicit cognition is developed and improved. It is important to note, though, that implicit cognition is not simply expertise. Even the novice basketball player has these implicit FTB responses; it’s just that her implicit cognition is not as finely tuned or as effective at achieving her goals, and the FTB components may not be as tightly linked. What makes cognition implicit on this view is that noticing a feature, feeling tension, and acting to alleviate that tension is a process that is typically spontaneous, non-conscious, and not easily articulated. Thus, implicit cognition encompasses both highly skilled expert behavior and relatively unskilled behavior. One final note of clarification: Each of the FTBA components occur on a continuum of consciousness, explicitness, and control. Some components may sometimes be somewhat conscious, somewhat articulatable, or somewhat subject to control. The distinction between implicit and explicit cognition is not a neat dichotomy. This is not a bug in the account; it is a feature. Many findings on implicit cognition demonstrate just this kind of messiness with respect to conscious awareness, voluntary control, and articulation (Sloman 2014). In what follows, I will presuppose this continuity between explicit and implicit cognition and focus on implicit social cognitive processes that cluster on the implicit end of the spectrum.

3.  Implicit Social Cognition The basketball example may, at first, seem quite dissimilar from the kind of implicit cognition posited by social cognition theorists. However, the cognitive components that underlie athletic skills also underlie various other cognitive processes (Michael et al. 2014). Brownstein applies his account of implicit cognition to detective work, implicit bias, aesthetic experience, among many other examples. In this section, I will apply this account to a sampling of implicit social cognition examples. I will argue that this account promises to unify disparate theorizing about implicit social cognition.

3.1 Mentalizing Mentalizing is the attribution of mental states.2 We can attribute mental states to ourselves or others, either in the past, present, future, or counterfactual scenarios. Most of the focus of the empirical and philosophical literature on mentalizing is on the attribution of mental states to others. All of the main general theories of mentalizing hold that mental state attribution can be implicit and probably is implicit much of the time (Spaulding 2015). According to these theories, we understand others’ behavior by attributing mental states and, on that basis, explaining and predicting behavior. The Theory Theory, one of the main accounts of mentalizing, holds that mental state attribution occurs through a process of tacit theorizing, relying a rich body of tacit background knowledge about mental states and behavior (Blackburn 1992; Davies and 326

Implicit Social Cognition

Stone 1995a; Nichols and Stich 2003). The Simulation Theory, another major theory of mentalizing, holds that mental state attribution is the result of (sometimes) tacitly mentally putting yourself in a target’s situation and imagining how you would think, feel, and act in that situation, and then attributing those thoughts, feelings, and behavior to a target (Davies and Stone 1995b; Heal 1998; Goldman 2006). Though they disagree on the process, both Theory Theory and Simulation Theory propose that mentalizing can and often does occur implicitly. By this, they mean that the process (theorizing, simulating, or some combination of both for hybrid theories) and the product of mentalizing (an explanation or prediction) may be unconscious, unarticulated, and proceed without voluntary control. Proponents of 4-E cognition, that is, embedded, embodied, extended, enactive cognition, have challenged the Theory Theory and the Simulation Theory’s notion of implicit mentalizing. Shaun Gallagher (2001, 2005) and Dan Zahavi (2011), among others, argue that there is no evidence that we unconsciously attribute mental states and explain and predict behavior. Gallagher, for instance, argues that mental state inferences may be subconscious, but the product, explanation or prediction, would have to be conscious and phenomenologically assessable. However, careful introspection reveals there is no evidence of explanation and prediction in typical social interactions. Thus, he argues, implicit mentalizing does not occur in ordinary social interactions. On this view, we mentalize only when we are in an unfamiliar situation, or when we encounter very odd behavior (Gallagher 2005: 208–215). Gallagher and other 4-E proponents maintain that implicit explanation and prediction is a non-sensical idea. Explanation and prediction must be deliberative, conscious cognitive acts (Gallagher 2005: 215). What these critiques get right is that unconscious explanation and (perhaps to a lesser extent) unconscious prediction sound odd. Our ordinary notions of explanation and, to some extent, prediction, bring to mind the careful deliberations of scientists. This connotation is unfortunate but not accidental. In the early days of the mentalizing literature, discussion of “theory of mind” were anchored in the Deductive-Nomological model of theories, which holds that theories deduce predictions and explanations. That is, with a general statement of the theory, a statement of the environmental conditions, and probably some auxiliary assumptions about measuring instruments, scientists literally deduce a statement about what has happened (an explanation) or what will happen (a prediction). Although there are few proponents of the Deductive-Nomological model anymore, the language of explanation and prediction stuck around in discussions of theory of mind (what I call mentalizing). In other works, I argue that interpretation and anticipation are more appropriate terminology than explanation and prediction (Spaulding 2015, 2018). Not only do these terms better reflect the phenomena that theories of mentalizing are aiming to explain, they do not connote conscious, deliberative cognition like explanation and prediction do. Thus, the objection to implicit mentalizing in the end turns on a terminological confusion that we can easily resolve. Pluralistic folk psychology, another recent challenge to the Theory Theory/Simulation Theory accounts of mentalizing, argues that most views of mentalizing too narrowly focus on belief and desire attribution and explanation and prediction (Andrews et al. 2020). Unlike 4-E challenges, pluralistic folk psychology accounts do not typically hold that mentalizing rarely occurs (Wolf et al. 2021). Rather, proponents of pluralistic folk psychology typically advocate for expanding the scope of mentalizing.3 According to pluralistic folk psychology, mentalizing involves the attribution of various mental representations, including propositional attitudes, emotions, moods, sensations, character traits, stereotypes, etc. (Westra 2017a, 2017b). These are interwoven with the frameworks, scripts, and schema we employ for understanding social interactions (Andrews 2008). Mentalizing is not limited to explanation and prediction on this view, either. We attribute mental states normatively, in order to socially regulate or mind shape others 327

Shannon Spaulding

(McGeer 2007; Zawidzki 2013). Our mentalizing functions also to boost our own self-esteem, solidify in-group connections, and dismiss out-group members, especially in the context of conflict and competition. Our own motivations influence the strategies for mentalizing – for example, egocentric projection for in-group members, stereotyping for out-group members – the mental state attributions we make, and what we do with these attributions (Spaulding 2017, 2018). Importantly, much of this occurs implicitly. That is, we are often not aware of, in direct control of, or able to clearly articulate the motivations, strategies, attributions, or uses to which we put these attributions. Now that we have on the table various ideas about how mentalizing works, we can ask the question: how does implicit mentalizing work? Consider Brownstein’s FTBA account. Recall that what makes cognition implicit on this view is that each of the FTBA components occur on a continuum of consciousness, explicitness, and control. The features we notice, the tension we experience, and the psychological or physical dispositions this generates range from totally unconscious, ineffable, and incontrollable to somewhat conscious, somewhat articulatable, or somewhat subject to control. Subjects notice a feature, which triggers a targeted low-level tension and a behavioral response aimed at alleviating the felt tension. Importantly, the features we notice, the strength and valence of the tensions we experience, the directed behavioral response we produce are all a function of our situation, goals, and cares. A feature may be salient in one context for one individual, but not another. The same is true for the tension we experience and the behavioral responses we generate. With respect to social cognition, the salient features are social in nature. Given the scope of what can be socially relevant, just about anything could be a salient social feature, for example, motor movements, speech, clothing, nearby artifacts, etc. Registering the salient feature triggers an immediate felt tension. This could be simply a low-level, unarticulated feeling of something looking not quite right, or it could scale up to a fully conscious feeling of puzzlement or confusion. There are many degrees in between these two extremes, and the character of the tension in any given case depends on many factors, including an individual’s history, beliefs, motivations, and the situational context. The tension triggers a behavior. Behavior here is understood quite broadly and includes psychological and physical dispositions. The subject may do a double take to get better view of the salient features and context. She may mentally categorize the person, behavior, or event in terms of a familiar script, trait, or stereotype. She may instinctively label the person with a character trait or generate a plausible inference about a target’s mental states. She may also mentally rehearse other times she has been in a similar situation and project the result of that simulation to the target. Again, the psychological and physical responses are a function of a variety of factors, including the situational context and our own motivations and cares. If successful, the behavior will eliminate the felt tension. If unsuccessful, it will not, and the subject may go on to try out other mental or motor behaviors. Brownstein’s account is well suited to explain implicit mentalizing, in its traditional theorizing or simulating guise, but also in other forms. It is distinctively well suited to explain motivated mentalizing (Spaulding 2017, 2018), an aspect of pluralistic folk psychology described earlier. Motivational factors drive people to notice certain features of people, behavior, and events. These motivational factors influence who we identify as part of our in-group or as part of an out-group. In certain contexts, and with certain goals, an individual may be considered part of a relevant in-group but in other context or with other goals she may be part of an outgroup. In conjunction with in-group/out-group status, our motivations influence our approach to mentalizing (and whether we choose to mentalize at all). For instance, when we are motivated to solidify in-group ties, we may egocentrically project our own mental states on a target. When we are in competition with the target, we may stereotype or (for extreme out-groups) 328

Implicit Social Cognition

interpret them in non-mentalisic dehumanized frameworks, for example, unthinking brutes or unfeeling automata. Motivational factors also shape what we do with the mentalistic inferences we make, that is, whether we explain, justify, dismiss, normatively regulate, or predict a target’s behavior. Each element of the FTBA account is shaped by situational and motivational factors.

3.2 Empathy Brownstein’s account is well suited to explain another significant domain of social cognition, as well: empathy. Empathy is recognizing another person’s emotions and, in response, sharing those emotions. Like mentalizing, empathy ranges from fully conscious and deliberative to implicit. We can work hard to understand another person’s perspective and feel their emotions from their point of view. We can also sometimes simply “see” what others are feeling and spontaneously share their emotions with little or no voluntary control over our empathic response. Sometimes, we may recognize others’ emotions and come to share those emotions in a way that is not obvious to conscious reflection. We simply find ourselves in happier moods around cheerful people, laughing at jokes that others are laughing at, or saddened by others’ sadness.4 Also like mentalizing, empathizing is deeply shaped by motivational factors (Zaki 2014; Weisz and Zaki 2018). Our motivations drive us to avoid or approach engagement with others emotions. For example, when empathizing will lead to our own suffering, involve material costs (like money or resources), and when it will interfere with competition, we are motivated to avoid exposure to others’ emotions. However, empathy also has positive benefits for an individual. When empathizing leads to positive affect, strengthening of social bonds, and demonstrating socially desirable responses, we are motivated to attend to and share others’ emotions. We carry out these motivations to approach or avoid engaging with others emotions through various regulatory strategies, which may be selected and implemented consciously, explicitly, and voluntarily or non-consciously, in an unarticulated way, and with little direct control (Zaki 2014). These regulatory strategies include selecting our situations, for example, choosing a route for our walk home, selecting a television program to watch, choosing news stories to read or figures and organizations to “follow” on social media. Another regulatory strategy is attentional modulation. When we are in a situation in which we are motivated to avoid engaging with others’ emotions, we can literally look away, focus on aspects of the situation that are unrelated to the emotion (e.g., what someone is wearing, features in the background), or simply think about something other than the emotions on display in front of us. When we are motivated to engage with others’ emotions, we focus our attention on the target’s face, what she is saying or doing, and tune out distracting information. Finally, we may carry out our motivation to approach or avoid engaging with others’ emotions by appraising the target. For instance, we may judge that the target is not really suffering; she is just being dramatic. Or we may think to ourselves that the target “had it coming” as a way of avoiding empathizing. When we want to empathize, we may regard the subject as innocent, fragile, or even martyr-like as a way to foster empathy.5 We frequently empathize, and we often are not consciously aware of, in full control of, or able to articulate the various aspects of our empathizing, and these features make empathy another domain of implicit social cognition that we ought to analyze. Brownstein’s account of implicit cognition is well suited to explain many aspects of implicit empathy. In particular, analyzing our implicit empathic responses in terms of FTBA accurately captures the ways in which the situational context and our motivations influence the unfolding of our empathic responses. The context and our motivations determine which features of situation, behavior, or person are salient, the nature of the tension one experiences in response to the target’s display of emotion, the behavior one engages in (situation selection, attentional 329

Shannon Spaulding

modulation, appraisal), and whether the tension is alleviated by the behavioral response. In cases where we are motivated to avoid engaging with others’ emotions, the tension will be eliminated to the extent that we can shift our circumstances or focus away from the target’s emotions or, failing that, negatively appraise the target. The tension will be successfully eliminated when we are motivated to empathize when we are appropriately situated and focused on the target’s emotions and positively appraise the target. Implicit empathy exhibits FTBA cognitive components, is context sensitive like implicit cognition, and driven by our personal motivations. Thus, this is another example of implicit social cognition that fits neatly in the FTBA framework.

3.3  Ontogenetic Development The previous two examples of implicit social cognition focus on the mature cognitive capacities of adults. The final example I will consider concerns implicit social cognition in young children. Clearly, young children do not have the exactly the same concepts or cognitive capacities of adults, whether the subject is implicit or explicit cognition. However, I shall argue that the structural aspects of young children’s implicit cognition can be fruitfully analyzed in terms of Brownstein’s FTBA account. First, we must address an ambiguity in the notion of implicit social cognition in the developmental literature. Sometimes the term refers to the kind of tasks or methodology used to investigate children’s cognitive capacities, and sometimes it refers to the nature of the cognitive processes. Implicit tests of social cognition do not require explicit verbal articulation to pass the tests. These include anticipatory looking, active helping, and violation of expectation. In each of these methodologies, a subjects’ nonverbal behavior (looking, helping, gaze time) indicate her expectations and her inferences about a target’s goals. However, implicit tests do not imply implicit cognition. One could, in principle, employ explicit cognition for an implicit test insofar as one could consciously, deliberatively represent (perhaps even in inner speech, for older children) the events in the test. The notion of implicit social cognition that is relevant for our purposes concerns the nature of the cognitive processes themselves. While we should take care to distinguish implicit methodologies from implicit cognition in what follows, it is likely that passing these implicit tests will involve some elements of implicit cognition. Recall that on the account of implicit cognition we are using here, the implicit/ explicit distinction is continuous, not dichotomous. Implicit cognition is characteristically less conscious, less explicit, and less subject to voluntary control. Thus, while we may not be able to decisively establish that passing an implicit social cognition test requires cognition that is nonconscious, unarticulated, and non-voluntary, in some experiments we may have good reason to think that subjects’ cognition is less conscious, less explicitly articulated, or more spontaneous. In particular, the spontaneous behavioral responses (anticipatory looking and helping) suggest implicit cognition. With that preliminary out of the way, we can discuss the experimental evidence. There is a great deal of controversy over when infants and young children develop and employ mental state concepts. However, there are some widely replicated findings that implicate some level of sensitivity to others’ mental states early on in development. Between 12 and 18 months, children begin to spontaneously help other people without being instructed to do so. In a variety of contexts, including familiar and novel situations, children will distinguish accidents from intentional actions, infer a person’s goal, and help them achieve it. For instance, they will pick up out-of-reach objects and bring them to the person who dropped them, help put things away by holding open a cabinet door, use a newly acquired skill to open a box when a person clumsily fails to retrieve an object from inside, correct an adult’s action when it is going 330

Implicit Social Cognition

wrong, warn against possible mistakes, etc.6 Thus, starting at about one year old, children exercise a robust and flexible capacity to infer others’ goals and act to help them achieve their goals. These implicit tests of young children’s social cognitive abilities do not imply the use of implicit social cognition, as I clarified earlier. However, given the nature of the tests, the age of the children, and the non-dichotomous notion of implicit cognition employed in this chapter, it is likely that children are employing implicit rather than explicit social cognition in some respects. In particular, the spontaneous behavioral responses and the inability of very young children to explicitly articulate (even in inner speech) rationales for behavior indicate that children’s mental processing of these scenes is more spontaneous and tacit and thus on the implicit end of the continuum. We can employ Brownstein’s FTBA account of implicit cognition as a framework to interpret these findings. Recall that what makes cognition implicit on this view is that noticing the feature, experiencing a tension, and acting to relieve the tension are typically spontaneous, unarticulated, and/or non-conscious. In these experiments, infants first notice a salient feature. For example, in the active helping paradigms, infants notice the experimenter’s interrupted trajectory of behavior. The experimenter may drop something, get blocked, or reach for and fail to grab an object. The salient feature presents to the children as something to be acted upon. Noticing the dropped toy, blocked path, or mistake generates a subtle tension. The children look longer at the agent and are spontaneously motivated to help the agent, for example, by retrieving an object or opening a cabinet, removing an obstacle, etc. This spontaneous helping ought to alleviate the felt tension for the child.7 What’s left unanswered here is the nature of the psychological dispositions in the FTBA cluster. The physical response involves longer looking times and spontaneous helping. But the real controversy in this field concerns the psychological dispositions triggered by noticing the salient feature and experiencing a subtle feeling that something is not quite right. Do infants and young children infer knowledge, or belief, or intention? Do they infer some simpler mental construct that is not a proper propositional attitude? Do they try to match what they are observing to a known pattern of behavior? Answering these questions is not a simple task. There is not a singular crucial experiment that will determine one answer to these questions correct and all other answers wrong. Rather, we have to carefully weigh the evidence and make a judicious inference to the best explanation about what kind of psychological dispositions best explains the bulk of the data. Given the uncertainty about some of the data and critiques of the data, I do not think we are in a position to say what the best explanation of the psychological dispositions is. Despite not knowing the exact nature of the psychological dispositions in children’s implicit social cognition, the FTBA framework is useful here. It helps us articulate a structure for implicit social cognition and helps to show that – whatever the nature of the psychological dispositions – implicit social cognition is not an unusual posit. In fact, children’s implicit social cognition operates in the same way as implicit empathy, implicit mentalizing, and implicit cognition more broadly.

4. Conclusion Theorists in many sub-domains of social cognition posit implicit social cognitive processes. However, there is little coordination amongst these sub-domains as to what counts as implicit social cognition. I have argued that we can use a general account of implicit cognition – Michael Brownstein’s FTBA account – to unify these disparate claims about implicit social cognition. I  applied this account to mentalizing, empathy, and children’s socio-cognitive development. The FTBA framework is specific enough to articulate the similarities in these various domains 331

Shannon Spaulding

but broad enough that it does not apply only to one sub-domain or another. In fact, it is meant to explain implicit cognition in general, not just various sub-domains of implicit social cognition. Thus, the FTBA account proves to be extremely useful in understanding the nature of implicit social cognition. However, this account does not answer every question about implicit social cognition. As I  mentioned earlier, the account does not settle the nature of the psychological dispositions activated in response to felt tension. As a result, it does not resolve the debate between the Theory Theory and the Simulation Theory. Nor does it validate or disprove the pluralistic folk psychology claim that mentalizing is a much broader phenomenon than simply theorizing or simulating. Nor does it resolve the debate about the development of children’s social-cognitive skills. Brownstein’s FTBA account of implicit cognition is a framework for having these debates, not a solution to the central questions of these debates. As a framework, this account of implicit cognition turns out to be quite useful. Still, there are two concerns about its application to social cognition. The first concern regards the commitment to affective cognition. On this view, all implicit cognition has an affective component (Duncan and Feldman Barrett 2007). The strength and valence of implicit attitudes can vary greatly. Sometimes, the affective component is just an unarticulated feeling of something being amiss or a sense of perceptual unquiet. Other times, the affective component can be quite vivid and conspicuous for a subject. The view that all cognition is affective is not, as far as I am aware, incompatible with any particular claims about social cognition in the sub-domains I have discussed. It does not beg any questions against any particular accounts. In fact, most social cognition theorists do not directly discuss the scope of affective cognition at all. However, it is a substantial claim about cognition in general that social cognition theorists may resist for independent reasons. It is, thus, an open question for discussion whether this commitment is acceptable. The second concern regarding FTBA and social cognition may be a bit more problematic than the first. Brownstein maintains that implicit attitudes – the representations of implicit ­cognition – are inferentially impoverished and insensitive to the logical form of evidence. Once implicit attitudes are activated (by noticing a salient feature with specific imperatival quality), the rest of the components automatically unfold in a sequence. Because of this automatic sequential unfolding, implicit attitudes exhibit a limited range of potential inferential patterns. If that’s right, then implicit attitudes cannot figure into practical reasoning like propositional attitudes do. This does not imply that implicit attitudes cannot change. As I discussed earlier, implicit cognition can evolve and improve over time based on positive feedback (e.g., reduced tension) or negative feedback (e.g., inability to move on to the next task). But, they are not directly responsive to logically contrary evidence. On Brownstein’s construal, implicit attitudes are not simply non-conscious versions of propositional attitudes. The reason this may be problematic for implicit social cognition is that, in some subdomains, theorists treat implicit social cognition as simply an unconscious version of explicit social cognition. Take mentalizing as an example. According to the Theory Theory, we make inferences to the best explanation about a target’s mental states (based on situational context, folk psychological knowledge, and background information) and generate an explanation or a prediction about a target’s behavior. When mentalizing is implicit, this process runs roughly the same way except that it unfolds without any need for voluntary control or conscious awareness. In other words, implicit mentalizing often is assumed to be unconscious practical reasoning about a target’s mental states and behavior. That does not mesh well with Brownstein’s view of implicit attitudes. It is an open question how we ought to resolve this tension. We could, of course, reject Brownstein’s account of implicit cognition. Or, we could accept it and think harder about the 332

Implicit Social Cognition

character of implicit social cognition. Perhaps implicit mentalizing is not simply unconscious practical reasoning. Alvin Goldman’s account of low-level simulation (Goldman 2006: Ch. 6; de Vignemont 2009) is ahead of the field in terms of offering a nuanced account of implicit mentalizing. He explicitly articulates the ways in which low-level simulation, which is implemented by mirror neurons on his view, differs from high-level simulation, which is implemented by enactive imagination on his view. It may be the case that the rest of the field needs to catch up and articulate similarly nuanced distinctions between high-level and low-level social cognition. A final alternative is to accept Brownstein’s framework but reject the claim that implicit attitudes in social cognition cannot enter into practical reasoning. How would this work? One option is to argue that the FTBA components in social cognition are not as “sticky” as they are in other domains, such as athletic skill or implicit bias. If the FTBA components do not follow in a tight sequence, there is a wider range of possible inferential patterns. How wide this range needs to be in order to qualify as practical reasoning is a difficult question to answer. But, perhaps we do not need a fine-grained answer to that question in order to make space for implicit practical reasoning in social cognition. A second option for this response is to challenge the data that Brownstein presents in defense of the idea that implicit attitudes are inferentially impoverished and insensitive to relevant evidence. There is not enough space here to present such a case, but it is worth noting that the data that bear on this matter are complicated, often ambiguous, and controversial.8 If we were to opt for this response, the implications would apply far beyond implicit social cognition. If successful, the argument would allow implicit attitudes from any domain to potentially enter into practical reasoning. There is much interesting work left to do in figuring out the nature of implicit social cognition. Brownstein’s account of implicit cognition is a good framework in which to ask the important and difficult questions about how implicit social cognition works.9

Related Topics Chapters 1, 2, 7, 8, 16, 17, 21, 24, 26, 29

Notes 1 See Brownstein et al. (2020) for an overview. 2 Over the decades and across disciplines, the attribution of mental states has had various names. These include folk psychology, theory of mind, mindreading, mentalizing, cognitive empathy, perspective taking, etc. I opt for mentalizing because, based on recent interdisciplinary survey data it seems to be the term most theorists recognize and employ to describe the mental state attributions that underlie social interactions. 3 Although, some proponents of pluralistic folk psychology maintain that in thinking about social cognition, we should focus less on the attribution of mental states and more on the various other tools that enable mentalizing. See, for example, Andrews (2008) and Zawidzki (2013). 4 It is important to distinguish emotional contagion – the mere sharing of emotions, of which even newborns are capable – and empathy. Both involve a subject sharing a target’s emotions, but empathy proper requires that the subject recognize the target’s emotion, whereas emotional contagion does not. For implicit empathy, we may not consciously label the emotion or voluntarily choose to infer an emotion, and we may not voluntarily choose or be consciously aware of sharing the emotion. 5 The fact that our personal motivations shape our empathic responses has led to some critiques of empathy as a poor guide to moral decision making. Thinkers like Paul Bloom (2017) and Jesse Prinz (2011a, 2011b) argue that our empathic responses are too idiosyncratic, subject to contextual influences, and shaped by individuals’ goals and cares to reliably guide our moral decision making toward the good. Empathy, they argue, leads to more biased and harmful moral decision making. My own view is that empathy is subject to these critiques, but it is far from alone in this regard. Social cognition itself is deeply

333

Shannon Spaulding shaped by our personal motivations, in-group/out-group dynamics, and biases that serve to reach our goals and solidify group affiliations and boundaries. See Spaulding (2018) for a defense of this view. 6 See Warneken (2015) for a review of these findings. 7 We can use this same FTBA framework to analyze anticipatory looking and violation of expectation studies on infant social cognition. The structure of the explanation will look the same for each kind of implicit social cognitive task. Given the sometimes spotty replication record of anticipatory looking and violation of expectation methodologies for infant social cognition (Dörrenberg et al. 2018; Powell et al. 2018), I will not rest much on these findings. My overall aim is not to establish that infants have or lack certain concepts or cognitive capacities but rather to show that the FTBA account can be utilized to explain implicit cognition of various sorts. 8 See Del Pinal and Spaulding (2018) and Spaulding (2021) for a discussion of these data. In a work in progress, I address some of the concerns about evaluating the stability and evidence-sensitivity of implicit attitudes. 9 Thanks to Michael Brownstein, Peter Railton, and Evan Westra for helpful feedback on the ideas in this chapter.

References Andrews, K. 2008. “It’s in your nature: a pluralistic folk psychology”. Synthese, 165: 13–29. Andrews, K., Spaulding, S., and Westra, E. 2020. “Introduction to folk psychology: pluralistic approaches”. Synthese. https://doi.org/10.1007/s11229-020-02837-3. Baillargeon, R., Scott, R., and He, Z. 2010. “False-belief understanding in infants”. Trends in Cognitive Sciences, 14: 110–118. Blackburn, S. 1992. “Theory, observation and drama”. Mind & Language, 7: 187–230. Bloom, P. 2017. Against empathy: the case for rational compassion. New York: Random House. Brownstein, M. 2018. The implicit mind: cognitive architecture, the self, and ethics. New York: Oxford University Press. Brownstein, M., Madva, A., and Gawronski, B. 2020. “Understanding implicit bias: putting the criticism into perspective”. Pacific Philosophical Quarterly, 101: 276–307. Davies, M., and Stone, T. 1995a. Folk psychology: the theory of mind debate. Oxford: Blackwell. Davies, M., and Stone, T. 1995b. Mental simulation: evaluations and applications. Oxford: Blackwell. Del Pinal, G., and Spaulding, S. 2018. “Symposium on Del Pinal and Spaulding, ‘Conceptual centrality and implicit bias’ ”. Philosophy of Brains. https://philosophyofbrains.com/2018/04/23/symposium-ondel-pinal-and-spaulding-conceptual-centrality-and-implicit-bias.aspx. de Vignemont, F. 2009. “Drawing the boundary between low-level and high-level mindreading”. Philosophical Studies, 144: 457–466. Dörrenberg, S., Rakoczy, H., and Liszkowski, U. 2018. “How (not) to measure infant Theory of Mind: testing the replicability and validity of four non-verbal measures”. Cognitive Development, 46: 12–30. Duncan, S., and Feldman Barrett, L. 2007. “Affect is a form of cognition: a neurobiological analysis”. Cognition and Emotion, 21: 1184–1211. Gallagher, S. 2001. “The practice of mind – theory, simulation or primary interaction”. Journal of Consciousness Studies, 8: 83–108. Gallagher, S. 2005. How the body shapes the mind. New York: Oxford University Press, USA. Gibson, J. J. 1986. The ecological approach to visual perception. Hillsdale: Lawrence Erlbaum. Goldman, A. I. 2006. Simulating minds: the philosophy, psychology, and neuroscience of mindreading. New York: Oxford University Press. Heal, J. 1998. “Co-cognition and off-line simulation: two ways of understanding the simulation approach”. Mind & Language, 13: 477–498. McGeer, V. 2007. “The regulative dimension of folk psychology”. In D. Hutto and M. Ratcliffe, eds., Folk psychology re-assessed. Dordrecht: Springer: 137–156. Michael, J., Christensen, W., and Overgaard, S. 2014. “Mindreading as social expertise”. Synthese, 191: 817–840. https://doi.org/10.1007/s11229-013-0295-z. Nichols, S., and S. Stich. 2003. Mindreading: an integrated account of pretence, self-awareness, and understanding other minds. Oxford: Oxford University Press. Onishi, K. H., and R. Baillargeon. 2005. “Do 15-month-old infants understand false beliefs?”. Science, 308: 255–258.

334

Implicit Social Cognition Powell, L. J., Hobbs, K., Bardis, A., Carey, S., and Saxe, R. 2018. “Replications of implicit theory of mind tasks with varying representational demands”. Cognitive Development, 46: 40–50. https://doi. org/10.1016/j.cogdev.2017.10.004. Prinz, J. 2011a. “Against empathy”. The Southern Journal of Philosophy, 49: 214–233. Prinz, J. 2011b. “Is empathy necessary for morality”. Empathy: philosophical and psychological perspectives, 1: 211–229. Sloman, S. A. 2014. “Two systems of reasoning, an update”. In B. Gawronski and Y. Trope, eds., Dualprocess theories of the social mind. New York: Guilford Press: 107–120. Spaulding, S. 2015. “Phenomenology of social cognition”. Erkenntnis, 80: 1069–1089. https://doi. org/10.1007/s10670-014-9698-6. Spaulding, S. 2017. “Do you see what I see? How social differences influence mindreading”. Synthese, 195: 4009–4030. Spaulding, S. 2018. How we understand others: philosophy and social cognition. New York: Routledge. Spaulding, S. 2021. “Beliefs and biases”. Synthese. https://doi.org/10.1007/s11229-021-03129-0. Thompson, J. R. 2012. “Implicit mindreading and embodied cognition”. Phenomenology and the Cognitive Sciences, 11: 449–466. Warneken, Felix. 2015. “Precocious prosociality: why do young children help?”. Child Development Perspectives, 9: 1–6. Weisz, Erika, and Jamil Zaki. 2018. “Motivated empathy: a social neuroscience perspective”. Current opinion in psychology, 24: 67–71. Westra, E. 2017a. “Character and theory of mind: an integrative approach”. Philosophical Studies, 175: 1217–1241. Westra, E. 2017b. “Stereotypes, theory of mind, and the action-prediction hierarchy”. Synthese, 196: 2821–2846. Wolf, J., Coninx, S., and Newen, A. 2021. “Rethinking integration of epistemic strategies in social understanding: examining the central role of mindreading in pluralist accounts”. Erkenntnis. https://doi. org/10.1007/s10670-021-00486-7. Zahavi, D. 2011. “Empathy and direct social perception: a phenomenological proposal”. Review of Philosophy and Psychology, 2: 541–558. Zaki, J. 2014. “Empathy: a motivated account”. Psychological Bulletin, 140: 1608. Zawidzki, T. W. 2013. Mindshaping: a new framework for understanding human social cognition. Cambridge, MA: MIT Press.

335

26 THE DEVELOPMENT OF IMPLICIT THEORY OF MIND Hannes Rakoczy

The most fundamental aim of developmental cognitive science is to describe and explain trajectories of cognitive ontogeny. What is the starting state with respect to the conceptual repertoire and abilities of a subject? What is the mature state? What are intermediate stages on the way from the initial to the mature state? And what are mechanisms of transition? A recurring theme in many accounts of developmental cognitive science is the transition from earlier implicit to later explicit forms of cognition. In a given domain, how does development progress from implicit forms of representing matters in that domain to explicit ones? In this chapter, we will focus on Theory of Mind as a case study. At the outset, we start with a relatively vague and pre-theoretic notion of the implicit-explicit distinction – roughly to the effect that explicit capacities are those that figure in flexible ways in inference, rational action planning and linguistic expression, and implicit ones are those that fall short of these characteristics in some way or other. Provisionally, we accept the premises that the implicit-explicit distinction is a unitary and sharp one, and that there is a tight correspondence between capacities and tasks such that there are explicit tasks that tap explicit capacities, and implicit tasks that tap implicit capacities. In the course of the chapter, the pre-theoretic notion will then be sharpened, and more nuanced distinctions will be introduced and developed. The simplified assumption of a 1:1 correspondence of competence and task will finally be discussed and questioned. The structure of the chapter is as follows: Section 1 gives an overview of the current state of research on Theory of Mind development, reviewing findings from the last four decades on explicit Theory of Mind, and more recent work on earlier capacities from implicit tasks. Section  2 focuses on the main conceptual question: If there was solid evidence for early competence from implicit tasks, what would this show with regard to the underlying cognitive processes and their development? How would the capacities tapped in implicit tasks relate – in terms of cognitive architecture and development – to those tapped in explicit tasks? Section 3 focuses on the main empirical question: Is there such solid evidence? Original findings as well as a more recent replication crisis are reviewed. Section 4, finally, discusses future directions for investigating implicit and explicit Theory of Mind in more fine-grained ways.

336

DOI: 10.4324/9781003014584-33

The Development of Implicit Theory of Mind

1.  Implicit and Explicit Theory of Mind: The Current State of Affairs Theory of Mind (ToM) is the conceptual framework with which we describe and understand others and ourselves as rational agents with a subjective view on the world. At the heart of ToM lies the ascription of propositional attitudes like beliefs and desires and thus a form of meta-representation (Dennett 1978; Perner 1991): We meta-represent how agents subjectively represent the world as it appears to them (beliefs) and as they want it to be (desires); and we predict and explain their rational actions accordingly (typically, rational agents perform actions that, according to their beliefs, would further their ends). Propositional attitudes are subjective in several respects: Different agents can have representational access to different objects and situations (Adam believes that he has apples in his bag, Eve believes that she has pears in hers); agents can misrepresent a given situation (Adam believes the fruit on the tree are apples whereas in fact they are pears); and agents represent a given situation always from some perspectives, under some aspects, and not under others so that different agents may view the same situation in different ways (Adam may believe that he has an apple tree in front of him while failing to represent it as the tree of knowledge; Eve may believe about the same tree that it is the tree of knowledge without representing it as an apple tree . . .).

1.1  The Development of Explicit Theory of Mind: The Standard Picture When and how does meta-representational ToM develop? Decades of research that addressed this question with explicit tasks have yielded a clear and consistent picture: The capacity for meta-representation and for understanding the subjectivity of propositional attitudes emerges in protracted ways over the preschool years. Even in the first two years of life, children ascribe simple mental states to others and track, for example, what they perceive or intend. Young children are thus sometimes said to operate with a “perception-goal folk psychology”. But this early conceptual framework is fundamentally limited in its cognitive sophistication. It allows children to understand only very basic forms of subjectivity (different agents may perceive or aim at different things), but falls short of a truly meta-representational understanding of subjective, aspectual, potentially inaccurate representations. These limitations are overcome in the course of a crucial conceptual transition around age four (Wellman et al. 2001). From around this age, children begin to master what has become the litmus test for operating with a fully-fledged “belief-desire psychology” and understanding misrepresentation: so-called False Belief (FB) tasks (Wimmer and Perner 1983). In FB tasks, a story protagonist holds a belief that turns out to be false (e.g., she puts an object in box 1, which is then transferred in her absence to box 2), and the child is explicitly asked to predict what the protagonist will do (e.g., where she will look for the object). Young children typically answer incorrectly on the basis of the actual state of affairs (e.g., protagonist will go to where the object really is) whereas children from age four typically answer correctly that the agent will act on the basis of her subjective belief, irrespective of whether it is true or false (e.g., go to the location where she believes the object to be even if it is not there). The very same developmental patterns emerge in superficially very diverse tasks that share a conceptual deep structure in that they all tap meta-representation. Furthermore, performance across these tasks is highly consistent and correlated: Competence emerges in a systematic package such that children tend to become competent at solving different tasks at the same time (Perner and Roessler 2012). Finally, this emerging competence is closely linked to the development of domain-general cognitive and linguistic capacities.

337

Hannes Rakoczy

The standard interpretation of this rich body of empirical evidence has been that the fouryear transition marks a kind of conceptual revolution (Gopnik and Astington 1988; Perner 1991; Rakoczy 2017). Children at this point acquire the conceptual apparatus for meta-representing that agents represent that something is the case; paradigmatically, they come to form metarepresentational beliefs that another agent believes that p. Even though meta-representation is much wider and includes, for example, desires about desires, beliefs about desires etc., in the present chapter we will follow standard practice in the field and focus on beliefs about beliefs as the paradigmatic form of meta-representational ToM.

1.2  Challenges to the Standard Picture New research in the last 15–20 years has challenged this standard interpretation. This research has used nonverbal, implicit measures that do not require subjects to answer questions or engage in high-level action planning. Rather it has tapped looking behaviour, neural signatures, priming or other forms of more or less spontaneous behaviour as dependent measures (for review, see Baillargeon et al. 2016; Baillargeon et al. 2010; Scott and Baillargeon 2017). In violation of expectation (VoE) studies, children first see sequences of events like those in explicit tasks. For example, a protagonist puts an object O in box 1, which is then transferred to box 2 in the presence (true belief (TB) condition) or absence of the protagonist (FB condition). In the test phase, they then see how the protagonist acts either consistently with her beliefs (searches for the object in box 1 in FB/ box 2 in TB) or inconsistently (searches for the object in box 2 in FB/ box 1 in TB). Results of several experiments since the seminal Onishi and Baillargeon (2005) study suggest that children much younger than four (even infants) look longer to, and thus seem surprised by, belief-inconsistent events. In particular, they look longer when the protagonist acts inconsistently with her false belief even if that means that she searches for the object where it really is. Results from spontaneous interaction studies suggest that children from one to two years respond differently and appropriately to an interaction partner as a function of her true or false belief (e.g., D. Buttelmann et al. 2009; Knudsen and Liszkowski 2012; Southgate et al. 2010). For example, in one set of studies the protagonist put object A in box 1, and object B in box 2. The objects were then swapped in her presence (TB) or absence (FB), and the protagonist pointed to box 1 and asked the child, ambiguously, “Can you give it to me?” (Southgate et al. 2010). Children tended to give her the object from box 1 in the TB, but the object from box 2 in the FB condition – which indicates that they took into account the agent’s belief in order to disambiguate what she meant. The broadest body of evidence comes from anticipatory looking (AL) tasks. In such tasks, like in VoE studies, subjects see standard scenarios in which a protagonist forms a true/false belief (e.g., as to whether an object is in box 1 or box 2). Rather than tapping post-hoc looking at (in-)consistent outcomes, a subject’s direction of gaze (towards box 1/box 2) is recorded as an indicator of expectations regarding what the protagonist will do (where she will go to search). These tasks, in contrast to VoE and interaction tasks, are suitable for, and have been used with, a wide variety of populations, across the lifespan, across species and across typical versus atypical conditions. Regarding lifespan development, results from several studies suggest that infants, preschoolers, older children and adults all show belief-consistent AL patterns (e.g., Schneider et al. 2012; Southgate et al. 2007; Surian and Geraci 2012). Two recent comparative studies suggest that perhaps apes (Krupenye et al. 2016) and even monkeys (Hayashi et al. 2020) show analogous AL patterns. And clinical studies suggest that children and adults with Asperger

338

The Development of Implicit Theory of Mind

Syndrome do not spontaneously engage in the same kind of AL patterns (e.g., Schuwerk et al. 2016; Senju et al. 2009). Complementary findings come from studies indicating that young children, before they can solve explicit FB tasks, show different peripheral physiological responses (e.g., pupil dilation) or different neurophysiological signatures to events in which agents act in belief-incongruent as compared to belief-congruent ways (Hyde et al. 2018; Southgate and Vernetti 2014).

2.  What Would Solid Evidence From Implicit Theory of Mind Tasks Really Show? Now, let us assume for the moment that these findings are robust, reliable and replicable (whether this is really the case will be the topic of section 3). If so, what would they mean? What kind of underlying cognitive processes and capacities would they be evidence for? Minimally, what such findings would show is that infants (or adults1) are representationally sensitive to events that involve agents forming (true or false) beliefs – as indicated in their looking and related responses. Something in the infant, we could say, clearly represents something about a situation in which an agent has a belief. But does that amount to the infant holding a meta-representational belief that this agent believes that p? This question actually breaks down into two sub-questions: First, does the infant hold meta-representational beliefs (rather than some simpler kind of representational states) about the agent? Second, does the infant ascribe to the agent beliefs (rather than some simpler kind of states)? These two sub-questions help to map the theoretical territory nicely. Strong nativist early competence accounts answer emphatically “yes” to both sub-questions (e.g., Carruthers 2013; Leslie 2005); skeptical sub-mentalizing and related accounts answer “not at all” to both (e.g., Heyes 2014); and intermediate conceptual change and two-systems accounts answer “not quite” to both (e.g., Apperly and Butterfill 2009; Perner 1991). While the radical responses (2x “yes” vs. 2x “no”) of nativist vs. skeptical accounts seem relatively straightforward, the more nuanced and intermediate 2x “not quite”-responses of conceptual change and two-systems views appear more in need of explanation. Conceptual change accounts assume that children’s ToM develops gradually. Basic mental state concepts (e.g., for perception and goal-directed action in the form of a “perception-goal folk psychology”) develop first, with subsequent refinement and acquisition of more sophisticated concepts (e.g., for subjective beliefs and desires) (Gopnik and Astington 1988; Perner 1991). Relatedly, two-systems accounts assume that ToM is not necessarily a unitary capacity that develops uniformly. Rather, basic ToM processes may be phylogenetically more ancient and develop earlier ontogenetically; these processes may continue to operate rapidly and more or less automatically, with little need for central cognitive resources, throughout the lifespan. Fully-fledged ToM processes, in contrast, may ontogenetically develop in more protracted ways, on the basis of linguistic experience and central cognitive resources. While the fully-fledged ToM processed do involve the metarepresentational ascription of beliefs and other propositional attitudes to other agents, the more basic processes may only involve simpler forms of keeping track of simpler forms of mental states of other agents (Apperly and Butterfill 2009). Both conceptual change and two-systems accounts would thus take intermediate positions regarding the answers to the two types of questions: Implicit tasks may indicate that infants form some kinds of representations of others’ representational states – but that does not necessarily mean that they hold fully-fledged metarepresentational beliefs about others’ fully-fledged beliefs. In the following, we will address the two sub-questions in more detail in turn.

339

Hannes Rakoczy

2.1  Would Solid Findings From Implicit Tasks Show That Infants Form Meta-representational Beliefs? Infants’ looking behaviour in the studies described earlier, if solid, would clearly show that the infant, or something in the infant, representationally tracks something about the agent’s epistemic condition – to put it as neutrally as possible. But would it show, less neutrally, that the infant forms a meta-representational belief . . . (. . . about the agent’s belief)? The standard nativist response is strongly affirmative and goes like this (Baillargeon et  al. 2010; Carruthers 2013; Leslie 2005): Looking time and related “implicit” tasks tap the core competence for ToM that is more or less innate and realized in a more or less modular architecture (perhaps in a “Theory of Mind Module”). The competence involves the use of metarepresentation, and generally the very same conceptual resources as those used later in life in explicit tasks. That “implicit” and “explicit” tasks diverge so massively, with success in the latter lagging years behind the former, does not mean that the two types of tasks tap into different kinds of conceptual competence. Rather, both types of tasks tap the very same conceptual competence – meta-representation – but the “explicit” tasks are artificially difficult because of additional extraneous task demands. In addition to ToM, these tasks require sophisticated linguistic, executive and other competencies. Young children’s failure in standard explicit tasks before age four does not indicate any form of ToM competence deficit, but merely performance limitations caused by the extraneous task demands. In terms of conceptual development, there thus need not be any form of fundamental conceptual change. Even infants form meta-representational beliefs. What develops is nothing about the meta-representational competence itself, but merely how this competence gets integrated with other processes and thus, as a consequence, translates into performance in verbally or otherwise more taxing tasks. This becomes particularly clear when considering the relation of the core meta-representational ToM competence and executive function. A large body of evidence from the last two decades has documented intimate relations between ToM performance in standard explicit tasks and executive function ((Carlson and Moses 2001; Devine and Hughes 2014). The nativist interpretation of this relation comes as a pure “expression account”: Executive function develops substantially over the preschool years and is crucial for “expressing” the ToM core competence – for translating it into performance – in tasks that are taxing in terms of inhibition and the like. Standard explicit FB tests are paradigms of such tasks: In predicting what the mistaken agent will do on the basis of her subjective belief, one’s own perspective, and what is taken to be objective truth (the two usually coincide, of course) have to be put aside. According to nativist expression accounts, early ToM capacities are thus implicit in the following sense: The infant operates with representational states of exactly the same type and with exactly the same conceptual content as later explicit states; it just so happens that this content cannot yet be expressed verbally or put to use in complex tasks. Such expression accounts contrast with emergence accounts according to which executive function is not just an extraneous add-on that helps to express the core meta-representational competence that is already in place, but part of the very meta-representational competence itself that emerges over time. Similar points hold regarding the well-established relation between ToM (tapped in explicit tasks) and language: Nativist accounts merely assign an expressive role to language, whereas other accounts consider language to have a more substantial role in the emergence and constitution of meta-representational competence itself. Nativist accounts thus move from the premise “infants are representationally sensitive to belief-involving situations” to the conclusion “infants form meta-representational beliefs”. Alternative accounts such as conceptual change and dual-process views would point out in response that this move involves a potential inferential gap. What it neglects is that the logical 340

The Development of Implicit Theory of Mind

geography is much more complex (see also Burge 2010, 2018). Yes, robust findings from implicit tasks would show some representational sensitivity in infants (and adults). But there are many ways of being representationally sensitive without forming fully-fledged propositional attitudes (in the present case, meta-representational beliefs). A starting intuition is the following: It is one thing for something to be information in a system; yet another for it to be information to the system (e.g., Clark and Karmiloff-Smith 1993; Cummins 1983). There have been many attempts to explicate this intuition, most of them closely related to one version or another of implicit-explicit distinctions. In the following, I will focus on two of them. The first distinction is between propositional attitudes (beliefs, in particular) and more basic subdoxastic states (Davies 1989; Stich 1978). Cognitive science postulates many types of representational states, but most of them are quite different from the representational states we ascribe at the personal level in our folk psychology. Take, for example, states that represent time in us in some way. Many representational states involved in speech perception track time, for instance in the form of “voice onset time” (VOT) that distinguishes different phonemes from each other. Now, both the state in me that tracks a VOT of, say, 80 ms, and my personal level belief “80 ms is a remarkably short time interval” are states that are about, that represent time in some sense. But they differ in fundamental and crucial ways: Beliefs, in contrast to subdoxastic states, are conceptualized, inferentially integrated and potentially conscious. A second, closely related distinction centers around the accessibility of different types of representational states (Block 1995):2 A given state is access conscious if its content is inferentially promiscuous (i.e., potentially available and integrated in open-ended flexible ways typical of conceptual thought), and available for the rational control of action planning and (typically) for the rational control of language. Again, my occurrent belief that “80 ms is a remarkably short time interval” is inferentially integrated and flexibly available for thought and action, and thus access-conscious, in ways that my speech perceptual tracking of 80 ms VOT in principle is not. From a functionalist point of view, simpler (subdoxastic, non-access conscious) representational states differ from fully-fledged propositional attitudes in their much more restricted functional profiles. That a belief is inferentially promiscuous and available in flexible and open-ended ways for reasoning, planning and action is not an accidental, extraneous fact about it – beliefs are not subdoxastic states that merely happen to be expressible in more varied circumstances. Rather, it is an essential feature of what makes it a belief. From this functionalist perspective, we can make clearer sense of two intuitions: that there is some kind of continuity in the development from simple to more complex states while at the same time there are deep qualitative differences between them. According to functionalism, a given state (in humans, typically neural in nature) realizes or implements a given mental state if it realizes the corresponding functional profile. Now, assume a given (neural) state of kind N finally, after some developmental history, comes to realize the functional profile of mental state of kind M (say, beliefs). Then, in the mature state of the system, N is the core realizer of M – given its functional relations to other elements of the system (the former and the latter together are the total realizers of M; (Shoemaker 1981). For the description of cognitive development, this means the following: There possibly is cognitive continuity in the sense that a given state N is in place early in development that eventually will turn out to be the core realizer of M. But in the early stages the right kinds of functional relations have not been established yet to make N realize M. N only comes to be the core realizer of M over time, once the right kinds of functional relations get established. Turning to the findings from implicit ToM tasks with these distinctions at hand, the following picture emerges: Results from these tasks would indicate simple (subdoxastic) representational states in the infant that track belief-involving situations. But as long as these representational states merely reveal themselves in the guidance of looking behavior, without any evidence that 341

Hannes Rakoczy

they embody anything like the functional profile of beliefs (conceptualization, inferential integration, availability for reasoning and rational action control),3 there is no reason to assume the states in question are the very same kinds of states tapped later in explicit verbal tasks. Rather, the following alternative picture seems more accurate: Over development, there is substantial conceptual change. The cognitive trajectory goes from subdoxastic representations of others’ epistemic situation in infancy to fully-fledged meta-representation from around age four. This also has consequences for describing neurocognitive development in this domain. That a given neural structure that is involved in explicit ToM reasoning later in life (for example, the temporoparietal junction) is already involved in infants’ responses to belief-involving events in looking times (e.g., Hyde et al. 2018), does not mean that the very same cognitive processes (meta-representational judgments) are in operation – in contrast to some nativist interpretations of such findings (e.g., Scott, Roby and Baillargeon in press). Rather, this may simply mean that what will turn out to be a core realizer of true meta-representation later in ontogeny is present earlier but still lacks the right kind of functional connections that would turn it into a (meta-representational) belief. What happens over development neuro-cognitively, then, is not so much local change in this or that area, but functional connection between a given area and others (in particular, those related to language, executive function and more central cognition in general) so that the core area plus its functional connections come to constitute the total realizer of the capacity in question. More recent findings on the neurocognitive changes underlying the transition to explicit ToM are highly compatible with such a picture (Grosse Wiesmann et al. 2017). According to this alternative picture of conceptual change and two-systems-views, early ToM capacities are thus implicit in quite a different sense from that envisaged by nativism. They are not the same capacities as the later ones, merely limited in their expressibility. Rather, they are qualitatively different, subdoxastic states rather than fully-fledged beliefs (see Frankish 2009).

2.2  Would Solid Findings From Implicit Tasks Show That Infants Ascribe Beliefs to Agents? The looking patterns in violation-of-expectation and anticipatory looking FB tasks, if solid, would minimally indicate that infants (and adults) are sensitive to situations in which an agent acts according to her true/false belief. The nativist early competence interpretation of the data goes further: What the looking data suggest, according to nativism, is that infants are not just sensitive to belief-involving situations, but represent beliefs as such. They operate with a concept of “belief ” (and other propositional attitudes); in fact, with the very same concept of belief that older children and adults have. It is just that the infants cannot use this concept as freely and flexibly yet. In contrast, conceptual change and two-systems accounts claim that this nativist line of argumentation, again, involves a potential inferential gap. It fails to do justice to the fact that there are many ways of keeping track of belief-involving situations without using a full-blown concept of “belief ” and thus without ascribing beliefs proper to agents. The notion of “belief ” is a very complex one4 that may simply be beyond the conceptual reach of infants. But simpler proxies can do many of the jobs of belief ascription in more primitive ways. One particular suggestion along these lines goes as follows (Apperly and Butterfill 2009; Butterfill and Apperly 2013): Infants may simply keep track of who registered which kind of information. This can get you quite far. You can engage in level-I perspective-taking (understanding who has seen what; Flavell et al. 1981); understand who has and who hasn’t got knowledge (in the sense of 342

The Development of Implicit Theory of Mind

information access) about some states of affairs (Phillips et al. 2020); and even understand simple forms of outdated/inaccurate informational relations (as a proxy of false beliefs): An agent may have registered that an object O was at place 1, but failed to register that O then moved to place 2, and thus acts on her outdated registration of O at place 1. But what such tracking of simpler representational “belief-like” states lacks is an appreciation not just of what agents have represented, but how they did so. Beliefs are essentially subjective or aspectual states in the sense that they are individuated in conceptually fine-grained ways: It matters fundamentally how the agent represents the object or states of affairs under consideration. Oedipus believes he ought to marry Yocasta, but he certainly does not believe he ought to marry his mother – even though, of course, Yocasta is his mother (unbeknownst to him). The more basic tracking of belief-like states, however, does not admit of such fine-grained, aspectual distinctions that are critical for any fully-fledged propositional attitude concepts. The informational connections of agents to objects and situations that the more basic system can track are purely relational, like non-epistemic seeing, for example (Dretske 1969). If Oedipus registers Yocasta, he thereby registers his mothers. As a consequence, reports about such informational relations are extensional (if agent A registers that O is in location L, and if O is identical to Z, then A registers that Z is in L as well), in contrast to the intensionality of propositional attitude reports (if A believes that O is in L, and O is identical to Z, then it does not necessarily follow that A believes that Z is in L – A may be as unaware about the O=Z identity as Oedipus is about the Yocasta=my mother identity). This two-system account thus predicts clear signature limits of early, implicit ToM capacities: Infants (and adults), in their looking behaviour, can master level-I perspective taking situations, and some FB tasks (those that can be solved by merely keeping track of who has registered what); but they cannot master level-II perspective taking problems, nor FB tasks that require the ascription of aspectual beliefs proper (regarding how an agent represents a situation). Nativism and conceptual change/two-systems views thus make clearly competing predictions with regard to the scope and limits of early ToM capacities: Nativist accounts assume that infants should be able, in principle, to solve all kinds of perspective-taking and FB tasks. In practice, any given limitations that may nevertheless arise in infants’ performance should have nothing to do with the content of the tasks, but only with extraneous (e.g., linguistic or inhibitory) tasks demands. In contrast, conceptual change/two-systems views posit characteristic content-related signature limits such that infants fail all tasks to do with the subjectivity and aspectuality of how agents represent situations. Now, what does the empirical evidence say vis-à-vis these competing positions and predictions? Unfortunately, we currently do now know. The pattern of existing evidence that speaks to this question is complex, mixed and inconclusive. There is evidence from looking time and interaction studies both for (e.g., Edwards and Low 2017; Fizke et al. 2017; Low et al. 2014; Low and Watts 2013; Oktay-Gür et al. 2018; Surtees et al. 2012) and against the signature limits predicted by the two-systems view (e.g., F. Buttelmann et al. 2015; Elekes et al. 2017; Kampis and Kovács 2022; Scott and Baillargeon 2009). More systematic and comprehensive research is thus needed in the future to test for such signature limits. In summary, what evidence from merely implicit ToM tasks would show with regard to the underlying cognitive capacities and processes is very much contested. Strong nativist accounts claim such evidence would show infants have meta-representational beliefs about other agents’ beliefs. Conceptual change and two-systems views concede that such evidence would indicate some form of ToM, some representational sensitivity to others’ representational states; but would claim that the states in question need not be fully-fledged meta-representational beliefs, nor need they involve the ascription of fully-fledged beliefs. 343

Hannes Rakoczy

3.  Is There Such Evidence? All of these foregoing debates arise in response to the question: How should we interpret findings from looking time and related implicit ToM tasks? They presuppose the reliability and validity of these findings. But the premise presupposed – that the findings are reliable and valid – has recently come under empirical attack in the course of a serious replication crisis. For all of the implicit measures reported earlier, many more recent studies with infants, children and adults, often with bigger samples than the initial studies, have either failed to replicate the original findings (thus putting into question their reliability), or have suggested that the effects vanish under more stringent conditions (thus putting into question their validity). A comprehensive and detailed review of this complex empirical situation goes beyond the scope of the present chapter (for such reviews, see Baillargeon et al. 2018; Barone et al. 2019; Kulke and Rakoczy 2018; Poulin-Dubois et al. 2018; Rakoczy 2022). But here is a short summary: Regarding VoE studies, initial positive findings came from relatively few labs. In addition, it has recently been noted that these findings are difficult to interpret since across these initial studies, no consistent set of methodological parameters (such as inclusion criteria, end-of-trial criteria etc.) has been used (Rubio-Fernández 2019). Independent replication results have been mixed, with some successful, some mixed, and some failed replications of original findings (Dörrenberg et al. 2018; Powell et al. 2018; Yott and Poulin-Dubois 2016). Similarly, with regard to interaction studies, the robustness of original findings is currently under dispute: Some studies did replicate original findings at least partially (Fizke et al. 2017; Király et  al. 2018); other studies failed to replicate original findings and thus question their reliability (Dörrenberg et al. 2018; Dörrenberg et al. 2019; Grosse Wiesmann et al. 2016; but see Rubio-Fernandez et al. 2021; Wenzel et al. 2020); finally, yet other studies replicated some original findings but questioned their validity: They produced additional evidence to suggest that these findings need not indicate what they were taken to indicate, rich ToM (Priewasser et al. 2018). As noted earlier, the broadest body of original evidence comes from AL studies. Similarly, for this measure we also have the biggest and most systematic corpus of replication data. In addition, replication data from AL methods are most straightforward to interpret since direct (in contrast to merely conceptual) replications are possible given these studies run in completely automated ways (on eye-tracking machines). Several large-scale replication studies with several hundreds of children and adults, many of them as direct replications as possible, with exactly the same original stimuli and methods, yield a relatively coherent yet disappointing picture: Original findings could largely not be replicated (e.g., Burnside et al. 2018; Dörrenberg et al. 2018; Kulke et al. 2019; Kulke and Rakoczy 2019; Kulke, Reiß, et al. 2018; Kulke, von Duhn, et al. 2018; Kulke, Wübker, et al. 2019; Schuwerk et al. 2018). This also includes a failed selfreplication attempt by the original authors of Southgate et al. (2007), one of the first and most influential AL studies (Kampis et al. 2021). There were two exceptions to this pattern: One condition (“FB1”) from one study (Southgate et al. 2007) could mostly be replicated but is so highly ambiguous that it is impossible to interpret by itself. Another study (Low and Watts 2013; location condition) also stood out in that it could be replicated; but follow-up replication studies showed that this effect vanished once crucial confounds were removed and thus suggest, in terms of validity, that the original task did not measure what it was designed to measure (ToM) (Kulke, von Duhn, et al. 2018). Finally, another crucial question concerns the relation of different implicit measures to each other. If they all tap the same underlying capacity – implicit ToM – then they should all converge and correlate. Such patterns have been widely observed in the case of explicit ToM tasks: 344

The Development of Implicit Theory of Mind

Despite massive differences in surface features, formats and measures, there is substantial correlation, and thus convergent validation between explicit ToM tasks (for review, see Perner and Roessler 2012). In contrast, several recent studies tested for such correlations between implicit tasks. Neither within a given type of task (such as AL, Kulke, Reiß, et al. 2018; Kulke, von Duhn, et al. 2018) nor between types of tasks were there any systematic correlations (e.g., Dörrenberg et al. 2018; Poulin-Dubois and Yott 2017; Yott and Poulin-Dubois 2016). All in all, in light of the complex and inconclusive empirical situation it is currently neither clear whether initial evidence from implicit ToM tasks is robustly replicable (reliability) nor whether it actually measures what it is supposed to measure (validity). Clearly, however, interpreting complex patterns of existing original and replication evidence in post-hoc ways is only of limited epistemic value. What is needed is a concerted effort to look and move ahead. Fortunately, exactly this is now happening: A large-scale international consortium has recently constituted itself under the umbrella of the ManyBabies initiative (Frank et al. 2017). This consortium involves original authors as well as authors of replication studies and many other experts in ToM research and brings together scientists from all theoretical backgrounds. In a true case of “adversarial collaboration” (Mellers et al. 2001), the group collectively agrees upon competing predictions made from diverging theoretical perspectives and how to test them against each other in large-scale, multi-lab, preregistered replication and validation studies. In due time, hopefully, the findings from these large-scale studies will thus allow us to reach firmer conclusions about the (non-)existence of solid evidence from various measures for various forms of early implicit ToM (Schuwerk et al. 2022).

4. Conclusion In this chapter, we have discussed the conceptual question how solid evidence from implicit ToM tasks would need to be interpreted theoretically, and the empirical question whether there is such evidence. With regard to the empirical question, the current situation is very much inconclusive, but progress is on the way in the form of the collaborative replication studies of the ManyBabies consortium. With regard to the conceptual question, I have tried to argue that the space of theoretical options is bigger and more complex than often assumed. If findings from implicit tasks turn out to be robust, this would present solid evidence that infants are representationally sensitive to others’ epistemic situations. But this in itself would neither mean that they form meta-representational beliefs (rather than some simpler representational states) nor that they ascribe fully-fledged beliefs (rather than some simpler representational states). Future work in this area certainly will need to map out this conceptual territory (of representational states about representational states that do not yet amount to beliefs about beliefs) in clearer and more fine-grained ways; and to devise new tasks that allow us to locate the cognitive capacities of a given creature within the space ranging from more basic to fully-fledged forms of (meta-) representational functioning. Another challenge for future theory building and experimentation in this area will be to overcome simplistic presumptions of a 1:1 correspondence of task and process such that a given type of (implicit) task taps certain types of (implicit) processes whereas other types of (explicit) tasks tap other types of (explicit) processes. Research on implicit vs. explicit processes in other domains of cognitive science has found ways to overcome and move beyond such assumption of the “process purity” of types of tasks. In memory research, in particular, so-called processdissociation procedures (Jacoby 1991) have been devised in order to isolate implicit and explicit memory processes in different kinds of direct and indirect tasks. The background assumption here is that, under suitable design conditions, it can be formally spelled out in precise ways that 345

Hannes Rakoczy

and how different kinds of processes play into different kinds of tasks to which degrees. Such an approach has only recently been adopted to the study of some aspects of adult ToM (Todd et al. 2017; Todd et al. 2019). Hopefully, future developmental and cognitive ToM research will adopt and refine such more nuanced approaches to the study of implicit vs. explicit cognition across the board.5

Notes 1 Implicit tasks have widely been used with infants before they master explicit ones in principle. But implicit tasks have also been used with adults in situations where they are supposedly not consciously aware of engaging in any ToM reasoning, for example where they are not asked to reason about a protagonist’s belief but their eyes, so to speak, engage in AL nonetheless (Schneider et al. 2012). In the following, for reasons of simplicity I will often refer to infants, but similar questions apply for the adult data. 2 Block distinguishes between two notions of consciousness: phenomenal (what it is like) and access consciousness (roughly, whether a given representational state is accessible to the subject for use in reasoning, planning, action and language). According to Block, while the two mostly converge, they are conceptually not wholly overlapping and there are thus rare cases of dissociations in both directions. I will here ignore these complications and merely focus on access consciousness. 3 Actually, when it comes to the interpretation of interactive tasks, things are more complicated. These tasks, intuitively, even if they do not document fully-fledged flexible conceptual thought, at least go beyond mere looking time tasks in that they involve some use of the requisite information for action planning (see Carruthers 2013). In fact, if infants revealed seemingly meta-representational deliberation and planning capacities in interactive tasks in the way older children and adults do, with the only exception that they are not yet able to express these capacities verbally, this would be convincing evidence that they operated with something very close to (access-conscious) beliefs about other agents’ beliefs. But so far, the interactive capacities of infants in such tasks have been very limited (not to mention the fact, discussed in the next section, that the reliability and validity even of these very limited interactive tasks has recently come into question). Against this background, and given space limitation, I am here ignoring these complications and focus on looking time as main indicator of early implicit ToM (see, e.g., Newen and Wolf (2020) for further distinctions between looking time and interactive studies on early FB understanding). 4 Beliefs are conceptually structured, holistically related, normatively governed etc., to name just some of those complexities (for details of these arguments, see Burge 2018; Butterfill and Apperly 2013). 5 Thank you for helpful comments on a previous version to Natalie Bleijlevens, Isa Garbisch, Feride Nur Haskaraca Kizilay, Marina Proft, Lydia Schidelko, Britta Schünemann, Rowan Titchener and Lisa Wenzel. I am deeply grateful to Robert Thompson and an anonymous reviewer for incredibly detailed, constructive and helpful feedback on the first version of this chapter.

References Apperly, I. A., and Butterfill, S. A. 2009. “Do humans have two systems to track beliefs and belief-like states?”. Psychological Review, 116: 953–970. https://doi.org/10.1037/a0016923 Baillargeon, R., Buttelmann, D., and Southgate, V. 2018. “Invited commentary: Interpreting failed replications of early false-belief findings: methodological and theoretical considerations”. Cognitive Development, 46: 112–124. https://doi.org/10.1016/j.cogdev.2018.06.001 Baillargeon, R., Scott, R. M., and Bian, L. 2016. “Psychological reasoning in infancy”. Annual Review of Psychology, 67: 159–186. Baillargeon, R., Scott, R. M., and He, Z. 2010. “False-belief understanding in infants”. Trends in Cognitive Sciences, 14: 110–118. https://doi.org/10.1016/j.tics.2009.12.006 Barone, P., Corradi, G., and Gomila, A. 2019. “Infants’ performance in spontaneous-response false belief tasks: a review and meta-analysis”. Infant Behavior and Development, 57: 101350. https://doi. org/10.1016/j.infbeh.2019.101350 Block, N. 1995. “On a confusion about a function of consciousness”. Behavioral and Brain Sciences, 18: 227–247. Burge, T. 2010. Origins of objectivity. Oxford: Oxford University Press.

346

The Development of Implicit Theory of Mind Burge, T. 2018. “Do infants and nonhuman animals attribute mental states?”. Psychological Review, 125: 409–434. https://doi.org/10.1037/rev0000091 Burnside, K., Ruel, A., Azar, N., and Poulin-Dubois, D. 2018. “Implicit false belief across the lifespan: non-replication of an anticipatory looking task”. Cognitive Development, 46: 4–11. https://doi. org/10.1016/j.cogdev.2017.08.006 Buttelmann, D., Carpenter, M., and Tomasello, M. 2009. “Eighteen-month-old infants show false belief understanding in an active helping paradigm”. Cognition, 112: 337–342. Buttelmann, F., Suhrke, J., and Buttelmann, D. 2015. “What you get is what you believe: eighteen-montholds demonstrate belief understanding in an unexpected-identity task”. Journal of Experimental Child Psychology, 131: 94–103. http://dx.doi.org/10.1016/j.jecp.2014.11.009 Butterfill, S. A., and Apperly, I. A. 2013. “How to construct a minimal Theory of Mind”. Mind and Language, 28: 606–637. https://doi.org/10.1111/mila.12036 Carlson, S. M., and Moses, L. J. 2001. “Individual differences in inhibitory control and children’s theory of mind”. Child Development, 72: 1032–1053. Carruthers, P. 2013. “Mindreading in infancy”. Mind and Language, 28: 141–172. https://doi.org/10.1111/ mila.12014 Clark, A., and Karmiloff-Smith, A. 1993. “The cognizer’s innards: a philosophical and psychological perspective on the development of thought”. Mind and Language, 8: 487–519. Cummins, R. 1983. The nature of psychological explanation. Cambridge, MA: MIT Press. Davies, M. 1989. “Tacit knowledge and subdoxastic states”. In A. George, ed., Reflections on Chomsky. Oxford: Blackwell: 131–152. Dennett, D. 1978. “Beliefs about beliefs”. Behavioral and Brain Sciences, 4: 568–570. Devine, R. T., and Hughes, C. 2014. “Relations between false belief understanding and executive function in early childhood: a meta-analysis”. Child Development, 85: 1777–1794. Dörrenberg, S., Rakoczy, H., and Liszkowski, U. 2018. “How (not) to measure infant Theory of Mind: testing the replicability and validity of four non-verbal measures”. Cognitive Development, 46: 12–30. https://doi.org/10.1016/j.cogdev.2018.01.001 Dörrenberg, S., Wenzel, L., Proft, M., Rakoczy, H., and Liszkowski, U. 2019. “Reliability and generalizability of an acted-out false belief task in 3-year-olds”. Infant Behavior and Development, 54: 13–21. https://doi.org/10.1016/j.infbeh.2018.11.005 Dretske, F. 1969. Seeing and knowing. Chicago: University of Chicago Press. Edwards, K., and Low, J. 2017. “Reaction time profiles of adults’ action prediction reveal two mindreading systems”. Cognition, 160: 1–16. http://dx.doi.org/10.1016/j.cognition.2016.12.004 Elekes, F., Varga, M., and Király, I. 2017. “Level‐2 perspectives computed quickly and spontaneously: evidence from eight‐ to 9.5‐year‐old children”. British Journal of Developmental Psychology, 35: 609–622. https://doi.org/10.1111/bjdp.12201 Fizke, E., Butterfill, S., van de Loo, L., Reindl, E., and Rakoczy, H. 2017. “Are there signature limits in early theory of mind?”. Journal of Experimental Child Psychology, 162: 209–224. https://doi. org/10.1016/j.jecp.2017.05.005 Flavell, J. H., Everett, B. A., Croft, K., and Flavell, E. R. 1981. “Young children’s knowledge about visual perception: further evidence for the Level 1-Level 2 distinction”. Developmental Psychology, 17: 99–103. Frank, M. C., Bergelson, E., Bergmann, C., Cristia, A., Floccia, C., Gervain, J., . . . Yurovsky, D. 2017. “A  collaborative approach to infant research: promoting reproducibility, best practices, and theorybuilding”. Infancy, 22: 421–435. https://doi.org/10.1111/infa.12182 Frankish, K. 2009. “Systems and levels: dual-system-theories and the personal-subpersonal distinction”. In J. S. B. T. Evans and K. Frankish, eds., In two minds: dual processes and beyond. New York: Oxford University Press: 89–107. Gopnik, A., and Astington, J. W. 1988. “Children’s understanding of representational change and its relation to the understanding of false belief and the appearance-reality distinction”. Child Development, 59: 26–37. Grosse Wiesmann, C., Friederici, A. D., Singer, T., and Steinbeis, N. 2016. “Implicit and explicit false belief development in preschool children”. Developmental Science, 20: e12445. https://doi.org/10.1111/desc.12445 Grosse Wiesmann, C., Schreiber, J., Singer, T., Steinbeis, N., and Friederici, A. D. 2017. “White matter maturation is associated with the emergence of Theory of Mind in early childhood”. Nature Communications, 8: 14692. https://doi.org/10.1038/ncomms14692 Hayashi, T., Akikawa, R., Kawasaki, K., Egawa, J., Minamimoto, T., Kobayashi, K., . . . Hasegawa, I. 2020. “Macaques exhibit implicit gaze bias anticipating others’ false-belief-driven actions via medial prefrontal cortex”. Cell Reports, 30: 4433–4444. https://doi.org/10.1016/j.celrep.2020.03.013

347

Hannes Rakoczy Heyes, C. 2014. “Submentalizing: I am not really reading your mind”. Perspectives on Psychological Science, 9: 131–143. Hyde, D. C., Simon, C. E., Ting, F., and Nikolaeva, J. I. 2018. “Functional organization of the temporalparietal junction for theory of mind in preverbal infants: a near-infrared spectroscopy study”. The Journal of Neuroscience, 38: 4264–4274. https://doi.org/10.1523/jneurosci.0264-17.2018 Jacoby, L. L. 1991. “A process dissociation framework: separating automatic from intentional uses of memory”. Journal of Memory and Language, 30: 513–541. https://doi.org/10.1016/0749-596X(91)90025-F Kampis, D., Kármán, P., Csibra, G., Southgate, V., and Hernik, M. 2021. “A two-lab direct replication attempt of Southgate, Senju and Csibra (2007)”. Royal Society Open Science, 8: 210190. https://doi. org/10.1098/rsos.210190 Kampis, D., and Kovács, Á. M. 2022. “Seeing the world from others’ perspective: 14-month-olds show altercentric modulation effects by others’ beliefs”. Open Mind: 1–19. https://doi.org/10.1162/ opmi_a_00050 Király, I., Oláh, K., Csibra, G., and Kovács, Á. M. 2018. “Retrospective attribution of false beliefs in 3-year-old children”. Proceedings of the National Academy of Sciences, 115: 11477–11482. https://doi. org/10.1073/pnas.1803505115 Knudsen, B., and Liszkowski, U. 2012. “18-month-olds predict specific action mistakes through attribution of false belief, not ignorance, and intervene accordingly”. Infancy, 17: 672–691. Krupenye, C., Kano, F., Hirata, S., Call, J., and Tomasello, M. 2016. “Great apes anticipate that other individuals will act according to false beliefs”. Science, 354: 110–114. https://doi.org/10.1126/science.aaf8110 Kulke, L., Johannsen, J., and Rakoczy, H. 2019. “Why can some implicit Theory of Mind tasks be replicated and others cannot? A  test of mentalizing versus submentalizing accounts”. PLoS ONE, 14: e0213772. https://doi.org/10.1371/journal.pone.0213772 Kulke, L., and Rakoczy, H. 2018. “Implicit Theory of Mind – an overview of current replications and nonreplications”. Data in Brief, 16(Supplement C): 101–104. https://doi.org/10.1016/j.dib.2017.11.016 Kulke, L., and Rakoczy, H. 2019. “Testing the role of verbal narration in implicit theory of mind tasks”. Journal of Cognition and Development, 20: 1–14. https://doi.org/10.1080/15248372.2018.1544140 Kulke, L., Reiß, M., Krist, H., and Rakoczy, H. 2018. “How robust are anticipatory looking measures of Theory of Mind? Replication attempts across the life span”. Cognitive Development, 46: 97–111. https://doi.org/10.1016/j.cogdev.2017.09.001 Kulke, L., von Duhn, B., Schneider, D., and Rakoczy, H. 2018. “Is implicit theory of mind a real and robust phenomenon? Results from a systematic replication study”. Psychological Science, 29: 888–900. https://doi.org/10.1177/0956797617747090 Kulke, L., Wübker, M., and Rakoczy, H. 2019. “Is implicit Theory of Mind real but hard to detect? Testing adults with different stimulus materials”. Royal Society Open Science, 6: 190068. https://doi. org/10.1098/rsos.190068 Leslie, A. M. 2005. “Developmental parallels in understanding minds and bodies”. Trends in Cognitive Science, 9: 459–462. https://doi.org/10.1016/j.tics.2005.08.002 Low, J., Drummond, W., Walmsley, A., and Wang, B. 2014. “Representing how rabbits quack and competitors act: limits on preschoolers’ efficient ability to track perspective”. Child Development, 85: 1519–1534. https://doi.org/10.1111/cdev.12224 Low, J., and Watts, J. 2013. “Attributing false beliefs about object identity reveals a signature blind spot in humans’ efficient mind-reading system”. Psychological Science, 24: 305–311. https://doi. org/10.1177/0956797612451469 Mellers, B., Hertwig, R., and Kahneman, D. 2001. “Do frequency representations eliminate conjunction effects? An exercise in adversarial collaboration”. Psychological Science, 12: 269–275. Newen, A., and Wolf, J. 2020. “The situational mental file account of the false belief tasks: a new solution of the paradox of false belief understanding”. Review of Philosophy and Psychology, 11: 717–744. https:// doi.org/10.1007/s13164-020-00466-w Oktay-Gür, N., Schulz, A., and Rakoczy, H. 2018. “Children exhibit different performance patterns in explicit and implicit theory of mind tasks”. Cognition, 173: 60–74. https://doi.org/10.1016/j. cognition.2018.01.001 Onishi, K. H., and Baillargeon, R. 2005. “Do 15-month-old infants understand false beliefs?”. Science, 308: 255–258. https://doi.org/10.1126/science.1107621 Perner, J. 1991. Understanding the representational mind. Cambridge, MA: MIT Press. Perner, J., and Roessler, J. 2012. “From infants’ to children’s appreciation of belief ”. Trends in Cognitive Sciences, 16: 519–525. https://doi.org/10.1016/j.tics.2012.08.004

348

The Development of Implicit Theory of Mind Phillips, J., Buckwalter, W., Cushman, F., Friedman, O., Martin, A., Turri, J., . . . Knobe, J. 2020. “Knowledge before belief ”. Behavioral and Brain Sciences, 44: 1–37. https://doi.org/10.1017/s0140525x20000618 Poulin-Dubois, D., Rakoczy, H., Burnside, K., Crivello, C., Dörrenberg, S., Edwards, K., . . . Ruffman, T. 2018. “Do infants understand false beliefs? We don’t know yet – a commentary on Baillargeon, Buttelmann and Southgate’s commentary”. Cognitive Development, 48: 302–315. https://doi.org/10.1016/j. cogdev.2018.09.005 Poulin-Dubois, D., and Yott, J. 2017. “Probing the depth of infants’ theory of mind: disunity in performance across paradigms”. Developmental Science, 21: e12600. https://doi.org/10.1111/desc.12600 Powell, L. J., Hobbs, K., Bardis, A., Carey, S., and Saxe, R. 2018. “Replications of implicit theory of mind tasks with varying representational demands”. Cognitive Development, 46: 40–50. https://doi. org/10.1016/j.cogdev.2017.10.004 Priewasser, B., Rafetseder, E., Gargitter, C., and Perner, J. 2018. “Helping as an early indicator of a theory of mind: mentalism or teleology?”. Cognitive Development, 46: 69–78. https://doi.org/10.1016/j. cogdev.2017.08.002 Rakoczy, H. 2017. “In defense of a developmental dogma: children acquire propositional attitude folk psychology around age 4”. Synthese, 194: 689–707. https://doi.org/10.1007/s11229-015-0860-8 Rakoczy, H. 2022. “Foundations of theory of mind and its development in early childhood”. Nature Reviews Psychology. https://doi.org/10.1038/s44159-022-00037-z Rubio-Fernández, P.  2019. “Publication standards in infancy research: three ways to make violationof-expectation studies more reliable”. Infant Behavior and Development, 54: 177–188. https://doi. org/10.1016/j.infbeh.2018.09.009 Rubio-Fernandez, P., Southgate, V., and Király, I. 2021. “Pragmatics for infants: commentary on Wenzel et al. (2020)”. Royal Society Open Science, 8: 210247. https://doi.org/10.1098/rsos.210247 Schneider, D., Bayliss, A. P., Becker, S. I., and Dux, P. E. 2012. “Eye movements reveal sustained implicit processing of others’ mental states”. Journal of Experimental Psychology: General, 141: 433–438. Schuwerk, T., Jarvers, I., Vuori, M., and Sodian, B. 2016. “Implicit mentalizing persists beyond early childhood and is profoundly impaired in children with autism spectrum condition”. Frontiers in Psychology, 7: 1696. https://doi.org/10.3389/fpsyg.2016.01696 Schuwerk, T., Kampis, D., Baillargeon, R., Biro, S., Bohn, M., Byers-Heinlein, K., . . . Rakoczy, H. 2022. “Action anticipation based on an agent’s epistemic state in toddlers and adults”. Child Development (InPrinciple Acceptance of Registered Report Stage 1: Study Design). Schuwerk, T., Priewasser, B., Sodian, B., and Perner, J. 2018. “The robustness and generalizability of findings on spontaneous false belief sensitivity: a replication attempt”. Royal Society Open Science, 5: 172273. https://doi.org/10.1098/rsos.172273 Scott, R. M., and Baillargeon, R. 2009. “Which penguin is this? Attributing false beliefs about object identity at 18 months”. Child Development, 80: 1172–1196. Scott, R. M., and Baillargeon, R. 2017. “Early false-belief understanding”. Trends in Cognitive Sciences, 21: 237–249. https://doi.org/10.1016/j.tics.2017.01.012 Scott, R. M., Roby, E., and Baillargeon, R. (in press). “How sophisticated is infants’ theory of mind?”. In O. Houdé and G. Borst, eds., Cambridge handbook of cognitive development. Cambridge: Cambridge University Press. Senju, A., Southgate, V., White, S., and Frith, U. 2009. “Mindblind eyes: an absence of spontaneous theory of mind in Asperger syndrome”. Science, 325: 883–885. Shoemaker, S. 1981. “Some varieties of functionalism”. Philosophical Topics, 12: 93–119. Southgate, V., Chevallier, C., and Csibra, G. 2010. “Seventeen-month-olds appeal to false beliefs to interpret others referential communication”. Developmental Science, 13: 907–912. Southgate, V., Senju, A., and Csibra, G. 2007. “Action anticipation through attribution of false belief by 2-year-olds”. Psychological Science, 18: 587–592. Southgate, V., and Vernetti, A. 2014. “Belief-based action prediction in preverbal infants”. Cognition, 130: 1–10. http://dx.doi.org/10.1016/j.cognition.2013.08.008 Stich, S. 1978. “Beliefs and subdoxastic states”. Philosophy of Science, 45: 499–518. Surian, L., and Geraci, A. 2012. “Where will the triangle look for it? Attributing false beliefs to a geometric shape at 17 months”. British Journal of Developmental Psychology, 30: 30–44. https://doi. org/10.1111/j.2044-835X.2011.02046.x Surtees, A. D. R., Butterfill, S. A., and Apperly, I. A. 2012. “Direct and indirect measures of level‐2 perspective‐taking in children and adults”. British Journal of Developmental Psychology, 30: 75–86. https:// doi.org/10.1111/j.2044-835X.2011.02063.x

349

Hannes Rakoczy Todd, A. R., Cameron, C. D., and Simpson, A. J. 2017. “Dissociating processes underlying level-1 visual perspective taking in adults”. Cognition, 159: 97–101. https://doi.org/10.1016/j.cognition.2016.11.010 Todd, A. R., Simpson, A. J., and Cameron, C. D. 2019. “Time pressure disrupts level-2, but not level1, visual perspective calculation: a process-dissociation analysis”. Cognition, 189: 41–54. https://doi. org/10.1016/j.cognition.2019.03.002 Wellman, H., Cross, D., and Watson, J. 2001, “Meta-analysis of theory-of-mind development: the truth about false belief ”. Child Development, 72: 655–684. Wenzel, L., Dörrenberg, S., Proft, M., Liszkowski, U., and Rakoczy, H. 2020. “Actions do not speak louder than words in an interactive false belief task”. Royal Society Open Science, 7: 191998. https://doi. org/10.1098/rsos.191998 Wimmer, H., and Perner, J. 1983. “Beliefs about beliefs: representation and constraining function of wrong beliefs in young children’s understanding of deception”. Cognition, 13: 103–128. https://doi. org/10.1016/0010-0277(83)90004-5 Yott, J., and Poulin-Dubois, D. 2016. “Are infants’ theory-of-mind abilities well integrated? Implicit understanding of intentions, desires, and beliefs”. Journal of Cognition and Development, 17: 683–698. https://doi.org/10.1080/15248372.2015.1086771

350

PART 7

Memory

27 IMPLICIT MEMORY Sarah K. Robins

Implicit memory is the use of memory without awareness of the activity of remembering. Distinguishing between memory’s implicit and explicit forms matters to both memory scientists and philosophers. There are, however, differences in where researchers in each field draw the distinction – between ways of recollecting or types of memory storage, respectively. Each proposal faces a similar problem: our view of explicit memory has shifted over time, making it difficult to understand implicit memory by contrast. These shifts illuminate key features of explicit memory, while an account of implicit memory remains elusive. A fruitful way forward, I propose, is to stop treating implicit and explicit memory as opposites, considering instead a range of implicit features of memory.

1.  Historical Background The 1980s have been described as a golden age for research into implicit memory in cognitive psychology and neuroscience (Schacter 1992). During this time, memory scientists demonstrated various ways that implicit and explicit recollection could be dissociated. Researchers would give participants a set of items to study – words, images, etc. – and then test their recollection of these items with both explicit and implicit measures. Explicit measures probed memory directly: participants would be asked to recall as many items from the previous set as they could, or to identify previously seen items from within a larger set. Explicit memory performance exhibited a familiar pattern of retention. Initially, participants remembered many items, but performance declined quickly. After only a few days, many could no longer recall or recognize several items, and after a week, failed on most. Implicit measures, in contrast, typically avoided any mention of memory. Instead, participants would be asked to engage in a seemingly unrelated activity, such as producing words from presented stems (e.g., TAB to TABLE) or fragments (e.g., A_A_IN to ASSASSIN), or reading words aloud as quickly as possible. In these tasks, researchers compared participant performance (speed and accuracy) across previously presented and novel items. Participants generally performed better on items from the previous set, indicating some form of retention. This retention effect was not as notable as the initial performance on explicit measures, but held constant over the weeklong period during which explicit performance declined (see Schacter 1987 for review). These differences in

DOI: 10.4324/9781003014584-35 353

Sarah K. Robins

explicit and implicit performance compelled many researchers to view them as distinct memory processes (Roediger 1990). Researchers also extended this framework to persons with amnesia – those who have significant explicit memory deficits as a result of neuropsychological trauma or injury. Persons with amnesia cannot remember their past experiences. When performing tasks of the sort described earlier, they perform poorly on the explicit measures, as expected. Their implicit task performance, however, remained intact (for a review, see Shimamura and Squire 1986). These results led many researchers to argue that explicit and implicit forms of memory are not only distinct processes, but are supported by distinct brain regions. In this way, work on implicit memory played a critical role in the development of the Multiple Memory Systems framework, the project of offering a taxonomy of memory, which occupied cognitive psychology and neuroscience for much of the late 1980s and 90s. The Multiple Memory Systems framework divides memory into separate forms, each of which is presumed to involve a distinct neurocognitive system. There have been changes to the taxonomy over the years (compare, for example: Schacter and Tulving 1994; Schacter et al. 2000; Squire 2004), but all versions begin from an initial sorting of memory into two sub-types: declarative and non-declarative. Declarative memory involves the explicit recall of facts and experiences – semantic and episodic memory, respectively. Non-declarative memory involves non-explicit forms of recollection, including the forms of priming and retention presented earlier, as well as procedural memory (habits and skills) and forms of conditioning. Accounts of memory systems differ over whether they take implicit memory to refer to all forms of nondeclarative memory or only to the kind of information retention described in the preceding studies, excluding procedural memory and conditioning. Implicit memory has not received much attention in philosophy, at least not under that description. There is, however, considerable interest in knowledge-how: “the kind of knowledge you have when it’s truly said of you that you know how to do something – say, ride a bicycle” (Fantl 2017). This form of knowledge is often thought to be importantly different from knowledge of facts, often referred to as knowledge-that, and so philosophers have offered various accounts of knowledge-how and its distinction from knowledge-that. Anti-intellectualist views consider knowledge-how to be a fully distinct form of knowing, one best characterized as an ability (Hawley 2003) or disposition (Ryle 1949). Intellectualists, on the other hand, recognize some differences between knowing how to ride a bicycle and knowing that the bicycle was invented in 1817, yet they argue that both forms of knowing involve knowledge of facts (Stanley and Williamson 2001). Viewed from a wide angle, the discussion of implicit memory in psychology and of knowledge-how in philosophy appear to be tracking the same, or at least highly similar, cognitive phenomena. Indeed, researchers in both fields have encouraged the association (e.g., Cohen and Squire 1980; Wallis 2008). There are, however, several dissimilarities that are important to keep in mind when making these comparisons. First, in these respective inquiries, psychologists and neuroscientists are concerned with memory; philosophers are concerned with knowledge. There is overlap between memory and knowledge – some knowledge is likely stored in memory – but on most accounts they’re fully dissociable (i.e., we can have knowledge from sources other than memory; it is possible to remember X while lacking the justification or some such to count as knowing X). As philosophy of memory has grown into its own distinctive subfield in recent years, the concerns about memory have grown beyond its role in preserving knowledge (Bernecker and Michaelian 2017). Second, many researchers have recently noted the lack of straightforward inferences from empirical findings about implicit memory to the intellectualist/anti-intellectualist debate about knowledge-how 354

Implicit Memory

(Drayson and Schwartz 2019; De Brigard 2019). Third, interest in implicit memory amongst cognitive psychologists and neuroscientists has waned in the decades since its golden age, as researchers have become more interested in declarative memory – especially episodic memory, its errors, and its connections to other cognitive faculties like imagination (Schacter 2019). Research into knowledge-how, in contrast, has been experiencing a resurgence in philosophy (Pavese and De Brigard 2019). Finally, and most challenging for any attempt at rapprochement, memory scientists and philosophers working on these topics have come to focus on distinct phenomena. For scientists studying implicit memory, the focus is on the activity of implicit recollection. For philosophers, the concern is either with the possession of knowledge-how or the storage of memory in implicit and explicit forms. These are distinct ways of conceiving of implicit memory. Given this divergence, I discuss each separately here.

2.  Implicit Recollection Over time, psychologists and neuroscientists have shifted how they think about implicit memory – from viewing it as a distinct memory system to viewing it as a form of recollection. Current definitions focus on retrieval, contrasting the implicit reactivation or use of previously learned information with explicit recollection. Cubelli and Della Sala offer the following as a consensus definition: A memory is implicit when the learned information is retrieved and used without awareness of remembering it and with no reference to the learning phase. (2020: 345) Here “implicit” refers to how retrieval happens, not what is being retrieved. This allows for implicit recollection across memory’s various forms; on this view, there can be implicit recollection of episodic, semantic, and procedural memories. Episodic memory is memory for particular personal experiences. I might have an episodic memory of where I parked my car in a lot, which I could recall implicitly by simply walking toward my car, without any effortful attempt to recall where I parked. Semantic memory is memory for facts or general pieces of knowledge. Suppose I am working on a crossword puzzle, and fail to notice it is one I have completed before. Owing to its nascent familiarity, I complete it more quickly than average, having implicitly remembered its content. Procedural memories include habits and skills, like the ability to knit. When I sit down to knit a scarf, for example, I remember how to do so implicitly by engaging in the required movements, with little attention to the steps or details. This is in contrast to times when knitting requires procedural memory to be recollected explicitly, as when encountering a difficult pattern, new type of yarn, etc. Cubelli and Della Sala’s definition involves two key elements: 1) lack of awareness of what’s retrieved and used, and 2) the absence of reference to the previous experience or learning event. The examples of implicit retrieval offered earlier contain these features. However, they each do so in different ways, highlighting ambiguities in the definition. Consider the first component of the definition: lack of awareness. What, exactly, is the awareness that is missing in cases of implicit remembering? It doesn’t seem that I am fully unaware of what is being remembered in any of the sample cases. When finding my car, completing the crossword, and knitting the scarf, I make use of the remembered information to complete the activity. What these cases lack seems more like awareness of remembering. Indeed, Cubelli and Della Sala caution against treating implicit as synonymous with unconscious because “the 355

Sarah K. Robins

memory content is the object of consciousness” (2020: 345), even if the person making use of that content does not recognize it as a memory or cannot articulate the content. Characterizing the feature as a lack of awareness of remembering does seem to capture at least some of what is going on in each of these cases of implicit recollection. When a particular answer to the crossword springs to mind, for example, I am aware of the content, but not aware of it as a memory of encountering the clue before. To get clear on how implicit recollection works, we want to pin down this lack of awareness more precisely. A definition that appeals to the absence of X is most useful when there is a clear sense of X in play. The X in question here is awareness of remembering, and that poses a problem: there is no clear sense available of what this awareness involves. This question may not strike psychologists and neuroscientists as particularly pressing, given that the experimental protocols they use generally require providing participants with overt instructions and information about the tasks involved. Participants are presumably aware of their remembering because the task involves asking them to remember. When we shift, however, to thinking about implicit memory beyond experimental settings and how to characterize awareness of remembering more generally, the question becomes more urgent – and more difficult to answer. The difficulty does not stem from a lack of attempts to characterize the feeling of remembering. Amongst philosophers and psychologists, there is a long tradition of theorizing about remembering’s distinctive feature(s), which distinguish it from other psychological states. A variety of such features been proposed: familiarity (Locke 1694/1979), a feeling of pastness (Russell 1912), vivacity (Hume 1739), spontaneity (Furlong 1951), intimacy (James 1890), etc. Each candidate feature has encountered considerable opposition, over the nature of the proposed feature (e.g., Reid’s (1785) challenge to Hume’s concept of vivacity) or the feature’s ability to capture all and only instances of explicit remembering (e.g., Furlong’s spontaneity doesn’t easily capture cases of effortful retrieval). Even without a settled account of the form of awareness distinctive to remembering, some have argued that the experiential component is essential to remembering, such that there could be no such thing as implicit memory (Klein 2015; Moyal-Sharrock 2009). Many other contemporary philosophers have taken the opposite approach, avoiding the appeal to any awareness condition in their accounts of remembering. Instead, they require only that the retained content is represented at the time of remembering. As noted earlier, this alone will not distinguish explicit and implicit forms of recollection, both of which have the retained content as the focus of awareness/representation. When contemporary philosophers have proposed adding further features to distinguish remembering, they have avoided appeals to awareness, opting instead to use functional characteristics of the psychological state (e.g., Debus’ (2010) epistemic relevance condition). This may be effective for distinguishing remembering in one respect – the functional role of memory seems quite different from that of perception or imagination or desire. The functional characterization looks less promising, however, for distinguishing between remembering’s implicit and explicit forms. Let’s turn now to the second component of Cubelli and Della Sala’s (2020) definition: the lack of reference to the previous event or learned information. Like the previous, it is contrastive. Understanding this feature of implicit memory will again require us to start with an examination of the explicit case from which it is being opposed. The definition implies that explicit remembering involves a “learning phase” that is absent in implicit remembering. This alone does not make clear where the reference is located in explicit cases. In the cue presented to spark remembering? In the mind of the rememberer? In the experimental design and setup? 356

Implicit Memory

The last of these options is the most straightforward. It is easy to measure and fits with standard ways of performing tests of implicit and explicit remembering. Implicit remembering tasks are those that probe retention without any mention of memory, where participants are engaged in a task that is not obviously tied to remembering but nonetheless allows the experimenter to look for effects of retaining prior information. In explicit remembering tasks, participants are asked to remember, recognize, or recall. The task is overtly about memory and the instructions given must make some kind of reference to the learning event – for example, was X on the list of items you saw yesterday? Which of these items was previously presented? How many items from yesterday’s list can you recall? Using task features to distinguish which instances of recollection make reference to a learning phase works well in experimental contexts, but it is difficult, if not impossible, to extend beyond experimental settings so as to track implicit and explicit remembering in everyday interactions. To see the point, consider some cases of explicit remembering in day-to-day life. Do they always reference the previous event where the remembered information was learned or experienced? This happens at least some of the time – I recall that I am supposed to bring a bottle of wine to the dinner and, in so doing, I see in my mind the host’s face as she made the request. But there are also plenty of times when this does not happen. I can remember that I  am supposed to bring wine to dinner without remembering who told me that, or when. I can even remember past experiences, representing aspects of the experience in great detail, without locating the experience at a particular point in time. More generally, psychologists distinguish source memory (memory for where, when, and how information was acquired) from item memory (Johnson 1992). Reference to the learning event is treated as a distinct form of memory, not an aspect of explicit remembering. The definition of implicit recollection used in contemporary psychology and neuroscience relies on a contrast with explicit recollection. It may be possible to draw this contrast clearly in experimental contexts, where the procedures used to conduct studies provide explicit recollection with the requisite features. The aim of this research, however, is to provide insight into the memory processes at work in our everyday lives. But explicit recollection outside of experimental contexts often lacks the features that would be required for this extension. Those who continue to support a distinction between implicit and explicit forms of recollection must either refine their definition or accept its limited scope.

3.  Implicit Memory Storage Philosophical interest in implicit memory also relies on the contrast between implicit and explicit. Instead of drawing the contrast between forms of recollection, philosophers make a distinction between what is stored in memory, and how. Here is Bernecker’s account of the philosophical distinction: You explicitly remember that p if this representation is actually present in your mind in the right sort of way, for example, as an item in your memory box or as a trace storing a sentence in the language of thought. To implicitly remember that p your mind may not contain a representation with that content. The contents of implicit memories . . . have never previously been tokened and don’t inhabit our long-term memory. (2010: 29) The intended distinction is clear enough and parallels the distinction philosophers have drawn between explicit and implicit forms of other mental states, like belief (Lycan 1986; Audi 1994). 357

Sarah K. Robins

Drawing the contrast in this way requires some account of explicit mental representation, but does not compel the endorsement of any particular view. As Bernecker illustrates earlier, explicit memory requires not only storage of an explicit representation, but also a previous experience where that representation was tokened. An explicit memory representation is a representation that was at some point occurrent, then stored and made available for subsequent reactivation in remembering. Explicit memory storage does not require remembering, of course; there are plenty of items in memory that could be remembered, but are never called upon. The contrast between explicit and implicit memory storage is best illustrated with an example. Suppose I received a yellow bike for my eighth birthday. For the purposes of this example, let’s treat the case as a semantic memory with the content “I received a yellow bike for my eighth birthday,” but the point could also be made with an episodic memory. An explicit semantic memory of this fact requires that this content was, at some point in my past, an occurrent representation – maybe as the birthday party was happening, or when talking about the party afterward with a friend – and also that this representation is stored in my memory, in a format that retains the content for later retrieval/reactivation. I could also have an implicit memory of this fact about my eighth birthday. In this case there would be no explicit representation of the form “I received a yellow bike for my eighth birthday,” either because no such representation was ever tokened (e.g., maybe I did not pay much attention to presents at my own birthday party) or because the representation was not stored or lost over time through the process of forgetting. Implicit memory of this fact about the yellow bike is still possible if it can be inferred from other facts that are part of my explicit memory/stored beliefs. My memory store may include the following explicitly represented beliefs: as a child my favorite color was yellow, my birthday is in May, and I broke my arm by falling off my bike while learning to ride the summer I was eight (indicating I received the bike shortly before this incident), etc. This would allow me to infer, and thus implicitly remember, that I received a yellow bike for my 8th birthday. The distinction between explicit and implicit forms of memory storage has been straightforward for philosophers of memory because of the long-standing and widespread commitment to the idea that explicit remembering requires a memory trace (De Brigard 2014a; Robins 2017). A memory trace is a stored representation of a past experience or piece of information, held in the mind of the rememberer since the acquisition event. Its subsequent reactivation is necessary (although on most views not sufficient) for remembering. Characterized as such, it meets the requirements for explicit memory storage. Implicit memory is then a derivative form, where the information is retained more diffusely, without a corresponding trace. Recent work in the philosophy of memory has challenged the commitment to memory traces. Philosophers are increasingly inclined to endorse constructivist or simulationist views of memory, which reject trace requirements on remembering (e.g., Michaelian 2016). They propose instead that representations in remembering are constructed from a general network of information about the past. Rather than store individual representations of gifts received at past birthdays, for example, I  store general information about birthday parties and gifts and the like, amalgamated across experiences. On this view, all cases of remembering look at least somewhat like the case of inferential implicit memory described earlier. This shift in characterizing memory storage is thought to have many advantages: it better accounts for the frequency of memory errors, which are well documented in cognitive psychology (Loftus 2003) and better aligns philosophical work on memory with the trend toward constructivism in memory science (Schacter 2019). Even philosophers of memory who remain committed to a role for memory traces have adjusted their accounts of memory traces to achieve similar ends, characterizing them as distributed (Sutton 1998; Bernecker 2010), dispositional (De Brigard 2014b), or content-free (Hutto and Peeters 2018). 358

Implicit Memory

The move to a trace-free or modified trace theory may be advantageous in some respects, but changing the view of explicit memory in this way blurs the distinction between implicit and explicit forms of memory storage. If memory traces are distributed or without content, then they do not meet the requirements for explicit storage. If memories are constructed rather than stored, then all memories are implicit rather than explicit. These implications of current trends in theorizing have gone mostly unnoticed, largely as a result of the lack of direct attention to implicit memory amongst philosophers of memory.

4.  Partially Implicit Memory The previous two sections addressed psychological and philosophical approaches to implicit memory. They each focused on a distinct aspect of the memory process – recollection vs. storage, respectively – but they both conceived of implicit and explicit memory as opposites, attempting to draw a clear distinction between them. As I illustrated in the preceding discussions, neither proposed distinction has been clearly articulated or easily maintained. Moving forward it thus seems productive to explore other ways to conceive of implicit memory, which highlight implicit features of memory. In this section, I draw out a few such possibilities from recent work in the philosophy of memory. In Memory and the Self (2016), Mark Rowlands offers an account of personal, autobiographical memory that includes a central role for a phenomenon he dubs “Rilkean memory.” Named for observations made by the poet Rainer Maria Rilke, Rilkean memories are those that have transformed from an explicit to implicit form over time. Rilkean memories have become so implicit, in fact, that they no longer include the original details and content. Instead, they register only as a behavior or feeling. They are cases where “the act of remembering becomes divorced from what is remembered” (Rowlands 2016: 73). Rilkean memories may be embodied, as when one returns to visit their childhood home and instinctively avoids walking too near a closet they feared was haunted as a child. They can also be affective, as when encountering a scent conjures an ineffable emotional response, but without any explicit recognition of its connection to a romantic encounter from much earlier in life. As Rowlands explains, these memories defy characterization within the frameworks standardly used: procedural vs. declarative, semantic vs. episodic, voluntary vs. involuntary – even implicit vs. explicit. The failure of existing distinctions to capture Rilkean memory does not make these memories any less implicit. Instead, they offer an opportunity to expand our conception of implicit memory in ways that reflect memory’s dynamics. A memory might be implicit at one time, but not at another. Reflecting on memory’s ability to change over time, we can also consider cases that move in the other direction, from implicit to explicit. In Remembering from the Outside (2018), Chris McCarroll defends the view that observer memories – memories of one’s own experience, represented from an outside perspective – can be genuine memories. To establish this counterintuitive view, McCarroll illustrates several ways that observer memories can be formed. One such route involves using information that is implicitly available during an experience to build an observer perspective of that experience in subsequent remembering. McCarroll observes, “perceptual experience is profoundly copious, colorful, and occasionally chaotic” (2018: 53). Our experience of an event involves an array of multimodal details, embedding – implicitly – a range of features about the world around us. When we return to these experiences in remembering, the previously implicit details can be used to construct an explicit representation of the past experience from an observer’s standpoint. During their first performance on stage, for example, a person may be highly attuned to sounds throughout the venue and the facial expressions of audience members. When recalling this debut later in life, those features may lend themselves to 359

Sarah K. Robins

a representation of the experience where the person sees herself on stage rather than from the perspective she occupied during the performance. For other memories, some features may remain implicit throughout – even when the rest of the memory is fully explicit, and indeed highly salient. Barbara Montero (2020) discusses sensational memory and its role in our recollection of intensely physical experiences. She focuses especially on painful experiences, like childbirth, and the commonly professed inability to remember how it felt. This inability to directly access the prior sensation is notably odd, in comparison to the vivid detail with which persons are often able to remember many other features of such experiences. One can remember that the experience was painful and possibly even descriptions of that sensational experience, but the sensational experience itself remains an implicit feature of the memory. Aspects of memory may also be implicit in other ways: in our habits, rituals, and performances; in our relationships, social practices, and monuments. Merlin Donald (1991) dubbed these external forms of retention exograms and John Sutton has marshalled them in service of an extended and collective view of remembering. Sutton identifies an array of such exograms: “other people, scrabble tiles, theater architecture, cocktail glasses, slide rules, incised sticks, shells, languages, moral norms, knots, codes, maps, diagrams, fingers, monuments, software devices, rituals, rhythms and rhymes, and roads” (2010: 214). Precisely because of the ways these differ from the operations of our mind and our internal memory capacities, he argues, these features of the world contribute to what and how we remember in a range of implicit ways. They prompt, unify, and update our memories, in roles that are subtle but critical to what and how we recall.

Conclusion This entry has surveyed a range of possible ways to understand implicit memory, each of which requires further exploration and elaboration. Considering how implicit memory can be understood and distinguished from explicit memory remains an important task for surveying the full range of memory phenomena and illuminating the contrast with memory’s more canonical, explicit form. Going forward, it is sensible to advocate for the distinction between implicit and explicit memory, even as we continue the search for what implicit memory might be.

Related Topics Chapters 1, 3, 14, 15, 28, 29, 31

References Audi, R. 1994. “Dispositional belief and a disposition to believe”. Nous, 28: 419–434. Bernecker, S. 2010. Memory: a philosophical study. Oxford: Oxford University Press. Bernecker, S., and Michaelian, K., eds. 2017. Routledge handbook of the philosophy of memory. London: Routledge. Cohen, N. J., and Squire, L. R. 1980. “Preserved learning and retention of pattern analyzing skill in amnesia: dissociation of knowing how and knowing that”. Science, 210: 207–209. Cubelli, R., and Della Sala, S. 2020. “Implicit memory”. Cortex, 125: 345. De Brigard, F. 2014a. “The nature of memory traces”. Philosophy Compass, 9: 402–414. De Brigard, F. 2014b. “Is memory for remembering? Recollection as a form of episodic hypothetical thinking”. Synthese, 191: 1–31. De Brigard, F. 2019. “Know-how, intellectualism, and memory systems”. Philosophical Psychology, 32: 720–759. Debus, D. 2010. “Accounting for epistemic relevance: a new problem for the causal theory of memory”. American Philosophical Quarterly, 47: 17–29.

360

Implicit Memory Drayson, Z., and Schwartz, A. 2019. “Intellectualism and the argument from cognitive science”. Philosophical Psychology, 32: 662–692. Donald, M. 1991. Origins of the modern mind. Cambridge, MA: Harvard University Press. Fantl, J. 2017. “Knowledge how”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Fall 2017 edn. https://plato.stanford.edu/archives/fall2017/entries/knowledge-how/. Furlong, E. J. 1951. A study in memory. New York: Thomas and Sons. Hawley, K. 2003. “Success and knowledge-how”. American Philosophical Quarterly, 40: 19–31. Hume, D. 1739/1978. A treatise of human nature. Ed. L. A. Selby-Bigge. Oxford: Clarendon. Hutto, D. D., and Peeters, A. 2018. “The roots of remembering: radically enactive recollecting”. In K. Michaelian, D. Debus, and D. Perrin, eds., New directions in the philosophy of memory. London: Routledge: 97–118. James, W. 1890. The principles of psychology. London: Macmillan. Johnson, M. K. 1992. “MEM: mechanisms of recollection”. Journal of Cognitive Neuroscience, 4: 268–280. Klein, S. B. 2015. “What memory is”. Wiley Interdisciplinary Reviews: Cognitive Science, 6: 1–38. Locke, J. 1694/1979. Essay concerning human understanding. Ed. P. H. Nidditch. Oxford: Clarendon. Loftus, E. F. 2003. “Make-believe memories”. American Psychologist, 58: 864–873. Lycan, W. G. 1986. “Tacit belief ”. In R. Bogdan, ed., Belief: form, content, function. Oxford: Oxford University Press. McCarroll, C. J. 2018. Remembering from the outside: personal memory and the mind. Oxford: Oxford University Press. Michaelian, K. 2016. Mental time travel: episodic memory and our knowledge of the personal past. Cambridge, MA: MIT Press. Montero, B. G. 2020. “What experience doesn’t teach: pain amnesia and a new paradigm for memory research”. Journal of Consciousness Studies, 27: 102–125. Moyal-Sharrock, D. 2009. “Wittgenstein and the memory debate”. New Ideas in Psychology, 27: 213–227. Pavese, C., and De Brigard, F. 2019. “Editor’s introduction to special issue: memory and skill”. Philosophical Psychology, 32: 585–587. Reid, T. 1785/1951. Essays on the intellectual powers of man. Ed. A. D. Woozley. London: MacMillan and Co. Robins, S. K. 2017. “Memory traces”. In S. Bernecker and K. Michaelian, eds., Routledge Handbook of the Philosophy of Memory. London: Routledge: 76–87. Roediger, H. L. 1990. “Implicit memory: retention without remembering”.  American Psychologist, 45: 1043–1056. Rowlands, M. 2016. Memory and the self: phenomenology, science, and autobiography. Oxford: Oxford University Press. Russell, B. 1912. The problems of philosophy. London: Williams and Norgate. Ryle, G. 1949. The concept of mind. London: Hutchinson & Co. Schacter, D. L. 1987. “Implicit memory: history and current status”. Journal of Experimental Psychology: Learning, Memory, and Cognition, 9: 39–54. Schacter, D. L. 1992. “Understanding implicit memory: a cognitive neuroscience approach”. American Psychologist, 47: 559–569. Schacter, D. L. 2019. “Implicit memory, constructive memory, and imagining the future: a career perspective”. Perspectives on Psychological Science, 14: 256–272. Schacter, D. L., and Tulving, E. 1994. “What are the memory systems of 1994?”. In D. L. Schacter and E. Tulving, eds., Memory systems 1994. Cambridge, MA: MIT Press: 1–38. Schacter, D. L., Wagner, A. D., and Buckner, R. L. 2000. “Memory systems of 1999”. In E. Tulving and F. I. M. Craik, eds., The Oxford handbook of memory. New York: Oxford University Press. 627–643. Shimamura, A. P., and Squire, L. R. 1986. “Memory and metamemory: a study of the feeling-of-knowing phenomenon in amnesic patients”. Journal of Experimental Psychology: Learning, Memory, and Cognition, 12: 452–460. Squire, L. R. 2004. “Memory systems of the brain: a brief history and current perspective”. Neurobiology of Learning and Memory, 82: 171–177. Stanley, J., and Williamson, T. 2001. “Knowing how”. Journal of Philosophy, 98: 411–444. Sutton, J. 1998. Philosophy and memory traces: Descartes to connectionism. Cambridge: Cambridge University Press. Sutton, J. 2010. “Exograms and interdisciplinarity: history, the extended mind, and the civilizing process”. In R. Menary, ed., The extended mind. Cambridge, MA: MIT Press: 189–225. Wallis, C. 2008. “Consciousness, context, and know-how”. Synthese, 160: 123–153.

361

28 MEMORY DURING FAILURES OF RECALL Information That Is Forgotten Is Not Gone Anne M. Cleary Why do people forget, and are there forms of memory still available for use when forgetting occurs? The answer to this question could potentially aid in developing methods of preventing or circumventing forgetting, which could be useful in a wide range of contexts, including for students at exam time, for eyewitnesses attempting to recall details of a crime scene, for patients who need to remember to take their medication, and even for rehabilitation strategies for patients of who have experienced memory impairment. Although it may seem, at an intuitive level, that when we have forgotten something, it is completely gone from memory forever, usually that is not the case. Usually, the information is still present in memory, and it is simply failing to be accessed at the moment. In some of the early work to suggest this, researchers compared memory performance on a cued recall test – whereby a cue is presented, such as the category name for the to-be-recalled information (e.g., “fruit” for the to-be-recalled study list words “apple,” “banana,” “pear,” and “plum”) – with performance on a free recall test (whereby no cue is presented and the person must simply try to freely recall as many items from the study list as possible). Even though participants received identical encoding conditions, those who received a cued recall test retrieved significantly more of the encoding-phase items from memory than those who received a free recall test (Tulving and Pearlstone 1966). As the encoding conditions were identical, this pattern suggests that more information had been encoded into memory than could be retrieved in the group that received the free recall test. That is, there was information in memory that was failing to be accessed in that group. This is known as retrieval failure. Instead of being caused by the decay or disappearance of information from memory, forgetting in this case was due to a failure to retrieve the memory from its place within the system. Importantly, this work suggests that the memory can still be present – just inaccessible (hence, Tulving and Pearlstone distinguish between accessibility and availability). Among memory researchers today, it is now thought that the mechanisms of forgetting involve competition among memory representations for access in response to whatever cues are available at the moment (e.g., Kuhl and Wagner 2009). When a memory cannot be accessed at the moment, it is thought to be because other memories are competing for access and getting in the way of that particular memory coming to mind. This means that there is the possibility that this particular memory will come to mind at a later point in time, which is one reason why

362

DOI: 10.4324/9781003014584-36

Memory During Failures of Recall

taking breaks or temporarily switching tasks can useful for generating a relevant idea for solving a problem (e.g., Smith et al. 2015).

How Unrecalled Memories Exert an Influence Besides the possibility that a memory that fails to consciously come to mind at one moment can still come to mind at a later time under the right circumstances (e.g., Smith et al. 2015), there are other ways that an unrecalled memory can exert an influence. One way is through eliciting a metacognitive feeling or sensation alerting the person to the unrecalled memory’s presence within the memory system. Another is through exerting effects on a person’s behavior or task performance, about which the person is completely unaware. Both of these possibilities are elaborated on later.

Consciously Detecting a Memory During Its Retrieval Failure The Tip-of-the-Tongue State An example of a consciously detectable feeling or sensation that alerts a person to an unrecalled memory’s presence is the tip-of-the-tongue (TOT) state (Brown 2012; Schwartz 1999, 2001, 2002; Schwartz et al. 2000). During a TOT state, the experiencer feels the presence of a word in memory despite a failure to call that word to mind. While the feeling itself is conscious (the person is aware of the TOT feeling itself), conscious access to the word in question is failing to occur. An interesting metacognitive phenomenon in its own right, the TOT state is not merely an illusory sensation of the presence of something in memory. Some external validation exists for the TOT’s ability to actually signal the presence of the target word in memory. For example, people are more likely to recognize the target when presented with it later (e.g., Cleary et al. 2021) and may even sometimes have access to related information (see Brown 2012, or Schwartz 2002, for reviews). Thus, the feeling itself can be a valid indicator of the presence of a currently inaccessible word in memory, and the detection that the person experiences is conscious. People not only consciously report on the feeling or sensation of the TOT, but also can strategically use the presence of the TOT state to make choices and to guide behavior (Cleary et al. 2021; Metcalfe et al. 2017). For example, Cleary et al. (2021) examined participant decisions and performance in a risk-involved testing situation in which participants first attempted to provide a short answer to a general knowledge question. If retrieval of the short answer failed, the participants could decide whether or not to take the risk involved in requesting to see a set of multiple-choice options. Opting to be presented with the multiple-choice options was risk-involved because if the wrong answer was selected from among the options, a point would be deducted from the score. However, there was also the potential to gain points, as a correct selection would result in a gain of one point. Choosing not to be presented with the multiple-choice options would result in no point gain or loss. Cleary et al. found that participants strategically used the presence versus the absence of a TOT state to decide when to take the risk of being presented with the multiple-choice options versus not. Specifically, participants were more likely to request the multiple-choice options when in a TOT state than when not. Moreover, the strategy was an advantageous one, as TOT states were predictive of selecting the correct answer from among the multiple-choice options. This points toward the idea that not only are TOT states a consciously detectable indication of the presence of a memory that cannot be retrieved, but they are also able to guide decisions and behavior in ways suggested by some metacognitive theorists (e.g., Nelson and Narens 1990).

363

Anne M. Cleary

Detecting Familiarity Another example of a consciously detectable signal or sensation that alerts a person to an unrecalled memory’s presence is familiarity. People can detect the increased sense of familiarity that emerges from cues that resemble earlier-encoded items, even when those earlier-encoded items cannot be recalled from memory. This is known as recognition without recall (e.g., Cleary 2004; Cleary et al. 2009; Cleary et al. 2016; Ryals and Cleary 2012). An example of the recognition without recall effect is as follows. Participants view a study list of words (e.g., pitchfork, elegant) in an encoding phase. The participants are subsequently presented with a test list containing cues; half of the cues graphemically resemble studied words (e.g., potchbork, erligant) and half do not (e.g., fronnel, poxen). The task is a cued recall task with an additional component involving a cue familiarity judgment. For each cue, the participant is to attempt to use the cue to recall a word from the study list that resembles it. Even if no word can be recalled, the participant is to rate how familiar the cue itself seems using a scale of zero (very unfamiliar) to ten (very familiar). Recognition without recall is the finding that, among cues for which recall of the corresponding studied item fails, higher familiarity ratings are given than among cues that do not resemble a studied item (e.g., Cleary 2004; Ryals and Cleary 2012). For example, if the cue “potchbork” fails to elicit recall of the graphemically similar studied item “pitchfork,” it will tend to receive higher familiarity ratings than a cue that does not correspond to a studied item. What is thought to occur during familiarity-detection is the ability to discern the relative strength of a familiarity signal. According to mathematical models of familiarity-detection, the signal itself emerges from feature-matching process that occurs between a test item (a probe) and all of the items that were recently stored in memory (the memory traces). Importantly, the feature-matching process is global in nature, such that the features in the test item are matched on a feature by feature basis to all of the items stored in memory (e.g., see Clark and Gronlund 1996, for a review). The recognition without recall effect seems to abide by this global matching principle in that when a test cue overlaps in features with multiple unrecalled studied items, familiarity ratings are even higher than when the test cue overlaps in features with only one studied item (e.g., Cleary et al. 2016; Huebert et al. in press; Ryals and Cleary 2012). For example, Ryals and Cleary found that cue familiarity ratings were higher when a test cue like “potchbork” corresponded to four different unrecalled studied items (e.g., “pitchfork,” “patchwork,” “pocketbook,” and “pullcork”) instead of just one studied item (e.g., “pitchfork”) that happened to go unrecalled. It is worth noting that the preceding description assumes at least some degree of mapping between the cognitive process of familiarity itself (i.e., as described by global matching models) and the metacognitive process of familiarity-detection. Tulving (1989) pointed out that the assumption that memory performance and memory as a conscious experience are strongly correlated (what he termed the doctrine of concordance) may not always hold. Although the cognitive process underlying recognition without recall may be well-described by global matching models, how people metacognitively detect relative increases or decreases in familiarity during recall failure and whether this metacognitive familiarity-detection maps onto models of the familiarity process remains an important question. Does a familiarity signal emerge into consciousness with varying possible degrees of intensity? If so, do people metacognitively make decisions such as where to place criteria along that familiarity signal intensity continuum in order to behaviorally indicate that varying intensity? In short, how well do theories of familiarity’s operation as a process map onto conscious subjective experiences of memory? The ability to detect familiarity during recall failure that occurs during recognition without recall may be the same as that which gives rise to the occasional real-life feeling or sensation of 364

Memory During Failures of Recall

recognizing someone or something as familiar despite being unable to pinpoint the source of that familiarity. The most common example of this real-life phenomenon is what has become known as the butcher-on-the-bus phenomenon (Brown 2020), which may be a misnomer or miscredited (MacLeod 2020), but refers to the example of encountering one’s butcher while traveling on the city bus – there may be a feeling of recognition for the person that cannot be placed when the person is encountered outside of the usual context. It is as yet unclear if the familiarity-detection tapped through the recognition without recall paradigm involves the same conscious metacognitive sense described by the butcher-on-the-bus example, and as described later, some research even points toward the possibility that familiarity-detection from features in a similar paradigm might actually be due to unconscious recognition processes (Ryals et al. 2011); if so, then participants might not necessarily be aware of when their familiarity ratings are higher versus lower or that their ratings are systematically differentiating familiarity levels based on objective manipulations of feature-overlap to studied items. Future research should further aim to delineate these two possibilities.

Déjà vu Another possible example of a consciously detectable feeling or sensation that alerts a person to an unrecalled memory’s presence – and one that relates to familiarity-detection – is déjà vu. Déjà vu is the feeling of having experienced something before despite knowing evidence to the contrary (this is actually a new experience). The idea that déjà vu is a kind of metacognitive familiarity during recall’s absence is a long-held theoretical explanation for déjà vu (Brown 2004; Cleary and Brown 2022), and one that is supported by empirical evidence (e.g., Cleary et al. 2009; Cleary et al. 2012; Cleary and Claxton 2018; Cleary et al. 2018). In a variant of the recognition without recall paradigm, Cleary (2012) had participants view a series of scenes at study through immersive virtual reality. Later on in the test phase, participants viewed a series of new scenes, half of which shared the same spatial layout as a studied scene and half of which did not. Among test scenes that failed to elicit recall of their corresponding spatially mapped studied scenes, a higher likelihood of reporting déjà vu was found than among test scenes that did not map onto any studied scene. Like with the recognition without recall effect (Cleary 2012; Ryals and Cleary 2012), global matching might be involved. This is because the greater the degree of match there was between a test scene and a forgotten scene in memory, the greater the likelihood of reporting déjà vu, which was shown in a second experiment by Cleary (2012) in which they included some of the exact scenes from the study phase among the test scenes. When participants failed to recognize that they had already explored those particular scenes (incorrectly calling them new scenes), they were even more likely to report experiencing déjà vu for those scenes. Thus, the more similar a seemingly novel scene is to one in memory that fails to be retrieved, the more likely it may be that a sense of déjà vu will occur. The conscious subjective experience of déjà vu is thought to differ from standard familiaritydetection in that déjà vu involves a sense of impossibility about the perceived familiarity – the situation feels familiar yet it seems impossible for it to be so, leading the feeling to be perplexing (Brown 2004; Cleary and Brown 2022). This is part of why the recognition without recall paradigm has been used to study déjà vu, as it presents a means of juxtaposing familiarity and novelty. In the paradigm, the cues are novel, yet may still seem familiar insofar as they resemble something that was recently encountered. In the variant of the recognition without recall paradigm that has been used to study déjà vu (Cleary 2012; Cleary and Claxton 2018; Cleary et al. 2018), this juxtaposition of familiarity and novelty occurs through the fact that the test scenes are typically novel within the context of the experiment, but some may seem familiar 365

Anne M. Cleary

due to their familiar spatial layout that maps onto a recently viewed scene. When recall fails, presumably the person experiencing the recall failure is left merely with a sense of familiarity with the scene possibly also accompanied by a recognition of its novelty. Though some have suggested a role of conflict detection in déjà vu (Urquhart et al. 2021), the extent to which this juxtaposition of seemingly opposing feelings or sensations is part of the conscious metacognitive experience that is déjà vu has not yet been fully explored and is a ripe area for future research.

Are There Varying Degrees of Conscious Awareness of Memories? Given that recall can fail and that during such instances, feelings or sensations of memory are sometimes still consciously detectable, a worthwhile question concerns whether memories can elicit varying degrees of conscious awareness of their presence on a spectrum. In a new framework from which to approach the study of memory, Rubin (2021) suggests a dimensional approach as an alternative to the standard hierarchical categorical approach of separating memory first into explicit and implicit categories of memory, then separating explicit memory into the categories of episodic and semantic memory and implicit memory into categories such as skills, priming, classical conditioning, etc. In Rubin’s dimensional approach, there are three primary dimensions of memory, each existing on a continuum: 1) The degree to which the memory involves a scene, 2) the degree to which the memory involves self-reference, and 3) the degree to which the memory is explicit vs. implicit. Lying at the high end of all three continua would tend to characterize conscious, recollective experiences of prior episodes or events that one has experienced. Rubin argues that déjà vu might be an example of a type of memory lying at the high end of the scene continuum and at the high end of the self-reference continuum, but lying at the implicit end of the implicit-explicit continuum. This raises an intriguing question concerning at what point a memory is exerting an influence completely outside of conscious awareness (i.e., at the extreme implicit end of the implicit-explicit continuum) versus is eliciting a consciously detectable feeling or sensation of memory despite a lack of full-blown conscious recollection (i.e., somewhere in between the implicit and explicit ends of the continuum). Given that déjà vu is itself a conscious experience, it seems unlikely to be completely implicit. However, if indeed memories vary along a spectrum going from completely unconscious/unaware to a full blown recollective experience, then it is possible that the aforementioned feelings of memory (déjà vu, familiarity-detection, and TOT experiences) may reside somewhere along that continuum partway between implicit and explicit. The feeling or sensation itself might be said to be explicit while its source (i.e., the actual item or the prior experience itself) remains inaccessible to consciousness. That said, if a form of memory resides at the extreme implicit end of the implicit-explicit continuum, an implication is that it exerts its influence completely outside of the person’s awareness.

Influencing a Person’s Behavior Outside of Awareness: Implicit Memory Although there is reason to believe that non-conscious memory systems exist, for which the goal is not to enable conscious recall but rather to learn information that can rapidly be applied and used (e.g., see Seger 1994, for a review of implicit learning), there is also reason to suspect that memories that are currently consciously inaccessible (due to either mechanisms of forgetting or to occluded perceptual information that limits conscious access to the representation, described later) may have the potential to exert an influence on behavior completely outside of one’s awareness. An example is implicit memory (Roediger and McDermott 1993; Schacter 366

Memory During Failures of Recall

et  al. 1993). In an implicit memory task – of which there are many variants – participants generally study a list of items followed by a test of memory that is indirect in nature. For example, the test might be what is called a perceptual identification task, in which each test item is rapidly flashed (such as for 30–50 ms) and masked (with a pre- and post-mask of symbols or lines meant to hinder the ability to identify the rapidly flashed item). For the participant, the task is to attempt to identify as many items as possible on the test. Typically, the test list of these rapidly flashed, masked items will be a random ordering of items that were studied and items that were not studied on an earlier list. Note that although there is a mix of studied and nonstudied items, there is no requirement to remember or to identify whether a given test item was studied or not. Thus, the participant does not need to consciously draw upon memory for earlier-presented study list items, yet participants will tend to identify significantly more studied than non-studied items from their rapidly flashed, masked presentation. Evidence points toward this type of memory being able to take place at an unconscious level: People can be unaware of the fact that they are identifying more studied than non-studied items. The best example of this is shown among amnesic patients, who, despite being unable to consciously recall having even been presented with an earlier study list, can usually still demonstrate the tendency to identify more studied than non-studied items in an indirect task such as the perceptual identification task (Schacter et al. 1993). An interesting manifestation of implicit memory is shown in a study by Eich (1984). Participants engaged in a dichotic listening task in which they were to verbally shadow (repeat out loud) the message being played in one ear (the attended ear) while ignoring a simultaneous message occurring in the other ear (the unattended ear). In the unattended ear, pairs of words were spoken such that uncommon spellings of homophones (e.g., REEL) would be primed by an immediately preceding word (e.g., MOVIE). Following the dichotic listening task, participants were given an old-new recognition test for which they needed to attempt to discriminate words heard in the unattended ear during the earlier dichotic listening task from new words not played at all during the dichotic listening task. Following the recognition test, participants were then given a spelling test in which they were asked to listen to words and spell them. Despite being unable to discriminate on the recognition test which words had been heard vs. unheard during the earlier dichotic listening task, participants tended to use the uncommon spellings for the homophones (e.g., REEL instead of REAL) that had been primed during the earlier dichotic listening task. This finding suggests that a form of implicit memory was taking place in which the participants were presumably unaware of the recent basis for their choice of spelling. Another interesting manifestation of implicit memory is a finding demonstrated by KunstWilson and Zajonc (1980). These researchers showed that when stimuli were presented in such a perceptually impoverished format at encoding that participants could not discriminate between studied and non-studied items on a later recognition test, participants indicated stronger liking for the stimuli that were studied than for the stimuli that were not. As participants were unable to discern which items were studied versus not studied, the impact of the memory for earlier-presented items – manifesting in the form of increased preference or liking – appears to be unknown to participants; although they may know what the prefer, they are likely unaware that a recent unretrieved memory is the cause. Note that there are some critical differences between this paradigm and the recognition without recall paradigm described earlier. Whereas in the study by Kunst-Wilson and Zajonc (1980) participants showed no ability to discriminate studied from unstudied items, in the recognition without recall paradigm, participants show an ability to discriminate cues corresponding to studied items from cues corresponding to non-studied items despite failing to recall the earlier studied items in question (Cleary 2004; Cleary et al. 2009; Cleary 2012; Cleary et al. 367

Anne M. Cleary

2016; Ryals and Cleary 2012). Thus, it is likely that whereas the type of memory shown by Kunst-Wilson and Zajonc is unconscious (occurring completely outside of participant awareness that their preferences are being influenced by unretrieved memories), the type of memory manifesting in recognition without recall could plausibly be occurring at least somewhat within participants’ awareness. As mentioned, one way that discrimination during recall failure could take place within awareness is through reliance on a consciously detectable familiarity signal that is higher on average for cues corresponding to studied items than for cues corresponding to non-studied items. In this case, the familiarity signal itself is available to awareness; sometimes the signal may feel stronger and other times weaker. It is as yet unknown if this type of signal, perhaps when barely perceptible, might also underlie liking ratings in the study by KunstWilson and Zajonc (1980), unbeknownst to the participant. If a signal of memory exists on a continuum, then it is conceivable that the it might range from being so subtle as to not be consciously detectable but able to influence behavior to being detectable as a feeling or sensation of memory to enabling or prompting full-blown recollective experience. To tie the notion of implicit memory back to Rubin’s (2021) dimensional approach to memory, implicit memory would likely be any type of memory that falls at the extreme implicit end of the implicit-explicit continuum. For example, priming on the aforementioned perceptual identification task that occurs outside of awareness would likely be at the extreme implicit end of the implicit-explicit continuum (while also being low on the scene continuum and low on the self-reference continuum). As another example, and to contrast implicit forms of memory with déjà vu, whereas the increased familiarity and reports of déjà vu shown during scene recall failure for scenes that share the same spatial layout as an earlier viewed scene (e.g., Cleary 2012; Cleary and Claxton 2018; Cleary et al. 2018) likely reflect consciously detectable familiarity residing somewhere between implicit and explicit on the presumed implicit-explicit continuum, implicit memory involving scenes might take place as follows. If participants were asked to navigate through a novel scene that shared the same spatial layout as an unrecalled earlier viewed scene and, despite feeling no déjà vu and no sense of familiarity for the scene, unknowingly followed the exact same navigational path taken in the earlier toured scene, this might be an example of a form of memory at the extreme implicit end of the implicit-explicit continuum while being high on the scene continuum and possibly even high on the self-reference continuum (as there is a firstperson perspective to navigating scenes).

On the Relationship Between Implicit Memory, Fluency, and Awareness During implicit memory, a person demonstrates an influence of an encoded memory through an effect of that memory on behavior, despite an apparent lack of awareness of the memory’s role in that behavior. Examples include being able to identify more perceptually degraded items if they had been earlier studied than if they had not (Roediger and McDermott 1993; Schacter et al. 1993), and demonstrating a greater liking or preference for items that had been earlier encoded so shallowly as to be non-discernable from new items on a recognition test (KunstWilson and Zajonc 1980). In contrast, during subjective feelings or sensations of memory during retrieval failure, like the detection of familiarity for example, there is thought to be at least some degree of metamemorial awareness occurring. It is as yet unclear what exactly distinguishes these two manifestations of memory during the absence of conscious recall. Although some have suggested that the feeling of familiarity might arise from detection of increased fluency or ease of processing (e.g., Jacoby et al. 1997; Jacoby and Whitehouse 1989), which occurs 368

Memory During Failures of Recall

during implicit memory insofar as studied items are more easily identifiable later on even if their previous encounter is unrecalled, fluency explanations have difficulty accounting for some of the metacognitive sensations of memory that occur during the absence of conscious recall. The reason is that a large amount of the information that we encounter day to day should be fluently processed – we identify words, people, objects and so on effortlessly and frequently – yet most of the time, we do not have the sensations of memory illustrated by the butcher-on-the-bus example or the déjà vu experience. In an effort to address this, Whittlesea and Williams (1998) wrote an article entitled, “Why Do Strangers Feel Familiar, but Friends Don’t? A Discrepancy-Attribution Account of Feelings of Familiarity.” In their discrepancy-attribution account (see also, Whittlesea and Williams 2001), Whittlesea and Williams proposed that experiences like the butcher-on-the-bus example occur when a sense of fluency is surprising within its context. So, for example, one’s butcher might be as fluently processed on the bus as in the butcher shop, but outside of the usual familiar paired context of the butcher shop, that fluency is surprising. Thus, in the context of the bus, the fluency with which the butcher’s face is processed is surprising, and it is ultimately that surprising sense of fluency that produces the metacognitive familiarity experience referred to as the butcher-on-the-bus phenomenon. Note that the same explanation could apply to déjà vu: Perhaps the déjà vu experience only occurs when a sense of familiarity is surprising because it is occurring within a novel context. However, research thus far does not seem to support the discrepancy attribution hypothesis. The discrepancy-attribution hypothesis was largely based on a previous finding that structurally regular nonwords (e.g., “hension”) were surprisingly fluent to people. The argument was that whereas everyday fluent things are not surprisingly fluent, when fluency is detected in a surprising context, that is when the metacognitive experience of an odd sense of familiarity will occur. For example, a word (e.g., peacock) will tend to be processed fairly fluently, but that fluency will not be surprising, because the fluency can be attributed to the fact that it is a word. In contrast, a structurally regular nonword (e.g., hension) will be surprisingly fluent, because there will be a discrepancy between its fluency of processing and the expectation about its fluency – its fluency will not have an easy attribution because it is not an actual word. So, the fluency of a structurally regular word should give rise to an increased subjective sense of familiarity. To test this, Whittlesea and Williams (1998) predicted that participants should show higher false alarms on a recognition test to structurally regular nonwords like “hension” than to either actual words (e.g., “peacock”) or to structurally irregular nonword (e.g., pnertap). While they did find this pattern (Whittlesea and Williams 1998, 2001), later research (Cleary et al. 2007) showed that the higher false alarm rate to the “hension” items was actually due to a stimulus confound. Specifically, the pool of structurally regular nonwords that Whittlesea and Williams had used had a higher degree of inter-item similarity among them, or “interstimulus similarity.” Given that similarity to a studied item is a factor that will increase false alarm rates, this was a confound. Cleary et al. showed that it was the interstimulus similarity that was driving the higher false alarm rates for the structurally regular nonwords rather than “surprising fluency.” Once the interstimulus similarity confound was controlled for, the obtained pattern was different, and was actually consistent with what they termed the “structural regularity hypothesis,” according to which, memory should generally be better for structurally regular items. In their article “False Memories Are Not Surprising: The Subjective Experience of an Associative Memory Illusion,” Karpicke et al. (2008) followed up on Whittlesea’s (2002) assertion that false memories might also be accompanied by a surprising sense of fluency (see also Whittlesea et al. 2005) and found no support for the idea that false memories were surprising to participants. To date, there is little direct empirical support for the discrepancy-attribution 369

Anne M. Cleary

hypothesis. Thus, the precise relationship between implicit memory and more conscious sensations of familiarity during recall failure has yet to be fully understood. It is possible that these are based on different mechanisms altogether, or that there are simply varying degrees of implicit versus explicit memory, as proposed by Rubin (2021) in his dimensional approach to memory. It is also possible that there is merit to the discrepancy-attribution hypothesis, and that new research paradigms will bear this out. For example, Schwartz and Metcalfe (2011) made similar arguments to the discrepancy attribution account regarding tip-of-the-tongue states, pointing out that mere retrieval failure is not enough to produce a tip-of-the-tongue state, and that it may be when an inordinate amount of time is spent trying to retrieve (as opposed to quickly failing to retrieve or quickly succeeding at retrieval) that a tip-of-the-tongue state arises. In this way, there may be a discrepancy between one’s retrieval process expectation (i.e., an expectation of a fairly rapid success or failure) and the experience (i.e., a surprisingly long attempt that fails to turn up the word) when a tip-of-the-tongue state occurs. It is also possible that the subjective metacognitive experience involved in the butcher-onthe-bus example arises from different mechanisms than the familiarity-detection shown in the recognition without recall paradigm. The finding that people give higher cue familiarity ratings to cues that resemble unrecalled studied items than to cues that do not resemble any studied items could be due to unconscious memory processes. Even though it is an explicit memory task and participants are being asked to give a judgment based on their sense of familiarity in relation to the earlier presented study list, it is theoretically possible for such ratings to be differentiating the cues based on processes occurring outside of participants’ conscious awareness as opposed to being based on a conscious judgment of familiarity. Some evidence pointing toward this possibility is the finding that a neural signature thought to be associated with unconscious processes in recognition memory paradigms (Voss and Paller 2009) has been shown to discriminate feature-based cues coming from studied items and feature-based cues coming from unstudied items during recall failure in a similar paradigm to the recognition without recall paradigm (Ryals et al. 2011). Some have suggested that unconscious processes may play more of a role in recognition memory than has been generally acknowledged or investigated among memory researchers (e.g., Cleary 2012; Voss et al. 2012). If Rubin (2021) is correct that the implicit-explicit distinction lies on a continuum, then there may be an unclear boundary between the contribution of an unconscious process to a recognition decision (or other type of decision, such as likeability) and that of a vague, consciously detectable feeling or sensation like subtle familiarity. That said, the fact that a variant of the recognition without recall paradigm is associated with metacognitive reports of déjà vu (e.g., Cleary et al. 2009; Cleary 2012; Cleary and Claxton 2018; Cleary et al. 2018) suggests that the familiarity detected in the recognition without recall paradigm is at least in some cases associated with a conscious metacognitive experience. Future research should examine whether subjective metacognitive feelings of memory (like subjective feelings of familiarity, déjà vu, and even tip-of-the-tongue experiences) are able to be differentiated from manifestations of memory that are completely outside of awareness and how these forms of memory might interrelate.

Summary, Conclusions, and Future Directions Because forgetting of information is often due to retrieval failure (e.g., Kuhl and Wagner 2009; Tulving and Pearlstone 1966), when forgetting occurs, the information is often still present in memory. The fact that information can still be present in memory despite a failure to retrieve it raises the possibility that memories can exert an influence on a person even when failing to be 370

Memory During Failures of Recall

consciously retrieved. Indeed, there appear to be two general categories of ways in which memory can manifest during an absence of conscious retrieval. The first is the conscious detection of a memory during its retrieval failure, as occurs when one experiences a tip-of-the-tongue state for an unretrieved word or a feeling of familiarity for a situation whose past encounter cannot be identified. In these cases, the unretrieved memory is somehow able to signal its presence in the form of a feeling or sensation of memory. The second way in which memory can manifest during an absence of conscious retrieval is through affecting a person’s judgments or behavior outside of the person’s apparent awareness, as occurs during implicit memory. The relationship between these two forms of memory – and how they might differ – is not yet fully understood. A possible fruitful avenue for future research exploration on this issue is Rubin’s (2021) new dimensional framework for the study of memory, according to which the implicit-explicit distinction lies on a continuum that intersects with two other continua: 1) the extent to which the memory involves a scene and 2) the extent to which the memory involves self-reference. In approaching the implicit-explicit distinction from the perspective that the distinction lies on a continuum, future research should attempt to delineate the boundary between unconscious and conscious contributions of memory that occur during the absence of conscious recollective experience.1

Related Topics Chapters 1, 2, 16, 17, 21, 22, 23, 27, 31

Notes 1 I am extremely grateful to Bennett Schwartz and Joseph Neisser for their invaluable comments on an earlier draft of this chapter. Their comments and suggestions greatly improved the chapter.

References Brown, A. S. 2004. The déjà vu experience. New York: Psychology Press. Brown, A. S. 2012. The tip of the tongue state. Hove, UK: Psychology Press. Brown, A. S. 2020. “The butcher on the bus experience”. In A. M. Cleary and B. L. Schwartz, eds., Memory quirks: the study of odd phenomena in memory. London: Routledge. Clark, S. E., and Gronlund, S. D. 1996. “Global matching models of recognition memory: how the models match the data”. Psychonomic Bulletin and Review, 3: 37–60. Cleary, A. M. 2004. “Orthography, phonology, and meaning: word features that give rise to feelings of familiarity in recognition”. Psychonomic Bulletin and Review, 11: 446–451. Cleary, A. M. 2012. “On the contribution of unconscious processes to recognition memory”. Cognitive Neuroscience, 3: 210–211. Cleary, A. M., and Brown, A. S. 2022. The déjà vu experience. 2nd edn. London: Routledge. Cleary, A. M., and Claxton, A. B. 2018. “Déjà vu: an illusion of prediction”. Psychological Science, 29: 635–644. Cleary, A. M., McNeely-White, K.L, Huebert, A. M., and Claxton, A. B. 2018. “Déjà vu and the feeling of prediction: an association with familiarity strength”. Memory, 29: 904–920. Cleary, A. M., McNeely-White, K. L., Russell, S. A., Huebert, A. M., and Hausman, H. 2021. “The tip-of-the-tongue state as a form of access to information: use of tip-of-the-tongue states for strategic adaptive test-taking”. Journal of Applied Research in Memory and Cognition, 10: 131–142. Cleary, A. M., Morris, A. L., and Langley, M. M. 2007. “Recognition memory for novel stimuli: The structural regularity hypothesis”. Journal of Experimental Psychology: Learning, Memory, and Cognition, 33: 379–393. Cleary, A. M., Ryals, A. J., and Nomi, J. N. 2009. “Can déjà vu result from similarity to a prior experience? Support for the similarity hypothesis of déjà vu”. Psychonomic Bulletin and Review, 16: 1082–1088.

371

Anne M. Cleary Cleary, A. M., Ryals, A. J., and Wagner, S. M. 2016. “Recognition during recall failure: semantic feature matching as a mechanism for recognition of semantic cues when recall fails”. Memory and Cognition, 44: 50–62. Eich, E. 1984. “Memory for unattended events: remembering with and without awareness”. Memory and Cognition, 12: 105–111. Huebert, A. M., McNeely-White, K. L., and Cleary, A. M. (in press). “Can cue familiarity during recall failure prompt illusory recollective experience?”. Memory  & Cognition. https://doi.org/10.3758/ s13421-021-01248-0 Jacoby, L. L., and Whitehouse, K. 1989. “An illusion of memory: false recognition influenced by unconscious perception”.  Journal of Experimental Psychology: General, 118: 126–135. https://doi. org/10.1037/0096-3445.118.2.126 Jacoby, L. L., Yonelinas, A. P., and Jennings, J. M. 1997. “The relation between conscious and unconscious (automatic) influences: a declaration of independence”. In J. D. Cohen and J. W. Schooler, eds., Scientific approaches to consciousness. Mahwah, NJ: Erlbaum: 13–47 Karpicke, J. D., McCabe, D. P., and Roediger, H. L. 2008. “False memories are not surprising: the subjective experience of an associative memory illusion”. Journal of Memory and Language, 58: 1065–1079. Kuhl, B. A., and Wagner, A. D. 2009. “Forgetting and retrieval”. In G. G. Berntson and J. T. Cacioppo, eds., Handbook of neuroscience for the behavioral sciences. Vol. 1. Hoboken, NJ: John Wiley and Sons Inc: 586–605. https://doi.org/10.1002/9780470478509.neubb001031 Kunst-Wilson, W. R., and Zajonc, R. B. 1980. “Affective discrimination of stimuli that cannot be Recognized”. Science, 207: 557–558. https://doi.org/10.1126/science.7352271 MacLeod, C. M. 2020. “The butcher on the bus: a note on familiarity without recollection”. History of Psychology, 23: 383–387. https://doi.org/10.1037/hop0000178 Metcalfe, J., Schwartz, B. L., and Bloom, P. A. 2017. “The tip-of-the-tongue state and curiosity”. Cognitive Research: Principles and Implications, 2: 31. Nelson, T. O., and Narens, L. 1990. “Metamemory: a theoretical framework and some new Findings”. In G. H. Bower, ed., The psychology of learning and motivation. Vol. 26. Cambridge, MA: Academic Press: 125–173 Roediger, H. L., and McDermott, K. B. 1993. “Implicit memory in normal human subjects”. In H. Spinnler and F. Boller, eds., Handbook of neuropsychology. Vol. 8. Amsterdam: Elsevier: 63–131. Rubin, D. C. 2021. “A  conceptual space for episodic and semantic memory”. Memory and Cognition. Advance online publication. https:// doi.org/10.3758/s13421‐021‐01148‐3 Ryals, A. J., and Cleary, A. M. 2012. “The recognition without cued recall phenomenon: support for a feature-matching theory over a partial recollection account”. Journal of Memory and Language, 66: 747–762. Ryals, A. J., Yadon, C. A., Nomi, J. S., and Cleary, A. M. 2011. “When word identification fails: ERP correlates of recognition without identification and of word identification failure”. Neuropsychologia, 12: 3224–3237. Schacter, D. L., Chiu, C. Y. P., and Ochsner, K. N. 1993. “Implicit memory: a selective review”. Annual Review of Neuroscience, 16: 159–182. Schwartz, B. L. 1999. “Sparkling at the end of the tongue: the etiology of tip-of-the-tongue Phenomenology”. Psychonomic Bulletin and Review, 6: 379–393. Schwartz, B. L. 2001. “The relation of tip-of-the-tongue and retrieval time”. Memory and Cognition, 29: 117–126. Schwartz, B. L. 2002. Tip-of-the-tongue states: phenomenology, mechanism, and lexical retrieval. Mahwah, NJ: Lawrence Erlbaum Associates. Schwartz, B. L., and Metcalfe, J. 2011. “Tip-of-the-tongue (TOT) states: retrieval, behavior, and experience”. Memory and Cognition, 39: 737–749. Schwartz, B. L., Travis, D. M., Castro, A. M., and Smith, S. M. 2000. “The phenomenology of real and illusory tip-of-the-tongue states”. Memory and Cognition, 28: 18–27. Seger, C. A. 1994. “Implicit learning”.  Psychological Bulletin,  115: 163–196. https://doi. org/10.1037/0033-2909.115.2.163 Smith, S. M., Gerkens, D. R., and Angello, G. 2015. “Alternating incubation effects in the generation of category exemplars”. Journal of Creative Behavior, 51: 95–106. Tulving, E. 1989. “Memory: performance, knowledge, and experience”. European Journal of Cognitive Psychology, 1: 3–26. https://doi.org/10.1080/09541448908403069 Tulving, E., and Pearlstone, Z. 1966. “Availability versus accessibility of information in memory for words”.  Journal of Verbal Learning and Verbal Behavior, 5: 381–391. https://doi.org/10.1016/ S0022-5371(66)80048-8

372

Memory During Failures of Recall Urquhart, J. A., Sivakumaran, M. H., Macfarlane, J. A., and O’Connor, A. R. 2021. “fMRI evidence supporting the role of memory conflict in the déjà vu experience”. Memory, 29: 921–932. https://doi. org/10.1080/09658211.2018.1524496. Voss, J. L., Lucas, H. D., and Paller, K. A. 2012. “More than a feeling: pervasive influences of memory without awareness of retrieval”. Cognitive Neuroscience, 3: 193–226. Voss, J. L., and Paller, K. A. 2009. “An electrophysiological signature of unconscious recognition memory”. Nature Neuroscience, 12: 349–355. Whittlesea, B. W. A. 2002. “False memory and the discrepancy – attribution hypothesis: the prototypefamiliarity illusion”. Journal of Experimental Psychology: General, 131: 96–115. Whittlesea, B. W. A., Masson, M. E. J., and Hughes, A. D. 2005. “False memory following rapidly presented lists: the element of surprise”. Psychological Research, 69: 420–430. Whittlesea, B. W. A., and Williams, L. D. 1998. “Why do strangers feel familiar, but friends don’t? A discrepancy-attribution account of feelings of familiarity”. Acta Psychologica, 98: 141–165. Whittlesea, B. W. A., and Williams, L. D. 2001. “The discrepancy – attribution hypothesis. II. Expectation, uncertainty, surprise, and feelings of familiarity”. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27: 14–33.

373

PART 8

Learning and Reasoning

29 IMPLICIT REASONING Thomas Sturm and Uljana Feest

1. Introduction The idea that reasoning can be implicit, or that there are implicit inferences involved in reasoning, is as plausible as it is problematic. On the one hand, there are good reasons to think that not all (if, indeed, even most) of the inferential steps in reasoning are explicit. On the other hand, the idea that there is implicit reasoning might be taken to embody an intellectualist or rationalist myth about how the mind really works that needs to be replaced by better accounts. These conflicting tendencies have a long history at the intersection between philosophy and psychology. In this chapter, we tease apart some philosophical themes touched upon by this literature. These themes concern the ways in which the notions of implicit and reasoning are construed. In our discussion, we rely on three widely held (albeit not universally accepted) minimal assumptions concerning the concept of reasoning: (a) the concept cannot be explicated independently of that of inferences – there is no reasoning without inferences of one sort or another; (b) inferences require premises and conclusions to be represented in a propositional format; (c) the rules connecting premises and conclusions are potentially normative – agents can consciously follow or violate them, and one can apply judgments of rightness of wrongness to inferences.1 It is important not to overstate what these assumptions claim. For instance, we are not committed to any particular normative system, such as particular systems of logic, theories of probability, and the like. More importantly, the question of whether there can be such a thing as implicit reasoning crucially turns on how assumption (c) is interpreted: if we take it to mean that inferential processes can only be reasoning processes if they are carried out consciously and with normative evaluations in mind, there seems to be an unresolvable tension between the notions of implicitness and reasoning. However, we will argue for a weaker interpretation of this assumption, according to which it is only necessary that it is possible for agents to meet the conditions described in (c), such that, when asked, agents can make explicit what their reasons for a certain belief or decision are, how they derive the belief from the reasons, and why they view the rules governing their inferences as correct. Using this modal reading of assumption (c), we will argue that reasoning can be implicit and, thus, that the seeming tension can be resolved. In the following, we will develop our argument by describing and evaluating some prominent accounts of the ways in which the notion of implicitness as applied to reasoning has been analyzed in the history of philosophy and psychology. In section 2, we begin by pointing to DOI: 10.4324/9781003014584-38 377

Thomas Sturm and Uljana Feest

prominent sources (roughly from Aristotle to Immanuel Kant) for the idea that inferences can be regarded as having implicit elements when they rely on unstated premises and/or when these unstated premises are unconscious. We also discuss historical figures (like David Hume and John Stuart Mill) who were critical of the idea of reasoning as requiring inferences from propositionally characterized premises to conclusions, and in response proposed associationist models of reasoning. Section 3 introduces Hermann von Helmholtz’s influential notion of “unconscious inference”, a position that tries to integrate aspects of both of these historical alternatives. In section 4, we further unpack the notion of implicit to show that it has two distinct ideas attached to it, that is, that of being unconscious and that of being automatic. While section 4 focuses on the issue of how the mind processes contents (automatically vs. deliberatively), section 5 considers the question of what is the format in which the content itself is represented (as associations or as propositions). We show that the automaticity of some cognitive processes does not preclude them from being reasoning processes unless there is a deep categorical or insuperable chasm between implicit and explicit processes. Section  6 considers this question more directly by addressing prominent “dual-systems” accounts – especially that of Daniel ­Kahneman – which have posited a strict distinction between types of processes. We will discuss the questions of whether the implicit processes posited by dual-system accounts can be regarded as reasoning processes at all, and whether the implicit nature of some cognitive processes prevents them from being responsive to norms of rationality. As we will argue, some implicit processes can indeed be instances of reasoning: because they can be made explicit and then also be evaluated critically just like standard, conscious or deliberate instances of reasoning.

2.  An Overly Abbreviated Prehistory of ‘Implicit Reasoning’ Theorists in the Aristotelian tradition noted the existence of so-called enthymematic arguments: arguments where certain premises are, for various reasons, missing (Rapp 2010). For instance, “Beyonce is mortal because she is a human being” leaves out the premise that all human beings are mortal. “He cannot be trusted since he is a politician” leaves out the premise that no politician can be trusted. “If it doesn’t fit, you must acquit”, stated by the lawyer Johnnie Cochrane in his notorious defense of O.J. Simpson against the charge of murder along with claiming that a certain glove did not fit Simpson’s hand, leaves out – among others – premises such as that to have murdered his wife, Simpson must have worn that glove or that it must have fit his hand (which are controversial premises). In some cases, premises are left out because one takes the audience to know the relevant premises and to be able to supplant them for themselves in thought. In other cases, one views, or tries to convince others, that a missing premise is so plausible or probable that it need not be mentioned. This enthymematic conception fits a standard dictionary meaning of “implicit”: implied but not clearly enunciated. It applies primarily to external, public reasoning or argumentation. Since Leibniz, the idea that not all our mental representations are conscious has become widely accepted. Likewise, Kant emphasized that, indeed, most representations are held unconsciously, or as he put it, “only a few points on the vast map of our mind are illuminated” (Kant 1798: 247). When considering the act of reasoning, this can be seen as extending the aforementioned meaning of “implicit” to internal (subvocal) reasoning, where one often is not conscious of all premises necessary to support one’s preferred conclusion. Since inferences are concatenations of a kind of cognitive representation – the premises and conclusions that are the building blocks of inferences – these inferences often contain unconscious elements and are therefore implicit in a second sense. This is one basic meaning of “implicit” in current psychological literature that studies implicit processes of thinking or cognition, implicit biases, and so on, often 378

Implicit Reasoning

in connection to studies of reasoning (cf. e.g., Macchi et al. 2016). In addition, the concept of implicitness in this current context becomes wedded to other aspects, such as automaticity, lack of deliberation, or deficiencies in rationality. While this historical background makes the idea of implicit reasoning plausible, the conception developed therein is nevertheless problematic. In the tradition from Aristotle to Leibniz to Kant and beyond, the assumption has been that since reasoning requires inference, implicit internal reasoning must be composed of the same elements and have basically the same structure as the inference that occurs in explicit reasoning: first, it must connect different judgments (as premises and conclusions) and, secondly, do this by applying rules, for instance, of logic or probability theory. But are there good reasons to insist that internal reasoning shares this structure, or are these assumptions antiquated holdovers from overly intellectualist or rationalist views of the mind? Whether inferential models of internal reasoning are literally correct or, to put it differently, psychologically realistic is a matter with a long history too. Since at least the early modern period, we find two main accounts of the processes underlying judgments (cf. Hatfield 2002). The basic cases considered in these accounts usually concerned perceptual judgments, and the basic question was what processes underlie these judgments. On the one side, there is the rationalist tradition (to which of course also René Descartes belongs), which posits processes of rational inference, this time from sensory input to perceptual output. For instance, to explain size or distance perception, it was assumed that the mind computes them by using innate geometrical rules. On the other side stand authors in the traditions of British empiricism and associationism such as George Berkeley, David Hume, and John Stuart Mill. They analyzed perceptual judgments as deriving not from other judgments, but from non-propositional “impressions” or “sensations”. In opposition to the rationalist tradition, they argued that the connections between sensory inputs and perceptual judgments or beliefs were essentially based on processes of association. Thus, they held that these connections were not using rules of logic or geometry and did not justify or guarantee the truth of the conclusion. Notoriously, Hume argued even that our judgments about the external world cannot be justified by rationalist standards. We can only explain, by the laws of association, such as similarity or spatiotemporal contiguity of sensations, why we humans unavoidably come to form certain beliefs. This model might or might not invoke the assumption of unconscious, implicit mental processes – the associationists just mentioned did not consider this issue clearly.

3.  Unconscious Inferences As seen, one plausible reading of the term “implicit” equates it with “unconscious”. A prominent figure who popularized (though by no means invented; see Hatfield 2002) the notion of “unconscious inferences” (unbewusste Schlüsse) was the German physicist and physiologist Hermann von Helmholtz (1821–1894) (Helmholtz 1867). His account was intended to explain perception, not reasoning, but it may still serve as a valuable starting point to identify some questions about psychological theorizing that posits unconscious inferences/processes. In line with the associationists, Helmholtz argued that our sense organs register simple sensations from the external world, which function as “signs” for the three-dimensional objects of our perceptual judgments. This raises questions about the processes that need to occur for such perceptual judgments to be formed. Helmholtz answered these questions by positing that judgments are the result of inferences from simple sensations to perceptions. Since we are not conscious of making such inferences, but only of their conclusions (perceptual experiences), he called them “unconscious inferences”. He argued that they operate in accordance with syllogisms with the major premise stating lawful connections between sensations and perceptions. In contrast 379

Thomas Sturm and Uljana Feest

with older associationist theories, Helmholtz’s likened perceptual to scientific judgments, which implied for him that the subject is active in forming such judgments (for details, see Hatfield 1990: 195–208). In turn, he argued that the major premise was established inductively as the result of a learning process (see also Hatfield 2002; Patton 2018). While Helmholtz thought that unconscious inferences are syllogistic, and include lawlike connections in their premises, what distinguished his view from the rationalists was that he claimed that the inferences leading to perception are governed by association and memory. For Helmholtz, the word “inference” (as opposed to mere mechanism) was warranted by his contention that this process involved a psychological activity. However, the question of whether genuine inferences can be drawn unconsciously is crucial here. A critic might claim that if one takes the notion of inference to require that premises are all represented in a propositional format and also that inferences are guided by rules in a controllable manner (where agents can actively follow or violate the rules, and can evaluate their inferences as right or wrong), then the processes described by Helmholtz here cannot be genuinely inferential. Even more questionable would be to slide from talk of unconscious inferences to talk of unconscious reasoning.2 This chapter does not aim to provide an answer to the question of whether Helmholtz’s account of unconscious inference was ultimately successful, or whether it can be regarded as describing processes of reasoning. Leaving aside such interpretive issues we may say, however, that his account raises in an exemplary way two questions, namely (a) what is the underlying machinery that make inferences possible in general, and (b) do unconscious inferences (if there is such a thing) require a specialized process or system, separate from the one used for conscious or explicit inferences? Following Hatfield (2002), we refer to these related issues as the cognitive machinery problem. Helmholtz, like the rationalist and empiricist/associationist accounts before him, was not committed to a specific stance on whether the inferential mental processes posited by him required a specialized system. However, in recent literature on judgment and decision making there are accounts of the inferential machinery that explicitly posit the existence of separate systems along the lines of a conscious/unconscious divide. We find this notion in dualsystem or dual-process approaches (see Evans and Stanovich 2013). Dual-system approaches are of particular interest in the context of this chapter when they are applied to processes of reasoning. The most prominent version of this is currently held by Kahneman (2011) who distinguishes between two systems or types of processes: system 1 operates with heuristics that are fast, automatic, unconscious, and require less effort and resources than the processes distinctive of system 2, which are slow, deliberate, and conscious. We will discuss the dual system hypothesis in section 6, specifically focusing on the issue of whether the system-1 processes posited by this hypothesis can be regarded as processes of reasoning, and – if so – whether they can be regarded as implicit reasoning. The crucial question, we will argue, is whether there is any ground for supposing that the contents operated on in system 1 are non-propositional and that the processes operative are outside the reach of conscious/deliberate evaluation. Before addressing the issue of how dual-system accounts respond to this question, we will disentangle various aspects of the notion of implicitness as used in recent (philosophy of) psychology and philosophy of mind. Those aspects pertain, on the one hand, to the question what is meant by the word “implicit” (section 4) and, on the other hand, on the question to the question of what kinds of entities (e.g., processes, representations) are taken to be implicit (section 5).

4.  Implicit Processes: Automatic Versus Deliberate When we think about the notion of implicitness in contrast to that of explicitness, what comes to mind is not only that implicit processes are unconscious (as opposed to conscious) but also 380

Implicit Reasoning

that they are automatic (as opposed to deliberately controlled). For example, a cognitive process might be set in motion by specific stimuli, resulting in a mental state (e.g., a perceptual judgment) or a behavior (e.g., pulling the trigger of a gun) even though the person in question might know the perceptual judgment to be false and/or disavow the behavior it can give rise to. Cases like these are discussed in the literature on perceptual illusions and (more recently) in some writings about implicit biases. Take, for example, the fact that many police officers in the United States are more likely to use their guns when confronted with a black suspect than when confronted with a white suspect under otherwise identical circumstances. A common analysis of this is that they operate with a stereotype that classifies black men as dangerous, leading them to the judgment of a particular person as dangerous. Now, in such cases both the mentally represented stereotype and the resulting judgment of an individual (or behavior towards the individual) may well be unconscious. In this vein, early work on implicit biases characterized such biases as consciously inaccessible or not consciously accessed (see Greenwald and Banaji 1995). However, some scholars of implicit bias have resisted the conclusion that this is the relevant sense of “implicit”, arguing instead that it is the automaticity of the underlying processes that makes them implicit, regardless of whether we are conscious of those judgments (e.g., Gawronski and Bodenhausen 2006). If so, then we can expect to see divergences between automatically triggered judgments and the judgments we arrive at on deliberative grounds.3 This would not only account for well-known experimental dissociations between the results on explicit and implicit tests but also suggests that even if we make a deliberative effort, our automatic processes cannot always be overridden by our conscious beliefs. If such automatic processes do exist (and are contenders for being classified as implicit), what exactly is their underlying mechanism? Within psychology, the distinction between controlled and automatic processes goes back to Shiffrin and Schneider (1977) and was taken up particularly in social psychology (Bargh et al. 1996). In the literature about implicit (social) cognition, many theorists take automatic processes be due to semantic associations between concepts (Brownstein 2019) and posit that concepts get activated by processes of spreading activation (Collins and Loftus 1975). So, we have some idea about what makes these processes automatic. But, again, can associative/automatic processes be regarded as reasoning processes? Picking up on our assumption (c) in the introduction of this chapter, we suggest that the automaticity of these processes as such does not necessarily preclude a process from being a reasoning process, as long as it is possible for a human subject to consciously explicate and deliberatively evaluate the process in question. However, we have also formulated two other conditions for reasoning, namely that it involves inferences, where the premises and conclusions are represented in a propositional format. With this in mind, we now turn to a second version of the implicit/explicit distinction, according to which there are not only two distinct types of processes but those processes also operate over two distinct kinds of cognitive representations, namely associative vs. propositional ones, respectively (e.g., Strack and Deutsch 2004; Gawronski and Bodenhausen 2006), where the former are “implicitly” represented contents. If such associative contents do indeed exist, and if automatic processes operate over them, then it seems like a stretch to refer to these processes as inferences (and hence, it seems like a stretch to view them as playing a role in reasoning). In the next section, we consider some recent discussions over whether such implicit contents do indeed exist.

5.  Implicit Contents: Associationist Versus Propositional Having covered the issue of the processes (automatic vs. deliberate) operating over cognitive representations, the next question concerns the issue of how the content is represented and whether there are one or two distinctive representational formats. Here, too, we encounter the 381

Thomas Sturm and Uljana Feest

terminology of “associative vs. propositional”. However, in this case, it does not concern types of (automatic or deliberate) modes of processing but rather the question of whether specific beliefs or attitudes are propositional attitudes (i.e., attitudes we have about semantic content that is represented in a propositional format) or whether they are mere behavioral dispositions, which can be triggered by content that is represented in a mere associative format. Take, for example, again, the currently much studied case of racial prejudice. The prejudice that blacks are violent could either be a specific attitude (e.g., a belief) toward the proposition that blacks are violent, or it could be a tendency to associate specific traits (e.g., black, violent). The issue at hand is whether there are separate systems where different types of contents are represented (one associative and one propositional) or whether there is only one kind of representational format, with different kinds of processes operating over them. Within philosophy, a position to the effect of the former possibility has been articulated by Tamar Gendler’s distinction between beliefs and aliefs, where aliefs manifest behaviorally as if they were guided by beliefs, but in fact are not (Gendler 2008a, 2008b). According to her, “to have an alief . . . is . . . to have an innate or habitual propensity to respond to an apparent stimulus in a particular way. It is to be in a mental state that is . . . associative, automatic and arational” (2008b: 557). Gendler thus describes certain kinds of mental states as due to, or even constituted by, the automatic processes mentioned earlier: to have aliefs, according to her, is to be in mental states that associatively trigger specific types of responses. As in the case of empiricist accounts of associated processes, the question is whether such mental states can figure in genuine reasoning processes, and/or whether such associative processes can constitute genuine reasoning processes, especially since they are, on Gendler’s own analysis, arational, that is, not sensitive to evidence or accountable to norms of reasoning. Brownstein and Madva claim that Gendler overstates the extent to which automatic processes are norm-insensitive. Thus, they argue that “aliefs can be norm-sensitive in virtue of their responsiveness to affective states of disequilibrium. Responsiveness to such affective states is flexible, self-modifying, and capable of error” (Brownstein and Madva 2012: 428). Even so, however, it seems like a stretch to refer to such “affective states of disequilibrium” as being open to reasoning (as opposed to, say, merely being adapted to the relevant environment). Genuine reasoning, we have posited earlier, requires rules that connect premises and conclusions, rules that are potentially normative. On the other end of the spectrum, psychologist de Houwer (2014) and philosopher Mandelbaum (2015) posit that all contents, insofar as they are driving behavior or involved in inferential relations are necessarily propositional, and that it is only because of this that they can be sensitive to evidence and figure in inferences. While de Houwer allows for reasoning processes to be automatic/unconscious (as laid out in the previous section), he holds that the underlying content must have been consciously and propositionally entertained during the learning process: “The processes by which the proposition arises into consciousness or the processes by which it is evaluated as true might well be unconscious, but the proposition itself at some point in time needs to be entertained consciously” (de Houwer 2014: 532). This account thereby rejects the notion of two separate systems in which content is represented in two formats. It thus seems better suited to allow for the relevant cognitive contents to figure in genuine reasoning, as defined by us in this chapter. To sum up the point of this and the previous section: If we take it to be a necessary condition for reasoning that it somehow involves deliberate and norm-sensitive inferences over propositional contents, then both the notions of (a) implicit (unconscious/automatic) inferences and (b) non-propositionally acquired and represented contents might easily seem like oxymorons. However, as the discussion of de Houwer and Mandelbaum shows, we can also imagine implicit 382

Implicit Reasoning

(unconscious/automatic) processes that are habitualized inferences, and that operate with premises and conclusions that were at one point consciously and propositionally entertained and that can be consciously explicated and evaluated when needed. In order for this kind of implicit reasoning to be possible, however, there cannot be a strict and categorical distinction between two modes of processing and representation as has been claimed by advocates of dual systems accounts. We therefore now turn to a more detailed analysis of dual-system accounts.

6.  Dual Systems: Against the Chasm Dual-systems theories (often also called dual-process theories) of judgment and decision making became widely accepted since the 2000s (cf. Evans 2003; Evans and Stanovich 2013; Kahneman 2011; for a broader historical perspective, see Frankish and Evans 2009). As the name indicates, such theories distinguish between types of causes of judgments and decisions. There exist many versions of such an account, but they agree on several assumptions. As mentioned earlier (section 3), they typically assert that there are two types of cognitive processes: system-1 processes are said to be unconscious, fast, automatic, intuitive, and guided by heuristics, while system-2 processes are said to be conscious, slow, deliberate, effortful, and guided by norms of inference such as those from logic, probability, or rational choice – the so-called “standard picture of rationality” (Stein 1996). Table 29.1 shows a still more differentiated list of typical correlations between features of cognition and the two systems. A general question is whether these correlations really hold water, both empirically and ­conceptually. Evans and Stanovich (2013) provide a discussion of typical objections. What

Table 29.1 Clusters of Attributes Frequently Associated With Dual-Process and DualSystem Theories of Higher Cognition (Evans and Stanovich (2013: 255)) Type 1 process (intuitive)

Type 2 process (reflective)

Defining features Does not require working memory Autonomous

Requires working memory Cognitive decoupling Mental simulation

Typical correlates Fast High capacity Parallel Nonconscious Biased responses Contextualized Automatic Associative Experience-based decision making Independent of cognitive ability

Slow Capacity limited Serial Conscious Normative responses Abstract Controlled Rule-based Consequential decision making Correlated with cognitive ability

System 1 (old mind)

System 2 (new mind)

Evolved early Similar to animal cognition Implicit knowledge Basic emotions

Evolved late Distinctively human Explicit knowledge Complex emotions

383

Thomas Sturm and Uljana Feest

matters here is that dual-system accounts partly map the distinction between systems directly onto those between the non-rational and the rational, and between the implicit and the explicit: system-1 processes are said to be a rational and implicit, whereas system-2 processes are viewed as rational processes and explicit. Which of the two types of processes is activated depends on the circumstances. For instance, system 1 can inhibit system 2 when we are under time pressure, do not pay attention, and so on. System 2 can be used to correct system-1 intuitive judgments when these go obviously wrong, and motivating subjects by incentives (say, money) to judge and decide more carefully might activate system 2. However, just stating that normative correction by system-2 processes is possible does not really show how that is done, nor does it clarify what the norms are. If system-1 processes in themselves are uncontrolled or automatic, how could they respond to normative interventions of system 2? In line with this worry, most defenders of dual-system accounts do not speak of system-1 processes as reasoning. For example, Kahneman and Tversky, the leading advocates of the heuristics-and-biases (HB) approach claimed that heuristics, by virtue of operating in uncontrolled, non-deliberative ways, cannot be applied in a deliberative fashion and hence their operation cannot constitute a form of reasoning (Kahneman et al. 1982). However, some dual-system theorists describe both types of processes as being instances of reasoning (e.g., Frankish 2016: vii.). This might be interpreted as either due to conceptual carelessness, or to the idea that system-1 processes are instances of reasoning but not rational in a certain sense – namely as irrational, because they do not satisfy norms of, for example, the standard picture of rationality. That is, system-1 processes would be bad reasoning, but reasoning still. However, this would lead into trouble, since the notion of bad reasoning implies the presence of invalid or unsound inferences. In turn, this would imply that system-1 processes would be inferential and non-inferential at the same time, automatic and non-automatic, and so on. Fortunately, there is a third option. We can apply the modal language of our condition (c) here to avoid a deep chasm between types of processes of judgment and decision making. Thus, it must be possible for an agent to employ rules consciously and deliberately, and to view the rules as norms which one can critically assess. In fast or intuitive judgments, we do not (typically) reason in such ways, but, in principle, we can always do so. For instance, with training or critical discussion, the processes leading to such judgments can be made explicit and then be subjected to normative evaluation. Of course, that such situations exist is an empirical matter; but it requires little if any investigation to support this. Because of Kahneman’s significant influence in current cognitive and social science, it is worthwhile to consider his stance towards dual-system accounts more closely. Originally, he and Tversky did not subscribe to such theories. They claimed that all our judgments and decisions are equally to be explained by heuristics – understood as simple rules of thumb that can lead to valid judgments and decisions, but often also to mistakes, such as “base rate fallacy”, “overconfidence bias”, or “hindsight bias” (Kahneman et al. 1982; Gilovich et al. 2002). As they wrote, heuristics “produce both valid and invalid judgments” (Kahneman and Tversky 1996: 582). That is to say, no rules of logic or probability or rational choice can ever explain human judgment and decision making; all of our judgments and decisions are to be explained by heuristics. Call this the generality thesis of the HB program. Critics of the HB program mounted considerable evidence against the generality thesis. To begin, the responses of subjects can, after all, be improved. People can be taught to respond in ways that conform to, say, the conjunction rule of probability theory or to base rates, for instance by using more transparent task formats, such as frequentist formats. Their behavior is then “substantively” rational, that is, satisfies a normative rule of logic of probability (e.g., Rips and Marcus 1977; Gigerenzer 1991). Does that already show that subjects implicitly use 384

Implicit Reasoning

the relevant formal rules, and that the cognitive processes causing their judgments are therefore implicit inferences guided by the standard model of rationality? Not necessarily; perhaps subjects come to the right results but not in the right way. People can be lucky. Still, one can train people to understand and use relevant rules, so that their reasoning becomes guided by them. In such cases, subjects’ inferences are not only in conformity with, but actually use rules of the standard picture (e.g. Gigerenzer and Hoffrage 1995). Then, their behavior can then be said to be both substantively and procedurally rational (Hooker and Streumer 2004). Not only their reasoning outputs are correct; their very reasoning processes are. For a while, such reasoning must be performed at a conscious and controlled level, hence explicitly; but after extended training, what was once explicit can become an unconscious, implicit habit. Evidence that people sometimes carry out genuine reasoning led Kahneman to accept a dual-systems account (Kahneman and Frederick 2002; Kahneman 2011). Only when the generality thesis of the early HB program is given up is the door open for letting norms of good reasoning do their work. (Note: Here, we need not commit ourselves to a specific normative account of rationality; but see Gigerenzer and Sturm 2012; Sturm 2019, 2021.) However, dual-systems accounts also face problems. Most importantly, their strict distinction between what is explicit and inferential/rational, on the one hand, and what is unconscious, automatic, and non-inferential/a rational, on the other hand, is unconvincing. This is important for our purposes since – as we have laid out earlier – the possibility of implicit reasoning hangs on there not being such a strict distinction. Let us therefore highlights three (related) problems of dual-systems accounts. (1) Some studies have shown that human infants and great apes are “intuitive statisticians” (cf. Rakoczy et al. 2014): bonobos or orangutans seem to track statistical information (e.g. regarding the relative frequency distribution in populations) and are thus able to draw inferences from populations to randomly drawn samples. This is hard to reconcile with the dual-systems approach since – on the one hand – it seems unlikely that infants and primates get the right results by means of conscious and deliberate inferences, while – on the other hand – these findings defy the expectation that system-1 processes typically violate standard norms of rationality. (2) Advocates of dual-systems accounts use central concepts, such as that of intuition, inconsistently in a revealing way. Mostly they use it to describe “snap” judgments: the fast responses of subjects in cognitive tasks, i.e., the processes that take place in system 1. However, Kahneman and Tversky also sometimes speak of an intuition of axioms of mathematics, probability, or rational choice theory (Tversky and Kahneman 1983: 344, 1986: S251f.), i.e., rules that can be known and used in the conscious and deliberate fashion associated with system 2. Thus, whereas in dual-systems theories “intuition” is often reserved for “a rational” and implicit judgment and decision making (or its outcomes), i.e., judgment and decision making that does not make use of norms of, e.g., the standard account of rationality, “intuition” is also used in a different meaning, where it is a legitimate and fully conscious, hence explicit basis of rational knowledge (Sturm 2014). (3) A general point has also been critically discussed in the literature: Why assume that heuristics necessarily operate in a non-deliberative, uncontrolled, hence implicit way? Consider a view of heuristics like Gigerenzer’s “bounded rationality”, according to whom heuristics are not necessarily lazy and error-producing shortcuts but can sometimes perform as well as, or even better than, more costly optimization strategies (Gigerenzer et al. 1999; Gigerenzer and Sturm 2012). Gigerenzer and his collaborators have emphasized that – just as we can learn to use rules of logic or probability (up to the point where their use can become 385

Thomas Sturm and Uljana Feest

unconscious or even quasi-automatic) – we can learn that and when certain heuristics are computationally less costly and safe enough to use. Then, we can and should use them consciously, deliberately, and thus explicitly (Kruglanski and Gigerenzer 2011). These arguments speak against the tenability of a strict distinction between separate systems and, at least in part, for our view that genuine reasoning can be implicit if it is in principle possible to make the inferential processes – whether guided by rules of the standard picture of rationality, or by heuristics – explicit and submit them to critical, normative assessment. Neither is implicitness per se a challenge to the normativity of reasoning, nor should one accept that the frequently unconscious or quasi-automatic operation of heuristics per se prohibits them from being used normatively, given appropriate considerations and circumstances. To accommodate these ideas and develop them further, studies of implicit reasoning should go beyond many traditional assumptions, be they coming from the rationalist, the empiricist/associationist, or the dual-systems approaches.

7. Conclusion We have provided a historically informed, philosophical analysis of the notions of implicit inference and reasoning. Central to our analysis were the assumptions that reasoning requires inferences, inferences require premises and conclusions to be represented in a propositional format, and that it must be possible that the rules connecting premises and conclusions be guided consciously and controlled by some normative standards. All too often, those who claim that many of the processes that bring about judgments and decisions are implicit have denied that such processes can be rational, or instances of genuine reasoning. This is certainly directed against excessive uses of rationalist models of unconscious inferences. However, this opposition is at least sometimes grounded in unclear or overly demanding notions of what reasoning or rationality are. First, for a process of thought to be reasoning, it need not always be explicit: logical and probabilistic inferences can be or become unconscious and automatic. Second, judgments and decisions can be explicitly – consciously and deliberatively – guided by heuristics. The border between what is implicit and what is explicit in reasoning is highly permeable, and should be so, given what learning and critical thinking require. This defense of rationalist models of judgment and decision making does not imply that non-rationalist models, such as associationist, never provide correct empirical explanations of judgments and decisions. But the normativity of reasoning and rationality implies that rationalist models might sometimes be providing correct explanations too.4

Related Topics Chapters 1, 2, 3, 7, 11, 16, 30, 31

Notes 1 It should be noted that we do not maintain that reasoning is nothing but the drawing of inferences according to (logical, probabilistic, or other formal) rules. For instance, reasoning also involves considerations of assessing the quality and applicability of such rules and, moreover, of relevance (see e.g., Grice 2001). Nonetheless, we presuppose that attempts to explicate the concept of reasoning independently of that of inference (e.g., Mercier and Sperber 2017) are not convincing. 2 Note that this question is not specific to Helmholtz but is raised by empiricist accounts in general. However, the issue is brought to the fore in Helmholtz’s work because he speaks so vividly of “unconscious inferences”.

386

Implicit Reasoning 3 See Feest (2020) for an overview of the various accounts of “implicit” in the literature about implicit bias. 4 We wish to thank Robert Thompson and two anonymous referees for valuable criticism and suggestions. Both authors contributed equally to this chapter.

References Bargh, J., Chen, M., and Burrows, L. 1996. “The automaticity of social behavior”. Journal of Personality and Social Psychology, 71: 230–244. Brownstein, M. 2019. “Implicit bias”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Fall 2019 edn. https://plato.stanford.edu/archives/fall2019/entries/implicit-bias/. Brownstein, M., and Madva, A. 2012. “The normativity of automaticity”. Mind and Language, 27: 410–434. Brownstein, M., Madva, A., and Gawronski, B. 2019. “What do implicit measures measure?”. Wiley Interdisciplinary Reviews: Cognitive Science, 10: e1501. Collins, A. M., and Loftus, E. F. 1975. “A spreading-activation theory of semantic processing”. Psychological Review, 82, 407–428. De Houwer, J. 2014. “Why a propositional single-process model of associative learning deserves to be defended”. In J. W. Sherman, B. Gawronski and Y. Trope, eds., Dual processes theories of the social mind. New York: Guilford: 530–541. Evans, J. 2003. “In two minds: dual-process accounts of reasoning”. TRENDS in Cognitive Sciences, 7: 454–459. Evans, J., and Stanovich, K. 2013. “Dual-process theories of higher cognition: advancing the debate”. Perspectives on Psychological Science, 8: 223–241. Feest, U. 2020. “Construct validity in psychological tests – the case of implicit social cognition”. European Journal for the Philosophy of Science, 10: 4. https://doi.org/10.1007/s13194-019-0270-8 Frankish, K. 2016. “Foreword”. In L. Macchi, M. Bagassi and R. Viale, eds., Cognitive unconscious and human rationality. Cambridge, MA: MIT Press: vii–x. Frankish, K., and Evans, J. 2009. “The duality of mind: An historical perspective”. In J. Evans and K. Frankish, eds., In two minds: dual processes and beyond. Oxford: Oxford University Press: 2–29. Gawronski, B., and Bodenhausen, G. 2006. “Associative and propositional processes in evaluation: an integrative review of implicit and explicit attitude change”. Psychological Bulletin, 132: 692–731. Gendler, T. S. 2008a. “Alief and belief ”. The Journal of Philosophy, 105: 634–663. Gendler, T. S. 2008b. “Alief in action (and reaction)”. Mind and Language, 23: 552–585. Gigerenzer, G. 1991. “How to make cognitive illusions disappear: beyond heuristics and biases”. European Review of Social Psychology, 2: 83–115. Gigerenzer, G., and Hoffrage, U. 1995. “How to improve Bayesian reasoning without instruction: frequency formats”. Psychological Review, 102: 684–704. Gigerenzer, G., and Sturm, T. 2012. “How (far) can rationality be naturalized?”. Synthese, 187: 243–268. Gigerenzer, G., Todd, P. M., and the ABC Research Group. 1999. Simple heuristics that make us smart. New York: Oxford University Press. Gilovich, T., Griffin, D., and Kahneman, D., eds. 2002. Heuristics and biases: the psychology of intuitive judgment. New York: Cambridge University Press. Greenwald, A. G., and Banaji, M. R. 1995. “Implicit social cognition: attitudes, self-esteem, and stereotypes”. Psychological Review, 102: 4–27. Grice, P. 2001. Aspects of reason. Oxford: Oxford University Press. Hatfield, G. 1990. The natural and the normative: theories of spatial perception from Kant to Helmholtz. Cambridge, MA: MIT Press. Hatfield, G. 2002. “Perception as unconscious inference”. In D. Heyer and R. Mausfeld, eds., Perception and the physical world: psychological and philosophical issues in perception. New York: John Wiley and Sons: 115–143. Helmholtz, H. v. 1867. Handbuch der physiologischen Optik. Leipzig: L. Voss. Hooker, B., and Streumer, B. 2004. “Procedural and substantive practical rationality”. In A. Mele and P. Rawlings, eds., The Oxford handbook of rationality. Oxford: Oxford University Press: 57–74. Kahneman, D. 2011. Thinking, fast and slow. London: Penguin. Kahneman, D., and Frederick, S. 2002. “Representativeness revisited: attribute substitution in intuitive judgment”. In T. Gilovich, D. Griffin, and D. Kahneman, eds., Heuristics and biases: the psychology of intuitive judgment. New York: Cambridge University Press: 49–83.

387

Thomas Sturm and Uljana Feest Kahneman, D., Slovic, P., and Tversky, A., eds. 1982. Judgment under uncertainty: heuristics and biases. New York: Cambridge University Press. Kahneman, D., and Tversky, A. 1996. “On the reality of cognitive illusions”. Psychological Review, 103: 582–591. Kant, I. 1798/2007. Anthropology from a pragmatic point of view. In I. Kant, Anthropology, history, and education. Eds. G. Zöller and R. Louden. Cambridge: Cambridge University Press. Kruglanski, A., and Gigerenzer, G. 2011. “Intuitive and deliberate judgments are based on common principles”. Psychological Review, 118: 97–109. Macchi, L., Bagassi, M., and Viale, R., eds. 2016. Cognitive unconscious and human rationality. Cambridge: MIT Press. Mandelbaum, E. 2015. “Attitude, inference, association: on the propositional structure of implicit bias”. Noûs, 50: 629–658. Mercier, H., and Sperber, D. 2017. The enigma of reason. Cambridge, MA: Harvard University Press. Patton, L. 2018. “Helmholtz’s physiological psychology”. In S. Lapointe, ed., Philosophy of mind in the nineteenth century. London: Routledge: 96–116. Rakoczy, H., Clüver, A., Saucke, L., Stoffregen, N., Gräbner, A., Migura, J., and Call, J. 2014. “Apes are intuitive statisticians”. Cognition, 131: 60–68. Rapp, C. 2010. “Aristotle’s rhetoric”. In E. Zalta, ed., The Stanford encyclopedia of philosophy. Spring 2010 edn. https://plato.stanford.edu/archives/spr2010/entries/aristotle-rhetoric/. Rips, L. J., and Marcus, S. L. 1977. “Supposition and the analysis of conditional sentences”. In M. A. Just and P. A. Carpenter, eds., Cognitive processes in comprehension. Hillsdale, NJ: Erlbaum: 185–220. Shiffrin, R., and Schneider, W. 1977. “Controlled and automatic human information processing: perceptual learning, automatic attending, and a general theory”. Psychological Review, 84: 127–190. Stein, E. 1996. Without good reason: the rationality debate in philosophy and cognitive science. Oxford: Oxford University Press. Strack, F., and Deutsch, R. 2004. “Reflective and impulsive determinants of social behavior”. Personality and Social Psychology Review, 8: 220–247. Sturm, T. 2014. “Intuition in Kahneman and Tversky’s psychology of rationality”. In L. Osbeck and B. Held, eds., Rational intuition: philosophical roots, scientific investigations. Cambridge: Cambridge University Press: 257–286. Sturm, T. 2019. “Formal versus bounded norms in the psychology of rationality: toward a multilevel analysis of their relationship”. Philosophy of the Social Sciences, 49: 190–209. Sturm, T. 2021. “Towards a critical naturalism about bounded rationality”. In R. Viale, ed., Routledge handbook of bounded rationality. London: Routledge: 73–89. Tversky, A., and Kahneman, D. 1983. “Extensional versus intuitive reasoning: conjunction fallacy in probability judgment”. Psychological Review, 90: 293–315. Tversky, A., and Kahneman, D. 1986. “Rational choice and the framing of decisions”. Journal of Business, 59: S251–S278.

388

30 IMPLICIT KNOWLEDGE OF (PARTS OF) LOGIC, AND HOW TO MAKE IT EXPLICIT Keith Stenning and Michiel van Lambalgen

1. Introduction The processing of narrative requires the hearer to take in an utterance of the current discourse, retrieve knowledge relevant to its interpretation in the context developed so far, and compute the resulting ‘discourse model’ of the updated story. The products of narrative are among the most conspicuous monuments to human reasoning. But this reasoning process itself is usually an implicit process: the results of the reasoning become explicit, but the processes remain inaccessible, even to the reasoning agent. So much so, that they have not generally been accepted by psychology as reasoning: they are typically dismissed as ‘mere psycholinguistics’. Reasoning is supposed to be deliberative, by definition. The current chapter proposes a change in this landscape that takes the implicit reasoning processes of narrative interpretation as paradigmatic cases of reasoning, starting out from a ‘logic of narrative’. Our aims are threefold: first to show that processing of narrative can be viewed as the application of a (‘non-monotonic’) logic; and to give a computational explanation how and why the process proposed can be implicit and not available to consciousness. Secondly, to consider what this means for Dual Process Theory – the part of the psychology of reasoning that distinguishes between implicit processes (System 1) and deliberative processes (System 2) (sometimes referred to as Type 1 and Type 2, respectively). Thirdly, we present novel data on illiterate subjects’ narrative processing, showing that from the perspective of a logic of narrative, illiterate reasoning is much closer to that of the educated than has been conventionally assumed.

2.  Logic and Discourse Reasoning, explicit as well as implicit, is inextricably linked to the capacity to understand natural language discourse, which can be defined roughly as a linked set of utterances, where the links can be given by grammatical constructions such as tense, aspect, and anaphoric relations but also by so-called rhetorical relations,1 which may occur overtly or covertly. Successful discourse comprehension results in a discourse representation structure (Kamp and Reyle 1993), or as it is called in psycholinguistics, a situation model (Zwaan and Radvansky 1998). Understanding discourse aims at achieving coherence of the information provided by the discourse, semantic memory and episodic memory, and nonverbal information supplied by perception. Optimal DOI: 10.4324/9781003014584-39 389

Keith Stenning and Michiel van Lambalgen

coherence results if there is a single situation model in which all the aforementioned pieces of information are jointly represented. One can think of this model as containing a graph that represents verbally encoded dependence relations. In a classic formulation: Cohesion occurs where the interpretation of some element in the discourse is dependent on that of another. The one presupposes the other in the sense that it cannot be effectively decoded except for recourse to it. When this happens, a relation of cohesion is set up, and the two elements, the presupposing and the presupposed, are thereby potentially integrated in the text. (Halliday and Hasan 2014: 4) To achieve coherence – and its representation as a situation model – is non-trivial, and generally requires inference. It may seem strange that the construction of a model requires inference. Ever since Aristotle, inference is supposed to be a relation between premises and a conclusion, and (restricting attention to classical logic for the moment) an inference is valid if any model of the premises is a model of the conclusion. On this view, constructing models is involved only in showing a proposed inference is not valid. An example will help to show how the two senses of inference (inference to a model and inference to a conclusion) are related. Suppose someone utters ‘Max fell’, and follows up with ‘John pushed him’. According to the first sense of inference, the task is to construct a situation model representing the order of events. No explicit relation between the clauses is provided, and the order of events has to be inferred. This can be achieved by searching semantic memory with the query ?precedes(e,d), whose intended interpretation is ‘for which events e,d ranging over {push, fall} do we have that e precedes d?’ Suppose this search returns the general principle causes(x,y) → precedes(x,y). The initial query then reduces to the query ?causes(e,d). Searching memory starting from the query just obtained may lead to the retrieval of the causal connection pushing causes falling. We then set d := push, e := fall, and we obtain by substitution (the technical term is ‘unification’) that the event ‘John pushed Max’ precedes ‘Max fell’. The situation model is now determined completely. This type of inference, involving backtracking from a query (which we shall often refer to as the ‘goal’) is a basic tool of ‘logic programming’.2 To see that this is a familiar inference, we may reverse the sequence of steps to obtain pushing causes falling 390

Implicit Knowledge of Logic, and How to Make It Explicit

and causes(x,y) → precedes(x,y) (the variables x,y are assumed to be universally quantified) imply precedes(push,fall). The inference used is thus an instance of the familiar pattern (usually called ‘universal instantiation’) all A are B c is A ⇒ c is B The difference between the two notions of inference (to a model vs. to a conclusion) can now be stated provisionally as follows. Classical logical inference takes premises as given. By contrast, inference to a situation model for a discourse involves positing a goal, and a systematic backtracking search for suitable premises and unifications which make the goal satisfiable.3 However, it is very well possible that an individual is competent in inference to a situation model, while the same inference presented in explicit verbal form turns out to be surprisingly difficult. The experiment on reasoning in illiterates reported in section 4 will provide a case in point. Earlier we mentioned in passing that a logic of narrative must be ‘non-monotonic’, which means that some inferences may have to be withdrawn when the discourse continues, and what was perceived as a full stop turns out to be merely a pause. ‘Max fell. John pushed him, or rather what was left of him, off the bridge into the river.’ is a grisly example: the order of events is now congruent to the order of utterances. Moving from toy examples to real-life discourse, van Lambalgen and Hamm (2004) presents a theory of non-monotonic processing of tense and aspect, based on the same principles. To construct a discourse model in which the events corresponding to tensed verbs are related as indicated by the discourse (‘indicated’ because one typically needs to go beyond the discourse to grasp these relations): (i) one formulates a goal (informally, ‘what are the temporal relations between the events?’), and (ii) backtracking from the goal guides the retrieval of relevant material from semantic memory. Implicit reasoning here is the searching of semantic memory in the service of interpretation. The evidence for this theory is indirect but suggestive. It starts with the observation that there is a privileged relationship between propositional logic programming and neural networks: recurrent neural networks can simulate computations with finite propositional logic programs (compare Hoelldobler and Kalinke (1994) and Stenning and van Lambalgen (2008: Ch. 8)). The reasoning used in the ‘Max fell . . .. ’ example is prima facie not propositional, but a mathematical trick gets us out of the woods: discourse models are typically concerned with relations in time, and it so happens that the customary mathematical representations of time enjoy so-called quantifier elimination (Hodges 1993: Ch. 8), which entails that the only relevant logical operations are the propositional operations negation, conjunction and disjunction, operating on a finite set of atomic sentences.4 The caveat in the endnote notwithstanding, the suggestive link between the inference mechanism of logic programming and neural networks has led us to predict ERP correlates of various forms of defeating a default expectation generated by a situation model; these predictions were subsequently verified, see Baggio et al. (2010); 391

Keith Stenning and Michiel van Lambalgen

Baggio and van Lambalgen (2007); Baggio et al. (2008, 2012). The same formalism together with an assumption on the neural architecture of people with autism spectrum disorder, namely impaired inhibitory neural processes, yielded predictions on discourse processing in people with ASD; for the results see Pijnacker et al. (2009); Stenning and van Lambalgen (2019). These successful predictions lend some credibility to the idea that the formalism is a high-level description of some implicit representations and processes – which shows that implicitness can go hand in hand with intricate combinatorial structure. Lest the reader thinks that making a reasoning pattern explicit is tantamount to exhibiting it in verbal form, we provide an example of an intermediate form of implicitness. Suppose the discourse ‘Max fell. John pushed him’, is uttered with the intention to deceive, for example, to falsely blame John for Max’ demise. Imagine the real sequence of events was ‘John pushed Max against the wall. Tony hit him. Max fell’, and during interrogation Tony tries to exculpate himself by uttering ‘Max fell. John pushed him’. This type of manipulation requires an awareness of the inferences that the interrogator is likely to make, based on the information available to her, and the (possibly erroneous) beliefs that will result from these. Importantly, while discourse processing is for the most part implicit in the sense that the comprehender is not aware of the inferences that result in a situation model, attributing the conclusion of an inference to someone else exhibits a more explicit mastery of inference, even though it may not yet rise to the level of full explicitness. This view is in line with developmental studies such as Sodian and Wimmer (1987), which introduced the following experimental paradigm. Imagine that you are presented with a container filled with red balls. A ball is then removed from this container and placed in an opaque bag without your seeing which ball was transferred. Despite the absence of perceptual input, you readily infer (using universal instantiation) that the ball in the bag is red. Moreover, because you are aware that this conclusion can be inferred, you recognise that another person who also did not see the transfer will make the same inference you did and will know the colour of the ball in the bag without having seen it. The result of the experiment was that four-year-olds said the other person did not know the colour of the ball, as opposed to six-year-olds, who said the other person did know. The experimenters concluded from these data that the younger children did not attribute the capacity for acquiring knowledge through inference to the other person because they lack an explicit representation of this inference – even though they were able to apply the inference. For expository purposes we have so far kept inference to a situation model separate from inference as it occurs in reasoning tasks presented in written form. But this skates over an obvious difficulty: the reasoning task in written form needs to be comprehended by the reasoner just like any other discourse – which means one must associate a situation model to it, preferably unique. By contrast, typical verbal logical reasoning tasks involve argument patterns whose validity depends not on a single model, but on a set of models: an argument is valid if the conclusion is true on every model of the premises, not valid if there is model of the premises that makes the conclusion false. Comprehending a reasoning task does not involve quantification over models, but puts a premium on models which make premises and conclusion true, whereas checking validity puts a premium on a search for counterexamples. This means that there are (at least) two logics at play when a reasoner engages with a reasoning task: logic for reasoning to an interpretation, and logic for reasoning from the interpretation produced by the first process. We have explored the consequences of this observation in Stenning and van Lambalgen (2008: Chs. 7, 9), which deal with Byrne’s suppression task (Byrne 1989).

392

Implicit Knowledge of Logic, and How to Make It Explicit

3.  How Does a Logic of Narrative Help to Understand Dual Processing? The short answer is ‘by providing a positive account of what System 1 does’: it supports narrative processing. Tversky and Kahneman (1983), working in decision theory, gave System 1 processing a heuristic role. Our proposal shares the characteristic of non-monotonicity with heuristics, but goes beyond this and proposes that the purpose of System 1 is to subserve reasoning to interpretations. In cases where subjects do not produce the expected answers in a reasoning task, they may not be trying and failing to turn System 1 off – they may not have set System 2 deduction as their goal in the first place. This new perspective turns erroneous ‘belief bias’ into ‘recruitment of relevant general knowledge’ – the lifeblood of narrative processing. Adding premises recruited from memory to the given premises is a forbidden move in classical logical deduction because classical logic has a contrasting goal: unassailable proof. More recently there have been interesting developments in Dual Process theories, lessening the gap between System 1 and System 2 (the collection De Neys (2017) is a good starting point). De Neys and others propose that, when called on to do System 2 classical logic, we do have ‘logical intuitions’: fast automatic effortless System 1 foreshadowings of aspects of what the deliberative System 2 might later deliver. The contrast between the following syllogisms taken from de Neys (De Neys 2012; De Neys et al. 2010) illustrate the notion of logical intuition: (1) (conflict condition) All vehicles have wheels. Boats are vehicles. So boats have wheels. (2) (control condition) All vehicles have wheels. Bikes are vehicles. So bikes have wheels. The standard view in Dual Process Theory analyses these as a valid syllogisms in classical logic (which they are), and views the interference of the second premise of the first syllogism with subjects’ responses as System 1 intruding upon System 2 reasoning (‘belief bias’). De Neys’ innovation here is to show that intuitions of the validity of the syllogism may co-exist with the subjects’ conflicting belief-bias responses: [O]ne basic procedure has been to simply look at peoples’ response latencies. A number of studies reported that people need typically more time to solve the conflict than the control versions. Now, clearly, the only difference between the two versions is whether the cued heuristic response is consistent with the traditional normative principles or not. If people were mere heuristic thinkers that did not take these normative considerations into account, they should not process the two types of problems any differently. Hence, the latency findings support the idea that people are sensitive to the traditional normative status of their judgment. (De Neys 2012: 30) In other words, the observed slowing down in the conflict condition compared to the control condition must be explained by the assumption that the subject had System 1 fast classical logical intuitions of validity which conflicted with their belief biased System 1 judgement of nonvalidity. This finding led to his proposal that such fast intuitions about classical logic are present in a parallel processing stream from the onset of processing in System 1.

393

Keith Stenning and Michiel van Lambalgen

Now let us consider the crux of de Neys’ argument more carefully: “the only difference between the two versions is whether the cued heuristic response is consistent with the traditional normative principles or not”. This sentence raises two important issues: (i) whether ‘traditional normative principles’ indeed entail that the first syllogism is valid tout court, and (ii) whether there are no other factors slowing down response on the first syllogism. As to (i), if the universal premises are read as defaults (say, ‘all typical vehicles have wheels’) then one might well take boats to be atypical vehicles, and the conclusion does not follow. The experimenter has few means at her disposal to force the classical reading of the premises. As regards (ii), if our considerations in section 2 are correct, then the subject’s main task is to establish coherence of the premises by means of the construction of a discourse model. This requires much more cognitive effort for the first syllogism because the first premise needs to be expanded to a conditional which allows exceptions, and boats have to be recognised as exceptions. Subjects are perhaps wholly operating in System 1 and System 2 does not get a look-in. The crucial empirical question is: what is the subject’s goal? Not what her goal should be according to the experimenter: but what her goal actually is. If interpreting narrative is the subject’s understanding of her goal, notice that introducing new information becomes the required importation of relevant general knowledge to achieve integration of an intended model. De Neys’ observations of interferences using brain imaging can then be viewed as directly analogous to Pijnacker’s observations of ERP traces of abnormality conditions quoted earlier (Pijnacker et al. 2009), except that de Neys uses syllogistic materials instead of suppression task materials.

4.  Narrative Reasoning in Illiterates With these concepts of implicit and explicit in hand, we will now revisit the received view of reasoning in illiterates whose canonical source is work by Luria (1976) and Scribner (1997). Luria’s field study seemed to show that illiterate subjects are rather obstinate in their refusal to draw conclusions from premisses supplied by the experimenter, as in this famous example (Luria 1976: 108): E: 

In the Far North, where there is snow, all bears are white. Novaya Zemlya is in the Far North and there is always snow there. What color are the bears there? S:  I don’t know what color the bears are there, I never saw them. ··· E:  But what do you think? S:  Once I saw a bear in a museum, that’s all. E:  But on the basis of what I said, what color do you think the bears are there? S:  Either one-coloured or two-coloured .  .  . [ponders for a long time]. To judge

from the place, they should be white. You say that there is a lot of snow there, but we have never been there!

Here Luria is talking to an illiterate peasant from Kazakhstan. He attributed the peasant’s refusal to answer to an overriding need to answer from personal knowledge only, interfering with the deductive inference. The peasant is the perfect empiricist. A careful analysis of this example and related data shows, however, that the situation is considerably more complicated. For one thing, the subject does draw the inference when he says ‘To judge from the place, they should be white’ – but he refuses to consider this an answer to the question posed by the experimenter. One cannot blame him. Why should someone be 394

Implicit Knowledge of Logic, and How to Make It Explicit

interested in truth relative to assumptions whose source is the experimenter? If the subject has no personal knowledge of their truth he cannot vouch for the truth of the conclusion, and furthermore, if the experimenter professes to know the assumptions are true, why would he need to ask the subject to draw the conclusion? Scribner (1997) working in Liberia, reports the same type of empiricist response as obtained by Luria. For instance, the reasoning problem whose classical solution requires universal instantiation and modus tollens ‘All Kpelle men are rice farmers; Mr. Smith5 is not a rice farmer. Is he a Kpelle man?’ led to the following dialogue between subject and experimenter: S:  I don’t know the man in person. I have not laid eyes on the man himself. E:  Just think about the statement. S:  If I know him in person, I can answer that question, but since I do not know

him in person

I cannot answer that question. Scribner’s problem minimises, but does not completely eliminate, the elements of the problem with which the subject cannot have been familiar – it’s just Mr. Smith. But then it is not clear that the subject fails to understand the inference pattern. If the interpretation of ‘all’ in the subject’s situation model happens to be ‘all typical’, then the subject needs to know whether Mr. Smith is typical as a matter of logic, not because the subject is bound to what he has experienced. It seemed to us that a new experiment was needed, in which logical ability was gauged using not classical logical inference patterns in standard form, but pieces of discourse which must be shown to be coherent. Our hypothesis was that subjects will have no trouble drawing the relevant inferences in such contexts. In order to show that logical ability is really inherent in discourse comprehension, and not a consequence of, for example, of few years of schooling, we used illiterate subjects. Our hypothesis entailed that there would be few if any refusals to answer that are characteristic of Luria’s data. The experiment took place in the UNHCR-managed refugee camp Kiryandongo (Uganda), which is home to 60,000 refugees, almost all originating from South Sudan.6 There were 63 subjects, all illiterate. The reasoning problems were read to the participants, several times if need be, and the participants were asked to supply a motivation for their answers, which were written down.

4.1  Modus Tollens (MT) The first experimental condition is similar to Scribner’s modus tollens example, except that ‘Mr. Smith’ has been replaced by a common first name: All people who own a house pay house tax. Okello does not pay house tax. Does he own a house? Also, the ‘yes’ answer to the conclusion question is prominent – since almost all refugees in Kiryandongo own a house. This circumstance could invite four kinds of answers and reasonings to justify these: • •

answering ‘I don’t know’ answering ‘no’ using modus tollens 395

Keith Stenning and Michiel van Lambalgen

• •

answering ‘no’ and explaining why Okello doesn’t have a house answering ‘yes’ and furnishing an explanation of why Okello doesn’t pay tax

The first type of answer is the one consistently found by Luria: ‘I don’t know this Okello, so how can I answer the question?’, but in our sample only one subject chose this answer. This is a striking difference with the experiments of Luria and Scribner: subjects did not engage in metacognitive reasoning about why they refused to answer. This invites the question why these responses were so rare. The reasoning experiment was embedded in an economic survey conducted in the camp, with questions relating to economic behaviours such as saving money or starting a business. The subjects answered these questions spontaneously and elaborately, and it may well be that they thought the reasoning questions were an integral part of the economic survey, and hence had to be answered in the same way. Three out of 63 subjects offered option 2, the correct answer (according to classical logic): ‘no’, with a justification referring to the premises only, disregarding the information that almost all inhabitants of the camp live in a house they own.7 • • •

If he does not pay means he doesn’t have a house. Only those [who] have [a] house pay [tax]. There are many Okellos, some have houses, some don’t. In this case, he doesn’t.

About one-third of subjects answered ‘no’ (consistent with modus tollens, although there is no reference to its premises) while realising that a ‘no’ answer is in conflict with the fact that almost everybody in Kiryandongo owns a house. But this means these subjects were at some level aware of the applicability of modus tollens, and in order to salvage consistency of the situation model sought ways to block the inference by introducing exceptions or abnormalities: • • •

Okello does not have a house, he’s at school. He is a drunkard, can’t afford to build a house. He is too old to afford [a house].

The logical form of the general fact about house ownership in Kiryandongo is then something like (H) For all x, if x lives in Kiryandongo and x is not an exception, then x owns a house. These subjects understood that (H) cannot be used in a modus tollens inference about x if x is an exception. The remainder of subjects offered ‘yes’ as an answer, and opted for consistency with their background knowledge. Like the ‘nay’ sayers in category 3, their verbal responses showed that they were aware that the logical form of the conditional premise must be enriched with a variable for exceptions to render modus tollens inapplicable. Examples are • •

He owns a house he sleeps in. Only that [he] may be too poor to pay tax. He is stubborn, he wants his wife to pay [tax].

Summarising, some 35% of subjects answered consistently with modus tollens, and excepting three, all of these subjects blocked the use of (H). The remainder of the subjects blocked the application of modus tollens. In both cases, subjects’ handling of exceptions showed an awareness of their role in creating a coherent situation model: blocking an inference may be as important as drawing the inference. 396

Implicit Knowledge of Logic, and How to Make It Explicit

4.2  Implicit Reasoning: Sketch of a Computation The next task elicited a wide spectrum of answers (though not the response typically obtained by Luria), and provides a window on implicit computations. If Poni wants to see her boyfriend then she goes to Kampala. In Kampala, she takes a bus to Mbarara. Poni wants to see her boyfriend. Does he live in Mbarara? This task is considerably more complex than the first task, since it involves a number of presuppositions that the subject must identify, and hence a spectrum of answers depending on which presuppositions are satisfied. These issues can be made explicit by the type of reasoning we used in the example discourse ‘Max fell. John pushed him’, although it is considerably more involved. Exhibiting the full computation (motivated by the theory of discourse comprehension detailed in van Lambalgen and Hamm (2004)) will be useful in distinguishing between implicit and less implicit subcomputations. For starters, the second premise is problematic for coherence because it is not clear whether the journey from Kampala to Mbarara is part of Poni’s plan to see her boyfriend; that trip could have a different purpose altogether. The data reflect this: 14 out of 63 subjects chose Kampala as the place where the boyfriend lives, as against 16 subjects who opted for Mbarara. Typical answers are • •

She at first went to Kampala, why go to Mbarara. The first destination should be the one. Mbarara is her last destination.

There were five subjects who couldn’t make a choice between Kampala and Mbarara. Four subjects treated this task as a search problem: Poni looks for her boyfriend in Kampala first, and travels to Mbarara if she cannot find him. It must be noted though that these 39 = (14 + 16 + 5 + 4) subjects do not contemplate scenarios in which Poni continues her journey beyond Mbarara, to a place not mentioned in the reasoning task: this is known as the closed world assumption. Some four subjects did not apply the closed-world assumption and answered versions of •

She is just loitering, the man isn’t in any of the two places.

A handful of subjects answered variants of ‘I don’t know’, meaning that they were unable to perform the computation outlined here – but that number is surprisingly low. But the elephant in the room is that, the third premise notwithstanding, it is nowhere explicitly said that by traveling as she does, Poni is fulfilling her intention to see her boyfriend. And indeed, some subjects have other ideas: • •

She is just unfaithful, the man is [in] [Kiryandongo] here, she wants to get a new man in Mbarara. She doesn’t want people to know where he [is]. But he is just here.

That is, Poni calculates that her trip will lead ‘people’ into thinking her boyfriend lives in Mbarara (or Kampala), and the subjects who gave this answer simultaneously represent both Poni’s scheming and the thinking of the deceived ‘people’. This involves an explicit awareness that inference is a source of (possibly erroneous) belief, as discussed in section 2. The question posed to the subjects was only whether the boyfriend lives in Mbarara. However, from the subjects’ answers it was plain they interpreted the task as concerned with the 397

Keith Stenning and Michiel van Lambalgen

resolution of the more general question: ‘where does Poni’s boyfriend live?’ This can be represented formally as what is known as a query ?Loc(A,Y,t), Bf(P,A), where P stands for Poni, A for Poni’s boyfriend, Bf(P,A) means A is Poni’s boyfriend, and for variable Y,t, Loc(A,Y,t) means A is at location Y at time t.8 Visit(P,A,t) means that Poni visits her boyfriend at time t. The result of answering a query should be a situation model which contains at least the location of Poni, the record of her trip, whether she has visited her boyfriend, and if so, where. The computation of the situation model involves retrieving material from memory; for instance (1), which is a general principle about co-location, and (2), which is a general principle about transportation. Retrieval is part of implicit processing, assumed to be available to each subject. The other clauses represent data mentioned in the enunciation of the reasoning problem. We assume that every subject who understands the problem can access these representations. (1) Loc(P,Y,t) ∧ Visit(P,A,t) → Loc(A,Y,t) (‘if Poni is at location Y at time t and visits her boyfriend at that time, then her boyfriend is also at location Y ’) (2) X ≠ Y’ ∧ Loc(P,X,r) ∧ r < s ∧ Travel(r,P,X, Y’,s) → Loc(P, Y’,s) (‘if Poni is at location X at time r, and leaves X at that time to go to Y’ ≠ X, and she arrives there at time s, then she is at Y’ at time s’) (3) Loc(P,C,d0) (‘Poni is at the camp at initial time d0’) (4) Travel(d0,P,C,K,d1) (‘at time d0 Poni departs from the camp to go to Kampala and arrives at time d1’) (5) Travel(d’1,P,K,M,d2) (‘at time d’1, Poni leaves Kampala to go to Mbarara and arrives at time d2’) (6) Travel(d0,P,C,M,d2) (from (4), (5) and the assumption d1 = d’1 by transitivity, which expresses that (4) and (5) are parts of a single journey (not broken up by a stay in Kampala), with one overarching goal, to visit the boyfriend) (7) for some d, Visit(P,A,d) (the visit indeed takes place) Using the technique of logic programming the computation proceeds without supervision until it hits a fork in the road. The first steps go like this. The query is solved by backwards reasoning: given a query one looks for a conditional whose consequent can be matched to the query by a suitable identification of constants with variables (or variables with variables); and analogously for facts. This procedure is called unification and belongs to implicit processing. Starting from the query ?Loc(A,Y,t), Bf(P,A), we apply clause (1) and obtain a new query ?Loc(P,Y,t), Loc(A,Y,t), Bf(P,A), Visit(P,A,t). Applying (2) to this new query, we obtain ?Loc(P,X,r), Loc(A,Y,t), X ≠Y, r < t, Bf(P,A), Visit(P,A,t), Travel(r,P,X,Y,t), Loc(P,Y,t).

398

Implicit Knowledge of Logic, and How to Make It Explicit

Using (3) and unification, we reduce this to ?X ≠ Y, d0 < t, Bf(P,A), Visit(P,A,t), Travel(d0,P,C,Y,t), Loc(P,Y,t), Loc(A,Y,t). So far, so implicit; but from here on the computation branches, which may require explicit choices. If a subject doesn’t have transitivity – that is, (6) fails because (4) and (5) are journeys with different goals – then the subject can use (4) only and we obtain by unification ?Y = K, t = d1, Loc(P,K,d1), Visit(P,A,d1), Loc(A,K,d1), which specifies a situation model, containing a possible answer to the original question: namely, at time d1 Poni visits her boyfriend in Kampala. It will be obvious that there is no mention of Mbarara here; indeed, Poni’s trip from Kampala to Mbarara was hard to integrate coherently for some subjects, because they realised the second leg of the journey could not be derived, i.e. incorporated in the situation model: •

He is in Kampala, why go to Mbarara? She must have escaped from her mum just to loiter.

If a subject really doesn’t have a representation of transitivity, then the entire computation of the situation model containing destination Kampala, is unsupervised, hence implicit. If the subject does have transitivity, using (6) one obtains ?Y = M, t = d2, Loc(P,M,d2), Visit(P,A,d2), Loc(A,M,d2). This second possible situation model enforces that at time d2 Poni visits her boyfriend in Mbarara. Again, this computation is implicit. An interesting case occurs when the subject explicitly entertains both possibilities and combines the two situation models into a single plan. •

She might not have found him in Kampala so she proceeds to Mbarara. (4 subjects)

As we have seen, a sophisticated type of reply attributes to Poni the intention that other people execute the computation just given so that they come to believe that the boyfriend is either in Kampala or Mbarara, while in reality Poni’s trip aims at concealing her boyfriend’s whereabouts. This amounts to these subjects adding epistemic modalities so that Poni can compute her belief about other people’s belief. Space does not allow us to provide the technical details.

5.  Implications for Implicitness Computing coherent situation models following the algorithm outlined in the previous section is largely unsupervised and implicit; but it is nonetheless a very expressive and inference-driven process, a far cry from the associative processes that are typically held to underpin System 1. The main problem in processing narrative is the recruitment of relevant general knowledge (such as preceding clauses (1) and (2)) and we conjecture that the neural networks for logic programs mentioned in section 2 can perform this function as an autonomous implicit process which also executes the very considerable amount of reasoning that is involved. This is what is sometimes referred to as ‘inference by retrieval’ – an active constructive inferential process. The results of that reasoning in working memory are then accessible to consciousness. Linguistic

399

Keith Stenning and Michiel van Lambalgen

communication generally provides our standard of what can be made explicit but actually rides on an apparatus that is implicit and inaccessible but rich in structure, almost entirely hidden, except in its results.

Related Topics Chapters 7, 19, 20, 29

Notes 1 For example: explanation, elaboration, (necessary or sufficient) condition, goal/plan, motivation, intention, inference . . . (Asher and Lascarides 2003). 2 We will see a more substantial use of this technique in section 4.2. 3 Reversing the sequence of steps – adding the premises and unifications retrieved in the search – yields a valid classical inference. Usually, there will be a unique smallest model of the premises returned by the search, smallest in the sense that it only contains entities which are forced to exist by the rules produced by the search. Due to the presence of implicit reasoning steps (for example, universal instantiation), inference to a situation model can in principle be evaluated by classical norms. 4 This is not a solution to the so-called binding problem however; in a nutshell, how to represent the binding pattern in an atomic formula R(a,b)? 5 This is not a Kpelle family name. 6 These data (part of a larger study) were collected (with the help of a translator), transcribed and analysed by Stijn Verberne in the period October–December 2019. The data will be made available on www. researchgate.net/profile/MichielLambalgen/research. 7 Even among literates endorsement rates of modus tollens are low, typically around 70%. 8 The data suggest Bf(P,A) must also be relativised to time, but we won’t go there.

References Asher, N., and Lascarides, A. 2003. Logics of conversation. Cambridge: Cambridge University Press. Baggio, G., Choma, T., van Lambalgen, M., and Hagoort, P. 2010. “Coercion and compositionality: an ERP study”. Journal of Cognitive Neuroscience, 22: 2131–2140. https://doi.org/10.1162/jocn.2009.21303 Baggio, G., and van Lambalgen, M. 2007. “Processing consequences of the imperfective paradox”. Journal of Semantics, 24: 307–330. https://doi.org/10.1093/jos/ffm005 Baggio, G., van Lambalgen, M., and Hagoort, P. 2008. “Computing and recomputing discourse models: An ERP study”. Journal of Memory and Language, 59: 36–53. https://doi.org/10.1016/j.jml.2008.02.005 Baggio, G., van Lambalgen, M., and Hagoort, P. 2012. “Does language processing evidence support compositionality?”. In W. Hinzen, E. Machery, and M. Werning, eds., The Oxford handbook of compositionality. Oxford: Oxford University Press: 657–674. Byrne, R. 1989. “Suppressing valid inferences with conditionals”. Cognition, 31: 61–83. De Neys, W. 2012. “Bias and conflict: a case for logical intuitions”. Perspectives on Psychological Science, 7: 28–38. De Neys, W., ed. 2017. Dual process theory 2.0. London: Routledge. De Neys, W., Moyens, E., and Vansteenwegen, D. 2010. “Feeling we’re biased: autonomic arousal and reasoning conflict”. Cognitive, Affective, and Behavioral Neuroscience, 10: 208–216. Halliday, M. A. K., and Hasan, R. 2014. Cohesion in English. London: Routledge. Hodges, W. 1993. Model theory. Cambridge: Cambridge University Press. Hoelldobler, S., and Kalinke, Y. 1994. “Towards a massively parallel computational model of logic programming”. In Proceedings ECAI94 workshop on combining symbolic and connectionist processing. ECAI: 68–77. Kamp, H., and Reyle, U. 1993. From discourse to logic. Dordrecht: Kluwer. Luria, A. 1976. Cognitive development: its social and cultural foundations. Cambridge, MA: Harvard University Press. Pijnacker, J., Geurts, B., van Lambalgen, M., Buitelaar, J., Kan, C., and Hagoort, P. 2009. “Defeasible reasoning in high-functioning adults with autism”. Neuropsychologia, 47: 644–651. Scribner, S. 1997. Mind and social practice. Cambridge: Cambridge University Press.

400

Implicit Knowledge of Logic, and How to Make It Explicit Sodian, B., and Wimmer, H. 1987. “Children’s understanding of inference as a source of knowledge”. Child Development, 58: 424–433. Stenning, K., and van Lambalgen, M. 2008. Human reasoning and cognitive science. Cambridge, MA: MIT Press. Stenning, K., and van Lambalgen, M. 2019. “Reasoning and discourse coherence in autism spectrum disorder”. In R. Byrne and K. Morsanyi, eds., Thinking, reasoning and decision making in autism. London: Routledge: 135–155. Tversky, A., and Kahneman, D. 1983. “Extensional versus intuitive reasoning: the conjunction fallacy in probability judgment”. Psychological Review, 90: 293–315. van Lambalgen, M., and Hamm, F. 2004. The proper treatment of events. Oxford: Blackwell. Zwaan, R., and Radvansky, G. 1998. “Situation models in language comprehension”. Psychological Bulletin, 123: 162–185.

401

31 WHAT IS IT LIKE TO LEARN IMPLICITLY? Arnaud Destrebecqz

Classical conditioning is certainly one of the best known and most studied psychological phenomena. Any student of psychology learns from the first year on that when two stimuli appear rapidly one after the other, repeatedly and systematically, the presentation of the first tends to modify and generally improve the behavioral response to the second. The general public is well aware of the anecdote of Pavlov’s dog who, after being exposed to numerous bell/food associations, will start salivating upon hearing the bell even before the food is presented to it. It is therefore quite surprising that any doubts remain concerning the exact nature of the mechanisms underlying this form of learning. Yet, in the case of human associative learning, an ongoing debate in the literature concerns the role of consciousness in learning. To develop this point, let’s consider another canine example: we can easily learn that if someone rings the doorbell then the dog will bark. From then on, we will be able to react more and more quickly to the barking of the dog (by telling him to be quiet for example) after hearing the sound of the doorbell. Such a scenario is in fact compatible with two types of interpretation. Either our behavior improves because learning allows us to anticipate the barking of the dog after having heard the bell, or because the systematic repetition of the bell/barking association has led to the strengthening of the associative link between the representations of these two stimuli in the cognitive system in such a way that the hearing of the sound will automatically lead to the activation of the barking representation and, therefore, to the improvement of the behavioral response to this second stimulus. According to the first interpretation, learning amounts to forming a rule-like proposition such as “When the doorbell rings the dog will bark”, so that the conscious expectancy for the barking is initiated by the occurrence of the doorbell. It must be noted that it is a conscious expectancy that can be measured on any subjective rating scale. According to the second interpretation, learning is a mandatory and automatic consequence of the repeated pairing of the two events. In this view, behavioral improvement does not require conscious expectancy for the barking after the bell, but rather results from automatic activation between the associated stimuli. In the case of human learning, learners are generally conscious of the relationships between the stimuli they are presented with. Moreover, it also seems to be the case that participants must pay attention to the relevant stimuli and be aware of their relationships in order to observe a behavioral change during the course of learning. Researchers have therefore claimed that 402

DOI: 10.4324/9781003014584-40

What Is It Like to Learn Implicitly?

human associative learning can be accounted for with a single cognitive process that establishes a propositional representation of the contingencies between the associated stimuli (Lovibond and Shanks 2002; Mitchell et al. 2009). It must be noted, however, that associative learning, of which Pavlovian conditioning is probably the best-known example, is by no means limited to our species It is a ubiquitous phenomenon that has also been demonstrated and studied in-depth in several taxons, even in invertebrate species such as honeybees. It has been shown that, in the case of honeybees, learning occurs because an associative link between the two to-be-associated stimuli (in general an odorous stimulus and a sugar solution) has been established and strengthened at the neural level each time the two stimuli were presented together or in close succession (Giurfa 2007). This phenomenon constitutes a typical example of classical conditioning, in which the bee learns to associate an unconditioned stimulus (the sugar solution) with a conditioned stimulus (the odorous stimulus), so leading to the occurrence of a conditioned response. The magnitude of the conditioned response to the conditioned stimulus is therefore a function of the strength of the associative link between the conditioned and the unconditioned stimuli. Research has therefore disclosed the existence of two distinct learning mechanisms that roughly correspond to the two interpretations given earlier, but there is no indication that these two processes are mutually exclusive. Indeed, we can imagine that the systematic repetition of the successive presentation of the two stimuli leads (1) learners to develop conscious expectancies and, at the same time, (2) to automatically strengthen the associative link between these stimuli. The possibility of unconscious learning, as well as the question of how conscious and unconscious learning processes interact with each other, has been the subject of much fierce debate in the literature. It appears that research in the field is particularly difficult because, on the one hand, no task is process-pure, involving only conscious or unconscious learning processes (Jacoby 1991) and, on the other hand, consciousness cannot be turned-off (Cleeremans 2014). It therefore seems impossible to devise an experimental situation in which learners would not be able to acquire any piece of conscious knowledge about the regularities of the material even if they were more complex than the simple associations I mentioned in this chapter so far. Given these challenges, many methodological innovations have been envisioned to precisely measure what has actually been learned and whether or not the acquired knowledge can be consciously accessed. To further complicate matters, these methods differ in their definition and operationalization of consciousness, and there is no consensus in the literature on what constitutes the best criterion to differentiate between conscious and unconscious learning. In the next section, I describe how researchers have addressed the issue of implicit learning, and try to summarize where we currently stand on this issue.

Implicit Learning: An Attempt to Make a Long Story Short Research about implicit learning, that is, broadly defined, learning that takes place in the absence of intention to learn and in such a way that the acquired knowledge cannot be easily verbalized, was initiated more than fifty years ago with the publication of Arthur Reber’s (1967) article about artificial grammar learning. In an initial experiment, participants were presented with a series of letter strings, such as TPPTXVS. Each sequence was presented for five seconds before the participant was asked to reproduce it on a sheet of paper. This phase continued until the participant could reproduce each string correctly. Even though this task was presented as a memory experiment, for half of the participants (the Experimental Condition), the strings were generated through a finite-state grammar 403

Arnaud Destrebecqz

as in Chomsky and Miller (1958) with the five letters P, S, T, V, X as the vocabulary and a set of rules of sentence construction as the grammar. For the other half of the participants (the Control Condition), a set of random strings was generated from the same five letters. While performance improved in both groups across the task, the numbers of transcription errors decreased more in the Experimental than in the Control condition as participants in the second group needed almost twice as many trials as participants in the first group to achieve the learning criterion. Interestingly, experimental participants, who were not informed about the existence of rules, almost completely lacked verbalizable knowledge about the structure of the material. To find more about the learning process, Reber conducted a second experiment in which, after the training phase, participants were told that the training sequences followed a set of “grammatical rules”, albeit they were given no details about the rules themselves. They were then asked to classify a set of new strings as grammatical or ungrammatical based on what they had learned from the training set. In this latter task, participants reached an average of almost 70% of correct classifications even though they did not report any explicit or verbalizable strategies. According to Reber (1967: 863), they “developed a strong sensitivity to the lawfulness that existed in the stimulus array. The process by which this efficient responding is developed is what is meant by implicit learning.” This (somewhat circular) initial definition of implicit learning subsequently generated immense curiosity among researchers who tried to clarify the nature of implicit learning. The central questions addressed in this field of research concern, on the one hand, the nature of the acquired representations and, on the other hand, the extent to which such knowledge is available to conscious awareness. Most recent implicit learning studies roughly follow the same scenario. First, participants are exposed to a structured material of some kind. Next, discrimination performance between regular (structured) and irregular stimuli is assessed in a second task so as to measure that learning took place (see Figure 31.1, Task 1 and Task 2). Then, learning is further assessed in a direct, forced-choice task, such as a recognition or a generation task in which participants are explicitly instructed to use their knowledge of the training material to perform the task (Task 3). This task is used to measure precisely and as completely as possible what the subjects have learned consciously. This phase is of course crucial in order to make sure that the amount of implicit learning is not overestimated. Finally, an additional subjective test is used (Task 4) in which participants are asked to indicate themselves whether some piece of knowledge can be qualified as conscious knowledge.

Table 31.1 The different tasks generally used in implicit learning studies. For each task (starting from Task 2), the figure indicates the empirical question asked and examples of the tasks used in the literature. Task 1

Task 2

Task 3

Task 4

Exposure/ Learning phase

Discrimination/ Test phase Did participants learn?

Objective measure of conscious learning Is the acquired knowledge above the objective threshold? Direct tasks (recognition, generation . . .)

Subjective measure of conscious learning Is the acquired knowledge above the objective threshold?

Grammaticality judgment task

404

Confidence ratings Feeling of warmth Wagering . . .

What Is It Like to Learn Implicitly?

Such an experimental apparatus is used in order to reach the best assessment as possible of the conscious or unconscious nature of the acquired knowledge. Sensitivity to the lawfulness of the learning material, as Reber (1967) put it, demonstrated by participants in Task 2 may indeed have resulted from different sources of knowledge. Indeed, the performance of the participants in this later discrimination task can be based on a knowledge of the rules of the grammar itself or on more fragmented knowledge, for example on the memorization of fragments of the sequences presented during Task 1 but without having an abstract knowledge of the rules. Moreover, whatever the level of abstraction of the acquired knowledge, the representations of the material developed during learning may be more or less accessible to consciousness. At first, implicit learning was mostly viewed as the acquisition of abstract information about a structured environment (Lewicki 1986; Reber 1989). Such a conclusion was based on the fact that participants were generally unable to articulate the rules of the grammar. Based on the observation of a dissociation between what participants could do and what they could say, it seemed fair to conclude that learning had occurred unconsciously. That interpretation only holds, of course, if participants’ performance is indeed based on the abstraction of the rules of the grammar itself. However, a series of later studies have convincingly shown that the classification of the novel sequences at test could perfectly be accounted for by the explicit memorisation of short sequences, small groups of two or three letters, which, when combined, constitute the longer strings displayed during the learning phase (Perruchet and Pacteau 1990, 1991; Shanks and St. John 1994). As these small letter chunks could successfully be recognized in a subsequent direct recognition phase, there seemed to be no reason to attribute learning to unconscious learning processes rather than to explicit memorization. In general, above chance performance in Task 3 is the norm, rather than the exception. The objective measure of conscious learning is often reliably above zero in such a way that the hypothesis of a powerful implicit learning system was gradually put into question and that the very idea of the unconscious acquisition of novel information was increasingly contested. However, such a radical conclusion is questionable as there is no evidence that performance in an objective measure exclusively reflects the influence of conscious knowledge. As the American philosopher John Searle (1998) pointed out, consciousness is an ontologically subjective phenomenon and objective measures only refer to the ability to choose accurately under forced choice conditions (Dienes and Seth 2010b). It is therefore quite conceivable that participants respond accurately despite the knowledge underlying this performance failing to being fully conscious. They might, for instance, perform above the chance level even though they feel like they are guessing the correct responses. In other words, knowledge would be above an objective threshold, as it can be disclosed in a direct task, but below a subjective threshold as the participant claims that he/she does not know that he/she holds that knowledge (Cheesman and Merikle 1986; Dienes and Berry 1997). As the exclusive use of an objective measure, such as a direct task, does not distinguish between conscious and unconscious knowledge, researchers began using additional subjective measures (Task 4) to distinguish between conscious and unconscious knowledge. Previously, the quest for implicit learning consisted in looking for a dissociation between learning and consciousness measures by using a direct task (Task 3). In this new perspective, it consists in identifying whether some of the acquired knowledge that influenced performance in the direct task, and that is therefore above the objective threshold, nevertheless remains below a subjective threshold. Knowledge is said to be below the objective threshold when discrimination is at chance in Task 2. Hence, the acquired knowledge is under the subjective threshold when a participant does not know that he knows some features of the training material or, in other words, when 405

Arnaud Destrebecqz

he has no metaknowledge. Learning would then be unconscious in the sense that its successful use is not accompanied by the participants’ ability to identify that they possess this knowledge. Different methods have been worked out, such as measuring the feeling of warmth of participants when performing the direct task (Wierzchoń et al. 2012) or to have them wager money on their responses (Dienes and Seth 2010a; Persaud et  al. 2007). One of the most popular techniques, however, consists in asking participants to express their confidence when making a decision in the direct task (Dienes et al. 1995). This can be achieved, for instance, by estimating how confident participants are on a graded scale or by classifying their response as a “guess” or as a “know” response. Dienes and colleagues proposed two approaches to assess whether knowledge was unconscious, that is, above the objective threshold but below the subjective threshold, the guessing and the zero-correlation criteria. The guessing criterion consists in measuring performance when participants claim that they are guessing. If discrimination performance is nevertheless above chance level, then there is an argument to claim that learning was unconscious. The zero-correlation criterion consists in measuring the correlation between performance and confidence. For each trial carried out in Task 3, participants can be asked to indicate how confident they are after each response. Two measures are then collected in close succession. If both measures are correlated, there is no reason to claim that learning was unconscious but if performance is dissociated from confidence, it can then be assumed that knowledge is, at least subjectively, unconscious. Although this methodological innovation is undoubtedly a step forward, in particular insofar as it takes into account the irreducibly subjective and private nature of consciousness, it is nevertheless not entirely satisfactory since no performance pattern indisputably demonstrates the presence of unconscious knowledge. Indeed, if performance is higher than chance while participants claimed to guess, or if there is no link between performance and confidence, it is always possible for the skeptic to argue that this does not indicate the influence of unconscious knowledge but rather the lack of exhaustivity of the direct task or an overly conservative interpretation of the instructions of the subjective measure. Participants could indeed report that they are guessing whereas they actually rely on conscious, but unsettled, fragmentary knowledge held with low confidence. Moreover, confidence ratings could, at least partially, reflect unconscious influences so that the correlation between confidence and performance may not be an exclusive measure of conscious learning (Timmermans and Cleeremans 2015). In other words, the use of forced-choice tasks (instead of mere verbal reports for instance) to measure conscious learning may be at the cost of the so-called exclusiveness assumption (Reingold and Merikle 1988), according to which the test of awareness must be sensitive only to the relevant conscious knowledge. Unfortunately, the most sensitive tests of awareness are also the most likely to be contaminated by implicit knowledge (Neal and Hesketh 1997). In sum, both objective and subjective measures may be differently influenced by both conscious and unconscious influences – a problem known as the contamination problem ( Jacoby et al. 1997; Overgaard 2015). As a consequence, proponents of implicit learning tend to interpret dissociations as indicating unconscious knowledge while skeptics can always attribute the same patterns of results to the fact that the objective and subjective measures are differently sensitive to the influence of conscious knowledge. Although the field has undoubtedly progressed over the last fifty years, particularly with regard to the operational definition of consciousness and the sensitivity of the tasks used to measure it, other problems persist. Rather than discussing all the potential difficulties here, I will highlight two of them. First, questioning people about their own experience is very likely to influence or even modify the processes that we are trying to measure. Consciousness cannot be turned off, so participants will always look for (and probably find at least some) regularities 406

What Is It Like to Learn Implicitly?

when presented with complex material. We can therefore always question the relevance of the knowledge identified after the learning phase because it is impossible to ensure that the former is at the basis of the latter. Second, the contamination problem, coupled with the absence of a consensual measure to distinguish conscious from unconscious influences, does not allow us to easily consider how the field could progress in order to achieve a universally accepted response regarding the possibility of unconscious learning. Research on implicit learning therefore has the characteristics of what Thomas Kuhn describes as a pre-paradigmatic science (Kuhn 1970): a lack of consensus on the interpretation of the available data, and a systematically contradictory approach of opposing schools of thought, one attempting to experimentally demonstrate the existence of implicit learning while the other essentially aims at putting this demonstration into question. It does not mean that no progress can be achieved but it entails that there is no consensus about what would constitute an indisputable proof of unconscious learning around which every scientist in the field could agree on. We therefore find ourselves in the paradoxical situation where some defend the idea that implicit learning is a ubiquitous phenomenon influencing almost every aspect of cognition (Cleeremans et al. 2019) while others consider that “robust and replicable instances of unconscious learning have failed to emerge in the experimental literature, consistent with the view that awareness is a necessary condition for all forms of learning” (Shanks 2010). In support of the former, we observe, for example, that social intuition can be based on implicit processes and guide behavior without explicitly learning the underlying rules (Heerey and Velani 2010; Norman and Price 2012). Numerous studies have also highlighted the ability of 8-month-olds to learn the statistical regularities present in a sequence of meaningless syllables or geometric shapes after only a few minutes of exposure (Bertels et al. 2022; Kirkham et al. 2002; Saffran et al. 1996). Other studies have highlighted the preserved learning abilities of patients or elderly people whose explicit cognitive functions have strongly declined (Bennett et  al. 2007; Witt et  al. 2002). By contrast, in systematic studies of implicit learning in adult participants “the current evidence for the lack of conscious knowledge about the study material is weak at best” (Perruchet 2008). The current state of research in the field of implicit learning suggests that a change of approach is needed to enable the field to move beyond the methodological conundrum described earlier. Pierre Perruchet made such an attempt in a series of studies that take it as a starting point that learning may result from both conscious and unconscious processes. In this original paradigm, rather than trying to induce the unlikely acquisition of exclusively unconscious knowledge, the two processes are pit against each other in order to demonstrate an unconscious influence on learning. This original paradigm deserves attention because it proposes a solution to the methodological and conceptual problems described earlier deemed convincing by both defenders and opponents of the idea of unconscious learning.

The Perruchet Effect The “Perruchet effect” (for a recent and extensive review, see Perruchet 2015) demonstrates a dissociation between participants’ reports and their overt behavior in the simplest form of associative learning. As indicated in the beginning of this chapter, human associative learning may involve two mechanisms: a conscious reasoning process that results in rule-like knowledge and an automatic, potentially unconscious, strengthening between the representations of the to-beassociated stimuli. These two processes are difficult to disentangle because they tend to make similar empirical predictions, that is, that the repeated pairing of the to-be-associated stimuli, and therefore the associative strength, goes with an increase in the expectation of the second 407

Arnaud Destrebecqz

stimulus after the occurrence of the first one. In the first implementation of the Perruchet’s paradigm (Perruchet 2015), an eye-blink conditioning study involving a partial reinforcement schedule, a tone (E1) occurred on each trial but was followed by an air-puff (E2) directed towards the participant’s cornea in only half of the trials. The (pseudo-)randomized sequence of trials comprised series (or runs) of consecutive reinforced trials (E1– E2) and runs of consecutive non-reinforced trials (E1-alone) of various lengths. During the inter-trial interval (ITI), participants had to provide a subjective evaluation of their expectancy concerning the occurrence of the air-puff on the next trial. Results showed that expectancies followed the gambler’s fallacy (GF) (Burns and Corpus 2004): The expectancy for E2 decreased when the length of the preceding reinforced run1 increased, but increased when the length of the preceding nonreinforced run increased. Importantly, the probability of occurrence of the CR (i.e., an eyeblink before the air-puff) increased when the tone and the air-puff were frequently paired in the previous trials. Reciprocally, the probability of occurrence of the CR decreased as the length of the preceding non-reinforced run increased, even though participants reported expectancies for an air-puff increased. These results are important because, in this case, participants’ behavior cannot be explained by conscious expectancy. Indeed, it is precisely when expectations are lowest that the strongest conditioned response is observed. Behavior cannot therefore be attributed to the effect of propositional knowledge but rather to the associative strength between the two stimuli. In their critical review of the role of awareness in Pavlovian conditioning, Lovibond and Shanks (2002: 8) also acknowledge that “Perruchet’s (1985) study provides the strongest evidence to date for a dissociation between eyeblink conditioning and expectancy”. Such a dissociation runs against the idea that consciousness is necessary for the establishment of conditioning and puts therefore into question a purely propositional model of learning in which link formation processes would simply be nonexistent. A similar pattern of results was later observed in cued reaction time (RT) tasks requiring a voluntary motor response, that one could expect to be more related to subjective expectancies (Barrett and Livesey 2010; Destrebecqz et al. 2010; Livesey and Costa 2014; Perruchet et al. 2006). In these studies, a tone (E1) was emitted on each trial, and participants had to quickly react to a visual target (E2) presented after the tone in half of the trials. Participants were also required to provide their expectancy of the target during the ITI. Results showed that, here also, expectancies followed the GF, but that RT decreased with the number of previous tone – target associations. These results therefore also exhibit a dissociation, suggesting that the decrease in RT should not be attributed to a larger expectancy for the target, but rather to the increase of the associative strength between E1 and E2. In a recent study, we further demonstrated that the Perruchet dissociation in the cued RT task cannot be attributed to non-associative factors such as a decrease in vigilance or a mere motor priming (Destrebecqz et al. 2019). We instead observed a small but reliable influence of the strength of the tone-target association on participants’ performance when these potential non-associative confounds were controlled for.2 Performance in this task does not only reflect associative strength, however. In line with the idea that consciousness cannot be turned off, our results also clearly demonstrated a reliable influence of conscious expectancies on performance. This influence was revealed by the comparison between participants who showed the GF and those who showed a “hot hand fallacy” (HH). This fallacy is the opposite of the GF, as these HH participants tend to increasingly expect a continuation of a given run when its length increases. Remarkably, the RT slope was reliably steeper in HH than in GF participants. This pattern of results clearly demonstrates an influence of expectancies on performance. Indeed, while for GF participants expectancies and strength 408

What Is It Like to Learn Implicitly?

counteract each other, in HH participants the effects of the associative strength and expectancies add to each other and both contribute to the reduction of the RT. In these conditioning and RT tasks, however, participants were fully informed about the E1-E2 association before the experiment. The Perruchet effect therefore does not demonstrate the unconscious acquisition of associative knowledge. That being said, it certainly represents a pattern of results that cannot be accounted for by a single propositional model of learning and strongly supports a dual-process model that also takes into account a link formation mechanism (Perruchet 2015). Such a link formation mechanism, however, cannot be equated with an unconscious learning process. Learning theories may also differ regarding this issue but, in contrast with the propositional view, a link formation mechanism does not require contingency awareness to operate. As an example, the neural mechanisms of the link formation process has been precisely described in conditioning in bees and, in that case, no one considers that learning to be accompanied by the awareness of the to-be-learned association. In the case of human associative learning, the relationship between a link formation mechanism and awareness has been the object of different speculations but, as pointed out by Perruchet (2015) in his review, it is very much possible that one stimulus (E1) evokes a vivid and precise representation of another one (E2) even though that second stimulus is not expected to occur. What is crucial here is that such a conscious representation of E2 and conscious expectancies cannot be attributed to a single learning process as they can be manipulated independently and influence behavior in opposite directions. This conclusion is in line with other approaches of cognitive functions (De Neys 2006; Greene 2014; Osman 2004) and cognition in general (Kahneman 2011; Sun 2001), according to which cognitive performance and behavior can be seen as the output of the interplay between two processes: one automatic, potentially unconscious, process that extracts the regularities of the environment as they have been presented to the learning organism and another, conscious, rational process that forms abstract and explicit representations of the world. In the next section, I describe dual-system views of implicit learning and discuss the way in which these models relate to current theories of consciousness.

What Does Implicit Learning Research Tell Us About Consciousness and the Nature of Implicit Processes? In their taxonomy of experimental paradigms for studying consciousness Frith et  al. (1999) describe implicit learning as an experimental paradigm in which behavioral changes are not accompanied by a concurrent change in subjective experience. In this chapter, however, I have argued that the literature suggests that we are far from a scientific consensus over whether we can learn unconsciously, that is, without a concurrent change in subjective experience. Indeed, it is generally observed that people can describe at least part of the regularities they have been presented with during a learning episode, even when those regularities have been presented only incidentally. In the previous section, I argued that the Perruchet effect supports a two-process model. Cleeremans’ Radical Plasticity Thesis (Cleeremans 2011), a model of consciousness based on the idea that conscious experience is the result of a redescription mechanism of inherently unconscious neural processes (Clark and Karmiloff-Smith 1993), offers a perspective that is compatible with a two-systems view. In that framework, representations are conceived as patterns of activation distributed over processing units, and availability to consciousness depends on what Cleeremans calls the quality of representation, a graded dimension defined over stability in time, strength, and distinctiveness. 409

Arnaud Destrebecqz

These three dimensions can be defined as follows: (1) stability refers to how long a representation can be maintained active during processing, (2) strength refers to the number of processing units involved in a representation and on their respective activation level, and (3) distinctiveness of a representation is inversely proportional to the extent of overlap that exists between that representation and other similar representations within a network. Quality of representation is gradually and necessarily strengthened during processing and even weak, low-quality representations are capable of influencing processing, for instance through associative priming mechanisms. However, those representations, characteristic of implicit cognition, and because of their poor quality, are only weakly available to form the contents of consciousness precisely because of their poor quality. How does the Radical Plasticity Thesis account for conscious cognition and how does it relate and interact with these low-quality representations? Cleeremans hypothesizes that fully conscious representations are conscious representations in virtue of the fact that they are strong enough to become an object of representation for a second-order system. In other words, to be fully conscious of something, one has to form a meta-representation that holds, as its representational content, that one is conscious of being in that state (Brown et al. 2019; Rosenthal 1997). Within Cleeremans’ framework, the interplay between conscious and unconscious processes involves both gradual changes and all-or-none transitions between conscious and unconscious representations. Quality of representation is a graded dimension, which captures the fact that the subjective experience associated with being in a given state is also graded. Indeed, after the systematic and repeated presentation of two stimuli (E1 and E2) in close temporal succession, the presentation of E1 will affect the behavioral response to, but also the subjective experience associated with E2, even if one remains unaware of their association. It is only when a representation has gained sufficient strength, stability, and distinctiveness that it may become the target of a meta-representation. In Cleeremans’ own terms, the system may then ‘realize’, if it is so capable, that is, if it is equipped with the mechanisms that are necessary to support self-inspection, that it has learned a novel partition of the input; that it now possesses a new ‘detector’ that only fires when a particular kind of stimulus, or a particular condition, is present. (2011: 7) In other words, these conscious representations have specific properties and are part of a second representational system which is connected but different from the first-order system as it conveys a higher-order attitude toward the first-order knowledge. Another dual-system view of learning that is worth mentioning is the Unexpected Event Hypothesis (Esser et al. 2022; Rünger and Frensch 2008). In this view, implicit and explicit learning also depend on different sets of representations but there is no necessary direct relationship between the two. Implicit and explicit representations of a given association are based on different and independent mechanisms. The model also posits that implicit learning would occur incidentally following exposure to a structured material through associative mechanisms leading to a continuous improvement of the responses to the stimuli. The crucial point is that the second mechanism leading to the acquisition of explicit knowledge is triggered by explicitly but unexpectedly noticing a change in one’s own behavior. Such a view, one that has also been put forward in the field of (implicit) memory research (Kelley and Jacoby 1990; Wicklund 1975), considers that awareness arises in response to a disruption in the flow of performance. If, for instance, E1 is followed by another stimulus than E2, with which it was previously paired, this may cause an error or slow down the response in such a way that participants may explicitly 410

What Is It Like to Learn Implicitly?

experience this disruption. According to the Unexpected Event Hypothesis, when a mismatch is detected between expected and experienced performance, it triggers an attributional process to adjust one’s prediction and form a new, conscious, representation that takes into account this novel information. Importantly, the explicit knowledge may or may not be based on the implicit knowledge. Unlike Cleeremans’ redescription process, the monitoring process put forward by the Unexpected Event Hypothesis is not directly connected with the first-order system and can lead to the development of a representation of the structure of the task that is different from the one produced by the associative learning mechanisms at work within the implicit system. As rightly pointed out by Esser et al. (2022), these two dual systems views of implicit learning are related to the two most prominent theories of consciousness, the Global Workspace Theory and The Higher Order Thought Theory. According to the Global Workspace Theory (Baars 2005; Dehaene et al. 2011), representations are conscious as soon as they become available to all the specialized cognitive subsystems implemented in the brains. In this view, information processing in the brain’s subsystems, which support specific tasks such as association-based learning for instance, is encapsulated and occurs unconsciously and efficiently in parallel. Depending on the context of the task at hand and on the strength of the representations stored in the subsystems, modules are selected, by top-down or bottom-up processes, to enter the Global Workspace and the information they contain is made available to all the subsystems. Esser et al. suggest that the detection of a hiatus between the predicted and the actual subjective experience by the monitoring process assumed by the Unexpected Event Hypothesis can trigger the amplification mechanism that allows access to consciousness in the Global Workspace Theory. According to the advocates of this theory, the implicit knowledge inscribed within the implicit subsystems then becomes conscious in virtue of the fact that it can be shared within the global workspace but it is also conceivable that another system, mobilized within the same global workspace, proceeds to a new, conscious, learning of the regularities previously acquired within an encapsulated sub-module. Cleeremans’ Radical Plasticity Theory, by contrast, is grounded within a concurrent theory of consciousness, the so-called Higher-Order-Thought (HOT) theories of consciousness, as well as within the implicit-explicit distinction advocated by (Dienes and Perner 1999) and according to which conscious knowledge is “attitude-explicit” in the sense that the first-order representation it stands for (e.g., “E1 predicts E2”) is the target of a second-order, metarepresentation (e.g., “I  know that E1 predicts E2”, “I  believe that E1 predicts E2” .  .  .). In Cleeremans’ framework, both the first- and second-order representations are the product of continuously operating learning mechanisms and increase progressively in strength, stability in time and distinctiveness. There is a debate in the literature about the extent to which the gradual or sudden features of access to consciousness provide a potential refutation basis for the models of learning and consciousness discussed so far. The notion of gradual processing is central to Cleeremans’ model, but he also emphasizes the notion of a qualitative threshold that distinguishes conscious and unconscious knowledge. Indeed, in order for an internal state to reach consciousness it must necessarily be redescribed in order to produce a meta-representation of the latent knowledge it stands for. An important point is that for this process to take place such an internal state must be available for redescription, and for this to be the case it must have certain characteristics such as enough strength, stability over time and distinctiveness, which, as I said, are gradually increasing. In Cleeremans’ model, aside from being unconscious, implicit representations are therefore those that lack strength, stability over time and/or distinctiveness. Consistent with the Global Workspace Model, the Unexpected Event Theory emphasizes the non-linearity of access to consciousness during learning, but this theory does not exclude 411

Arnaud Destrebecqz

the possibility of gradual learning of regularities following repeated exposure to a structured material. In other words, these two models of learning share a common vision in that they are both compatible with a first-order process that includes an associative strength reinforcement mechanism. Although they hold distinct views about the defining features of conscious representations, the two models also converge on the need to hypothesize two independent learning systems. Even though it might seem like a truism, both of these models, as well as the general models of consciousness with which they can be associated, are based on the idea that unconscious processing is a reality that can be measured using appropriate methods. As I indicated in the first part of this chapter, however, such methods are far from unanimous in the scientific community. One exception is the Perruchet effect, which implements a dissociation that is difficult to account for without making the assumption of an associative learning mechanism whose effect on behavior is dissociated from that exerted by a conscious propositional mechanism. The approach initiated by Perruchet (1985) does not share the “empirical intractability” (Perruchet 2015: 124) characteristic of the dissociation approach derived from Reber’s (1967) seminal studies. It does not make it possible to demonstrate the acquisition of unconscious knowledge per se, but it does offer a more generally accepted paradigm and new research possibilities that could advance the field. As stated by Shanks (2010: 295), “the Perruchet effect represents perhaps the most clear-cut current evidence for independent implicit and explicit learning systems.” Although studies exploiting this paradigm remain scarce, many variations are possible. For example, we can mention the manipulation of the probability of appearance of the different imperative stimuli, their strength of association with the preparatory stimuli or the delay between the onset of the preparatory and imperative stimuli. All of these experimental variations can have differential effects on the conscious and unconscious components of learning and therefore help to refine our understanding of the nature of the mechanisms that learning models will need to account for. From then on, the general models of consciousness will also benefit from the development of such a research program.3

Related Topics Chapters 3, 7, 8, 9, 10, 17, 21, 27, 28, 29

Notes 1 A reinforced run is a series of consecutive reinforced (E1-E2) trials. A non-reinforced run is a series of consecutive non-reinforced trials (E1 alone). 2 The moderate effect size of this additional associative component is not surprising, for two reasons. First, the partial reinforcement schedule necessarily limits the strength of the associative link that can be built. Second, reaction times to the target are mostly determined by a motor priming effect whether the target is successfully associated with the predictive sound or not. There is therefore only a limited opportunity for the associative strength to further improve performance. 3 I would like to thank Axel Cleeremans for his comments and suggestions on a previous version of this chapter and J. Robert Thompson for his tact and his patience.

References Baars, B. J. 2005. “Global workspace theory of consciousness: toward a cognitive neuroscience of human experience”. Progress in Brain Research, 150: 45–53. Barrett, L. C., and Livesey, E. J. 2010. “Dissociations between expectancy and performance in simple and two-choice reaction-time tasks: a test of associative and nonassociative explanations”. Journal of Experimental Psychology. Learning, Memory, and Cognition, 36: 864–877.

412

What Is It Like to Learn Implicitly? Bennett, I. J., Howard, J. H., Jr, and Howard, D. V. 2007. “Age-related differences in implicit learning of subtle third-order sequential structure”. The Journals of Gerontology. Series B, Psychological Sciences and Social Sciences, 62: P98–P103. Bertels, J., San Anton, E., Boursain, E., Bulf, H., and Destrebecqz, A. 2022. “Visual statistical learning in infancy: discrimination of fine-grained regularities depends on early test trials”. Infancy: The Official Journal of the International Society on Infant Studies, 27: 462–478. Brown, R., Lau, H., and LeDoux, J. E. 2019. “Understanding the higher-order approach to consciousness”. Trends in Cognitive Sciences, 23: 754–768. Burns, B. D., and Corpus, B. 2004. “Randomness and inductions from streaks: ‘gambler’s fallacy’ versus ‘hot hand’ ”. Psychonomic Bulletin and Review, 11: 179–184. Cheesman, J., and Merikle, P. M. 1986. “Distinguishing conscious from unconscious perceptual processes”. Canadian Journal of Psychology, 40: 343–367. Chomsky, N., and Miller, G. A. 1958. “Finite state languages”. Information and Control, 1: 91–112. Clark, A., and Karmiloff-Smith, A. 1993. “The cognizer’s innards: a psychological and philosophical perspective on the development of thought”. Mind and Language, 8: 487–519. Cleeremans, A. 2011. “The radical plasticity thesis: how the brain learns to be conscious”. Frontiers in Psychology, 2: 86. Cleeremans, A. 2014. “Connecting conscious and unconscious processing”. Cognitive Science, 38: 1286–1315. Cleeremans, A., Allakhverdov, V., and Kuvaldina, M. 2019. Implicit learning: 50 years on. London: Routledge. Dehaene, S., Changeux, J.-P., and Naccache, L. 2011. “The global neuronal workspace model of conscious access: from neuronal architectures to clinical applications”. In S. Dehane and Y. Christen, eds., Characterizing consciousness: from cognition to the clinic? Heidelberg: Springer: 55–84. De Neys, W. 2006. “Dual processing in reasoning: two systems but one reasoner”. Psychological Science, 17: 428–433. Destrebecqz, A., Perruchet, P., Cleeremans, A., Laureys, S., Maquet, P., and Peigneux, P.  2010. “The influence of temporal factors on automatic priming and conscious expectancy in a simple reaction time task”. Quarterly Journal of Experimental Psychology, 63: 291–309. Destrebecqz, A., Velde, M. V., Anton, E. S., Cleeremans, A., and Bertels, J. 2019. “Saving the Perruchet effect: a role for the strength of the association in associative learning”. Quarterly Journal of Experimental Psychology, 72: 1379–1386. Dienes, Z., Altmann, G. T. M., Kwan, L., and Goode, A. 1995. “Unconscious knowledge of artificial grammars is applied strategically”. Journal of Experimental Psychology. Learning, Memory, and Cognition, 21: 1322–1338. Dienes, Z., and Berry, D. 1997. “Implicit learning: below the subjective threshold”. Psychonomic Bulletin and Review, 4: 3–23. Dienes, Z., and Perner, J. 1999. “A theory of implicit and explicit knowledge”. Behavioral and Brain Sciences, 22: 735–808. Dienes, Z., and Seth, A. 2010a. “Gambling on the unconscious: a comparison of wagering and confidence ratings as measures of awareness in an artificial grammar task”. Consciousness and Cognition, 19: 674–681. Dienes, Z., and Seth, A. K. 2010b. “Conscious and the unconscious”. In G. Koob, M. Le Moal, and R. Thompson, eds., Encyclopedia of behavioral neuroscience, Amsterdam: Elsevier: 322–327. Esser, S., Lustig, C., and Haider, H. 2022. “What triggers explicit awareness in implicit sequence learning? Implications from theories of consciousness”. Psychological Research, 86: 1442–1457. Frith, C., Perry, R., and Lumer, E. 1999. “The neural correlates of conscious experience: an experimental framework”. Trends in Cognitive Sciences, 3: 105–114. Giurfa, M. 2007. “Behavioral and neural analysis of associative learning in the honeybee: a taste from the magic well”. Journal of Comparative Physiology. A, Neuroethology, Sensory, Neural, and Behavioral Physiology, 193: 801–824. Greene, J. D. 2014. “Beyond point-and-shoot morality: why cognitive (neuro)science matters for ethics”. Ethics, 124: 695–726. Heerey, E. A., and Velani, H. 2010. “Implicit learning of social predictions”. Journal of Experimental Social Psychology, 46: 577–581. Jacoby, L. L. 1991. “A  process dissociation framework: separating automatic from intentional uses of memory”. Journal of Memory and Language, 30: 513–541. Jacoby, L. L., Yonelinas, A., and Jennings, J. 1997. “The relation between conscious and unconscious (automatic) influences: a declaration of independence”. In J. Cohen and J. Schooler, eds., Scientific approaches to consciousness. Hillsdale, NJ: Lawrence Erlbaum Associates: 13–47.

413

Arnaud Destrebecqz Kahneman, D. 2011. Thinking, fast and slow. New York: Macmillan. Kelley, C. M., and Jacoby, L. L. 1990. “The construction of subjective experience: memory attributions”. Mind and Language, 5: 49–68. Kirkham, N. Z., Slemmer, J. A., and Johnson, S. P. 2002. “Visual statistical learning in infancy: evidence for a domain general learning mechanism”. Cognition, 83: B35–B42. Kuhn, T. S. 1970. The structure of scientific revolutions. Chicago: University of Chicago Press. Lewicki, P. 1986. “Processing information about covariations that cannot be articulated”. Journal of Experimental Psychology. Learning, Memory, and Cognition, 12: 135–146. Livesey, E. J., and Costa, D. S. J. 2014. “Automaticity and conscious control in single and choice response time versions of the Perruchet effect”. Quarterly Journal of Experimental Psychology, 67: 646–664. Lovibond, P. F., and Shanks, D. R. 2002. “The role of awareness in Pavlovian conditioning: empirical evidence and theoretical implications”. Journal of Experimental Psychology. Animal Behavior Processes, 28: 3–26. Mitchell, C. J., De Houwer, J., and Lovibond, P. F. 2009. “The propositional nature of human associative learning”. The Behavioral and Brain Sciences, 32: 183–246. Neal, A., and Hesketh, B. 1997. “Episodic knowledge and implicit learning”. Psychonomic Bulletin and Review, 4: 24–37. Norman, E., and Price, M. C. 2012. “Social intuition as a form of implicit learning: sequences of body movements are learned less explicitly than letter sequences”. Advances in Cognitive Psychology, 8: 121–131. Osman, M. 2004. “An evaluation of dual-process theories of reasoning”. Psychonomic Bulletin and Review, 11: 988–1010. Overgaard, M. 2015. “Consciousness research methods: the empirical ‘hard problem’ ”. In M. Overgaard, ed., Behavioural methods in consciousness research. Oxford: Oxford University Press: 3–6. Perruchet, P. 1985. “A pitfall for the expectancy theory of human eyelid conditioning”. Pavlovian Journal of Biological Sciences, 20: 163–170. Perruchet, P. 2008. “Implicit learning”. In J. Byrne, ed., Cognitive psychology of memory. Vol. ii. Oxford: Elsevier: 597–621. Perruchet, P.  2015. “Dissociating conscious expectancies from automatic link formation in associative learning: a review on the so-called Perruchet effect”. Journal of Experimental Psychology. Animal Learning and Cognition, 41: 105–127. Perruchet, P., Cleeremans, A., and Destrebecqz, A. 2006. “Dissociating the effects of automatic activation and explicit expectancy on reaction times in a simple associative learning task”. Journal of Experimental Psychology: Learning, Memory, and Cognition, 32: 955–965. Perruchet, P., and Pacteau, C. 1990. “Synthetic grammar learning: implicit rule abstraction or explicit fragmentary knowledge?”. Journal of Experimental Psychology. General, 119: 264–275. Perruchet, P., and Pacteau, C. 1991. “Implicit acquisition of abstract knowledge about artificial grammar: some methodological and conceptual issues”. Journal of Experimental Psychology. General, 120: 112–116. Persaud, N., McLeod, P., and Cowey, A. 2007. “Post-decision wagering objectively measures awareness”. Nature Neuroscience, 10: 257–261. Reber, A. S. 1967. “Implicit learning of artificial grammars”. Journal of Verbal Learning and Verbal Behavior, 6: 855–863. Reber, A. S. 1989. “Implicit learning and tacit knowledge”. Journal of Experimental Psychology. General, 118: 219–235. Reingold, E. M., and Merikle, P. M. 1988. “Using direct and indirect measures to study perception without awareness”. Perception and Psychophysics, 44: 563–575. Rosenthal, D. M. 1997. “A Theory of Consciousness”. In N. Block, O. J. Flanagan, and G. Guzeldere, eds., The nature of consciousness. Cambridge, MA: MIT Press: 729–754. Rünger, D., and Frensch, P. A. 2008. “How incidental sequence learning creates reportable knowledge: the role of unexpected events”. Journal of Experimental Psychology. Learning, Memory, and Cognition, 34: 1011–1026. Saffran, J. R., Aslin, R. N., and Newport, E. L. 1996. “Statistical learning by 8-month-old infants”. Science, 274: 1926–1928. Searle, J. R. 1998. “How to study consciousness scientifically”. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 353: 1935–1942. Shanks, D. R. 2010. “Learning: from association to cognition”. Annual Review of Psychology, 61: 273–301. Shanks, D. R., and St. John, M. F. 1994. “Characteristics of dissociable human learning systems”. The Behavioral and Brain Sciences, 17: 367–395. Sun, R. 2001. Duality of the mind: a bottom-up approach toward cognition. Hillsdale, NJ: Lawrence Erlbaum Associates.

414

What Is It Like to Learn Implicitly? Timmermans, B., and Cleeremans, A. 2015. “How can we measure awareness? An overview of current methods”. In M. Overgaard, ed., Behavioural methods in consciousness research. Oxford: Oxford University Press: 21–46. Wicklund, R. A. 1975. “Objective self-awareness”. In L. Berkowitz, ed., Advances in experimental social psychology. Vol. 8. New York: Academic Press: 233–275. Wierzchoń, M., Asanowicz, D., Paulewicz, B., and Cleeremans, A. 2012. “Subjective measures of consciousness in artificial grammar learning task”. Consciousness and Cognition, 21: 1141–1153. Witt, K., Nühsman, A., and Deuschl, G. 2002. “Intact artificial grammar learning in patients with cerebellar degeneration and advanced Parkinson’s disease”. Neuropsychologia, 40: 1534–1540.

415

INDEX

Note: Page numbers in italic indicate a figure and page numbers in bold indicate a table on the ­corresponding page. racist 313 – 314; see also propositional attitudes; psychological attitudes automaticity 4 – 5, 7; and addiction 283 – 290; automatic versus deliberate processes 380 – 381; and dual process theory 105 – 112; and implicit bias 116 – 120; and implicit learning 402 – 403, 407 – 409; and implicit mechanisms in action 271 – 275; and implicit reasoning 378 – 386; measuring and modeling 45 – 48; and race 314 – 315 awareness 6; focal 192 – 194; and implicit memory 366 – 370; self-awareness 5 – 6, 171 – 172, 177, 227, 231, 302, 315; subsidiary 192 – 194

action: and addiction 283 – 287, 290 – 291; adjustments in 274 – 275; and expertise 203 – 206; and habit 169 – 170, 176 – 179; and implicit beliefs 218 – 219; implicit mechanisms in 271 – 277; and implicit mental states 80 – 82; and implicit processing 129 – 130, 149 – 150; intentional 2, 192, 272, 330; and social cognition 314 – 321, 336 – 342; and tacit knowledge 192 – 195; see also behavior active representation 250; versus dormant 36 – 37 addiction 282 – 291 agency: and addiction 282 – 291; ascribing beliefs to agents 342 – 343; and implicit mechanisms in action 271 – 277; and phenomenology 300 – 308; sense of 276 – 277 aliefs 36, 61 – 62, 382 analysis, levels of 95 – 97 approach bias 287 – 290 articulability of thought 7 – 8, 69 – 71, 74 – 76, 79, 84 assessment: of implicit cognition in relation to addiction 282 – 288, 291 associationism 117 – 118; associationist versus propositional contents 381 – 383 associations: and addiction 284 – 290; and implicit bias 115 – 123; and implicit beliefs 220 – 221; and implicit learning 402 – 403, 407 – 411; and implicit reasoning 378 – 383; measuring and modeling implicit cognition 44 – 52; and unconscious mentality 61 – 63 attentional bias 287 – 289 attitudes 115 – 116; explicit 45, 95 – 97, 116 – 121, 221, 319; implicit 45, 95 – 97, 117 – 121, 147, 150, 218 – 223, 285, 314, 319 – 321, 332 – 333 ;

background (Husserl) 168, 174 – 175, 178 – 179 Bayesian approaches 130 – 132, 137 – 139, 264 behavior 1 – 8; and addiction 283 – 291; behavior/ testimony mismatch 79 – 82, 85; discordant 3 – 4, 61 – 62; and expertise 203 – 206; and failures of recall 363 – 368; and habit 168 – 170, 176 – 179; and implicit beliefs 218 – 222; and implicit learning 407 – 412; implicit mechanisms in 271 – 277; and implicit mental states 79 – 85; and implicitness 34 – 40, 56 – 65, 69 – 76, 95 – 98; and implicit memory 366 – 368; and implicit processing 109 – 111, 116 – 123, 129 – 130, 137 – 138, 149 – 150; and implicit reasoning 381 – 385; and implicit self-knowledge 231 – 232; intentional 2, 192, 271 – 275, 324 – 332; and language 237 – 238, 249 – 256; predicting 51 – 52, 83, 326; skilled 79 – 85; and social cognition 313 – 321, 336 – 343; and tacit knowledge 192 – 195; and unconscious inference 158 – 162

416

Index belief: closure of 79, 219; and cognitive penetration 144 – 151; and dual process theory 110 – 112; and expertise 202 – 211; false beliefs 337 – 340, 342 – 343; and habit 168 – 179; implicit beliefs 82, 215 – 223, 232; and implicit bias 115 – 122; and implicit knowledge 392 – 393, 397 – 399; and implicit memory 357 – 358; and implicitness 33 – 39, 44, 60 – 65, 79 – 85; and implicit reasoning 377 – 382; and implicit self-knowledge 226 – 232; and language processing 247 – 248, 248, 251 – 254; metarepresentational 338 – 345; and social cognition 313 – 314, 325 – 328; and tacit knowledge 182 – 189, 191 – 200; and Theory of Mind 340 – 343; and unconscious inference 155 – 164 belief attributions 216, 218, 222; see also mental state attributions bias: approach bias 287 – 290; attentional bias 287 – 289; cognitive bias modification (CBM) 282, 287 – 291; explicit bias 319; and imaginings 121 – 123; implicit bias 4, 36 – 39, 45 – 47, 61 – 63, 95 – 97, 115 – 123, 149 – 150, 221 – 222, 232, 313 – 314, 317 – 320; malleability of 51; racial bias 48, 51, 315 – 319 blame, epistemic 147 – 149 body, the 159 – 160, 172 – 173, 186 – 187, 209, 313 – 316, 321 – 322; habitual body 176 – 177, 321; see also embodiment Brownstein, Michael 5, 115 – 116, 221 – 222, 324 – 333, 381 – 382

34 – 38; and implicit mental states 83 – 84; and language processing 251 – 252, 256 – 257; and the levels metaphor 93 – 97; and predictive processing 130 – 133, 136 – 138 concept possession 80 – 81, 83, 85 – 86 conceptual clarity 247, 254 – 255, 258, 303 conceptual content 70 – 71, 73, 227 – 228, 340 conceptual/nonconceptual distinction 69 – 76, 227 – 228, 231 conceptual representation 93; versus unconscious 35 – 36 conceptual thought 341, 346n3 conscious awareness of memories 366; see also memory consciousness 3 – 6, 56 – 68; and implicit learning research 409 – 412; levels of 97 – 99; life of 169 – 171; pre-reflective self-consciousness 302 – 303 conscious processes 6, 58, 93, 159, 273 conscious representation 35 – 36, 63, 409 – 412 conscious states 57 – 59, 229 – 230, 251, 254 conscious thought 26, 58, 173 constructs 45 – 49, 115 – 123 content: conceptual 70 – 71, 73, 227 – 228, 340; implicit 147, 260 – 262, 381 – 383; nonconceptual 69 – 76, 227, 231; representational 8, 26, 34 – 41, 57, 61, 65 – 66, 79 – 86, 122 – 123, 130 – 133, 175, 187, 192, 200, 215 – 223, 227 – 228, 252, 271 – 272, 355v359, 378, 381 – 382 context: contextual variation of bias 51; and cultural permeation 316 – 317; socially contextualized interactions 314 – 316 contrast-class-ification 5, 21 – 25, 22, 24 control 3 – 5, 7; and addiction 282 – 291; and cognitive penetration 149 – 151; and Dual Process Theory 105 – 108; and implicit bias 115 – 121; and implicit mechanisms in action 271 – 277; and implicit mental states 84 – 85; and implicit reasoning 380 – 386; and implicit social cognition 324 – 332; and implicit Theory of Mind 341 – 342; and the levels metaphor 91 – 99; measuring and modeling 45 – 48; and phenomenology 300 – 308; and predictive processing 127 – 129; and race 313 – 315 control, levels of 93 – 95 coping, skillful 177 corrections during action 274 – 275 ‘craftwork with integrity’ 210 – 211 cultural permeation 316 – 317, 319

causal perspectivism 161 – 164 change, study of 205 – 206 chasm between implicit and explicit processes 378, 383 – 386, 383 Chomsky, Noam 16 – 17, 93, 96 – 98, 237 – 245, 404 classification: contrast-class-ification 5, 21 – 25, 22, 24 closure of belief 79, 219 cognition: explicit 1 – 8, 21 – 27, 90 – 93, 96 – 97, 117 – 118, 123, 326, 330; implicit 1 – 8, 21 – 27, 90 – 92; social 313 – 321, 324 – 333, 336 – 346 cognitive bias modification (CBM) 282, 287 – 291 cognitive influences 144 – 146, 150 – 151; implicit 147 – 149 cognitive penetration 144 – 151, 317n1 cognizing (Chomsky) 237 – 245 coherence (in discourse) 389 – 390, 394 – 399 collective, the: collective knowledge 202 – 211; and language 208 – 209 Collins, Harry 197 – 200, 202 – 211 competence/performance distinction 37, 239 – 243 composition, levels of 92 – 93 computation and computational accounts of cognition: and implicit knowledge 389 – 391, 397 – 399; and implicit mental representation

Davies, Martin 137 – 138, 199, 223n6, 227, 254, 257, 324 – 327 declarative knowledge 250 – 251, 253 – 254, 256, 258 declarative representations 253 déjà vu 365 – 366

417

Index and implicit mental representation 34; and implicit self-knowledge 226, 232; and pragmatic inference 263 – 266; and predictive processing 136; and tacit knowledge 183 – 184, 199; versus implicit 227 – 228 explicit measures 272, 276, 283, 353 – 354 explicit predictive processing 127 – 128; core tenets of 128 – 129; and explicit representations 134 – 136, 135; implicit Bayesian inference in 137 – 139; implicit representation in 132 – 134; intellectual 130 – 131; radical 131 – 132 explicit process 93 – 99, 127 – 139, 274, 345, 378 explicit reports 276 explicit representation 16; and implicit beliefs 217 – 219; and implicit knowledge 392; and implicit learning 409 – 410; and implicit memory 358 – 359; and implicit mental representation 33 – 37; and implicit selfknowledge 228; and the levels metaphor 93; and predictive processing 128, 134 – 136, 135; principles of grammar 255 – 257 explicit self-knowledge 229 – 231 explicit task 94, 336 – 340 explicit theory of mind 337 – 339 explicit thought 11, 34, 173

deliberate processes: versus automatic 380 – 381 deliberative thought 2 – 4, 61 – 64 deliberative reasoning 2 – 3, 378 – 386, 389, 393 democracies, pluralist 210 – 211 description: personal levels of 251 – 253; subpersonal levels of 251 – 253 development: ontogenetic 330 – 331; of Theory of Mind 336 – 350 direct measures 45 – 48, 116, 282 – 283 direct reports 9; see also verbal reports discourse model 389 – 392, 394 – 395, 397 dispositions 36 – 41, 74 – 76, 133 – 134, 215 – 218, 250 – 258, 331 – 332 distributed knowledge 202 – 211 division of labour 207 – 208, 207 dormant representation 36 – 37 dual process theory (DPT) 4 – 5, 105 – 112; and addiction 283, 290; coherence of 106 – 107; and implicit bias 105 – 112, 115, 118, 120 – 121; and implicit knowledge 389; and implicit learning 409 – 410; and implicit mental representation 36, 39; and implicit reasoning 378, 380, 383 – 386, 383; and implicit Theory of Mind 339 – 343; and the levels metaphor 94; a logic of narrative 393 – 394; measuring and modeling 46; System-1/System-2 processes 94, 380, 383, 384 – 385, 389, 393, 394; Type-1/Type-2 processes 4 – 5, 105 – 112, 380, 383, 389, 393, 399

failures of recall 362 – 371; see also memory False Belief (FB) tasks 337 – 340, 342 – 343 familiarity 364 – 365 fluency 203, 206 – 211, 368 – 370 focal awareness 192 – 194 folk psychology 1 – 2, 86, 221, 251 – 254, 327 – 328, 332, 337 – 341 forgetting see failures of recall; memory formats, propositional 377, 380 – 382, 386 forms of life 198, 202, 203 fragmentation: of belief 219 – 222; of mind 79 – 87, 120

embodiment 168 – 170, 176 – 177; embodied perception 170, 172 – 174; embodied racism 320 – 321; embodied social interaction 313 – 321 empathy 187 – 188, 329 – 330 enactive accounts 25, 127, 131, 169, 315, 327, 333 epistemic blame 147 – 149 epistemic threats 147 – 149 epistemology 199 – 200 error: error detection 274 – 275; immunity to error through misidentification 226 – 227, 229 existing representations 37 experience: of agency 271 – 277; collective and distributed knowledge 202 – 211; perceptual 60, 70 – 71, 90, 94, 145 – 149, 158 – 159, 161 – 164, 173, 176 – 177, 379; pre-reflective 168, 171 – 172, 229, 300 – 308; unconscious inference in 155 – 164 expertise 202 – 211 explication 204 – 205 explicit attitudes 45, 95 – 97, 116 – 121, 221, 319; explicit bias 319 explicit cognition 1 – 8, 21 – 27, 90 – 93, 96 – 97, 117 – 118, 123, 326, 330 explicit judgment 170, 276 – 277 explicit knowledge 389 – 400; and expertise 204 – 207; and implicit knowledge 410 – 411;

games 209 Gendler, Tamar 3, 80, 221, 382 generative linguistics 237 – 239 goal-directedness 272 – 274 grammar: and cognizing 237 – 245; and the conceptual/nonconceptual distinction 69 – 73, 76; and expertise 204; grammatical knowledge 241 – 242; and implicit learning 403 – 405, 404; and implicit mental representation 36; and language processing 248 – 254; and the levels metaphor 93, 96 – 98; principles of 255 – 257; and tacit knowledge 237 – 245; and unconscious mentality 56 habit 168 – 169; and the concept of horizon 178 – 179; and embodied perception 172 – 174; the habitual body 176 – 177, 321; and the life

418

Index implicit/explicit distinction 10, 21 – 25, 33 – 41, 90 – 99, 115 – 120, 127 – 139, 220, 251 – 254, 326, 330, 336, 353 – 357, 380 – 386 implicit influencers 149 – 150 implicit judgment 385 implicit knowledge 71 – 73; of logic 389 – 400; in pragmatic inference 259 – 266; versus explicit 227 – 228; see also tacit knowledge implicit learning 402 – 412 implicit measures 18, 44 – 46; and addiction 283, 288, 291; and implicit bias 116 – 117, 119; and implicit mechanisms in action 272, 275 – 276; and social cognition 317, 338, 344; and memory 353 implicit mechanisms 271 – 277 implicit memory 353 – 360, 366 – 368; and awareness 368 – 370; and fluency 368 – 370; storage 357 – 359 implicit mental representation see implicit representation implicit moral judgment 111 – 112; see also moral judgment implicitness 399 – 400; and the conceptual/ nonconceptual distinction 69 – 76; and the levels metaphor 90 – 99; measuring and modeling 44 – 53; and mental representation 33 – 41; and mental states 79 – 87; and unconscious mentality 56 – 66; varieties of 222 – 223 implicit predictive processing 127 – 128; core tenets of 128 – 129; and explicit representations 134 – 136, 135; implicit Bayesian inference in 137 – 139; implicit representation in 132 – 134; intellectual 130 – 131; radical 131 – 132 implicit processes 6 – 7; and addiction 284 – 289; automatic versus deliberate 380 – 381; and cognitive penetration 144 – 151; and dual process theory 105 – 112; and implicit bias 115 – 123; implicit cognitive processes 289 – 290; and implicit knowledge 389, 399; and implicit learning research 407, 409 – 412; and implicit mechanisms in action 271, 274, 277; and implicit reasoning 378; and language processing 247 – 258, 248; and the levels metaphor 93 – 99; and predictive processing 127 – 139 implicit reasoning 377 – 386; sketch of a computation 397 – 399 implicit recollection 355 – 357 implicit reports 8 – 9, 215 – 222, 271 – 276 implicit representation 150, 219, 253, 392, 411; common ways of demarcating 33 – 41; in predictive processing 132 – 134 implicit self-knowledge 226 – 232 implicit social cognition 324 – 333 implicit tasks 336, 339 – 343, 345, 354 implicit Theory of Mind 336 – 346 implicit thought 11, 116 – 123

of consciousness 169 – 171; and passive synthesis 174 – 176; and self-awareness 171 – 172; and skillful coping 177; see also skill Helmholtz, Hermann von 56, 132, 155 – 164, 378 – 380 history: of implicit memory 353 – 355; of implicit reasoning 378 – 379; of unconscious mentality 57 – 58 horizons 168, 174 – 176, 178 – 179 Husserl, Edmund 168 – 169, 301, 303, 307; the concept of horizon 178 – 179; embodied perception 172 – 174; the habitual body 176 – 177; the life of consciousness 169 – 171; passive synthesis 174 – 176; self-awareness 171 – 172; skillful coping 177 IAT (Implicit Association Test) 9; and addiction 285 – 289; and implicit beliefs 220 – 221; and implicit processing 117, 121, 149; measuring and modeling 44 – 45, 49 – 52; and social cognition 314 illiterates, narrative reasoning in 394 – 399 imaginings: and bias 121 – 123 imitation games see Turing Test immunity to error through misidentification 226 – 227, 229 implicature 263 – 265 implicit aspects of embodied social interaction 313 – 321 Implicit Association Test see IAT (Implicit Association Test) implicit associative response (IAR) 285 – 287 implicit attitudes 45; and addiction 285; and cognitive penetration 147, 150; and implicit beliefs 218 – 223; and implicit bias 117 – 121; and the levels metaphor 95 – 97; and social cognition 314, 319 – 321, 332 – 333 implicit Bayesian inference 137 – 139 implicit beliefs 82, 215 – 223, 232 implicit bias 4, 36 – 39, 45 – 47, 61 – 63, 95 – 97, 149 – 150, 221 – 222; and processing 115 – 123; and racist attitudes 313 – 314, 317 – 320; and self-knowledge 232 implicit cognition 1 – 8, 21 – 27, 90 – 92; and addiction 282 – 291; and cognitive penetration 144 – 151; and the conceptual/ nonconceptual 69 – 76; and dual process theory 105 – 112; implicit social cognition 324 – 333; measuring and modeling 44 – 53; and unconscious mentality 56 – 66 implicit cognitive influence 147 – 149 implicit cognitive processes 289 – 290 implicit contents 34 – 41, 69 – 76, 80 – 86, 147, 260 – 262, 355 – 358, 381 – 383

419

Index 248; linguistics 137, 237 – 268; and pragmatic inference 259 – 268; and tacit knowledge 237 – 246 learning: and dual process theory 108 – 109; and implicit knowledge 389 – 401; and implicit reasoning 377 – 388; learning implicitly 402 – 412 levels 90 – 99; of analysis 95 – 97; and consciousness 97 – 99; of control 93 – 95; higher 90 – 92, 99, 129, 145; lower 84, 90 – 93, 97 – 98, 129; Marr’s account of 95 – 97; metaphor 90 – 99; personal 72 – 76, 79 – 87, 183 – 188, 250 – 258, 341; of size 92; subpersonal 72 – 73, 76, 79 – 87, 199, 250 – 258 linguistic processing see language logic 389 – 400; closure of belief under logical consequence 79

indirect measures 44 – 50, 53, 73, 121 – 123, 221, 283 – 285, 288; and implicit bias 116 – 118; process modeling of 49 indirect reports 44 – 45 infants 340 – 343; see also development inference 36 – 37, 62 – 64, 128 – 132, 137 – 139, 327 – 332, 377 – 386, 390 – 399; Helmholtz on 155 – 164; pragmatic 259 – 266; unconscious 379 – 380 inferential integration 26, 39, 62 – 63, 341 – 346 influence: of cognition on perception 144 – 151; of memory on behavior 366 – 368; of memory on thought 363 – 366 information: in retrieval 362 – 371; self-related 227 integration, inferential 26, 39, 62 – 63, 341 – 346 intellectualism 4, 26 – 27, 74 – 76, 377 – 379; ‘intellectualist’ predictive processing 127 – 132, 134 – 139 intentional content see content, representational intentionality 7, 56 – 57, 64 – 66, 177 – 178; unconscious 58 – 61 intentions 145 – 149, 273 – 274; intentional action 2, 192, 272, 330 interactional expertise: and the division of labor 207 – 208, 207; and imitation games 209; and language practice 206 – 207 interaction see social interaction interpretation: implicit cognition 46; unifying interpretative difference 63 – 66 introspection 36, 64, 86, 98, 122, 146, 272 – 274 intuitive reasoning 105 – 110, 383 – 385, 383 intuitive thought 4

know-how 15 – 17, 25, 39 – 40, 69 – 70, 73 – 76, 136, 139, 195, 199, 241 – 243; versus know-that 17, 25, 74, 243 – 245; and self-knowledge 231 knowing: and expertise 202 – 211; and habit 168 – 179; and implicit beliefs 215 – 223; and implicit self-knowledge 226 – 232; and tacit knowledge 182 – 189, 191 – 200; and unconscious inference 155 – 164 knowledge 202 – 211, 240 – 245; declarative 250 – 251, 253 – 254, 256, 258; explicit 34, 136, 183 – 184, 199, 204 – 207, 226 – 228, 232, 263 – 266, 389 – 400, 410 – 411; grammatical 241 – 242; implicit 71 – 73, 227 – 228, 259 – 266, 389 – 400; practical 73 – 76, 241; procedural 69, 231, 250, 253, 256, 258; tacit 182 – 189, 191 – 200, 203 – 205, 237 – 245; see also selfknowledge

malleability of bias 51 Marr, David 11, 56, 66, 73, 93, 95 – 97, 164 Marr’s three levels 95 – 97 measures: and constructs 48; direct 45 – 48, 116, 282 – 283; explicit 272, 276, 283, 353 – 354; implicit 18, 44 – 46, 116 – 117, 119, 272, 275 – 276, 283, 288, 291, 317, 338, 344, 353; indirect 44 – 50, 53, 73, 116 – 118, 121 – 123, 221, 283 – 285, 288; nonverbal 15, 215 – 223; and operating conditions 47 – 48; verbal 15, 215 – 223 mechanism: implicit mechanisms in action 271 – 277; levels of 92 – 93 memory 353 – 373; assessment of addictionrelevant memory associations 284 – 287; and failures of recall 362 – 371; implicit 353 – 360; influence of 363; retrieval failure 363 – 366; storage 357 – 359 mentality, unconscious 56 – 66 mentalizing 324 – 333, 339 mental state attributions 79 – 87, 328; personallevel versus subpersonal-level approaches to 81 – 87 mental representation see implicit representation; explicit representation metaphor 260 – 262 meta-representational beliefs 338 – 345 misidentification, immunity to error through 226 – 227, 229 mismatch, behavior/testimony 79 – 80 modeling and models 44 – 53; discourse 389 – 392, 394 – 395, 397 modus tollens (MT) 395 – 396 moral judgment 105 – 112 moral reasoning 111 – 112

language 206 – 207, 211; and cognizing 237 – 246; and the collective 208 – 209; language practice 206 – 207, 211; language processing 247 – 258,

narrative: and dual processing 393 – 394; narrative reasoning in illiterates 394 – 399 nonconceptual/conceptual distinction 69 – 76

judgment: explicit 170, 276 – 277; implicit 385; moral 105 – 112

420

Index pragmatic inference 259 – 266 predicting behavior 51 – 52, 83, 254, 326, 340 predictive processing 127 – 128, 163 – 164, 249; core tenets of 128 – 129; and explicit representations 134 – 136, 135; implicit Bayesian inference in 137 – 139; implicit representation in 132 – 134; intellectual 130 – 131; radical 131 – 132 pre-reflective: experience 168, 171 – 172, 229, 300 – 308; feeling 276; level 321 privileged access 80 – 81 procedural dispositions 253 procedural knowledge 69, 231, 250, 253, 256, 258 processes: automatic versus deliberate 380 – 381; conscious 6, 58, 93, 159, 273; explicit 93 – 99, 127 – 139, 274, 345, 378; implicit 6 – 7, 93 – 99, 105 – 112, 115 – 123, 127 – 139, 144 – 151, 247 – 258, 271, 274, 277, 284 – 290, 378, 380 – 381, 389, 399, 407, 409 – 412; predictive 127 – 139; unconscious 97, 271, 370, 407, 410 processing 115 – 123; and cognitive penetration 144 – 151; and dual process theory 105 – 112; higher levels of 90 – 91, 99, 129; implicit bias 115 – 123; linguistic 247 – 258, 248; lower levels of 84, 98, 129; predictive 127 – 139, 163 – 164, 249 process modeling 49 propositional attitudes 34, 56, 70, 215 – 225, 243, 252 – 253, 327, 331 – 332, 337, 339, 341 – 343, 382; and implicitness 34, 56, 70; and implicit reasoning 382; and language 243, 252 – 253; and social cognition 327, 331 – 332, 337, 339, 341 – 343 propositional contents 381 – 383; see also content propositional formats 377, 380 – 382, 386 propositionalism 118 – 120 psychological attitudes 44 – 45, 115 – 117, 221 – 223, 252 – 253, 331 – 332 psychology: folk 1 – 2, 86, 221, 232, 251 – 254, 324, 327 – 328, 332, 337 – 341; physiological 157; pure 157 psychopathology 300 – 308

nonconceptual content 69 – 76, 227, 231 nonconceptual representation 42n5 nonconceptual self-awareness 227 nonconceptual thought 70 – 73, 227 – 231 nonconscious states see unconscious states nonverbal measures 15, 215 – 223 nonverbal reports 3 – 4, 285 ontology 199 – 200 operating conditions 46 – 47 operating principles 46 – 48 organization, levels of 92 – 93 outcomes, treatment (addiction) 288 – 291 partially implicit memory 359 – 360 passive synthesis 174 – 176 penetration, cognitive 144 – 151, 317n1 perception: and cognitive penetration 144 – 151; and the conceptual/nonconceptual distinction 70 – 71; and dual process theory 111 – 112; embodied 172 – 174; and expertise 202 – 211; and habit 168 – 179; Helmholtz’s analysis 158 – 161; and implicit beliefs 215 – 223; and implicit mechanisms in action 275 – 276; and implicit mental states 80 – 84; and implicit reasoning 379 – 380; and implicit selfknowledge 226 – 232; and implicit Theory of Mind 337 – 341; and language processing 255; and the levels metaphor 90, 94; the mind in 155 – 158; perceptual theory 161 – 164; and phenomenology 302 – 304; and predictive processing 127 – 133, 136 – 139; and race 315, 318 – 320; and tacit knowledge 182 – 189, 191 – 200; and unconscious inference 155 – 164; and unconscious mentality 60; visual 80 performance: competence/performance distinction 239 – 243 permeation, cultural 316 – 317 Perruchet effect 407 – 409 personal-level: approaches to attributing implicit mental states 79 – 87; content ascription 72 – 73; descriptions 251 – 253; knowledge 74 – 76; processing 72 – 73, 81 – 87, 250 – 258 perspectivism, causal 161 – 164 phenomena, implicit 92 – 97; characterization of 4 – 8, 21 – 27 phenomenology 168 – 179, 184, 193, 229, 300 – 308, 315; of embodied perception 172 – 174; see also agency, sense of physiological psychology 157 pluralist democracies 210 – 211 Polanyi, Michael 14 – 15, 168, 172 – 174, 177, 179 – 180, 182 – 189, 191 – 194, 196, 199, 200, 204, 212 potentially inferred representations: versus existing 37 practical knowledge 73 – 76

Quadruple Process model 50 – 51, 50; applying 51 – 52 race 313 – 321 racial bias 48, 51, 315 – 319 racism, embodied 320 – 321 racist attitudes 313 – 314 ‘radical’ predictive processing 131 – 132 rationality 37, 80 – 83, 86, 108 – 110, 230, 264, 284, 378 – 379, 383 – 386 rational learning 108 – 109 rational speech acts 263 – 265

421

Index coping 177; and tacit knowledge 182 – 189, 191 – 194; see also know-how; procedural knowledge social cognition: implicit social cognition 115, 324 – 333, 381; and race 313 – 321; see also mentalizing; Theory of Mind social interaction 313 – 321 socialisation, expertise as 210, 210; and human intelligence 211 socially contextualized interactions 314 – 316 speech acts: rational 263 – 265; recognition 262 – 263 sub-doxastic 199, 341 – 342 subpersonal: approaches to attributing implicit mental states 79 – 87; content ascriptions 72; levels of description 251 – 253; levels of processing 72 – 73, 81 – 87, 91, 111, 250 – 258, 301, 316 subsidiary awareness 192 – 194 synthesis, passive 174 – 176

reaction time 18, 93, 117, 220, 249 – 250, 272, 274, 277, 282, (RT) measures 285, 288, 408, 408n2 reasoning: deliberative 2 – 3, 378 – 386, 389, 393; implicit 377 – 388, 397 – 399; and implicit knowledge 389 – 401; intuitive 105 – 110, 383 – 385, 383; and learning implicitly 402 – 412; moral 111 – 112; narrative reasoning in illiterates 394 – 399 recall, failures of 362 – 371; see also memory recognition: speech acts 262 – 263 recollection, implicit 355 – 357 recovery: implicit cognition and addiction 288 – 290 reflection 168 – 172, 231 – 232, 273 – 274, 301 – 302, 313 – 316; reflective mind 110; reflective selfawareness 171 – 172 reports: direct 9; explicit 276; implicit 8 – 9, 215 – 222, 271 – 276; indirect 44 – 45; nonverbal 3 – 4, 285; self 3, 44, 116, 150, 272; verbal 61 – 63, 215, 219, 222, 228, 271, 406 representation: active 36 – 37, 250; as behavioral disposition of functional architecture 37 – 38; conceptual 35 – 36, 93; conscious 35 – 36, 63, 409 – 412; declarative 253; distinct 37 – 38; dormant 36 – 37; existing 37; explicit 16, 33 – 37, 93, 128, 134 – 136, 135, 217 – 219, 228, 255 – 257, 358 – 359, 392, 409 – 410; implicit 33 – 41, 132 – 134, 150, 219, 253, 392, 411; and meta-representational beliefs 338 – 345; nonconceptual 42n5; potentially inferred 37; in predictive processing 132 – 136; propositional 403; tacit 41n1; unconscious 35 – 36, 410 representationalism 215; problems for 216 Ryle, Gilbert 13, 13n12, 14, 16, 26, 26n19, 74 – 75, 136, 162, 168, 194 – 199, 231

tacit 14 – 17, 33, 69, 71, 72, 128, 168, 171, 178, 182 – 189, 191 – 200, 202 – 211, 217, 227, 237, 239 – 241, 243, 253, 257, 300 – 302, 305, 326, 331 tacit knowledge 182 – 189, 191 – 200, 203 – 204, 237 – 245; attributions of 72 – 73; and explication 204 – 205; transfer of 206 tacit representation 41n1 tacit self-awareness 171 – 172 tasks: explicit 94, 336 – 340; False Belief (FB) tasks 337 – 340, 342 – 343; implicit 336, 339 – 343, 345, 354 testimony: behavior/testimony mismatch 79 – 80; see also verbal reports Theory of Mind (ToM) 18, 231, 327, 336 – 346 thought: articulability of 7 – 8, 69 – 71, 74 – 76, 79, 84; conceptual 341, 346n3; conscious 26, 58, 173; deliberative 2 – 4, 61 – 64; explicit 11, 34, 173; implicit 11, 116 – 123; intuitive 4; nonconceptual 70 – 73, 227 – 231; unconscious 56 – 58, 314 threats, epistemic 147 – 149 tip-of-the-tongue (TOT) state 363 transfer of knowledge 202, 206, 211 transformation, explication by 205 treatment outcomes (addiction) 288 triple process theory (TPT) 110 trust 206 Turing Test 209 – 211 two systems see dual process theory Type 1 and Type 2 processes see dual process theory typicality effects 265 – 266

science: levels of 92; in pluralist democracies 210 – 211; usefulness as a field site 204 self-awareness 5 – 6, 177, 231, 302, 315; nonconceptual 227; reflective 171 – 172; tacit 171 – 172 self-consciousness, pre-reflective 302 – 303 self-knowledge 226 – 232 self-related information 227 self-reports 3, 44, 116, 150, 272 sense-physiology 157 skill: and the concept of horizon 178 – 179; and embodied perception 172 – 174; and expertise 202; the habitual body 176 – 177; and implicit mechanisms in action 272 – 273; and implicit memory 354 – 355; and implicitness 52, 74, 93, 97; and implicit self-knowledge 231 – 232; and implicit social cognition 325 – 326, 330 – 333; and the life of consciousness 169 – 171; and passive synthesis 174 – 176; and self-awareness 171 – 172; skilled behavior 79 – 85; skillful

unconscious inference 155 – 164, 379 – 380 unconscious intentionality 59 – 61

422

Index variation, contextual 51 verbal measures 15, 215 – 223 verbal reports 61 – 63, 215, 219, 222, 228, 271, 406

unconscious mentality 56 – 66 unconscious processes 4, 6, 55 – 66, 97, 271, 370, 403, 407, 410 unconscious representation 35 – 36, 410 unconscious states 6, 36, 56, 58 – 59, 62, 251 – 252, 254 unconscious thought 56 – 58, 314

Wittgenstein, Ludwig 183, 192, 194 – 199, 202 – 205, 230

423