New Perspectives on Epistemic Closure (Routledge Studies in Epistemology) [1 ed.] 0367610205, 9780367610203

Citation preview

New Perspectives on Epistemic Closure

This volume brings together new research on the topic of epistemic closure from both leading philosophers and emerging voices in epistemology. It connects epistemic closure principles to related themes in epistemology such as scepticism, dogmatism, evidentialism, epistemic logic, and modal epistemology. Epistemic closure is of central importance to contemporary epistemology, so much so that no epistemology is complete without an answer to the question of where it stands on the issue. The chapters in this book touch on the central themes of closure and transmission and argue for and against different closure and transmission principles. The contributors address issues such as whether knowledge and justification are closed under deductive entailment; whether scepticism can be properly contained by restricting closure principles; whether justification for a set of premises can fail to transmit across inference to a conclusion; Moore’s Paradox; and which theories of knowledge—contextualism, contrastivism, or relevant alternatives epistemology—emerge from denying closure. New Perspectives on Epistemic Closure will be of interest to scholars and advanced students working in epistemology. Matthew Jope is Leverhulme Trust Early Career Fellow at the University of Edinburgh. His main area of research is epistemology, with a focus on closure, scepticism, trust, and risk. Previously, he held postdoctoral positions at the University of Barcelona and the University of Glasgow, and before that, he was a Marie Curie Early Stage Researcher at the University of Edinburgh where he completed his PhD. His work has appeared in the Journal of Philosophy, Synthese, and Erkenntnis. Duncan Pritchard FRSE is UC Distinguished Professor of Philosophy at the University of California, Irvine, and Director of the Center for Knowledge, Technology & Society. His main area of research is epistemology, and he has published widely in this area, including the monographs Epistemic Luck (2005), The Nature and Value of Knowledge (co-authored, 2010), Epistemological Disjunctivism (2012), and Epistemic Angst: Radical Skepticism and the Groundlessness of Our Believing (2015). His most recent book is Scepticism: A Very Short Introduction (2019). In 2007, he was awarded the Philip Leverhulme Prize. In 2011, he was elected to a Fellowship of the Royal Society of Edinburgh. In 2013, he delivered the annual Soochow Lectures in Philosophy in Taiwan.

Routledge Studies in Epistemology

Edited by Kevin McCain, University of Alabama at Birmingham, USA, and Scott Stapleford, St. Thomas University, Canada

The Philosophy of Group Polarization Epistemology, Metaphysics, Psychology Fernando Broncano-Berrocal and J. Adam Carter The Social Epistemology of Legal Trials Edited by Zachary Hoskins and Jon Robson Intellectual Dependability A Virtue Theory of the Epistemic and Educational Ideal T. Ryan Byerly Skeptical Invariantism Reconsidered Edited by Christos Kyriacou and Kevin Wallbridge Epistemic Autonomy Edited by Jonathan Matheson and Kirk Lougheed Epistemic Dilemmas New Arguments, New Angles Edited by Kevin McCain, Scott Stapleford and Matthias Steup Proposition and Doxastic Justification New Essays on Their Nature and Significance Edited by Paul Silva Jr. and Luis R.G. Oliveira Epistemic Instrumentalism Explained Nathaniel Sharadin New Perspectives on Epistemic Closure Edited by Matthew Jope and Duncan Pritchard For more information about this series, please visit: https://www. routledge.com/Routledge-Studies-in-Epistemology/book-series/RSIE

New Perspectives on Epistemic Closure

Edited by Matthew Jope and Duncan Pritchard

First published 2023 by Routledge 605 Third Avenue, New York, NY 10158 and by Routledge 4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN Routledge is an imprint of the Taylor & Francis Group, an informa business © 2023 selection and editorial matter, Matthew Jope and Duncan Pritchard; individual chapters, the contributors The right of Matthew Jope and Duncan Pritchard to be identified as the authors 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. ISBN: 978-0-367-61020-3 (hbk) ISBN: 978-0-367-61231-3 (pbk) ISBN: 978-1-003-10476-6 (ebk) DOI: 10.4324/9781003104766 Typeset in Sabon by codeMantra

Contents

List of Contributors

vii



2 Modest Foundations for Perceptual Knowledge

24

K R I S TA L AW L O R

3 Closure, Warrant Transmission, and Defeat

43

MONA SI M ION

4 Modal Epistemology and the Logic of Counterfactuals

58

T I MOT H Y W I LL I A MSON

5 The Skeptical Paradox and the Generality of Closure (­and Other Principles)

77

Y U VA L AV N U R

6 Dogma, Defeat, and Transmission

97

C H R ISTOPH K EL P

7 In Defence of Closure DU NCA N PR I TC H A R D

113

vi Contents

8 No Spindly Brown Grass: Knowledge, Closure, and Subjective Distinguishability

131

G E N I A S C H Ö N B AU M S F E L D

9 Underdetermination and Closure: Thoughts on Two Sceptical Arguments

147

M A RT I N S M I T H

10 Closure and Transmission Again

163

CR ISPI N W R IGH T

Index

191

Contributors

Yuval Avnur is Professor of Philosophy and Director of the Humanities Institute at Scripps College, Claremont Matthew Jope is Leverhulme Trust Early Career Fellow at the University of Edinburgh Christoph Kelp  is Senior Lecturer in Philosophy and Director of the COGITO Epistemology Research Centre at the University of Glasgow Krista Lawlor is Henry Waldgrave Stuart Memorial Professor of Philosophy at Stanford University Duncan Pritchard FRSE  is UC Distinguished Professor of Philosophy and Director of the Center for Knowledge, Technology & Society at the University of California, Irvine Genia Schönbaumsfeld  is Professor of Philosophy at the University of Southampton Mona Simion is Professor of Philosophy at the University of Glasgow Martin Smith is Reader in Philosophy at the University of Edinburgh Timothy Williamson FBA, FRSE is Wykeham Professor of Logic at the University of Oxford and A. Whitney Griswold Visiting Professor of Philosophy at Yale University Crispin Wright FBA, FRSE, FAAAS, MAE is Global Distinguished Professor of Philosophy at New York University and Professor of Philosophical Research at the University of Stirling

Introduction Matthew Jope and Duncan Pritchard

This volume brings together new research on the topic of epistemic closure, an issue that has been at the forefront of epistemological discussion since the late 1960s. The general thought behind closure is that deductive inference preserves epistemic properties like knowledge and justification. Very roughly, to know that P and to deduce Q from P is to know that Q. By reasoning deductively from known premises, we guarantee knowledge of our conclusions. Such principles strike us as intuitively compelling. How could one fail to know something that one has competently deduced from known premises? Such principles also feature in the background of countless arguments throughout the history of philosophy. Indeed, we might suppose that it is the paradigm way to argue for some claim by showing that it is a logical consequence of some other set of claims which we know to be true. And yet, despite their pervasiveness in philosophical argumentation and prima facie plausibility, closure principles have come into question in recent decades. It is now just over 50 years since the publication of Fred Dretske’s seminal paper, ‘­Epistemic Operators,’ which sparked the debate about epistemic closure. Dretske offered a series of examples that purportedly demonstrate the failure of closure, the general structure of which involves an ordinary, quotidian proposition such as those animals are zebras that entails a conclusion which is the denial of an error possibility, such as those animals are not cleverly disguised mules. Since it is possible to know the former while not being in a position to know the latter, Dretske argues we ought to give up unrestricted epistemic closure. One benefit of doing so is that it would neutralise a particularly troubling form of sceptical argumentation that appeals to closure in order to show that we lack knowledge of many ordinary, everyday propositions, on account of the fact that we cannot, it seems, know the denials of the sceptical error possibilities that are entailed by these everyday claims. There has also been extensive debate about the relevance of such principles to a range of other issues, such as the problem of easy knowledge, epistemic entitlement, Moore’s paradox, various epistemic puzzles (­like the preface and lottery paradoxes), and a number of topics in formal epistemology. This volume brings together important new perspectives DOI: 10.4324/9781003104766-1

2  Matthew Jope and Duncan Pritchard on epistemic closure, from both established researchers and emerging voices in epistemology. In the opening chapter, “­I ntuitive Closure, Transmission Failure and Doxastic Justification,” Matthew Jope puts pressure on the claim that the Moorean inference can be diagnosed as exhibiting transmission failure but not closure failure. To do this, he begins by noting that the kind of closure principle typically used to establish this diagnosis has, in the recent literature on closure, been shown to be subject to several counterexamples. With a reformulated principle in play, Jope then goes on to distinguish between closure and transmission principles for both propositional and doxastic justification, arguing that while it may be possible for these principles to come apart for propositional justification, it seems much harder to see how they could come apart for doxastic justification. Jope considers the conservative account of the structure of justification needed to support the transmission failure diagnosis of Moore and attempts to develop a corresponding conservative account of the structure of doxastic justification, showing us that any such account will either lead to failure of closure or else lead to the implausible conclusion that certain types of justified beliefs are immune from bad basing. In “Modest Foundations for Perceptual Knowledge,” Krista Lawlor considers a problem for modest foundationalism about perceptual justification. While it seems plausible that many of our perceptual beliefs are justified without the need for further justified beliefs, this modest foundationalist view seems to licence an excess of justification, as it seems to warrant empirically ambitious or even outlandish beliefs that might be inferred from ordinary starting points. Lawlor develops a commonsense view of warrant transmission that helps the modest foundationalism block the excess. Furthermore, the view brings to light interesting differences between warrant transmission and closure. In “­Closure, Warrant Transmission, and Defeat,” Mona Simion considers ­Neo-​­Moorean responses to scepticism, distinguishing between what she labels ‘­concessive’ and ‘­radical’ versions. She argues that the former concede too much to the sceptic, while the latter are unable to say anything about what it is that seems in some sense reasonable about the sceptical position. In response, Simion offers a novel account of what she terms radical ­neo-​­Mooreanism which denies there is anything wrong with the Moorean inference while at the same time being able to explain the source of the intuition that there is something epistemically good about the sceptical position. Although the sceptic is in violation of epistemic norms by not believing in the conclusion of Moore’s inference, they are nonetheless obeying a c­ ontrary-­​­­to-​­duty obligation that arises given the commitments that follow from their unjustified sceptical beliefs. In “­Modal Epistemology and the Logic of Counterfactuals,” Timothy Williamson addresses the counterfactual conditional analysis

Introduction  3 of knowledge due to Robert Nozick, according to which closure for knowledge fails as a result of the failure of closure for counterfactual conditionals that are assumed to be necessary for knowledge, such as safety and sensitivity. Nozick’s account relies on a classic semantics of counterfactuals. Williamson argues that this classic semantics for counterfactuals should be rejected on the grounds that it fails to do justice to the role of ‘­would’ as an independently meaningful modal operator contributing to the compositional semantics. He then proposes an alternative, compositional semantics for counterfactual conditionals which takes these to be contextually restricted strict conditionals. On this new semantics, principles such as strengthening the antecedent, which failed on Nozick’s account, turn out to be valid after all. The implications for safety and sensitivity are discussed and it is shown that, given this new semantics, a qualified closure principle for knowledge holds after all. In “­The Sceptical Paradox and the Generality of Closure (and Other Principles),” Yuval Avnur considers a sceptical c­losure-​­based paradox for justification. He argues for a solution to this paradox that rejects the standard formulation of closure on the grounds that it applies universally from any proposition P to any other proposition Q that follows deductively from P. Avnur argues that this universal formulation of closure ought to be replaced by a restricted closure principle that excludes deduction to a special class of propositions. This class of propositions is determined by a bias in our notion of justification which skews towards ­anti-​­sceptical assumptions. Avnur then explores some wider implications for this solution, relating to its application to other forms of sceptical paradoxes. In “­Dogma, Defeat and Transmission,” Christoph Kelp considers the ­closure-​­based sceptical paradox and the dogmatism paradox together. Kelp argues that what these paradoxes share in common is a reliance on the principle of warrant transmission and attempts to find a unified strategy for dealing with both kinds of paradox. He considers strategies for solving the paradox that preserve the principle of transmission by appealing to the notion of defeat, and other strategies that abolish the principle of transmission by appealing to epistemic modals or the notion of background warrant; however, he finds these solutions face a range of problems. Kelp then goes on to explore the prospects for a novel ­transmission-​­abolitionist account that appeals to the distinction between ­question-​­settling inquiry and ­question-​­begging inquiry, where the latter gives rise to transmission failure. In “­I n Defence of Closure,” Duncan Pritchard defends a closure principle for knowledge. Pritchard begins by arguing that the most plausible formulation of such a principle must be understood as a diachronic, competent deduction principle. He then highlights the difference between local and global error possibilities, arguing that these are best kept apart when it comes to discussions of how to defend closure. Pritchard offers

4  Matthew Jope and Duncan Pritchard two strategies for defending knowledge closure against arguments that involve local error possibilities. Both strategies appeal to a distinction between favouring and discriminating epistemic support, with one of the strategies in addition appealing to the kind of factive epistemic support defended by epistemological disjunctivism. He then moves on to discuss strategies for dealing with global error possibilities, outlining in detail his preferred approach which makes use of the Wittgensteinian notion of a hinge commitment. Prichard’s claim is that such commitments are neither in the market for knowledge nor the kinds of things that can feature as the conclusions of ­closure-​­style inferences. Consequently, they are inapplicable to ­closure-​­based inferences and therefore do not present counterexamples to closure. In “No Spindly Brown Grass: Knowledge, Closure and Subjective Distinguishability” Genia Schönbaumsfeld examines Fred Dretske’s case for the rejection of epistemic closure. Schönbaumsfeld shows that Dretske espouses a conception of perceptual reasons that Schönbaumsfeld labels the ‘Reasons Identity Thesis’. This thesis is widely endorsed in epistemology, but Schönbaumsfeld argues that it concedes too much to the sceptic unnecessarily. Moreover, the argument Dretske offers in defence of a related claim—­ which appeals to the thought that we are unable to subjectively discriminate between good and bad pairs of perceptual cases—does not offer a good reason to adopt the Reasons Identity Thesis. Schönbaumsfeld then goes on to offer an alternative conception of perceptual reasons that manages to avoid the problems that arise for Dretske’s account, and which does not make such concessions to the sceptic. In “Underdetermination and Closure. Thoughts on Two Sceptical underdetermination-​­ based Arguments” Martin Smith considers the ­ sceptical argument alongside the ­closure-​­based sceptical argument and argues that neither of these two classic forms of scepticism are as philosophically interesting as is commonly supposed. Beginning with the underdetermination argument, Smith addresses the motivation behind one of its key premises and argues that this assumption begs the question against the ­non-​­sceptic. Moving on to the ­closure-​­based argument, Smith considers the motivation for one of its key premises, arguing that it assumes a principle that, while stopping short of begging the question, is nonetheless easily rejected by the ­non-​­sceptic. Smith concludes that these two forms of sceptical argument do not, contrary to received philosophical wisdom, expose deep or hidden tension in our orthodox epistemological theorising. Rather, they rest on assumptions that a­ nti-​ s­ ceptics are under no pressure to accept. Finally, in “­Closure and Transmission Again,” Crispin Wright revisits the debate regarding the status of c­ losure-​­style ‘­transmission principles’ (­a debate that he instigated in his earlier work), along with the related issue of transmission failure. He argues against the recent trend

Introduction  5 to formulate closure principles diachronically, a move that is held to capture what is at issue in the transmission principle. Wright maintains that we should stick with the original distinction between (­­non-​­diachronic) closure and transmission, and that dynamic closure principles for knowledge in particular are inconsistent with a rationality constraint on ­knowledge-​­acquisition. He offers a template for transmission failure, which he illustrates with four widely discussed cases from the philosophical literature: Moore’s ‘­Proof of an External World’; McKinsey’s ‘­proof’ of the incompatibility of content externalism with privileged ­ rains-­​­­in-­​­­a-​­vat; and s­ elf-​­knowledge; Putnam’s ‘­proof’ that we are not b Dretske’s famous ‘­Zebra’ case. In light of his discussion of transmission failure, Wright also critically discusses Duncan Pritchard’s recent attempts to defend (­diachronic) closure for knowledge.

1 Intuitive Closure, Transmission Failure, and Doxastic Justification1 Matthew Jope

1 Introduction: Closure vs Transmission Some philosophers have argued that knowledge is not closed under known logical entailment. Roughly speaking, the thought is that there exist counterexamples to the claim that if one knows that P, and knows that P entails Q, then one knows that Q.2 One such alleged counterexample is the infamous Moorean inference: Moore Here is a hand. If this is a hand, then the external world exists. Therefore, the external world exists. If Moore is indeed a case of closure failure, then one might know that there exist things such as hands, and know that this entails that there is an external world, while failing to know that there exists an external world. The ­closure-​­denialists claim that the principle of epistemic closure fails to hold unrestrictedly, as exemplified by the Moorean inference. Others have argued that the rejection of epistemic closure is a philosophical abomination. Indeed, for some authors, closure has an almost axiomatic status in epistemology.3 If in the end the rejection of closure is the correct theory of knowledge to adopt, we should not lose sight of the fact that the proposal is highly revisionary, forcing us to abandon principles we routinely take for granted. An alternative diagnosis of what is wrong with Moorean reasoning holds that it is not closure that fails but a related though logically stronger principle of transmission of justification or warrant. An inference transmits justification if it presents a route to acquiring justification for its conclusion. Transmission thus appears stronger than closure in the sense that it entails but is not entailed by closure. Closure might hold, but transmission fail, in cases where one has justification for both premise and conclusion while the latter obtains independently of having carried out the inference. It follows that transmission will fail in cases where one cannot possess justification for the premise unless one already has justification for the conclusion. Those who diagnose Moore as exhibiting DOI: 10.4324/9781003104766-2

Intuitive Closure, Transmission Failure, Doxastic Justification  7 transmission failure argue that this is exactly what we find. They hold that it is only if one already has some justification to believe that there is an external, ­mind-​­independent reality to which one’s perceptual experiences correspond that one can acquire justification for beliefs based on perception. If so, one cannot acquire justification for the conclusion of Moore by carrying out the inference, on pain of a kind of epistemic circularity. Nonetheless, closure is respected because so long as one has justification for the premise, one is guaranteed to have some (­­inference-​ ­independent) justification for the conclusion. One purported benefit of this approach is that by diagnosing Moore in terms of a failure of transmission but not closure we are able to avoid the kind of epistemological revisionism associated with the c­ losure-​­failure diagnosis. By rejecting the logically stronger principle of transmission, so the thought goes, we can safeguard the logically weaker principle of closure. The aim of this chapter is to put pressure on the transmission failure diagnosis by arguing that it is committed to rejecting closure after all. I will show that the formulation of closure used to draw a distinction with transmission is an outdated formulation that, in more recent work on closure, has been shown to be vulnerable to several counterexamples (­Section 2). With a more intuitive formulation of closure in play, I will then go on to show that it is much harder to see how there can be cases of transmission failure that are not also cases of closure failure. In order to do this, I will distinguish between closure and transmission for different epistemic properties, arguing that while we may be able to make sense of how closure and transmission of propositional justification could come apart (­Section 3), it is much more difficult to see how closure and transmission of doxastic justification could come apart (­Section 4). The upshot is that if transmission of doxastic justification fails for the kinds of reasons given by those who defend transmission failure, it looks as though closure for doxastic justification must fail also.

2 The Intuitive Closure Principle The general idea behind closure is that epistemic operators such as knows that and probable that penetrate to the known consequences of a proposition. The first to introduce the idea into the philosophical literature, Fred Dretske formulates closure in the following way. Naïve Closure4 If S knows that P, and knows that P entails Q, then S knows that Q.5 There are, however, several problems with this initial formulation. For example, one might satisfy the antecedent but not the consequent of Naïve Closure simply by failing to form a belief in Q. Such a case presents a relatively trivial counterexample to Naïve Closure. Further

8  Matthew Jope uninteresting counterexamples include cases where, although one does have a belief that Q, this belief fails to count as knowledge because it is formed on a bad basis, or cases in which one loses the knowledge or justification for P while carrying out the inference. For reasons such as these, more recent discussions around closure tend towards strengthening the antecedent of the principle, formulating closure diachronically, with an explicit emphasis on how closure is meant to capture one way that we may legitimately extend our knowledge through inference. For example, Duncan Pritchard writes that the reason we ought to find epistemic closure so compelling is that “­such principles attempt to codify how one might legitimately extend one’s knowledge via competent deduction from what one already knows” (­2016, 13). I will borrow the term Intuitive Closure from Williamson, to refer to the diachronic formulation of knowledge closure which is meant to capture this idea that inference is a way to extend our knowledge.6 Intuitive Closure If S knows that P, and S competently deduces Q from P, thereby coming to believe that Q on the basis of the competent deduction while retaining their knowledge that P, then S knows that Q. This formulation avoids the problems associated with the naïve formulation by stipulating that S has competently deduced Q from P. What makes Intuitive Closure so intuitive is the thought that deduction is a paradigm way of growing one’s knowledge base. Williamson argues from the intuition that “­deduction is a way of extending one’s knowledge” to the Intuitive Closure principle (­2000, 117). Echoing Williamson, John Hawthorne argues that “­T he core idea behind closure is that we can add to what we know by performing deductions on what we already know” (­2005, 29). Likewise, in trying to formulate a satisfying closure principle, Steven Luper takes us to be trying to capture the intuition that “­we can extend our knowledge by recognizing, and accepting thereby, things that follow from something that we know” (­Luper, 2016). And in a similar spirit, Duncan Pritchard articulates the force of the intuitiveness of Intuitive Closure by asking “­How could one draw a competent deduction from one’s knowledge … without thereby coming to know the deduced conclusion?” (­2016, 14).7 One thing that is interesting to note here is that, if we take these remarks at face value and understand closure as a dynamic, ­knowledge-​ ­extension principle, the line between closure and transmission starts to blur. Recall that the transmission failure diagnosis discussed above involves the thought that in certain cases closure holds while transmission fails because the justification one has for Q is independent of both the justification one has for P and of the inference from P to Q. For this diagnosis to make sense, Wright has to appeal to a static formulation of closure

Intuitive Closure, Transmission Failure, Doxastic Justification  9 which states merely that “­whenever there is warrant for the premises of a valid argument there is warrant for the conclusion too” (­2000, 141).8 In order to get the result that he needs, the version of closure Wright appeals to must allow for cases where one has warrant for the conclusion that is independent of the warrant one has for the premise(­s). This is not a dynamic principle along the lines of Intuitive Closure but rather more like the static Naïve Closure formulation. Recall that Intuitive Closure stipulates that a subject base their belief in Q on the competent deduction. With this restriction in play, it is hard to see how Intuitive Closure will hold in alleged cases of transmission failure in which the justification for the conclusion is independent of the justification for the premises and thus independent of the deduction from those premises. This somewhat undermines the benefits of the transmission failure diagnosis that Wright offers because the claim that this strategy can retain closure by rejecting the stronger principle of transmission now comes with a caveat: the kind of closure principle that it can retain is a static formulation. It remains to be seen whether the transmission failure diagnosis can retain the kind of dynamic formulation of closure along the lines of Intuitive Closure. If it has to reject these closure principles as well, the transmission failure framework for dealing with problems such as diagnosing Moorean reasoning is more theoretically costly than is commonly acknowledged. In order to fully appreciate the problem, it will be necessary to have a precise formulation of transmission that we can compare to Intuitive Closure. In attempting to articulate a comparative transmission principle, there are a number of ambiguities that we need to address. Firstly, there is the fact that while the above closure principles, following Dretske and others, concern knowledge, discussions around transmission and its failure, following Crispin Wright, typically involve justification or warrant. To assess whether closure and transmission can indeed come apart in the way that Wright suggests, we ought to be clear about which epistemic property we have in mind. We will need therefore to articulate transmission principles for knowledge and for justification. Secondly, the term transmission failure appears to be ambiguous between a number of distinct phenomena, including inferences that are unable rationally to overcome doubt about their conclusions, inferences that are inapt to provide their conclusions with the same justification enjoyed by their premises, inferences that are unable to strengthen an already justified conclusion, and inferences that are unable to provide a ­fi rst-​­time justification for their conclusions. Most of the literature on transmission and transmission failure has tended to focus on this last sense of transmission, including most of Wright’s discussion on the topic.9 For this reason, I will also restrict the focus onto transmission in this sense of an inference providing a ­fi rst-​­time justification for its conclusion. Finally, an additional complication that arises is the fact that philosophers distinguish between two kinds of justification: propositional and

10  Matthew Jope doxastic. Propositional justification refers to that which one has justification to believe. Doxastic justification refers to the justificatory status of actual beliefs. Propositional justification for a proposition P does not necessarily entail that one’s belief that P is justified. For example, I presumably have propositional justification to believe that oranges are edible. However, doxastic justification is not guaranteed. I may believe this proposition on a bad epistemic basis (­i.e. not on the propositional justification that I in fact have) and therefore fail to have a justified belief despite my having justification to hold this belief. This raises the question of whether transmission of justification is best understood in terms of propositional or doxastic justification. Wright himself seems to have something like the former notion in mind when he talks about transmission. For example, in clarifying the concept of transmission, he writes “­to acquire a warrant for the premises of a valid argument and to recognise its validity is to acquire […] a warrant to accept the conclusion” (­2000, ­140–​­141; emphasis added).10 Again, the tendency is for the debate to follow Wright in this regard, as Moretti and Piazza concur: “­Most epistemologists investigating transmission and transmission failure broadly identify the epistemic property capable of being transmitted with propositional justification” (­2018). Nonetheless, we at least want to ask questions about doxastic justification too. After all, it does not seem trivial that if there is transmission failure of propositional justification then there will be transmission failure of doxastic justification, and we surely do want to know whether justified belief transmits. Moreover, one might even go further and say that doxastic justification is the more interesting notion because any interest in whether we would have justification to believe some proposition ultimately derives from our interest in whether we could justifiably believe it. Some authors are sensitive to this distinction between closure and transmission for different types of justification. Most notably, Nicholas Silins (­2005) and Chris Tucker (­2010) seem to agree that the property we ought primarily to be interested in is that of justified belief. To quickly recap, we now have three kinds of epistemic property about which we want to ask questions: knowledge, propositional justification, and doxastic justification. And we have two kinds of epistemic principle that we want to compare: closure and transmission. Given that the question that we want an answer to is whether closure and transmission can come apart for each of our three kinds of epistemic properties, what we really have are three questions on our hands: Q1: Can closure and transmission for propositional justification come apart? Q2: Can closure and transmission for doxastic justification come apart? Q3: Can closure and transmission for knowledge come apart?

Intuitive Closure, Transmission Failure, Doxastic Justification  11 With the questions laid out in this way, it becomes easier to appreciate that there are a variety of different ways that the transmission failure diagnosis could be understood, and thus a variety of different ways in which closure and transmission could potentially come apart. Earlier on, we noted some reservations over whether closure and transmission could come apart once we clarify an ‘­intuitive’ closure principle. However, the Intuitive Closure principle we formulated was a principle that concerns knowledge, while Wright’s characterisations of transmission failure seem to indicate a principle of propositional justification. What we will need in order to substantiate the claim that closure and transmission can come apart are contrasting closure and transmission principles for each of our epistemic properties. We have already seen that much of the literature on transmission failure has focused on transmission of propositional justification or warrant, i.e. on Q1. Wright’s conception of transmission seems to be about the acquisition of warrant “­to accept” conclusions (­2002, 332).11 Supposing we grant for the sake of argument his account of transmission failure, we might then wonder what follows with respect to Q2 and Q3. Does Wright’s account of the structure of justification itself entail corresponding failures of transmission for justified belief and knowledge? One reason for thinking that we want answers to these questions in addition to questions about propositional justification is that most of the initial debates around the possibility of counterexamples to closure focused on knowledge closure. Insofar as the transmission failure diagnosis is supposed to offer an alternative to the closure failure diagnosis, it would be somewhat unsatisfying if it had nothing to say about the knowledge questions. Discussion of the knowledge questions can of course be avoided by embracing a wholesale knowledge scepticism. This is in fact close to views Wright has espoused under the label of “­The Russellian Retreat” (­1991, 88).12 If we give up the knowledge questions and retreat to questions of justification, then we can simply ignore Q3, but we’ll still want an answer to Q2. For this reason, I will focus on answering Q2.13 In order to do so, it is going to be helpful to clarify the relevant closure and transmission principles. Since those already enjoy much discussion with respect to propositional justification, it is going to be easier to start there and then move on to adapt them into principles for doxastic justification.

3 Transmission Failure of Propositional Justification In order to check whether transmission and closure might come apart for propositional justification, we first need to settle on a closure principle for propositional justification. Fortunately, we have the Intuitive Closure principle for knowledge, which can be easily translated into a corresponding principle for propositional justification.

12  Matthew Jope ­P-​­Closure If S has justification to believe P, and S competently deduces Q from P, while retaining their justification for P, then S has justification to believe Q. There are a couple of differences between Intuitive Closure and ­P-​ ­Closure. Obviously, Intuitive Closure concerns knowledge while ­P-​ C ­ losure concerns propositional justification. But additionally, ­P-​­Closure weakens the antecedent, dropping the part that required S to have come to believe that Q on the basis of the competent deduction. The reason for this should be clear. It was introduced to the closure principle to avoid trivial counterexamples in which the subject lacks knowledge because they had formed no belief or formed a belief on a bad basis. But given that propositional justification for a proposition P does not require that one have formed any belief in P, this part of the principle is not necessary. I take it that ­P-​­Closure is highly intuitive. However, I will not attempt to establish the truth of this epistemic closure principle because it will be sufficient to assume it is true in order to answer the question we are interested in, namely whether closure and transmission for propositional justification can come apart such that closure holds and transmission fails. There are several ways we can formulate our corresponding principle of transmission. Wright sometimes leans towards formulating transmission in terms of acquiring a justification for a proposition on the basis of a competent deduction. Here is one formulation he uses: (­T1) To acquire a warrant for the premises of a valid argument and to recognise its validity is to a­ cquire—​­perhaps for the first ­time—​­a warrant to accept the conclusion.14 Moretti and Piazza on the other hand formulate transmission, not in terms of ­justification-​­acquisition, but instead by stipulating in virtue of what it is that one’s justification for a proposition obtains. Here is their formulation: (­T2) If (­i) one has justification for P and (­ii) one competently deduces Q from P then (­iii) one has justification for Q in virtue of the satisfaction of (­i) and (­ii).15 The differences between these two ways of formulating transmission are perhaps not, ultimately, so substantial. T2 is, however, preferable for current purposes because it is more in line with the ‘­competent deduction’ formulation of Intuitive Closure and will therefore be easier to compare. One thing that we do need to flag is that both principles will be prone to one of the kinds of trivial counterexample that we saw above

Intuitive Closure, Transmission Failure, Doxastic Justification  13 for Naïve Closure in which, while carrying out the inference, the subject loses their justification for P. So, the transmission principle we require ought to stipulate that the subjects retain their justification for P while carrying out the inference. The transmission principle we will use to compare with ­P-​­Closure is thus the following: ­P-​­Transmission If (­i) S has justification to believe P and (­ii) S competently deduces Q from P, while retaining their justification for P, then (­iii) S has justification to believe Q in virtue of the satisfaction of (­i) and (­ii). We are now in a position to compare ­P-​­Closure and ­P-​­Transmission. On the face of it, it looks like these could indeed come apart in the way that Wright suggests. ­P-​­Closure simply says that if one has propositional justification for P and competently deduces Q from P, then one has propositional justification for Q. It doesn’t specify where the justification for Q comes from and neither does it make reference to the formation and basing of a belief, so there’s no reason to think that it’s incompatible with the kinds of cases Wright has in mind in which one satisfies the consequent by virtue of possessing prior justification for Q that is independent of the inference. The next thing to do is to check whether our principles can accommodate the diagnosis of ­transmission—​­but not ­closure—​­failure in Moore. This diagnosis follows from a ‘­conservative’ account of the structure of justification. According to conservatives such as Wright, certain propositions function like presuppositions of inquiry (­2004, 2014). The justification that I have to believe that I have hands depends in part on there being a corresponding perceptual experience and in part on further justification to presuppose that I am not dreaming, that I am not a brain in a vat, that there is a ­mind-​­independent reality, and so on. Conservatives think the acquisition of ordinary evidential justification can only take place within a framework in which there is prior justification to presuppose that, roughly, things are as they appear to be.16 The notion of presupposition here is a weak one in the sense that it does not require that one form an actual belief in these presuppositions. For example, Wright suggests that the correct way to understand our commitment towards these propositions is a kind of implicit trust (­2004, 1 ­ 92–​­193). Nonetheless, conservatism requires that there already be justification in place for these presuppositions. And for this reason, transmission has to fail in all cases where there is an inference from an evidentially justified proposition P to a presupposition for the method of inquiring into P based on the kind of evidence that can be brought to bear on P. Q cannot be justified in virtue of the justification for P if justification for Q is among the background conditions that make P justified in the first place. For example, Wright argues that prior background justification for there is

14  Matthew Jope an external world is necessary in order to acquire justification for here is a hand on the basis of the relevant visual experience. On the supposition that this account of justification is correct, it follows that we have a case in which ­P-​­Closure is satisfied but ­P-​­Transmission is not. One might have justification for here is a hand and also for there is an external world, and so satisfy the antecedent and consequent of ­P-​­Closure. But given that justification for there is an external world is among the conditions that make here is a hand justified in the first place, it won’t be the case that one has justification for there is external world in virtue of the justification one has for here is a hand, and ergo ­P-​­Transmission fails.

4 Transmission Failure of Doxastic Justification We have seen how Wright’s account of transmission failure works for propositional justification such that it has the result that closure (­for propositional justification) and transmission (­for propositional justification) could come apart. So, the answer to Q1 above seems to be ‘­yes’, at least granted a certain conception of the structure of justification and an accompanying account of default entitlement for presuppositions. What about Q2? As before, we’ll need to specify the relevant closure and transmission principles for doxastic justification first before examining whether they come apart. Again, we can adapt our Intuitive Closure principle for knowledge into a corresponding principle for doxastic justification: ­ -​­Closure D If S’s belief that P is justified, and S competently deduces Q from P, thereby coming to believe that Q on the basis of the competent deduction, while retaining their justified belief that P, then S’s belief that Q is justified. The first thing to note is that ­D -​­Closure, like Intuitive Closure but unlike ­P-​­Closure, contains the strengthened antecedent. This is because doxastic justification, like knowledge but unlike propositional justification, requires one to have formed a belief in the relevant proposition. A weaker formulation of a closure principle for doxastic justification would be subject to precisely the same kinds of trivial counterexamples faced by the Naïve Closure principle for knowledge in which one fails to form a belief that Q, or the belief is formed on a bad basis, or one loses the justification for the premises while carrying out the inference. A transmission principle for doxastic justification will be similar to the above ­D -​­Closure but will stipulate the conditions in virtue of which the belief that Q is justified. In line with the above discussions of transmission, its consequent ought to state that not only is S’s belief that Q justified, but it is justified in virtue of competent deduction from justified premises.

Intuitive Closure, Transmission Failure, Doxastic Justification  15 ­D -​­Transmission If (­i) S’s belief that P is justified and (­ii) S competently deduces Q from P, thereby coming to believe that Q on the basis of P, while retaining their justified belief that P, then S’s belief that Q is justified in virtue of the satisfaction of (­i) and (­ii). Formulating the principle in this way mirrors the way that transmission principles for propositional j­ustification—​­including the above ­P-​ ­Transmission—​­are formulated, stipulating the conditions under which the deduced belief is justified. Notice, though, that one important way that ­D -​­Transmission differs from ­P-​­Transmission is that it includes the strengthened antecedent, which was not needed for ­P-​­Transmission because doxastic justification requires the formulation of a belief while propositional justification does not. With these closure and transmission principles for doxastic justification now in play, we are able to ask whether they can come apart in the way needed to maintain the transmission failure diagnosis of Moore. At first glance, the principles look similar, and to some degree, it may seem as though satisfying ­D -​­Closure guarantees the satisfaction of ­D -​ ­Transmission. After all, in insisting that S’s belief that Q is formed on the basis of deduction from P, ­D -​­Closure seems to imply that justification for Q is inherited inferentially. If so, it is hard to see how there could be daylight between these two principles.17 Though it is tempting to read ­D -​­Closure in this way, all that its consequent strictly requires is that the resulting Q ­ -​­belief is justified. It is compatible with ­D -​­Closure that this justification is not transmitted via inference but is instead independent of it. For example, it is compatible with ­D -​­Closure that the justification for one’s Q ­ -​­belief is a matter of default entitlement. This means that, granted an account of default doxastic justification, it ought to be possible for these principles to come apart in cases where a belief that Q is justified (­and thus satisfies ­D -​­Closure) but not in virtue of deduction from justified premises (­thus violating ­D -​­Transmission). Though the principles themselves allow room for such a possibility, the question we need to consider is whether we can make epistemological sense of such a case. That is, whether we can make sense of a conservative account of doxastic justification that allows for the possibility of a belief that is justified independently of the inferential basis on which it is held. In order to maintain the transmission failure diagnosis of Moore, what is needed is an account of how the above two principles could come apart. I will argue that this is much harder to establish in the case of doxastic justification than it was earlier on for propositional justification. To appreciate the problem, we will first assume that ­D -​ ­Closure holds. This assumption is both r­ easonable—​­since ­D -​­Closure is extremely p ­ lausible—​­and necessary if the transmission failure project is to be vindicated. After all, a key benefit of the transmission failure

16  Matthew Jope strategy is that it promises to avoid a range of problems for the ­Dretske-​ ­style denial of closure for knowledge, but by denying closure for doxastic justification, we would simply reintroduce many of these same problems but for justified belief rather than knowledge. Furthermore, to suggest that closure holds for propositional justification but fails for doxastic justification is to entertain the thought that there could be justification to believe a proposition that could not be justifiably believed. It is unclear whether this a coherent possibility. It is tempting to think that if a proposition could in principle not be justifiably believed, then there could not be justification to believe it. All this is to say that we may reasonably assume that ­D -​­Closure holds for the purposes of assessing whether ­D -​­Transmission holds. Assuming that ­D -​­Closure holds, is it possible for ­D -​­Transmission to fail? If so, then the thought must be something like this. One can satisfy closure by having justified beliefs in both the P ­ -​­and Q ­ -​­propositions, but where the latter belief is justified independently of the deductive inference. A challenge for such an account is then to spell out what it is in virtue of which a belief in a proposition such as there is an external world could be justified if not inferentially from a justified belief in a proposition such as here is a hand.18 Perhaps a similar conservative story could be told here to the one told above for propositional justification. Recall that the answer to the corresponding question for propositional justification was that there could be a type of default warrant or justification to presuppose the existence of the external world. This default epistemic status is in no way a type of evidence but, according to Wright, is to be understood as a kind of epistemic entitlement or epistemic right (­2004, 2014).19 It is our right to trust implicitly that certain sceptical scenarios are false. ‘­I mplicit’ in the sense that such trust need involve no belief in, nor indeed any awareness of, the proposition in question. Nonetheless, the entitlement to presuppose that there is an external world needs to be in place in the structure of justification in order for one to initially acquire justification for here is a hand, and it is for this reason that propositional justification fails to transmit across the inference from here is a hand to there is an external world. If a similar story is going to work for doxastic justification, then it will be necessary to make sense of how it is that a belief in a proposition such as there is an external world could be justified independently of deductive inference from a belief in a proposition such as here is a hand. On standard accounts, doxastic justification is explained in terms of propositional justification. The thought is roughly that a belief is justified if and only if it is a belief in a proposition for which there is justification and it is based on that which propositionally justifies its content. 20 So, a belief in a proposition such as it is raining outside is justified if and only if (­a) there is justification to believe the proposition it is raining outside and (­b) the belief is based on whatever justifies this p ­ roposition—​­a

Intuitive Closure, Transmission Failure, Doxastic Justification  17 perceptual experience as of it raining outside, say. For a belief in a ­Q -​ ­proposition to be (­doxastically) justified it needs to meet these twin conditions. We can grant that condition (­a) is met given the assumption that propositional justification is closed. How about condition (­b)? One difficulty here is that the kind of propositional justification that obtains for Q ­ -​­propositions is not of the ordinary, evidential kind, but rather the default entitlement kind, and it is not immediately obvious that an epistemic ­right—​­an ­entitlement—​­is the correct sort of thing on which to base a belief. As discussed, Wright seems to understand the notion of an entitlement to Q along the lines of a right to trust that Q rather than a justification to believe that Q. In a very strict sense, then, even closure does not hold for doxastic justification, let alone for transmission. However, the same is true of the conservative account of propositional justification in the sense that evidential justification to believe is not closed. 21 Wright manages to rescue closure for propositional justification only by defending a disjunctive account of justification according to which a proposition can be justified either evidentially or by default. Perhaps, then, the more charitable way to a corresponding account of doxastic justification is to see doxastic justification as similarly disjunctive. Building off Wright’s disjunctive conception of propositional justification, we may suppose that doxastic justification comes in two varieties: either it is a matter of a belief that P being based on that which justifies P, or it is a matter of implicit trust in P where there is an epistemic right to trust. With the disjunctive account of doxastic justification now on the table, we can try to make sense of what a conservative account of doxastic justification might look like. One may justifiably believe that (­P) here is a hand only if one antecedently justifiably trusts that (­Q) there is an external world. This promises to secure closure in the sense that one may have a justified attitude towards both premise and conclusion. Transmission fails because the trust that one has towards Q is not justified in virtue of the inference but in virtue of our epistemic entitlements to take for granted our a­ nti-​­sceptical commitments. Notice, however, that although this promises to secure closure in this disjunctive sense, it will not secure closure for justified belief. For example, ­D -​­Closure will fail on this disjunctive account. This is because ­D -​­Closure specifically involves the formation of a belief—​­not merely an implicit ­trusting—​­which is based on an inference. So, in order for ­D -​­Closure to hold it is necessary that, in forming a belief that Q on the basis of an inference from a justified belief that P, the resulting belief that Q is itself justified. And this is precisely what the conservative account of the structure of doxastic ­justification—​­which we have just tried to develop by analogy with the conservative account of the structure of propositional ­justification—​­is committed to denying. We are entitled merely to trust or presuppose that there is an external world, but we are not justified in going further than this and forming a bona fide belief. If one does happen to go

18  Matthew Jope beyond merely trusting that Q and forms a belief that Q, transmission failure guarantees that the resulting belief will not be justified. Furthermore, notice that if we suppose that belief is necessary for knowledge, this means that we cannot know that Q. With respect to the Moorean inference, this means the original ­closure-​­failure diagnosis is vindicated: one can merely trust in Moore’s conclusion, which is presumably not compatible with knowledge. This is a bad result insofar as one of the key benefits of the transmission failure diagnosis was supposed to be that it avoids the epistemological revisionism associated with the rejection of closure for knowledge. A natural response that a conservative could offer here is to hold that the foregoing argument attempts to make too much of the distinction between trust and belief. Conservatives introduce this distinction as a way to mirror, attitudinally, two distinct types of epistemic support: evidential and ­non-​­evidential. But if we suppose instead that irrespective of the type of epistemic support in play the corresponding attitude is the same, then the space opens up to avoid the above problem and countenance a form of closure for doxastic justification after all. On this second way of looking at things, beliefs can be justified either evidentially or ­non-​­evidentially. In the former case, the justification has to be earned, while in the latter case the justification is unearned; however, in both cases, the propositional attitude is the same. If one satisfies the antecedent of ­D -​­Closure, then one is guaranteed to satisfy the consequent by justifiably believing the ­Q -​­proposition in either of these two ways. Thus, ­D -​­Closure holds. ­D -​­Transmission, on the other hand, will fail in cases where one’s belief in the ­Q -​­proposition is justified by default because it will be false that the belief is justified in virtue of having deduced it from anything. Transmission failure of doxastic justification occurs, according to this proposal, because certain inferences are such that their conclusions may be justifiably believed by default, independently of the inferential basis on which they are held. The account of default doxastic justification needed to support this diagnosis would be highly revisionary, but arguably no more so than the more familiar entitlement theoretic claim that there can be default propositional justification. The bigger problem that this account faces, beyond the revisionary nature of default justification in general, is that beliefs justified by default would seem to be immune from bad basing. A belief that is justified by default is justified in virtue of its content alone, so the thought would go. That is, certain beliefs enjoy the status of being justified not in virtue of the basis on which they are held but in virtue of the fact that they are beliefs in certain propositions that themselves enjoy default epistemic support. But notice that if the actual inferential basis has no bearing on the justificatory status of one’s belief in the conclusion of the Moorean inference, then one could base this belief on any inference, or indeed any basis, without affecting the belief’s justificatory status. It looks like

Intuitive Closure, Transmission Failure, Doxastic Justification  19 whether the inference is valid, for example, has no relevance to the justificatory status of the belief.22 To appreciate the point, consider the following alternative to Moore’s inference: Unhappy Moore I would be very unhappy if the external world did not exist. I do not wish to be unhappy. Therefore, the external world exists. What to say about one who reasons in this way? Clearly the reasoning is invalid. And yet, because the conclusion of Unhappy Moore is the same as that of Moore—​­and is therefore a matter of default ­entitlement—​­the account under consideration seems committed to saying that one who reasons in this way, thereby forming a belief in the conclusion, is justified in doing so. If Moore fails to transmit justification to its conclusion in virtue of the fact that one who believes its conclusion is justified by default in so b ­ elieving—​­that is, justified independently of the inferential basis on which the belief is ­held—​­then it seems clear that one could reason instead in accordance with Unhappy Moore and be justified as a matter of default too. This is a highly counterintuitive result. Invalid reasoning should not be a way to arrive at justified beliefs. To reiterate the general problem here, it appears as though the defender of transmission failure of doxastic justification has, broadly speaking, two options. The first option is to deny that Q ­ -​­type propositions such as there is an external world can ever be justifiably believed, offering instead an account of how they may nonetheless be justifiably trusted. As we noted above, this has the consequence that closure fails for justified belief. One who goes beyond mere trust in the conclusion of the Moorean inference and forms an actual belief will not be justified in so believing. Moreover, if belief and justification are necessary for knowledge, then it seems as though this first option entails the failure of closure for knowledge. Thus, option one fails to deliver as advertised the result that by rejecting transmission we can safeguard closure. The second option is to accept that ­Q -​­type propositions can be justifiably believed as a matter of default entitlement. The consequence of this is that ­ -​­type propositions is independent the justificatory status of beliefs in Q of the basis on which they are held, which would seem to immunise such beliefs from bad basing. This gets the result that closure holds but transmission fails for Moore’s inference, but at the cost that Moore may just as well have formed his belief in the existence of the external world on anything, including invalid reasoning, guessing, wishful thinking, and so on, which I submit to be a reductio of this second option. A third ­ -​­type propositions can be justifiably option is of course to accept that Q believed and that the basis on which they are believed is relevant to their justificatory status, but that is simply to accept that transmission holds after all and thus not available to defenders of the transmission failure

20  Matthew Jope approach. The upshot is that there seems to be no way to establish transmission failure for doxastic justification that does not collapse into either a rejection of closure for doxastic justification or the absurdity that certain beliefs are immune from bad basing.

5 Conclusion One of the purported benefits of the transmission failure diagnosis of the Moorean inference is that it offers a way to safeguard closure. The aim of this chapter has been to push back against this claim by arguing firstly that it relies on an outdated formulation of closure that we have independent reason to reject, and secondly, by arguing that once we shift focus onto transmission and closure for doxastic justification, and bear in mind a more Intuitive Closure principle, it becomes much harder to see how to make sense of an account of transmission failure that does not entail failure of closure also.23

Notes 1 The work that went into this chapter was made possible by funding from the European Union’s Horizon 2020 Research and Innovation programme under Grant Agreement no. 675415 and a Leonardo Grant for Researchers and Cultural Creators from the BBVA Foundation. 2 Very roughly speaking, as the discussion in Section 2 will make clear. 3 See, for example, Stewart Cohen (­1999) and also Richard Feldman (­1995). 4 The reason for labelling this Naïve Closure will be made clear below. 5 Dretske (­1970). 6 For various formulations of diachronic closure principles, see Williamson (­2000, 117), Hawthorne (­2005, 29), Silins (­2005, 90), Pritchard (­2016  & this volume). The formulation used here most closely resembles Pritchard’s formulation, which he simply labels ‘­T he Closure Principle’ (­2016, 13). I borrow the label, however, from Williamson in order to distinguish it from the rejected formulations above. 7 Although Pritchard defends something close to Intuitive Closure, he nonetheless argues that we cannot have knowledge of the conclusion of the Moorean inference for the reason that it belongs to a special class of p ­ ropositions—­​ hinges—​­ that are neither in the market for rational doubt nor for rational ­­ belief (­2016, ­63–​­66). See Jope (­2019) for a response to Pritchard’s argument. 8 See also, Wright (­2002, 332; 2003, 57). 9 As indeed Luca Moretti and Tommaso Piazza confirm in their Stanford Encyclopedia of Philosophy entry on transmission: “­Much of the extant literature on epistemic transmission concentrates on examples of transmission of ­fi rst-​­time justification” (­2018). 10 And again, for example, drawing the c­ losure-​­transmission distinction in a subsequent article, Wright seems to have in mind something more along the lines of propositional rather than doxastic warrant: Closure says that whenever there is warrant for the premises of a (­known) valid argument, there is warrant for the conclusion too. Transmission

Intuitive Closure, Transmission Failure, Doxastic Justification  21 says more: roughly, that to acquire a warrant for the premises of a valid argument and to recognise its validity is thereby to ­acquire—​­perhaps for the first ­time—​­a warrant to accept the conclusion. (­2002, 3 ­ 31–​­332) 11 It is possible to overstate the sense in which Wright’s account is an account of propositional justification. His formulations of closure and transmission principles are clearly principles about warrant or justification to believe or to accept propositions rather than about the epistemic status of agent’s doxastic ­attitudes—​­as the quoted passages indicate. However, I do not mean to suggest that Wright explicitly intends for his account to apply to propositional warrant or justification rather than doxastic. Instead, all that I am saying here is that the most plausible interpretation of his account is of an account of propositional warrant or justification and that we might do well to inquire how this account extends to doxastic warrant or justification. 12 Wright again refers to the idea of the Russellian retreat in (­2012, 473) and seems to espouse something like the view again in (­2017, 2 ­ 8–​­29). 13 For a discussion, see Silins (­2005). 14 (­Wright, 2000, ­140–​­141). 15 (­Moretti & Piazza, 2018). 16 This is a statement of conservatism about perceptual justification. But we can also think about conservatism for other sources or methods of belief formation such as memory, testimony, and so on. Elsewhere, I have argued that a commitment to conservatism in one of these domains entails a commitment to conservatism in others (­2021b). 17 Thanks to Krista Lawlor for pointing out this worry. 18 One natural idea here is to appeal to an externalist condition such as safety or reliability. Wright has argued elsewhere that such strategies, despite showing some initial promise, are ultimately unstable (­2008). In response to Wright, I have argued elsewhere that the instability argument fails on the grounds that it slips in certain internalist assumptions about evidence which the externalist is under no pressure to accept (­2021a). 19 Elsewhere, I have argued that the notion of a default epistemic right leads to a dilemma concerning the apposite degree of warranted confidence one ought to take in a proposition on the basis of an entitlement (­Jope, 2021c). 20 This standard account of doxastic justification has come under pressure from John Turri who argues that we ought to run the analysis in the other direction, understanding propositional justification in terms of doxastic (­2010). For present purposes, nothing hinges on taking a position on this. I stick with the standard account simply because it is standard, but the general point will work just as well if we take doxastic justification to be the more primitive notion. 21 Though this is unambiguously the view Wright advances in his 2004 entitlement article, two caveats are needed. Firstly, although the disjunctive account of warrant entails that justification to believe is not closed, the reason for this is one that, at a certain point, Wright seems not entirely wedded to. If indeed the concept of ‘­belief’ is evidentially controlled, then justification to believe is not closed. Wright, however, tells us “­I do not myself know whether the notion of belief is actually so tightly evidentially controlled” (­2004, 176). Rather, he seems to prefer to grant this as an assumption, meaning that the claim that ­justification-­​­­to-​­believe is not closed is ultimately conditional on this assumption. Secondly, in his subsequent 2014 article, in response to concerns regarding the alchemy and leaching problems, Wright’s

22  Matthew Jope considered view seems to be that there is closure for ­justification-­​­­to-​­believe after all, however the resulting notion of belief in Q ­ -​­type propositions is one whose evidential credentials “­depends for its force on our antecedent reason to trust in the truth of those very same propositions, so that the ultimate authority we have for accepting them depends on the rationality of that trust” (­2014, 235). 22 Perhaps additionally we might want to say that there ought to be no defeaters for the justification, though for the present discussion this caveat seems irrelevant. 23 I would like to thank the following people for helpful feedback on this chapter: Aidan McGlynn, Duncan Pritchard, Christoph Kelp, Martin Smith, and Crispin Wright.

References Cohen, S. (­1999). ‘­Contextualism, Skepticism and the Structure of Reasons.’ Philosophical Perspectives, 13: 5 ­ 7–​­89. Dretske, F. (­ 1970). ‘­ Epistemic Operators.’ The Journal of Philosophy, 67: ­1007–​­1023. Feldman, R. (­1995). ‘­I n Defense of Closure.’ The Philosophical Quarterly, 181: ­487–​­494. Hawthorne, J. (­2005). ‘­T he Case for Closure.’ In M. Steup, J. Turri, & E. Sosa (­eds.), Contemporary Debates in Epistemology (­p­­p.  ­26–​­42). Malden, MA: Blackwell. Jope, M. (­2019). ‘­Closure, Credence and Rationality: A Problem for N ­ on-​­Belief Hinge Epistemology.’ Synthese, 198: ­3565–​­3575. Jope, M. (­2021a). ‘­On the Alleged Instability of Externalist ­A nti-​­Skepticism.’ The Journal of Philosophy, 118(­1): ­43–​­50. Jope, M. (­2021b). ‘­T he Symmetry Problem for Testimonial Conservatism.’ Synthese, 199: ­6149–​­6167. Jope, M. (­2021c). ‘­Trust and Confidence: A Dilemma for Epistemic Entitlement Theory.’ Erkenntnis, https://­doi.org/­10.1007/­­s10670-­​­­021-­​­­0 0478-​­7. Luper, S. (­2016). ‘­Epistemic Closure.’ The Stanford Encyclopedia of Philosophy. Zalta (­ed.), URL = https://­plato.stanford.edu/­archives/­win2018/­entries/­­closure​­epistemic/. Moretti, L.,  & Piazza, T. (­2018). ‘­Transmission and Transmission Failure in Epistemology.’ The Stanford Encyclopedia of Philosophy. Zalta (­ed.), URL = https://­plato.stanford.edu/­archives/­win2018/­entries/­­transmission-­​­­justification​­warrant/. 2016). Epistemic Angst. Princeton, NJ: Princeton University Pritchard, D. (­ Press. Silins, N. (­2005). ‘­Transmission Failure Failure’, Philosophical Studies, 126: ­71–​­102. Tucker, C. (­2010). ‘­W hen Transmission Fails.’ Philosophical Review, 4: ­497–​­529. Turri, J. (­2010). ‘­On the Relationship between Propositional and Doxastic Justification.’ Philosophical and Phenomenological Research, 80(­2): ­312–​­326. Silins, N. (­2005). ‘­Transmission Failure Failure.’ Philosophical Studies, 126: ­71–​­102. Williamson, T. (­2000). Knowledge and Its Limits. Oxford: Oxford University Press.

Intuitive Closure, Transmission Failure, Doxastic Justification  23 Wright, C. (­1991). ‘­S cepticism and Dreaming: Imploding the Demon.’ Mind, 100: ­87–​­116. ——— (­2000). ‘­Cogency and Q ­ uestion-​­Begging: Some Reflections on McKinsey’s Paradox and Moore’s Proof.’ Philosophical Issues, 10(­1): ­140–​­163. ———​­(­ 2002). ‘(­­ Anti-​­ )­ S ceptics Simple and Subtle: G. E. Moore and John McDowell.’ Philosophy and Phenomenological Research, 65(­2): ­330–​­348. ———​­(­2004). ‘­On Epistemic Entitlement: Warrant for Nothing (­and Foundations for Free?).’ Proceedings of the Aristotelian Society, Supplementary Vol (­78): ­167–​­212. ——— (­2008). ‘­­Internal-​­External: Doxastic Norms and the Defusing of Skeptical Paradox.’ The Journal of Philosophy, 105: ­501–​­517. ——— (­2012). ‘­Replies Part IV: Warrant Transmission and Entitlement.’ In Coliva, A. (­ed.), Mind, Meaning and Knowledge; Themes from the Philosophy of Crispin Wright (­p­­p. ­451–​­486). Oxford: Oxford University Press. ——— (­2014). ‘­On Epistemic Entitlement (­II): Welfare State Epistemology.’ In D. Dodd,  & E. Zardini (­eds.), Scepticism and Perceptual Justification (­p­­p. ­213–​­247). New York: Oxford University Press.

2 Modest Foundations for Perceptual Knowledge Krista Lawlor

Modest foundationalism holds that at least some of our perceptual beliefs are basic, in that they are not epistemically supported by further beliefs (­Lyons 2017). If one sees a squirrel on the bird feeder, then one may be justified in believing that there is a squirrel on the birdfeeder, without the need for further justified beliefs. In particular, one need not have further justified beliefs about the goodness of one’s evidence, the reliability of one’s senses, or the veridicality of one’s experience. Modest foundationalism accords with common sense. Because such “­evidential beliefs” are not required, unreflective or immature subjects count as justified in their perceptual beliefs.1 This is a good thing. But modest foundational views run into a p ­ roblem—​­on such modest foundations odd and ungainly epistemic edifices can sprout up. This general problem takes specific form in the problems of easy knowledge (­Cohen 2000, 2002) and bootstrapping (­Vogel 2008). If one can gain perceptually justified beliefs without further evidential beliefs, then what’s to stop one from going on to infer an evidential belief? If upon seeing a hand one is justified in believing that here is a hand, why not go on to conclude with Moore that that there is an external world to which one’s senses give one access, or given enough such perceptual encounters, that one’s senses are reliable? The question is how to construct a modest foundational account of justification for garden variety perceptual beliefs without creating an excess of justification for evidential or other “­ heavyweight” beliefs (­Dretske 2005). Let’s call this the problem of excess justification. The modest foundationalist about perceptual justification needs to avoid giving us too much of a good thing. In this chapter, I develop a commonsense account of how perceptual justification is transmitted across inferences which I hope will solve the modest foundationalist’s problem. In Section §1, I provide background about the problem of excess justification. What’s needed to block excess justification is a more nuanced understanding of how perceptual warrant transmits from the premises to the conclusion of an inference. In §­2 –​­5, I build an account of how the transmission of perceptual justification works. The account I offer blocks problematic inferences. Of course, DOI: 10.4324/9781003104766-3

Modest Foundations for Perceptual Knowledge  25 some people think there’s no such thing as too much of a good thing, and in §6, I consider some of their potential objections to my account.

1 Modest Foundationalism about Perceptual Warrant Let’s start with background about the problem of excess justification. The recent dialectic between Crispin Wright and Jim Pryor brings out the main issues we will be concerned with. Wright’s work on the transmission of perceptual warrant is concerned to avoid excess justification; to that end, he identifies a problematic sort of reasoning from ordinary perceptual beliefs. Here’s a case to illustrate:2 Zebra: ­Z ebra-​­1. [Visual experiences as of a zebra] ­Z ebra-​­2. That’s a zebra in the pen. ­Z ebra-​­3. That’s not a mule cleverly disguised as a zebra. This sort of reasoning strikes many as problematic. Intuitively, Z ­ ebra-​­3 is not justified from the perceptual belief Z ­ ebra-​­2. Wright’s diagnosis is that the reasoning in Zebra is problematic because ­Zebra-​­3 must be warrantedly assumed for ­Zebra-​­1 to count in favor of ­Z ebra-​­2. Given that this is so, warrant from one’s perceptual belief cannot be warrant for the conclusion Z ­ ebra-​­3. Following Jim Pryor, let’s call the problematic structure of justification exhibited by Zebra reasoning, “­Background Warrant structure” (­Pryor 2012, 273). By identifying this structure of justification, Wright addresses the problem of excess justification: Z ­ ebra-​ 3 ­ is not justified by the premises, and we can say why. Pryor (­2012), however, argues that not all reasoning from perceptual experiences to ambitious conclusions has this problematic structure. In particular, he claims that the following argument does not fail in the way Zebra does: Moore: ­Moore-​­1. [Visual experiences as of hands] ­Moore-​­2. I have two hands. ­Moore-​­3. I’m not a disembodied handless spirit, who is merely hallucinating hands. This might seem like a perfect example of excess justification: how can one get perceptual justification for M ­ oore-​­3? But according to Pryor, the reasoning in Moore is fine as it stands. He defends the reasoning in Moore in two steps. Step 1: Pryor begins by distinguishing mediate and immediate warrant: mediate warrant for a belief owes at least in part to warrant one has to believe something else, while immediate warrant does not owe

26  Krista Lawlor even in part to the warrant one has to believe something else (­2012, 275). Pryor plausibly claims that some perceptual beliefs have immediate warrant. (­They are warranted by perceptual experiences or, as Pryor has it, phenomenal seemings.) This ­view—​­that warrant for perceptual beliefs is gained without need of other warranting belief(­s)—​­just is modest foundationalism.3 Pryor notes that it follows from Step 1 that some perceptual beliefs are warranted in a way that that does not exhibit the problematic Background Warrant structure. If modest foundationalism is true, then some perceptual beliefs are immediately warranted (­2012, 276), and that in turn means some perceptual beliefs are justified without reasoning, so perforce, some perceptual beliefs are warranted in a way that that does not exhibit the problematic Background Warrant structure.4 This is all fine as far as it goes. But it is important to note that Pryor’s Step 1 only takes us so far in defense of the Moorean argument. Modest foundationalism secures that ­Moore-​­2 is warranted by ­Moore-​­1. Pryor also has to show that ­Moore-​­3 can licitly be arrived at. Here, it is not enough to show that Moore does not exhibit the problematic Background Warrant structure; we also have to see that Moore is good reasoning. After all, there could be some other problem with the justification exhibited by Moore. So, Pryor needs a second step in his defense of Moore. Pryor responds, “­I don’t think anyone has yet identified a pattern of ­transmission-​­failure that we have any consensus Moorean reasoning exemplifies” (­2012, 273). Pryor himself believes that the reasoning in Moore is good, and that some kind of cognitive illusion explains the discomfort people often feel with it.5 This response is of course not a positive defense of Moorean reasoning but an attempt to shift the burden of proof to those who find the reasoning in Moore problematic. The second step in Pryor’s defense of Moore, then, is to treat the inference from M ­ oore-​­2 to ­Moore-​­3 as innocent until proven guilty. If we accept Moore as a good argument, the resulting view is what Pryor calls “­Moorean dogmatism”—​­the idea that “­the Moorean reasoning is capable of transmitting warrant to ­Moore-​­3” (­2012, 276). According to the Moorean dogmatist, you can warrantedly believe you are not a disembodied handless spirit, who is merely hallucinating hands, on the strength of your visual experiences as of hands. I believe that this represents an implausible excess of j­ustification—​­just the sort of excess of justification that I hope to show the modest foundationalist is not committed to. Contra Pryor, I think we can identify a mistake in the reasoning in Moore. Moreover, I believe we can retain modest foundationalism about perceptual justification and identify the problem elsewhere in Moorean reasoning. In other words, Pryor’s step 1 is fine, but step 2 is a mistake. The task ahead is to see how we can retain modest foundationalism, while avoiding Moorean dogmatism. Before embarking on our task, a

Modest Foundations for Perceptual Knowledge  27 couple of notes. First, while Crispin Wright at one point rejected modest foundationalism about perceptual warrant, recently his view has evolved. He now accepts modest foundationalism for unreflective beliefs, while rejecting it for reflective ones: …it seems we think that the acquisition of basic perceptual knowledge, or other forms of perceptual warrant, doesn’t require the kind of epistemological ‘­policing’ of ­authenticity-​­conditions …. But it is different when we adopt a stance in which we undertake to scrutinize our claims to perceptual knowledge, or warrant. Then we seem obliged either to take a positive view of any a­ uthenticity-​­condition that may be entered into the conversation, or to qualify our claim to warrant. (­2014, 219 my emphasis) Since Wright endorses modest foundationalism for unreflective perceptual beliefs, he also faces the problem of excess justification.6 The problem is to explain how to adopt modest foundationalism without bringing Moorean dogmatism in its wake. In what follows, I will assume that our untutored reaction that there is something wrong with the reasoning in Moore is correct; there is an excess of justification in Moore. Some philosophers disagree, as we have seen, and hold that warrant does transmit from your perceptual belief to the claim that you are not the victim of global illusion; these philosophers account in other ways for the oddities of Moore.7 For the bulk of the chapter, I set Moorean dogmatist accounts aside, though I will return in Section §6 of the chapter to consider some potential dogmatist objections to my more modest view. How might one be a modest foundationalist about perceptual justification, while avoiding the excess of Moorean dogmatism? We need to say what goes wrong with the transmission of justification across such inferences as Moore. And to do that, we need a fuller account of how perceptual justification transmits or fails to transmit from perceptual beliefs to further conclusions.

2 Content The key to my account of how perceptual justification transmits begins with a story about c­ ontent—​­a story about what you see and what is justified by your visual experience. For my purposes, any account of content will do that allows us to more finely discriminate contents than is afforded by standard propositions. For ease of exposition, I will build on some insights of Fred Dretske’s ­contrastivist-​­style account of perceptual content. I emphasize that to tackle the problem we don’t need to be contrastivists per se, and

28  Krista Lawlor I am not here advocating for contrastivism.8 Any of several ­fi ne-​­grained accounts of perceptual contents may enable us to give a sufficiently nuanced account of how perceptual justification transmits. An example of Dretske’s (­1972) will get us started. Watching a soccer game, you see a blur of activity, and someone whom you take to be Clyde shoots and scores. Your attention is squarely on the ball. It is a near thing, but you carefully observe the ball’s trajectory, and see the shot is good. You infer that it was Clyde’s ­shot—​­you believe he is the forward, and it was the forward who had the ball. You say: If Clyde hadn’t made the last shot, we would’ve lost. Dretske notes that the information you convey with your statement is captured by including a contrast with relevant alternatives, a contrast we often bring out in speech by employing emphasis (­rising tone/­pitch): If Clyde hadn’t made the last shot [instead of bungling it, or its being blocked], we would’ve lost. In your epistemic circumstances, you would not make a statement with a focus on Clyde’s participation: If Clyde hadn’t made the last shot, we would’ve lost. In fact, this statement would not reflect your commitments at all. You do not mean to say that if Clyde (­not Fred or James) hadn’t made the last shot, we would’ve lost. In your m ­ ind—​­rightly ­enough—​­it’s the making of the shot that matters to the win, not whether it was Clyde who made the shot or not. The first point we notice is that one can have different sources of information and justification for distinct elements within the situation one sees. Looking at the situation on the field, you infer it was Clyde and visually discriminate that the shot was made. One might instead discriminate that it was Clyde and infer that the shot was made. (­Your friend Charley, sitting next to you at the game, might be focusing her attention on the person with the ball, working hard to discriminate whether it is Clyde or not. She might take your word for it that the shot was good or hear the roar of the crowd and say: Clyde made the last shot.) We already know that two subjects may be justified in believing the same proposition on different bases. I can see what you infer. The point here is that even when two people see the same situation and are both aptly described as seeing that P (­that Clyde made the last shot), they can have different epistemic justification for what they see. We can make all this a bit more precise, in terms of situations and relevant alternatives (­K ratzer 2002):

Modest Foundations for Perceptual Knowledge  29 Let s be a situation, a particular part of our world. We can think of propositions as sets of possible situations, whose job it is to classify situations, by describing some situations and not others, or if you will, being true of some situations and not true of others. Vision and other sense modalities give us epistemic access to situations. The general picture is this. If S sees that P, then S has a de re belief of a situation s, which situation is described by the proposition that P.9 Seeing that P puts one in a position to eliminate a set of alternative descriptions (­that Q, that R, …) of the situation s.10 If one’s visual experience puts one in a position to eliminate all the relevant alternatives to P, then one is in a position to know that P on the basis of seeing s. If one’s visual experience puts one in a position to eliminate only some of the relevant alternatives to P, then one has to eliminate the remainder by other means (­inference, testimony, memory) in order to be in a position to know that P on the basis of seeing s. What Dretske’s case shows is that we need to add some structure to this general picture. Dretske’s case shows that two subjects can see that P, their different visual experiences eliminating different sets of relevant alternatives to P, while relying on other means to eliminate the remainder. Your visual experience eliminates that the shot was bungled or blocked, and your background beliefs eliminate that it was anyone other than Clyde; your friend Charley’s visual experience eliminates that it was James or Fred, and testimony eliminates for her that the shot was missed. If these are all the relevant alternatives in the situation at hand, then you both know that Clyde made the shot. How elimination works is a matter for further discussion.11 For now, let’s say this. Your perceptual experience e of the situation s discriminates elements of s, in virtue of the elements looking a particular way. Your experience can be described in terms of the way these elements look: you have the visual experience as of a shot’s being made. We use the “­as of” formula to signal that your experience has qualitative content, which content discriminates, without being otherwise committal about the existence or identity and individuation conditions of the items that cause your experience. Your experience as of a shot’s being made is inconsistent with the alternative descriptions of the situation, that the shot was bungled or blocked, and so eliminates these alternatives. Your experience of as a shot’s being made is consistent with the description that the shot was made, so does not eliminate this alternative; in fact, if you have the requisite concepts, and all the relevant alternative descriptions are eliminated, we can use the proposition that the shot was made to characterize your perceptual belief. Let’s say that complete propositional justification is justification sufficient for being in a position to know P, should it remain undefeated. Then we can translate the above claim about being in a position to know that P into a claim about propositional justification: a visual experience

30  Krista Lawlor of s that puts one in a position to eliminate all the relevant alternatives not describable by P thereby equips one with complete propositional justification for P about s. An experience that puts one in a position to eliminate only some of the relevant alternatives not describable by P equips one with partial propositional justification for P about s.12 In Dretske’s case, the two subjects’ visual experiences provide (­distinct kinds of) partial justification for believing that Clyde made the shot.

3 Common Sense about the Transmission of Perceptual Justification Our next task is to use this machinery to examine how perceptual justification transmits across inferences. Back to Dretske again. Dretske notes that logically equivalent statements that have different focal alternatives will behave differently in inferences. For instance, while these statements are logically equivalent (­true in all the same worlds), [1] Clyde made the last shot [2] Clyde made the last shot the difference in their foci can result in a difference in truth conditions for statements that embed them. The statement schema [3] If [S], we would have lost is true or false, depending on which of [1] or [2] is embedded at [S] as the antecedent. It is true that [4] If Clyde hadn’t made the last shot, we would have lost because we needed that goal, but it’s false that [5] If Clyde hadn’t made the last shot, we would have lost because anyone’s making that goal would have given us the win. Embedding statements can see an effect on their truth conditions, owing to their sensitivity to focal alternatives. This fact has an effect on arguments from such statements. Reasoning from [4] as a premise may be cogent (­i.e. not fallacious) where reasoning from [5] is not, even if the validity of the argument is unaffected by the substitution of [4] by [5]. Suppose you reason as follows: [4] If Clyde hadn’t made the last shot, we would have lost We didn’t lose

Modest Foundations for Perceptual Knowledge  31 So, [2] Clyde made the last shot This argument has true premises and is valid. Suppose you reason instead: [4] If Clyde hadn’t made the last shot, we would have lost We didn’t lose So, [1] Clyde made the last shot This is bad reasoning, but the argument is not invalid. What’s wrong with this reasoning, says Dretske, is that “[a]nyone willing to accept the first premise as true does so on the understanding that Clyde was the fellow taking the shot. To elevate an assumption or presupposition made in the premises to the conclusion of one’s argument is, if you will, to commit a pragmatic fallacy” (­1972, 434). Dretske’s diagnosis of a fallacy, when he says that “­anyone accepting the first premise” [4] is “­presupposing” it was Clyde, does not require that in reasoning through the argument that’s the best epistemic position anyone could be in. Your friend Charley might accept the first premise but she isn’t stuck presupposing that it was ­Clyde—​­after all, she saw him kick the ball. What Dretske is getting at is this: if, as in our example, you have justification for the premise partly from perception and partly from another ­source—​ ­in our example, a background ­belief—​­then you cannot be justified in reasoning from this premise to this very background belief. Dretske’s case can be used to examine some features of the transmission of perceptual justification. Let’s start by thinking a bit more about what it is for justification to be transmitted when a person makes an inference. Here is a first pass: Where the premise and conclusion are empirical, transmission requires the epistemic work done by one’s evidence in justifying the premise(­s) should be the work that is capitalized on in justifying the conclusion. Now, I have characterized the work done by one’s evidence in terms of the elimination of relevant alternatives. So, transmission of justification occurs if the elimination of alternatives to the premise is capitalized on in justifying the c­ onclusion—​­that is to say, these alternatives should stay eliminated across the inference, and that this is so should be of some epistemic ­value—​­it should do some work in justifying the conclusion.13 This understanding accords with commonsense about good arguments. To illustrate, let’s think about a simple case, where you reason from [2] Clyde made the shot to [1] Clyde made the shot

32  Krista Lawlor You are justified in [2], because your visual experience eliminates the alternatives that the shot was bungled or blocked. But in order to have perceptual justification for [1] that Clyde made the shot, your perceptual experience must eliminate different alternatives, namely, that it was the doing of Fred, or James. Your perceptual experience does not eliminate these alternatives. So far as perceptual justification goes, then, there is no transmission. Again, transmission requires that the epistemic work done in justifying the premise is work that is capitalized on in justifying the conclusion. Let me digress a bit here to note that even if we don’t want to say that your visual experience provides justification for believing that Clyde made the shot, we might allow that your visual experience provides some epistemic support for this belief. Epistemic support is provided, since your belief that the shot was made could after all help put you in a position to believe that Clyde made the shot, were you to have additional beliefs. Here’s a suggestion about how to sustain the distinction we need. In our case, your visual experience as of the shot being made eliminates that it was bungled or blocked. Seeing the shot provides justification for your belief that the shot was made, since all the relevant alternatives to this proposition in the situation at hand are eliminated. Seeing the shot also provides partial justification for the belief that Clyde made the shot. Your visual experience eliminates some of the relevant alternatives to this belief. You have further support for believing that Clyde was the agent from a background belief, or testimony or what have you. So, what about the belief that Clyde made the shot? In this case, the epistemic role of visual experience is to justify the background belief that the shot was made. This belief is implied by the belief that Clyde made the shot. So, your visual experience supports the belief that Clyde made the ­shot—​­its role is to justify the background belief that the shot was made. But your visual experience doesn’t itself justify that Clyde made the shot. End digression. Setting complexities about epistemic support aside, the point we want to keep in view is just this: your perceptual justification for thinking that the shot was made does not transmit to the conclusion that it was Clyde who made the shot; perceptual justification transmits when one’s perceptual evidence makes possible the elimination of alternatives to the premise(­s), and the elimination of these alternatives is what justifies the conclusion. Let’s return to the transition that we’d like explained.

4 From Warrant Transmission to Transmission Failure We want a diagnosis of what goes wrong in Moorean reasoning. Recall Moore: ­Moore-​­1. [Visual experiences as of hands] ­Moore-​­2. I have two hands.

Modest Foundations for Perceptual Knowledge  33 ­Moore-​­3. I’m not a disembodied handless spirit, who is merely hallucinating hands. Here are some things we can now say about key moments in the argument: i You see your hands and believe I have two hands. Your perceptual experience eliminates such alternative descriptions of the situation before you as that these are prosthetics, or someone else’s hands. So, you have complete propositional justification for your perceptual belief in ­Moore-​­2 . ii You infer that I’m not a disembodied handless spirit, who is merely hallucinating hands. Whether justification transmits from ­Moore-​ 2 ­ to this claim depends on just what one is committed to by such a claim, which in turn depends on the relevant alternatives to the claim. It’s plausible that the thing one would want ruled out h ­ ere—​ a­ t least one relevant a­ lternative—​­is just that one is a disembodied spirit merely hallucinating.14 Let’s suppose this describes your situation, and that you take it that a relevant alternative to the inferred conclusion in this case is just that you are the victim of such an illusory experience. This new alternative is not eliminated by your perceptual experience. So justification from perceptual experience does not transmit to this conclusion. On this diagnosis, you have complete propositional justification for your perceptual belief that here are two hands, because your experience discriminates having two hands from the relevant alternatives. Your inference to the a­ nti-​­skeptical conclusion, however, does not transmit this j­ustification—​­you do not have perceptual justification for the a­ nti-​ s­keptical conclusion. What this means is that in the absence of other sources of justification for the a­ nti-​­skeptical conclusion, you have no justification for the belief that I’m not a disembodied handless spirit, who is merely hallucinating hands. I want to emphasize that you might have other sources of justification for this ­belief—​­for instance, coherence with other of your most firmly held metaphysical beliefs. But you will not have perceptual justification for this belief from the experience of seeing your hands. This fact does not owe to some undetected weakness in your perceptual justification, but to the demanding nature of the task of justifying the ­anti-​­skeptical conclusion.

5 Modest Foundationalism without Dogmatic Mooreanism Now, we’ve seen how an account of transmission of perceptual justification makes possible modest foundationalism without Mooreanism dogmatism. The view is modest foundationalist in holding that one does

34  Krista Lawlor not need justification for the denial of skeptical hypotheses in order to possess justification for one’s perceptual beliefs. And it rejects Moorean dogmatism, in holding that perceptual justification does not transmit to the denial of skeptical hypotheses. Let me briefly situate where the position stands in relation to other positions in recent debates about warrant transmission. The view is like Neta (­2004) in defending a sort of contextualism about perceptual knowledge, but it differs from Neta’s in that I suppose the evidence that one has is fixed, whether one considers the skeptical hypothesis or not. What changes with the introduction of the skeptical hypothesis is the amount of work one’s evidence has to do for one to be justified, I claim, not its status as evidence. The view also has some similarities to Silins (­2008). There are important differences, too: Silins affirms what he calls “­Rationalism” and he denies “­Strong Liberalism”; I have my doubts about “­Rationalism” and uphold “­Strong Liberalism.” Briefly, here’s what these disagreements are about. Silins holds that we are in some way or other justified in believing that we are not Brains in Vats (­hereafter, “­­not-​­BIV”). He labels the idea that one has justification for ­not-​­BIV, “­Rationalism.” He does not argue for Rationalism, but suggests some account can be given. On my account, Rationalism does not need to be true, while Silins is committed to it, since he accepts unrestricted justification closure (­2008, 128). I reject unrestricted justification closure. (­More about this below.) Silins’ explanation of why justification does not transmit in the Moorean inference is probabilistic. The core idea is that one’s perceptual experience as of a hand lowers the probability of ­not-​­BIV, and so perceptual experience cannot justify the proposition ­not-​­BIV. There are limits, as he himself notes, to what the probabilistic argument can show. It cannot show that justification fails to transmit to the belief that there is an external world, because the denial of this proposition does not explain our having the experiences we do (­­p.  127). So, the limits of the probabilistic account join with his acceptance of unrestricted justification closure and result in his being committed to our being justified in this belief, if we are perceptually justified in ordinary perceptual beliefs. On my view, we need not be committed to having justification for this belief, or any other “­metaphysically heavyweight” proposition, whenever we are justified in ordinary perceptual beliefs.15

6 Objections Moorean dogmatists hold that perceptual justification for ­garden-​­variety perceptual beliefs transmits to the denial of a global skeptical hypothesis (­e.g. that I am not now the victim of global illusion). Here are a few objections to my view that Moorean dogmatists might make, along with my replies.

Modest Foundations for Perceptual Knowledge  35 6.1  Closure of Justification I have resisted the claim that justification for your perceptual belief (­here’s a hand) transmits to the denial of skeptical hypotheses (­I am not the victim of globally illusory experience). I have also argued that one need not have antecedent justification for the denial of such skeptical hypotheses, before one can have justification for one’s perceptual belief. So, the view I defend tolerates the failure of closure of justification under entailment: CJ: [J(­φ) & (­φ ⇒ ψ)]→ J(­ψ) Read “­φ ⇒ ψ” as “­φ deductively entails ψ,” and so ψ is an a priori knowable consequence of φ. Now, some will argue for Moorean dogmatism this way: we should tolerate no failures of closure principles such as ­CJ—​­they are “­intuitively compelling.” But if CJ holds, and it is also true that ordinary perceptual beliefs entail the denial of skeptical hypotheses, then one is only justified in ordinary perceptual beliefs if one is also justified in the denial of skeptical hypotheses. A paradox! Better to accept Moorean dogmatism than to (­absurdly) deny CJ or to (­equally absurdly) deny that we have justification for ordinary perceptual beliefs. My reply is this. There are more ways out of the skeptical paradox driven by CJ. Our choice is not between either (­i) accepting CJ and being unhappy dogmatists16 or (­ii) denying an intuitively compelling epistemic principle. A third option is to accept a more carefully stated closure principle: CJ*: [J(­φ) & (­φ ⇒ ψ) & Restriction(­φ, ψ)] → J(­ψ) For example, we might restrict closure by requiring that the consequent, ψ, introduces no new relevant alternatives than φ introduces.17 In good cases, where the conclusion introduces no new alternatives and CJ* is satisfied, justification transmits from premise to conclusion. In bad cases, CJ* is not satisfied and justification fails to transmit from premise to conclusion. Here is a good case: you see a blue cube on your desk, and your experience eliminates relevant alternatives (­say, that it is a sphere or pyramid, that it is red or green); having no reason to consider other defeating hypotheses, you are justified in believing that it is a blue cube. You reason from it’s a blue cube to the conclusion it’s a cube. You are justified in this conclusion. Here is a bad case: you see a blue cube, believe that it is a cube and reason from its being a cube to your not being presented with a hologram of a cube. What motivates CJ*? A restricted closure principle such as CJ* does not generate the skeptical paradox of justification, alluded to above.

36  Krista Lawlor Moreover, we can save the intuition that some such closure principles apply: whatever intuitive appeal CJ has derives from good cases, which cases support CJ* as well. These facts motivate thinking CJ* superior to CJ. Finally, it’s worth remembering that in cases where ψ is an empirical proposition (­it’s not a zebra, but a mule disguised to look like a zebra), justification requires you have pertinent evidence in order for the proposition to be justified. CJ* enforces this requirement.18 With closure principles available that have no dogmatic upshot, the ­closure-​­based argument for Moorean dogmatism is very weak. 6.2  Evidential Power of Perception A Moorean dogmatist might say: (­i) we grant that with the inference to the denial of the skeptical hypothesis, more alternatives are relevant, but also (­ii) one’s evidence is sufficient to eliminate these skeptical alternatives. One’s perceptual experience of one’s hands contains in it that which eliminates the proposition that one is the victim of global illusion. The first reply to the dogmatist is simply to ask how one’s perceptual experience as of hands could possibly eliminate the skeptical alternative that what is before one is only an illusory hand. Here the dogmatist might appeal to disjunctivism and deny that one’s experience as of hands is the totality of one’s evidence. Rather, one’s perceptual evidence is one’s perceptual experience plus the fact that here are hands. Thus, the totality of one’s perceptual evidence does not match the skeptical alternative19, so the alternative is eliminated. Consequently, one’s perceptual evidence puts one in a position to know that one is not the victim of global illusion. I haven’t space to go into the difficulties that disjunctivism brings in its wake. There are sizable difficulties arising from the idea of expanding one’s evidence so that one’s evidence is external to one’s epistemic perspective.20 At the very least, I note that this gambit drives up the cost of being a Moorean dogmatist. 6.3  Common Sense as Moorean? Adam Leite (­2015) argues as follows: … I have hands and a body and am sitting at a desk typing. Moreover, there aren’t any brains in ­vats – ​­virtual reality programming is not yet that good. These considerations are decisive evidence that I am not a BIV; in fact they entail that I am not. … In ordinary life, we would not hesitate for a moment to regard the presence of hands and computers as evidence for all sorts of things. Considerations about the world are routinely counted as evidence. And the crucial point here is that such considerations eliminate

Modest Foundations for Perceptual Knowledge  37 precisely the alternatives … relevant to the question of whether one knows that one is not a BIV, etc. In the first paragraph quoted, Leite seems to have in mind the idea that ­non-​­perceptual grounds help to establish that I am not a BIV. One does not just see that I am not a BIV, according to Leite, but one’s justification for this claim rests on additional beliefs about the limits of technology, etc. For all I have said in this chapter, Leite may be correct about the ­anti-​­skeptical argument to be made on the basis of such commonsense beliefs. In this chapter, I am concerned with perceptual justification for the ­anti-​­skeptical conclusion, of the kind that Moore claims to have. With that concern in mind, let’s focus instead on the idea in the second quoted paragraph, namely, the “­crucial point” that “­the presence of hands eliminates precisely the alternatives relevant to whether one knows that one is not a BIV.” This is an idea which a Moorean dogmatist might happily seize on. In reply, I suggest that there are two ways it could work out that our perceptual evidence is sufficient to eliminate the relevant alternatives to the ­anti-​­skeptical conclusion. (­i) We could say that the relevant alternatives expand upon making the inference to the denial of the skeptical hypothesis, but also claim that one’s evidence is nonetheless sufficient to eliminate these newly relevant skeptical alternatives. I’ve considered this option just above (­under “­Evidential power of perception”), so let me consider the second way justification could transmit. (­ii) We could instead say that the only relevant alternatives to the a­ nti-​­skeptical conclusion are the same alternatives that are relevant for one’s ordinary beliefs. If both of these claims I am not the victim of global illusion I have a computer in front of me, I have hands and a body, … have the same relevant alternatives, then Leite is correct that warrant transmits from the latter to the former. I believe this way of defending Moorean dogmatism is implausible for two reasons. First, we have as yet no reason to suppose that relevant alternatives for the conclusion of Moore are the same alternatives relevant for the premises. (­Compare, if all the relevant alternatives to the claim that I am ­demon-​­deceived and merely dreaming all of this are just the same g­ arden-​­variety alternatives one eliminates when one is justified in believing one hasn’t taken any sleeping pills, then one can ­ emon-​­deceived and merely dreaming be justified in believing one is not d all of this, merely by remembering one hasn’t consumed anything this morning. But what reason is there for thinking the same alternatives are relevant to each claim?) Second, if we were to stipulate that what count as relevant alternatives is fixed by the premises, that would ensure the epistemic work done in justifying premises would always be sufficient to justify conclusions. But the result would be that far too many inferences count as ­warrant-​­transmitting. This way leads to an explosion of

38  Krista Lawlor justification, where one counts as justified in many conclusions on very little evidence. 21 It’s hard to make good on Leite’s claim that “­the presence of hands eliminates precisely the alternatives relevant to whether one knows that one is not a BIV.”

7 Resisting Moorean Dogmatism, Saving Modest Foundationalism Modest foundationalism delivers justification for our ­ garden-​­ variety perceptual beliefs. Modest foundationalism does not commit us to Moorean dogmatism. In resisting Moorean dogmatism, we acknowledge the limits of our perceptual justification. Mooreanism dogmatism is blocked by commonsense features of the transmission of perceptual justification across an inference. Transmission of perceptual justification requires that the epistemic work done by one’s evidence in justifying the premise(­s) be the work that is capitalized on in justifying the conclusion. If we stick to this understanding of transmission, we don’t suffer an excess of justification for overly ambitious or heavyweight conclusions.

Notes 1 Foundationalist will argue about what supports one’s basic beliefs. Internalists insist that the support be internal to the perceiving ­subject—​­perhaps in the form of an experience or a perceptual “­seeming”; externalists require something more than experience to support perceptual ­belief—​­such as reliable experience. (­S ee citations in Lyons, ibid.) 2 See Wright (­2004, 2014). This framing of the case is a ­hat-​­tip to Fred Dretske’s (­1970, 2005) work on epistemic closure. 3 Pryor calls the view that some perceptual beliefs are immediately warranted “­dogmatism about perception” (­2012, 276). 4 Could a similar response be made in defense of Zebra? Possibly. For the sake of argument, Pryor is prepared to grant that Zebra is an instance of Background Warrant failure. 5 While “­there are some respects in which these arguments are persuasively crippled … [i]n terms of their justificatory structure, though, I think these arguments have nothing to be ashamed of” (­Pryor 2004, 362). 6 What should be the response to such scrutiny? Wright’s response is “­conservative”—​­that is, one must make a positive epistemic defense of what one takes to be the case. One must have something positive to say in favor of taking it that your visual system is functioning satisfactorily, or even, that one is not the victim of global illusion. With respect to the latter, given the difficulty of imagining how to defend oneself, the idea of entitlement comes into its own: The promise of the notion of epistemic entitlement is that it offers to return a positive response to the question: How can it be epistemically rational to repose confidence in an ­authenticity-​­condition for whose satisfaction one has no evidence or other form of cognitive warrant? (­2014, 221)

Modest Foundations for Perceptual Knowledge  39 I am moved by Wright’s argument that if we are to persist in taking ourselves to be justified in our perceptual beliefs, we should have something to say in response to prima facie reasonable scrutiny, even if scrutiny extends to raising questions about one’s presuppositions. However, I think we have more options than finding something positive to say in favor of the epistemic credentials of one’s presuppositions. As a reasonable person, one may be obliged to consider questions about the reliability of one’s senses, or about whether or not one is the victim of global illusion. In response, however, one need not produce any sort of positive epistemic grounds for taking one’s senses to be reliable. One might instead assert the reasonableness of ignoring the alternative. The skeptical interlocutor, if reasonable, is under symmetrical constraints to be schooled in the matter of what to worry about and when. 7 For instance, one might hold that one is justified in the belief that one is not the victim of illusion, but considering the argument, doubt enters one’s mind and so one’s doxastic justification is destroyed. (­See Pryor ibid; Alston 1986; Silins 2005). Neta (­2007) argues just the ­opposite—​­that justification doesn’t transmit, but rational conviction does. The literature on warrant transmission is large. See the essays in Nuccetelli (­2003), Moretti and Piazza (­2013), and Tucker (­2012). 8 I choose Dretske’s contrastivist for ease of exposition. We do not need all of contrastivism’s (­S chaffer 2007) commitments (­for example, knowledge as irreducibly t­ hree-​­place relation) in order to solve the problem for modest foundationalism. 9 This is what Dretske (­1969) calls “­epistemic seeing,” which requires that one possess the concepts requisite for believing the proposition that P. 10 Note that Kratzer would say that s “­exemplifies P” which involves s being the minimal situation describable by P. I set the issues here to one side. 11 Lewis (­1996) provides a story that works in his possible worlds framework: the experience e I have now does not match the experience I would have in the world w*, so e eliminates w*. Given the ­situation-​­based account I favor, this idea of matching experiences in worlds won’t do as an account of elimination. Thanks to ­A nna-​­Sara Malmgren for discussion. 12 Going forward, I’ll suppress “­about s” and “­propositional” but should be understood to mean propositional justification throughout unless otherwise noted. Propositional (­as opposed to doxastic) justification is roughly justification one’s belief would have were one to form a belief on its basis. 13 We might instead say, no new alternatives are relevant to the conclusion that are not eliminated already in light of the grounds for the premise(­s). This is a more permissive view of warrant transmission. The idea here would be that what grounds the premises is sufficient to ground the conclusion, whether or not those grounds are made use of in exactly the same way. I favor the slightly more demanding notion of warrant transmission. 14 It is not necessary that the negation of one’s claim is a relevant alternative to the claim (­Holliday 2013), though often it is. Also, it is important that on my proposal relevance is not given a psychological construal, in terms say of salience or the like. If it were, the worry (­voiced by Wright 2012, 459) arises that this introduces a kind of relativity into the assessment of every argument in such a way as to “­obliterate the distinction between transmissive and ­non-​­transmissive arguments,” unless one could say more about the sorts of salience one might countenance, and what epistemic constraints on it might be. 15 Note that Schaffer (­2004) develops a contrastivist account similar to the position I am arguing for in this chapter. Schaffer shows how, by using a

40  Krista Lawlor relevant alternatives account of knowledge, we can say Moore knows he has hands as opposed to stumps, say, but that Moore does not know he has hands as opposed to ­hand-​­images. The difference between us is this: Schaffer’s focus is knowledge closure and not warrant transmission, and his principles are built to capture ­non-­​­­justification-​­related features of knowledge, some of which, interestingly, only show themselves in a contrastivist analysis. (­For instance consider Schaffer’s principle C ­ ontract-​­q: (­Kspq1 & (­q2 → q1) & {q2} ≠ 0) → Espq2). Roughly glossed: knowledge is closed as long as the contrast proposition does not expand (­take in more logical space), and if it contracts, the proposition does not become the empty set or null proposition. What makes Schaffer’s principle intuitively compelling in part is the idea that if the contrast proposition contracts to the null proposition, one no longer has a genuine question on one’s hands. This is the sort of case where one might expect knowledge to be conserved, Schaffer argues, but no part of his argument for this claim concerns the transmission of warrant under these conditions.) In short, contrastivism about knowledge is silent about what happens when a person performs a deduction or thinks through the inference from one claim to the other. What can be expected from inference as such, as an epistemic process? How does it work in the good case where warrant does transmit and fail to work in the bad case where it doesn’t? For this, one needs an account of warrant transmission. 16 Schiffer (­2004, 182) is an unhappy dogmatist; he holds that the concept of justification is a bit of a mess, and “­there is nothing for it but to count our disbeliefs in skeptical hypotheses as justified and be done with it.” Schiffer also thinks that modest foundationalism is ­unmotivated—​­the idea that one does not need justification for the denial of skeptical hypotheses, when those hypotheses are defeaters of one’s perceptual belief, is not acceptable to him (­2004, 180). 17 Recall, I am not advocating contrastivism in this chapter. One might instead restrict closure to cases where no new subject matter is introduced. See Lewis (­1988), Gemes (­1994), Yablo (­2017), Fine (­2016). Restricting closure principles seems to be required for empirical knowledge; I make no claims about whether such restrictions have a place in other epistemic domains, such as mathematical knowledge. 18 McGlynn (­2014, 180) argues that these two “­compelling” principles ClosOR: If one has justification for P and one knows that P entails P or ­Q —​­where the ‘­or’ is read here and throughout as ­inclusive—​­then one has justification for P or Q Equivalence principle: if one knows a priori that P and Q are equivalent and one has justification for P, then one has justification for Q. ensure the following: Let P be that here are hands, and Q that there is an external material world, and grant that Q is entailed by P. Q is equivalent to P or Q. So, by ClosOr, if one has justification for P, one has justification for P or Q, and by Equivalence, one also has justification for Q. If one has perceptual justification for here are hands, then one also has justification for there is an external material world. Wright is moved by McGlynn’s argument, and so entertains a position combining (­a) liberalism (­i.e. modest foundationalism) about authenticity conditions for ­fi rst-​­order perceptual belief (­Wright 2012, ­473–​­4), as well as (­b) transmission of warrant from such beliefs to a­ nti-​­skeptical conclusions (­Wright 2014, 233). 19 Note that the envisioned disjunctivist reply only handles “­­world-​­side skepticism.” That is, skepticism generated by a hypothesis about whether the world is as it presents itself as being. Skepticism also arises with hypotheses

Modest Foundations for Perceptual Knowledge  41 about my experience, and whether it is not veridical, despite the presence of a world of things external to my mind. See Austin (­1946, 158). 20 For a good compact discussion of these difficulties, see Schiffer (­2009). 21 For a related idea problem about “­knowledge explosion,” see Klein (­2000, 220) and criticism by Cohen (­1999) and Brueckner (­1998).

Works Cited Alston, William P. 1986. “­Epistemic Circularity.” Philosophy and Phenomenological Research 47: ­1–​­39. Austin, J. L. 1946. “­Other Minds.” Supplement to the Proceedings of the Aristotelian Society 20: 1 ­ 48–​­87. Brueckner, Anthony. 1998. “­K lein on Closure and Skepticism.” Philosophical Studies: An International Journal for Philosophy in the Analytic Tradition 98: ­139–​­51. Cohen, Stewart. 1999. “­ Contextualism, Skepticism, and the Structure of ­Reasons.” Noûs 33: ­57–​­89. —​­—​­—.​­​­ 2000. “­Contextualism and Skepticism.” In Skepticism, edited by E. Sosa and E. Villanueva, ­94–​­107. Boston, MA: Blackwell. —​­—​­—​­. 2002. “­Basic Knowledge and the Problem of Easy Knowledge.” Philosophy and Phenomenological Research 65 (­2): ­309–​­29. https://­doi.org/­10.1111/­ j.­1933-​­1592.2002.tb00204.x. Dretske, Fred. 1969. Seeing and Knowing. Chicago, IL: University of Chicago Press. —​­—​­—​­. 1970. “­Epistemic Operators.” Journal of Philosophy 67: ­1007–​­23. —​­—​­—.​­​­ 1972. “­Contrastive Statements.” Philosophical Review 81 (­4): ­411–​­37. —​­—​­—​­. 2005. “­T he Case against Closure.” In Contemporary Debates in Epistemology, edited by M. Steup and E. Sosa, 1 ­ 3–​­25. Oxford: Blackwell. Fine, Kit. 2016. “­A ngellic Content.” Journal of Philosophical Logic 45 (­2): ­199–​ ­226. https://­doi.org/­10.1007/­­s10992-­​­­015-­​­­9371-​­9. Gemes, Ken. 1994. “­A New Theory of Content I: Basic Content.” Journal of Philosophical Logic 23 (­6): ­595–​­620. Holliday, Wes. 2013. Knowing What Follows: Epistemic Closure and Epistemic Logic. Amsterdam: ILLC Publications. Klein, Peter. 2000. “­Contextualism and the Real Nature of Academic Skepticism.” In Skepticism, edited by E. Sosa and E. Villanueva, ­108–​­116. Boston, MA: Blackwell. Kratzer, Angelika. 2002. “­Facts: Particulars or Information Units?” Linguistics and Philosophy 25 (­5/­6): ­655–​­70. Leite, Adam. 2015. “­W hy Don’t I Know I’m Not a Brain in a Vat?” Philosophy and Phenomenological Research 90 (­11): ­205–​­13. Lewis, David. 1988. “­Relevant Implication.” Theoria 54 (­3): ­161–​­74. —​­—— ​­ .​­​­ 1996. “­Elusive Knowledge.” Australian Journal of Philosophy 74: ­549–​­67. Lyons, Jack. 2017. “­ Epistemological Problems of Perception.” In Stanplato.stanford. ford Encyclopedia of Philosophy, Spring 2017. https://­ edu/­archives/­spr2017/­entries/­­perception- ​­episprob/. McGlynn, Aidan. 2014. “­On Epistemic Alchemy.” In Scepticism and Perceptual Justification, edited by D. Dodd and E. Zardini, ­173–​­88. Oxford: Oxford University Press.

42  Krista Lawlor Moretti, Luca and Tommaso Piazza. 2013. “­Transmission of Justification and Warrant.” In Stanford Encyclopedia of Philosophy, Winter 2013. https://­plato. stanford.edu/­archives/­win2013/­entries/­­transmission-­​­­justification-​­warrant/. Neta, Ram. 2004. “­Perceptual Evidence and the New Dogmatism.” Philosophical Studies 119 (­1/­2): ­199–​­214. —​­—​­—​­. 2007. “­Fixing the Transmission: The New Mooreans.” In Themes from G. E. Moore: New Essays in Epistemology and Ethics, edited by Susana Nuccetelli and Gary Seay, ­62–​­83 Oxford: Clarendon Press. Nuccetelli, Susana, ed. 2003. New Essays on Semantic Externalism and S­ elf-​ ­Knowledge. Cambridge: MIT Press. Pryor, James. 2004. “­W hat’s Wrong with Moore’s Argument?” Philosophical Issues 14 (­Epistemology): ­349–​­78. Schaffer, Jonathan. 2004. “­From Contextualism to Contrastivism.” Philosophical Studies: An International Journal for Philosophy in the Analytic Tradition 119 (­1/­2): 7 ­ 3–​­103. —​­—​­—​­. 2007. “­Closure, Contrast, Answer.” Philosophical Studies 133: ­233–​­55. Schiffer, Stephen. 2004. “­Skepticism and the Vagaries of Justified Belief.” Philosophical Studies 119 (­­1–​­2): ­161–​­84. —​­—​­—​­. 2009. “­Evidence = Knowledge: Williamson’s Solution to Skepticism.” In Williamson on Knowledge, edited by Greenough, P and Pritchard, D., 5 ­ –​­11. Oxford: Oxford University Press. Silins, Nicholas. 2005. “­Transmission Failure Failure.” Philosophical Studies 126: ­71–​­102. —​­—​­—​­. 2008. “­Basic Justification and the Moorean Response to the Skeptic.” In Oxford Studies in Epistemology, edited by Tamar Gendler and John Hawthorne, 2:­108–​­42. Oxford: Oxford University Press. Tucker, Chris. 2012. “­Transmission and Transmission Failure in Epistemology.” In Internet Encyclopedia of Philosophy. http://­w ww.iep.utm.edu/­transmis/. Vogel, Jonathan. 2008. “­Epistemic Bootstrapping.” Journal of Philosophy 105 (­9): ­518–​­539. Wright, Crispin. 2004. “­Warrant for Nothing (­and Foundations for Free)?” Aristotelian Society Supplement 78: ­167–​­212. —​­—​­—​­. 2012. “­Replies.” In Mind, Meaning  & Knowledge: Themes from the Philosophy of Crispin Wright, edited by Annalisa Coliva, 3 ­ 77–​­486. Oxford: Oxford University Press. —​­—​­—​­. 2014. “­On Epistemic Entitlement (­I I): Welfare State Epistemology.” In Scepticism and Perceptual Justification, edited by D. Dodd and E. Zardini, ­213–​­47. Oxford: Oxford University Press. Yablo, Stephen. 2017. “­Precis of Aboutness.” Philosophical Studies 174 (­3): ­771–​­77.

3 Closure, Warrant Transmission, and Defeat Mona Simion

1 Introduction* According to what I will dub ‘­ Concessive’ ­ Neo-​­ Mooreanism, warrant transmits in M ­ oorean-​­style a­ nti-​­sceptical inferences; however, the sceptic is rationally obscured from coming to believe the conclusion of Moore’s argument in virtue of psychological ­higher-​­order defeat (­e.g. Pryor 2004, 2012). In contrast, ‘­Radical’ ­Neo-​­Mooreanism (­e.g. Williamson 2000, 2007) claims that warrant transmits in the Moorean inference, closure for knowledge holds unrestrictedly, and nothing is wrong with the Moorean argument, dialectically or otherwise; furthermore, Radical ­Neo-​­Mooreanism charges the sceptic with epistemic malfunction for resisting Moore’s argument.1 This chapter argues that both Concessive and Radical ­ Neo-​ ­Mooreanism remain unsatisfactory and develops a novel N ­ eo-​­Moorean view. The view falls squarely within the Radical ­Neo-​­Moorean camp, in that it holds that closure holds unrestrictedly, warrant transmits through Moore’s inference, and that there is nothing w ­ rong  – ​­epistemically or ­dialectically – ​­with Moore’s argument. Nevertheless, the account is superior to extant Radical ­Neo-​­Mooreanisms in explanatory power: it explains both the precise variety of epistemic failure exhibited by the sceptic and the intuition of reasonableness when it comes to the sceptic’s resistance to Moore’s argument. It does so in terms of epistemic functions and c­ ontrary-­​­­to-​­duty obligations. In Section 2, I briefly outline the issue at stake and the N ­ eo-​­Moorean responses that I deal with. In Section 3, I discuss Concessive N ­ eo-​ M ­ ooreanism and reject the defeat claim. In Section 4, I develop my own Radical ­Neo-​­Mooreanism, and in Section 5, I conclude.

2 Two ­Neo-​­Mooreanisms Moore (­1939) sees his hands in front of him and comes to believe that HANDS: ‘­Hands exist’ based on his extraordinarily reliable perceptual belief formation processes. Moore’s belief is warranted, if any beliefs

DOI: 10.4324/9781003104766-4

44  Mona Simion are: Moore is an excellent believer. Indeed, Moore knows that hands exist. In spite of his laudable epistemic ways, Dretske (­1970) thinks Moore shouldn’t feel free to do whatever it pleases him to do with this belief, epistemically speaking; in particular, Dretske thinks that, in spite of his warranted belief that HANDS, Moore should refrain from reasoning to some propositions he knows to be entailed by HANDS, such as, for instance, WORLD: ‘­There is an external world.’ He thinks that this is an instance of closure failure for knowledge: we don’t always know the stuff that we know our knowledge to entail. In better news, conversely, that’s why the sceptic is wrong to think that my not knowing that I’m not a brain in a vat implies that I don’t know any of the ordinary things I take myself to know. 2002, 2003, 2004) agrees: Moore shouldn’t reason to Wright (­ WORLD from HANDS. However, that’s not because closure fails, but because the stronger principle of warrant transmission fails: the problem here, according to him, is not that we sometimes fail to know the stuff that we know is entailed by what we know. Rather, the issue is that the warrant Moore has for HANDS fails to transmit to WORLD. Compatibly, though, Moore may still be entitled to believe WORLD on independent grounds. If Moore is entitled to believe HANDS, then perhaps he must also be entitled to believe WORLD. But it doesn’t follow that his warrant to believe WORLD is his warrant to believe HANDS. Rather, it may be that Moore needs to be independently entitled to believe WORLD to begin with, if he is to be entitled to believe HANDS. Many philosophers are on board with rejecting at least one of these ­principles – ​­be it merely warrant transmission or closure as well. At the same time, since closure and warrant transmission constitute a bedrock of our epistemic ­ways – indeed, ​­ crucial vehicles for expanding our body of ­knowledge – ​­one cannot give them up without a working restriction recipe: if closure and/­or warrant transmission don’t hold unrestrictedly, when do they hold? It is fair to say the jury is still out on this front, and a satisfactory restriction recipe does not seem to be within easy reach. 2 That being said, several philosophers take the alternative route of resisting the failure claims altogether, and thus fully dismiss the data: according to them, closure and warrant transmission are too important a theoretical tool to be abandoned on grounds of misguided intuitions. They reject the intuition that something fishy is going on in Moore’s argument and argue that scepticism is just an instance of cognitive malfunction: the sceptic’s cognitive system malfunctions in that it fails to get rid of her unjustified sceptical beliefs in favour of the justified Moorean conclusion. I call these people ‘­Radical ­Neo-​­Mooreans.’ Here is Williamson: Our cognitive immunity system should be able to destroy bad old beliefs, not just prevent the influx of bad new ones. But that ability sometimes becomes indiscriminate, and destroys good beliefs too. (­2007: 681)

Closure, Warrant Transmission, and Defeat  45 I like Radical N ­ eo-​­Mooreanism. The majority reaction to this move, however, is that it is less than fair to the sceptic; indeed, this view (­intuitively unfairly) categorizes scepticism, without qualification, in the same normative boat with other epistemic malfunctions, such as, for instance, wishful thinking. It is undeniable, though, that in the case of the sceptic, but not in the case of the wishful thinker, we think that there is something r­ easonable – even ​­ if not quite r­ ight – ​­about their resistance to Moore’s argument. This intuitive difference cries for an explanation. At the other side of the ­Neo-​­Moorean spectrum, we find concessive ­Neo-​­Mooreans (­e.g. Pryor 2004, 2012); these philosophers accept both closure and transmission in Moorean inferences and try to come up with alternative explanations of the data: i.e. with an alternative account of what is intuitively amiss with Moore’s argument. In the next section, I look closer at the Concessive N ­ eo-​­Moorean explanation of this datum.

3 Against Concessive ­Neo-​­Mooreanism According to Jim Pryor (­2004), while Moore is right to reason from HANDS to WORLD, he wouldn’t be very convincing were he to do so in conversation with a sceptic. The problem behind the intuitive fishiness of Moore’s reasoning pattern is pragmatic, not epistemic: It is lack of dialectical force, not lack of warrant transmission, that’s triggering the uneasiness intuition. In the cases of alleged failure of closure and/­or transmission, warrant transmits, but the argument fails dialectically due to psychological ­higher-​­order defeat.3 The sceptic about WORLD will not be convinced by Moore’s argument in its favour from HANDS. Here is Pryor: For a philosopher with such beliefs [i.e. sceptical beliefs], it’d be epistemically defective to believe things just on the basis of her ­experiences—​­even if those experiences are in fact giving her categorical warrant to so believe. (­2012, 286) Why would it be thus epistemically defective? According to Pryor, the sceptic’s unjustified sceptical beliefs rationally obstruct her from believing based on Moore’s argument, via psychological defeat. In particular, Pryor thinks that Moore’s argument gives the sceptic propositional justification for the conclusion, but it fails to generate doxastic justification, due to the psychological defeat generated by the sceptic’s previously acquired sceptical beliefs. Since the sceptical beliefs are not justified, according to Pryor, they don’t defeat the propositional justification generated by Moore’s argument. They do, however, rationally obstruct the sceptic from justifiably believing the conclusion of Moore’s argument, and in this they defeat the sceptic’s doxastic justification.

46  Mona Simion The point then, in a nutshell, is that even though it transmits warrant, the Moorean argument fails to convince the rational sceptic in virtue of the conflict between the Moorean claims and the sceptic’s previously held beliefs. The sceptic has propositional justification, but does not have doxastic justification, for HANDS and WORLD. In what follows, I will take issue with this claim at several junctures. First and foremost, though, it is worth clarifying what exactly the content of the sceptical beliefs that allegedly do the defeating work here is. I want to start off by noting that it is implausible to think that the sceptical belief at stake in the literature is (­or should be) something like ­non-​­WORLD: ‘­The external world does not exist.’ After all, what we are talking ­about –​ ­and the philosopher that is worth engaging ­with – ​­is a reasonable sceptic, who e.g. believes in ­underdetermination – ​­i.e. thinks that, for all she knows, she may well be a brain in a vat, not someone who is anxiously fully confident that they’re a brain in a vat. The reasonable sceptic that is worth engaging with thinks that, for all the evidence she has supports, there may well be no external world. If so, the reasonable sceptic will, at best, have a.5 credence that ­Non-​­WORLD, or else she will suspend belief on the issue. Not much will hang on this below, but since I am interested in being maximally charitable to Concessive N ­ eo-​­Mooreanism, I will, for the most part, discuss the reasonable sceptic rather than the maximally anxious sceptic in what follows. Everything I will say, though, will apply mutatis mutandis to the anxious sceptic as well. Now here is a widely endorsed thesis in philosophy: justification is normative. The following is an attractive way of capturing this thought: One’s ­ϕ -​­ing is prima facie practically, morally, epistemically, etc. justified if and only if one prima facie practically, morally, epistemically, etc. permissibly ϕs. Plausibly enough, then, one’s belief that p is epistemically justified if and only if one epistemically permissibly believes that p. Justifiers are considerations that support belief, in that, if all else goes ­well – ​­i.e. proper basing, no defeat, good processing, etc. – enough ​­ justifiers render a belief epistemically permissible. Where does defeat fit in this picture? Just like justification, defeat is a normative category, in that it affects the permissibility of belief. Unlike justification, however, its function is to counter rather than sup­ elief – they ​­ contribute to rendering port believing. If justifiers support b it ­permissible – ​­defeaters contribute to rendering it impermissible. It is plausible, then, to think that defeat is the archenemy of justification: if justification is normative with a positive ­valence – ​­in that it renders belief ­permissible  – ​­(­full) defeat is normative with a negative valence, in rendering belief impermissible. In ­reasons-​­talk, if you wish, justifiers are normative reasons for belief, while defeaters are normative reasons against believing. Now, let’s go back to Pryor’s account of what goes on in the exchange between Moore and the sceptic. Recall: according to Pryor, even though

Closure, Warrant Transmission, and Defeat  47 Moore’s argument does provide the sceptic with propositional justification, it fails to provide her with doxastic justification, in virtue of her unjustified sceptical beliefs defeating the latter, but not the former. As such, according to Pryor, the sceptic’s belief that HANDS (­and WORLD) based on Moore’s argument would be rendered unjustified via defeat. The problem with this picture is that it’s not clear how an unjustified belief can have defeating force to begin with. To be clear, I am not claiming that we do not often resist information we are presented with because of our previously held unjustified beliefs. Indeed, we often resist information presented to us for bad reasons, e.g. due to wishfully believing that it is not true (­think, for instance, of cases of partisanship in virtue of friendship, or cases of people in abusive relationships that refuse to acknowledge the abuse, etc.). The question at stake when it comes to defeat, though, is not one concerning the possibility of resistance to evidence, but of permissibility: since justification and defeat are normative, they can only be instantiated in cases in which permissibility is at stake. Cases of wishful thinking are paradigmatic cases in which the hearer is, to use Pryor’s term, ‘­obscured’ from believing information that is presented to them, due to their wishes. Clearly, though, wishful thinking cases are impermissibility cases: the hearer should not, as a matter of fact, resist the testimony in question, even though they do. Again, to follow Pryor’s terminology, these are cases where the believer is not ‘­rationally obscured’ from forming said beliefs, but merely ‘­obscured.’ Or to put it in reasons terms, their unjustified, wishful ­thinking-​­based beliefs are mere motivating reasons for resisting testimony, but not normative reasons. If all this is so, the question that arises is: is the sceptic being ‘­rationally obscured,’ as Pryor would have it, from adopting a belief based on Moore’s testimony by her previously held unjustified sceptical beliefs, or rather, just like the wishful thinker, merely ‘­obscured’ from so doing? Since defeat is a normative category, and since, by Pryor’s own stipulation, the sceptic’s sceptical beliefs are unjustified, it would seem as though they do not qualify as defeaters proper, but rather as mere ­ on-​­normative motivating reasons for resisting Moore’s argument. The n cannot defeat the normative: motivating reasons cannot outweigh normative reasons normatively. Just because I wish really hard to steal your purse, it does not follow that it is permissible to steal your purse: my motivating reasons, no matter how strong, in favour of stealing, cannot outweigh the normative reasons against stealing, since they don’t factor into the overall permissibility calculus to begin with. Why, then, is it intuitive, and according to Pryor, right to think that, ­ on-​­p (­or a doubt about whether p, or once one has adopted a belief that n a.5 credence that ­non-​­p), it would be importantly epistemically defective to adopt a subsequent belief that p? Take the following standard case of ­higher-​­order defeat: I come to believe that the walls in your studio

48  Mona Simion are white but illuminated by a red light to look red. Subsequently, upon arriving at your studio, it seems problematic for me to adopt the belief ‘­The wall in front of me is red’ based on my corresponding perceptual experience as of a red wall. Why is this so? In particular, why is it that, even if we stipulate that my initial belief that the wall is white and illuminated to look red is unjustified, it would seem that, now that I hold it, I shouldn’t just trust my perceptual experience? Maybe the answer to this question has something to do with the order in which the beliefs have been acquired; that is, maybe a difference in extant doxastic states is an epistemologically significant difference. Indeed, Pryor himself alludes to an answer along these lines. According to him, were the sceptic to believe based on Moore’s testimony that HANDS, and thereby WORLD, her belief would be irrational, because it would not cohere with her previously held sceptical beliefs. According to Pryor, since irrationality precludes justification, were the sceptic to believe what Moore says, her belief would also be unjustified: I will count a belief as rational when it’s a belief that none of your other beliefs or doubts rationally oppose or rationally obstruct you from believing. […]A rational commitment is a hypothetical relation between your beliefs; it doesn’t “­detach.” That is, you can have a belief in P, that belief can rationally commit you to believe Q. and yet you be under no categorical requirement to believe Q. Suppose you believe Johnny can fly. This belief rationally commits you to the belief that someone can fly. If you’re not justified in believing that Johnny can fly, though, you need not have any justification for the further belief. You may even have plenty of evidence and be fully justified in believing that no one can fly. But your belief that Johnny can fly still rationally commits you to the belief that someone can fly. Given your belief about Johnny, if you refrain from believing that someone can fly, you’ll thereby exhibit a rational failing. (­Pryor 2004: ­363–​­64) Since rational failings are incompatible with justification, Pryor takes it that this hypothetical type of normativity that he associates with ­rationality – of ​­ the form ‘­if you believe that p, then you are rationally ​­ affect the permissibility of belief committed to believing that q’ – will tout court: were the sceptic to believe what Moore tells her, her belief would be ­irrational – ​­since she is antecedently committed to believing the ­opposite – ​­and thereby unjustified. There are two problems with this normative assessment, though. First and foremost, note that there are two ways of resolving cognitive dissonance due to holding two conflicting beliefs B1 and B2: one can abandon either B1 or B2. Coherence doesn’t tell us which one we should choose: it merely tells us that one needs to go.4 There are two ways of proceeding

Closure, Warrant Transmission, and Defeat  49 in cases in which one is presented with information B2 that runs counter to one’s extant belief B1: one can resist adopting B2, or, alternatively, one can abandon B1. Again, coherence doesn’t recommend any particular course of action: it just tells us we need to choose between them. One thing that Pryor could reply at this juncture is: time makes a difference, epistemically. The previously held belief takes precedence over the incoming information; this is what explains why the sceptic is rational to resist Moore’s argument. The question that arises, though, is: why should we think that time is of such devastating epistemological significance? Why is it, just because the sceptical belief precedes Moore’s testimony temporally, that we should think that it also gets normative priority? After all, consider the following pair of cases (­adapted from Jessica Brown 2018):5 Case 1: A reliable testifier A, who knows that p, asserts that p. At the very same time as receiving A’s testimony, the hearer also receives contrary testimony from another reliable testifier, B, that ­not-​­p. Case 2: We slightly change Case 1 so that the testimony from B arrives just a bit later than the testimony from A, but for whatever reason the hearer does not form any belief about p before the testimony from B arrives. In the cases, the evidentiary and doxastic situation is constant: one testimony item for, one against p, and no difference in mental states. Clearly, the time difference will not make any epistemic difference: in both Case 1 and Case 2, the hearer has equally strong evidence for and against p. She should suspend belief. But now consider: Case 3: Differs from Case 2 only in the following respect: as a result of receiving A’s testimony, the hearer forms the belief that p before receiving B’s testimony. Now, note that there is no temporal difference before Case 2 and Case 3. As such, by the lights of the philosopher who believes that time can make an epistemic difference, there should be no difference in epistemic assessment either. But if there is no epistemic difference between Cases 1 and 2, nor any epistemic difference between Cases 2 and 3, it follows that there is no epistemic difference between Case 1 and Case 3 either. If so, what the hearer should do in both cases is suspend, rather than give priority to the first belief she formed and dismiss the second. Let’s take stock: we have seen that considerations pertaining to coherence cannot explain why we should think that the sceptic is rational to resist Moore’s argument: coherence is indifferent between resisting Moore’s argument and abandoning the previously held sceptical belief. We have also seen that time does not make an epistemic difference either. If so, just because a belief is antecedently held, it does not follow it takes epistemic priority. All of this suggests that the sceptic has no epistemic normative reason to give priority to her sceptical belief and thereby resist Moore’s argument.

50  Mona Simion Furthermore, recall that on Pryor’s view, Moore’s argument is justification conferring, while the sceptical belief is unjustified. If so, there is epistemic normative reason for the sceptic to adopt the conclusion of Moore’s argument, and no epistemic normative reason to hold on to the sceptical ­belief – ​­albeit, of course, the sceptic may well have a merely motivating reason to do so. All in all, it would seem, the sceptic ought (­epistemically) to abandon her sceptical belief and adopt the conclusion of Moore’s argument. The Concessive ­Neo-​­Moorean solution to the sceptical puzzle is wrong: Moore’s argument, while it may well often fail to convince the sceptic, that’s not because it lacks dialectical power, but rather because the sceptic is epistemically impermissibly resisting its conclusion, in virtue of her previously held unjustified sceptical beliefs.

4 A New Radical ­Neo-​­Mooreanism Let’s take stock: we’ve seen that Radical ­Neo-­​­­Mooreanism – ​­claiming that the sceptic’s resistance to Moore’s argument is an instance of epistemic ­malfunction – ​­is thought by many to fail to offer a fully satisfactory explanation of the datum, in that it places the sceptic in the same boat with wishful thinkers, epistemically speaking. However, intuitively, we find the sceptic reasonable, even if wrong, when she resists Moore’s inference. Concessive ­Neo-​­Mooreanism does better on this front: according to this philosopher, the intuition of epistemic permissibility concerning the sceptic’s resistance to Moore’s argument is to be explained in terms of psychological defeat: Moore’s argument is warrant conferring, but dialectically defective. Alas, on closer investigation, this account was shown to run into normative trouble: given that the sceptical belief is unjustified, it remains unclear why the sceptic should favour it over the warranted conclusion of the Moorean argument. In what follows, I will develop a new ­Neo-​­Mooreanism. My view falls squarely within the Radical ­Neo-​­Moorean camp, in that it takes transmission to hold in Moorean inferences and finds no fl ­ aw – ​­epistemic or ­dialectical  – ​­with Moore’s argument. However, as opposed to extant Radical ­Neo-​­Mooreanisms, it does predict that there is something epistemically good about the sceptic’s doxastic response, that sets her apart from believers merely displaying ­full-​­on cognitive malfunctions, such as wishful thinking. Here is how I think about these things (­Simion, Forthcoming a, b): Evidence consists of facts. They can be facts about the world around us or mere facts about a subject’s psychology. My having a perception as of a table in front of me is a psychological fact; it (­pro tanto, prima facie) supports the belief that there is a table in front of me. So does the fact that there is a table in plain view in front of me.

Closure, Warrant Transmission, and Defeat  51 In my view, evidence consists of facts that are knowledge indicators, in that they enhance closeness to knowledge: it consists of facts that one is in a position to know, and that increase one’s evidential p ­ robability – i.e. ​­ the probability on one’s total body of ­evidence – ​­of p being the case. The fact that there is a table in front of me is a piece of evidence for me that there is a table in front of me. It is a knowledge indicator: it raises the probability on my evidence that there is a table in front of me, and I’m in a position to know it. Not just any psychological facts will constitute evidence that there is a table in front of me: my having a perception as of a table will fit the bill in virtue of having the relevant indicator property. Perceptions are knowledge indicators; the fact that I have a perception as of p is a fact that I am in a position to know and that increases my evidential probability that p is the case. The fact that I wish that there was a table in front of me will not fit the bill, even if, unbeknownst to me, my table wishes are strongly correlated with the presence of tables: wishes are not knowledge indicators, for they don’t raise my evidential probability of p being the case. For the same reason, mere beliefs, as opposed to justified and knowledgeable beliefs, will not be evidence material; they lack the relevant indicator property. Here is the view in full: Evidence as Knowledge Indicators: a fact e is evidence for one for a proposition p just in case one is in a position to know e, and one’s evidential probability that p is the case conditional on e is higher than one’s unconditional evidential probability that p is the case. Or, more formally, and where P stands for the probability on one’s total body of evidence: Evidence as Knowledge Indicators: a fact e is evidence for p for S iff S is in a position to know e, and P(­p/­e) > P(­p). Conversely, defeaters are indicators of ignorance: they are facts that one is in a position to know, and that lower one’s evidential probability that p is the case: Defeaters as Ignorance Indicators: a fact d is a defeater for S’s evidence e for p iff S is in a position to know d and S’s evidential probability that p conditional on e&d is lower than S’s evidential probability that p conditional on e. Or, more formally: Defeaters as Ignorance Indicators: a fact d is a defeater for S’s evidence e for p iff S is in a position to know d, and P(­p/­e&d) < P(­p/­e). What is it for me to be in a position to know e? Plausibly, a certain availability relation needs to be instantiated. The fact in question needs to be at hand for me in my epistemic environment: at hand qualitatively (­it needs to be the type of thing a creature like me can access and process), quantitatively (­it needs to remain within the amount of things a creature like me can access and process at one particular time),

52  Mona Simion and environmentally (­it needs to be easily available in m ­ y – ​­internal or e­ xternal – ​­epistemic environment, i.e. in my mind, or in my physical and social surroundings). On my view, a fact F being such that I am in a position to know it has to do with the capacity of my properly functioning ­knowledge-​­generating processes to take up F: Being in a Position to Know: S is in a position to know a fact F iff S has a cognitive process with the function of generating knowledge that can (­qualitatively, quantitatively, and environmentally) easily uptake F in cognizers of S’s type. Some evidence and defeaters I take up with my belief formation machinery, while some I fail to take up, although I should. What grounds this ‘­should,’ in my view, is proper epistemic functioning.6 Because they are knowledge indicators, pieces of evidence are warrant makers: they are the proper inputs to our processes of belief formation, and when we have enough thereof, and the processes in question are properly functioning in all other ways, the resulting belief is epistemically warranted. In turn, when our belief formation processes either fail to take up knowledge indicators that they could have easily taken up, or they take them up but fail to output the corresponding belief, they are malfunctioning. A subject S’s belief formation process P is malfunctioning epistemically if S has sufficient evidence supporting p that is available to be taken up via P and P fails to output a belief that p. The proper function of belief formation processes, then, on my view, is input dependent: failing to take up the right i­ nputs – ​­whether it occurs by taking up the wrong inputs, or by failing to take up the right i­ nputs –​ i­s an instance of malfunctioning. One illuminating analogy here is the proper functioning of the lungs: as opposed to functional traits whose proper function is not ­input-​ d ­ ependent (­e.g. hearts can function properly in vats with orange juice,7 even though they fail to pump blood), what it is for our lungs to function properly is, partly, for them to take up the right amount of the right stuff, i.e. oxygen, from the environment. Lungs that fail to do so are improperly f­ unctioning – ​­whether they fail via taking up carbon dioxide, or by just failing to take up easily available oxygen. Our cognitive systems are not like our hearts, they are like our lungs: their proper functioning is ­input-​­dependent. To see this, note that cognitive systems that take up wishes as inputs are instantiating malfunctioning, just like lungs that take up carbon dioxide. Just like the lungs, then, our cognitive systems can also malfunction by not taking up easily available proper inputs. On my view, then, one way in which our belief formation processes can fail to function properly is by failing to take up easily available evidence. Going back to our sceptic: just like the wishful thinker, on this view of evidence and defeat, the sceptic has no epistemic reason to believe in her preferred sceptical hypothesis. There are no knowledge indicators

Closure, Warrant Transmission, and Defeat  53 available to her to this effect. There are no facts that raise the evidential probability of the sceptical hypotheses within her reach. Furthermore, Moore’s assertion that HANDS provides the sceptic with evidence that there are hands, since Moore’s testimony to this effect is a knowledge indicator. Also, since the sceptic’s sceptical belief is not a knowledge indicator, it does not qualify as a defeater for HANDS. In this, the sceptic is in double breach of ­justification-​­conferring epistemic norms: she has unjustified sceptical beliefs, and she resists knowledge indicators on offer because of them. The sceptic does not have defeaters for HANDS; rather, she has mere motivating reasons to this effect: ­facts – ​­i.e. the fact that she believes ­Non-​­WORLD/­doubts ­WORLD – that ​­ lead her to unjustifiably reject HANDS. What is it, then, that explains our intuition of reasonableness in the sceptic case, and the lack thereof in the case of the wishful thinker? Recall: According to the view developed here, the sceptic ought not hold sceptical beliefs to begin with, ought to come to believe that WORLD based on Moore’s argument, and thereby ought to draw the inference to WORLD with Moore and abandon her antecedently held sceptical beliefs. If she fails to do all that, she is in breach of the ­justification-​ ­conferring epistemic norm: her resistance to Moore’s argument is epistemically impermissible. Now, here is, however, a ­well-​­known fact about norms, generally speaking: sometimes, when we engage in impermissible actions, this gives rise to ­contrary-­​­­to-​­duty obligations. Consider the following normative claims: 1 It ought to be that John does not break the neighbour’s window. 2 If John breaks the neighbour’s window, it ought to be that he apologizes. 3 is a primary obligation, saying what Jones ought to do unconditionally. In contrast, (­2) is a ­contrary-­​­­to-​­duty obligation about (­in the context of 1) what Jones ought to do conditional on his violating his primary obligation. (­1) is a norm of many sorts: social, prudential, moral, and a norm of politeness. Should John break the neighbour’s window, there would be nothing good about it. That being said, John would be even worse off if, should he break the neighbour’s window, he would also fail to go and apologize to the neighbour. Our functionalist normative schema has the resources needed to explain ​­ the type that govthis datum: i­ nput-​­independent proper ­functioning – of erns h ­ earts – remains ​­ a dimension of functional evaluation in its own right, independently of whether the general proper functioning of the trait in question is i­ nput-​­dependent or not: just like we can ask whether a heart is doing what it’s supposed to do with the stuff that it takes ­up – ​­be it blood or orange j­uice – ​­we can also ask whether the lungs are doing

54  Mona Simion the stuff that they’re supposed to do with the stuff that they have taken ­up – ​­be it oxygen or carbon dioxide. There’s going to be an evaluative difference, then, between two pairs of lungs that are both improperly functioning simpliciter – i.e. ​­ in the i­nput-​­dependent sense, in that they take up the wrong kind of stuff from the ­environment – ​­in terms of how they process their input gas: are they carrying the input gas through the respiratory system, and subsequently through the lining of the air sacs, to the blood cells? The pair of lungs who do are better than the pair of lungs who don’t, in that, even though strictly speaking both are malfunctioning overall, the former are at least displaying ­input-​­independent proper functioning. What explains our intuition of reasonableness in the sceptic’s case, I claim, is not an epistemic norm simpliciter, but rather an epistemic ­contrary-­​­­to-​­duty imperative: now that the sceptic is in breach of the ­justification-​­conferring epistemic norm, short of abandoning her unjustified belief, the next best thing for her to do is embrace the commitments following from her unjustified beliefs and reject the commitments that follow from their negation. The next best thing for the sceptic, now that she believes/­has a.5 credence that n ­ on-​­WORLD and rejects HANDS, both impermissibly, short of abandoning her impermissible beliefs, is to reject whatever follows from HANDS. The sceptic’s cognitive system, just like the wishful thinker’s, and just like lungs taking up carbon dioxide from the environment, is overall malfunctioning on several counts: it takes up improper inputs (­her sceptical beliefs) and rejects excellent inputs (­Moore’s testimony that HANDS). That being so, though, the sceptic does something right in terms of ­input-​­independent functioning: it processes the (­bad) stuff that she takes up in the right way. Her cognitive system would be even worse were she, now that she believes/­has a.5 credence that ­non-​­HANDS, go ahead and infer that WORLD. Before I close, I would like to consider a possible objection to my view. So far, I have been assuming, with Pryor and Williamson, that the sceptic’s sceptical beliefs/­doubts are unjustified. One could worry though: Doesn’t my view of evidence allow for the (­reasonable) sceptic to have ­induction-​­based evidence for her.5 credence that ­non-​­World? After all, the sceptic could reason as follows: (­1) When I can’t tell the difference between pears and apples, I can’t come to know that there’s an apple in front of me; (­2) When I can’t tell the difference between John and his twin brother Tim, I can’t come to know that John is in front of me……. (­3) Therefore, when I can’t tell the difference between x and y, I can’t come to know that x is the case. (­4) I can’t tell the difference between WORLD and n ­ on-​­WORLD, therefore (­5) I don’t know that WORLD. In turn, if the sceptic believes that (­5), on pain of Moorean paradoxicality, she can’t believe that WORLD. Two things about this: first, crucially, the envisaged sceptic is wrong, (­3) is notably too strong: I can come to know that there’s a pear in front

Closure, Warrant Transmission, and Defeat  55 of me in a world where there are no apples, or where apples are extremely rare, even if I can’t tell the difference between pears and apples. That being said, of course, (­3) may well be justified inductively, which would lead to (­5) being justified inductively. Second, though, note that Moorean paradoxicality, just like incoherence, tells us nothing about which of the two beliefs should be abandoned: it merely predicts that one needs to go. Why think WORLD needs to go, rather than (­5)? Furthermore, notice that in everyday testimonial cases, it’s the previously held ignorance belief that should be abandoned: I believe I don’t know whether you are 32 years old, you tell me that you are 32 years old, I thereby come to know that you are 32 years old and abandon my belief that I don’t know that you are 32 years old. That’s how it normally goes. Last attempt: maybe the sceptic’s inductively justified belief that she can’t tell the difference between WORLD and ­non-​­WORLD acts as an undercutting defeater for Moore’s testimony that HANDS? This could work. The problem, though, is that undercutting defeaters need to exhibit particular strength properties in order to successfully undercut. For instance, my t­ hree-­​­­year-​­old’s testimony that Dretske is wrong about closure failure, because he took a hallucinogenic drug before writing ‘­Epistemic Operators,’ will not successfully undercut my evidence that closure fails sourced in Dretske’s chapter. Why not? My ­three-­​­­year-​­old is just not a very reliable testifier on the ­issue – ​­not reliable enough to undercut Dretske’s written testimony, at any rate. The testimony from my ­three-­​­­year-​­old does not lower my evidential probability conditional on Dretske’s testimony that closure fails. If so, what would need to happen in the case of the sceptic for her i­nduction-​­based sceptical belief to undercut Moore’s testimony would be that the former is weighty enough, epistemically. Why, though, think that the sceptic’s induction has such devastating epistemic effects against Moore’s testimony? Also, recall that the inductive argument only warrants the reasonable sceptical belief ‘­I don’t know that WORLD,’ not the anxious sceptical belief that ‘­­non-​ ­WORLD.’ Of course, though, the former is much weaker than the latter, and thus with much less defeat power.8

5 Conclusion This chapter has developed a novel, functionalist variety of Radical ­Neo-​­Mooreanism. I have argued with Williamson that, just like the wishful thinker, the sceptic is displaying epistemic malfunction in rejecting Moore’s testimony. On my account, that is because her cognitive processes fail to pick up knowledge indicators. I have also shown, however, that the intuition that there’s something reasonable about the sceptic who resists going through Moore’s inference is right: the sceptic is in compliance with a c­ ontrary-­​­­to-​­duty obligation akin to i­nput-​ ­independent well functioning.

56  Mona Simion To be clear: this account does not make any concessions to the sceptic in terms of j­ustification-​­conferring epistemic ­norms – ​­i.e. primary epistemic obligations: no justification for sceptical beliefs, nor any defeat against Moore’s testimony, is instantiated at the context. The account merely explains why we find the sceptic reasonable (­albeit wrong) to resist Moore’s inference from HANDS to WORLD: she is in compliance with her c­ ontrary-­​­­to-​­duty epistemic obligations. Now that she’s broken the window, as it were, the sceptic might as well go ahead and apologize.

Notes * This research has received funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation programme (grant agreement No 948356, KnowledgeLab project, PI: Mona Simion). 1 Other closure and warrant ­ transmission-​­ preserving responses come e.g. from hinge epistemologists (­e.g. Pritchard 2015) and are motivated by e.g. the thought that our doxastic attitudes towards hinge propositions such as ‘­T he external world exists’ are not beliefs, and thereby not the stuff that makes the proper target of closure and transmission principles. See e.g. Jope (­2019) and Simion et al. (­2019) for discussion. 2 But see Kelp (­2019) for my favourite proposal. 3 To my knowledge, the first to have introduced the category of psychological (/­doxastic) defeat is Jennifer Lackey (­e.g. 2006: 438). The first and now considered the classic view on the nature of defeat in epistemology is due to Pollock (­1986). For excellent recent work on defeat, see Brown and Simion (­2021). 4 See also Simion (­Forthcoming, b) and Graham and Lyons (­2021) for similar points. 5 See also Goldberg (­2021). 6 See e.g. Graham (­2012), Millikan (­1984), Simion (­Forthcoming, b). 7 I borrow this from Graham (­2012). 8 Thanks to Chris Kelp for pressing me on this.

References Brown, J. (­ 2018). Fallibilism: Evidence and Knowledge. Oxford: Oxford U ­ niversity Press. Brown, J. and Simion, M. (­2021). Reasons, Justification, and Defeat. Oxford: Oxford University Press. Dretske, F. (­1970). Epistemic operators. Journal of Philosophy 67: ­1007–​­23. Goldberg, S. (­2021). The normativity of knowledge and the scope and sources of defeat. In J. Brown and M. Simion (­eds.), Reasons, Justification, and Defeat. Oxford: Oxford University Press, ­18–​­39. Graham, P. (­2012). Epistemic entitlement. Nous 46/­3: ­4 49–​­82. Graham, P. and Lyons, J. (­2021). The structure of defeat: Pollock’s evidentialism, Lackey’s framework, and prospects for reliabilism. In J. Brown and M. Simion (­eds.), Reasons, Justification, and Defeat. Oxford: Oxford University Press, ­39–​­69.

Closure, Warrant Transmission, and Defeat  57 Jope, M. (­2019). Closure, credence and rationality: a problem for n ­ on-​­belief hinge epistemology. Synthese. https://­doi.org/­10.1007/­­s11229-­​­­019-­​­­02153-​­5. Kelp, C. (­2019). Inquiry and the transmission of knowledge. Philosophy and Phenomenological Research 99: ­298–​­310. Lackey, J. (­2006). Knowing from testimony. Philosophy Compass 1/­5: ­432–​­48. Millikan, R. (­1984). Language, Thought, and Other Biological Categories. Cambridge, MA: MIT Press. Moore, G. E. (­1939). Proof of an external world. Proceedings of the British Academy 25: 2 ­ 73–​­300. Pollock, J. (­1986). Contemporary Theories of Knowledge. Savage, MD: Rowman and Littlefield. Pritchard, D. (­2015). Epistemic Angst: Radical Skepticism and the Groundlessness of Our Believing. Princeton, NJ: Princeton University Press. Pryor, J. (­2000). The skeptic and the dogmatist. Nous 34: ­517–​­49. Pryor, J. (­2004). What’s wrong with Moore’s argument? Philosophical Issues 14: Epistemology: ­349–​­78. Pryor, J. (­2012). When warrant transmits. In A. Coliva (­ed.) Mind, Meaning, and Knowledge: Themes from the Philosophy of Crispin Wright. Oxford: Oxford University Press, ­270–​­303. Simion, M. (­Forthcoming, a). Resistance to evidence and the duty to believe. Philosophy and Phenomenological Research. Simion, M. (­Forthcoming, b). Resistance to Evidence. Cambridge: Cambridge University Press. Simion, M., Schnurr, J. and Gordon, E. (­2019). Epistemic norms, closure, and ­no-​­belief hinge epistemology. Synthese, Online First. https://­doi.org/­10.1007/ ­­s11229-­​­­019-­​­­02165-​­1. Williamson, T. (­2000). Knowledge and Its Limits. Oxford: Oxford University Press. Williamson, T. (­2007). Knowledge and scepticism. In F. Jackson and M. Smith (­eds.), The Oxford Handbook of Contemporary Philosophy. Oxford: Oxford University Press, 6 ­ 81–​­700. Wright, C. (­2002). (­­Anti-​­)­sceptics simple and subtle: Moore and McDowell. Philosophy and Phenomenological Research 65: 3 ­ 30–​­48. Wright, C. (­2003). Some reflections on the acquisition of warrant by inference. In Susana Nuccetelli (­ed.), New Essays on Semantic Externalism and S­ elf-​ ­Knowledge. Cambridge: MIT Press, 5 ­ 7–​­77. Wright, C. (­2004). Warrant for nothing (­and foundations for free)? Aristotelian Society Supplement. Vol. 78/­1: ­167–​­212.

4 Modal Epistemology and the Logic of Counterfactuals Timothy Williamson

Modal epistemology, as understood here, is epistemology done using modal conditions to characterize epistemic states: for example, a modal connection between belief and truth to characterize actual knowledge. By contrast, the epistemology of modality is the epistemology of epistemic states whose content concerns modal matters: for example, knowledge that something is necessary or possible. Modal epistemology in the tradition descending from Fred Dretske (­1970) and Robert Nozick (­1981) has focussed on modal connections expressed by counterfactual conditions, such as Sensitivity: if the proposition weren’t true, you wouldn’t believe it (­the simplest version). Sensitivity to a truth in that sense has been claimed necessary for knowing that truth. Notoriously, such counterfactual conditions on knowledge seem to undermine plausible principles of epistemic closure: sometimes you know each premise of an argument, and believe the conclusion by competent deduction from the premises, yet you fail to know the conclusion, for although you are sensitive to each premise, you are not sensitive to the conclusion, so the alleged necessary condition for knowledge fails. Classic accounts of the semantics of counterfactual conditionals, in particular those of Robert Stalnaker (­1968) and David Lewis (­1973), vindicate such apparent counterexamples to epistemic closure, in a way intended by proponents of the sensitivity condition. The contrapositive of the sensitivity condition has also been used to define a proposed modal condition on knowledge, Counterfactual Safety: if you believed the proposition, it would be true (­again, the simplest version). Sensitivity and Counterfactual Safety are usually thought to work in quite different ways. Section 1 discusses in more detail the relation between the logic and semantics of counterfactuals on one hand and counterexamples to epistemic closure on the other. Section 2 expounds a more natural approach to the logic and semantics counterfactual conditionals. Section 3 explains why the alleged counterexamples to epistemic closure are illusory on this alternative approach, and how the illusions arise. It also explains why Sensitivity and Counterfactual Safety are in principle equivalent. Section 4 formulates closure principles for those conditions validated DOI: 10.4324/9781003104766-5

Modal Epistemology and the Logic of Counterfactuals  59 by the alternative semantics. Section 5 concludes with a methodological moral.

1 The Standard Picture One is sensitive to A just in case if A were not true, one would not believe A. More complex versions of sensitivity qualify ‘­believes’ with ‘­by method M’ or ‘­on basis B’ or the like, and even variation in the proposition believed. The argument of this chapter can be adjusted to allow for such qualifications; for simplicity, they will usually be omitted. We formalize the operator ‘­one believes that …’ as Bel, negation as ¬, and the counterfactual conditional with antecedent A and consequent C as A > C (‘­If A were true, C would be true’). Thus ‘­one is sensitive to A’ becomes ¬A > ¬Bel[A]. For convenience, ‘­A’ takes sentence position in formulas but name position in their English paraphrases, where it acts as a (­variable) name of the proposition expressed by the corresponding sentence. Stalnaker and Lewis formulate their semantic accounts of counterfactual conditionals within the framework of possible worlds semantics. Sentences are true or false at (­possible) worlds; an ­A-​­world is a world at which A is true. A counterfactual conditional A > C is true at a world w if and only if the A-​­worlds closest to w are C-​­worlds. If there are no A-​­worlds (­A is impossible), A > C is true vacuously. Stalnaker stipulates that there is always a unique closest A-​­world; Lewis does not. Closeness is similarity on the relevant dimension; much can be said as to what that dimension is, but for present purposes we take it as vaguely understood. Thus, putting the pieces together, one is sensitive to A if and only if, at the closest worlds at which A is not true, one does not believe A. According to the principle of Strengthening the Antecedent, if A+ entails A, then A > C entails A+ > C. For example, since ‘­Exactly forty people were at the party’ entails ‘­At least forty people were at the party’, ‘­If at least forty people were at the party, the room was crowded’ entails ‘­If exactly forty people were at the party, the room was crowded’. In the present setting, we can define entailment by stipulating that D entails E if and only if every D-​­world is an E-​­world. The ­Stalnaker–​­L ewis view predicts that Strengthening the Antecedent is invalid. For let w, x, and y be worlds such that x is closer than y to w. Let A be true just at x and y, A+ true just at y (­so A+ entails A), and C true just at x. Thus the closest ­A-​­world to w is x, which is also a C-​­world, so A > C is true at w. But the closest A+-​­world to w is y, which is not a C-​­world, so A+ > C is not true at w. Thus A > C does not entail A+ > C, so Strengthening the Antecedent fails. This phenomenon leads to failures of epistemic closure, especially with respect to sceptical scenarios, just as Nozick and other proponents of a Sensitivity condition intended. Let H be the proposition that I have hands. In the actual world I know H in the normal way. As a corollary, H is true and I believe H. I am also

60  Timothy Williamson sensitive to H, for in the closest ¬H-​­worlds to the actual world, I lost my hands in a nasty accident, of which I am well aware, and so do not believe that I have hands, so ¬H > ¬Bel[H] is true at the actual world. Now let BIV be the proposition that I am a handless brain in a vat who believes that it has hands. Thus H entails ¬BIV. We may assume that in all relevant worlds in which I believe H, I also believe ¬BIV by competent deduction from H, for our present interest is not in failures of epistemic closure which merely reflect my deductive limitations. However, I am not sensitive to ¬BIV, for in the closest ¬¬BIV-​­worlds, in other words BIV-​­worlds, I believe H, and so also believe ¬BIV, so BIV > ¬Bel[¬BIV] (­omitting a double negation) is false at the actual world. Consequently, although I know H, and H entails ¬BIV, I do not know ¬BIV. This is so even though ¬BIV is true (­since H is true), and I believe ¬BIV by having competently deduced it from H, which I already know. The ­non-​­closure of Sensitivity here is the main reason for the n ­ on-​­closure of knowledge, even though other details have to be checked too, since Sensitivity is not the only condition on knowledge, and we require competent deduction, not just entailment in itself. Nozick ingeniously used this pattern to explain the appeal of scepticism without surrendering to it: the sceptic rightly denies that we know that we are not in the sceptical scenario, but wrongly concludes that we fail to know ordinary truths incompatible with our being in the sceptical scenario, because the sceptic wrongly assumes epistemic closure. Although Nozick mainly followed Lewis’s semantics for counterfactual conditionals, in one respect he was unorthodox. Lewis was sympathetic to the Strong Centering assumption that each world is the closest to itself. Thus if A and C are both true at w, so is A > C, for the uniquely closest A-​­world to w is w, which is also a C-​­world. Nozick demurred. In addition to truth, belief, and sensitivity, his analysis of knowledge required a fourth condition: ‘­Not only is p true and S believes it, but if it were true he would believe it’ (­Nozick 1981: 176). He denied that the conjunction made the conditional redundant (­as Strong Centering implies), insisting that a counterfactual is true only where the consequent is true at close (­but not maximally close) worlds at which the antecedent is true. The symmetry between his third and fourth conditions, ¬A > ¬Bel[A] and A > Bel[A], was an attractive feature of his analysis, in Nozick’s eyes: if the fourth condition were redundant, the symmetry would be an illusion. Ernest Sosa has used something like the converse of Nozick’s fourth condition, Bel[A] > A, to formulate a version of another putative modal condition on knowledge, Counterfactual Safety: if one believed A, it would be true (­Sosa 1999, 2003, 2007, 2009; for an approach to safety not employing counterfactual conditionals see Williamson 2000). Like Nozick, Sosa insists that the conjunction of the antecedent and consequent does not make the counterfactual conditional redundant.

Modal Epistemology and the Logic of Counterfactuals  61 The Sensitivity condition ¬A > ¬Bel[A] is just the contrapositive of the Counterfactual Safety condition Bel[A] > A. Thus if counterfactual conditionals obeyed the rule of Contraposition (­A > C entails ¬C > ¬A), Sensitivity would be equivalent to Counterfactual Safety (­since double negations cancel). But the S­ talnaker–​­Lewis account predicts the failure of Contraposition. For let w, x, and y be worlds such that x is closer than y to w. Let A be true just at x and y, and C true at every world except y. Thus the closest A-​­world to w is x, which is also an C-​­world, so A > C is true at w. But the closest ¬C-​­world to w is y, which is an A-​­world, so ¬C > ¬A is not true at w. Thus Contraposition is invalid. We can illustrate the failure of Contraposition on the ­Stalnaker–​­Lewis semantics with or without Strong Centering in terms of the sceptical scenario. The Counterfactual Safety conditional Bel[¬BIV] > ¬BIV is true in the good case: at all worlds close to the actual world (­including itself) in which I believe that I am not a handless brain in a vat who believes that it has hands, I am indeed not a handless brain in a vat who believes that it has hands, for I am not a brain in a vat in any close world. But the contrapositive of that Counterfactual Safety conditional is the Sensitivity conditional BIV > ¬Bel[¬BIV] (­omitting a double negation), which was seen above to be false on the semantics. Conversely, on the ­Stalnaker–​­Lewis semantics, in some other cases the Sensitivity conditional is true while the Counterfactual Safety conditional is false. For example, let A be ‘­I am the Pope’. At every close world, I neither am the Pope nor believe that I am the Pope. Thus the Sensitivity conditional ¬A > ¬Bel[A] is actually true. However, in the nearest possible worlds in which I do believe that I am the Pope, I am merely deluded, and still not the Pope. Thus the Counterfactual Safety conditional Bel[A] > A is false on the ­Stalnaker–​­Lewis semantics (­with or without Strong Centering). Still, given Weak Centering (­the principle that any world is one of the closest to itself), when A is believed and the Sensitivity conditional holds, so does the Counterfactual Safety conditional on the ­Lewis–​­Stalnaker semantics. Unlike Strong Centering, Weak Centering is quite compatible with Nozick’s unorthodox treatment of counterfactuals. To check the connection, suppose that both ¬A > ¬Bel[A] and Bel[A] hold at w. There are two possibilities. (­i) A holds at all the closest worlds to w. Since Bel[A] holds at one of them by Weak Centering, by the semantics Bel[A] > A holds at w. (­ii) ¬A holds at one of the closest worlds to w. By the semantics, at those worlds, ¬Bel[A] holds wherever ¬A holds, so A holds wherever Bel[A] holds; since Bel[A] holds at one of those worlds, Bel[A] > A holds at w. Either way, the advertised connection holds. Although the semantics does not make Counterfactual Safety equivalent to Sensitivity, it does make Counterfactual Safety, like Sensitivity, violate closure under entailment. For example, suppose that I am non-​­ poodles, but poor good at distinguishing between poodles and ­

62  Timothy Williamson at distinguishing between dogs and ­non-​­dogs: there are many wolves around, all of which I take to be dogs. Let P say that there is a poodle in front of me and D say that there is a dog in front of me. Thus P entails D (­and I know the entailment). Plausibly, P satisfies counterfactual safety: Bel[P] > P is true. But, in the same circumstances, D may well violate counterfactual safety: Bel[D] > D may well be false. Thus Counterfactual Safety is not closed under entailment. However, the ­non-​­closure of Counterfactual Safety differs from the ­non-​­closure of Sensitivity: the former, unlike the latter, depends on the gap between entailment in itself and competent deduction. For let us restrict attention to worlds in which (­i) if I believe that there is a poodle in front of me, I also believe by competent deduction from that belief that there is a dog in front of me and (­ii) if I believe that there is a dog in front of me, I do so by competent deduction from my belief that there is a poodle in front of me. If we pretend away all other worlds, Bel[P] and Bel[D] are true at exactly the same worlds and so have the same counterfactual implications; since P entails D, Bel[P] > P entails Bel[D] > D, for the ­Stalnaker–​­Lewis approach does at least make counterfactual implication closed under entailment in the consequent, for a fixed antecedent. Of course, this argument does not show that Counterfactual Safety is really closed under entailment. Rather, it suggests that we may be able to define some more restricted kind of ­belief-­​­­on-­​­­a-­​­­given-​­basis for which the appropriate analogue of Counterfactual Safety is closed under competent deduction. That suggestion is followed up in Section 3. By contrast, no such rehabilitation of closure is feasible for Sensitivity, on the ­Stalnaker–​­Lewis approach. For example, suppose that we restrict attention to worlds in which the analogues of (­i) and (­ii) hold: (­i*) if I believe that I have hands, I also believe by competent deduction from that belief that I am not a handless brain in a vat who believes that it has hands and (­ii*) if I believe that I am not a handless brain in a vat who believes that it has hands, I do so by competent deduction from my belief that I have hands. If we pretend away all other worlds, Bel[H] and Bel[¬BIV] are true at exactly the same worlds, as are their negations ¬Bel[H] and ¬Bel[¬BIV]. But that does not help, for they occur in the consequents of the Sensitivity conditionals. The corresponding antecedents are ¬H and BIV, which are not susceptible to such doctoring and still pick out different sets of worlds, even amongst those which satisfy both (­i*) and (­ii*). Thus, even under those restrictions, the ­Stalnaker–​­Lewis approach still predicts that Sensitivity is not closed under entailment.

2 An Alternative Semantics for Counterfactual Conditionals Following Nozick, modal epistemology uses counterfactual conditionals, which are standardly contrasted with plain indicative conditionals.

Modal Epistemology and the Logic of Counterfactuals  63 The difference is crucial. The counterfactual conditional ‘­If there weren’t a poodle in front of me now, I wouldn’t now believe that there was a poodle in front of me’ is a sensible thing for an expert on poodles to say; by contrast, the indicative conditional ‘­If there isn’t a poodle in front of me now, I don’t now believe that there is a poodle in front of me’ would be a weird thing for her to say. In some sense, the counterfactual permits one to develop scenarios in which some of what one actually knows (­for instance, that there is a poodle in front of her now) does not obtain, without thereby treating that actual knowledge as doubtful, while the indicative conditional does not. That allows one to evaluate the consequent of the counterfactual without the wrong kind of interference from one’s background knowledge (­for instance, that she does now believe that there is a poodle in front of her). In English, the two kinds of conditional statement are distinguished syntactically, as in the indicative ‘­If X is the case, Y will be the case’ versus the counterfactual ‘­If X was the case, Y would be the case’. Some other languages leave the distinction to be tacitly understood from the conversational context, but the explicit marking gives useful clues about the underlying structural difference. Oddly, the difference is usually described as one between the indicative and counterfactual conditionals, as though it were between two readings of ‘­if’ itself, sometimes formalized as → and □→. But why locate the difference in a word in common between the two forms of sentence, when there are visible differences elsewhere, in the verb phrases? For example, in the antecedent, ‘­is the case’ contrasts with ‘­was the case’; in the consequent, ‘­will be the case’ contrasts with ‘­would be the case’. The simplest default hypothesis is that ‘­if’ has exactly the same meaning in the two types of sentence, which combines with the distinct meanings of the different verb forms to give the different meanings of the two types of sentence within the standard framework of compositional semantics, which determines the meaning of a complex expression as a function of the meanings of its simpler constituents and the way in which they are put together. The default hypothesis gains immediate support from the observation that ‘­would’ has a life of its own, independent of ‘­if’. Syntactically, ‘­would’ is of course the past tense of ‘­will’. For example, when two people part, someone might predict ‘­T hey will never meet again’. Later, after the prediction has turned out correct, one might end an account of their parting with ‘­T hey would never meet again’. But ‘­would’ also has what linguists call a ‘­fake past’ use, on which it ranges over relevant counterfactual possibilities instead of past times. For example, when you report ‘­Mary is not at home’, I might comment ‘­She wouldn’t be’. There is no suppressed ‘­if’ clause. Nor am I talking about the past; rather, I am subsuming her absence under a modal generalization over a range of more or less close possibilities: it is no accident that she is not at home; perhaps she is always out for a walk at this time of day, or whatever.

64  Timothy Williamson In such cases, ‘­would’ functions as a contextually restricted local necessity operator. The natural hypothesis is that ‘­would’ functions that way in counterfactual conditionals too. Thus ‘­If X was the case, Y would be the case’ is the result of applying a contextually restricted local necessity operator, realized as a fake past tense, to ‘­If X is the case, Y will be the case’ (­the ‘­will’ here may also be a fake future, as in ‘­Mary will be out walking now’). Consequently, ‘­If X was the case, Y would be the case’ is true as uttered in a given context if and only if at every possibility relevant in that context, ‘­If X is the case, Y will be the case’ is true. To complete the compositional semantics of such modalized conditionals, we need an independent semantics of ‘­if’ itself. The simplest hypothesis is that ‘­if’ is just the ­truth-​­functional material conditional, so that the corresponding counterfactuals are just contextually restricted strict conditionals. Thus ‘­If X was the case, Y would be the case’ is true as uttered in a given context if and only if at every possibility relevant to that context at which X is the case, Y is also the case. Where ⊃ is the material conditional (‘­if’) and □ is the contextually restricted necessity operator, we can formalize the ‘­would’ conditional with antecedent A and consequent C as □(­A ⊃ C). Elsewhere, I have developed such an approach in detail, and shown that it has the resources to answer the salient kinds of objection likely to be brought against it (­Williamson 2020, also for references to the literature on conditionals). I will not try to summarize all that argumentation here, but just mention a few features of special significance for its application to modal epistemology.1 On the ­non-​­modal side, the material interpretation of the English word ‘­if’ faces numerous apparent counterexamples, which might be compelling were it not for the numerous apparent counterexamples to any rival interpretation. Fortunately, the evidence is not as anarchic as it seems. Our practice of using conditionals can be understood as built on a simple heuristic for assessing them (­Williamson 2020: 19): Suppositional Rule

Take an attitude unconditionally to ‘­I f A, C’  just in case you take it conditionally to C  on the supposition A.

For example, accept ‘­If A, C’ unconditionally just in case you accept C conditionally on the supposition A; reject ‘­If A, C’ unconditionally just in case you reject C conditionally on the supposition A. In probabilistic terms, your unconditional probability for ‘­If A, C’ should equal your conditional probability for C on A. The Suppositional Rule is closely related to a test originally proposed by Frank Ramsey for assessing conditionals (­Ramsey 1929: 143). There is considerable evidence that it is our primary means of assessing conditionals. Naturally enough, we also use it in assessing conditionals under further background suppositions, in

Modal Epistemology and the Logic of Counterfactuals  65 common between the two sides of the Suppositional Rule. However, the Rule is not our only means of assessing conditionals. For example, one might accept ‘­If A, C’ on the testimony of a trusted informant, without bothering to assess C on the supposition A. The alert reader may have noticed that the material interpretation of ‘­if’ does not fully validate the Suppositional Rule. For instance, on the material interpretation of ‘­if’, ‘­N N is the Prime Minister’ entails ‘­If NN has just died, NN is the Prime Minister’. Thus, if one is confident that NN is the Prime Minister, one should also be confident that if NN has just died, NN is the Prime Minister. But, of course, one’s natural response to ‘­If NN has just died, NN is the Prime Minister’ is to reject it, just as the Suppositional Rule predicts, because one rejects ‘­N N is the Prime Minister’ on the supposition ‘­N N has just died’. Thus one might expect our reliance on the Suppositional Rule to refute the material interpretation of ‘­if’, and support some other interpretation which does fully validate the Rule. But things are not so simple. For the Suppositional Rule is actually inconsistent, and so is not fully valid on any interpretation of ‘­if’. A quick way to detect the inconsistency is by noting that the Rule implies the standard introduction and elimination rules for the conditional in systems of natural deduction, conditional proof, and modus ponens. For, by the ­right-­​­­to-​­left direction of the Suppositional Rule, if C is conclusively accepted on the supposition A and background assumptions, then ‘­If A, C’ should be conclusively accepted on those background assumptions, which is equivalent to the rule of conditional proof. Conversely, by the l­ eft-­​­­to-​­right direction of the Suppositional Rule, if ‘­If A, C’ is conclusively accepted on background assumptions, then C should be conclusively accepted on the supposition A and those background assumptions, which is equivalent to the rule of modus ponens. Together, the rules of ­ ell-​­known conditional proof and modus ponens for a conditional are w to force its equivalence to the material conditional. Thus the Suppositional Rule makes ‘­if’ equivalent to the material conditional. But we saw previously that the Suppositional Rule also has consequences inconsistent with the equivalence of ‘­if’ to the material conditional. Thus the Suppositional Rule itself is inconsistent. Naturally, one wonders how an inconsistent rule can be our primary way of assessing conditionals. Won’t that make our whole practice of using ‘­if’ collapse? A useful analogy is with our practice of using the words ‘­true’ and ‘­false’. Our primary means of assessing predications of ‘­true’ and ‘­false’ is by disquotation. We take the same attitude to ‘“­Dubrovnik is in Croatia” is true’ as to ‘­Dubrovnik is in Croatia’, and the same attitude to ‘“­Dubrovnik is in Croatia” is false’ as to ‘­Dubrovnik is not in Croatia’. Such moves are essential to our practice of using ‘­true’ and ‘­false’. Nevertheless, the disquotation rules are inconsistent, as the Liar and other

66  Timothy Williamson semantic paradoxes show. Despite the inconsistency, our whole practice of using ‘­true’ and ‘­false’ does not collapse: it is a going concern. The inconsistencies are too marginal to do serious damage. Philosophers who regard the disquotational rules as somehow analytic may be tempted to treat their validity as a precondition for ‘­true’ and ‘­false’ to be so much as meaningful. But that would be a mistake: the rules are invalid, because inconsistent, yet ‘­true’ and ‘­false’ are still meaningful words of English by any reasonable standard. Nor is any particular restriction built into the rules as we use them, to preserve consistency; the semantic paradoxes strike normal speakers of English like a bolt from the blue. Although research in philosophical logic may succeed in identifying more qualified rules for ‘­true’ and ‘­false’ which are valid without exception, that is a scientific discovery, not the articulation of something which normal speakers of English knew all along. Similarly, despite the inconsistency of the Suppositional Rule, our whole practice of using ‘­if’ does not collapse: it is a going concern. The inconsistency is not serious enough to undermine the practice, though it may be less marginal than the semantic paradoxes. Despite the Rule’s invalidity, ‘­if’ is still a meaningful word of English by any reasonable standard. The Rule is not somehow analytic. Nor is any particular restriction built into the Rule as we use it, to preserve consistency; the paradoxes of material implication strike normal speakers of English like a bolt from the blue. Although research in philosophical logic may succeed in identifying more qualified rules for ‘­if’ which are valid without exception, that is a scientific discovery, not the articulation of something which normal speakers of English knew all along. We do better to regard the Suppositional Rule as a heuristic for assessing conditionals in the psychologists’ sense, a rule of thumb, normally quick and easy to apply, and reliable enough for practical purposes, but not perfectly reliable. Human cognition relies on many such heuristics. From the user’s perspective, they need not wear their status as mere heuristics on their sleeve, especially when we have no better heuristic to check them by. That they are mere heuristics may be something else we have to discover, perhaps as a corollary from discovering that they are inconsistent. In such cases, we cannot simply read the semantics of the word off the heuristic for its application. Instead, we must work out what semantics for the word makes the best sense of the practice, including all the relevant heuristics. For ‘­if’ they include the primary heuristic, the Suppositional Rule, but also secondary heuristics, such as reliance on conditional testimony. There are clear theoretical reasons for taking the material interpretation of ‘­if’ to make the best sense of our practice of using the word, in terms of both simplicity and charity (­Williamson 2020: ­89–​­121). Since those reasons are not available to normal speakers of English except on complex theoretical reflection, the semantics of their own word ‘­if’ is not transparent to them.

Modal Epistemology and the Logic of Counterfactuals  67 A consequence of this view is that, in cases where the Suppositional Rule gives the wrong ­result—​­for example, when they accept ‘­N N is the Prime Minister’ but reject ‘­If NN has just died, NN is the Prime Minister’—​­they are typically in no position to recognize the error and correct it. They apply the Suppositional Rule correctly, and there is no special clue that it may not be giving the right result, so the case gives the illusion of being a counterexample to the material interpretation. Such predicaments are part of the unrecognized cost of relying on fallible heuristics. Fortunately, heuristics on which we continually rely, such as the Suppositional Rule, also tend to have major benefits, otherwise in the long run they would hardly survive. In particular, the Suppositional Rule enables us to extract and apply information about various sorts of ­non-​ ­accidental connection encoded in our cognitive dispositions, by using imagination in ­reality-​­oriented ways to make offline assessments of some sentences on the supposition of others. By contrast, if we try to evaluate the material conditional in the disjunctive form ¬A ∨ C or the negated conjunctive form ¬(­A ∧ ¬C) by direct t­ruth-​­functional calculation, we get stuck unless we know the ­truth-​­values of both constituent sentences. For instance, we can easily and correctly assess the conditional ‘­If he is sitting, he is not running’ even when we know neither whether he is sitting nor whether he is running. In general, the Suppositional Rule does not make us overestimate the probabilities of conditionals, since the probability of the material conditional A ⊃ C is always at least as high as the conditional probability of C on A. Of course, we may overestimate the probability of the material conditional if we already overestimate the conditional probability, but the Suppositional Rule is not to blame for that. In what follows, we will treat the material interpretation of ‘­if’ as a working hypothesis, together with the defence of it just briefly sketched, including the Suppositional Rule as a heuristic. We must be correspondingly critical in our ­pre-​­theoretical assessments of conditionals, since they will usually rely on implicit applications of the rules. For modal epistemology, our main interest is in conditionals modalized with ‘­would’. How do we assess them? If we simply reuse the Suppositional Rule as for plain indicative conditionals, ‘­would’ becomes redundant, which it clearly is not: as examples above showed, it makes a difference. The fake past ‘­distances’ the supposed situation in which the antecedent holds from actuality, and the contextual restriction on ‘­would’ qualifies the supposed situation, not actuality. The Suppositional Rule can then be applied with respect to the supposed situation, as it were within the scope of the fake past. The overall way of assessing modalized conditionals feels quite like assessment by the Suppositional Rule, but it has a subtle counterfactual twist (­Williamson 2020:­ 189–​­213 has details). We can conveniently describe it in terms like those for

68  Timothy Williamson the original Suppositional Rule, but with ‘­counterfactually’ inserted to mark the difference. The contextual restriction in the semantics of ‘­would’ has pragmatic effects too when we assess counterfactuals. For counterfactually supposing the antecedent modifies the context: it makes new possibilities relevant. Consider a context before the counterfactual with antecedent A and consequent C was entertained in which no A-​­worlds were relevant. Once we entertain the counterfactual and start assessing it in the usual way, we counterfactually suppose A, and assess C on that supposition, which tends to make A-​­worlds relevant. When A is possible, there are A-​­worlds, and at least some of them will become relevant in the modified context. This effect is no part of the semantics, for ‘­would’ applies equally well to unconditional sentences. It is just pragmatic, but it plays a major role when we confront examples. 2 The alternative approach to the semantics of counterfactuals has significant implications for their logic. The next section focusses in particular on the repercussions for modal epistemology.

3 Applications to Modal Epistemology The S­ talnaker–​­Lewis approach to the semantics of counterfactuals invalidates the principle of Strengthening the Antecedent, as Section 1 noted: sometimes, when A counterfactually implies C, and A+ entails A, A+ fails to counterfactually imply C. Nozick’s modal epistemology exploits this phenomenon. Consider again (­1)–​­(­4): 1 2 3 4

I do not have hands. I am a handless brain in a vat who believes that it has hands. If I did not have hands, I would not believe that I had hands. If I was a handless brain in a vat who believed that it had hands, I would not believe that I had hands.

In normal circumstances, on Nozick’s view, although (­3) is true, and (­2) (­the antecedent of (­4)) entails (­1) (­the antecedent of (­3)), (­4) is not true. Indeed, the opposite counterfactual (­5) to (­4) is uncontroversially true, since its antecedent entails its consequent: 5 If I was a handless brain in a vat who believed that it had hands, I would believe that I had hands. Because (­4) and (­5) have the same possible antecedent and contradictory consequents, they seem to be inconsistent; therefore, since (­5) is true, (­4) is false. Thus Strengthening the Antecedent fails. To complete the picture, we may further assume that in all the worlds at issue in which I believe (­6) (­which is contradictory to (­1)), I

Modal Epistemology and the Logic of Counterfactuals  69 competently deduce (­7) (­which is contradictory to (­2)) and believe (­7) on that basis: 6 I have hands. 7 I am not a handless brain in a vat who believes that it has hands. Thus, given (­5), (­8) holds too: 8 If I was a handless brain in a vat who believed that it had hands, I would believe that I was not a handless brain in a vat who believed that it had hands. Compare (­9): 9 If I was a handless brain in a vat who believed that it had hands, I would not believe that I was not a handless brain in a vat who believed that it had hands. Because (­8) and (­9) (­like (­4) and (­5)) have the same possible antecedent and contradictory consequents, they seem to be inconsistent; therefore, since (­8) is true, (­9) is false. But (­9) is in effect Nozick’s Sensitivity condition for me to know that I am not a handless brain in a vat who believes that it has hands. Thus, on his view, the falsity of (­9) excludes me from knowing that I am not a handless brain in a vat who believes that it has hands. By contrast, the Sensitivity condition for me to know that I have hands is (­3), which is true in normal circumstances. Nozick’s modal epistemology allows me to know that I have hands, but even if I competently deduce that I am not a handless brain in a vat who believes that it has hands and believe it on that basis, I cannot thereby know that I am not a handless brain in a vat who believes that it has hands.3 The invalidity of Strengthening the Antecedent for counterfactuals is central to Nozick’s strategy, for if it were valid, the falsity of (­4) would entail the falsity of (­3): the insensitivity would extend to my ordinary belief that I have hands, and conceding the impossibility of knowing that one is not in a sceptical scenario would not enable one to retain the possibility of ordinary knowledge. On the alternative approach of Section 2 to the semantics of counterfactuals, Strengthening the Antecedent is valid. Formally, the counterfactual with antecedent A and consequent C is interpreted as the restricted strict conditional □(­A ⊃ C). Thus Strengthening the Antecedent becomes the principle that if A+ entails A, then □(­A ⊃ C) entails □(­A+ ⊃ C), where entailment is understood as unrestricted strict implication. If A+ entails A, every A+-​­world is an A-​­world. If, in addition, □(­A ⊃ C) is true, every contextually relevant A-​­world is a C-​­world. Consequently, every contextually relevant A+-​­world is a C-​­world, so □(­A+ ⊃ C) is true. Thus the alternative approach validates Strengthening the Antecedent.

70  Timothy Williamson In particular, Nozick’s strategy fails because (­3) entails (­4), on the alternative semantics. However, that it validates Strengthening the Antecedent looks like a problem for the alternative semantics. For the natural ­pre-​­theoretic assessments of (­3) and (­4) are that (­3) is true and (­4) false. That challenge neglects the contextualist aspect of the semantics. In an ordinary context, no worlds in which I am a handless brain in a vat who believes that it has hands are relevant; thus (­2) is false in all relevant worlds. The relevant worlds are much closer to actuality: in those in which I lack hands, I am straightforwardly aware of lacking hands, and do not believe that I have hands, so (­3) is true. But (­4) is also true in the ordinary context, vacuously, because in every relevant world the antecedent (­2) of the embedded material conditional is false, so the material conditional itself is true. But once we start entertaining (­4), and assessing it in the usual suppositional way, we change the context, by making relevant some worlds in which I am a handless brain in a vat who believes that it has hands. Those worlds falsify the antecedents and verify the consequents of the material conditionals embedded in (­3) and (­4). Consequently, both (­3) and (­4) are false in the new context. Thus the apparent failure of the entailment from (­3) to (­4) was a mere artefact of the unnoticed shift from the ordinary context in which we considered (­3) to the special epistemological context created by considering (­4). In any fixed content in which (­3) is true, (­4) is also true. The entailment holds. The contextualist account gains support from what happens when we reconsider (­3) immediately after considering (­4) (­this is an example of what are known in the literature as ‘­reverse Sobel sequences’). Now (­3) can easily look false, because we can still easily count as relevant the newly added worlds in which I am a handless brain in a vat who believes that it has hands, and they falsify the material conditional embedded in (­3). Which worlds count depends on the extrinsic context, not on the intrinsic semantics.4 How does this critique of Nozick’s modal epistemology compare with longstanding contextualist objections to his approach? It has in common with them the complaint that Nozick’s apparent cases of ­non-​­closure are artefacts of unnoticed ­context-​­shifting, and the suggestion that a suitably qualified principle of epistemic closure will hold without exception in each fixed context. However, the conclusion usually drawn is that Nozick’s counterfactual analysis of knowledge is incorrect, because it endorses those cases of ­non-​­closure (­compare Luper 1984; BonJour 1987; DeRose 1995; Roush 2005). On the present account, by contrast, what generates the ­non-​­closure is not the counterfactual analysis of knowledge but rather Nozick’s application of it, which ignores the contextual shiftiness of his counterfactuals themselves. In principle, an epistemologist could retain the letter of Nozick’s analysis, but reinterpret it as already an implicitly contextualist account of ‘­know’, fully consistent (­for all that Nozick has shown) with a suitably qualified principle of epistemic closure holding without exception in each fixed context.

Modal Epistemology and the Logic of Counterfactuals  71 For myself, I endorse neither Nozick’s analysis nor contextualism about ‘­know’ (­Williamson 2000, 2005). Nevertheless, counterfactual conditions are of epistemological interest in their own right, and we need to understand how they really work, once contextual effects are properly controlled for. Section 4 will show how to formulate appropriate closure principle for suitable variants of Sensitivity and Counterfactual Safety, without prejudice to their relation to knowledge. Another respect in which traditional modal epistemology relies on the ­Stalnaker–​­Lewis semantics concerns the distinction between Counterfactual Safety and Sensitivity, as Section 1 noted. Simple versions of those conditions are often formalized as contrapositives: for Counterfactual Safety, Bel[A] > A, for Sensitivity, ¬A > ¬Bel[A]. But the approach of Section 2 makes contrapositive counterfactuals logically equivalent. For A > C and ¬C > ¬A are analysed as □(­A ⊃ C) and □(¬C ⊃ ¬A) respectively, which are logically equivalent, because A ⊃ C and ¬C ⊃ ¬A are ­truth-​­functionally equivalent and □ is a normal modal operator. Thus Contraposition is valid. As with Strengthening the Antecedent, the validity of Contraposition seems to pose a problem for the alternative semantics, since it faces apparent counterexamples (­L ewis 1973: 35). We can illustrate them from modal epistemology. Imagine Con the conspiracy theorist, though with no great loyalty to any one such theory. He currently believes that he has been kidnapped by aliens and taken to another planet. Here is a Counterfactual Safety condition: 10 If Con believed that he was on Earth, he would be on Earth. A plausible ­pre-​­theoretic assessment of (­10) is that it is true: had Con believed a different conspiracy theory instead, set on Earth, he would still have been on Earth. The contrapositive (­11) of (­10) gives a Sensitivity condition: 11 If Con was not on Earth, he would not believe that he was on Earth. A plausible p ­ re-​­theoretic assessment of (­11) is that it is false: if Con was put in a rocket and sent away from Earth, he would assume that he was being held somewhere on Earth as part of an elaborate conspiracy. As before, the challenge neglects the contextualist aspect of the semantics. In an ordinary context, no worlds in which Con is not on Earth are relevant, so (­10) is true. Then (­11) is also true in the ordinary context, vacuously. Hence (­10) and (­11) have the same ­truth-​­value in the ordinary context. But once we start entertaining (­11), and assessing it in the usual suppositional way, we change the context, by making relevant some worlds in which Con is not on Earth. If, in those worlds, he believes that he is on Earth, (­11) is false, but so too is (­10). Hence (­10) and (­11) also

72  Timothy Williamson have the same ­truth-​­value in the new context. Thus the apparent failure of the equivalence between (­10) and (­11) was a mere artefact of the unnoticed shift from the ordinary context in which we considered (­10) to the ­wider-​­horizon context created by considering (­11). In any fixed content, (­10) and (­11) have the same ­truth-​­value. The equivalence holds. Again, the contextualist account gains support from what happens when we reconsider (­10) immediately after considering (­11). Now (­10) can easily look false, because we can still easily count as relevant the newly added worlds in which Con is not on Earth. Which worlds count depends on the extrinsic context, not on the intrinsic semantics. On reflection, like apparent counterexamples to Strengthening the Antecedent, apparent counterexamples to Contraposition emerge as mere artefacts of c­ontext-​­shifting; they support the contextualist account of counterfactuals on which both principles are valid with respect to any fixed context. Thus, in their present forms, Sensitivity and Counterfactual Safety are logically equivalent. Any failure of equivalence is pragmatic: considering them naturally creates different contexts, owing to their different antecedents, so Sensitivity in its natural context is not equivalent to Counterfactual Safety in its natural context. Nevertheless, in any one context, Sensitivity is equivalent to Counterfactual Safety. Which contexts a philosopher wishes to propose them for is another matter.

4 An Epistemic Closure Principle for a Counterfactual Condition Section 3 explained how an apparent case of n ­ on-​­closure for Sensitivity dissolves into a mere artefact of ­context-​­shifting. Assuming the alternative semantics for counterfactuals, this section formulates and validates closure principles for both Sensitivity and Counterfactual Safety; since they are equivalent, the same principle does for both. We do not assume any particular relation between either Sensitivity or Counterfactual Safety and knowledge. For simplicity, we start with the case of s­ingle-​­premise closure for Counterfactual Safety. The formula Bel[C / A] means ‘­One believes C on the basis of believing A’. Thus Bel[C / A] entails both Bel[C] and Bel[A]. The formula SafeBel[A] abbreviates the Counterfactual Safety condition □(­B el[A] ⊃ A), which can be paraphrased as ‘­If one believed A, A would be true’, or more briefly as ‘­A is safe to believe’. Similarly, the formula SafeBel[C / A] abbreviates the qualified Counterfactual Safety condition □(­B el[C / A] ⊃ C), which can be paraphrased as ‘­If one believed C on the basis of believing A, C would be true’, or more briefly as ‘­C is safe to believe given A’. We can now formulate a s­ ingle-​­premise closure principle for Counterfactual Safety:

Modal Epistemology and the Logic of Counterfactuals  73 ­SPC- ​­CS

If A entails C, then SafeBel[A] entails SafeBel[C / A].

Thus if A is safe to believe, then whatever A entails is safe to believe given A. In other words, if A entails C, and A would be true if one believed A, then C would be true if one believed C on the basis of believing A. Naturally, SafeBel[A] and SafeBel[C / A] must be understood with respect to the same context. We can establish ­SPC- ​­CS thus. Assume SafeBel[A] and that A entails C. Suppose that in a relevant world w one believes C on the basis of believing A. Thus in w one believes A. Given SafeBel[A], A is true in every relevant world in which one believes A, and so in particular A is true in w. But A entails C, so C is true in w too. Thus C is true in every relevant world in which one believes C on the basis of believing A, which is SafeBel[C / A]. QED. The premise SafeBel[A] is essential to the validity of ­SPC- ​­CS, even given that A entails C. Without it, one might believe C on the basis of believing A at a relevant world w even though both A and C were false at w. We can easily generalize ­SPC- ​­CS to a m ­ ulti-​­premise closure principle for Counterfactual Safety. The formula Bel[C / A1, …, An] means ‘­One believes C on the basis of believing A1, …, An’. Thus Bel[C / A1, …, An] entails both Bel[C] and Bel[A1], …, Bel[An]. Correspondingly, SafeBel[C / A1, …, An] abbreviates the qualified Counterfactual Safety condition □(­B el[C / A1, …, An] ⊃ C], which can be paraphrased as ‘­If one believed C on the basis of believing A1, …, An, C would be true’, or more briefly as ‘­C is safe to believe given A1, …, An’. We can now formulate the ­multi-​­premise closure principle for Counterfactual Safety, of which ­SPC- ​­CS is simply the special case for n = 1: ­M PC-​­CS

If A1, …, An jointly entail C, then SafeBel[A1], …, SafeBel[An] jointly entail SafeBel[C / A1, …, An].

Thus if each of A1, …, An is safe to believe, then whatever they jointly entail is safe to believe given them. In other words, if A1, …, An jointly entail C, and for each i Ai would be true if one believed Ai, then C would be true if one believed C on the basis of believing A1, …, An. Naturally, SafeBel[A1],  …, SafeBel[An], and SafeBel[C / A1,  …, An] in ­MPC-​­CS must all be understood with respect to the same context. The proof of ­MPC-​­CS is omitted, since it is an obvious generalization of the proof of ­SPC- ​­CS. The premises SafeBel[A1], …, SafeBel[An] are essential to the validity of ­MPC-​­CS, just as the premise SafeBel[A] is essential to the validity of ­SPC- ​­CS, even given the corresponding entailment, and for a similar reason.

74  Timothy Williamson Of course, since Counterfactual Safety is equivalent to Sensitivity with respect to a given context, a corresponding version of ­multi-​­premise closure holds for Sensitivity too. It says that if A1, …, An jointly entail C, and for each i one would not believe Ai if Ai was not true, then one would not believe C on the basis of believing A1, …, An if C was not true (­for any fixed context). The corresponding s­ingle-​­premise closure principle for Sensitivity is just the special case where n = 1. The validity of such ­multi-​­premise closure principles for natural versions of both Sensitivity and Counterfactual Safety suggests that treating either condition as necessary for knowledge would not jeopardize a corresponding ­multi-​­premise closure principle for knowledge too. If they are objectionable conditions on knowledge, it must be for some other reason. Of course, Dretske and Nozick regarded the failure of natural closure principles for knowledge as the right result. If the proposed conditions on knowledge do not deliver that result, once the counterfactual conditionals in them are properly understood, an epistemologist in their spirit might still stipulate that those counterfactual conditionals are to be assigned the meanings Dretske or Nozick wanted, perhaps those which some version of the S­ talnaker–​­Lewis approach assigns them, even if those meanings diverge from their actual meanings in English. Subtler means to the same end would be by stipulating that any occurrence of ‘­would’ in the analysis of knowledge is to range over just those worlds which Dretske or Nozick would have expected it to range over. However, such manoeuvres make vivid the unnaturalness involved in resisting an appropriate closure principle for knowledge.5

5 Methodological Moral In this paper, we have seen how treacherous counterfactual conditionals can be as servants of philosophical theorizing. We already had grounds to suspect them in that role, given the ‘­conditional fallacy’: they may fail to serve their purpose because realizing the antecedent could have unintended ­side-​­effects and alter the very thing we intended to probe for (­Shope 1978). The conditional fallacy does not depend on counterfactual conditionals’ contextual shiftiness, though the two may interact: the ­side-​­effects may themselves depend on context. The contextual shiftiness introduces another parameter which the theorist needs to keep under control. The content of a theoretical claim made with a counterfactual conditional varies with the context in which the claim is made or interpreted. Of course, the theorist can in principle stipulate a canonical context which the counterfactual is to be interpreted as if uttered in, but it may be hard in practice to filter out unconscious effects from the different context in which the discussion is actually taking place. It might be easier and safer to make the intended ­sub-​­domain of worlds explicit, rather

Modal Epistemology and the Logic of Counterfactuals  75 than relying on our shifty implicit understanding of the counterfactual. That is in effect to dispense with the counterfactual conditional itself. Might a contextualist in epistemology welcome the contextual shiftiness of counterfactual conditionals, using them in theoretical formulations to inject the desired shiftiness? That possibility was briefly raised in Section 3. The trouble is that although ‘­would’ is shifty, its shifts may not stay within the required lines. Epistemological contextualists typically envisage ‘­know’ and other terms of epistemic appraisal as varying on quite specific dimensions, such as a scale of epistemic standards. By contrast, ‘­would’ varies with contextual relevance, which in turn depends on whatever happens to be salient to participants in the conversation. It may well include factors which by most standards count as n ­ on-​­epistemic. Such uncontrolled contextual shiftiness is unlikely to favour theoretical insight. Counterfactual conditionals are treacherous indeed. In our key theoretical formulations, we may do better to stick to plain material ‘­if’ and more overtly restricted modal operators, keeping the action on the table rather than under it.6

Notes 1 For other semantic accounts of counterfactual conditionals as strict conditionals see von Fintel (­2001) and Gillies (­2007). However, they rely on a framework of dynamic semantics and do not provide a compositional semantics derived from independent semantic accounts of ‘­if’ and ‘­would’. ­ on-​­pragmatically 2 von Fintel (­2001) and Gillies (­2007) explain such effects n by building them into a dynamic semantics. However, such an account is unduly inflexible; in reality, the effects are not mandatory (­Williamson 2020: 228). Moreover, it is more explanatory to derive such effects from more general considerations, as here, rather than building them into the semantics by hand. 3 The underlying point still holds once Nozick’s further complications, such as relativization to methods, have been factored in. 4 For early resistance to Nozick’s approach on related lines see Wright (­1983). Sarah Moss (­2012) has responded on behalf of the ­Stalnaker–​­Lewis approach to appeals to reverse Sobel sequences; see Williamson (­2020: ­223–​ 2 ­ 28) for a reply. 5 See Dretske (­2013a) and (­2013b) and Hawthorne (­2013) for a debate on the merits of closure for knowledge. Williamson (­2009) discusses further complexities for versions of Safety which vary the proposition at issue. 6 Thanks to participants for discussion of a predecessor of this paper at the 2019 Edinburgh workshop on epistemic closure, and to an anonymous referee for useful comments on an earlier version of this paper.

References Bonjour, Laurence. 1987: ‘­Nozick, externalism, and skepticism’, in Luper 1987: ­297–​­313. DeRose, Keith. 1995: ‘­Solving the sceptical problem’, Philosophical Review, 104: ­1–​­52.

76  Timothy Williamson Dretske, Fred. 1970: ‘­Epistemic operators’, Journal of Philosophy, 67: ­1007–​­1023. Dretske, Fred. 2013a: ‘­T he case against closure’, in Steup et al. 2013: 2 ­ 7–​­40. Dretske, Fred. 2013b: ‘­Reply to Hawthorne’, in Steup et al. 2013: 5 ­ 6–​­59. Gillies, Anthony. 2007: ‘­Counterfactual scorekeeping’, Linguistics and Philosophy, 30: ­329–​­360. Hawthorne, John. 2013: ‘­T he case for closure’, in Steup et al. 2013: ­40–​­56. Lewis, David. 1973: Counterfactuals. Oxford: Blackwell. Luper, Steven. 1984: ‘­T he epistemic predicament: knowledge, Nozickian tracking, and skepticism’, Australasian Journal of Philosophy, 62: ­26–​­50. Luper, Steven (­ed.). 1987: The Possibility of Knowledge: Nozick and His ­C ritics. Totowa, NJ: Rowman and Littlefield. Luper, Steven (­ed.). 2003: The Skeptics. Farnham: Ashgate Publishing. Moss, Sarah. 2012: ‘­On the pragmatics of counterfactuals’, Noûs, 46: ­561–​­586. Nozick, Robert. 1981: Philosophical Explanations. Oxford: Clarendon Press. Ramsey, Frank. 1929: ‘­G eneral propositions and causality’, reprinted in Hugh Mellor (­ed.), Foundations: Essays in Philosophy, Logic, Mathematics and ­ 33–​­151. Economics. London: Routledge and Kegan Paul: 1 Roush, Sherrilyn. 2005: Tracking Truth: Knowledge, Evidence and Science. Oxford: Oxford University Press. Shope, Robert. 1978: ‘­T he conditional fallacy in modern philosophy’, Journal of Philosophy, 75: ­397–​­413. Sosa, Ernest. 1999: ‘­How to defeat opposition to Moore’, Philosophical Perspectives, 13: ­141–​­152. Sosa, Ernest. 2003: ‘­Neither contextualism nor skepticism’, in Luper 2003: 1 ­ 65–​­182. Sosa, Ernest. 2007: A Virtue Epistemology: Apt Belief and Reflective Knowledge, volume I. Oxford: Oxford University Press. Sosa, Ernest. 2009: Reflective Knowledge: Apt Belief and Reflective Knowledge, Volume II. Oxford: Oxford University Press. Stalnaker, Robert. 1968: ‘­A theory of conditionals’, American Philosophical Quarterly Monographs, 2: ­98–​­112. eds.). 2013: Contemporary Steup, Matthias, John Turri, and Ernest Sosa (­ Debates in Epistemology, 2nd ed. Oxford: ­Wiley-​­Blackwell. von Fintel, Kai. 2001: ‘­Counterfactuals in a dynamic context’, in Michael Kenstowicz (­ed.), Ken Hale: A Life in Language. Cambridge: MIT Press: ­123–​­152. Williamson, Timothy. 2000: Knowledge and Its Limits. Oxford: Oxford University Press. Williamson, Timothy. 2005: ‘­Contextualism, ­subject-​­sensitive invariantism and knowledge of knowledge’, Philosophical Quarterly, 55: ­213–​­235. Williamson, Timothy. 2009: ‘­Probability and danger’, The Amherst Lecture in Philosophy, 4: 1 ­ –​­35. http://­w ww.amherstlecture.org/­williamson2009/. Williamson, Timothy. 2020: Suppose and Tell: The Semantics and Heuristics of Conditionals. Oxford: Oxford University Press. Wright, Crispin. 1983: ‘­Keeping track of Nozick’, Analysis, 43: ­134–​­140.

5 The Skeptical Paradox and the Generality of Closure (­and Other Principles) Yuval Avnur

In this essay, I defend a solution to a skeptical paradox. The paradox I focus on concerns epistemic justification (rather than knowledge) and skeptical scenarios that entail that most of our ordinary beliefs about the external world are false. This familiar skeptical paradox hinges on a “closure” principle. The solution is to restrict closure, despite its first appearing as a fully general principle, so that it can no longer give rise to the paradox. This has some extra advantages. First, it suggests a general strategy that provides solutions to other versions of the paradox, not just those that depend on closure. Second, it clarifies the relation between the paradox and other kinds of skeptical problem.

1 A Skeptical Paradox The paradox I have in mind involves propositional justification, or what you have justification to believe (­even if you don’t believe it, or don’t believe it in a justified way), and skeptical scenarios that entail the falsity of your ordinary, contingent beliefs about your environment that you base on your senses. Two standard examples are that you are a brain in a vat in an otherwise empty world, and that you are a victim of Descartes’ demon. Let ‘­SK’ stand for your favorite such scenario, and let ‘­P ’ stand for your favorite ordinary proposition whose negation is entailed by S­ K—​ m ­ ine is [there are tables]. Here is the paradox presented as an argument: Skeptical Premise: You lack justification to believe that ~SK Connecting Premise: If you lack justification to believe that ~SK, then you lack justification to believe that P. Paradoxical Conclusion: You lack justification to believe that P. The conclusion is “­paradoxical” in the sense that its negation is so plausible (­for many ordinary beliefs), that the fact that any remotely plausible premises seem to lead to such a conclusion makes the argument a paradox. The paradox set, then, contains the premises and the negation of the Paradoxical Conclusion. The Connecting Premise connects DOI: 10.4324/9781003104766-6

78  Yuval Avnur our ignorance about SK to our ordinary beliefs. Some closure principle about justification will motivate this premise, since SK entails ~P. But, if we were considering a different kind of skeptical scenario, such as that you are currently dreaming, which does not entail ~P, we’d need a different Connecting Premise. This will inevitably depend on some other general epistemic principle.

2 The Premises Why do the premises of the paradox seem plausible? 2.1  The Skeptical Premise The Skeptical Premise is plausible because justification of the relevant, epistemic sort, requires some good evidence or argument, while there seems to be none for ~SK. Consider: Evidentialism: if you lack good evidence that some proposition is true, then you lack (­epistemic) justification to believe that proposition.1 Why is Evidentialism plausible? Arguably, it is due to the very nature of belief and evidence. To believe is to take something to be true, and evidence for something is simply an indication of its truth. Justification is a positive status. How could taking something true be justified if nothing indicates its truth? Perhaps there are reasons for taking something to be true that do not themselves indicate the truth of that thing, so they are not evidence. That you threatened to tickle me until I cry otherwise, for example, could be reason to believe whatever you say. But that is not the sort of reason at stake in the paradox, for it is obvious that I could have this sort of reason to believe ~SK, and it is obvious I have the other, ­evidence-​­requiring, “­evidential” or “­epistemic” sort of justification to believe P. So, we can take Evidentialism to define the kind of justification that the paradox is about, rather than being a substantive principle about all justifications. Evidential justification is what makes both the Skeptical Premise plausible and the Paradoxical Conclusion paradoxical. The term ‘­good’ in Evidentialism needs explication. What counts as good evidence arguably depends on the kind of proposition at issue. I take no stand here, on behalf of Evidentialism, on what good evidence is for, say, mathematical propositions and propositions about one’s own mental states. The propositions at issue are contingent propositions about the external world, such as P and ~SK. When it comes to such propositions, the important qualification introduced by ‘­good’ is that not just any indication of the truth of something is good evidence for that thing. That there are tables is a good indication that there are tables,

The Skeptical Paradox and the Generality of Closure  79 but it is not good evidence that there are tables. Good evidence requires, at least, independence from what it is evidence for. I do not offer an account of evidential goodness and independence, for lack of space and because Evidentialism is not my focus. My point is merely to explain Evidentialism’s plausibility, and to point out that the case for it seems fully general, since it applies to all beliefs, given their nature. Given Evidentialism’s initial plausibility, why is the Skeptical Premise plausible? We seem to lack good, empirical evidence for ~SK, even though we have sensory experiences whose contents entail ~SK. The problem with appealing to such evidence is not that it is q ­ uestion-​­begging against a skeptical position, 2 or that it problematically assumes that the Paradoxical Conclusion is false. We aren’t currently debating a skeptic, and we are assuming that the Paradoxical Conclusion must be false, in approaching this argument as a paradox. Rather, the problem is that, intuitively, the appeal to empirical evidence seems to depend on ~SK, in a way that runs afoul of the “­independence” or “­goodness” condition above. To appeal to empirical evidence is ultimately to appeal to how things seem or look. If you were neutral (­or negative) about ~SK, or about whether things are the way they look, then why would you appeal to how things look, empirically, as evidence for how things are? Instead, in appealing to how things look as evidence for something, you seem to be assuming that things are (­likely to be) the way they look. This at least suggests that any empirical evidence we might appeal to will depend on ~SK. And this suggests that such empirical evidence for ~SK is therefore not independent or good. Of course, many philosophers reject this suggestion, for a variety of reasons (­including a rejection of the internalist assumptions lurking throughout these paragraphs).3 But they have been aptly criticized elsewhere and are not my target here. On the other hand, a priori arguments for ~SK seem doomed, a priori, since they are attempts to establish a specific, contingent truth by “­pure thought” alone. That hasn’t stopped many philosophers from trying, but such attempts, such as Putnam’s (­1981) appeal to semantic externalism and appeals to an “­inference to the best explanation” (­e.g. Vogel 2005), have been aptly criticized elsewhere and are not my target either. To be clear, I have not attempted to refute the most popular strategy for solving the paradox, which is to reject the Skeptical Premise. Rather, I have merely presented an initial case for the Skeptical Premise, in order to motivate a closer look at the Connecting Premise. 2.2  The Connecting Premise: Closure The Connecting Premise seems plausible, in part, for the same reason that we seem to lack independent evidence for ~SK: empirical claims, such as P, seem to somehow depend on ~SK, and so if you lack justification to believe ~SK, there goes the basis of your justification to believe

80  Yuval Avnur P. But, setting this ­often-​­neglected intuition in favor of the Connecting Premise aside, it is often observed that the Connecting Premise derives its plausibility from some general principle such as closure, along with some innocuous assumptions. What is closure? Roughly: if you have justification to believe something, then you have justification to believe any known entailment of that thing. There are various formulations of this principle, and various problems for them. Some formulate the principle in terms of competent deduction, rather than known entailment.4 But the version that I think is most defensible is this: Closure: For all propositions A and B: If you have justification to believe the proposition [A & (­if A then B)], then you have justification to believe B. Here it might help to note that the propositional justification in the consequent amounts to this: you have what it takes, epistemically, to form a belief that B. This is compatible with your never coming to believe it, and with your coming to believe it in a way that is not (­doxastically) justified. Propositional justification is the abstract, basic notion of having sufficient epistemic support, in principle available, for a belief. But this need not be due to an inference from A, as a deductive formulation of the principle would suggest. Closure, or some variant of it, is held by most epistemologists to be not only plausible, but obviously true, and almost has the status of an axiom. This explains why the majority of ­anti-​­skeptical strategies target the Skeptical Premise, and the need, of those who do deny Closure, to come up with an explanation for the mass and intense appeal of Closure. The focus on Closure might seem puzzling given the bigger picture. If we were considering a different sort of skeptical scenario, one which does not entail ~P, then Closure would be irrelevant. To see why, consider the auxiliary assumption required for Closure to imply the Connecting Premise. The Connecting Premise says nothing about your awareness of the logical relation between P and ~SK. So, to derive the Connecting Premise from Closure, we must assume that whenever you have justification to believe P, you realize that P entails ~SK, so that you have justification to believe [P & (­if P then ~SK)]. Only now can we derive the Connecting Premise, since the relevant instance of the consequent of Closure is that you have justification to believe ~SK. According to Closure, then, if you lack justification to believe ~SK, the relevant instance of the antecedent of Closure is false: you lack justification to believe [P & (­if P then ~SK)]. According to the assumption, you therefore lack justification to believe P. In other words, if you lack justification to believe ~SK, given the assumption and Closure, then you lack justification to believe P. That’s the Connecting Premise. But now consider, instead of SK, the scenario in which you are dreaming. We should not assume

The Skeptical Paradox and the Generality of Closure  81 that you ever have justification to believe [if P then you’re not dreaming], since you presumably realize that you might be dreaming while asleep at a table. The assumption we used above is no good. In that case, we need some principle other than Closure to connect the dreaming scenario to P in a way that implies the relevant version of the Connecting Premise.5 For now, let us continue to focus on Closure, keeping in mind that the same strategy defended below should apply just as well to other such general principles, which support other versions of the paradox. Next, I describe the two main arguments for Closure. 2.2.1  The Argument from Deduction Here is one argument for Closure. Suppose you have justification to believe, about some propositions R and S, [R & if R then S]. This is an instance of the antecedent of Closure. Either your justification to believe R depends on some prior justification to believe S, or it does not. Suppose it does. Then an inference to S would be circular or ­question-​ ­begging. But, given (­as we are supposing) that you have justification to believe R, you must already have justification to believe S, as this is (­in part) what your justification to believe R depends on. So, in this case, you already have justification to believe S. Suppose instead that your justification to believe R does not depend on some justification to believe S. Then, the inference, from [R & (­if R then S)] to S, could provide justification to believe S. For, how could that inference fail to provide justification for S, given that your justification to believe the premise does not depend on any justification to believe S? It would not, in this case, be a circular or ­question-​­begging inference. You therefore have justification to believe S; you have what it takes, epistemically, to justifiedly conclude that S, even if you never do so. So, either way, you have justification to believe S, which is the consequent of Closure. Since its consequent can be derived from its antecedent for two arbitrary propositions, Closure must be true. Notice that this argument depends crucially on the idea that either your justification to believe R rests on justification to believe S, or else deducing S from [R & (­if R then S)] produces justification. There is no other way that a deductive inference can fail, according to this reasoning. But, as I explain in Section 4, this is dubious and is the reason the argument from deduction fails. 2.2.2  The Argument from Coherence A different kind of argument for Closure claims that any closure failure would make justification incoherent, or make possible that an incoherent state be justified. The best way to make this case is to claim that any epistemic notion for which Closure fails is probabilistically incoherent, in the sense of failing to conform to the probability calculus. Return to R

82  Yuval Avnur and S, from above, and recall that you have justification to believe [R & (­if R then S)]. In that case, if you are not as confident that, or assign as high a probability to, S as you do R, you are probabilistically incoherent. This might naturally be taken to mean that you would not have justification to believe R in such a case, or that any justification you did have would be undermined by your failure to believe S. Failure to believe (­with at least the same confidence) S would make you irrational, so you must have some justification to believe S.6 This argument is tempting but not valid, and at best supports a restricted version of Closure, as I explain in Section 4. But the arguments from deduction and coherence constitute the main case for Closure and, accordingly, the Connecting Premise.

3 Closure Denial The most popular strategy is to deny the Skeptical Premise, but my purpose here is to deny Closure, and thereby reject the Connecting Premise. There are different ways to deny the Connecting Premise. Here is a brief, critical summary of the alternatives to the one I defend. 3.1  Closure Fails Even in ­Non-​­Skeptical Cases The (­historically) first, and still most widely known, denial of Closure is Dretske’s (­1970, 2005). I will not go into the details, because I think it is the least promising approach. Closure fails, according to this view, even in relatively ordinary, or at any rate n ­ on-​­skeptical situations. For example, suppose you see a bottle on the table (­in normal circumstances), and it looks like there’s wine in it. You can have justification to believe that there’s wine in the bottle on the basis of the way it looks, while lacking justification to believe that the liquid in the bottle is not colored w ­ ater—​ i­t would look just the same if it was full of colored water instead. But such examples are implausible. You do seem to have at least some justification to believe such things as that the bottle does not have colored water in it. It’s just that this justification does not derive from inference from its looking like wine. Rather, your justification to believe that it is not water comes from background considerations, perhaps inductive information from past experiences with bottles and drinking. If you really lacked justification to believe that it is not water, then, intuitively, you would lack justification to believe it is wine. It seems that Dretske conflated different kinds of justification, concluding from the fact that you don’t have justification to believe it is not water by inference from how things look, that you lack any kind of justification. But independent, background justification is in place in all such ordinary cases; I’m unaware of any plausible ­non-​­skeptical instances of Closure failure, for this reason.

The Skeptical Paradox and the Generality of Closure  83 3.2  Mixed Closure Wright (­2004) and Coliva (­2015, ch. 3) both deny Closure, at least interpreted to be about evidential justification. They accept, in its place, a similarly general principle that mixes different epistemic statuses. The initial impetus for this strategy is that Closure for evidential justification is false. As Wright (­2004: 178) points out, this should not be surprising. He cites D ­ retske-​­style cases, such as the zebra case: at the zoo, your visual evidence supports that it is a zebra, but the same visual does not support that it is not a cleverly disguised mule. In other words, the same visual evidence justifies belief that it is a zebra, but not belief that it is not a cleverly disguised mule, even if you realize that zebras are not cleverly disguised mules. Specific (­in this case visual) evidential relations are not even apparently closed under known entailment. This is a crucial insight, and we can add a further point. In a formal or “­Bayesian” framework, evidence is a matter of confirmation (­where e confirms q just in case the probability of q given e is higher than the prior probability of q). But we know independently that confirmation closure, also known as Hempel’s “­special consequence principle,” fails.7 In other words, the “­makes more probable” relation is not closed under known entailment, and being supported by evidence arguably just is being made more probable by that evidence. So, why should the status of being supported by evidence be closed under known entailment? Not only are specific sorts of evidence unclosed, evidence in general is unclosed. Wright’s response to this was to preserve a general closure principle for a category, warrant, that includes evidential justification and entitlements. Briefly, an entitlement is a n ­ on-​­evidential, positive status that applies to trusting “­cornerstone” propositions, of which ~SK is one. While he offers a few different accounts of entitlement, one in particular is held to be relevant to Cartesian skeptical paradoxes of the sort at issue here, namely, entitlement of “­cognitive project” (­2004: 205; sections 5 and 6). This entitlement derives from the fact that, when investigating some empirical matter, such as p, we must presuppose ~SK, have no reason to doubt it, and would be engaging in circular reasoning if we were to try to establish it within this investigation. Setting aside various worries about entitlement,8 the suggestion is that a general closure principle can be saved if it is construed to be about ­warrant—​­the status of having either evidential justification or entitlement. Coliva (­2015) instead suggests that ~SK, and other cornerstones, or “­hinges” as she calls them, are necessary assumptions of epistemic rationality. Specifically, the assumption that ~SK is necessary for the acquisition of perceptual warrants (­or justification due to the evidence of the senses). The subject’s assumption that ~SK is therefore “­mandated” by epistemic rationality and is partly constitutive of epistemic rationality itself (­129). I find much to agree with here. But Coliva’s approach to

84  Yuval Avnur Closure is, like Wright, to posit a fully general version of it that is about, not evidential justification, but “­extended” epistemic rationality, which includes both evidential justification and those assumptions necessary for the acquisition of that justification. And both P and ~SK are epistemically rational in that extended sense, so mixed closure is satisfied. Below, I offer some criticism of the idea that assuming ~SK is necessary for perceptual warrants. For now, consider an objection to the general strategy of mixing ­non-​­evidential with evidential statuses in order to produce a general closure principle, due to Pedersen (­MS).9 Pedersen posits an obviously gerrymandered property, ‘­­blib-​­blob’: “­Something is a ­blib-​­blob if and only if it is divisible by 2 with no remainder, or has two hydrogen atoms and an oxygen atom, or is valid, or is blue, or loves soccer.” It’d be unsatisfying to try to vindicate a general mathematical principle, say, by replacing it with a ­blib-​­blob principle. Unless entitlement and rational mandates can be shown to be epistemically significant in the same way that evidential justification i­s—​­to bear on the likely truth of the proposition they apply ­to—​­warrant (­Wright’s) and rationality (­Coliva’s) will be ­blib-­​­­blob-​­like properties, with nothing unifying them. In that case, Closure for warrant and rationality is not worth defending, for it is trivial to come up with some disjunctive property, one of whose applications is to justified beliefs, such that that property is closed. Why should that make the failure of closure for justification any less problematic? On the other hand, the appeals to warrant and rationality were made precisely because it cannot be determined that ~SK has a property that is directly epistemically significant in the way that evidential justification is. The whole point of appealing to a n ­ on-​­evidential status is that, apparently, there is no positive epistemic status to attach to ~SK. So, the objection goes, the strategy of mixing Closure is bound to be unsatisfying. It should be noted, in defense of Wright and Coliva’s strategies, that entitlement and rational mandates are clearly closely related to evidential justification. It seems clear that what they have offered is not exactly like a b ­ lib-​­blob closure. Nevertheless, Pedersen’s objection presses us to go further, either in explaining how the ­non-​­evidential statuses are similar enough to the evidential ones for the purpose of solving the paradox, or in explaining why it is not so bad, after all, if Closure for evidential justification fails to hold. Here I pursue that second path. My main difference from the mixed strategy is that I doubt that some positive, n ­ on-​ ­epistemic status can adequately explain why Closure and the arguments for it fail. 3.3  Restricting Closure Finally, then, we come to the strategy that I think is most promising. On this view, Closure applies to a large set of cases, but it is not fully general, as it initially appears. Restricting Closure in this way should ­ on-​­skeptical exceptions to Closure, avoid accommodate the absence of n

The Skeptical Paradox and the Generality of Closure  85 Pederson’s objection, and explain why the arguments for Closure fail.10 Rather than posit a general, mixed principle in order to compensate for the failure of Closure for evidential justification, as Wright and Coliva do, this strategy posits no such general principle to replace Closure. Instead, it refutes the arguments in favor of Closure directly: we shouldn’t have ever thought there is a fully general principle here. A similar strategy has been suggested for other great paradoxes. For example, some solutions to Russell’s paradox (­including Russell’s) proceed by restricting the principle of properties and sets, so that it does not apply generally to all properties. And some solutions to the liar paradox restrict principles governing the truth predicate that initially appear fully general. The present strategy is similar, with Closure as the apparently general principle. Restricting Closure requires the following steps, each of which is described below, in Sections 4.­1–​­4.3. First, distinguish a special set of propositions which includes ~SK (­and, preferably, other skeptical scenarios). The more natural and n ­ on-​­ad hoc this set is, the better. Second, exclude this set of propositions from Closure, by restricting its quantifier ‘­For all’ to n ­ on-​­special propositions, and show how this explains the failure of the arguments for Closure. Finally, attempt to explain why unrestricted Closure may have seemed plausible even though it is false. The result is that we can deny the Connecting Premise, because it rests on an instance of Closure that concerns ~SK, and we thereby avoid the Paradoxical Conclusion.

4 How to Restrict Closure This is the main, positive section of the chapter, in which I develop the solution to the paradox. 4.1  The Special Set The special set of propositions is determined by the very notion of epistemic justification. Justification is not a neutral (­or “­formal”) notion, it is biased towards some contingent contents, and those contents are the special propositions to be excluded from ordinary epistemic principles.11 Any theory of justification on which the Paradoxical Conclusion is false, so that many of our perceptual beliefs are justified, is biased in this way. For example, some, called ‘­Dogmatists,’ hold that it’s looking to you as if P, and your lacking any relevant defeaters, is sufficient for (­prima facie) justification to believe P.12 The notion of justification thereby facilitates an assumption that things are generally the way they look, unless you have evidence otherwise. You m ­ ay—​­it is good t­ o—​­take your perceptual beliefs to be faring well with respect to the truth, simply by virtue of their being based on perception, so the implicit starting assumption you may help yourself to even if implicitly, according to this

86  Yuval Avnur way of evaluating beliefs, is that you are probably not massively deceived. Others, called ‘­Conservatives,’ hold that you have, by default, a (­prima facie) justification to believe, or accept, or trust, or assume, that things are the way they look.13 So as long as this justification is undefeated by other evidence, you have justification to believe that P if it looks like P. The difference between the two is that, for Conservatives, your perceptual justification to believe P is mediated by, or dependent on, this other justification to assume that things are the way they look; for Dogmatists, your justification to believe P is immediate. Both have in common the idea that, one way or another, the way we evaluate beliefs has a b ­ uilt-​­in bias toward the proposition that we are not massively deceived, since a justified subject may operate doxastically on that starting assumption. Why is the notion of justification biased in this way? This question is important in order to show that the special category I will soon distinguish on the basis of this bias is not ad hoc. Whether or not we ever occurently entertain it, we typically all assume, and assume that everyone else assumes, that we are not massively deceived, or, a little more precisely, that we can arrive at the likely truth through the use of our faculties. Call this assumed proposition ‘­T,’ for Truth. We evaluate beliefs according to this starting assumption.14 Similarly, a population in which everyone assumes that all Germans are cruel might be expected to employ a term such as “­boche.”15 The primary way of evaluating beliefs is to see if they are justified, and this notion of justification applies to what our faculties ultimately produce (­absent defeaters), and implies (­from the subject’s perspective) that the justified belief is likely true. Thereby, the notion is biased toward T: if your faculties deliver P, then all else equal you have what it takes, from this epistemic perspective, to believe P with justification. Why do we all assume T, and assume that everyone else assumes it? Briefly, any creature with beliefs and sources of information must find itself assuming T when epistemically evaluating beliefs for whether they are supported by those sources and must be unable to produce independent basis for believing T. For, any appeal to an informational basis for believing, and hence for believing T, must assume T. Our evaluation of beliefs must assume something, T, that cannot be positively evaluated. This explains why we use such a notion of justification and provides a defense of, perhaps a reason for, using it.16 Unlike Wright, I do not appeal to this observation in order to show that we are entitled to trust T. Rather, my point is only to show that it is not arbitrary, or a mere matter of convention, that our notion of justification displays a bias toward T. I note, too, that if we did not evaluate beliefs on the basis of an assumption that T, the Paradoxical Conclusion would not seem so intensely paradoxical in the first place. In other work, I defend and detail the idea that justification is loaded with this bias, and explain how it gives rise to the paradox.17 My purpose here is to show how this simple observation can also serve to distinguish a special set of propositions which are naturally excluded from principles

The Skeptical Paradox and the Generality of Closure  87 such as Closure. When it comes to our beliefs about the world around us, T implies ~SK and the negation of any other skeptical scenario that initially seems to raise a skeptical paradox. A bias toward the proposition that you are capable of getting at the likely truth about your environment by the use of your faculties is, all else equal, also a bias toward the proposition that your senses are not massively deceiving you, because otherwise you could not get at the truth about it. (­This is so regardless of whether the subject, whose justification is in question, ever explicitly entertains SK.) And the various skeptical scenarios are just ways for your senses to be massively deceiving you. So, justification is biased toward T, and this means that perceptual justification is biased toward ~SK. The bias of justification, which includes ~SK, is merely prima facie; it sanctions a starting assumption on the part of the thinker. The bias can be overwhelmed by certain courses of evidence. One can imagine cases in which one gains justification to believe some skeptical scenario. For example, one might find, at the bottom of one’s visual field, messages purporting to be from the crew running the vat that one’s brain is in (­in ­ rains-­​­­in-­​­­a-​­vat [BIV] scenario). These messages might make predicthe b tions about the course of one’s experiences which come true. Surely, with the messages’ reliability established, this is some good evidence that one is a BIV and may even provide justification to believe it. If it is possible to be a BIV, then it is possible for the scientists to help you to justifiedly believe it in some such way. In such a case, one need not, and should not believe, assume, or accept ~SK, even though some of one’s beliefs, such as the belief that SK, are justified. The observation that one’s course of experience could generate justification to believe SK constitutes an objection to two other, recent rejections of the Connecting Premise. The first is Coliva (­2015, ch. 4), according to which propositions such as ~SK are mandated by epistemic rationality, because they must be assumed in order for us to have perceptual justification at all. That one could have perceptual justification that one is a BIV while presumably no longer assuming that one is not a BIV seems to refute the claim that the latter assumption is required for perceptual justification. One can still rationally respond to perceptual evidence, even when one no longer assumes ~SK. Another recent attempt is Pritchard (­2016, ch. 4.1). On this view, propositions such as ~SK (­as well as what he calls ‘­uber hinges’ such as T) are not proper, “­­knowledge-​­apt” beliefs at all, and for this reason, Closure does not apply to them, and the Connecting Premise is false. ~SK is not a ­knowledge-​­apt belief because it is entirely unresponsive to rational considerations.18 But, as we’ve just seen, it is possible to rationally come to believe some SK, so ~SK cannot be disqualified as a possible content of belief in this way. The rational thing to do in the case described above is to reject ~SK, and I see no reason to suppose one couldn’t do the rational thing there. Our usual commitment to ~SK is also rationally responsive in a more temporary way. Recall Descartes’ observation in

88  Yuval Avnur the first meditation: “­T he result [of considering the skeptical scenario of dreaming] is that I begin to feel dazed, and this very feeling only reinforces the notion that I may be asleep.” He later observes, as does Hume after him, that such effects don’t last long. But these are still responses, albeit temporary, to rational considerations.19 Since our attitudes about SK are not unresponsive to rational considerations, there seems to be no reason to legislate that they aren’t proper beliefs. Notably, in the above scenario, one is still employing the notion of justification that is biased toward T. After all, it is by the use of one’s faculties, including reasoning and understanding as well as sensory experience, that one arrives at the conclusion that one’s senses are no longer to be trusted as before, and that is precisely what generates justification to believe that one is a BIV. 20 So, what is the special set distinguished by this account of justification’s bias? The notion of justification at issue in the paradox, which makes sense of the repugnance of the Paradoxical Conclusion, is biased toward some propositions, including T and ~SK, and those propositions are the special ones. We have seen that one might, though, come to learn that SK is t­ rue—​­perhaps one could even come to learn, for some specific SK, that SK is false. But that is the result of further investigation, not due to the a priori nature of justification itself. Even if, for example, we find some successful a priori argument for ~SK, the notion of justification is still biased toward some propositions, and it is these propositions that are special, and excluded from principles like Closure, as I explain next. 4.2  Exclude the Set from Closure In Section 2, I described the arguments from Deduction and Coherence. We can now appreciate where they go wrong when it comes to the bias loaded into justification, and why, at best, they instead support: Closure*: For all ­non-​­special propositions A and B: If you have justification to believe the proposition [A & if A then B], then you have justification to believe B. This is Closure restricted. It applies to everything aside from what justification itself is biased toward. To be clear, Closure* does not imply that one cannot, if conditions are right, get justification to believe B. Rather, Closure* is the closest thing to Closure that can be successfully argued for a priori, or independently of any particular course of experience or argument for B. The instance of Closure that is of interest here involves ~SK as B, in the consequent. And I do not deny that one could undergo experiences, or come up with a successful a priori argument, to produce inferential justification to believe ~SK. The point rather is that the a priori arguments that appeared to support Closure and hence the

The Skeptical Paradox and the Generality of Closure  89 Connecting Premise really only (­at best) support Closure*, which does not support the Connecting Premise. So, there is no in principle support for the Connecting Premise, and this is enough to solve the paradox. If one gained some justification to believe ~SK, then the Skeptical Premise could be denied instead (­the antecedent of the Connecting Premise would then be false). Either way, the paradox would be solved. The task now is to explain why the arguments given for Closure fail and support Closure* instead. 4.2.1  Failed Inferences and Deduction Begin with the argument from deduction, from Section 2.2.1. What is wrong with the following, N ­ eo-​­Moorean inference, which illustrates why we lack independent evidence for ~SK? NM1: I have hands (­look, here they are!) NM2: If I have hands, then ~SK (­because SK entails that I have no hands) So, NM3: ~SK Wright famously coined the term ‘­transmission failure’ for this sort of case. Roughly, the idea is that transmission failure occurs when a premise, such as NM1, depends for its justification in part on some justification (­or “­warrant” from entitlement, as Wright terms it) one already has for the conclusion, NM3. So, the argument cannot provide any new justification to believe the conclusion. ­ eo-​­Moorean inference exhibit this sort of transmission But does the N failure? We are currently considering the strategy according to which the Skeptical Premise is true, so there is no justification to believe ~SK, and thus no justification for NM3 for NM1 to depend on. (­A nd we are setting aside Wright’s “­entitlement” to trust NM3.) Justification for NM1 is, as the dogmatists say, immediate. So, no, the ­neo-​­Moorean inference does not exhibit this failure. Instead, the ­Neo-​­Moorean inference fails to generate justification due to a different reason: the very notion of ­justification—​­of being supported by (­perceptual) ­evidence—​­is biased toward ~SK, or NM3. In regarding NM1 as a premise fit to be used in an argument in which you are trying to determine whether NM3 is true, you thereby regard NM1 as justified (­even if not explicitly in those terms). This is a primary function of the notion of justification, to determine which propositions you should infer the truth from. But, to regard any empirical premise as justified is to use a notion that is biased toward ~SK/­NM3, which is the conclusion of this argument. That is no way to earn justification for something,

90  Yuval Avnur by appeal to a notion that is biased toward it! You may as well appeal to the notion of boche to establish that Germans are cruel. Of course, there is nothing special about NM1 or hands; this happens because the conclusion, NM3, is loaded into the very notion of justification, which is how we evaluate a premise’s fitness for use in an argument. In appealing to any premise in such an argument, you commit yourself to the claim that that premise is justified, but justification is biased toward ~SK, so no premise can be used in such a direct argument to generate or specify a justification to believe ~SK (­where ‘­direct’ means to exclude the sort of odd cases, mentioned above, involving things like messages in one’s visual field, the reliability of which might be established without an assumption that ~SK, though it would still require the assumption that T). 21 This explanation of why the ­Neo-​­Moorean inference fails does not commit us to the view that we have an antecedent justification to believe ~SK, on which our empirical premises depend for their justification, so it is not an appeal to Wright’s transmission failure. Rather, the inference fails because of how our notion of justification works. Return now to the question of why the propositions toward which the notion of justification is biased are to be excluded from Closure. The answer is that such propositions systematically evade the argument from deduction. That argument assumed that the only way for an inference to fail is for the conclusion to already be justified (­and for the premise’s justification to rest on that justification). Because justification is loaded with a bias toward ~SK, ~SK cannot be justified by any direct inference from empirical premises about the world. (­Contrast the case above, where the BIV lab sends you messages in your visual field or as voices in your mind. There, you appeal to sensory experiences, but not to objects in the world that are their contents.) But it is not the case, as the argument from deduction implies, that therefore ~SK must itself be justified, as if justification must apply to the propositions that it is biased toward. Rather, ~SK is a sort of blind spot of justification, in principle (­again, absent any bizarre sequence of experiences or a priori arguments for ~SK), because justification is biased toward it. So, there is more than one way for an inference to fail. 22 Objection: if ~SK cannot be justified in a direct way from empirical premises, shouldn’t one just discharge the assumption. Or stop believing it? Reply: No, that is not what epistemic justification requires; arguably, you should be at least as confident in ~SK as you are in propositions that entail it, as I explain in the next subsection. Still, the idea that a deductive inference fails only if the premise depends for its justification on the conclusion is actually a useful one with no apparent exceptions in ordinary life. It may hold perfectly well for n ­ on-​­special propositions, which justification is not biased toward. And, therefore, the argument seems plausible initially and supports a Closure principle applied only to ­non-​­special propositions. That is, it supports Closure*.

The Skeptical Paradox and the Generality of Closure  91 4.2.2  Coherence Requirements Next is the argument from coherence, described in Section 2.2.2. It is invalid, since it conflates necessary and sufficient conditions for justification. Recall that the argument is based on the observation that, if you are confident that [P & (­if P then ~SK)], and are not as confident in ~SK, you are probabilistically incoherent and hence (­with respect to these propositions) irrational. Surely justification follows what it is rationally required for you to believe, so that you must have justification to believe ~SK insofar as you have justification to believe [P & (­if P then ~SK)]. The problem with this argument is that, at best, conformity to the probability calculus is a necessary condition on one’s credence being justified.23 It is terrifically implausible that conformity to the probability calculus is a sufficient condition for being justified. If that were so, then I could have justification to believe that my table is conscious and wishes it were a squirrel, so long as I have coherent attitudes toward logically related propositions, for example, I also lack belief that it is not conscious. So, at most, all that the argument from coherence shows is that unless I also believe ~SK, I cannot have justification to believe P. That does not yet imply that I have justification to believe ~SK, because it may be that some beliefs for which one may lack justification are required for other beliefs to have justification. This is not as weird as it might initially sound. Arguably, as we’ve seen, having evidence is required for justification to believe anything. But the state of having evidence is not itself even a belief (­even if it requires one), and so it is not something you can have the same sort of epistemic justification for. The same is presumably true of being conscious, if conscious experience is required for perceptual justification. Perhaps some special beliefs are just like that: requirements for having justification, but not themselves things that have justification (­unless, again, some bizarre sequence of experience, or magnificent a priori proof, generates justification for them). Believing ~SK is a requirement for having perceptual justification about objects in your environment, absent evidence that SK is true, but it is not itself justified. ~SK plays this special epistemic role, of being an unjustified (­in the absence of evidence for or against it) requirement, because justification is biased toward it. So, the argument from coherence, for Closure, is invalid. But it may support Closure*. Grant that unless you believe ~SK, you lack justification to believe P. If justification itself is loaded with a bias toward ~SK, and if our use of this notion is explained by the fact that we all typically assume that ~SK, then this necessary condition makes sense, as long as you have justification for propositions like P (­and so lack justification for SK). It would make no sense for you to regard yourself as having justification to believe P while doubting what that notion is loaded with. Perhaps regarding yourself as having justification is not a necessary

92  Yuval Avnur condition on having justification. But if even an ideally conceptually sophisticated adult could not coherently regard her own belief that P as justified (­because she doesn’t even believe the starting assumption built into justification, and not because she has any evidence against it) were it to come up, then that suggests that no one, in similar circumstances, could be justified. This might explain, and at least accommodates, the idea that belief that ~SK, when ~SK comes up, is a requirement for having justification to believe ordinary things like P. 4.3  Explaining Closure’s Appeal What about Closure’s immediate, intuitive appeal? Setting aside special propositions, which never come up in everyday contexts and play a special role in our epistemic evaluation, Closure* and Closure are equivalent. So, it is not surprising that Closure seems initially true, especially given that Closure is the simpler of the two. It might be suggested, though, that Closure still seems more intuitive than Closure*, even once the theory behind restriction is taken into account. 24 Maybe so, but this comes with the territory of paradox. Had we restricted Evidentialism instead, it probably would have remained intuitive, too. This is the case with other paradoxes whose solutions require restricting general principles. After all, does it not still seem intuitive that, for every property, there is a set of all and only those things that have the property? And yet, we know that this is not only false, but ­self-​­contradictory!

5 Other Paradoxes and Other Problems 5.1  Other Paradoxes Other versions of the paradox ultimately appeal to some general principle in support of a connecting premise. Such general principles are motivated by consideration of everyday, or at any rate ­non-​­skeptical cases. But, since the notion of justification is biased toward the negation of any skeptical scenario that raises such a paradox, those general principles can be naturally and harmlessly restricted to apply only to the ­non-​ ­skeptical cases. So, any paradox will be ripe for a restriction solution. For example, some paradoxes appeal to this principle instead of Closure: Underdetermination: For all propositions, A and B: if you know that A and B are inconsistent, and your evidence does not support A more than B, then you do not have justification to believe A. 25 It would follow, letting P and SK be instances of A and B, and given that you know that P and SK are inconsistent, that if you lack evidential justification to believe ~SK, then you lack justification to believe P. For

The Skeptical Paradox and the Generality of Closure  93 if you had justification to believe P, your evidence would favor P over SK (­according to Underdetermination). And if your evidence supported P over SK, that would presumably provide justification to believe ~SK. So, Underdetermination seems to support the Connecting Premise. But, if justification is loaded with a bias toward ~SK, then P and ~SK could not be on equal ground before consideration of evidence. The reason you cannot appeal to some evidence for P as a basis of justification to believe P over SK, is that, in judging P to be justified by the evidence, you are already assuming a bias toward ~SK. So, ~SK should be excluded from Underdetermination for just the same sort of reason as Closure, and this explains the failure of Underdetermination: it should be restricted to ­non-​­special propositions. Accordingly, no connecting premise concerning ~SK can be supported by such a principle. For another example, consider paradoxes involving dream scenarios, which do not entail ~P. The bias toward T implies the negation of such scenarios in just the same way that it implies ~SK. If you can get at the truth by the use of your faculties, then you’re not dreaming. So, the strategy of restriction can be used to exclude other skeptical scenarios besides SK’s, and so to restrict whatever general principle connects such scenarios with our ordinary beliefs. 5.2  Other Problems The strategy of restricting closure removes one argument, in the form of a paradox, that purports to show that our notion of justification is empty or incoherent. But why should we use, and care about, this notion of justification rather than another, perhaps one that is not biased toward ~SK? This is a different p ­ roblem—​­just as “­W hy care about what’s moral?” is different from “­Is anything moral?” or “­Is morality coherent?” This is a practical question: which tool should we use to evaluate beliefs? Insofar as it is difficult to answer, this is a worthy skeptical problem.26 Another, related question is whether T is true. This is a metaphysical question about how we are causally connected to the world, but one of profound epistemological importance. If it turned out that some skeptical scenario was actually playing out, then our notion of justification, loaded as it is with a bias against the actual situation, would be an unfortunate concept for one to use. Is our notion defective in this way? The only way to answer this is to find out whether T. And no epistemological, a priori theory of justification can help here. How could conceptual legislation about ‘­justification’ determine whether, as it happens, T is true?27

Notes 1 Some philosophers use ‘­evidentialism’ to express an analysis of justification, which gives necessary and sufficient conditions. I use it here in a weaker sense, to specify only a necessary condition on being epistemically justified.

94  Yuval Avnur See Avnur (­2011: ­178–​­80) for discussion of ‘­epistemic’ and connection to the truth, in relation to the paradox. 2 See Wright (­1991: 89) on the “­adversarial stance,” and Pryor (­2000: 517) on the “­modest a­ nti-​­skeptical project.” 3 Two different, classic cases are Pryor (­2000) and Williamson (­2000). 4 See Williamson (­2000), Hawthorne (­2005), and Pritchard (­2016). 5 See Stroud (­1984, ch. 2). 6 In Avnur (­2012a), consider another version of the argument from coherence. 7 See Hempel (­1965, 31). 8 Wright himself considers an objection to entitlement, the “­leeching” problem (­209), which has remained a live objection even today. See Coliva (­2015, ch. 2, section 2) for a helpful survey of objections to Wright. See Avnur (­2011) for a more general objection to any appeal, in the context of skepticism, to a warrant that does not require evidence. 9 For Pedersen’s positive view, see Pedersen (­2020). 10 P.F. Strawson’s (­1985) view is at least compatible with this approach, though he never presented it in relation to Closure. See Avnur (­2012c, 2019) on how my view differs. 11 See Strawson (­1952) for a similar view about “­reasonable belief,” in response to the problem of induction. 12 Pryor (­2000, 2004), Huemer (­2001). 13 Wright and Davies (­2004), White (­2006). 14 See Goldman (­1993) for one development of this idea, and Avnur (­2019, section 2) for discussion. ­ 9–​­71). 15 The example is from Dummett (­1973), discussed in Brandom (­2000: 6 But, though this is a convenient example, the view that justification is loaded with a bias carries no commitment to inferentialism, as I explain in Avnur (­2019). For a similar use of the analogy between boche and justification, see Lange’s (­2008) discussion of induction. 16 See Avnur (­2012c, 2019) and Pritchard (­2016: ­97–​­8) for a Wittgensteinian version of this idea. 17 Avnur (­2019). ­ 1–​­2, ­101–​­2) on the exclusion of n ­ on-​­beliefs from clo 18 See Pritchard (­2016: 9 sure and on our seeming to ourselves to believe ~SK. 19 It is also debatable whether responsiveness to rational considerations is a necessary condition on belief. See Avnur (­2020) for some discussion. 20 One could even find oneself in a situation in which even T comes into doubt: perhaps you learn, by the use of your faculties, that you have no way to arrive at the truth: your evidence is contradictory and impossible to explain. This is like Hume’s ‘­consequent’ skeptical result, a sort of reductio situation, in which, though you start out assuming something, on that basis you learn that the assumption is wrong. 21 But see Barnett (­2014) for various issues that arise for this ­over-​­simplified claim about “­circular” arguments. 22 For more on this, see my (­2012a, 2012b). Coliva and I independently arrived at similar ideas on this second sort of transmission failure. See Coliva (­2012, 2015, ch. 3, section 3). Others have motivated other conditions for inference failure besides Wright’s. See Silins (­2008), Kung and Yamada (­2010), and Weisberg (­2010). 23 Though this is also implausible. See Christensen (­2004) for discussion. 24 Thanks to Adam Pautz for discussion. 25 See Brueckner (­1994), Cohen (­1998), Vogel (­2004), and Pritchard (­2005, 2016).

The Skeptical Paradox and the Generality of Closure  95 26 See Avnur (­2019) and Coliva (­2015, ch. 4.4, 4.5) on relativism and the “­Oblomovian” problem. 27 Thanks to Dominic Bailey, Adam Pautz, and Duncan Pritchard for discussion.

References Avnur, Yuval (­2011). An old problem for the new rationalism. Synthese 183 (­2):­175–​­85. Avnur, Yuval (­2012a). Closure reconsidered. Philosophers’ Imprint 12. Avnur, Yuval (­2012b). Mere faith and entitlement. Synthese 189 (­2):­297–​­315. Avnur, Yuval (­2012c). In defense of secular belief. Oxford Studies in Philosophy of Religion 4. Avnur, Yuval (­2019). The nature and limits of skeptical criticism. International Journal for the Study of Skepticism 9 (­3):­183–​­205. Avnur, Yuval (­2020). What is wrong with agnostic belief? In Gavin Hyman and Francis Fallon (­eds.), Agnosticism: Explorations in Philosophy and Religious Thought. Oxford: Oxford University Press. Avnur, Yuval (­ 2019). Justification as a loaded notion. Synthese 198 (­ 5):­ 4897–​­4916. Barnett, David James (­2014). What’s the matter with epistemic circularity? Philosophical Studies 171 (­2):­177–​­205. Brandom, R. (­2000). Articulating Reasons: An Introduction to Inferentialism. Cambridge: Harvard University Press. Brueckner, Anthony (­1994). The structure of the skeptical argument. Philosophy and Phenomenological Research 54 (­4):­827–​­35. Christensen, David (­2004). Putting Logic in Its Place: Formal Constraints on Rational Belief. Oxford: Oxford University Press. Cohen, Stewart (­1998). Two kinds of skeptical argument. Philosophical and Phenomenological Research 58 (­1):­143–​­59. Coliva, Annalisa (­2012). Varieties of failure (­of warrant transmission: what else?!). Synthese 189 (­2):­235–​­54. Coliva, Annalisa (­2015). Extended Rationality: A Hinge Epistemology. London, England: ­Palgrave-​­Macmillan. Dretske, Fred (­ 1970). Epistemic operators. The Journal of Philosophy 67 (­24):­1007–​­23. Dretske, Fred (­2005). The case against closure. In Ernest Sosa and Mathias Steup (­eds.), Contemporary Debates in Epistemology, 1 ­ 3–​­25. Malden, MA: Blackwell. Dummett, M. (­1973). Frege: Philosophy of Language. London: Duckworth. Goldman, Alvin I. (­1993). Epistemic folkways and scientific epistemology. Philosophical Issues 3:­271–​­285. Hawthorne, John (­2005). The Case for Closure. In Matthias Steup  & Ernest Sosa (­eds.), Contemporary Debates in Epistemology, ­26–​­43. Blackwell. Hempel, Carl. (­1965). Aspects of Scientific Explanation and Other Essays in the Philosophy of Science. New York: Free Press. Huemer, Michael (­2001). The problem of defeasible justification. Erkenntnis 54 (­3):­375–​­397. Kung, Peter and Yamada, Masahiro (­2010). A neglected way of begging the question. American Philosophical Quarterly 47 (­3):287.

96  Yuval Avnur Lange, M. (­2008). Hume and the problem of induction. In D. Gabbay, S. Hartmann and J. Woods (­eds.), Handbook of the History and Philosophy of Logic, Vol. 10: Inductive Logic, ­43–​­92. Hoboken: Elsevier. Pedersen, Nikolaj (­2020). Pluralist consequentialist a­ nti-​­scepticism. In P. Graham and N. J. L. L. Pedersen (­eds.), Epistemic Entitlement, ­297–​­326. Oxford: Oxford University Press. Pedersen, Nikolaj (­MS). N ­ on-​­evidentialist epistemology, closure, and the question of unity. Pritchard, Duncan (­2005). The structure of sceptical arguments. Philosophical Quarterly 55 (­218):­37–​­52. Pritchard, Duncan (­ 2016). Epistemic Angst: Radical Skepticism and the Groundlessness of Our Believing. Princeton, NJ: Princeton University Press. Pryor, James (­2000). The skeptic and the dogmatist. Noûs 34 (­4):­517–​­49. Pryor, James (­ 2004). What’s wrong with Moore’s argument?  Philosophical Issues 14 (­1):­349–​­378. Putnam, Hilary (­1981). Reason, Truth and History. New York: Cambridge U ­ niversity Press. Silins, Nicholas (­2008). Basic justification and the Moorean response to the skeptic. In Gendler, Tamar Szabo & Hawthorne, John (­eds.) Oxford Studies in Epistemology Volume 2, ­108–​­142. Oxford University Press. Strawson, P. F. (­1952). Introduction to Logical Theory. New York: Wiley. Strawson, P. F. (­1985). Scepticism and Naturalism: Some Varieties. New York: Columbia University Press. Stroud, Barry (­1984). The Significance of Philosophical Scepticism. Oxford: Oxford University Press. Vogel, Jonathan (­ 2004). Skeptical arguments. Philosophical Issues 14 (­1):­426–​­55. Vogel, Jonathan (­2005). The refutation of skepticism. In Steup Matthias and Sosa Ernest (­eds.), Contemporary Debates in Epistemology, ­72–​­84. Malden, MA: Blackwell. Weisberg, Jonathan (­2010). Bootstrapping in general. Philosophy and Phenomenological Research 81 (­3):­525–​­48. White, Roger (­ 2006). Problems for Dogmatism.  Philosophical Studies  131 (­3):­525–​­557. Williamson, Timothy (­2000). Knowledge and Its Limits. New York: Oxford University Press. Wright, Crispin (­1991). Scepticism and dreaming: imploding the demon. Mind 100 (­1):­87–​­116. Wright, Crispin and Davies, Martin (­2004). On epistemic entitlement. Aristotelian Society Supplementary Volume 78:­167–​­245.

6 Dogma, Defeat, and Transmission Christoph Kelp

1 Introduction The problem of scepticism is one of the most prominent problems in epistemology. In one of its guises, the one I will be working with here, it takes the form of a paradox. Here goes: SP1. I know (­by looking) that (­H ANDS =) I have hands. SP2. If I know that HANDS, then I know that I am not a (­BIV =) disembodied brain in a vat that is fed deceptive experiences as of a normal physical environment. SP3. I don’t know that I am not a BIV. Each of S­ P1–​­SP3 is highly plausible. Of course, I know that I have hands. If I know anything at all, that’s one of the things I know. I can see them at the end of my arm, move them at will, etc. That’s why SP1 is plausible. SP2 is motivated by the highly plausible principle (­Transmission =) that knowledge extends across competent deduction.1 Since I know that having hands entails not being a BIV, I can deduce that I am not a BIV from the fact that I have hands. So, assuming that I did go through the deduction, if I know that I have hands, I know that I am not a BIV. Finally, SP3 is also plausible. How could it be false? How could I know that I am not a BIV? After all, everything would seem exactly the same to me as it would were I to be a BIV. The dogmatism paradox is another important problem in epistemology. Here is one of its instances: DP1. I know (­by looking) that (­R ED WALL =) the wall before me is red. DP2. If I know that RED WALL, then I know that any evidence against RED WALL I may encounter is misleading. DP3. If I know that any evidence against RED WALL I may encounter is misleading, then I can justifiably disregard it. DP4. It is not the case that I can justifiably disregard any evidence against RED WALL I may encounter. 2

DOI: 10.4324/9781003104766-7

98  Christoph Kelp Again, all of ­DP1–​­DP4 are highly plausible. DP1 is another standard instance of perceptual knowledge. DP2 is motivated by Transmission. DP3 is motivated by the plausible principle (­Disregard =) that one can justifiably disregard evidence that one knows to be misleading.3 DP4 is intuitively highly plausible. For instance, when you tell me that the wall is white and illuminated by red light, I cannot justifiably disregard your testimony, at least not provided that the only evidence I have is my perception. Doing so would be highly irresponsible. Finally, consider the following two arguments: SA1. HANDS is true (­where my source for this is perception). SA2. If HANDS is true, then I am not a BIV. SA3. Hence, I am not a BIV. DA1. RED WALL is true (­where my source for this is perception). DA2. If RED WALL is true, then any evidence against RED WALL is misleading. DA3. Hence, any evidence against RED WALL is misleading. It is widely agreed that there is something suspect about these arguments. Very few people would be moved by these arguments in the direction of believing the conclusion. At the same time, given Transmission, anyone who knows the premises and competently deduces the conclusion from them, comes to know the conclusion. Given that knowledge makes belief permissible, the question arises why these arguments don’t move people in the direction of believing the conclusions. The question becomes even more pressing if we focus on those of us who have an interest in whether these conclusions hold and are currently considering whether they do; and, even more dramatically, if we grant that we have a duty to believe that p in cases such that (­we are considering the question whether p, have an interest in whether p, etc.) we will come to know that p by believing that p. Why would we miss out on the opportunity or even violate a duty here? This is also puzzling, especially once we note that, in the vast majority of cases, we are not equally hampered. Suppose, for instance, that we know that either Plum or Scarlett was the murderer. If we come to know that it wasn’t Scarlett, few of us would not be moved in the direction of believing the conclusion of the following argument: MA1. It was Plum or Scarlett. MA2. It wasn’t Scarlett. MA3. Hence, it was Plum. We are thus faced with two paradoxes and a puzzle (­henceforth also ‘­the triad’). For present purposes, I’d like to distinguish between two kinds of solution to them. One holds on to Transmission (‘­preservationism’), the

Dogma, Defeat, and Transmission  99 other abandons it (‘­abolitionism’). The central aim of this chapter is to show that the most popular versions of both kinds of solution run into trouble and to offer a better alternative. More specifically, I will show that the dogmatism paradox causes trouble for the most popular versions of both preservationism (­Section 2) and abolitionism (­Section 3). Section 4 takes on the constructive task of developing a better alternative.

2 Preservationism Note that what all members of the triad have in common is that they feature Transmission. One might think that this provides excellent reason for thinking that Transmission is false. And, indeed denying Transmission promises to offer a clean solution in all three cases, at least at first glance. On this view, the fact that I know that HANDS does not commit us to the claim that I also know that I am not a BIV. Likewise, the fact that I know that RED WALL does not commit us to the claim that I do not know that any evidence against RED WALL is misleading. And, finally, what’s suspect about the arguments ­SA1–​­SA3 and D ­ A1–​­DA3 is that, unlike M ­ A1–​­MA3, they don’t transmit knowledge. Provided that knowledge is required for permissible belief (­or provided that justification is thus required and that justification is also not transmitted), it is of no surprise that these arguments don’t move us in the direction of belief. In this way, one might think that champions of preservationism are on the back foot from the very beginning. At the same time, it is worth noting that preservationism is the most popular view in the literature. Since the vast majority of epistemologists are keen to avoid scepticism, they allow that we know many ordinary empirical propositions. Most importantly for present purposes, they allow that I have perceptual knowledge that HANDS. And, of course, preservationists will maintain that the argument from SA1 to SA3 is knowledge transmitting. This means that, according to preservationism, the culprit in the sceptical paradox is SP3.4 In addition, they will have to provide an error theory for SP3 and owe us an alternative explanation of why the argument from SA1 to SA3 looks suspect. But, for present purposes at least, I’d like to assume that they can deliver the goods on this front. What I would like to focus on instead is the dogmatism paradox. The reason for this is that it seems considerably less palatable to let the argument from DP1 to the negation of DP4 run its course. Allowing, contrary to intuition, that I know that I am not BIVs is not that bad. After all, it’s a victory against the sceptic. And that’s something worth having. Allowing, contrary to intuition, that I can justifiably disregard any evidence against RED WALL I may encounter is considerably worse. It’s not a victory against anyone. By the same token, denying DP4 is considerably less palatable.

100  Christoph Kelp Given that this is so, it may now look as though the only remaining option for champions of Transmission is to deny Disregard, which motivates DP3. But, again, there is a question as to why Disregard should fail here. This question is particularly difficult to answer once we take note of the fact that knowledge is factive. As a result, if I know that RED WALL and I know that RED WALL entails that any evidence against RED WALL I may encounter is misleading, I have a guarantee that any evidence against RED WALL I may encounter is misleading. But given that I have this kind of guarantee, why should it be that I can’t disregard the evidence in question justifiably? What’s more, there are many cases in which we justifiably disregard evidence we know to be misleading. This suggests that, even if we abandon Disregard, we will still need to countenance a restricted version of the principle. That, however, weakens the position of friends of Transmission ­vis-­​­­à-​­vis their foes, at least on the face of it. To see why, note that one of the main charges that friends of Transmission have levelled against their opponents is that they haven’t given a viable restricted version of Transmission (­e.g. Hawthorne 2004). This charge loses a considerable amount of its force once champions of Transmission have to concede that they are in the same boat in that they need to give a viable restricted version of Disregard. Giving a viable restricted version of a principle that comes out extensionally adequate is hard, and there is no particular reason for thinking that it should be easier for Disregard than for Transmission. What comes to light is that preservationists are not only on the back foot from the beginning, the dogmatism paradox also poses a difficult challenge for them in its own right. In this way, it may appear as though the prospects for preservationism are looking more and more dim. Fortunately, however, there is an elegant alternative solution to the dogmatism paradox that promises to enable preservationists to hold on to all of ­DP1–​­DP4. It appeals to the defeasibility of knowledge. Here is the famous passage from Gilbert Harman that first states it: The argument for paradox overlooks the way actually having evidence can make a difference. Since I now know that [RED WALL], I now know that any evidence that appears to indicate something else is misleading. That does not warrant me in simply disregarding any further evidence, since getting that further evidence can change what I know. In particular, after I get such further evidence I may no longer know that it is misleading. For having the new evidence can make it true that I no longer know that [RED WALL]; if I no longer know that, I no longer know that the new evidence is misleading. (­Harman 1973, 149) In other words, when I encounter evidence against RED WALL, my knowledge that RED WALL is defeated. Once I acquire such evidence, I

Dogma, Defeat, and Transmission  101 no longer know that RED WALL. But given that I no longer know this, I no longer know that any evidence against RED WALL is misleading, not even if Transmission holds. But if I don’t know that any evidence against RED WALL is misleading, then I cannot justifiably disregard it, not even if Disregard holds. Once we are clear that knowledge is defeasible, we can escape the paradox.5 It may be worth noting just how neat the defeat solution is for preservationists. First, it allows them to meet the challenge the dogmatism paradox poses for their view. Second, it also serves to level the playing field with abolitionists once more. The fact that Transmission features in all members of the triad is no longer reason to favour abolitionism. After all, what the defeat solution promises to establish is that the dogmatism paradox and the puzzle about the argument from DA1 to DA3 can be solved in an independently plausible and a ­cost-​­free way, simply by taking note of the defeasibility of knowledge. It is easy enough to see that, by the same token, it is of considerable importance for preservationists that the defeat solution comes out as viable. If it doesn’t, both the challenge and the reason for thinking that preservationism is at a dialectical disadvantage ­vis-­​­­à-​­vis abolitionism resurface. In the remainder of this section, I will develop an argument that the defeat solution doesn’t work at the end of the day.6 To begin with, let’s take a closer look at a couple of concrete cases involving defeat. DEFEAT. At t0, I am standing in front of a red wall. At t1, I have perceptual knowledge that RED WALL. At t2, A tells me that (­R ED LIGHT =) the wall before me is white illuminated by red light. While at t1, I know that RED WALL, at t2 I acquire new ­evidence – ​­A’s testimony that RED LIGHT – ​­which constitutes evidence against RED WALL. The new evidence defeats my knowledge that RED WALL with the result that, at t2, I don’t know that RED WALL. DEFEAT DEFEAT. At t0, I am standing in front of a red wall. At t1, I have perceptual knowledge that RED WALL. At t2, A tells me that (­R ED LIGHT =) the wall before me is white illuminated by red light. At t3, B tells me that (­FALSE TESTIMONY =) what A just told me is false. DEFEAT DEFEAT is just like DEFEAT up until t2. That is to say, at t1, I know that RED WALL and at t2 my knowledge is defeated by new evidence I acquire. The difference between the two cases is that there is a further development here. At t3, I acquire yet further new e­ vidence –​ ­B’s testimony that FALSE T ­ ESTIMONY – which ​­ constitutes evidence that A’s testimony is false. The further evidence defeats the defeater

102  Christoph Kelp provided by A’s testimony.7 As a result, at t3, I once again know that RED WALL. With these two cases of defeat on the table, let’s ask whether I can justifiably disregard the evidence provided by A’s testimony. Intuitively, in DEFEAT, the answer is ‘­no’. In particular, I do not justifiably disregard A’s testimony when doing so based on the conclusion of the following line of reasoning: REASONING 1. RED WALL is true. If so, the evidence against RED WALL provided by A’s testimony that RED LIGHT is misleading. So, the evidence against RED WALL provided by A’s testimony that RED LIGHT is misleading. Of course, DEFEAT combined with REASONING 1 is a paradigm instance of the dogmatism paradox. Unsurprisingly, champions of the defeat solution can handle it without difficulty. When I receive A’s testimony at t2, I acquire a defeater for my knowledge that RED WALL. Since I no longer know that RED WALL, I don’t come to know that A’s testimony is misleading and so won’t justifiably disregard it when doing so based on REASONING 1, not even if Transmission and Disregard are both true. What about DEFEAT DEFEAT? Here it is intuitive that I can justifiably disregard A’s testimony. In particular, I justifiably disregard A’s testimony when I do so based on the conclusion of the following line of reasoning: REASONING 2. FALSE TESTIMONY is true. If so, the evidence against RED WALL provided by A’s testimony that RED LIGHT is misleading. So, the evidence against RED WALL provided by A’s testimony that RED LIGHT is misleading. Thus far, we have two cases. One combines DEFEAT with REASONING 1 (­­D-​­R1) and the other combines DEFEAT DEFEAT with REASONING 2 (­­DD-​­R2). While, in ­D -​­R1, intuitively, I do not justifiably disregard the evidence provided by A’s testimony, my knowledge that RED WALL is defeated, and the case can be handled in the way envisaged by champions of the defeat solution without further trouble. In contrast, my knowledge that RED WALL isn’t defeated in ­DD-​­R2. At the same time, intuitively, I do justifiably disregard the evidence provided by A’s testimony. In this way, neither case means trouble for the defeat solution. To see why the defeat solution does face a problem in the end, consider a third case (­­DD-​­R3) which combines DEFEAT DEFEAT with the following reasoning: REASONING 3. I don’t care whether FALSE TESTIMONY is true. All I care about is, as I can see with my own eyes, that RED WALL

Dogma, Defeat, and Transmission  103 is true. If RED WALL is true, the evidence against RED WALL provided by A’s testimony that RED LIGHT is misleading. So, the evidence against RED WALL provided by A’s testimony that RED LIGHT is misleading. Note that, in ­DD-​­R3, I know that RED WALL. After all, since the case includes DEFEAT DEFEAT, I not only have a defeater for my knowledge that RED WALL but also a defeater defeater. At the same time, unlike in the other case that includes DEFEAT DEFEAT, i.e. ­DD-​­R2, I don’t dismiss the A’s testimony by reasoning from the proposition that FALSE TESTIMONY. Instead, my reasoning sidesteps the proposition that FALSE TESTIMONY entirely and proceeds from the proposition that RED WALL alone. The trouble is that when I disregard A’s testimony in this way, intuitively, I do not do so justifiably, no more so than in D ­ -​­R1. But since, in the present case, I know that RED WALL, and I disregard A’s testimony based on the conclusion of REASONING 3, then it follows from Transmission and Disregard that I do so justifiably. In this way, we have an instance of the paradox.8 Of course, this is bad news for the defeat solution. After all, unlike in the case of D ­ -​­R1, ­DD-​­R3 cannot be dealt with by appealing to the defeasibility of knowledge. After all, thanks to the fact that I have a defeater defeater here, my knowledge is just not defeated. But, of course, if the defeat solution doesn’t work for all instances of the paradox, there is excellent reason for thinking that it remains ultimately unsatisfactory after all.9

3 Abolitionism Given that the dogmatism paradox remains a thorn in the side of preservationism, let’s look into the prospects of alternatives that deny Transmission. There are two prominent versions of abolitionism. The first appeals to the nature of knowledge, the second to the structure of warrant. The first version of abolitionism is most closely associated with the work of Fred Dretske (­1970) and Robert Nozick (­1981). They defend a ­sensitivity-​­based account of knowledge according to which, roughly, to know that p is to sensitively believe that p. And, roughly, one’s belief that p is sensitive if and only if were p to be false, one wouldn’t believe that p. On this view, again roughly, Transmission fails when one knows the premises of one’s argument but the belief in the conclusion isn’t sensitive. Most importantly for present purposes, on the sensitivity approach, both problematic ­arguments – ​­i.e. the argument from SA1 to SA3 and the argument from DA1 to ­DA3 – ​­are instances of Transmission failure. I know that HANDS (­SA1) and that if so, I am not a BIV (­SA2). At the same time, my belief that I am not a BIV (­SA3) isn’t sensitive. If I were

104  Christoph Kelp a BIV, I would still believe that I wasn’t. Similarly, I know that RED WALL (­DA1) and that if so, any evidence against RED WALL I may encounter is misleading (­DA2). However, my belief that any evidence against RED WALL is misleading (­DA3) is not sensitive: If some such evidence were not misleading, I would still believe that it was. While versions of abolitionism that explain the failure of Transmission as grounded in the nature of knowledge can deliver the desired results for present purposes, it is fair to say that they have decreased in popularity in recent times. One important reason for this is that it has already proved very hard to pin down a viable account of the nature of knowledge. To pin down an account that is not only viable but also underwrites failures of Transmission in all and only the cases in which it should would be a truly Herculean feat. It is unsurprising that it hasn’t been accomplished to date. Since, in addition, I have discussed the s­ensitivity-​­based version of this approach in detail elsewhere (e.g. Kelp 2019, Kelp 2021a), I will set this version of abolitionism aside here. Instead, I would like to focus on the second kind of approach, which continues to enjoy a considerable degree of popularity, and which is most prominently championed by Crispin Wright (­e.g. 2004, 2007). The key idea here is to adopt an account of the structure of warrant according to which certain kinds of warrant are information dependent. Roughly, to say that a kind of warrant is information dependent is to say that one can acquire a warrant of this kind only if one already has a certain other kind of warrant. For instance, according to Wright, perceptual warrant is information dependent in this way. What this means is that perception is not a source of basic warrant for propositions about the external world. Rather, perception will be a source of warrant for propositions about perceivable objects only if one already has certain background warrants. On this view, then, perceptual warrants depend on background warrants. It is the information dependence of warrant that serves to explain why warrant doesn’t transmit across competent deduction, at least not in all cases. To see how, note first that what happens when warrant successfully transmits across deduction is that one acquires a warrant for the conclusion that depends on one’s warrant for the premises. Suppose I have a warrant that it was either Plum or Scarlett who did the deed and that it was not Scarlett. When I competently deduce that it was Plum, the warrant that I acquire for the conclusion of my deduction depends on the warrant I have for its premises. Now, suppose warrant is information dependent in the way envisaged. Suppose also that I have a perceptual warrant that p that depends on a background warrant for q. If I know that p entails q, as we may suppose I do, I can now competently deduce q from p. So, suppose I do that. If warrant successfully transmits from p to q, I now acquire a warrant for q that depends on my warrant that p. But since, ex hypothesi, my warrant

Dogma, Defeat, and Transmission  105 that p depends on my background warrant for q, it follows that I acquire a warrant for q that depends on itself. Since warrants cannot have this kind of circular structure, there is reason to believe that warrant doesn’t successfully transmit across my competent deduction in this case. With these points in play, let’s ask how champions of the present view explain what’s wrong with the argument from SA1 to SA3, i.e. SA1. HANDS is true (­where my source for this is perception). SA2. If HANDS is true, then I am not a BIV. SA3. Hence, I am not a BIV. Key to the answer is that they hold that among the background warrants I need to have for perception to give me warrant for propositions about perceivable objects is a warrant for the proposition that I am not a BIV. It is easy to see that it follows that warrant fails to transmit from the proposition that HANDS to the proposition that I am not a BIV. After all, my warrant that HANDS depends on the background warrant that I am not a BIV. If warrant successfully transmitted from the proposition that HANDS to the proposition that I am not a BIV, I would now acquire a warrant that I am not a BIV that depends on my warrant for the proposition that HANDS. But since my warrant for the proposition that HANDS depends on my background warrant for the proposition that I am not a BIV and since warrants cannot have this kind of circular structure, we get the result that warrant doesn’t successfully transmit across my competent deduction from the proposition that HANDS to the proposition that I am not a BIV.10 We now have a clearer view of how champions of the present approach aim to explain what’s wrong with the argument from SA1 to SA3. The question remains how they venture to handle arguments such as the one from DA1 to DA3, i.e. DA1. RED WALL is true (­where my source for this is perception). DA2. If RED WALL is true, then any evidence against RED WALL is misleading. DA3. Hence, any evidence against RED WALL is misleading. Note that this argument is suspect in just the same way as the argument from SA1 to SA3. Accordingly, we may expect that whatever explains what’s wrong with the argument from SA1 to SA3 will also explain what’s wrong with the argument from DA1 to DA3 and vice versa.11 On the present view, what’s wrong with the argument from SA1 to SA3 is that my reasoning fails to transmit warrant from the proposition that HANDS to the proposition that I am not a BIV because warrant is information dependent in such a way that I need to already have a warrant that I am not a BIV to acquire a warrant that HANDS.

106  Christoph Kelp If ­DA1–​­DA3 affords the same explanation, this means that my reasoning fails to transmit warrant from the proposition that RED WALL to the proposition that any evidence against RED WALL is misleading because my perceptual warrant that RED WALL is information dependent. More specifically, it is information dependent in such a way that I need to already have a background warrant that any evidence against RED WALL is misleading.
The trouble for the present view is that this is not very plausible at all. In particular, the claim that my perceptual warrant that RED WALL depends on a background warrant that any evidence against RED WALL is misleading enjoys little plausibility. It is simply incredible that looking at a red wall can give me perceptual warrant for thinking that RED WALL only if I already have a background warrant e.g. that someone’s testimony that RED LIGHT is misleading. If this isn’t immediately obvious, consider also: I am a detective investigating a certain crime. Let’s suppose that during my investigation I may come across evidence that incriminates the culprit as well as evidence that exculpates them. On the present proposal, if I am to acquire warrant that a certain suspect did the deed due to the incriminating evidence, I must have a background warrant that any exculpating evidence is misleading. At the same time, if I am to acquire warrant that the suspect did not commit the crime, I must have a background warrant that any incriminating evidence is misleading. But given that this is so, at the start of my investigation where it is entirely open what evidence I will come across, I will have to have a background warrant that any incriminating evidence is misleading and a background warrant that any exculpating evidence is misleading also. In that case, however, whatever evidence on whether the suspect did the deed I may come across, I will have a background warrant that this evidence is misleading. As a result, it is hard to see how I could acquire warrant on whether the suspect did the deed either way, no matter how my investigation develops. Couldn’t one say that the two warrants simply defeat each other with the result that I don’t have background warrant that whatever evidence on whether the suspect committed the crime I may come across is misleading? No. This is because the problem resurfaces, albeit for different reasons. If the two background warrants defeat each other, it is once again hard to see how I could acquire warrant on whether the suspect did it either way. Of course, now the problem is not that I have a background warrant that whatever evidence on whether the suspect did it I may come across is misleading. Rather, the problem is that, no matter what evidence I come across, I won’t acquire a warrant for what the evidence suggests, because I don’t have the requisite background warrant (­as it is defeated). I take these considerations to make a strong case that even if Wright’s information dependence thesis for warrant is true, it does not extend to background warrants that any evidence against the relevant target

Dogma, Defeat, and Transmission  107 propositions is misleading. By the same token, even if Wright’s appeal to the information dependence of warrant offers a way to deal with the problem of scepticism, it cannot be extended to provide a viable solution to the entire triad. In particular, it doesn’t serve to explain what’s wrong with arguments like the one from DA1 to DA3, and it doesn’t provide a satisfactory solution to the dogmatism paradox.

4 A New Abolitionism What comes to light is that a viable solution to both of our paradoxes and the puzzle remains to be found. The most popular approaches on both sides of the preservationism/­abolitionism divide remain ultimately unsatisfactory. In what follows, I will develop a version of abolitionism that fares better. More specifically, first, I will briefly rehearse an approach to Transmission failure and how it can deal with the sceptical paradox that I have developed elsewhere in more detail (­Kelp 2019, 2021a, b). I will then show how this approach can be extended to provide a promising solution to the dogmatism paradox also. Here is the approach to Transmission failure, in a nutshell. One key idea is that inquiry into whether p aims at settling the question whether p. Moreover, it is plausible that ­question-​­begging excludes ­question-​ s­ettling: you cannot settle the question whether p by means of an argument for p that is ­question-​­begging. Finally, it is widely agreed that knowledge that p/­­not-​­p is sufficient for attaining the aim of inquiry into whether p. Given that this is so, we get the result that believing that p based on an argument for p that is q ­ uestion-​­begging does not give one knowledge that p. If, in addition, one knows all the premises of the argument and believes that p based on competent deduction from these premises, we get the result that Transmission fails. We now have the conditions for a kind of case in which Transmission fails. Now, crucially, these conditions are satisfied in standard cases in which we go through the argument from SA1 to SA3. In standard cases, I know that HANDS, I know that HANDS entails that I am not a BIV, I competently deduce, and, on that basis, I come to believe that I am not a BIV. At the same time, it is widely agreed that the argument from SA1 to SA3 is q ­ uestion-​­begging.12 Given that this is so, standard cases in which we go through the argument from SA1 to SA3 are cases of Transmission failure. In this way, the approach offers an account of what’s wrong with the argument from SA1 to SA3 in terms of the failure of Transmission. We also get (­part of13) a solution to the sceptical paradox in terms of a denial of Transmission, which is what motivates SP2.14 With these points in play, let’s look at whether this approach to Transmission failure and scepticism can be extended to the dogmatism paradox and if so how. Fortunately, there is reason for optimism. To see this, recall that the argument ­DA1–​­DA3 is suspect in the same way as the

108  Christoph Kelp argument from SA1 to SA3. But if what makes the latter suspect is that it is ­question-​­begging, then the same goes for the former. Crucially, just as in the sceptical case, here too, the conditions for failure of Transmission are satisfied in standard cases. In standard cases, I know that RED WALL, I know that RED WALL entails that RED LIGHT is false, and from this, I competently deduce and, on that basis, come to believe that RED LIGHT is false. Since, just like the argument from SA1 to SA3, the argument from DA1 to DA3 is q ­ uestion-​­begging, standard cases in which we go through the argument from DA1 to DA3 are cases of Transmission failure. In this way, the approach offers an account of what’s wrong with the argument from DA1 to DA3 in terms of the failure of Transmission. We also get a solution to the dogmatism paradox in terms of a denial of Transmission, which is what motivates DP2. Before closing, I’d briefly like to look at how the present approach improves on the competition in both the preservationist and the abolitionist camps. First, since the view is a version of abolitionism, it does not require the defeat solution to offer a viable solution to the dogmatism paradox. After all, on this view, the problem lies with Transmission. Accordingly, it will come as no surprise that the account can handle the case that causes trouble for the defeat solution. Recall that here I not only have a defeater for my knowledge that RED WALL, but also a defeater defeater, with the result that I know that RED WALL. The problem for the defeat solution is that if I go on to dismiss A’s testimony that RED LIGHT as misleading by going through the REASONING 3, I will not do so justifiably. While this means trouble for the defeat solution, it is easy to see that the present approach fares better. After all, even though I know that RED WALL (­thanks to a defeater defeater), the argument I use to arrive to the conclusion that A’s testimony that RED ­ uestion-​­begging. On the present view, it LIGHT is misleading is still q will be an instance of Transmission failure. Since knowledge doesn’t transmit to the conclusion, the present approach can explain that when I dismiss A’s testimony that RED LIGHT as misleading, I do so no more justifiably than in the case in which I have a defeater but no defeater defeater. Second, the view can also improve on the background warrant approach to the failure of transmission of warrant. To see this, note that the present view simply doesn’t endorse a background warrant account of the structure of warrant. On the contrary, it can allow that we acquire warrants, including perceptual ones, without any background warrants being already in place. As a result, the trouble the dogmatism paradox made for the background warrant approach can be avoided. In order to acquire a perceptual warrant that RED WALL, I don’t already need to have a background warrant that any evidence against RED WALL I may acquire is misleading. Likewise, it is not the case that before starting an investigation into whether a certain suspect committed a certain crime,

Dogma, Defeat, and Transmission  109 I need to have a background warrant that any exculpating evidence is misleading and a warrant that any incriminating evidence is misleading.

5 Conclusion This chapter has investigated a triad of paradoxes and puzzles involving Transmission. I have argued that the dogmatism paradox in particular means trouble for the most prominent ways of preserving and abolishing Transmission. The key reason why the most popular version of preservationism runs into trouble is that the defeat solution remains ultimately unsatisfactory and the specific way in which it fails. On the abolitionist side, I focused here on approaches to failure of transmission of warrant that appeal to a background warrant account of the structure of warrant. I argued that even if these approaches offer a workable solution to the problem of scepticism, they remain unsatisfactory when it comes to the dogmatism paradox. This is because the key claim about the dependence of warrants on a background warrant that any evidence against what we have warrant for is misleading is problematic. Finally, I have developed an alternative view that combines a version of abolitionism without the problematic claim about the structure of warrant. I have argued that this view avoids the problems on both sides and can offer a unified account of the two paradoxes and the puzzle in terms of Transmission failure.

Notes 1 Famous defences of Transmission include Hawthorne (­2004) and Williamson (­2000). Transmission has been contested by Dretske (­1970) and Nozick (­1981). The related transmission of warrant principle is defended e.g. in Pryor (­2000, 2004) and rejected e.g. in Wright (­2004, 2007) and Daveis (­2003). 2 The dogmatism paradox was first presented by Kripke in an unpublished lecture to the Cambridge Moral Sciences Club. 3 It is worth noting, however, that Disregard has been challenged e.g. in Veber (­2004) and Ye (­2015). 4 Famous defences of this approach to scepticism include Moore (­ 1939), Pritchard (­2005, 2015), Sosa (­1999), and Williamson (­2000). 5 Prominent champions of (­versions of) the defeat solution include Conee (­2004), Ginet (­1980), Harman (­1973), Hawthorne (­2004), Sorensen (­1988), and Kelly (­2008). 6 See L ­ asonen-​­Aarnio (­2014) for a different attack on the defeat solution to the dogmatism puzzle and Brown (­2018) for a recent response. 7 If this isn’t obvious, note that we can stipulate that A knows C to be an omniscient and omnibenevolent god. 8 It may also be worth considering a version of the case in which the defeaters provided by both A and B’s testimony are external (­sometimes also ‘­propositional’ (­e.g. Bergmann 2006) or ‘­normative’ (­e.g. Lackey 2008), see e.g. Goldberg (­2018), Kelp (­2022), and Simion (­Forthcoming) for defences of external defeat). For instance, suppose I am a sexist: I don’t even listen to what women tell me, and A and B are both women. In this case, neither the

110  Christoph Kelp defeater nor the defeater defeater are even psychologically registered by me. Even so, the defeater defeats my knowledge that RED WALL and the defeater defeater reinstates it. In this case, I end up with knowledge that RED WALL. Of course, since I didn’t psychologically register the defeater provided by A’s testimony, I wouldn’t reason along the lines of REASONING 1 (­or 3). However, I might still reason along the following lines: REASONING 4. RED WALL is true. If so, then if what A just told me is evidence against RED WALL, then it’s misleading evidence. So, if what A just told me is evidence against RED WALL, then it’s misleading evidence. Given that I arrive at a conditional conclusion, can I still disregard A’s testimony? Perhaps. In any case, I can certainly acquire a disposition to disregard it upon learning that A’s testimony is evidence against RED WALL. Crucially, it is hard to deny that REASONING 4 does not justify the acquisition of a disposition to disregard A’s testimony here upon learning that it is evidence against RED WALL. At the same time, if Disregard holds, and if one can justifiably disregard evidence one knows to be misleading, then surely the following will also be the case: (­Disposition to Disregard =) for evidence, e, that such that one knows that if some condition C holds, e is misleading, one is justified in acquiring a disposition to disregard e upon learning that C holds. Again, we can generate an instance of at least a close cousin of the dogmatism paradox (­with Disposition to Disregard in place of Disregard). ­ D-​­R3, A 9 This argument against defeatists relies on the intuition that, in D doesn’t justifiably disregard A’s testimony. Couldn’t defeatists simply reject this intuition? There is reason to think that the answer is ‘­no’. If the intuition in D ­ D-​­R3 is rejected, it is hard to see how the defeatist approach could still continue to do any work in the dogmatism paradox. After all, if defeatists want to reject the intuition in ­DD-​­R3, they will need a viable error theory that explains exactly how our intuition was led astray here. The trouble is that it is hard to see what this error theory might look like such that it applies to ­DD-​­R3 but not to standard instances of the dogmatism paradox, such as D-R1. For if it does apply to standard instances as well, the defeat solution to the paradox becomes redundant. The error theory will already provide us with a solution to the dogmatism paradox, and one that works uniformly for all instances of the paradox. 10 Note that champions of the present view argue that warrant fails to transmit across competent deduction in certain cases. At the same time, I set up the discussion as concerning the question of whether knowledge transmits across competent deduction in the way envisaged by Transmission. This raises the question as to how the issue of the transmission of warrant relates to the issue of whether Transmission holds. There is reason to believe that if warrant fails to transmit across competent deduction, then Transmission fails also. To see this, note that, according to Transmission, if one comes to believe the conclusion of one’s argument based on competent deduction from the premises, and if one knows the premises, then one knows the conclusion. Now, if warrant fails to transmit across competent deduction, it is hard to see how this could be. After all, it is widely agreed that knowledge that p requires belief that p is based on a warrant for p. To come to know the conclusion, then, one must come to believe it based on a warrant for it. However, it is hard to see how one could come to believe the conclusion based on a warrant when one believes it based on a ­non-​­transmissive inference. Finally, note that the fact that champions of transmission of warrant failure may allow for warrant to be closed under competent deduction is ­by-­​­­the-​­by. True, if warrant is thus closed, it follows that one must have warrant for the

Dogma, Defeat, and Transmission  111 conclusion when going through the inference. However, to acquire knowledge of the conclusion, one’s belief in the conclusion must be based on this warrant. And, again, it is hard to see how this could be given that the belief is based on inference and that the inference is n ­ on-​­transmissive (­though see Greenough 2020 for discussion). 11 Or, to be more precise, we may expect this once we are clear that the defeat solution to the dogmatism paradox doesn’t work. 12 Even champions of Transmission and the related transmission of warrant principle acknowledge this point (­ Pryor 2000, Markie 2005, Pritchard 2007). 13 The reason it’s only part of the solution is that, in my view, SP3 (­I don’t know that I am not a BIV) is false also. I develop this account in more detail in Kelp (­2021a, b). 14 In the more detailed discussion of Transmission failure, I also show how a restricted version of Transmission can be preserved. See Kelp (­2019, 2021a) for more details.

References Bergmann, M. 2006. Justification without Awareness. Oxford: Oxford ­University Press. Brown, J. 2018. Fallibilism: Evidence and Knowledge. Oxford: Oxford University Press. Conee, E. 2004. Heeding misleading evidence. In Evidentialism: Essays in Epistemology, E. Conee and R. Feldman. New York: Oxford University Press, ­259–​­76. Davies, M. (­2003). The problem of armchair knowledge. In New Essays on Semantic Externalism and ­Self-​­Knowledge, edited by S. Nuccetelli. Cambridge, MA: MIT Press, ­23–​­55. Dretske, F. 1970. Epistemic operators. Journal of Philosophy 67: ­1007–​­23. Ginet, C. 1980. Knowing less by knowing more. Midwest Studies in Philosophy 5: ­151–​­62. Goldberg, S. 2018. To the Best of Our Knowledge: Social Expectations and Epistemic Normativity. Oxford: Oxford University Press. Greenough, P. 2020. Knowledge for nothing. In Epistemic Entitlement, edited by P.J. Graham and N.J.L.L. Pedersen. Oxford: Oxford University Press, 361–88​­. Harman, G. 1973. Thought. Princeton, NJ: Princeton University Press. Hawthorne, J. 2004. Knowledge and Lotteries. Oxford: Oxford University Press. Kelp, C. 2019. Inquiry and the transmission of knowledge. Philosophy and Phe­ 98–​­310. nomenological Research 99: 2 Kelp, C. 2021a. Inquiry, Knowledge, and Understanding. Oxford: Oxford U ­ niversity Press. Kelp, C. 2021b. Theory of inquiry. Philosophy and Phenomenological Research 103:­359–​­84. Kelp, C. 2022 (­in press). Proficiencies and Defeat. Philosophical Issues 32. Lackey, J. 2008. Learning from Words. Testimony as a Source of Knowledge. Oxford: Oxford University Press. Lasonen-​­ ­ Aarnio, M. 2014. The dogmatism puzzle. Australasian Journal of ­Philosophy 92: ­417–​­32.

112  Christoph Kelp Markie, P. 2005. Easy knowledge. Philosophy and Phenomenological Research 70: ­406–​­16. Moore, G. E. 1939. Proof of an external world. Proceedings of the British Academy 25: ­273–​­300. Nozick. R. 1981. Philosophical Explanations. Cambridge: Harvard University Press. Pritchard, D. 2005. Epistemic Luck. Oxford: Oxford University Press. Pritchard, D. 2007. How to be a ­neo-​­Moorean. In Internalism and Externalism in Semantics and Epistemology, edited by S. Goldberg. Oxford: Oxford University Press, 6 ­ 8–​­99. Pritchard, D. 2015. Epistemic Angst: Radical Skepticism and the Groundlessness of Our Believing. Princeton, NJ: Princeton University Press. Pryor, J. 2000. The skeptic and the dogmatist. Nous 34: 5 ­ 17–​­49. Pryor, J. 2004. What’s wrong with Moore’s argument? Philosophical Issues 14: ­349–​­378. Simion, M. Forthcoming. Resistance to evidence and the duty to believe. Philosophy and Phenomenological Research. Sorensen, R. 1988. Dogmatism, junk knowledge, and conditionals. The Philosophical Quarterly 38: ­433–​­54. Sosa, E. 1999. How to defeat opposition to Moore. In Philosophical Perspectives, edited by J. Tomberlin. Oxford: Blackwell, ­141–​­53. Veber, M. 2004. What do you do with misleading evidence? The Philosophical Quarterly 54: ­557–​­69. Ye, R. 2016. Misleading evidence and the dogmatism puzzle. Australasian Journal of Philosophy 94: ­563–​­75. Williamson, T. 2000. Knowledge and Its Limits. Oxford: Oxford University Press. Wright, C. 2004. Warrant for nothing (­and foundations for free)? Proceedings of the Aristotelian Society, Suppl. Vol. 78: ­167–​­212. Wright, C. 2007. The perils of dogmatism. In Themes from G.E. Moore. New Essays in Epistemology and Ethics, edited by S. Nuccetelli. Oxford: Oxford University Press, ­25–​­48.

7 In Defence of Closure Duncan Pritchard

The closure principle has been formulated in a range of different ways since it began to be discussed in the 1970s. The initial formulation, due to Fred Dretske (­1970), was essentially concerned with the idea that knowledge is closed under entailments, such that one can know what is entailed by what one knows, regardless of whether one’s knows or even believes the entailment in question.1 Dretske famously argued that such a principle did not hold universally. In particular, he held that closure fails because ‘­knows that φ’ is not a ‘­fully penetrating’ operator, like ‘­it is necessary that φ’, which ‘­penetrates’ through to all entailments. Interestingly, however, Dretske argued that it was also not a ‘­­non-​­penetrating’ operator either, like ‘­it is surprising that φ’, which offers no guarantee of penetration to entailments. Instead, he claimed that ‘­knows that φ’ is a ‘­­semi-​­penetrating operator’, such that it ‘­penetrates’ through some entailments but not others in a principled fashion. Dretske’s concern was to specify what this principled basis was that prevented certain kinds of operator penetrations to occur. The proposal that Dretske offers is now very familiar and can be found in some form in the work of several figures who reject closure in this sense. In broad terms, the idea is that some of our knowledge occurs, entirely legitimately (­i.e., not in a way that undermines its epistemic standing), against the background of certain presuppositions holding. It follows that one cannot use this knowledge in order to come to know these presuppositions, on pain of a kind of bootstrapping. 2 ­Closure-​­style inferences are thus usually fi ­ ne—​­and hence the ‘­knows that φ’ operator penetrates through to ­entailments—​­so long as it doesn’t concern inferences to these background presuppositions. Before we can meaningfully engage with the question of whether the closure principle should be restricted, however, we first need to identify which formulation is at issue. Dretske (­2005) himself in more recent work focusses on a reading of the closure principle whereby the entailment is known.3 This certainly makes closure much more compelling, as I take it there isn’t much of a temptation to regard unknown consequences of our knowledge as being in the market for knowledge purely in virtue of the fact that they are entailed by one’s knowledge. Aside from anything else, DOI: 10.4324/9781003104766-8

114  Duncan Pritchard there is nothing to ensure that one even believes such consequences, and yet belief is usually thought to be necessary for knowledge.4 While the idea that knowledge is closed under known entailment is more plausible, however, it is still prone to obvious counterexamples. In particular, the subject’s belief in the entailed proposition might have nothing whatsoever to do with her knowledge of the antecedent proposition and the relevant entailment. In such a case, why should her knowledge of the entailing proposition and the entailment have any bearing on whether she knows the entailed proposition? This issue highlights an important point about closure principles, which is that they are trying to capture a specific kind of b ­ elief-​­forming process. In particular, what we are interested in is whether knowledge can be extended via one’s knowledge of entailments to knowledge of the entailed propositions, and that means that we are thinking of the subject’s belief in the entailed proposition as being based on this inference. Since we are now describing an inferential process, this means that we need a diachronic formulation of the closure principle, rather than the kind of synchronic formulation that Dretske was considering. We thus get the following competent deduction articulation of closure that is now standard in the literature: Competent Deduction Closure If S knows that p, and S competently deduces from p that q, thereby forming a belief that q on this basis while retaining her knowledge that p, then S knows that q.5 Henceforth, by ‘­closure’, we will have in mind this specific formulation. A few remarks on this formulation are in order. First, the notion of competent deduction in play ensures that the subject isn’t simply making a mistake in undertaking the relevant inference, and hence, it is meant to presuppose that the target entailment both holds and is known to hold by the subject. Second, it is here stipulated that the belief in the entailed proposition is based upon the competent deduction from the subject’s knowledge of the entailing proposition, which thereby excludes the issue we just raised where the basis for belief has nothing to do with the subject’s knowledge of the entailing proposition and the entailment. Third, since inferences take place over time, we also need to stipulate that the subject retains her knowledge of the entailing proposition throughout, in order to exclude cases where the subject loses this knowledge during the inference (­perhaps because of the inference, in that the inference itself prompts her to doubt that she knows the entailing proposition). Finally, fourth, I want to highlight that the notion of knowledge that is in play here is rationally grounded knowledge. Depending on one’s wider epistemology, rationally grounded knowledge might well be the only kind of knowledge that there is.6 But we do not need to take sides in that issue.

In Defence of Closure  115 The crux of the matter is that any principle that is concerned with such reflective rational processes as competent deductions is clearly directed at the kind of knowledge that figures in the space of reasons. So construed, closure certainly seems highly intuitive. Given that competent deduction is a paradigmatically rational process, it is hard to see how employing such a process to extract a belief from what one already knows wouldn’t thereby ensure that this belief is also known. We thus have a formulation of closure which is such that it would be highly surprising to discover that it didn’t universally hold. With this in mind, I want to consider two kinds of counterexample that are typically posed against closure. Both of these counterexamples concern ­error-​ p ­ ossibilities the falsity of which is entailed by putatively known propositions, but whereas the first kind of e­ rror-​­possibility is local, the second kind is radical (­and so of relevance to the problem of radical scepticism). As we will see, it is important in the debate regarding the status of closure to keep local and global e­ rror-​­possibilities apart. Consider Dretske’s (­1970) famous counterexample to closure involving zebras and cleverly disguised mules. Our agent, let’s call her ‘­Zula’, is looking into a clearly marked zebra enclosure at a regular zoo and can see the z­ ebra-​­like creature inside. Does Zula know [Z]: that the creature she is looking at in the enclosure is a zebra? So long as circumstances are as described, and there is no epistemic ‘­funny business’ going on, then I think we would treat Zula as knowing this proposition. Note, however, that [Z] entails [­not-​­CDM]: that the creature Zula is looking at in the enclosure is not a cleverly disguised mule, and presumably, Zula will also know that this entailment holds. Accordingly, it seems that Zula can come to know that [­not-​­CDM] via closure by competently deducing it from her knowledge that [Z]. The difficulty, however, is that it doesn’t seem that Zula has an adequate epistemic basis for knowing [­not-​­CDM]. In particular, we can stipulate that Zula doesn’t have any special expertise that might be relevant in this regard, such as zoological training. Moreover, she hasn’t made any special checks of the creature either, such as closely inspecting it at close range. So, how then could she possibly be in a position to know [­not-​­CDM]? In particular, it seems that Zula isn’t able to discriminate between a zebra and a cleverly disguised mule. And yet it appears to follow from closure that her knowledge of [Z] enables her to come to know, via a competent deduction, [­not-​­CDM]. But how can it be that she can know that it is a zebra and not a cleverly disguised mule if she can’t discriminate between the two? Such cases naturally invite the following relevant alternatives line. In order to have knowledge, it is not required that one is able to rule out all possible e­ rror-​­possibilities, as that would be to set an unduly austere, infallibilist, requirement on knowledge.7 All that is required is that one is able to rule out the relevant ones, in some sense of ‘­relevance’ to be

116  Duncan Pritchard determined. The irrelevant alternatives, in contrast, can be reasonably presupposed as false and thus ignored. Intuitively, the CDM alternative is an irrelevant alternative when it comes to Zula’s knowledge (­in these conditions) of [Z], which is why she can know the latter even while not being in a position to exclude the former. This is thus why an unrestricted form of closure fails, in that if it were allowed, then one would be able to use closure to come to know an irrelevant alternative the falsity of which is presupposed in, and entailed by, one’s knowledge.8 There is also a variation on the relevant alternatives line that deserves note, which involves saving closure by going contextualist. In particular, given that we are working with a diachronic formulation of closure, it is possible to hold that the very consideration of an irrelevant ­error-​ ­possibility can thereby change the context and thus make that alternative relevant. On this view, Zula can know [Z] even though she doesn’t know [­not-​­CDM] just so long as she doesn’t entertain the latter. But once she starts to consider the entailment at issue, and hence the CDM alternative that it concerns, then this possibility now becomes relevant and her inability to exclude it ensures that she no longer knows the entailing proposition, [Z]. There is thus no single context in which Zula knows both [Z] and [­not-​­CDM], and hence, there is no need (­on this score at least) to deny closure.9 I think we can respond to this kind of case without either denying closure or being obliged to embrace contextualism. The problem arises from the fact that Zula is unable, in her present situation, to perceptually discriminate between a zebra and a cleverly disguised mule. This is why it is puzzling that Zula seems to be in a position, via closure, to know both [Z] and [­not-​­CDM], as this implies that she can tell the difference between the two scenarios. I think what is key here is to recognise that there is a way of knowing the difference which doesn’t imply that one can discriminate between the two scenarios at issue. If that’s right, then there needn’t be anything puzzling about Zula’s ability to know both [Z] and [­not-​­CDM], even though she cannot, ex hypothesi, perceptually discriminate between them. Consider the distinction between favouring as opposed to discriminating epistemic support. What Zula lacks is the discriminative ability to perceptually distinguish between the possibility that the creature before her is a zebra rather than a cleverly disguised mule. Someone with different perceptual abilities, like a zoologist, might be able to undertake such a discrimination, as might someone who has been able to make special checks of the creature. But neither applies to Zula. This needn’t mean that Zula is unable to know that the creature is a zebra rather than a cleverly disguised mule; however, it just means that if she does know this, it can’t be via her perceptual ability to discriminate between the two scenarios. It is possible to know that one scenario obtains rather than another, even though one cannot discriminate between them, via the possession

In Defence of Closure  117 of favouring epistemic support. In particular, if one has independent reasons for treating the alternative as being highly unlikely, then that will ordinarily suffice to enable one to know that the target scenario has obtained rather than the alternative. One can discount all kinds of implausible alternatives in this way, and thereby come to know that they don’t obtain. I know that the person at the mall who looks a bit like Kanye West is not really Kanye West because of a wide range of evidence that I possess which favours this believed scenario over the alternative. Why would Kanye be hanging out at the local mall? Would Kanye be dressed like that? Isn’t the real Kanye much shorter than this individual? And so on. This favouring evidence enables me to know that the person is a Kanye doppelgänger and not the real deal, even though I may be unable to perceptually discriminate between a good Kanye ­look-​­alike and the real thing. What goes for Kanye West and his ­look-​­alike in this scenario can equally apply in the Zula case. If Zula is an ordinary epistemic agent, then we would expect her to have all kinds of reasons at her disposal that would bear on the plausibility of the CDM hypothesis. She knows that this is a reputable zoo, and that reputable zoos don’t deceive their clientele. She knows that a deception of this sort would be hard to pull off, and expensive too, which makes it unlikely that anyone would attempt it. She knows that such a deception would be easily spotted by someone, even if it would trick her, and that there would be penalties for the zoo as a result. And so on. She thus has excellent rational support that favours her belief that the creature is a zebra over the CDM alternative. And if that’s right, then she can know both [Z] and [­not-​­CDM] even though she can’t perceptually discriminate between these two scenarios. One might object to the foregoing by claiming that it puts the threshold for perceptual knowledge implausibly high. Is it really required for perceptual knowledge that one has the kind of background reasons just noted? But notice that, on closer inspection, this is not a very demanding condition at all. To begin with, it doesn’t follow from the foregoing that in order for Zula to know [Z] she must first assemble a rational basis for excluding such scenarios as CDM. Indeed, there is no reason why she would have even considered such a scenario in normal circumstances. The claim is rather that it is only once Zula attempts to undertake the ­target-​ c­ ompetent deduction that such favouring epistemic support becomes relevant. Prior to this e­ rror-​­possibility coming into play, Zula can know [Z] even in the absence of such support, such as by possessing the discriminative capacities that would enable her to tell apart the scenario depicted in [Z] from relevant alternative scenarios (­that the creature before her is not a horse, say). In considering the relevant entailment, however, Zula is thereby o ­ bliged—​­as any agent would be when considering an inference of this ­kind—​­to consider whether she has grounds to dismiss this

118  Duncan Pritchard alternative, or whether she should instead reconsider her commitment to the entailing belief.10 This is the point that she should reflect on her background reasons and marshal the supporting favouring epistemic support. For any ordinary agent, however, such background reasons will be readily available, and hence they ought to be similarly available to Zula. That is, requiring that Zula has these background reasons in these conditions is not an austere epistemic requirement to impose. In contrast, if Zula really is the kind of agent for whom such background reasons are not a­ vailable—​­i.e., such that they cannot be summoned by reflection if required, without any further empirical investigation being ­needed—​ ­then it is not so clear that she has knowledge of the entailing proposition in the first place. At the very least, if she knows this proposition, then presumably she does so purely in virtue of exercising some sort of perceptual capacity, and not because her knowledge enjoys rational support.11 As we noted above, however, our concern in these deliberations over closure is precisely with rationally grounded knowledge. Once the distinction between favouring and discriminating epistemic support has been made explicit, then a further dialectical option becomes apparent. The kind of rational support that one gains for one’s belief via perception is standardly thought to be n ­ on-​­factive, such as considerations about the way things appear (­which obviously don’t entail how things are). There are epistemological proposals that reject this way of thinking about the reasons that ground perceptual knowledge, however, and maintain instead that a subject’s rational support for their perceptual knowledge in epistemically paradigmatic conditions can be factive. In particular, epistemological disjunctivism argues that in epistemically paradigmatic conditions one can have perceptual knowledge of a proposition, p, in virtue of being in possession of the factive reason that one sees that p (­where seeing that p entails p).12 It would take me too far afield to attempt to defend epistemological disjunctivism here, but I do want to note how treating it as a theoretical possibility bears on ­Zula-​­style cases. Epistemological disjunctivism doesn’t deny that Zula is unable to perceptually discriminate between a zebra and a cleverly disguised mule in the conditions that she is in. Factive reasons, however, provide a decisive kind of favouring epistemic support for one’s belief, in that they entail the falsity of alternative scenarios. In order to be the kind of epistemic agent that would be credited with factive rational support, Zula would also be likely to possess the kind of background reasons noted above, but crucially she wouldn’t require this background support in order to be able to know both [Z] and [­not-​­CDM], as the factive reason would suffice by itself. There would thus be no impetus in this case to deny closure as opposed to embracing Zula’s knowledge of [­not-​­CDM].13 The general line of reasoning that we have applied to the Zula case can be similarly applied, mutatis mutandis, to a range of other putative

In Defence of Closure  119 counterexamples to closure. In particular, it is generally applicable to any such case that essentially appeals, as the Zula case does, to a local ­error-​­possibility.14 More specifically, this response is applicable precisely because the ­error-​­possibility at issue, being local, doesn’t immediately call into question the favouring epistemic support that is available to the protagonist. This point straightforwardly applies when it comes to the background reasons that we noted above. Even if Zula has been given reasons for treating the CDM alternative seriously (­i.e., it has been rationally motivated), as opposed to merely having this alternative made explicit to her, it is still possible for Zula to legitimately appeal to background reasons in order to rationally dismiss it, as this background knowledge is not called into question by the CDM ­error-​­possibility. Matters become more complicated when it comes to the factive reasons defended by epistemological disjunctivism. This is because such factive reasons are only available to the subject in epistemologically paradigm conditions. Accordingly, it makes a difference whether the CDM ­error-​­possibility has been merely raised or rationally motivated, since in the latter case the subject is no longer in the right conditions in order to possess factive reasons.15 Nonetheless, so long as this ­error-​­possibility is merely raised, then the factive reasons are available to enable Zula to discount this alternative. Moreover, even if it is rationally motivated, it is still possible for Zula to appeal to her background reasons in the way just described in order to dismiss it. The upshot is that we can accommodate the intuitions in play in ­Zula-​­style cases involving local ­error-​­possibilities without denying such an intuitive principle as closure (­or else going contextualist). So what then, if anything, does provide the impetus for rejecting this principle? I think that for this we need to focus not on cases involving local ­error-​ p ­ ossibilities but on global ­error-​­possibilities, since on the face of it at least they call into question both one’s background reasons and also that one is in the kind of epistemic conditions that could possibly support factive reasons. In particular, the real case for denying closure concerns the kind of ­error-​­possibilities characteristically appealed to in the context of radical scepticism. Let’s follow convention and focus on the radical sceptical hypothesis ­ rain-­​­­in-­​­­a-​­vat being ‘­fed’ deceptive experithat one is a disembodied b ences [BIV], such that most of one’s beliefs are false. Let’s now imagine a subject who is in entirely normal epistemic conditions, and so is not a victim of the BIV hypothesis, nor is such a scenario even modally close. Our subject believes something mundane that entails [­not-​­BIV], such as that she is presently sitting down [E]. We would naturally attribute knowledge to our subject of [E]. And yet, if she does know [E], then since this obviously entails [­not-​­BIV], it follows that our subject can reason, via closure, to knowledge of [­not-​­BIV].

120  Duncan Pritchard Crucially, however, it is widely accepted that it is at least problematic to treat subjects as having knowledge of the denials of radical sceptical hypotheses. How could our subject come to know [­not-​­BIV]? Given that BIV depicts a scenario that is ex hypothesi indistinguishable from the subject’s ordinary experiences, what would she possibly appeal to in order to exclude it? In particular, the subject lacks a perceptual capacity to discriminate between everyday circumstances and the BIV alternative. Moreover, given the radical nature of the BIV hypothesis, it also calls the subject’s background reasons into question too, and hence, it does not seem appropriate for the subject to appeal to favouring epistemic support via this route. So, how then can it be possible for our subject to know both [E] and [­not-​­BIV], given that she has no way of distinguishing between the two scenarios at issue, either directly via a perceptual capacity or indirectly via appeal to favouring epistemic support? Radical sceptical arguments of this kind thus pose a challenge for closure, of a kind that is distinct from the difficulty that is presented by Z ­ ula-​­style cases involving local ­error-​­possibilities. So, how might we deal with this challenge to closure? I want to explore two options. The first is to revisit the idea of favouring epistemic support that we saw could be effective in dealing with ­Zula-​­style cases. This avenue might seem unpromising given the point we just made about how radical sceptical ­error-​­possibilities call into question one’s background reasons. Recall, however, that there was a second way of appealing to favouring epistemic support available, which concerned the factive reasons defended by epistemological disjunctivism. According to this proposal, if one is in epistemically paradigm conditions, then one can have available to one factive reasons in support of one’s perceptual knowledge. As we noted above, such factive reasons would provide one with a decisive form of favouring epistemic support to enable one to know the entailed proposition even when one lacks the relevant discriminative powers to exclude the target ­error-​­possibility. Could this strategy be applied to the radical sceptical case? One might initially find such an approach suspect. If the radical sceptical ­error-​­possibilities call into question one’s background reasons, then why don’t they also undermine one’s factive reasons? In particular, why don’t they entail that one is not in the kind of epistemically paradigm conditions that are the prerequisite of being in possession of factive reasons? There is, however, an important factor that is relevant here, which is whether the target ­error-​­possibility has been rationally motivated, or whether it is merely being considered by the subject. We noted above that different epistemic burdens are imposed on the subject in the different cases, especially with regard to factive reasons. In particular, if Zula is provided with good reasons for thinking that there is the kind of deception taking place that would motivate the CDM alternative, then it’s

In Defence of Closure  121 unlikely that she is in the kind of epistemically paradigm conditions that would be needed for her perceptual belief to enjoy factive rational support. If that’s right, then wouldn’t that point also neutralise the appeal to factive reasons in response to radical scepticism? Notice, however, that radical sceptical ­error-​­possibilities are precisely not rationally motivated. For one thing, it is hard to see what possible motivation could be given. What could possibly function as a reason for thinking that one’s beliefs are radically in error, such that it wouldn’t itself be called into question by the very ­error-​­possibility at issue? In any case, it is certainly not part of the employment of radical sceptical hypotheses as part of the radical sceptical puzzle to rationally motivate them. This would bring with it empirical commitments that the radical sceptic would not want to be saddled with. Moreover, radical scepticism of this kind, at least when presented in its strongest form, is a putative paradox, in that it is meant to be exposing deep tensions within our epistemological concepts.16 In presenting a paradox, however, one is meant to draw only on commitments that we all (­supposedly) share, and not advance controversial claims, much less controversial empirical claims, as that would undermine the thesis that what is under consideration is a genuine paradox. That’s why there is no contention as part of the radical sceptical paradox that we have any reason for thinking that we are the victim of a radical sceptical hypothesis. The thesis is just that such a scenario is possible and that we cannot exclude it, but that our failure to exclude it conflicts with our everyday knowledge (­where our commitment to closure is playing a large part in setting up that conflict). If radical sceptical hypotheses are not rationally motivated, however, but merely raised, then that alters the dialectical situation somewhat. In particular, there is no basis for treating the subject as not being in epistemically paradigm conditions, such that factive reasons are unavailable to her. Moreover, notice that the factive reasons advocated by epistemological disjunctivism are different to the sort of background reasons that might be appealed to in the context of radical scepticism. In the BIV case, for example, the latter might consist of considerations concerning the state of current technology that run counter to the BIV hypothesis. But what real purchase do these considerations really have on the plausibility of the BIV scenario? After all, these are considerations that are entirely compatible with the truth of the BIV hypothesis. The factive reasons that epistemological disjunctivism advances are very different in this regard. For what epistemological disjunctivism is precisely challenging is a key presupposition of the radical sceptical paradox to the effect that the rational support one’s beliefs enjoy, even in epistemically paradigm conditions, must always be ­non-​­factive. That is, radical scepticism is trading on the fact that even in ideal epistemic conditions the rational support one’s beliefs enjoy is compatible with those beliefs being massively false. If that’s right, then even if the radical

122  Duncan Pritchard sceptical hypotheses are not only false but also modally ­far-​­fetched, and even if no rational basis has been offered (­or could be offered) for taking them seriously, it would still remain that one lacks a rational basis for excluding them. In contrast, however, if one grants that one’s perceptual knowledge can be supported by factive reasons when one is in ideal epistemic conditions, and radical scepticism has offered no rational basis for thinking that one is not in such conditions, then it is open to one to appeal to one’s factive reasons to exclude radical sceptical scenarios. In particular, notice that one is now in a much better dialectical position than one would be if one could only appeal to background reasons. For unlike one’s background reasons, the factive reasons that support one’s perceptual knowledge actually entail the falsity of radical sceptical hypotheses.17 The upshot is that epistemological disjunctivism offers one way of dealing with the challenge to closure presented by the radical sceptical paradox. Rather than rejecting closure, one can instead embrace the idea that agents can have knowledge of the entailed proposition. This is not in virtue of having some mysterious perceptual capacity that enables them to discriminate between everyday life and the sceptical alternative, but rather due to the possession of favouring epistemic support. Moreover, unlike one’s background reasons, this favouring epistemic support does speak to the radical sceptical hypothesis at issue, by entailing its falsity (­as opposed to being entirely compatible with its truth). Whether one finds such a strategy plausible will, of course, depend on whether one finds epistemological disjunctivism credible (­and we haven’t argued for this thesis here). Nonetheless, the point is that on this epistemological proposal there is no special challenge raised for closure in the context of radical scepticism.18 While I hold that epistemological disjunctivism has an important part to play in responding to the problem of radical scepticism, ultimately I don’t think that it offers the right response to ­closure-​­based radical scepticism.19 The reason for this is that I’m independently convinced by Wittgenstein’s account of hinge commitments. In essence, this is the idea that it is in the nature of our system of rational evaluation that it presupposes that one has certain basic ‘­hinge’ commitments which are themselves immune to rational evaluation. It follows that one does not have specific reasons for regarding one’s hinge commitments as true. Moreover, these hinge commitments are not like hypotheses or assumptions, which one might endorse without having any commitment to their truth, but are rather claims that one is optimally certain of. Indeed, far from being concerned with theoretical theses, they instead primarily consist of entirely mundane claims, such as that one has two hands or that one’s name is s­ uch-­​­­and-​­such. One’s visceral certainty (­in normal conditions) in these claims is manifest in one’s actions, such that genuine (­as opposed to merely performative) doubt of them is simply impossible.

In Defence of Closure  123 There are various accounts of the Wittgensteinian notion of a hinge commitment available in the literature, though I take the former description to be broadly common ground among most of these views. Where they tend to depart from each other is in terms of what they add to this basic account. 20 On one prominent proposal, for example, our hinge commitments can be in the market for knowledge even despite the hinge role that they play in our rational evaluations. For although we cannot have any rational basis for thinking that they are true, there is nonetheless a kind of strategic rational basis for their ­truth—​­an epistemic entitlement, as some commentators put ­it—​­that accrues to them in virtue of how they are constitutive of our system of epistemic rationality. As we might put the point, if in order to be an epistemically rational subject at all one is required to have hinge commitments, then it must be epistemically rational to believe one’s hinge commitments even though, by definition, one cannot have a reason for thinking that these commitments are true. Accordingly, this opens up a way for them to count as knowledge, even though one has no reason for regarding them as true.21 I don’t find such an approach plausible, however. There are various reasons for this, but I want to focus on one core concern, which is that this way of thinking about hinge commitments doesn’t take on board the nature of the propositional attitude in play. Recall that our hinge commitments exhibit an optimal certainty that is manifest in our actions, even while being divorced from any rational basis for their truth. Wittgenstein characterises this arational certainty as being visceral, ‘­animal’; it is simply there ‘­like our life’. 22 Wittgenstein thus regards our hinge commitments as involving a distinctive kind of propositional attitude. 23 Given how broad our folk notion of belief i­s—​­which takes in a wide range of propositional attitudes, including acceptances and ­faith-​ l­ike commitments, as well as the kind of propositional attitude that is involved in having ­knowledge—​­one could truly describe our hinge commitments as beliefs. 24 But to do so glosses over important distinctions that are relevant here. In particular, our hinge commitments are not assumptions or hypotheses, such that one might be in some sense committed to them (­e.g., committed to act as if they are true), without thereby being committed to their truth. For the visceral certainty we have for our hinge commitments ensures that they are incompatible with agnosticism about the truth of the target proposition. More importantly for our purposes, our hinge commitments are also not beliefs in the specific sense of belief that is most relevant to ­epistemology—​­viz., that propositional attitude which is a constituent part of rationally grounded knowledge. Belief in this ­sense—­​­­K-​­apt belief, if you ­will—​­bears certain basic conceptual connections to reasons and truth that set it apart from other propositional attitudes that might fall under the broad folk conception of belief, including our hinge commitments. Here is one key consideration in this regard,

124  Duncan Pritchard which is that ­K-​­apt beliefs are such that one cannot retain one’s ­K-​­apt belief that p while recognising that one has no rational basis for the truth of p. If one retains one’s commitment to p even while recognising this lack of a rational basis, then whatever one’s propositional attitude in this regard, it cannot be a K ­ -​­apt believing, but must instead be some other propositional attitude (­such as a wishful thinking that p, for example). Crucially, however, one’s hinge commitments precisely fail this test for ­K-​­apt belief, in that they are ­all-​­out commitments to the truth of p that one retains even when one comes to appreciate that, as hinge commitments, one lacks a rational basis for them.25 That is, while one might profess doubt of one’s hinge c­ ommitments—​­as part of a philosophical investigation, for e­ xample—​­this will not be genuine doubt of them, as will be apparent from how one’s actions continue to manifest one’s hinge certainty. That our hinge commitments are not ­K-​­apt beliefs is important, since it highlights that they are simply not in the market for knowledge, as our commitment to them is not the kind of propositional attitude that could function as a constituent part of rationally grounded knowledge.26 It follows that attempts to show that they can be known via appeal to broadly strategic epistemic considerations are not going to work. For even if it is true that there is a positive epistemic story to tell about our hinge commitments on this front, the fact remains that our hinge commitments are simply not the kind of propositional attitude that could amount to knowledge. This issue is relevant to our present concerns because of how our commitment to the denials of radical sceptical hypotheses seem to be hinge commitments. That is, the general form of a hinge commitment is that it is the kind of consideration that would reveal wholesale and fundamental error on one’s part, such as being wrong about whether one has hands in normal conditions (­i.e., as opposed to the abnormal conditions of waking up after a serious car accident).27 If that’s right, then it follows that one is hinge committed to the falsity of radical sceptical hypotheses, as they directly concern the possibility of wholesale and fundamental error. Accordingly, if such propositions are not in the market for knowledge, then that seems to pose a challenge for the closure principle. One response to this puzzle might be to simply accept that a Wittgensteinian hinge epistemology entails the denial of closure, and some commentators have embraced that line. 28 Another alternative is to contend that one can have knowledge of one’s hinge commitments and thus that one doesn’t need to reject closure. Given the foregoing, one might think that I am embracing the first option over the second, but in fact I think that this is a false dilemma, in that our inability to know our hinge commitments is consistent with the closure principle. In order to see why this is the case, we need to remember how we formulated that principle above. Recall that it was argued that if we are to regard closure as a compelling principle that isn’t susceptible to straightforward counterexamples, then it is crucial that it is understood

In Defence of Closure  125 diachronically as a competent deduction principle. In particular, via c­ losure-​­based inferences, one thereby comes to acquire a belief in the entailed proposition that is based on the competent deduction. Since the idea is that closure delivers rationally grounded knowledge, we are clearly concerned with the acquisition of K ­ -​­apt belief in the entailed proposition. We have seen, however, that one simply cannot have a ­K-​ a­ pt belief in one’s hinge commitments, as the visceral certainly involved in the latter excludes this possibility. 29 Accordingly, there is no need to deny the closure principle in order to accommodate the unknowability of our hinge commitments, since there cannot be a ­closure-​­based inference involving our hinge commitments that generates the target counterexample. Rather than there being exceptions to the closure principle, it is instead simply inapplicable to our hinge commitments. The challenge posed by local and global ­error-​­possibilities to the closure principle is thus found to be illusory. To begin with, we have seen the importance of properly characterising what this principle amounts to, and thus what we are meaning to defend, as not every formulation of this principle is even prima facie plausible. Next, we need to distinguish between local and global e­ rror-​­possibilities, as we have seen that they pose distinct issues for the closure principle, and hence, the response we should mount is similarly different. In the case of local ­error-​­possibilities, we saw how favouring epistemic support can enable a subject to have knowledge of the entailed proposition in the relevant cases, either by appealing to background knowledge or by appealing to the factive reasons advocated by epistemological disjunctivism. In the case of the global ­error-​­possibilities raised as part of the problem of radical scepticism, the background knowledge is no longer applicable because it is called into question by the ­error-​­possibility in play (­in contrast to scenarios involving local ­error-​­possibilities). Nonetheless, there are still ways of responding to this version of the challenge. On the one hand, one can directly appeal to one’s factive reasons in order to maintain that one can have knowledge of the entailed proposition. On the other hand, one can opt for my favoured route of maintaining that one is hinge committed to the denials of radical sceptical hypotheses, and that such commitments are simply not in the market for knowledge. More specifically, one is unable to have a ­K-​­apt belief in one’s hinge commitments of the kind that is relevant to a c­ losure-​­based inference. Accordingly, far from closure failing in this case, closure is simply inapplicable. The upshot is that regardless of whether the e­ rror-​­possibility is local or global, there is no challenge to the closure principle, properly understood.30

Notes 1 Note that in what follows I will be restricting my attention to single premise formulations of closure.

126  Duncan Pritchard 2 As Dretske (­ 2005) puts the point in more recent work, our knowledge doesn’t penetrate to the ‘­heavyweight implications’ of what is known, where these concern the legitimate presuppositions of our knowledge. 3 See also Nozick’s (­1981) influential denial of the closure principle, which also treats knowledge as being closed under known entailments. 4 At the very least, it is usually thought that there is a ­belief-​­like propositional attitude that is a constituent of knowledge. (­T he reason for this caveat will become apparent below). 5 This is essentially the formulation of the closure principle put forward by Williamson (­2000, 117) and Hawthorne (­2005, 29). 6 That’s at least one way of thinking of the epistemic externalism/­internalism distinction as it is applied to knowledge, such that epistemic externalism about knowledge maintains that there can be knowledge that is not rationally grounded, whereas epistemic internalism about knowledge denies this. See Pritchard (­forthcoming) for more on this point. For further discussion of the epistemic externalism/­internalism distinction more generally, see Vahid (­2011). 7 Infallibilism of this kind is usually treated as an implausible position, though for an exception, see Unger (­1971, 1975). 8 Dretske (­1970, 1981) himself advanced this relevant alternatives line. For some useful discussions of Dretske in this regard, see Stine (­1976) and Cohen (­1988). 9 Strictly speaking, contextualism of this kind should be expressed metalinguistically, as being concerned with ‘­knows’, but we can set aside this complication for our purposes. For some key defences of contextualism of this variety, see Cohen (­1988, 1999, 2000, 2002), DeRose (­1995), and Lewis (­1996). 10 Note that I’m not suggesting here that an ­error-​­possibility figuring in a competent deduction is the only way to make it relevant in the applicable sense (­i.e., such that one needs to exclude it in order to have knowledge of the target proposition), only that this is sufficient to make it relevant. For example, just as there can be normative epistemic defeaters, which one needs to exclude even if one isn’t aware of them, so there can be ­error-​­possibilities that are relevant in the applicable sense, and so must be excluded, even if one isn’t aware of them (­on account of the fact that one ought to be aware of them, just as with normative epistemic defeaters). (­Note too that I’m also not suggesting that merely being aware of an ­error-​­possibility makes it relevant in the applicable sense, though for simplicity in what follows I will proceed as if this does suffice. Given that such a concession can only benefit my opponent, it shouldn’t be controversial). For further discussion of these issues, see Pritchard (­2010). See also endnote 11. 11 This is the line I advance in Pritchard (­2010) as part of what I call a ‘­­two-​ t­iered’ account of perceptual knowledge. In particular, while one can have knowledge purely in virtue of exercising certain perceptual capacities, once one entertains (­indiscriminable) ­error-​­possibilities like CDM, and so infers their denial, then favouring reasons are required. Such an account is thus distinct from both conservativism and dogmatism in this regard, at least as those two proposals are applied to local ­error-​­possibilities like CDM (­i.e., as opposed to the kind of radical sceptical scenarios we will consider below). It is distinct from the former since although it does demand the possession of epistemic support that bears on the entailed proposition insofar as one undertakes the relevant closure inference, it doesn’t deny that one can know the entailing proposition without it. And it is distinct from dogmatism in that it denies that one can know the entailed proposition via competent deduction

In Defence of Closure  127 closure absent the relevant favouring epistemic support. For a key defence of conservatism, see Wright (­2004). For a key defence of dogmatism, see Pryor (­2000). 12 Epistemological disjunctivism is rooted in the work of McDowell (­ e.g., 1995), but see, especially, Pritchard (­2012) for a full development and defence of this proposal. 13 Favouring reasons of the more familiar kind would still be required in n ­ on-​ ­paradigmatic epistemic conditions though, such as where the subject is in a context in which the CDM ­error-​­possibility has been rationally motivated (­and hence counts as a misleading defeater). 14 Such as, for example, the hypothesis that the table one is looking at is not red but rather white and illuminated by red ­light—​­see Cohen (­2002, 313). 15 See Pritchard (­2 011a, 2012, part 1) for further discussion of what such epistemologically paradigm conditions involve and in particular why they include both objective factors (­such as whether the ­error-​­possibility at issue is objectively likely to occur) and subjective factors (­such as whether the ­error-​­possibility at issue has been rationally motivated, even if it is in fact objectively unlikely to occur). Note that it follows from how we have set up the Zula case that the CDM alternative is objectively unlikely, as that’s required to ensure the security of the intuition that she knows [Z]. 16 I defend the idea that radical scepticism is best understood as a (­putative) paradox in more detail in Pritchard (­2014, 2015, part 1). See also Stroud (­1984) for an influential defence of this idea. 17 As I discuss in Pritchard (­2012, part 3), there might be dialectical limitations on the actual citing of factive reasons in a debate with the radical sceptic, but the point is that from a purely epistemological point of view, there is nothing amiss in treating factive reasons as providing the relevant favouring epistemic support. 18 See Pritchard (­2012, part 3) for a fuller development of such a ‘­­neo-​­Moorean’ response to radical scepticism. See also Pritchard (­2007, 2008). 19 I think the real ­anti-​­sceptical import of epistemological disjunctivism lies in responding to what I call u ­ nderdetermination-​­based radical scepticism, which turns on a different epistemic principle to closure and generates what I claim is a logically distinct radical sceptical argument. See Pritchard (­2015, parts 1 & 3) for the details. 20 For some of the key discussions of a Wittgensteinian hinge epistemology, see McGinn (­1989), Williams (­1991), ­Moyal-​­Sharrock (­2004), Wright (­2004), Coliva (­2010, 2015), and Schönbaumsfeld (­2016). I summarise my own views in this regard in Pritchard (­2015, part 2). For an overview of this debate, see Pritchard (­2011b, 2017). 21 The primary defence of this line can be found in the work of Wright (­e.g., 2004). See also Coliva (­2015) for an important development of this general type of proposal. 22 See Wittgenstein (­1969, §359 & §559). 23 This is why I refer to our hinge commitments rather than, as is more common in the literature, describing them as hinge propositions. The latter locution prompts us to focus on the particular propositional content that is in play, but that’s not what is distinctive about our hinges, which is rather the special kind of propositional attitude that we bear to these contents in appropriate conditions. 24 For a useful article exploring the different kinds of propositional attitudes that are typically treated as beliefs, see Stevenson (­2002).

128  Duncan Pritchard 25 See Pritchard (­2015, part 2) for further discussion of the propositional attitude involved in our hinge commitments. In particular, this attitude is sui generis, in that it is distinct from other kinds of propositional attitude in the vicinity of which we are familiar, such as ­K-​­apt beliefs, acceptances, hypotheses, assumptions, aliefs, and so on. As I also explain, on my account of hinge commitments, there is a sense in which they can be indirectly responsive to rational considerations even if they are not directly responsive, in virtue of how the content of one’s hinge commitments is determined by one’s set of ­K-​­apt beliefs, and these are in principle responsive to rational considerations in the usual way. 26 It follows that although our hinge commitments are unknown, this does not constitute an epistemic failing on our part, as they are not in the market for knowledge. Our inability to know our hinge commitments is thus akin to our inability to draw a ­circle-​­square. Relatedly, one is not ignorant of one’s hinge commitments either, even though they are unknown (­and not truly believed either, at least in the specific ­K-​­apt sense)—​­see Pritchard (­2021). 27 Indeed, on my account of hinge commitments, there is really only one overarching hinge commitment that we all s­hare—​­what I refer to as the über hinge commitment—​­which is the primitive certainty that one is not radically and fundamentally in error. All of one’s specific hinge commitments are held to be manifestations of the über hinge commitment. See Pritchard (­2015, part 2). 28 See, for example, Coliva (­2015). 29 ­Relatedly—​­although I haven’t had space to explore this contention h ­ ere—​ o ­ ne cannot acquire one’s hinge commitments via rational processes such as competent deduction anyway. See Pritchard (­2015, part 2), where I discuss the distinctive way in which one’s hinge commitments are acquired. 30 I am grateful to Yuval Avnur for helpful discussion of the issues covered in this chapter.

References Cohen, S. (­1984). ‘­Justification and Truth’, Philosophical Studies 46, ­279–​­96. ——— ​­(­1988). ‘­How to Be a Fallibilist’, Philosophical Perspectives 2, ­91–​­123. ——— ​­(­1999). ‘­Contextualism, Skepticism, and the Structure of Reasons’, Philosophical Perspectives 13, ­57–​­89. ——— ​­(­2000). ‘­Contextualism and Skepticism’, Philosophical Issues 10, ­ 4–​­107. 9 ——— ​­(­2002). ‘­Basic Knowledge and the Problem of Easy Knowledge’, Philosophy and Phenomenological Research 65, ­309–​­29. Coliva, A. (­2010). Moore and Wittgenstein: Scepticism, Certainty, and Common Sense, London: Palgrave Macmillan. ——— ​­(­2015). Extended Rationality: A Hinge Epistemology, London: Palgrave Macmillan. DeRose, K. (­1995). ‘­Solving the Skeptical Problem’, Philosophical Review 104, ­1–​­52. Dretske, F. (­1970). ‘­Epistemic Operators’, Journal of Philosophy 67, ­1007–​­23. ——— ​­(­1981). ‘­T he Pragmatic Dimension of Knowledge’, Philosophical Studies 40, ­363–​­78. ——— ​­(­2005). ‘­T he Case against Closure’, Contemporary Debates in Episte­ 3–​­26, Oxford: Blackwell. mology, (­eds.) E. Sosa & M. Steup, 1

In Defence of Closure  129 Hawthorne, J. (­2005). ‘­T he Case for Closure’, Contemporary Debates in Epistemology, (­eds.) E. Sosa & M. Steup, ­26–​­43, Oxford: Blackwell. Lewis, D. (­1996). ‘­Elusive Knowledge’, Australasian Journal of Philosophy 74, ­549–​­67. McDowell, J. (­1995). ‘­K nowledge and the Internal’, Philosophy and Phenomenological Research 55, ­877–​­93. McGinn, M. (­1989). Sense and Certainty: A Dissolution of Scepticism, Oxford: Blackwell. ­Moyal-​­Sharrock, D. (­2004). Understanding Wittgenstein’s On Certainty, London: Palgrave Macmillan. Nozick, R. (­ 1981). Philosophical Explanations, Oxford: Oxford University Press. Pritchard, D. H. (­2007). ‘­How to Be a N ­ eo-​­Moorean’, Internalism and Externalism in Semantics and Epistemology, (­ed.) S. Goldberg, ­68–​­99, Oxford: Oxford University Press. ——— (​­ ­2008). ‘­McDowellian N ­ eo-​­Mooreanism’, Disjunctivism: Perception, ­ 83–​­310, Oxford: Action, Knowledge, (­eds.) A. Haddock & F. Macpherson, 2 Oxford University Press. ——— ​­(­2010). ‘­Relevant Alternatives, Perceptual Knowledge and Discrimination’, Noûs 44, ­245–​­68. ——— ​­(­2011a). ‘­Epistemological Disjunctivism and the Basis Problem’, Philosophical Issues 21, 4 ­ 34–​­55. ——— ​­(­2011b). ‘­Wittgenstein on Scepticism’, The Oxford Handbook on Wittgenstein, (­eds.) O. Kuusela & M. McGinn, ­521–​­47, Oxford: Oxford University Press. ——— (​­ ­2012). Epistemological Disjunctivism, Oxford: Oxford University Press. ——— (​­ ­2014). ‘­S ceptical Intuitions’, Intuitions, (­ eds.) D. Rowbottom  & T. Booth, ­213–​­31, Oxford: Oxford University Press. ——— (​­ ­2015). Epistemic Angst: Radical Skepticism and the Groundlessness of Our Believing, Princeton, NJ: Princeton University Press. ​­ 2017). ‘­Wittgenstein on Hinge Commitments and Radical Scepticism in ——— (­ On Certainty’, Blackwell Companion to Wittgenstein, (­eds.) H.-​­J. Glock & J. Hyman, 5 ­ 63–​­75, Oxford: Blackwell. ——— ​­(­2021). ‘­Ignorance and Inquiry’, American Philosophical Quarterly 58, 1 ­ 11–​­23. ——— ​­(­forthcoming). ‘­Moderate Knowledge Externalism’, Externalism about Knowledge, (­ed.) L. R. G. Oliveira, Oxford: Oxford University Press. Pryor, J. (­2000). ‘­T he Skeptic and the Dogmatist’, Noûs 34, ­517–​­49. Schönbaumsfeld, G. (­2016). The Illusion of Doubt, Oxford: Oxford University Press. Stevenson, L. (­2002). ‘­Six Levels of Mentality’, Philosophical Explorations 5, ­105–​­24. Stine, G. C. (­1976). ‘­Skepticism, Relevant Alternatives, and Deductive Closure’, Philosophical Studies 29, 2 ­ 49–​­61. Stroud. B. (­1984). The Significance of Philosophical Scepticism, Oxford: Clarendon Press. Unger, P. (­1971). ‘­A Defence of Skepticism’, Philosophical Review 80, ­198–​­219. ——— ​­(­1975). Ignorance: A Case for Scepticism, Oxford: Clarendon.

130  Duncan Pritchard Vahid, H. (­2011). ‘­Externalism/­I nternalism’, Routledge Companion to Epistemology, (­eds.) S. Bernecker & D. H. Pritchard, ­144–​­55, London: Routledge. Williams, M. (­ 1991). Unnatural Doubts: Epistemological Realism and the Basis of Scepticism, Oxford: Blackwell. Williamson, T. (­2000). Knowledge and Its Limits, Oxford: Oxford University Press. Wittgenstein, L. (­1969). On Certainty, (­eds.) G. E. M. Anscombe & G. H. von Wright, (­tr.) D. Paul & G. E. M. Anscombe, Oxford: Blackwell. Wright, C. (­2004). ‘­Warrant for Nothing (­and Foundations for Free)?’, Proceedings of the Aristotelian Society (­supp. vol.) 78, ­167–​­212.

8 No Spindly Brown Grass Knowledge, Closure, and Subjective Distinguishability Genia Schönbaumsfeld

1 Introduction Conceptions of perceptual reasons in contemporary epistemology tend to be ­scepticism-​­driven. That is to say, epistemologists implicitly (­or explicitly) take the sceptical problem as a point of departure and then formulate a notion of perceptual grounds that, right from the start, makes serious concessions to scepticism.1 This constitutes, in effect, a ‘­lowest common denominator’ ­approach – the ​­ resulting view is whatever one might be left with after the sceptic (­or our sceptical alter ego) has done her work. But if that is the strategy, then it is not surprising that the outcome, to put it in Dretske’s memorable words, is that ‘­philosophy is a business where one learns to live with spindly brown grass in one’s own yard because neighbouring yards are in even worse shape’ (­Dretske 2005b: 43). In this chapter, I want to argue that to take such an approach is a mistake. We should rather proceed the opposite way. In other words, we should begin with a conception of perceptual reasons that fits the facts and our ordinary intuitions, and then explore what the ramifications for scepticism might be. We should not, ab initio, cede the ­high-​­ground to the sceptic and let her dictate the terms of engagement. I will proceed as follows. In order to show the ­scepticism-​­driven nature of much contemporary epistemology, I will make use of Dretske’s attempt to deny closure as a case study. For although most epistemologists have not been impressed by Dretske’s manoeuvre, the reasons that drive him to make it are widely shared. So, if I am right about the underlying diagnosis, the lesson will generalize and enable us to resist certain apparently compelling moves that in fact drive us right into the arms of scepticism. This is not yet to construct a full ­anti-​­sceptical strategy, but merely an attempt to undermine ways of thinking that are less compulsory than they might seem, and more ­scepticism-​­friendly than strictly required.

2 Scepticism and Closure In his paper, ‘­The Case Against Closure’, Dretske makes the following claims:

DOI: 10.4324/9781003104766-9

132  Genia Schönbaumsfeld Despite knowing that cookies are objective (­­mind-​­independent) objects, I can see (­roughly: tell by looking) that there are cookies in the jar without being able to see, without being able to tell by looking, that there are ­mind-​­independent objects…Maybe one has to know there are physical objects in order to see that there are cookies in the jar…, but one surely isn’t claiming to see that there are physical objects in claiming to see there are cookies in the jar. After all, hallucinatory cookies ‘­in’ hallucinatory jars can look exactly like real cookies in real jars. So one cannot, not by vision alone, distinguish real cookies from mental figments. One cannot see that the world really is the way it visually appears to be. (­Dretske 2005a: 14) This is quite a striking passage. Is it as s­ elf-​­evident as Dretske would like us to think that one ‘­surely isn’t claiming to see that there are physical objects in claiming to see there are cookies in the jar’? For, pace Dretske, one’s ordinary way of thinking about cookies does precisely seem to be that they are sweet, ­mind-​­independent physical objects. Imagine that you ask your friend whether she believes that there are cookies in the jar (­when there are), and she says ‘­yes’. Further imagine you then go on to ask her, ‘­do you believe that these cookies exist only in your mind?’, and I’m confident, she would say ‘­no’. Imagine further still that you then ask her whether she thinks that seeing that there are cookies in the jar is compatible with believing that she is hallucinating that there are, and I’m sure she would also say ‘­no’. Of course, one doesn’t settle philosophical questions by asking one’s friends. Nevertheless, what this exchange shows is that Dretske’s proposal appears quite r­ evisionary – ​­he is denying we would want to claim what most normal people, and, indeed, most philosophers (­when not mesmerized by scepticism) would want to say; namely, that to see that there are cookies in the jar is to see certain kinds of physical objects in the jar. So, why does Dretske want to say something that (­certainly at first glance, and perhaps later too) seems highly ­counter-​­intuitive? The reason is that he is already convinced by a certain sceptical argument. After all, he says, ‘­hallucinatory cookies “­in” hallucinatory jars can look exactly like real cookies in real jars. So one cannot, not by vision alone, distinguish real cookies from mental figments’. In other words, Dretske appears to be endorsing the conclusion of the ‘­I ndistinguishability Argument’, which runs as follows: 2.1  The Indistinguishability Argument Q1 In the bad case it only seems (­looks) to me as if P. Q2 I cannot subjectively distinguish P (­the good case) from the mere appearance as of P (­the bad case).

Knowledge, Closure, and Subjective Distinguishability  133 C1 If I cannot subjectively distinguish between P and the mere appearance as of P, I cannot know that P (­that I am in the good case). C2 I cannot know that P. That is to say, because I cannot visually distinguish a real cookie from the hallucination of ­one – ​­they both ‘­look’ the ­same – I​­ cannot know that the ‘­cookie’ in the jar in front of me is a real (­i.e. physical) cookie. This sounds like a sceptical conclusion. Interestingly enough, however, Dretske wants to persuade us that it is not, and this is where his (­in)­famous denying of the closure principle2 comes in. Dretske reasons as follows. When we perceive that P, P has certain ‘­heavyweight’ ­implications – such ​­ as that cookies are ­mind-​­independent physical ­objects – ​­that cannot themselves be perceived to be so, and which we consequently have no reason to believe, even though these implications are among the known consequences of P (­Dretske 2005a: 16).3 So, unless we deny that we need to know their ‘­heavyweight’ implications in order to know the ordinary things we think we know (­such as that there are cookies in the jar) – we ​­ have to accept the sceptical conclusion. In the face of this, Dretske contends, it is surely better to maintain that there are some things, namely, the ‘­heavyweight’ implications, that we needn’t know, despite the fact that we know that our ordinary knowledge depends on their truth (­Dretske 2005a: 17): ‘­The only way to preserve knowledge of homely truths, the truths everyone takes themselves to know, is…to abandon closure’ (­Dretske 2005a: 18).4 In other words, denying closure is a l­ast-​­ditch attempt to ward off the scepticism that, according to Dretske, would otherwise be inescapable. Let’s flesh out Dretske’s account a bit more. In his paper, ‘­Epistemic Operators’, where Dretske first developed these ideas, he explains that there are certain presuppositions associated with a statement, which, although their truth is entailed by the truth of the statement, are not part of what is operated on when we operate on the statement with one of our epistemic operators. The epistemic operators do not penetrate to these presuppositions. (­Dretske 1970: 1014)­5 For example, I go to the zoo and look at an enclosure labelled ‘­zebras’. I see animals that look like zebras. I conclude that there are zebras in the pen. However, can I know that these ‘­zebras’ are not mules cleverly disguised to look like zebras? No. That they are not mules cleverly disguised to look like zebras is a presupposition that I’m ruling out because it is highly unlikely. But can I know that this presupposition is true? No. My perceptual evidence, according to Dretske, is not sufficient to rule out this possibility (­Dretske 1970: 1016). Again, this sounds like a sceptical conclusion. So, how are we supposed to avoid scepticism? Dretske agrees with the sceptic that you cannot know

134  Genia Schönbaumsfeld that the zebras are not cleverly disguised mules. But he believes that this does not imply that we do not know that the animals are zebras, because he rejects the principle that gets us to this conclusion: the principle that if you do not know that Q is true, when it is known that P entails Q, then you do not know that P is true (­Dretske 1970: 1016). In other words, Dretske rejects the principle that knowledge is ‘­closed’ under known logical entailment. This, to put it mildly, generates some very odd consequences. It implies, for instance, that I can know that the zebra in the pen is not a mule, as my perceptual e­ vidence – ​­its looking like a z­ ebra – ​­speaks against the relevant alternative that it’s a mule. Nevertheless, I cannot know that the zebra is not a cleverly disguised mule, as my perceptual evidence does not speak against this scenario. Hence, I can know that the animal in the enclosure is not a mule, but I cannot know that it is not a cleverly disguised mule. And this seems not just to be an ‘­abominable conjunction’,6 but courts downright logical contradiction: I can both know and not know that the animal in the enclosure is a mule. Similarly, I can know, on Dretske’s conception, that the wall is red and not white, but I cannot know that the wall is not white cleverly illuminated to look red. This sounds like a case of wanting to have one’s logical cake and eat it too. One might try to alleviate the tension by confining oneself to ‘­looks’ talk. That is to say, if all that Dretske means by saying we know the animal is a zebra, not a mule, is that we know that the animal looks like a zebra, then there is no logical contradiction. For that the animal looks like a zebra is, of course, perfectly compatible with the animal in fact being a cleverly disguised mule. Unfortunately, however, if we take this route, it seems that we are not left with much knowledge at all. For how can something’s looking like a zebra be sufficient for knowing it’s a zebra? All sorts of things might look like a zebra and not be a zebra. My child’s toy zebra looks like a zebra and is not a zebra.7 A tiger might look like a zebra if my vision is obscured by branches and sufficiently far away, etc. Consequently, Dretske seems to be faced with the following dilemma: either his account is logically problematic (­I can both know and not know that an animal is not a mule), or we can’t know much more than how an object happens to look. And a sceptic would, naturally, be quite happy with either claim. Does Dretske’s later, ‘­­information-​­caused belief’ conception (­Dretske 1981) save him from being impaled on this dilemma? It appears not. For even if we were to agree that the animal’s looking like a zebra causes one to believe that the animal is a zebra and not a mule, it is still going to be the case that this information underdetermines whether the animal is a mule cleverly disguised to look like a zebra. Consequently, on this conception too, I’m going to come out as embracing both that I can know that the animal in the pen is a zebra and not a mule, but I cannot know that the animal is not a cleverly disguised mule. Hence, in order to avoid

Knowledge, Closure, and Subjective Distinguishability  135 this problem, Dretske would again have to retreat to ‘­looks’ ­talk – ​­all one can know is that the animal looks like a zebra; nothing more, nothing less. But, if so, he cannot really avoid the sceptical conclusion. If, strictly speaking, all we can know, on Dretske’s conception, is that it looks like there are cookies in the jar or that there are zebras in the pen, then the information that we are receiving from our perceptual experiences always underdetermines whether these experiences are of actual physical cookies or merely figments of our imagination. In other words, our perceptual experiences, on this view, will always be experiences of something which leaves it open whether these experiences are of actually existing physical objects or not. And this implies that whether we find ourselves in the good or the bad case, our perceptual reasons are the same8; they always underdetermine whether we are confronted by an actual physical object or are merely hallucinating it. In The Illusion of Doubt, I called this view, which is shared by the vast majority of contemporary epistemologists, the Reasons Identity Thesis (­R IT).9 Endorsement of this thesis has the following unwelcome implications for Dretske’s case against closure. Consider a standard c­losure-​­based sceptical argument: 2.2  Standard C ­ losure-​­Based Sceptical Argument (­BIV1) If I know I have two hands (­or that there is a cookie or zebra in front of me), then I know I’m not a ­brain-­​­­in-­​­­a-​­vat (­BIV). ((­BIV1) is motivated by the closure principle, ‘­If S knows that P, and S competently deduces Q from P thereby coming to believe that Q on this basis while retaining her knowledge that P, then S knows that Q’.) But, it seems, (­BIV2) I don’t know I’m not a BIV. So, (­BIV3) I don’t know I have two hands (­or that there is a cookie or zebra in front of me). Dretske wants to avoid (­BIV3) by denying the closure principle. That is to say, he contends that he can know that he has hands (­or that there are cookies in the jar) without this knowledge transmitting to whatever is logically entailed by this knowledge. But, as we have already seen, this in turn implies that Dretske must confine himself to knowledge of how things appear (­how they look); otherwise, he could neither avoid the logical contradictions discussed above, nor make plausible why knowledge of the presence of cookies does not confer knowledge of sweet physical objects (­since everyone agrees that cookies are sweet physical objects). If so, however, then the antecedent in (­ BIV1), on Dretske’s parsing, doesn’t mean what it ordinarily means. For, on Dretske’s reading, ‘­knowing one’s got hands (­cookies, etc.)’ means ‘­knowing that one has an experience as of a hand’ (­knowing that there is something that looks

136  Genia Schönbaumsfeld like a hand), not ‘­knowing that there is an actual physical hand there’ (­since this, on Dretske’s view, is a ‘­heavyweight implication’ that we cannot know). But, if so, (­BIV1) needs to be modified to reflect this fact. In other words, (­BIV1) should read: (­BIV1)*: If I know that I am having an experience as of there being a hand in front of me, then I know I’m not a BIV. If this is what (­BIV1) really means, however, then it generates the unfortunate consequence that the premise now turns out to be false: one cannot deduce that one is not a BIV from the presence of an experience as of there being a hand in front of one. Experiences as of hands are n ­ on-​ ­factive and don’t imply the presence of real, physical hands.10 If (­BIV1)* is false, though, it naturally cannot serve as a premise in a ­closure-​­based sceptical argument. Hence, once this is made clear, it becomes apparent that, on Dretske’s conception, no c­ losure-​­based sceptical argument can run and, therefore, that a denial of closure is entirely redundant.11 Given that Dretske is already starting with the claim that one can only have knowledge of h ­ and-​­like appearances (­experiences as of there being hands), never of the (­physical) hands themselves, and one cannot deduce something from a proposition that isn’t entailed by it, the denying of closure is a red herring. For to deny closure requires denying that one can ‘­deduce’ (­and thereby come to know) that one is not a BIV from the fact that one is seeing a physical hand, as only knowledge of a physical hand entails the falsity of the sceptical conclusion. But that is not what Dretske is denying. Dretske is denying that one can deduce that one is not a BIV from a h ­ and-​­like appearance, and this is both true and entirely consistent with the closure principle. In other words, the reason why I cannot, for example, reason from ‘­I see that there is a cookie in the jar’ to the presence of a physical cookie is not because closure fails (­as Dretske claims), but rather because Dretske has already accepted the view that, even in the good case, c­ ookie-​­experiences always fall short of providing knowledge of real (­physical) cookies (­i.e. because he has already accepted the Reasons Identity Thesis). Hence, all I can know, on this conception, is that I am having experiences as of there being cookies in front of me, but such n ­ on-​­factive experiences are of course not sufficient to ground knowledge of ‘­heavyweight’ cookies (­about the fact that there are real cookies there). In his epistemological ­self-​­profile (­Dretske 2010), Dretske comes very close to admitting this. He says: If I can’t simply assume that deceptive demons and cunning hoaxes are not possibilities my evidence need rule out, possibilities I can ignore…then I really can’t see what I purport to ­see – ​­that my wife is on the sofa. I can no longer see, just by looking, that she is there because these ways of her not being there would look exactly the same way to me…So closure fails in these special (­isolated) cases

Knowledge, Closure, and Subjective Distinguishability  137 because the key premise, P, is not available to those who want to use it to infer Q. (­Dretske 2010: 134) The latter sentence is revealing. If P is not available to those who want to use it to infer Q, then it is not that closure fails. Rather, the fact that I don’t know that P means that I can’t use knowledge of P to infer that Q. And this is exactly what one would expect; exactly what the closure principle says: if I don’t know that Q, this is because I don’t know that P. Given that all I can know, on Dretske’s view, is that I have an experience as of my wife’s being on the sofa, and this is compatible with being deceived about this experience, it follows that Dretske, as he himself admits, cannot know that his wife is really on the sofa (­it only looks/­seems/­ appears that way). Consequently, the blame for the sceptical implication lies squarely with Dretske’s antecedent endorsement of the Reasons Identity Thesis, not with the closure principle, which only comes into play if we have already established that knowledge of physical objects is possible. Since this is deemed impossible by Dretske ab initio, however, no knowledge of P is, in fact, available from which one could infer Q. For all one can know, on Dretske’s conception, is that it appears to one that ­P – ​­and from an appearance that P it is naturally not possible to deduce that Q. The closure principle, therefore, cannot furnish one with new empirical knowledge, but only informs one of what follows logically from what one already knows. Hence, competent deduction cannot give one any ­anti-​­sceptical knowledge that one did not already possess before (­one made the deduction) – it ​­ can only make one aware of the fact that the actual knowledge one possesses has a­ nti-​­sceptical implications. Conversely, if one’s putative knowledge that P is confined to knowledge of its merely seeming to one that P, then one cannot, in any case, derive an ­anti-​­sceptical conclusion from such an appearance, and so a denial of closure is obsolete. So, instead of scapegoating an innocent principle, what we need to do is to examine the reasons why Dretske (­and others) believe that we can never know more than how things look (­seem). That is to say, we need to subject the Indistinguishability Argument and the conception of perceptual reasons that it appears to thrust upon us to more detailed scrutiny, as it is acceptance of these notions, not of the closure principle, that pushes us towards scepticism.

3 Experiential Markers and Subjective Distinguishability As we have seen, Dretske’s master argument consists in claiming that by seeing cookies in a jar, we can never come to know that there really are sweet physical objects in the jar, as ‘­hallucinatory cookies “­in”

138  Genia Schönbaumsfeld hallucinatory jars can look exactly like real cookies in real jars. So one cannot, not by vision alone, distinguish real cookies from mental figments’ (­Dretske 2005: 14). Since actual cookies and hallucinations of cookies are phenomenologically indiscriminable, in other words, Dretske thinks that we can never know that we are confronted by real cookies rather than mental figments. Whatever else is wrong with this reasoning (­of which more below), the first thing to note here is that this conclusion is a bit quick. It hardly goes without saying that if I cannot tell the difference by ‘­looking’  –​ ­where I presume this means ‘­introspectively distinguishing on the basis of phenomenology alone’ – ​­I can never know that what I am looking at is real and not a hallucination. For even if I cannot tell the difference on the basis of how things appear to me, I could conduct further empirical investigations (­such as, say, checking for paint in the zebra case), or looking more closely; or consulting other, maybe better trained, people. So, perhaps Dretske is assuming that these routes are not available to me, because they are no more ‘­secure’ than looking ­is – where ​­ ‘­secure’ probably also means that I already need to have a way of phenomenologically distinguishing those cases from their ‘­fake’ counterparts. But this seems ­question-​­begging, if one is not already convinced of the conclusion of the Indistinguishability Argument. Second, it is entirely unclear whether what Dretske (­and the vast majority of epistemologists) thinks we n ­ eed – ​­subjective phenomenological ­discriminability – ​­would in fact help us, even if it were available. For if it wouldn’t, then we should stop thinking that not having subjective discriminability is, by itself, sufficient to make the Reasons Identity Thesis and the conclusion of the Indistinguishability Argument compulsory.12 So, let’s suppose, for the sake of argument, that an experiential marker were available that would enable us subjectively to distinguish a real cookie from a deception. To this end, let’s imagine that ­non-​­veridical perceptual experiences always come along with a pinkish tinge, so that whenever one appears to perceive something that is out of kilter with the way the world is, one’s experiences have this hue. For example, when I seem to see a cookie in a jar which is actually a hologram, I will experience the scene in front of me as if I were looking at it through ­rose-​­tinted glasses. This would alert me to the fact that the hologram only looks like a cookie, but is not in fact a cookie. Would this be a useful result? Contrary to what one might expect, it seems not, for the following reasons. If we want the marker in question to alert us to every occasion of something’s appearing other than how it actually is, then virtually all our perceptual experiences will end up coming with a ­rose-​­coloured hue. One might call this the problem of ‘­pinkish encroachment’. The reason for this is the fact that we have a unique point of view on the world, and every time we change our vantage point, things will look different to us. Large towers look small every time we move away from them; square

Knowledge, Closure, and Subjective Distinguishability  139 towers look round the greater the distance from which we view them. In order for everything not to appear pink to us for most of the time, therefore, the ­rose-​­tinted marker would, at the very least, have to be able to track how things seem to us under standard conditions, and only light up when we are straying from those. But even with such a mechanism built in, we would end up having a lot of ­rose-​­tinted experiences. For example, the marker would have to be ‘­on’ whenever we looked at something further away in our visual field, or in motion and so on. Furthermore, it would have to be able to ‘­guess’ which object in our environment we are focussing on, in order not to light up all the time, given that some objects in our visual field will look the way they are, whereas others, because they are further away, will not. Since such an ‘­intentional marker’ seems impossible, it is hard to see how any experiential marker could be ­fi ne-​ ­grained enough to convey useful information.13 What is more, without a distinction between how things look and how they really ­are – ​­the apparent ‘­gap’ exploited by the sceptically minded that the marker is supposed to ­close – ​­a creature with a point of view on a world could not learn anything about this world. An object, for instance, ought to look small from ­afar – ​­if it stayed large even though we were moving away from it, something would be seriously wrong. For it would imply that the object is growing as we are moving away, that our ­eye-​­sight is failing, that we are hallucinating, that distance is having an uncanny effect on objects, that my point of view in space can affect the nature of an object (­not just its appearance), etc. In other words, there are ways an object ought to look from a given perspective, despite that look’s not necessarily being indicative of what the object is really like. Since this is an ordinary feature of perception, it would be useless, and a case of more pinkish encroachment, if the experiential marker came on every time things looked just how they ought to look from one’s perspective. In order to get round these problems, it might, perhaps, seem a good idea to distinguish between ‘­local’ and ‘­global’ experiential markers. ‘­Local’ experiential markers that are supposed to inform us of deceptive appearances in our immediate environment don’t, as we have just seen, appear capable of serving the purpose they were originally drafted in for, as, due to pinkish encroachment, they are incapable of conveying much useful information. The marker might, on occasion, be able to pick up on a genuine perceptual ­illusion – ​­e.g. a hallucination induced by a ­drug – but ​­ since this marker would have come on many times before when nothing was amiss except that we had changed position and perspective, we would not be able to make much use of this information (­for we wouldn’t be able to distinguish this case from all the other cases where the marker had been present). But perhaps a ‘­global’ experiential marker that came on only if we were in radical or global sceptical scenarios, would be more promising.

140  Genia Schönbaumsfeld Imagine, for example, that in a B ­ IV-​­world all of our apparent perceptual experiences come with a pinkish hue, since none of them are veridical. Would this in fact help the BIV? Only if the BIV could come to know that experiential ­rose-​­tintedness indicates ­non-​­veridicality, and it is hard to see how the BIV could come to know this. For if the BIV were a BIV from birth and all its perceptual experiences were ­rose-​­tinted from the off, then it would regard r­ ose-​­tintedness as just a normal feature of perceptual experience (­as it would never have had any ­non-​­pink perceptual experiences). If, on the other hand, the BIV had previously been a normal person and woke up to find itself having ­rose-​­coloured perceptual experiences, then it would probably think something was wrong with its perceptual faculties, not entertain the idea that it had surreptitiously been envatted. Conversely, if the BIV suddenly became disenvatted and started having ­non-​­pinkish perceptual experiences, then it would similarly think that something had happened to its perceptual capacities, not that it had been released, all of a sudden, into the real world. So, whichever way we look at ­it – whether ​­ from a ‘­local’ or a ‘­global’ ­perspective –​ ­experiential markers don’t seem able to convey much information.14 If this is right, then it looks like the presence of experiential ­markers –​ ­the only conceivable way of fleshing out the notion of subjective introspective distinguishability (­discriminability on the basis of phenomenal content alone)  – ​­would not advance our case against scepticism, since these markers (­if such could be had) would not be able to ‘­tell’ us what we wanted them to. Consequently, we need to be very cautious about agreeing that we cannot know that the cookie in the jar is a real cookie, unless such markers (­ and subjective introspective distinguishability) were available, since it now turns out that such markers would be entirely useless to the task at hand.15

4 Implications for Dretske’s Argument and an Alternative Conception We have seen that the main thrust of Dretske’s case against closure is centred around the thought that seeing that there are cookies in the jar is not a reason to believe that there is a physical reality outside of one’s mind (­that there really are cookies in the jar), since, if one were a BIV, things would look exactly the same as if one weren’t: ‘­If your reasons for believing P are such that you might have them when P is false, then they aren’t good enough to know that P is true’ (­Dretske 2005b: 44). This may well be true, but the question is why one should accept the view that your reasons for believing ­P – ​­say, seeing that there are cookies in the j­ar – ​­are reasons that you would have even when P is false. For, given the foregoing considerations, this is not as obvious as Dretske seems to think, since one might believe instead that in cases where P is false, it would only appear as if one had the reasons to believe that P that one has

Knowledge, Closure, and Subjective Distinguishability  141 in the good case, not that, in actual fact, one has those reasons (­whether this misleading appearance means that one lacks any reason at all to believe that P is a further question, but not one that needs to be adjudicated here). That is to say, one might think instead that in cases where P is true, we have reason to believe that P, while in cases where P is false, we do not have reason to believe that ­P – it ​­ only seems that we do.16 Why does Dretske not consider this possibility? Again, because he is implicitly relying on the argument from subjective indistinguishability: unless I can ‘­tell’ merely by ‘­inspecting’ or ‘­consulting’ my immediate ­cookie-​­experience that it is in fact an experience of a real cookie and not, say, of a hologram, I cannot know that I am experiencing an actual (­physical) cookie. But, as we have already seen, this conclusion doesn’t follow unless one is already convinced of the Reasons Identity Thesis. Since one key argument for accepting this ­thesis – the ​­ absence, in perception, of experiential m ­ arkers – ​­has just collapsed (­see previous section), there is nothing that forces us to adopt this conception. Consequently, it remains a live option to think that because the good case is better than the bad case, it also furnishes one with much better reasons than its counterpart: in the good case, one’s ­cookie-​­experiences can be factive (­i.e. entail the presence of an actual physical object), while, in the bad case, this is not so. To perceive that there is a cookie in the jar is not, therefore, to perceive a something which leaves it open whether I am perceiving a cookie or am merely hallucinating it. For if I’m ‘­perceiving’ a ‘­mental cookie’, I’m not, of course, perceiving anything (­and certainly not a ‘­t ype of cookie’17). At best, I’m being appeared to in a c­ ookie-​­like manner. To think otherwise, is just to adopt what McDowell (­1998a, 1998b) calls the Highest Common Factor conception and I call the Reasons Identity Thesis: the notion that all that perception can give us is ­appearance-​­based reasons that are compatible with those appearances being massively deceptive. Since there is no good reason to accept this conception, however, we also don’t need to endorse Dretske’s claim that we can never ‘­perceive heavyweight implications to be so’. Rather, we could reason as follows. If I am in the good case, and I can come to know (­in virtue of my factive ­cookie-​­experience) that I am seeing a cookie in the jar, then I could also come to ­know – ​­in virtue of my factive ­cookie-­​­­experience – ​­that I am not seeing a ‘­­hologram-​­cookie’ in the jar, as seeing a ‘­­hologram-​­cookie’ is to be taken in by a simulation, not to perceive (­see) a sweet physical object.18 But coming to know that I am not seeing a ‘­­hologram-​­cookie’ is not, pace Dretske, an ‘­extra’, as it were ‘­heavyweight’ experience that I might have ‘­on top of’ seeing that there is a cookie there. Rather, if I am in the good case, and RIT is rejected, then seeing that there is a cookie there is already, all by itself, a suitably ‘­heavyweight’ proposition. That is to say, if my ­cookie-​­experience is factive, then it is already an experience of a physical cookie, and not, pace Dretske, an experience of something

142  Genia Schönbaumsfeld that leaves it open whether it is the experience of a physical cookie or of a ‘­­hologram-​­cookie’ (­since, on such a conception, no ­cookie-​­experience could ever be f­active  – that ​­ is to say, entail the presence of an actual physical cookie, not the mere appearance of one). Of course, Dretske is right that one cannot ‘­perceive’ that one is not perceiving a ‘­­hologram-​­cookie’, since all one can perceive is a cookie. But unless one has already ruled in advance that ‘­­cookie-​­perception’ can never give one knowledge of physical cookies, perceiving a cookie can, in the good case, be sufficient to give one knowledge of the presence of physical cookies even if one can’t subjectively distinguish between physical cookies and ‘­­hologram-​­cookies’.19 Consequently, one doesn’t, pace Dretske, need an additional reason to believe the ‘­heavyweight’ implication that one isn’t perceiving a ‘­­hologram-​­cookie’. In the good case, I can know ­non-​­inferentially that there is a cookie before me; I do not need to infer its existence from experiences that are neutral between being either experiences of actual cookies or ‘­­hologram-​­cookies’. What is more, Dretske’s argument to the contrary seems to be relying on the principle that if one perceives that P, and one knows that P entails Q, one cannot know that Q if Q is not itself something one can perceive. But that principle is false. If I know that Pierre murdered Maria, because I saw him kill her, I also know that Felipe, a friend of Maria’s, was not the murderer, although I cannot (­do not) ‘­perceive’ that Felipe was not the murderer (­since I have no perceptions about Felipe). Hence, although knowledge of Felipe’s not being the murderer is entailed by my having perceived that Pierre is the murderer, my knowledge of Felipe’s not being the murderer does not depend on my having ‘­perceived’ that Felipe is not the murderer. Similarly, in order to know that I’m seeing a real cookie when I am, I don’t need to be able to do two different things: to perceive a cookie and, on top of that, to perceive that a cookie is a physical object (­whatever that could so much as mean). So, if I am in the good case, no further evidence is required to license the move from ‘­I see that there is a cookie in the jar’ to ‘­I see that there is a physical cookie in front of me’, although in the corresponding bad case, additional evidence may be necessary (­e.g. if I find myself in ‘­­cookie-​ ­façade’ land). Therefore, contra Dretske, we have no reason to believe that there is a general failure of evidential transmission to what he calls ‘­heavyweight’ implications (­only that this may sometimes be the case). Finally, were we to grant that seeing that there is a cookie in the jar is never a reason to believe that we are seeing actual physical cookies in the jar (­since this is an ‘­unperceivable’, ‘­heavyweight’ implication), then it’s unclear why we should nevertheless regard this as a good reason to believe that there are cookies in the jar (­as opposed to anything, or nothing). An appeal to ‘­relevant alternatives’ is not going to head off this objection, as one doesn’t know what something is just because one can distinguish it from salient alternatives. 20

Knowledge, Closure, and Subjective Distinguishability  143 At this point, one might try a last defence measure: to construe present-​­ day contrastivism Dretske’s view as an early articulation of ­ (­Schaffer 2005). In other words, perhaps we should regard Dretske as saying that knowledge is ‘­contrastive’  – ​­that is, that there really is no such thing as knowing that there is a cookie simpliciter, for example, as knowing that P is in fact a ­three-​­place relation concerning a subject, a proposition, and a contrast class. Consequently, one can only know that there is a cookie in the jar relative to the relevant contrast ­class – ​­say, that there is a ­cookie-­​­­rather-­​­­than-­​­­a-​­pretzel in the ­jar – ​­and not that there is a cookie in the jar tout court. But to adopt such a strategy reaps few rewards. For, on the contrastivist’s proposal too, one will come out as not knowing very much. Hence, there doesn’t seem to be much ­pay-​­off for espousing such a revisionary stance about knowledge. To see this, let’s take a closer look at the contrastivist’s conception. According to Schaffer, contrastivism enables one to know that one has hands rather than stumps, but not to know that one has hands rather than ­vat-​­hands (­Schaffer 2005: 262): ‘­Since the existence of possibilities outside one’s discriminatory range does not imply the absence of any possibilities inside that range, sceptical doubts do not imply the absence of any ordinary knowledge’ (­Schaffer 2005: 263). Do they not, though? For what are we to make of this ‘­ordinary knowledge’? As we already saw at the beginning, ‘­ordinary’ people believe that ordinary cookies are physical cookies, and not simulations or hallucinations. Does Schaffer’s account take this on board any more than Dretske’s does? No. For if I cannot know that the cookie before me is a cookie and not a ­vat-​­cookie, as this lies outside my discriminatory range, then neither can I know that the cookie before me is a real (­i.e. physical) cookie, as opposed to the mere appearance of one. Consequently, just as on Dretske’s conception, whatever ‘­ordinary knowledge’ I can possess (­knowledge of ‘­­cookies-­​­­rather-­​­­than-​­pretzels’; ‘­­hands-­​­­rather-­​­­than-​­stumps’) will at best be knowledge of appearances only (­i.e. knowledge that never rules out the contrast class of deceptions or hallucinations). But, if so, we are back to square one and the sceptic (­or our sceptical alter ego) has the last laugh.

5 Conclusion In summary, in this chapter, I have argued for four main claims: 1 Acceptance of the Reasons Identity Thesis, which is widely endorsed in contemporary epistemology, is already to make substantial concessions to scepticism that are unwarranted. 2 Implicit endorsement of this thesis leads Dretske to deny ­closure – ​­a move that is a red herring and that won’t, in fact, preserve us from scepticism.

144  Genia Schönbaumsfeld 3 The absence of subjective introspective discriminability is not a good reason to adopt the Reasons Identity Thesis, as experiential markers, if, per impossibile, they were available, are not a means to avoiding scepticism. 4 An alternative, less revisionary, conception of perceptual reasons is available that avoids the problems generated by RIT and that does not sell out to scepticism from the outset. Of course, what I have done in this chapter constitutes only a first step towards a full ­a nti-​­sceptical strategy. 21 Nevertheless, I hope to have persuaded the reader that we do not, right from the start, need to concede that we are ‘­walled off’ from the facts and that knowledge of ‘­heavyweight’ propositions is in principle impossible. If this is right, we don’t need to settle for the ‘­spindly brown grass’ that Dretske speaks of: without the Reasons Identity Thesis throwing a spanner into the works from the very beginning, verdant regeneration can beckon still. 22

Notes 1 This seems true of fallibilism, contextualism, contrastivism, and accounts that deny closure. Without the sceptical problem as a motivating background force, many of these conceptions would lose their appeal (­i.e. would not strike us as independently plausible). 2 The closure principle states that ‘­If S knows that P, and S competently deduces Q from P thereby coming to believe that Q on this basis while retaining her knowledge that P, then S knows that Q’ (­I am here following Hawthorne 2005 and Pritchard 2012). 3 Or, to put it in 1970s speak: the epistemic operator ‘­to know’ does not penetrate to all the known logical consequences of what is known (­see Dretske 1970). 4 Compare Nozick (­1981: 242): ‘­I f our notion of knowledge was as strong as we naturally tend to think (­namely, closed under known entailment) then the sceptic would be right’. 5 Crispin Wright (­2002, 2014) has a similar conception: unless we presuppose that there is an external world, we cannot know any perceptual claim to be true. Wright puts this in terms of his ­I –­​­­II–​­III schema: I. My current experience is in all respects as if P. II. P III. There is an external world. With this schema in place, one can only know (­I I) if one already knows (­I II). See The Illusion of Doubt, chapter 2, for further discussion. 6 Conjunctions of the sort ‘­I know I have a hand, but I don’t know I’m not a handless BIV’ (­see DeRose 1995). DeRose makes this point against Nozick. 7 This is a point that Dretske should concede, given that he says that a decoy duck is not a duck (­Dretske 1981: 57)! 8 I take this to be an epistemological conclusion, not a metaphysical one, although it does have metaphysical ­implications – ​­namely the essentially inferential nature of perceptual knowledge. For more on this, see The Illusion of Doubt, chapter 3.

Knowledge, Closure, and Subjective Distinguishability  145 9 See, for example, Avnur (­2019), Burge (­2003), Conee (­2007), Coliva (­2012, 2014, 2015), McGinn (­1984), Millar (­1991), Nagel (­1986), Pollock (­1974), Pryor (­2000), Stroud (­1984), White (­2014), Wright (­2002, 2014). 10 Also see Wright (­2002, 2014). 11 I first argued this in chapter 1 of The Illusion of Doubt. 12 For more on this, see The Illusion of Doubt, chapter 1, as well as Pritchard (­2012). 13 If, on the other hand, one tried to ‘­correct’ the marker for every change brought on by changing perspective (­e.g. by moving), then it would virtually never be on. In other words, it seems that experiential markers would either have to be on all the time or never, and neither option is compatible with information transmission. 14 For more on this, see Schönbaumsfeld (­2021). 15 Now perhaps some would regard this as a way of compounding scepticism: If experiential markers cannot give one perceptual knowledge, nothing can. But that would be irrational. If experiential markers are not a means to the end they were envisaged for, then one ought to consider an alternative means to this end, rather than come to the conclusion that they are the o ­ nly – ​­albeit logically ­impossible – game ​­ in town. Since an alternative conception is available (­see the next section), there is no reason to throw in the towel at this point. 16 So, in such cases one might be blamelessly taking oneself to be in possession of a reason, even though one is not. But blamelessness alone does not suffice to turn an appearance (­of a reason) into reality. For good further discussion of this point, see Pritchard (­2012: ­42–​­5). 17 For ‘­hallucinated cookies’ are not tokens of the type ‘­cookie’. Dretske sometimes speaks as if they were. But if he literally meant that, the view would be absurd. 18 I am here leaving it open whether we can know that we are in this privileged situation, I am merely arguing that this is a possible alternative to Dretske’s conception and that he has not done enough to rule it out. 19 Compare Klein (­2004), who makes a similar point about zebras and cleverly disguised mules; see also Pritchard (­2012), parts II and III. 20 Let’s say I come across an object that I haven’t encountered before, which looks like a tool or piece of equipment. I clearly won’t find out what it is just because I know it’s not one of the relevant alternatives: a hammer, a ­chain-​ s­ aw, a pair of pliers etc. 21 See The Illusion of Doubt for the full picture. 22 I would like to thank Duncan Pritchard for helpful comments on an earlier draft.

References Avnur, Y. (­2019), ‘­T he Nature and Limits of Skeptical Criticism’, International Journal for the Study of Skepticism 9/­3, ­183–​­205. Burge, T. (­2003), ‘­Perceptual Entitlement’, Philosophy and Phenomenological Research 67, ­503–​­48. Coliva, A. (­2012), ‘­Moore’s Proof, Liberals and C ­ onservatives  – ​­Is There a (­Wittgensteinian) Third Way’, in A. Coliva (­ed.), Mind, Meaning and Knowledge (­Oxford: Oxford University Press), ­323–​­51. Coliva, A. (­2014), ‘­Moderatism, Transmission Failures, Closure and Humean Scepticism’, in D. Dodd and E. Zardini (­eds.), op. cit., ­248–​­76. Coliva, A. (­2015), Extended Rationality: A Hinge Epistemology (­London: Palgrave Macmillan).

146  Genia Schönbaumsfeld Conee, E. (­2007), ‘­Disjunctivism and ­A nti-​­Skepticism’, Philosophical Issues 17, ­16–​­36. DeRose, K. (­1995), ‘­Solving the Sceptical Problem’, Philosophical Review 104: 1, ­1–​­52. Dretske, F. (­1970), ‘­Epistemic Operators’, Journal of Philosophy 60, ­1007–​­23. Dretske, F. (­1981), Knowledge and the Flow of Information (­Boston: MIT Press). Dretske, F. (­2005a), ‘­The Case against Closure’, in Ernest Sosa and Matthias Steup (­eds.), Contemporary Debates in Epistemology (­Oxford: Blackwell), ­13–​­25. Dretske, F. (­2005b), ‘­Reply to Hawthorne’, in Ernest Sosa and Matthias Steup (­eds.), op. cit., ­43–​­6. Dretske, F. (­2010), ‘­Epistemological ­Self-​­Profile’, in Jonathan Dancy, Ernest Sosa and Matthias Steup (­eds.), A Companion to Epistemology (­Oxford: Blackwell), ­130–​­3. Hawthorne, J. (­2005), ‘­T he Case for Closure’, in Ernest Sosa and Mathias Steup (­eds.), op. cit., ­26–​­42. Klein, P. (­2004), ‘­Closure Matters: Academic Scepticism and Easy Knowledge’, Philosophical Issues (­Epistemology) 14: ­165–​­84. McDowell, J. (­1998a), ‘­Criteria, Defeasibility and Knowledge’, in J. McDowell (­eds.), Meaning, Knowledge and Reality (­Cambridge, MA: Harvard University Press), ­369–​­94. McDowell, J. (­1998b), ‘­K nowledge and the Internal’, in op. cit., ­395–​­413. McGinn, C. (­1984), ‘­T he Concept of Knowledge’, Midwest Studies in Philosophy IX, ­529–​­544. Millar, A. (­1991), Reasons and Experience (­Oxford: Clarendon Press). Nagel, T. (­1986), The View from Nowhere (­Oxford: Oxford University Press). Nozick, R. (­1981), Philosophical Explanations (­Oxford: Oxford University Press). Pollock, J. (­ 1974), Knowledge and Justification (­ Princeton, NJ: Princeton ­University Press). Pritchard, D. (­2012), Epistemological Disjunctivism (­Oxford: Oxford University Press). Pryor, J. (­2000), ‘­T he Skeptic and the Dogmatist’, Noûs 43, ­517–​­49. 2005), ‘­ Contrastive Knowledge’, in Tamar Szabo Gendler and Schaffer, J. (­ John Hawthorne (­eds.), Oxford Studies in Epistemology Volume 1 (­Oxford: Oxford University Press), ­235–​­71. Schönbaumsfeld, G. (­2016), The Illusion of Doubt (­Oxford: Oxford University Press). Schönbaumsfeld, G. (­2021), ‘­Subjective ­Distinguishability – ​­Who Needs it?’ in ­ 41–​­57. Midwest Studies in Philosophy 45, 2 Stroud, B. (­ 1984), The Significance of Philosophical Scepticism (­Oxford: Oxford University Press). White, R. (­2014), ‘­W hat Is My Evidence That Here Is a Hand?’, in D. Dodd and E. Zardini (­eds.), Scepticism and Perceptual Justification (­Oxford: Oxford University Press), ­298–​­321. Wright, C. (­2002), ‘­­Anti-​­Sceptics Simple and Subtle: G. E. Moore and John McDowell’, Philosophy and Phenomenological Research 65, ­330–​­48. Wright, C. (­2014), ‘­On Epistemic Entitlement (­I I): Welfare State Epistemology’, in D. Dodd and E. Zardini (­eds.), op. cit., 2 ­ 13–​­47.

9 Underdetermination and Closure Thoughts on Two Sceptical Arguments Martin Smith 1 Introduction Let P be some quotidian proposition that we would ordinarily take ourselves to ­know  – say ​­ the proposition that I am currently seated at my desk. Here is an uninteresting sceptical argument to the effect that I don’t know P: (­S1) If I know P then I must have rational support that favours P over any proposition that is inconsistent with P. (­S2) The only way that one can have rational support that favours a proposition X over any proposition that is inconsistent with X is if one’s current subjective experiences entail that X is true. (­S3) My current subjective experiences do not entail that P is true. (­S 4) Therefore, I don’t know P (­S1, S2, S3). Call this the simple sceptical argument. What makes the argument uninteresting is, of course, S­ 2 – ​­a premise that obviously stacks the deck in favour of scepticism, and that no ­non-​­sceptic should be willing to accept. A couple of points of clarification regarding this premise: Experiences don’t, strictly speaking, stand in any entailment relations. When we say that a proposition X is entailed by my subjective experiences what is meant is that X is entailed by a proposition E which provides a comprehensive description of those e­ xperiences  – the ​­ proposition that it currently seems to me that I am seated at a desk, staring at an illuminated laptop screen, listening to the hum of distant traffic and so on… One’s subjective experiences are meant to include things like perceptual seemings and apparent m ­ emories – ​­and perhaps we could also add (­occurrent) beliefs, desires, intentions, etc. We could even let one’s current subjective experiences include the totality of one’s current ­non-​­factive mental states, if one finds this a better category to work with. In any case, it’s obvious that my subjective experiences, so characterised, are not going to entail anything about the external world, other minds, the past, the future, etc. (­modulo certain externalist constraints on content1). If S2 holds, then the only propositions for which I could

DOI: 10.4324/9781003104766-10

148  Martin Smith have favouring rational support are going to be propositions that directly describe my current subjective experiences. What this argument seeks to do, in effect, is derive the implausible claim that we cannot know any propositions about the external world, other minds, the past or the future from the even more implausible claim that we can’t have rational support that favours any propositions about the external world, other minds, the past or the future over their alternatives. It may be that there are further, supplementary arguments that could be given for ­S2 –​ ­but, if so, then this is where the interest would lie. If no further argument is o ­ ffered – ​­if the sceptic just appeals to S2 as a bare ­premise – then ​­ this clearly begs the question against a n ­ on-​­sceptic. By pinpointing S2, I don’t mean to suggest that the other premises in this argument are necessarily beyond challenge. While S3 does seem obvious, given the way ‘­subjective experience’ has been characterised, there may be certain theories of knowledge which would predict that S1 is false. 2 In any case, whatever reasons there may be to reject S1, its role in this argument is certainly not amongst them. Given the ­question-​ b ­ egging nature of S2, the argument puts no pressure on S1.

2 The ­Underdetermination-​­Based Sceptical Argument Most epistemologists would, I imagine, regard the simple sceptical argument to be a rather poor example of the ­genre  – and ​­ would insist that there are plenty of sceptical arguments that are more subtle and interesting than this. Indeed, one common argumentative technique in epistemology is to defend a substantial view (­of knowledge, justification, evidence…) on the grounds that it is needed to answer some sceptical argument or other. I will be concerned here with two sorts of sceptical argument that have been particularly prominent in recent ­epistemology  – t​­ he ­underdetermination-​­based sceptical argument and the ­closure-​­based sceptical ­argument – both ​­ of which have been used to prop up substantial philosophical views. ­Underdetermination-​­based sceptical arguments have a long history, with Descartes’ sceptical argument in the Meditations often interpreted as an argument of this kind (­Yalçin, 1992). The ­closure-​­based sceptical argument is also, on occasion, associated with Descartes (­see Stroud, 1984, cha­p.  1) but, in its familiar modern form, stems from the work of Dretske (­1970) and Nozick (­1981, cha­p. 3, section II). The contemporary distinction between these two forms of argument is a product of a literature on the logical structure of sceptical arguments from the 1990s and 2000s (­Brueckner, 1994; Cohen, 1998; Byrne, 2004; Vogel, 2004; Pritchard, 2005) (­the present chapter owes much to this literature). The distinction plays a pivotal role in Pritchard’s recent Epistemic Angst (­2015), and the terminology that I will use in formulating these

Underdetermination and Closure   149 arguments is close to his (­though little, I think, hinges on the particular terminology that we choose, and I will note some variants along the way). Let BIV be the proposition that I am a disembodied brain floating in a vat of nutrients wired up to a supercomputer that monitors electrical output and supplies electrical input, stimulating full and rich perceptual experiences, perhaps like those I am currently undergoing… Let P, once again, be the (­incompatible) proposition that I am currently seated at my desk. The ­underdetermination-​­based sceptical argument, at a first pass, runs as follows: (­U1) I don’t have rational support that favours P over BIV. (­U2) If I know P then I must have rational support that favours P over BIV. (­U3) Therefore, I don’t know P (­U1, U2). In his initial discussion of this argument (­in chapter 2 of Epistemic Angst), Pritchard’s focus is largely upon U2. As well as offering a more detailed defence of the premise, he considers ways in which it might be modified in order to resist certain attempts to refute it.3 His initial remarks on U1, however, are rather brief, and worth quoting in full4: This claim seems undeniable. For given that the experiences had by the subject in the BIV case are subjectively indistinguishable from everyday experiences, then how is one to come by rational support for an everyday perceptual belief that epistemically favors this belief over an incompatible radical skeptical alternative? (­Pritchard, 2015, p30) Although Pritchard does reject U1 in the end (­a point I will come to), I want to look more closely at this short, but suggestive, rationale for the premise. What Pritchard seems to be proposing is that because P and BIV are both consistent with my subjective experiences, it follows that I can’t have rational support that favours the former over the latter. This suggests the following general principle, from which U1 might be derived: (­U0) For any two inconsistent propositions X and Y, if Y is consistent with one’s current subjective experiences then one doesn’t have rational support that favours X over Y. Even the name ‘­­ underdetermination-​­ based sceptical argument’ very much suggests this kind of motivation for U ­ 1 – ​­suggests that the reason I lack rational support favouring P over BIV is that the truth of the matter is ‘­underdetermined’ by my evidence (­construed subjectively).

150  Martin Smith But if U1 is really supposed to be supported by U0 then this exposes a serious problem in the argument. If I have rational support that favours X over an inconsistent proposition Y then, according to U0, Y must be inconsistent with my subjective experiences. As a result, if I have rational support that favours X over every proposition that is inconsistent with X then, according to U0, every proposition that is inconsistent with X would also have to be inconsistent with my subjective experiences, which is just to say that my subjective experiences would have to entail X. Though it may not look it at first, U0 is actually logically stronger than ­S2  – ​­the q ­ uestion-​­begging premise from the simple sceptical argument. This point is made quite clearly by Brueckner (­1994, ­pp834–​­835) –​ ­though he uses a different terminology, speaking of ‘­sensory evidence’ rather than ‘­subjective experiences’ or ‘­rational support’. As Brueckner points out, if the sceptic is assuming that one’s sensory evidence cannot favour one proposition X over an incompatible proposition Y in the event that it is consistent with both, then this is tantamount to assuming that one’s sensory evidence can only favour those propositions that it entails. To unearth this assumption in a sceptical argument is, as Brueckner puts it, an ‘­embarrassment’ for the sceptic. Any ­non-​­sceptic is committed to rejecting U0. In addition, it’s worth pointing out that U0 is subject to a number of compelling counterexamples. Suppose I throw a fair s­ ix-​­sided die and it lands just out of view. It seems obvious that I have rational support that favours the proposition that the die landed on a number between 1 and 5 over the proposition that the die landed 6, despite the fact that both propositions are consistent with my subjective experiences. Suppose someone tells me that it’s currently raining outside, and I have no reason to doubt their word. Once again, it seems obvious that I have rational support that favours the proposition that it’s raining over the proposition that it isn’t, even though these propositions are both consistent with my subjective experiences. 5 When expanded using U0, the ­underdetermination-​­based sceptical argument is every bit as uninteresting as the simple sceptical argument from Section 1. Perhaps we shouldn’t be too hasty in dismissing the argument, ­though – after ​­ all, there may be a better way to bolster U1. One thing that we might observe is that the BIV hypothesis is not only consistent with my current subjective experiences, it actually entails those ­experiences – ​­or could, at any rate, be easily formulated in such a way that it does. That is, we could spell out BIV as the proposition that I am a disembodied brain floating in a vat of nutrients supplied with electrical signals… and it currently seems to me that I am seated at a desk, staring at an illuminated laptop screen, listening to the hum of distant traffic… So, rather than using U0 to support U1, we could appeal to the following principle instead:

Underdetermination and Closure   151 (­U0*) For any two inconsistent propositions X and Y, if Y entails one’s subjective experiences then one doesn’t have rational support that favours X over Y. Something like this strategy is suggested by Cohen (­1998, p ­ p146–​­147), and may offer a more charitable interpretation of Pritchard’s remark as well (­after all, he does say ‘…the experiences had by the subject in the BIV case are subjectively indistinguishable from everyday experiences’). Ultimately, though, this second rationale for U1 is no more successful than the fi ­ rst – although ​­ U0* is logically weaker than U0, it still commits us to the q ­ uestion-​­begging S2. To see this, let E once again be a proposition describing the totality of my current subjective experiences. If there is a proposition Y that is inconsistent with X and consistent with E, then there is a further ­proposition – ​­namely Y ∧ E ­  – ​­that is inconsistent with X and which entails E. If I have rational support that favours X over every inconsistent proposition then, according to U0*, there are no propositions that are inconsistent with X and which entail E. By the above reasoning, it follows that there are no propositions that are inconsistent with X and consistent with E, which is just to say that E entails X. Substituting U0* for U0 fails, then, to inject any more interest into the ­underdetermination-​­based sceptical argument. It’s also worth noting that the counterexamples I levelled at U0 can easily be adapted as counterexamples to U0*6: If I throw a fair, s­ ix-​­sided die that lands just out of view, it seems obvious that I have rational support that favours the proposition that I’m currently having an experience as of just having thrown a die, etc. and the die landed on a number between 1 and 5, over the proposition that I’m currently having an experience as of just having thrown a die, etc. and the die landed 6. If someone tells me that it’s currently raining outside and I have no reason to doubt their word, it seems obvious that I have rational support that favours the proposition that I’m currently having an experience as of being just told that it’s raining, etc. and it is raining, over the proposition that I’m currently having an experience as of being just told that it’s raining, etc. and it isn’t raining. Though it may have an initial appeal, upon close scrutiny U0* is no more plausible a principle than U0. We have now looked at two strategies to support ­U1 – ​­using the principles U0 and U0* – and ​­ argued that they both beg the question against a ­non-​­sceptic. But what if the sceptic declines to offer any further support for U ­ 1 – ​­and just puts it forward in the hope that we’ll accept it as is? In one way, this looks like desperation on the sceptic’s p ­ art – ​­and could hardly serve to imbue the argument with greater interest. And yet, when putting forward an argument, a sceptic, like anyone else, has to rest content with some premises that are not derived from other things. So why not U1? Why should this need to be deduced from a general principle?

152  Martin Smith U1 states that our evidence fails to favour either one of two incompatible ­hypotheses – ​­P and BIV. When we encounter rival hypotheses in ­day-­​­­to-​­day l­ife – two ​­ people involved in a car accident give competing accounts of what happened, two websites give competing weather forecasts, etc. – we ​­ can of course judge that our evidence favours neither one, and we can do this without necessarily committing to any general principles about evidential favouring. But the two propositions that feature in U1 are not like this; the BIV hypothesis is artificially designed, and it is designed to have one very particular ­characteristic – ​­namely, to be consistent with all of our subjective experiences. In fact, in one important sense, all that there is to the BIV hypothesis is that it is inconsistent with many of our ordinary beliefs while being consistent with our subjective e­ xperiences –​ ­the rest is just filler that can vary from presentation to presentation.7 If U1 is simply offered as a bare premise, then there is no reason to accept it unless one is also prepared to accept the general claim captured by U0. Unless the sceptic is able to devise some better way of supporting U1, we are forced to conclude that the u ­ nderdetermination-​­based sceptical argument is, in the end, no more philosophically interesting than the simple sceptical argument. In fact, the two arguments turn out to rest upon the very same q ­ uestion-​­begging a­ ssumption  – it ​­ is just that the ­underdetermination-​­based argument succeeds in burying it deeper.

3 An Aside: Epistemological Disjunctivism As I mentioned in the previous section, the ­underdetermination-​­based sceptical argument plays a key role in Pritchard’s Epistemic Angst, where he uses it to argue his case for epistemological ­disjunctivism – ​­the view that, in normal cases of perceptual knowledge, one’s perceptual belief enjoys a kind of rational support that is both factive and reflectively accessible (­Pritchard, 2012, part 1, cha­p. 1, 2015, cha­p. 5). According to the disjunctivist, when I know, on the basis of perception, that I’m currently seated at my desk, the rational support for my belief is provided by the fact that I see (­and feel) that I’m seated at my d ­ esk – ​­which entails that I’m seated at my desk and which is (­supposedly) accessible via reflection. In this case, I do have rational support that favours P over BIV (­indeed rational support that entails P), and U1 is false. If, of course, I was hallucinating or I really was a brain in a vat, etc., then I wouldn’t have this kind of rational support for my perceptual belief, even though my subjective experiences may be exactly the same. But if I’m not in this kind of situation then, according to the disjunctivist, my rational support need not be limited to my subjective experiences, or to the states that I share with my hallucinating or envatted counterpart.8 Whatever one makes of epistemological disjunctivism, to adopt this view as a way of avoiding the u ­ nderdetermination-​­based sceptical argument would, I think, be a considerable overreaction. As suggested in the

Underdetermination and Closure   153 previous section, the most straightforward way for a ­non-​­sceptic to respond to the argument is to point out that one can have rational support that favours one proposition over another even when they are both consistent with one’s subjective ­experiences – ​­and to reject U0 and U1 on this basis. In making this point, we don’t need to appeal to d ­ isjunctivism – or ​­ to any substantial philosophical theory. This is not to say, of course, that epistemological disjunctivism is necessarily mistaken. Pritchard offers some independent considerations in support of the view, which I won’t consider here.9 The point that I would make here is that independent considerations are very much ­needed  – ​­the ­underdetermination-​­based sceptical argument, in and of itself, generates no support for the view.10 Furthermore, even if one does accept disjunctivism (­for independent reasons), one should still, I believe, respond to the ­underdetermination-​ ­based sceptical argument in the straightforward way suggested above. That is, even for an avowed disjunctivist, to invoke disjunctivism in responding to this argument is excessive. Suppose again that I’ve just rolled a fair ­six-​­sided die and it has landed out of view. Suppose a sceptic tells me that I have no rational support favouring the proposition that the die is showing between 1 and 5 over the proposition that the die is showing 6. Even an avowed disjunctivist, such as Pritchard, would presumably wish to resist this suggestion. But disjunctivism has no bearing on this kind of c­ ase – ​­it isn’t a case of putative perceptual knowledge, and I clearly don’t have any rational support that entails the proposition that the die is showing between 1 and 5. If the disjunctivist accepts that I do nevertheless have rational support that favours the ­1–­​­­5-​­proposition over the ­6 -​­proposition, then they already have all that they need to reject U0 and ­U1 – ​­epistemological disjunctivism is surplus to requirements.11 Pritchard does, in fact, explicitly accept that there is such a thing as ­non-​­conclusive ­favouring  – that ​­ it is possible to have rational support that favours a proposition X over a proposition Y without entailing X (­Pritchard, 2015, cha­p. 5, section 4, see also, Pritchard, 2012, part II). He argues, however, that ­non-​­conclusive favouring is simply not available when it comes to P and BIV, and that the only rational support that could favour P over BIV is rational support that entails P (­such as that supplied by disjunctivism). If this argument were effective, then it could be used to bolster U1 in such a way that it really would take a substantial philosophical theory, like disjunctivism, to resist it. Pritchard’s argument begins from the claim that one cannot appeal to any background evidence or beliefs in order to argue against the BIV ­hypothesis – because ​­ the hypothesis calls all of this evidence ‘­into question’ (­Pritchard, 2015, ­pp137–​­138). But what does it mean exactly for a hypothesis like BIV to call a piece of evidence or a background belief ‘­into question’? In explaining this, Pritchard falls back on something close to the idea that all of our subjective experiences are consistent with, or entailed by, BIV. He writes, for instance:

154  Martin Smith Raising the consideration that, for example, one has good reason for thinking that current technology could not support BIVs is clearly inappropriate in the context of the BIV skeptical hypothesis precisely because one would also believe that one had reasons of this sort if one were a BIV. (­Pritchard, 2015, p138) If this is all that is at work in the argument then, clearly, we are still going to need something like U0 or U0* in order to get to the desired conclusion. I won’t discuss this further ­here – ​­and, in any case, I don’t mean to definitively rule out the possibility of there being some kind of ­non-­​­­question-​­begging argument in support of U1. But if such an argument were offered, then this is the argument that would be doing the philosophical ‘­heavy lifting’  – a​­ nd not the ­underdetermination-​­based sceptical argument.

4 The ­Closure-​­Based Sceptical Argument The c­losure-​­ based sceptical argument, like the u ­nderdetermination-​ b ­ased sceptical argument, trades on the possibility of concocting ­hypotheses – ​­like ­BIV – ​­that are consistent with our subjective experiences, but inconsistent with propositions that we would ordinarily take ourselves to ­know  – ​­like P. Sticking with these two propositions, the ­closure-​­based argument runs as follows: (­C1) I cannot know ~BIV. (­C2) If I know P, then I can know ~BIV. (­C3) Therefore, I don’t know P. (­C1, C2) This argument has been widely discussed since the work of Dretske and Nozick and, from the very beginning, the primary focus has been on C2, which is motivated by a closure principle for ­knowledge – a​­ principle to the effect that one can always extend one’s knowledge by drawing competent (­single premise) deductive inferences. While there is some controversy over how to precisely formulate this closure principle, many epistemologists have settled on something like the following: If one knows a proposition X and knows that X entails Y and one competently deduces Y from X, while retaining knowledge of X, then one comes to know Y. (­Hawthorne, 2003, p34; Pritchard, 2015, p13) Since I know that P entails ~BIV and am perfectly capable of deducing the latter from the former, if I know P (­and can make the deduction without losing this knowledge) then, according to the closure principle, I must be in a position to know ~BIV, giving us C2.

Underdetermination and Closure   155 C1, on the other hand, is not usually discussed at length. Some who present the argument simply regard it as ‘­intuitive’ that I can’t know that I’m not a brain in a vat and leave it at that. Those who do say more to motivate the premise generally just labour the familiar point that the BIV proposition is consistent with all of our current subjective experiences. As Pritchard writes: The initial plank in the case for skepticism comes from the contention that one cannot know that one is not a BIV. Such a claim seems entirely compelling. After all, since the BIV scenario is ex hypothesi subjectively indistinguishable from normal perceptual conditions, it is hard to see how one might come to know such a thing. (­Pritchard, 2015, p12) In a similar vein, Nozick writes: You think you are seeing these words but could you not be hallucinating or dreaming or having your brain stimulated to give you the experience of seeing these marks on paper even though no such thing is before you? More extremely, could you not be floating in a tank while s­ uper-​­psychologists stimulate your brain electrochemically to produce exactly the same experiences as you are now having, or even to produce the whole sequence of experiences you have had in your lifetime so far? If one of these other things was happening, your experience would be exactly the same as it now is. So how can you know that none of them is happening? (­Nozick, 1981, p167) For a final, more recent, example, Kelp remarks: You don’t know that you are not a handless BIV. How could you? After all, everything would seem to you exactly as it would were you to be a normal handed person. (­Kelp, 2021, p376) On one interpretation, Pritchard, Nozick and Kelp are appealing to the following principle: (­C 0) For any proposition X, if ~X is consistent with one’s current subjective experiences then one cannot know X. But it follows immediately from C0 that the only way in which one can know a proposition X is if X is entailed by one’s subjective experiences (­see Vogel, 2004, ­pp426–​­427). That is to say, C0 commits us to an especially demanding kind of infallibilism about knowledge, and should be immediately rejected by any ­non-​­sceptic.12

156  Martin Smith As we’ve seen, it’s obvious that my subjective experiences are not going to entail anything about the external world, or other minds, or the past, or the future, etc. If C0 holds, then the only propositions which I could know are going to be those propositions that directly describe my current subjective experiences. C0 allows us, then, to derive a sceptical result directly, without the need for C2 or the closure principle. The idea that the closure principle may turn out to be an idle wheel in motivating scepticism is suggested by Brueckner (­1994) and Byrne (­2004) and defended in detail by Schönbaumsfeld (­2017). Schönbaumsfeld also notes that C1 is usually motivated by citing the ‘­subjective indistinguishability’ of propositions such as P and BIV, and argues that this motivation implies an underlying assumption that would effectively lead to scepticism all by itself.13 When C1 is based upon C0, the ­closure-​­based sceptical argument is devoid of interest. It would be convenient, in a way, if we could simply leave our analysis ­here – ​­and put the ­closure-​­based sceptical argument into the same category as the ­underdetermination-​­based sceptical argument (­and simple sceptical argument). But there are some further complications associated with the ­closure-​­based ­argument – ​­complications which do, in the end, give it more philosophical interest. As observed in Section 2, the BIV hypothesis could be formulated in such a way that it entails, and is not merely consistent with, one’s subjective experiences. This suggests another way of motivating C1: (­C0*) For any proposition X, if ~X entails one’s current subjective experiences then one cannot know X. This represents an ­ alternative  – ​­ and perhaps even more ­ faithful  –​ ­interpretation of Pritchard, Nozick and Kelp’s remarks. C0* is logically weaker than C0 and does not commit one to infallibilism about knowledge. Is a ­non-​­sceptic under pressure to accept it? One thing that we might observe right away is that C0* is somewhat reminiscent of U0* which, as we’ve seen, a ­non-​­sceptic is committed to rejecting. That is, a ­non-​­sceptic must affirm that one can have rational support that favours a proposition X over an inconsistent proposition Y, even if one’s subjective experiences are entailed by Y. From here, it seems a relatively small step to accept that one could know a proposition X even if one’s subjective experiences are entailed by ~X. As discussed in Sections 1 and 2, it’s plausible that rational favouring is one necessary condition for ­knowledge – ​­if one knows a proposition X then one must have rational support that favours X over ~X. But even if ~X entails one’s current subjective experiences, once we have rejected U0*, we should remain open to the idea that this condition could still be met – ​­that it could still be the case that one has rational support that favours X over ~X. As a result, unless there is some other condition on knowledge which is necessarily precluded in this kind of situation, we

Underdetermination and Closure   157 should also be open to the idea that one could still know ­X – ​­and, thus, we should reject C0*. While I think it is a small step for a ­non-​­sceptic to reject C0*, given their prior rejection of U0*, I wouldn’t go so far as to say that the use of this principle to support C1 begs the question against a ­non-​­sceptic. Furthermore, on this reconstruction of the c­ losure-​­based sceptical argument, the closure premise C2 will turn out to play a crucial role in deriving the sceptical ­conclusion – and, ​­ as a result, a ­non-​­sceptic could accept C0*, provided they are willing to reject the closure principle. This particular combination of views is, in fact, characteristic of one version of the theory of relevant alternatives. According to the theory of relevant alternatives, in order to know a proposition X, one must have ruled out all of the relevant alternatives to ­X – that ​­ is, all of the relevant propositions that are inconsistent with X. Relevant alternatives theorists disagree over what exactly is involved in ruling an alternative out, but may take the view that an alternative which entails one’s subjective experiences can never count as ruled out. Relevant alternatives theorists also disagree over what it is for an alternative to be relevant, but may take the view that ~X always counts as a relevant alternative to X. These commitments would deliver us C0* – ​­if ~X entails one’s subjective experiences then one cannot rule it out and, if it is a relevant alternative to X, then one is prevented from knowing X. This view would predict, then, that I cannot know ~BIV, but it need not lead to a general sceptical result. Consider again proposition P (­that I’m currently seated at my desk). Since ~P does not entail my subjective experiences, it is consistent with the present view that I have successfully ruled it out and, provided BIV and any other alternatives that do entail my subjective experiences are deemed irrelevant, it is consistent with my knowing P. It is for this reason, of course, that closure can fail on the theory of relevant alternatives.14 There are yet further ways that one could try to support C1. It seems clear, for instance, that I haven’t conducted any kind of inquiry or investigation into the possibility that I might be a brain in a ­vat – ​­I can’t point to a time at which I acquired justification for believing that this possibility doesn’t obtain, and I can’t offer any reasons that would reassure someone who takes the possibility seriously (­Wright, 2004; Smith, 2013). If we were prepared to endorse a principle like the following, then this would give us another motivation for C1: (­C0**) For any proposition X, if one has not acquired justification for believing X and/­or one can offer no reasons that would potentially convince a doubter, then one cannot know X. Once again, is a ­non-​­sceptic under pressure to accept such a principle? For any proposition X, if ~X entails all of one’s subjective experiences then it seems plausible that one could not legitimately claim to have acquired

158  Martin Smith justification for believing X or to have convincing reasons that one could offer a person who doubts X. As a result, if we deny U0*, then lacking acquired justification for believing X and being unable to offer convincing reasons in favour of X is still compatible with one having rational support that favours X over ~X. But if we have rational support that favours X over ~X, why should it matter whether we acquired our justification, or are able to convince a doubter? As above, unless there is some other condition on knowledge, aside from rational favouring, which is supposed to be blocked in this circumstance, we should be open to the idea that one could still know ­X – ​­and, thus, we should reject C0**.15 While C0* and C0** may not exactly beg the question against a ­non-​ ­sceptic, they should be regarded as contentious assumptions and, absent any other reasons for accepting C1, the premise should be regarded as weak. In spite of this, the argument does carry philosophical lessons. If one is attracted to principles like C0* and C0** (­and wishes to avoid scepticism) then the price of this is to deny C2 and the principle of closure. As discussed, there are theories of k ­ nowledge – like ​­ the theory of relevant ­alternatives – ​­which could offer this package of views. If, on the other hand, one wishes to preserve closure (­and avoid scepticism) then one will have to give up C0* and C0**. This too can serve as an impetus for developing a compliant theory of knowledge.16

5 Conclusion The best sceptical arguments, one might think, expose a deep or hidden tension within our ordinary thinking or orthodox philosophical theorising about knowledge (­Wright, 1984, ­pp429–​­430). Whether there are any sceptical arguments that accomplish this is not something that I take myself to have settled h ­ ere – ​­but the two arguments that I have surveyed do not I think live up to this kind of description. In the end, the primary tension at work in these arguments is the tension between the idea that we know a great deal, and the idea that the standards for knowledge or for rational support are extremely demanding. While that tension is, to some extent, present in our ordinary thinking and philosophical theorising, it is neither hidden nor deep.

Acknowledgements Thanks to Dominik Kauss and to an anonymous referee for helpful comments on earlier drafts of this material.

Notes 1 Such constraints may mean that certain very general propositions about the external world or other minds or the past are entailed by propositions about

Underdetermination and Closure   159 the contents of one’s ­non-​­factive mental states. If the latter propositions are included as facts about one’s ‘­subjective experiences’ then S2 is consistent with the former propositions being known. Whatever the truth about this view, however, it won’t offer any meaningful relief from the scepticism engendered by S2, and I won’t consider it further here. 2 There are, for instance, certain ‘­ externalist’ views on which knowledge doesn’t require any kind of rational ­basis – ​­views on which, say, a belief that is reliably formed, or a belief that appropriately ‘­tracks the truth’ may qualify as knowledge, even if a rational basis for the belief is lacking (­Pritchard, 2015, chapter 1, section 4, chapter 2, section 2). Externalists of this stripe may wish to reject S1. 3 In his discussion of U2, Pritchard is largely concerned to thwart externalists of the kind discussed in n2 who claim that not all knowledge is rationally grounded, and who would seek to dismiss the premise on this basis. As Pritchard points out, we could simply replace the notion of knowledge with the notion of rationally grounded knowledge and run the argument as before. Having made this substitution, the externalist would no longer have any complaint against U2, but would still need to find some response to the argument, or face a w ­ ide-​­ranging scepticism about rationally grounded knowledge. To claim that not all knowledge is rationally grounded is one t­hing – to ​­ accept that knowledge is never, or almost never, rationally grounded would be an externalism too extreme for most (­Pritchard, 2015, chapter 2, section 2). 4 Pritchard notes that this premise would be rejected by an ­abductivist  –​ ­someone who holds, roughly, that ordinary quotidian propositions like P provide a better explanation of our evidence than their sceptical alternatives like BIV. Pritchard offers a series of criticisms of this view (­Pritchard, 2015, chap 1, section 6) but, whatever we make of them, I think his focus on abductivism is somewhat misplaced. I will return to this in n11. ​­ 5 Further counterexamples to U0 are described by Goldman (­2007)  – who also notes that the principle is taken for granted in the ­underdetermination-​ ­ ased sceptical ­argument – ​­and by Markosian (­2014, p170). b 6 This is true also of Goldman and Markosian’s counterexamples to U0 as mentioned in the previous footnote (­see also Markosian, 2014, p171). 7 While any worthwhile sceptical hypothesis will be consistent with one’s subjective experiences, it is only ‘­global’ sceptical hypotheses that are inconsistent with a broad range of ordinary beliefs. ‘­Local’ sceptical hypotheses, in contrast, clash with only a few, targeted beliefs and leave many others untouched. While I only consider the global ­brain-­​­­in-­​­­a-​­vat hypothesis here, sceptical arguments can equally well be driven by local sceptical hypotheses (­see Smith, 2016). 8 Sceptical concerns are also at the forefront of McDowell’s case for epistemological disjunctivism (­see, for instance, McDowell, 1982, 1995) – ​­though McDowell is less forthcoming than Pritchard in explaining the sceptical problem(­s) that disjunctivism is allegedly needed to answer. 9 According to Pritchard, disjunctivism is supported by our ordinary epistemic practices, since these practices often involve appealing to factive reasons when our beliefs are challenged. In particular, as Pritchard points out, ‘­I can see that…’ ‘­I can hear that…’, etc. can, under the right circumstances, seem like perfectly sensible answers to ‘­W hy do you believe that…?’ ‘­How do you know that…?’ etc. While these considerations are mentioned in Epistemic Angst (­2015, cha­p. 5, section 3), most of the dialectical weight still appears to be borne by the u ­ nderdetermination-​­based sceptical argument. These considerations are perhaps given a more prominent role in Pritchard’s Epistemological Disjunctivism (­Pritchard, 2012, part I, section 2).

160  Martin Smith 10 Logue (­2011, section 1) suggests a variation on the u ­ nderdetermination-​ b ­ ased sceptical argument in which U1 is, in effect, broken down into two separate premises. Adapting Logue’s terminology slightly, in her version of the argument U1 becomes: (­U1a) If BIV were true then I would not have rational support that favours P over BIV, and (­U1b) The rational support that is available to me is the same whether P or BIV is true. (­W hile a few other minor adjustments are needed to get us to U3, the argument can, in effect, go through with U1a and U1b in place of U1). In a way, this reformulated argument seems ­tailor-​­made for a disjunctivist treatment, as U1b would immediately set off disjunctivist alarm bells. And yet, presenting the argument in this way can also highlight why the dispute between disjunctivists and their opponents may be something of a sideshow: Unless we can find a way to motivate U1a that does not presuppose U0 or U0* or something of this ilk, the a­ nti-​­sceptic should refuse to follow the argument any further, and will never even get so far as to consider the a­ nti-​­disjunctivist premise U1b. 11 This helps to illustrate why Pritchard is wrong to focus on abductivism in his initial defence of U1 (­as noted in n4). U1 should be rejected by anyone who thinks it possible for evidence to favour a proposition without entailing it. It doesn’t matter whether this favouring relation is spelled out in abductivist terms or in some other way. In fact, the ­non-​­sceptic is under no obligation to spell this relation out at all in order to respond to the ­underdetermination-​ ­ ased sceptical a­ rgument – it’s b ​­ existence is strongly supported by examples, and the sceptic has provided no reason whatsoever to doubt it. 12 There are several different ways of defining infallibilism about knowledge, not all of which will make scepticism inevitable. On one definition, for instance, infallibilism is the view that one can only know a proposition X if X is entailed by one’s evidence. If we adopted a generous view of when a proposition could count as part of one’s ­evidence – ​­one that allowed propositions about the external world, other minds, the future, the past, etc. to ­qualify –​ ­ on-​­sceptics (­for t­ hen we could accept this kind of infallibilism and still be n discussion see Brown, 2018, cha­p. 1). The kind of infallibilism forced by C0, however, is less forgiving, and leaves no scope for manoeuvres of this kind. 13 The underlying assumption that Schönbaumsfeld identifies is what she calls the ‘­default view’ of perceptual ­reasons – ​­the a­ nti-​­disjunctivist position that our perceptual reasons are limited to how things appear and, thus, are the same in both ‘­good’ cases of veridical perception and ‘­bad’ cases of radical deception. In light of the discussion in the last section, I don’t think that this default view is enough, by itself, to motivate C1 (­or a more general sceptical conclusion). A ­non-​­sceptic can perfectly well accept Schönbaumsfeld’s default view (­or stay neutral) and still reject C1. 14 Some relevant alternatives theorists have rejected the assumption that ~X must always count as a relevant alternative to X (­Stine, 1976; ­Blome-​ ­ illmann, 2015). This style of relevant alternatives theory can’t be used to T underwrite C0* and may, on the contrary, provide a rationale for rejecting ­it  – ​­and a way of responding to the ­closure-​­based sceptical argument that doesn’t involve rejecting closure. In any case, as I suggest in the main text, a n ­ on-​­sceptic has no obligation to offer a ‘­rationale’ for rejecting C0* –​ ­drawing the connection with the suspect U0* is enough to shift the dialectical onus onto the sceptic. 15 Instances of C0** that involve the negations of local sceptical hypotheses, of the kind discussed in n7, may be more intuitively compelling than instances that involve the negations of global sceptical hypotheses like BIV. While global hypotheses would typically lie beyond any kind of empirical

Underdetermination and Closure   161 investigation, it may be straightforward to describe a procedure for investigating a local ­hypothesis  – ​­and for potentially acquiring justification or convincing reasons for rejecting it. If such a procedure exists, it might be particularly awkward to insist that one can know the hypothesis to be false without having carried the procedure out. I won’t discuss this further ­here –​ ­but see Smith (­2016). 16 Another kind of response to the c­ losure-​­based sceptical argument, favoured by some contextualists and subject sensitive invariantists, involves conceding that ‘­I don’t know that I am not a brain in a vat’ (­usually) expresses a truth when uttered, while maintaining that ‘­I know that I am currently seated at my desk’ can also express a truth when sceptical considerations are far from one’s mind. One advertised advantage of this approach is that it can explain why premise C1 seems intuitive and the argument seems compelling, while still avoiding scepticism (­see, for instance DeRose, 1995). I think it is a mistake, however, to regard the ‘­intuitiveness’ of C1 as some sort of datum that an adequate treatment of scepticism needs to account for. For the kinds of reasons given at the end of section 2, I suspect that one will only find C1 intuitive to the extent that one is willing to accept certain general p ­ rinciples – principles ​­ that are very much in need of philosophical scrutiny. To give these issues the attention they deserve would, however, take me beyond the scope of this chapter. None of this is to say, in any case, that either contextualism or subject sensitive invariantism is necessarily ­incorrect – ​­merely that they are not needed for a satisfying response to the ­closure-​­based sceptical argument.

Bibliography ­Blome-​­Tillmann, M. (­2015) ‘­Solving the Moorean puzzle’ Philosophical Studies v172(­2), p ­ p493–​­514. Brown, J. (­2018) Fallibilism: Evidence and Knowledge (­Oxford: Oxford University Press). Brueckner, A. (­1994) ‘­T he structure of the skeptical argument’ Philosophy and Phenomenological Research v54(­4), ­pp827–​­835. Byrne, A. (­ 2004) ‘­ How hard are the sceptical paradoxes?’ Noûs v38(­2), ­pp299–​­325. Cohen, S. (­1998) ‘­Two kinds of skeptical argument’ Philosophy and Phenomenological Research v58(­1), ­pp143–​­159. DeRose, K. (­ 1995) ‘­ Solving the skeptical problem’ Philosophical Review v104(­1), p ­ p1–​­52. 1970) ‘­ Epistemic operators’ Journal of Philosophy v67(­24), Dretske, F. (­ ­pp1007–​­1023. Goldman, A. (­2007) ‘­T he underdetermination argument for ­brain-­​­­in-­​­­the-​­vat scepticism’ Analysis v67(­1), ­pp32–​­36. Hawthorne, J. (­2003) Knowledge and Lotteries (­Oxford: Oxford University Press). Kelp, C. (­2021) ‘­T heory of inquiry’ Philosophy and Phenomenological Research v103(­2), ­pp359–​­384. Logue, H. (­2011) ‘­T he skeptic and the naïve realist’ Philosophical Issues v21(­1), ­pp268–​­288. Markosian, N. (­2014) ‘­Do you know that you are not a brain in a vat?’ Logos and Episteme v5(­2), ­pp161–​­181.

162  Martin Smith McDowell, J. (­1982) ‘­Criteria, defeasibility and knowledge’ Proceedings of the British Academy v68, ­pp455–​­479. McDowell, J. (­1995) ‘­K nowledge and the internal’ Philosophy and Phenomenological Research v55(­4), ­pp877–​­893. Nozick, R. (­1981) Philosophical Explanations (­Cambridge, MA: Harvard University Press). Pritchard, D. (­2005) ‘­T he structure of sceptical arguments’ Philosophical Quarterly v55(­218), ­pp37–​­52. Pritchard, D. (­2012) Epistemological DIsjunctivism (­Oxford: Oxford University Press). Pritchard, D. (­ 2015) Epistemic Angst (­ Princeton, NJ: Princeton University Press). Schönbaumsfeld, G. (­ 2017) ‘­ T he ‘­ default view’ of perceptual reasons and ‘­­closure-​­based’ sceptical arguments’ International Journal for the Study of Skepticism v7(­2), p ­ p114–​­135. Smith, M. (­2013) ‘­Entitlement and evidence’ Australasian Journal of Philosophy v91(­4), ­pp735–​­753. Smith, M. (­2016) ‘­S cepticism by a thousand cuts’ International Journal for the Study of Skepticism v6(­1), ­pp44–​­52. Stine, G. (­1976) ‘­Skepticism, relevant alternatives and deductive closure’ Philosophical Studies v29(­4), p ­ p249–​­261. Stroud, B. (­1984) The Significance of Philosophical Scepticism (­Oxford: Oxford University Press). Vogel, J. (­2004) ‘­Skeptical arguments’ Philosophical Issues v14, ­pp426–​­455. Wright, C. (­1984) ‘­Facts and certainty’ Proceedings of the British Academy v71, ­pp429–​­472. Wright, C. (­2004) ‘­Warrant for nothing (­and foundations for free)?’ Aristotelian Society Supplementary Volume v78(­1), ­pp167–​­212. Yalçin, U. (­1992) ‘­Skeptical arguments from underdetermination’ Philosophical Studies v68(­1), ­pp1–​­34.

10 Closure and Transmission Again Crispin Wright

1 Closure: Dynamic or Static? In the standard vernacular of logicians, a property of propositions is said to be closed under logical consequence if and only if it applies to every proposition entailed by any proposition to which it applies. Examples are truth, necessity, possibility and likelihood; counterexamples are falsity, impossibility and exceptionality. A further set of counterexamples are epistemic: being known, being believed, being regarded as certain. The latter counterexamples arise if for no other reason than because we are not logically ­far-​­sighted enough explicitly to acknowledge every consequence of what we know, believe or are certain about. But one might yet be tempted to suppose that every consequence of what we know is at least knowable—​­indeed that every consequence of what is knowable may likewise be known. When it is pointed out that logical consequence itself may not always be a decidable relationship, one may then retreat to the thought that every recognisable consequence of anything knowable is likewise knowable, that is, to Knowability Closure If P is knowable and may be known to entail Q, then Q also is knowable. Knowability Closure is an epistemic closure principle pretty much in keeping with the logicians’ standard use of “­closure” — ​­it departs only in replacing the reference to logical consequence by one to recognisable logical consequence. However recent epistemology has departed significantly further from the logicians’ standard use of “­closure”. Consider Matthew Jope’s formulation of what he calls Intuitive Closure If S knows that P, and S competently deduces Q from P, thereby coming to believe that Q on the basis of the competent deduction while retaining their knowledge that P, then S knows that Q.1

DOI: 10.4324/9781003104766-11

164  Crispin Wright Intuitive Closure is a clone of what Duncan Pritchard calls Competent Deduction Closure If S knows that P, and S competently deduces from P that Q, thereby forming a belief that Q on this basis while retaining her knowledge that P, then S knows that Q. 2 These formulations add to the idea of Q’s being a recognisable logical consequence of P the stipulation that its being so is actually competently deductively recognised by the subject, and then add more: that if S bases a belief in Q on that fact ­then—​­assuming knowledge of P ­throughout—​ t­ hey will then enjoy knowledge of Q. The closure of knowability across knowable entailment has thus become a principle about knowledge acquisition. It is easy to miss the difference. Knowability Closure above says nothing about how or why Q may be known. It merely stipulates that the knowable consequences of known propositions should be knowable somehow. It says nothing about the process whereby they may be known or on what knowledge of them may be properly based. It is therefore quite consistent with the existence of cases where although P entails Q and both may be knowable, knowledge of Q cannot rationally be based on knowledge of P but must be independent of it (­at least if knowledge that P is envisaged as acquired in a particular way). This is arguably as it should be. Closure principles generally say nothing about the grounds for the entailed proposition’s possessing the property in question. They are not inheritance principles. The consequences of necessary propositions are, necessarily, one and all necessary. But there is no basis in that thought for the idea that when a necessary P entails Q, Q’s necessity is somehow sourced in that fact. That is a further issue. Closure may or may not be indicative of grounding. So why does Jope count as “­intuitive” a principle incorporating the additional strength of his proposal? Here is what he says: What makes Intuitive Closure so intuitive is the thought that deduction is a paradigm way of growing one’s knowledge base. Williamson argues from the intuition that “­deduction is a way of extending 2000, 117). one’s knowledge” to the intuitive closure principle (­ Echoing Williamson, John Hawthorne argues that “­T he core idea behind closure is that we can add to what we know by performing deductions on what we already know” (­2005, 29). Likewise, in trying to formulate a satisfying closure principle, Steven Luper takes us to be trying to capture the intuition that “­we can extend our knowledge by recognizing, and accepting thereby, things that follow from something that we know” (­Luper, 2016). And in a similar spirit, Duncan Pritchard articulates the force of the intuitiveness of

Closure and Transmission Again  165 intuitive closure by asking “­How could one draw a competent deduction from one’s knowledge … without thereby coming to know the deduced conclusion?”3,4 This is a distinguished list of authorities. But the quoted thoughts of Williamson, Hawthorne, and Luper converge on the idea, merely, that deduction is (­normally) a way of extending ­knowledge—​­which no one sane will ­dispute—​­not that each and every deduction is potentially at the service of knowledge extension. Pritchard, for his part, merely asks what is presumably intended as a rhetorical question. Jope continues: One thing that is interesting to note here is that, if we take these remarks at face value and understand closure as a dynamic, ­knowledge-​­extension principle, the line between closure and transmission starts to blur.5 That blurring, in my judgement, is not an upshot that we should welcome. There are two reasons why we should want the distinction between Closure and Transmission to be robust. First, while epistemologists are and should be properly concerned with the formulation and refinement of principles which constrain ­knowledge—​­or more ­generally—​­epistemic warrant acquisition, they also have matching reasons to be concerned with principles which properly constrain the management of our beliefs, with their c­ o-​­ordination and revision. And static closure principles have a strong case for inclusion in the latter category. It is, on the face of it, only a static closure principle, like Knowability Closure, to which the standard closure sceptical argument need appeal: should it be indeed beyond all knowledge that one is not an envatted brain in the usual scenario, it is only static closure that would be needed to draw the sceptical conclusion that any proposition that entails that one is not so afflicted is likewise beyond all knowledge. More generally, whatever the trend in the recent literature, there are good motives to explore n ­ on-​­dynamic principles which concern knowledge architecture and knowledge commitment. When does knowledge of some propositions ensure the possibility of knowledge of others? When must you know certain propositions if you are to (­claim to) know others? When Dretske (1970) and following him Nozick (1981), motivated by the thought that k ­ nowledge-​­conferring processes should, as such, be counterfactually sensitive to variation in the fact known, proposed that Closure actually fails, the suggestion was generally received as shocking. What shocked was the excuse seemingly offered for apparently egregiously irrational professions like “­I know I have hands but ­no-​­one can know whether they might not be a handless brain in a vat”, which,

166  Crispin Wright according to Dretske’s and Nozick’s theories, ought not to have seemed irrational. Readers will have to consult their own intellectual phenomenology, but I would suggest that the sense of irrationality provoked by such a claim has much less to do with the idea that in claiming knowledge that they have hands, the speaker thereby claims to be in position deductively to advance to knowledge that they are not a ­brain-­​­­in-­​­­a-​­vat than it has with the idea that in claiming knowledge their having hands, they commit themselves to the possibility of knowing that they are not a ­brain-­​­­in-­​­­a-​­vat, which accordingly cannot simultaneously rationally be disclaimed. The second reason why we should wish to retain a ­non-​­dynamic understanding of Closure, contrasted with Transmission, is this. Suppose, as I will in this chapter once again argue, that T ­ ransmission-​­failure is a real phenomenon. Suppose in particular that there are cases where knowledge is achievable both that P, and that P entails Q, but where there is no possibility of rationally basing knowledge of Q on recognition of those facts. Then there will be a question ­whether—​­if such a case is not simply a case where Q is unknown: in other words, a straight example of closure failure as Dretske and Nozick ­anticipated—​­it is nevertheless a condition on such a case arising that Q be known independently. If that is so, then that will be an argument for sustaining a static closure principle while rejecting any unrestricted dynamic principles such as Intuitive Closure and Competent Deduction Closure. What is unfortunate about an exclusive focus on dynamic closure principles is that it encourages oversight of that dialectic. Generally: when Closure (­for whatever epistemic operator) is formulated dynamically, as in Intuitive Closure and Competent Deduction Closure, there is of course, as Jope observes, foreseeable difficulty in explaining how it can hold when Transmission fails. But that is because dynamic closure is already, in effect, a principle of unrestricted transmission. If Transmis­ hether—​­and if so sion does indeed fail in certain cases, the question is w ­why—​­some form of static closure principle still holds, or whether none does. The tendency exemplified by Jope and the authors he cites is liable to make that important question inaudible. In what follows, I will review the case for rejecting unrestricted Transmission before returning to the question of the status of Static Closure principles. First we need to be clearer about this question:

2 What Is it for an Argument to Be Transmissive? What is it for a valid argument to be transmissive of epistemic warrant for its premises to its conclusion? I have explored various accounts in previous work6 but here I think it will be helpful to focus on one especially clear kind of diagnosis.

Closure and Transmission Again  167 It will call for some kit. First we need what I think, since the work of Pollock and others, is a standard notion of an undermining 7 defeater for a belief. Suppose we are in the business of accumulating evidence on the question whether or not a proposition P obtains. Let E be the evidence so far gathered and suppose we incline to regard E as defeasibly supportive of P. Let U state a potential underminer of E’s support for P: that is, (­we are of the opinion that) if U is true, then E should no longer be regarded as supportive of P. For example, let P be that you have contracted COVID 19, and let E be the positive outcome of a lateral flow test that you have just taken using a kit purportedly supplied by a ­well-​­known pharmaceuticals manufacturer based in San Diego, CA. We can count it as an additional part of your evidence that the kit is advertised at approximately 98% positive ­accuracy—​­false positives occur on average only in 1 of 50 trials. Now suppose you ­learn—​­this is ­U —​­that the Wolverhampton depot in which the kits are stored for distribution in the UK has inadvertently taken in a large number of fake kits. Presumably, this information should rationally be received as to some degree compromising E’s support for P. Note that whether something is an underminer of the significance of a given item of evidence may vary with collateral information. It may only be in the context of background theoretical knowledge that a given kind of evidence for a certain belief loses its supportive force. Up on the Cuillin Ridge in the Isle of Skye in thick mist, I get a steady compass signal (­E) indicative, in conjunction with the map, that I am indeed on track and should continue on in the same direction. Someone says, “­You know the rock around here is predominantly gabbro?” While I ought to regard this as undermining the seemingly encouraging compass reading, I am ignorant of the magnetic properties of gabbro. So I merely reply “­Yes”, and carry on trusting my compass undaunted. We can prescind from such ­information-​­relativity in the notion of an underminer in what follows. Second, we need a notion of ­open- ​­mindedness. Your ­open-​­mindedness about P, in the sense that concerns us here, requires i your having no opinion whether or not P; ii your not being agnostic about P in the sense of “­agnostic” frequently associated with the stance of religious agnosticism, that is a stance which holds e.g. that the existence of God is a matter on which no rational subject can in any circumstances can consider that they have a knowledgeable or even a justified view. Your ­open-​­mindedness about P involves readiness to accept or reject P, should appropriate evidence turn up but also, as we shall here understand it, iii your being of the opinion that so far you have no basis for a view, but not ruling out the possibility of new information that might mandate taking one.

168  Crispin Wright ­ otions—­​­­open-​­mindedness and undermining defeat of With these two n ­evidence—​­in place, we can elicit an evident but interesting corollary. Suppose you are presented with purported evidence E, for P, and suppose you are ­open-​­minded about whether something you recognise as a specific potential underminer, U, for E obtains. Note that we are not supposing that you are sceptical about ­U —​­doubtful that U is ­true—​­but merely that you consider that you so far have no basis for a view about whether or not U is true. Well, if U does obtain, then, as you recognise, E should not be regarded as supportive of P. So if you are ­open-​­minded about U, you ought rationally to be open minded about whether or not E should be regarded as supportive of P. More explicitly, here is the reasoning that underlies that thought. U, if true, is an underminer of E’s evidential force for P. So if E is supportive of P, U is false. So if your present state is one of open mindedness whether U is false or not, it cannot be that you are in epistemic position to regard E as supportive of P. There is an implicit appeal here to closure for whatever relevant epistemic notion is supposed to be at odds with o ­ pen-​­mindedness. But that seems perfectly intuitive. Consider an example. You are running a schoolroom physics experiment in which it is important that there be no leakage of pressure from the apparatus. If pressure is lost, = U, the significance of the findings for the hypothesis under test, = P, will be severely compromised. Suppose you are ­open-​­minded about whether pressure was lost; members of the class attached the clamps, you haven’t checked their work and reckon it quite possible it was slapdash. But you run the experiment, take the final readings, = E, and announce that they are indeed confirmatory of P. Can you rationally profess to have confirmed P although simultaneously open m ­ inded-​­about whether pressure was indeed lost? We have introduced the relevant options here in relation to cases where what is at issue is the suasive force of defeasible evidence. But the same basic framework applies equally to cases where P is based on some form of (­putatively) direct cognition: perception, episodic memory, or (­perhaps) ­non-​­inferential intellectual intuition, for example, where reliance on defeasible evidence is not involved. Even so, cognitive faculties of this kind have conditions of effective operation. And a claim based on the presumed operation of such a faculty may be undermined by finding that these conditions were compromised in one way or another. Perhaps the conditions under which the faculty was operating are unsuitable for its effective function. (­Suppose, for instance, you are in Barn Façade County.) Or perhaps the faculty itself was suffering some kind of internal malfunction. (­Perhaps you have taken an hallucinogenic substance of some kind.) Then, once again, you cannot rationally advance a claim, P, based on the presumed effective operation of such a faculty, and at the same time profess o ­ pen-​­mindedness about whether some condition,

Closure and Transmission Again  169 U, obtains which you acknowledge to be a potential underminer of the f­ aculty—​­a condition such that, if it obtains, the effective functioning of the faculty concerned will indeed be likely to be compromised. We still need two more notions. Where U is any underminer of the support of evidence E for P, let us call the negation of U an Authenticity Condition for E with respect to P. Likewise where U is an underminer of the capacity of faculty F to reliably determine whether P, let the negation of U be an Authenticity Condition for F with respect to P. Finally let us say that an argument {P ⇒ Q} is potentially cogent for a rational thinker who is presented with evidence E for P, or who achieves prospective verification of P by employing faculty F, just if acknowledgement of the evidential force of E for P, or the competence of F in the circumstances to determine whether P,8 is rationally consistent with o ­ pen-​­mindedness about Q. Intuitively, a cogent argument is one where acceptance of the grounds offered for its premises is consistent with anterior ­open-​­mindedness about its conclusion. Within the class of deductively valid arguments, it is only the cogent arguments, thus characterised, that are at the service of extending our knowledge. So we come finally to the crux: is there any other kind? Are there valid arguments for whose premises there are probative grounds, but which are nevertheless not cogent for a given thinker who is antecedently ­open-​ m ­ inded about their conclusion? Well yes, indeed there are! We now have all we need to articulate a template for one kind of Transmission failure.9 We merely have to construct a valid argument, {P ⇒ Q} where Q is an authenticity condition for the particular evidence, E, offered for P, or for the competence of F to determine whether P. In order for the argument to be potentially cogent, it has to be rationally possible to be ­open-​­minded about Q consistently with appreciation of the probative force of E for P. If Q is an authenticity condition for E with respect to P, that is exactly what is not rationally possible. The ­non-​­transmissiveness of an argument is accordingly not a feature purely of the contents of the premises and conclusion concerned but is relative to the informational context: it is relative to the character of the particular epistemic pedigree of the premises. ­Non-​­transmissive but valid arguments for a given thinker are arguments where, such is the thinker’s information, a commitment to the conclusion, whether or not based on independent cognitive achievement, is a rational precondition of their accrediting the evidence for the premises (­or the cognitive process involved in recognising them as true) as persuasive of their truth. You cannot consider yourself to have learned anything by following such an argument if prior persuasion of the conclusion is a rational precondition of accepting the premises on the particular grounds proffered. Here are a couple of quotidian examples:

170  Crispin Wright 2.1 ­Chiff-​­Chaffs and ­Willow-​­Warblers ­Suppose—​­to adapt an example of Austin’­s—​­that I am a less than expert but not totally incompetent bird watcher whose method of deciding whether or not a bird is a willow warbler is simply to judge on the basis of size, colour, movement and overall a­ ppearance—​­provided I am already confident it is not a c­ hiff-​­chaff. If, on the other hand, I regard that issue as open, then, the two species being so similar, I have no method except to ask an expert. Let the region be such that there are vanishingly few ­chiff-​­chaffs in the local bird ­population—​­indeed, no ­warbler-​­family birds of any kind other than ­willow-​­warblers. And suppose I believe this. Then the following argument: —​­ E:  That

bird’s size, colour, movement and general gestalt suggest it is a w ­ illow-​­warbler P:  That bird is a willow warbler Q:  That bird is not a c ­ hiff-​­chaff —​­is ­non-​­transmissive for me, though not for an expert birder who can distinguish the two species by sight. That is because, given my background information and limitations, I can rationally treat E as probative of P only if I am antecedently confident of Q. Note that the situation is not that I get a warrant for P in any case which then, as it were, gets stuck on the wrong side of the entailment from P to Q. Rather it is only in a context where I am antecedently confident of Q that I can rationally consider that I have good grounds for P. I cannot rationally appeal to the argument to enhance my confidence in Q. Another example 2.2 Twins Jessica and Jocelyn are identical twins with a disconcerting habit of matching their daily attire, hairstyle, etc. Both are well known to me but not so well that, seeing either by herself, I can be confident which of them it is. Consider this argument: E:  That girl looks exactly like P:  That girl is Jessica Q:  That girl is not Jocelyn

Jessica

Let the context be one where I am antecedently confident of Q: I know that Jocelyn is out of town to attend a campus ­fly-​­back. Then, seeing a girl approaching who looks exactly like ­Jessica—​­so I have ­E —​­I am fully warranted in concluding P. But I cannot then rationally refer this warrant across the entailment to Q to conclude that she is not Jocelyn. Rather

Closure and Transmission Again  171 again, it is only in the context of my antecedent knowledge about Jocelyn’s whereabouts that I can consider E supportive of P in the first place. One more case of a slightly different structure: 2.3  Town Clock I am at the Urbino summer school in Epistemology sitting out in the town square during the ­mid-​­morning break and wondering if I have time for another excellent Italian coffee before the 11.00 am session. I look across at the town hall clock and note that it says 10.51. So there is just about time. Can I trust it to be accurate enough? (­I know it is ­functioning—​­it chimed a few minutes ago.) I reason as follows: E:  The clock says 10.51 P:  So the time is 10.51 Q:  So (­from E and P together)

The clock is accurate on this occasion.

Thus (­idiotically) reassured, I catch the waiter’s eye … … These examples ought to be persuasive that rational belief management sometimes requires recognising that the order of the epistemic dependencies among one’s beliefs and suppositions inverts the order of their logical dependencies. But notice that we have taken the issues into the territory of rational belief management. Our question has been, in effect, can I always rationally transmit whatever grounds I have for certain beliefs to propositions that I recognise to be entailed by them? The answer manifestly is: N ­ o—​­not when the latter are authenticity conditions for the probative force of the grounds concerned for the former. But does this point abrade with dynamic closure principles for knowledge?

3 Does transmission failure, as now schematised in section 2, count against the validity of intuitive closure? Jope’s Intuitive Closure, recall, runs as follows: If S knows that P, and S competently deduces Q from P, thereby coming to believe that Q on the basis of the competent deduction while retaining their knowledge that P, then S knows that Q. Closure “­Dynamists”—​­Jope, Pritchard, Hawthorne, Williamson et al.—​ ­might argue that there is no tension between this kind of principle and the gist of the previous section. After all, the examples offered illustrate how a thinker’s ­open-​­mindedness about the conclusions of certain valid arguments, {P ⇒ Q},—​­a precondition of their potential ­cogency—​­ought

172  Crispin Wright rationally to have the effect of disabling a prima facie warrant for their premises. And in that case, the thinker won’t know that P in the first place, so that the antecedent of Intuitive Closure will not be satisfied. But not so fast: matters are more complex. First, we must be mindful that basing, as a real psychological phenomenon, can be rationally inappropriate: one can base a belief on considerations that are poorly selected to justify it. Biases, prejudices and favouritism offer a legion of examples where a thinker’s belief is based on considerations that are rationally inappropriate. Second, we ought to want to impose a rationality condition on knowledge in general. A belief should not count as knowledgeable, even if gifted by the circumstances of its ­formation—​­the method and prevailing ­conditions—​­in such a way as to ensure its truth, if the thinker is nevertheless irrational to hold it. (­If some of your prejudices turn out to be nomically true, that doesn’t place the beliefs they have driven you to hold in epistemic good standing.) The question is accordingly whether there are circumstances of the structure depicted by Intuitive Closure in which basing the belief in Q on P and knowledge of the deduction would result in a t­ rue—​­since the consequences of known propositions are t­ rue—​­but irrationally based belief. And the answer will depend on how relaxed we are about the degree of epistemic scrutiny required to know that P. Consider Town Clock again. Could I be said to know what time it is by looking at the Urbino clock? Ordinarily we should not scruple to allow that I could. No doubt that’s because it’s common knowledge that clocks are designed to tell the time, that municipal clocks are generally maintained so as to help the citizens rather than provoke confusion, etc. Generally it would seem in most contexts overly fastidious to insist on a background check on the clock in question before regarding its readings as ­knowledge-​­conducive. So allow that I can know what time it just by looking at the clock. Manifestly I can also know what the clock says just by looking at the clock. So, by Intuitive Closure, I can know the ­conjunction—​­that the clock says that ­ ccasion—​­just T and that T, and hence that the clock is accurate on this o by looking at the clock and basing my belief in the conjunction just on a ­conjunction-​­introduction step. But that is manifestly an irrational basis for confidence in the ­on-­​­­the-​­occasion accuracy of the clock. In general, Intuitive Closure will be in trouble whenever Q is an authenticity condition for the relevant method for acquiring knowledge of P of such a kind that ordinary standards, or a preferred philosophical account, of what it takes to get knowledge of P requires no independent scrutiny of Q which is, rather, permissibly “­taken for granted”, as we say, if considered at ­all—​­a “­prop” in the sense of fn. 9 above. In neither case would basing the belief in Q on the deduction amount to basing it on anything that corroborates it. So: where knowledge is required to be rational, Intuitive Closure as formulated is open to counterexamples and accordingly is false.10

Closure and Transmission Again  173 What might be true would be something along these lines: Intuitive Closure* If S knows that P by some method for which Q is not an authenticity condition, and S competently deduces Q from P, thereby coming to believe that Q on the basis of the competent deduction while retaining their knowledge that P, then S knows that Q. But I shall not explore further here whether further tweaks are needed.

4 Transmission ­Failure—​­Some Controversial Cases Let us now briefly review, in the light of the ­Transmission-​­failure template outlined in Section 2, four somewhat notorious and much discussed arguments: G.E. Moore’s11 (­widely ridiculed until quite recently) ‘­Proof of an external world’, Michael McKinsey’s12 apparent proof of the inconsistency of semantic externalism with privileged psychological ­self-​­knowledge, Hilary Putnam’s13 (­generally disbelieved) proof that we are not ­brains-­​­­in-­​­­a-​­vat, and Fred Dretske’s14 fraudulent zebras. First up is 4.1 Moore The status of Moore’s ‘­Proof’ appears to turn on what we write in for E. This reasoning, at least, is arguably n ­ on-​­transmissive: E:  My experience is in all respects as if there is a hand in front of my face P:  Here is a hand.

So, (­since hands are material objects existing in space) —​­ Q:  There Q  is

is an external material world.

manifestly an authenticity condition for the probative relevance of E for P. If, bizarrely, I am o ­ pen-​­minded about the existence of the external world, I ought to be o ­ pen-​­minded about the evidential relevance of my experience for claims about it.

There are two ways this account of the shortcoming in this version of Moore may be challenged. One is the perceptual Dogmatism stoutly defended by Jim Pryor15 On Pryor’s account, Q while indeed an ­anti-​­underminer for E ­vis-­​­­à-​­vis P, is not such that antecedent ­open-​ ­mindedness about it must rationally require o ­ pen-​­mindedness about the evidential force of E for P. What is true is that an antecedent doubt about ­Q —​­a tendency to disbelief of ­Q —​­defeats the evidential force of E for P. But mere o ­ pen-​­mindedness should not. Rather sense experience

174  Crispin Wright is properly regarded as default (­though defeasibly)—​­in ­post-​­­Philosophical evidentially supportive for claims Investigations jargon, criterially—​­ about the external world and may thus, absent defeating evidence or spoiling background considerations, be rationally responded to by the formation of appropriate external worldly beliefs. I shall not further discuss perceptual Dogmatism here.16 A second response, currently somewhat in the ascendant, is to dispute the very epistemic architecture of the proof when it is configured as in the ­3 -​­step form above. On this kind of response, it is wrong to interpret Moore’s train of thought as beginning with evidence conceived as internal to consciousness. Rather the sensory states on which perceptual justification is based should be conceived of as wide—​­as ­world-​­involving. When Moore holds up his hand in front of his face, the epistemic ground for his claim that “­Here is a hand” is directly constituted by his sensory awareness of the hand, an intake, in John McDowell’s phrase, of an aspect of “­the layout of reality”.17 In effect, therefore, the proper way of representing Moore’s proof is rather this: 4.2  Moore Lite E/ ­P:  Here is a hand. So Q:  There is an external

material world

When the architecture of the proof is so conceived there is, of course, no transmission failure. But it is a further question whether, as seems to be widely supposed, this a­ djustment—​­the resort to a factive notion of perceptual ­warrant—​­can be marshalled against perceptual scepticism without, in effect falling back on reasoning for which ­Transmission-​­failure is once again an issue. The point can best be bought out in a “­disjunctivist” setting. Consider this argument: 4.3 ­Ante-​­Moore E:  The

phenomenal character of my subjective experience is consistent both with its amounting to perceptual awareness of a hand in front of my face and with its being the product of a sustained hallucination. P:  I am perceptually aware of a hand in front of my face Q:  Here is a hand Here an endorsement of E is consistent with acknowledging, with metaphysical disjunctivism, that there is no one kind of experiential state of which both veridical perception and hallucination are species. Rather the one is, metaphysically essentially, a form of relational engagement with matters external while the other is but a play of light and colour in phenomenal consciousness. Let that be so. Still, the point hasn’t gone

Closure and Transmission Again  175 away that the one kind of experience can in principle be subjectively indistinguishable from the other. So if we represent the ultimate justification of the premise in Moore Lite—​­“­Here is a hand”—​­as correctly captured in its role as conclusion in ­Ante-​­Moore, then the issues about transmission failure are now going arise in the process of getting as far the premise of Moore’s Proof, not in the drawing of its metaphysically heavyweight conclusion. In general, it seems that any kind of reasoning that starts with a basis of evidence given by phenomenal consciousness and purports to underwrite conclusions about the very existence of a material reality is g­ oing—​­pace Pryor’s d ­ ogmatism—​­to be open to a challenge of T ­ ransmission-​­failure unless some case is made that that is a misconceived s­ tarting-​­place: that perceptual justification begins with factive states of worldly awareness. The sceptic will then gleefully rejoin that their challenge is in effect to justify the claim that we can legitimately start with the assumption that we do indeed enjoy such factive states. At this point the issue with Moore’s proof is not one of whether the reasoning is ­warrant-​­transmissive but whether we are ­entitled—​­and if so what entitles ­us—​­to lay claim to its premise.18 Now to McKinsey: Michael Mckinsey’s much discussed ‘­paradox’ purports to elicit a contradiction between the status of ordinary, ‘­privileged’ ­self-​­knowledge of one’s intentional attitudes and one form of semantic externalism. It is simply formulated: P(­1):  I believe that water is wet P(­2):  An agent has the concept

of water only if they, or others of their speech community have, historically, interacted with water. Hence Q:  Members of my speech community have, historically, interacted with water. The paradox is then that ordinarily effortless and immediate ­ self-​ k ­ nowledge of one’s basic intentional states can deliver knowledge of P(­1). So if a priori philosophical reflection can indeed corroborate a content externalism that entails P(­2), viz. that it is only if the concept of water picks out a real physical kind with which we have had historical interactions that it is a real ­truth-​­conditionally contributive concept, then it would appear to follow that by combining quotidian armchair psychological ­self-​­knowledge with some subtle armchair philosophy, I can deliver knowledge of a contingency of h ­ istory—​­pretty good going from the armchair! If you share my hunch, you will suspect that Mckinsey’s argumentation is somehow n ­ on-­​­­transmissive—​­that, on its assumptions, it somehow merely elicits a presupposition of the warrant for its premises, rather than furnishing a genuinely novel reason for believing its conclusion. But

176  Crispin Wright is that so? There are two obstacles in the way of getting a clear view of the matter. Transmission-​ First, a mechanical application of the template for ­ ­failure proposed in Section 2 would bid us ask whether o ­ pen-​­mindedness about McKinsey’s conclusion would rationally enforce open mindedness about the sufficiency of the warrants for its premises. But what are the warrants for its premises? Well, in the case of P(­2), the warrant is given by the philosophical considerations that are supposed to mandate the relevant ­content-​­externalism. One might, of course, have one’s doubts about how persuasive those considerations really are. But it doesn’t seem that ­open-​­mindedness about whether members of my speech community have ever historically interacted with water should enjoin open mindedness about the cogency of such philosophical argumentation as a species,—​­not in the way that ­open-​­mindedness about the very existence of an external world ought to entrain open mindedness about the probative force of phenomenal experience for claims about material objects. As for P(­1), well what in any case is our warrant for sincere ­self-​­ascriptions of basic intentional attitudes? I think we have no clarity about the cognitive machinery underlying avowals of the kind expressed by P(­1), and consequently no clarity about the conditions for its effective operation, or what circumstances might undermine it. The second awkwardness about trying to bring McKinsey under the ­Transmission-​­failure template is that o ­ pen-​­mindedness about the ­conclusion—​­“­Members of my speech community have, historically, interacted with water”—​­is itself an attitude that embeds the concept of water and thereby, in the presence of P(­2), entails the integrity of that concept which, from an externalist point of view, is itself already a commitment to the conclusion. You cannot rationally be o ­ pen-​­minded about Q in this case without denying P(­2)—​­so trivially you should wonder about the adequacy of the grounds for P(­2). But the kind of commitment to so wondering which the test is looking for is not supposed to go through a doubt about the t­ ruth-​­value of a premise, as opposed to a doubt about the adequacy of the type of grounds proffered for it. non-​ We could just give up on the suspicion that McKinsey is ­ ­transmissive. But I believe that would be premature. Here is another approach. ­Content-​­externalist views come with the possibility of certain kinds of illusion of content. Most of us would, for example, accept an externalist view of singular demonstrative content. When we use demonstratives, real external worldly semantic relations are required before we can succeed in thinking anything with a determinate truth condition. Suppose I am a Victorian lepidopterist on an Amazonian expedition, looking for rare and exotic species. Unmindful of the rather delicious stew of the local fungi I lunched on, I seemingly catch sight of a large specimen, brilliantly patterned in orange, purple and aquamarine, and think to myself, “­That is a hitherto undocumented

Closure and Transmission Again  177 species and will make me famous”. I do not think anything true. But nor is there any determinately false singular thought which is exactly what I t­hink—​­the nearest determinate false thought will be an existential thought: that there is a brilliantly coloured butterfly over there, etc. In so far as the content I intentionally think purports to be a demonstrative content, it is an illusory content. In an influential early discussion of these matters,19 Paul Boghossian introduced the example of Dry Earth. Dry Earth seems to its inhabitants (­intrinsic duplicates of us) just as Earth seems to us: it seems to them that there are rivers, lakes and rain, and a clear, odourless, potable liquid flowing from their domestic taps. But this is all illusory. There’s no such substance anywhere in their environment. What goes wrong when a Dry Earthling runs Mckinsey’s argument? Well clearly, the premise, P(­1), is not true: the embedded content, that water is wet, is, by externalist lights, an illusory content for the Dry Earthling, so the tokening of ‘­water’, no less than the tokening of ‘­that’ in the lepidopterist’s thought, is divested of determinate ­truth-​­conditionally contributive content. The Mckinsey Q is an authenticity condition not for the good standing of the process whereby P(­1) is seemingly confirmed but for there being anything determinate to be confirmed in the first place. That seems nearer the mark. But there is still the awkwardness that McKinsey Q, as formulated, presupposes the good standing of the embedded content in P, so that the relevant issue is silenced just by o ­ pen-​ mindedness about Q. I think there is no way around this except to ­ countenance the notion of what is, by externalist lights, a defective concept that can nevertheless feature in ersatz attitudinal states that, while contentually defective, are nevertheless genuinely explanatory of thought and action in routine ways. (­It is, after all, not as if all is dark within the Victorian lepidopterist’s head: suppose, e.g., he gets out his net and takes — just for the purposes of section 4.4 — a swipe at what is in fact thin air. How to explain his action?) So, let us adopt the convention of italicising a concept expression to indicate that it is to convey a ‘­concept’ which, in the good case, where external presuppositions of its purported content are met, will indeed make whatever kind of semantic contribution externalism proposes, but in the bad case will contribute only to an illusion of determinate thought. And let us say that a ‘­concept’ of this kind is validated just in case the relevant external conditions for its determinacy of content are indeed met. Then we can r­ e-​­run a Mckinseyish argument like this: 4.4 ­Ante-​­McKinsey P(­1minus):  I believe that water is wet P(­1):  I believe that water is wet P(­2):  An agent has the concept of water

only if they, or others of their speech community have, historically, interacted with water. ­Hence​­

178  Crispin Wright Q: 

Members of my speech community have, historically, interacted with water.

Now we have in effect equipped ourselves with the resources to identify an authenticity condition for McKinsey’s P(­1). It is: Q*: Members

of my speech community have, historically, interacted with a substance that validates water.

Note that, given that the only substance that can validate water is  … water (!), Q* is actually equivalent to Q. ­Open-​­mindedness about Q* rationally enjoins ­open-​­mindedness about the transition from P(­1minus) to P(­1). If I am o ­ pen-​­minded about the possibility that water, in my thought, is by externalist lights, contentually illusory, then I should be o ­ pen-​­minded about whether I can justifiably advance from the piece of ­self-​­knowledge expressed by P(­1minus) to McKinsey’s original P(­1). If this analysis of the issues is accepted, the suspicion that McKinsey merely concludes in a proposition that the justifiability of its premises presupposes, rather than generating a novel reason for accepting that proposition, is confirmed. But the explanation is not exactly that the actual reasoning McKinsey gave is ­non-​­transmissive. McKinsey itself, Like M ­ oore-​­Lite, is transmissive. Nevertheless the impact of the relevant kind of ­content-​­externalism is to complicate the justificational architecture of its premise P(­1) in such a way that (­i) the real item of pure ­self-​­knowledge, strictly so regarded, involved is rather P(­1minus), and that (­ii) there is an authenticity condition for the transition from that to McKinsey’s P(­1) —​ ­to wit: Q*—​­which is equivalent to the original conclusion Q. 4.5 Putnam Proof’, like Mckinsey’s, is fuelled by somewhat inchoate Putnam’s ‘­ ­content-​­externalist assumptions. He envisages a scenario in which creatures whose subjective mental lives are qualitatively indistinguishable from ours, and whose language is lexically and grammatically indistinguishable from English, have in fact always been brains in a vat, tended by computers and automatic machinery; moreover this description of their situation is actually a complete inventory of their world. Putnam takes it that, in this predicament, these thinkers are so causally isolated from their real environment that, on externalist assumptions, they could not so much as have the concept of ­brain-­​­­in-­​­­a-​­vat and thereby could not mean ­brain-­​­­in-­​­­a-​­vat by their A ­ nglo-​­form expression, “­­brain-­​­­in-­​­­a-​­vat”, or by any other. However we, in regular English, mean exactly ­brain-­​ ­­in-­​­­a-​­vat by “­­brain-­​­­in-­​­­a-​­vat”. It follows that we are not in the described p ­ redicament—​­that we are not brains i­ n-­​­­a-​­vat.

Closure and Transmission Again  179 Putnam’s argument makes assumptions that compromise its possible epistemological significance. His scenario has ­features—​­for instance, the brains’ historically complete causal insulation from brains or ­vats—​ w ­ hich are required to get the background c­ ontent-​­externalism to engage with it but which someone who wanted to fashion a ­brain-­​­­in-­​­­a-​­vat scenario for sceptical purposes could easily dance around, e.g. by working with the alternative scenario of overnight envatment by a mad scientist instead. In fairness, however, Putnam himself seems to have primarily conceived the argument not as a riposte to scepticism but as a challenge to a core idea of metaphysical r­ ealism—​­roughly, the notion that if our epistemic entanglement with the external world generates any true beliefs, that it does so is matter of pure contingency. I won’t comment further here on its effectiveness in that direction. 20 Our question is the limited one whether when one reasons as below, and assuming the premises are warranted, one generates a warrant for accepting the conclusion: Putnam: P(­1): 

In the language I speak, ‘­­brain-­​­­in-­​­­a-​­vat’ means ­brain-­​­­in-­​­­a-​­vat (­My language is disquotational) P(­2): In the language of b ­ rains-­​­­in-​­a vat in the described predicament, “­­brain-­​­­in-­​­­a-​­vat” could not mean ­brain-­​­­in-​­a vat (­by the absence of the causal connections which according to ­ content-​­ externalism would be required by such a content). So Q:  I am not such a ­brain-­​­­in-­​­­a-​­vat This is indeed suspiciously reminiscent of McKinsey. But let’s see. Here we have two independently supported premises conjunctively entailing the stated conclusion. The question about transmissiveness is therefore whether a rational thinker who is ­open-​­minded a­ bout—​­so, presumably, haunted by the possibility o ­ f—​­the conclusion’s being true should accordingly be ­open-​­minded about the acceptability of either premise on the grounds offered for them. It can seem o ­ bvious—​­and has, to a number of ­commentators—​­that in the presence of P(­2), P(­1) is q ­ uestion-​­begging; or, as we can more felicitously now express the charge, that anterior ­open-​­mindedness about Q should, if one accepts P(­2), rationally require ­open-​­mindedness about P(­1). After all, if you were a ­brain-­​­­in-­​­­a-​­vat, you would by P(­2) have no means of expressing the concept ­brain-­​­­in-​­vat. That, though, second thoughts should rapidly teach, is an addled train of thought. Suppose you are genuinely ­open-​­minded whether or not you are a ­brain-­​­­in-­​­­a-​­vat. Nevertheless, whether you are or not, it is presumably solid that your language, the language we are now working i­n—​ w ­ hether it is English or its vat ­homophone—​­allows for the expression of ­the—​­that is, your— ​­concept ­brain-­​­­in-­​­­a-​­vat. After all, this is the very concept about your falling under which you are ­open-​­minded. So P(­1) is unexceptionable whatever your predicament. Hence if you are rightly

180  Crispin Wright persuaded by ­content-​­externalist considerations that creatures in a certain scenario would lack any expression for a concept which you can express, that is just to say that you can do something which they cannot. So you are not one of them. QED What is true is that a ­brain-­​­­in-­​­­a-​­vat could presumably carry out token reasoning of exactly this form, and accordingly validly draw some conclusion or other. “­So how do I know that that is not what I have done?” You should know better than to ask! Your question is answered by the very reflection that you are not a ­brain-­​­­in-­​­­a-​­vat. I suggest accordingly that there is no ­clear-​­headed obstacle to the idea ­ arrant-​­transmissive. What that Putnam’s argument, unlike Mckinsey’s, is w exactly is its philosophical significance is, of course, quite another matter.21 Finally Zebras: Here is Dretske’s famous example, which he originally conceived, of course, as a failure of Closure: E:  The look of the animals, in a zoo cage marked “­Plains Zebra” P:  Those animals are Zebras Q: Those animals are not mules, their coats cleverly dyed and manes

coiffured to look exactly like zebras. This impresses as an absolute paradigm of T ­ ransmission-​­failure. I believe it has been a primary force in persuading many theorists of the reality of the phenomenon. Surely, if you are open minded about the possibility that the zoo has gone in for artful disguise and fraudulent labelling of its animals, you have to be ­open-​­minded about the significance of the appearance presented by its exhibits, in which case you cannot yet trust that E is indeed probative for P. This was, for a long time, my own reaction to the example. But philosophical appearances, no less than those of the animals in fraudulent zoos, can of course be deceptive. What is the logical form of the above Q? Presumably it is a negated conjunction: it is not the case that [Those animals are mules and those animals have had their coats cleverly dyed… etc.], so it is something of the form ~(­Fx & Gx). Hence it is tantamount, in classical logic at least 22 to a disjunction of negated disjuncts: ~Fx V ~Gx. But now the ­Transmission-​­failure diagnosis of Zebras hits a snag: surely any reasoning that warrants a certain conclusion must also warrant the disjunction of that conclusion with any proposition. The following reasoning is surely transmissive, notwithstanding the inconclusiveness of E: E:  The look of the animals, in a zoo P:  Those animals are Zebras Q:  These animals are not mules.

cage marked “­Plains Zebra”

Closure and Transmission Again  181 So how can it make a difference if we tack an arbitrary disjunct onto Q? The very tempting answer is: it cannot. What is happening, it may be suggested, when the argument is taken in a way that incites the intuition that Dretske intended, is that, by some quirk of the pragmatics of conversation, we receive a negated conjunction as though it implied the negation of its second conjunct. “­He’s not a very funny comedian” would usually, for example, be heard as saying that the gentleman in question is not very funny. Which is fine if the conversational common ground includes that he is a comedian. But it seems that in Dretske’s example, the same conversational habit kicks in even though it is not presumed that the animals in question are mules. We hear the conclusion of the argument as implying that the animals in question have not been cleverly disguised. And while that is indeed an authenticity condition for the transition from E to P, it is not actually entailed by the relevant Q. So, one may now be inclined to conclude, the impression that Dretske’s argument has given so many as being, if not an example of failure of closure, at least ­non-​­transmissive, is in a way correct: the argument does indeed fail to transmit justification from its premise to the conclusion that the animals in question have not been cleverly disguised. But it fails to do so because, so interpreted, it is invalid! However, having been persuaded of this reappraisal of Dretske’s example for the last decade or so since Aidan McGlynn first argued for it in a seminar at the Northern Institute of Philosophy in Aberdeen, 23 I have now24 come to think that it too is mistaken and that first thoughts were correct: for the impression that Zebras is indeed an instance of ­Transmission-​­failure is validated by the template for the phenomenon outlined in Section 2 above. Here’s why. There is nothing relevantly objectionable about representing the conclusion of Zebras as of the form, ~(­Fx & Gx), nor with converting that to the disjunction, ~Fx V ~Gx. Rather, very surprisingly, the thought that warrant for a proposition must constitute, ergo transmit, to warrant for any disjunction in which it is a disjunct is simply wrong. For suppose you are open minded about ~Fx V ~Gx. Then you should consider that you are in no position to exclude the possibility that that disjunction is false. But that is the possibility that each disjunct is false, so that Fx and Gx are both true. And that, in the case at hand, is the possibility that you are confronting cleverly disguised mules! But if you are o ­ pen-​­minded about that, you had better be open minded about the probative force of your visual evidence for the claim that they are zebras. Zebras can thus continue to enjoy its reputation as a masthead example of ­Transmission-​­failure.

5 Closure and Metaphysical “­Heavyweights” For any given cognitive project, where we aim to determine a view about P by scouting indirect evidence or bringing some direct cognitive capacity

182  Crispin Wright to bear, the authenticity conditions come in three broad categories. There will be some, first, which a properly responsible execution of the project will demand that we check independently and where failure to do so will be considered negligent and to compromise the significance of the results. Which conditions come into this category will often be in part a function of opportunity, stakes and interests: an untested assumption can be negligent in one context but permissibly taken for granted in another. Second, there will be, on the opposite side of the coin, what above I termed ­props—​­authenticity conditions which in context, it is considered rationally permissible to take for granted, and a check on which will be considered overly fastidious or even paranoid: demanding, for instance, wherever practicable, a character reference for any casual informant on some mundane matter, constantly checking ones watch against the Speaking Clock 25; poring over ­long-​­term analyses of the accuracy of Met office weather forecasts before being willing to plan a picnic on the basis of tomorrow’s forecast that it will be sunny and dry all day. Third, however, will be the metaphysical “­heavyweights” (­Hawthorne), the structural ‘­Hinges’, that feature in what I called, in earlier work, the ­I –­​­­I I–​­I II sceptical arguments illustrated in this table: Domain of Enquiry I E(vidence)

External World

Other Minds

The Past

Visual and Your gashed Recent proprioceptive shin and excavation evidence as twisted of a huge of a hand in ankle after fossilised front of my falling off reptilian face your bicycle skeleton II P Here is a hand You are A dinosaur died shocked and here many hurting millennia ago III Q There is an There are other The negation (Authenticity external minds of Russell’s Condition) world Five Minute Hypothesis26

Induction All Fs in a large random ­cross-​­sample of Fs prove to be G All F’s are G The World is inductively amenable

In each of the four columns there is at least a strong prima facie case that antecedent ­open-​­mindedness about Q would debar a rational subject with crediting the evidence of type E, variously illustrated, with probative force for any proposition of the type illustrated by the relevant P. If that is accepted, it follows that there is no getting to know the ­Q -​­propositions by the kind of inferential route indicated by the downwards vertical path in the respective columns, and h ­ ence—​­at least so the Sceptical argument ­runs—​­no getting to know them at all. How might we resist this conclusion?

Closure and Transmission Again  183 It can be resisted if it is possible, after all, to validate something like Dynamic Closure in full generality. If Dynamic Closure could be shown to hold, then we can rest content with ordinary standards of knowledge or justification for the ­P-​­type propositions as illustrated in the table and then appeal to Dynamic Closure to quash any inhibition about whether we can thereby know or justify the ­Q -​­propositions. Let me accordingly close by offering some reactions to Duncan’s Pritchard’s recent attempts27 to save Dynamic Closure in the face of the pressure exerted by the apparent counterexamples spawned by the Transmission Failure template schematised in Section 2. Pritchard is clear that his discussion concerns the rational formation of knowledge28 —​­a commendable departure from the cruder forms of knowledge externalism that have been prominent in the discussion of recent d ­ ecades—​­and that the question is accordingly how, in the awkward cases, a belief in Q can be rationalised on the basis of the evidence for P and the ­deduction—​­exactly what the intuition of ­Transmission-​ ­failure denies. His strategy is to offer different treatments for cases where the ­Q -​­proposition is not a metaphysically heavyweight hinge, taking Zebras as epitomising the latter type of case. Lest the reader feels that the status of that example remains controversial, notwithstanding the discussion in the previous section, I will here focus on the earlier example of Twins. E:  My

perception of a girl approaching who looks and dresses exactly like Jessica P:  That girl is Jessica Q:  That girl is not Jocelyn. Pritchard has two proposals for saving Dynamic closure in such a case. One is to construe the evidence concerned as factive, so that E becomes: E+:  My

perception of Jessica approaching.

The other strategy is to amplify the relevant evidence for P so that it includes sufficient background data to warrant Q: to let my evidence for P include, for example, the collateral information that Jocelyn is out of town today. Construing perceptual evidence as factive is, as noted earlier, the key thought of disjunctivist conceptions of perceptual justification. It is, as Pritchard acknowledges, highly controversial. I have no space here to evaluate further its promise as a weapon against perceptual scepticism. 29 The more immediate point is that it cannot plausibly generalise to prima facie T ­ ransmission-​­failure cases where the E ­ -​­place is occupied by defeasible evidence that P. For example, suppose I have misplaced my car keys and am going to be late unless I find them quickly:

184  Crispin Wright 5.1 Keys E:  Catherine tells me she thinks she saw them P:  My keys are on the shelf in the pantry Q:  Catherine is right on this occasion.

on the shelf in the pantry

Q is an authenticity condition for the transition from E to P: Open mindedness about Q will rationally require open mindedness about the evidential significance of E for P. But here there is no possibility of taking E to be factive—​­E can be true while P is false. What about the alternative of reconceiving my evidence for P to include background materials corroborative of Q? Perhaps Catherine has been able to bale out my forgetfulness many times in the past. Now I do indeed have evidence for Q, so a static version of Closure is comfortable with the example. But it is hard to see how amplifying the background evidence is this way addresses the concern about Dynamic Closure. For now the basis for my acceptance of Q is nothing to do with its deduction from E and P but is simply Catherine’s track record as a helpful St. Anthony, whereas Dynamic Closure requires a basis for accepting Q in the competent deduction. As in all relevant cases, the rationality of accepting P on the basis of E depends on prior acceptance of (­the likelihood) of Q. Dynamic Closure simply inverts, so misrepresents, the order of justification. What about cases when Q is metaphysically heavyweight? Here Pritchard takes an interestingly different tack. It is notable that there is again little if any prospect of construing the evidential bases in the columns for the Other Minds, The Past and Induction inferences schematised in the above table as factive.30 And amplification of background ­evidence—​­even prescinding from the misgiving just voiced about the relevance of the tactic in general in defence of dynamic ­closure—​­looks to be forestalled as well. When the issue is whether there are other minds at all, or whether the remote past was a reality, or whether the world continues, as in the past, to be inductively amenable, what background evidence might we possibly adduce? Pritchard’s ingenious response is to query whether, when it comes to “­H inges” of so fundamental and general a character, it is proper to think of our attitude of unconditional certainty as a species of belief. Dynamic closure, as he formulates it, talks about the knowledgeable status of a belief formed by competent deduction from a known proposition. If it is actually impossible to believe, properly so described, the conclusion of a competent deduction of a metaphysical heavyweight from a known n ­ on-​ m ­ etaphysical heavyweight, then the antecedent to Dynamic Closure, appropriately formulated, is, necessarily, never fulfilled in such a case. So the apparent counterexamples to the principle are finessed. What to make of this move? I have some sympathy for the idea that, as it features in ordinary thought, the notion of belief is somewhat generic,

Closure and Transmission Again  185 indifferent to the distinctions between religious faith, unevidenced trust, the kind of attitude that is generally reckoned to be an essential ingredient of knowledge, a­ nd—​­maybe the same t­hing—​­the kind of attitude that one may cautiously profess in lieu of claiming knowledge. In earlier work I proposed that our attitude to the heavyweight hinges should be regarded as one of unconditional but rational trust. That we know such propositions to be true is arguably an inappropriate idea if they indeed are out of reach of any probative cognitive achievement, lying “­apart from the route travelled by enquiry”, as Wittgenstein put it.31 In a similar vein, Pritchard proposes that we sharply distinguish the kind of attitudinal acceptance that is potentially a component of k ­ nowledge—​­what he terms ­K-​­belief—​­from other kinds of acceptance and contends that it is only and exactly K ­ -​­belief for which Dynamic Closure properly holds. That said, I would suggest that the attempt to protect Dynamic Closure in this way is vulnerable to a charge of creative accounting. At least it is so unless or until we have some independent account of the contrast between the two kinds of a­ ttitude—­​­­K-​­belief and the kind of acceptance we go in for with metaphysical H ­ inges— from ​­ which their differential behaviour with regard to Closure follows and is thereby independently explained. The fact is that a theorist who rejects Dynamic Closure on the kind of grounds rehearsed in the preceding is claiming that one cannot always earn a rational epistemic warrant for accepting a proposition by deducing it from something which one knows. It is no contradiction of the point of this claim to reply that, to the contrary, one can indeed always so acquire warrant for believing such a proposition, provided it is the kind of proposition which admits of belief in the first place. Absent some independent account of the relevant notion of belief and a justification for partitioning off the exceptions as ­non-​­believable, the claim that Dynamic Closure holds comes to little more than the claim that it ­holds—​­except when it doesn’t.

6 Conclusion It is not a good reason to maintain Dynamic Closure that we otherwise put in jeopardy our ability to advance our knowledge and justified beliefs by deduction. There is no such jeopardy provided we can corral and theorise the exceptions. The hypothesis I would propose is that they are one and all cases where the conclusion of a valid piece of reasoning encodes an authenticity condition for the warrant we take ourselves to have for one or more of the premises. When that is so, the reasoning ought rationally to be regarded as powerless to induce justified confidence in its stated conclusion since absent such confidence in advance, one rationally ought to lack confidence in the putative warrant for the premises. That, in a nutshell, is how and why transmission failure occurs,

186  Crispin Wright Cases where the premises are known are no exception to this possibility. Dynamic Closure, in full generality, is false provided knowledge of the conclusion of a valid inference from known premises is required to be rationally based thereon. Better learn to live with it.

7 Coda—​­Static Closure What of static closure principles, in particular Knowability Closure? This was formulated as follows: If P is knowable and may be known to entail Q, then Q also is knowable. It should be apparent that even this principle fails. One group of counterexamples will be cases where Q is both entailed by P and is a prop for the authenticity of the m ­ eans—​­evidence or cognitive p ­ rocess—​­whereby P may be known but which lies beyond all independent ­corroboration—​ ­the situation of the metaphysical heavyweights. Another theoretically possible type of counterexample will be where the authenticity condition, Q, again enjoys the status of a prop for the acquisition, by the relevant method in context, of knowledge that P but allows only of independent corroboration falling short of the strength required for knowledge. Dretske and Nozick did not orchestrate their respective discussions of knowledge closure around the distinction between static and dynamic versions. Nor, unless one believes that it is indeed constitutive of knowledge that it be counterfactually sensitive in their technical sense, did they offer cogent reason for rejecting either version of the principle. I conclude nevertheless that their instinct was sound. So long as knowledge by deduction is subject to a rational basis requirement, Dynamic Closure fails in full generality. And so long as knowledge generally is required to be the product of a specific cognitive achievement, Static Closure founders on the status of the Heavyweight Hinges.32

Notes 1 This volume, p. 8 2 This volume, p. 114. I’ll focus on Jope’s formulation in what follows. 3 Pritchard (­2016a), ­p. 14. 4 Page reference this volume, p. 8. 5 Page reference this volume, p. 8. 6 Wright (­2000) ­onwards — ​­see eight further entries in References. 7 John Pollock says “­undercutting”. Locus classicus: Pollock (­1974). 8 For ease of exposition, I will sometimes omit explicit reference to the “­faculty” case in the formulations to follow. 9 I conjecture, though I have no space to investigate the matter fully here, that this proposal has the resources to explain the intuitively convergent ideas about transmission failure put forward in their respective contributions by Yuval Avnur and Krista Lawlor in the present volume (­Avnur, pp. ­86–88; Lawlor, pp. ­30–33). Call any authenticity condition that for

Closure and Transmission Again  187 a certain range of cognitive projects is normally considered as something which may permissibly be taken for granted without special investigation (­if indeed investigation would be so much as possible), a prop. Then I think Avnur’s idea of certain kind of epistemic method’s being “­biased towards” the obtaining of a certain condition comes down to our treatment of that condition as a prop for investigations conducted by that method. Likewise, in consideration of Lawlor’s proposal in terms of relevant and irrelevant alternatives, I conjecture that in a wide range of cases, for a proposition Q, inconsistent with P, to be an irrelevant alternative to P for the purposes of an investigation whether P, is for its negation, ~Q, to be included among the props for the method of investigation concerned or the significance of the evidence it discloses. 10 And the same applies, naturally, to Pritchard’s Competent Deduction Closure. 11 Moore (­1939). 12 McKinsey (­1991). 13 Putnam (­1981), chapter 1. 14 Dretske (­1970, 2005). 15 Pryor (­2000). 16 I have responded to Pryor in Wright (­2007) and (­2012). 17 McDowell (­1994), p ­ . 26. 18 I am aware of course that Disjunctivists characteristically think they have the resources to address this issue. Their response is, roughly, that although BAD experiences can in principle be subjectively indistinguishable from normal perceptual experiences, an agent who is perceiving external material things in a normal way can, just on that account, know that they are, so that the sceptical argument transitions illicitly from “­A n agent in BAD need not be able to tell that they are” to “­A n agent in GOOD need not be able to tell that they are”. Compare, “­I f drunk, I need not be able to tell whether I am unfit to drive”, and “­I f sober, I will be able to tell whether I am unfit to drive”. There is strain in the comparison. Being drunk is phenomenologically, I am assured, nothing like normal sober consciousness. But being in one of those BAD states that is conceded to be phenomenologically indistinguishable from a possible episode of normal perceptual experience is a fortiori very much like such a normal episode! Yet although, while in the former, one would not be able to tell that one was, yet if in the latter, it would be clear that that was the situation. How? One way or another, what Disjunctivism needs to claim is that E in ­Ante-​ ­ oore is t­ endentious — ​­that E will be false if I am in GOOD: for, the claim M must be, it can be and normally is luminous to a normally perceptually functioning subject in GOOD that such is their situation. I have difficulty distinguishing that assertion from a Johnsonian argumentum ad lapidum. 19 Boghossian (­1997). 20 ­Brain-­​­­in-­​­­a-​­vathood a la Putnam is thus intended as a metaphor for the massive error and ignorance which metaphysical realism entails, on Putnam’s account, is metaphysically possibly our actual predicament. If it can be shown that, necessarily, we are not ­brain-­​­­in-­​­­a-​­vat, it is thereby intended to follow that we are, necessarily, in no such predicament, so that metaphysical realism is, necessarily, a misconception of the relationship between our thought and the real world. 21 A recent further discussion of the issues is Thorpe and Wright (­forthcoming). 22 —— ​­And in intuitionist logic too, if we take it that F and G are decidable. 23 See McGlynn (­2014), section 4 for his own expression of it.

188  Crispin Wright 4 ——​­nudged by Sven Rosenkranz. 2 25 Britain’s famous speaking clock was introduced in 1936 and still keeps perfect time for the 30 million people who use the service each year. 26 Namely, that the world did not spring into existence five minutes ago, albeit replete with apparent traces of a much more ancient history (­Russell, 1921, ­p. 159). 27 Pritchard (­2016a, 2016b) and this volume. 28 “­Finally … I want to highlight that the notion of knowledge that is in play here is rationally grounded knowledge. Depending on one’s wider epistemology, rationally grounded knowledge might well be the only kind of knowledge that there is”, this volume, p. 114. 29 See note 18 above. 30 Pace John McDowell who once wrote that. We should not jib at, or interpret away, the commonsense thought that, on those occasions that are paradigmatically suitable for training in the assertoric use of the relevant part of language, one can literally perceive in another person’s facial expression or his behaviour, that he is (­for instance) in pain, and not just infer that he is in pain from what one perceives. (­McDowell, 1978, ­p. 135) 31 On Certainty §88 (Wittgenstein, 1969). 32 Thanks to Sven Rosenkranz and to colleagues in the Stirling W ­ ork-­​­­in-​­Progress seminar for helpful comments and discussion, and to Matt Jope and Duncan Pritchard for giving me the opportunity to contribute to the present volume.

References Yuval Avnur (­ T his volume) “­ T he Skeptical Paradox and the Generality of ­Closure (­and Other Principles)”. Paul Boghossian (­1997) What the Externalist Can Know a Priori”, Proceedings of the Aristotelian Society 97 (­2) p­p. ­161–​­175. Fred Dretske (­1970) “­Epistemic Operators”, Journal of Philosophy 67 (­24) p­p. ­1007–​­1023. Fred Dretske (­2005) “­Is Knowledge Closed under Known Entailment? The Case against Closure” in Matthias Steup  & Ernest Sosa (­eds.), Contemporary Debates in Epistemology Oxford, UK: Blackwell, p­p. ­13–​­26. John Hawthorne (­2005) “­T he Case for Closure” in Matthias Steup & Ernest Sosa (­eds.), Contemporary Debates in Epistemology Oxford, UK: Blackwell, p­p. ­26–​­43. Matthew Jope (­T his volume) “­I ntuitive Closure, Transmission Failure and Doxastic Justification”. Krista Lawlor (­T his volume) “­Modest Foundations for Perceptual Knowledge”. Stephen Luper (­2016) ‘­Epistemic Closure’ in Zalta (­ed.), The Stanford Encyclopedia of Philosophy. URL = https://­plato.stanford.edu/­archives/­win2018/­entries/­­ closure-​­epistemic/. John McDowell (­1978) “­On ‘­T he Reality of the Past’” in C. Hookway & P. Pettit (­eds.), Action and Interpretation Cambridge: Cambridge University Press, p­p. ­127–​­144. ——— ​­(­1994) Mind and World Oxford: Oxford University Press. Aidan McGlynn (­2014) “­On Epistemic Alchemy” in D. Dodd  & E. Zardini (­eds.), Scepticism and Perceptual Justification Oxford: Oxford University Press, p­p. ­173–​­189.

Closure and Transmission Again  189 Michael McKinsey (­1991) “­­Anti-​­Individualism and Privileged Access”, Analysis 51 (­1) p­p. ­9 –​­16. G.E. Moore (­1939) “­Proof of an External World”, Proceedings of the British Academy 25 (­5) p­p. ­273–​­300. Robert Nozick (­1981) Philosophical Explanations Cambridge, MA: Harvard University Press. John Pollock (­1974) Knowledge and Justification Princeton, NJ: Princeton University Press. Duncan Pritchard (­2016a) Epistemic Angst Princeton, NJ: Princeton University Press. ———​­ (­2016b) “­Epistemic Angst”, Philosophical and Phenomenological Research 93 (­3) p­p. ­70–​­90. ——— ​­(­T his volume) “­I n Defence Of Closure”. James Pryor (­2000) “­T he Skeptic and the Dogmatist” Noûs 34 (­4) p­p. ­517–​­549. Hilary Putnam (­1981) Reason, Truth and History Cambridge: Cambridge University Press. Bertrand Russell (­1921) The Analysis of Mind London: G. Allen & Unwin. Joshua R. Thorpe & Crispin Wright (­Forthcoming) “­Putnam’s Proof Revisited” in Sanjit Chakraborty & James Conant (­eds.), Engaging Putnam Abingdon: Routledge, pp 63–88. Timothy Williamson (­2000) Knowledge and Its Limits Oxford: Oxford University Press. Ludwig Wittgenstein (­1969) On Certainty (­edited by G.E.M. Anscombe  & G.H. von Wright) New York and London: Harper Torchbooks. Crispin Wright (­2000) “­Cogency and Q ­ uestion-​­Begging: Some Reflections on McKinsey’s Paradox and Putnam’s Proof”, Philosophical Issues 10 p­p. ­140–​­163. ——— ​­(­2002) “(­­A nti-​­)­sceptics Simple and Subtle: G.E. Moore and John McDowell”, Philosophy and Phenomenological Research 65 (­2) p­p. ­330–​­348. ——— ​­(­2003) “­Some Reflections on the Acquisition of Warrant by Inference” in S. Nuccetelli (­ed.), New Essays on Semantic Externalism and ­Self-​­Knowledge Bradford: MIT, p­p. ­57–​­77. ——— ​­(­2004) “­Warrant for Nothing (­and Foundations for Free)?”, Proceedings of the Aristotelian Society, Supplementary Volume 78 p­p. ­167–​­245. ——— ​­(­2007) “­T he Perils of Dogmatism” in Susana Nuccetelli & Gary Seay (­eds.), Themes from G. E. Moore: New Essays in Epistemology and Ethics Oxford: The Clarendon Press, pp. 25–48. ——— ​­(­2011) “­McKinsey One More Time” in Anthony Hatzimoysis (­ed.), ­Self-​ ­Knowledge Oxford: Oxford University Press, p­p. ­80–​­104. ——— ​­(­2012) “­Replies Part IV: Warrant Transmission and Entitlement” in A. Coliva (­ed.), Mind, Meaning, and Knowledge. Themes from the Philosophy of Crispin Wright Oxford: Oxford University Press, p­p. ­451–​­486. ——— (­ ​­ 2014) “­On Epistemic Entitlement II: Welfare State Epistemology” in D. Dodd & E. Zardini (­eds.), Scepticism and Perceptual Justification Oxford: Oxford University Press, p­p. ­173–​­189.

Index

Alston, W. P. 39, 41 Austin, J. L. 41, 170 Avnur, Y. 3, 94, 95, 128, 145, 186, 188 Barnett, D. J. 94, 95 Bergmann, M. 109, 111 Blome-Tillmann, M. Boghossian, P. 177, 187, 188 Bonjour, L. 70, 75 bootstrapping 24, 113 brain-in-a-vat 5, 13, 34, 36–38, 44, 46, 60–62, 68–70, 77, 87–90, 97–99, 103–105, 107, 111, 119–121, 135–136, 143, 149–157, 159, 161, 165–166, 173, 178–180, 187 Brandom, R. 15, 95 Brown, J. 49, 56, 109, 111, 160, 161 Brueckner, A. 41, 94, 95, 148, 150, 156, 161 Christensen, D. 94, 95 Cohen, S. 20, 22, 24, 41, 94, 95, 126, 127, 128, 148, 151, 161 Coliva, A. 23, 42, 57, 83, 84, 87, 94, 95, 127, 128, 145, 189 Conee, E. 109, 111, 145, 146 conservatism 13, 21, 127 content externalism see semantic/ content externalism contextualism 22, 34, 71, 116, 126, 144, 161 contrastivism 28, 40, 143, 144 counterfactual conditionals 2–3, 58–64, 68–75 Davies, M. 94, 96, 111 defeat 2, 3, 43, 45–47, 50–53, 55–56, 101–103, 106, 108–109, 110, 111, 167–168

DeRose, K. 70, 75, 126, 128, 144, 146, 161 dogmatism 3, 26–27, 33–38, 40, 97, 99–103, 107–109, 110, 111, 126–127, 173–175 Dretske, F. 1, 4, 7, 9, 16, 20, 22, 24, 28, 30–31, 39, 41, 44, 55, 56, 58, 74, 75, 76, 82, 83, 95, 103, 109, 111, 113–115, 126, 128, 131–138, 140–144, 145, 146, 148, 154, 161, 165–166, 181, 186, 187, 188 Dummett, M. 94, 95 easy knowledge 1, 24 entitlement 1, 14–19, 21, 38, 83–84, 89, 94, 123 epistemological disjunctivism 4, 36, 118–122, 125, 127, 152–153, 159, 174, 187 evidentialism 78, 92, 93 Feldman, R. 20, 22, 111 Fine, K. 40, 41 Gemes, K. 40, 41 Gillies, A. 75, 76 Ginet, C. 109, 111 Goldberg, S. 56, 109, 111, 112, 129 Goldman, A. I. 94, 95, 159, 161 Gordon, E. 57 Graham, P. 56, 96, 111 Greenough, P. 42, 111 Harman, G. 100, 109, 111 Hawthorne, J. 8, 20, 22, 42, 75, 76, 94, 95, 96, 100, 109, 111, 126, 129, 144, 146, 154, 161, 164–165, 171, 182, 188 Hempel, C. 83, 94, 95

192 Index hinge epistemology 4, 20, 56, 83, 87, 122–125, 127, 128, 182–186 Holliday, W. 39, 41 Huemer, M. 94, 95 Jope, M. 2, 20, 21, 22, 56, 164–166, 171, 188 justification: doxastic 2, 7, 10–11, 14–20, 21, 39, 45–49; propositional 2, 7, 9–18, 20, 21, 29–30, 33, 39, 45–47, 77, 80 Kelp, C. 3, 22, 56, 107, 108, 109, 111, 155–156, 161 Klein, P. 41, 145, 146 Kratzer, A. 28, 39, 41 Kung, P. 94, 95 Lackey, J. 56, 57, 109, 111 Lange, M. 94, 96 Lasonen-Aarnio, M. 109, 111 Leite, A. 36–37, 41 Lewis, D. 39, 40, 41, 58–62, 68, 71, 74, 75, 76, 126, 129 Liberalism 2, 34, 40 Logue, H. 160, 161 Luper, S. 8, 22, 70, 75, 76, 164–165, 188 Lyons, J. 24, 38, 41, 56 Markie, P. 111, 112 Markosian, N. 159, 161 McDowell, J. 23, 57, 127, 129, 141, 146, 159, 162, 187, 188, 189 McGinn, M. 127, 129 McGinn, C. 145, 146 McGlynn, A. 22, 40, 41, 181, 187, 188 McKinsey, M. 175–179, 187, 188, 189 Millar, A. 145, 146 Millikan, R. 56, 57 Moore, G. E. 2, 6–7, 13, 15, 19, 22, 23, 24–27, 32–33, 37, 40, 42, 43–50, 53, 57, 76, 109, 112, 128, 146, 173–175, 178, 187, 189 Mooreanism/Neo-Mooreanism 2, 6–7, 9, 18–20, 26–27, 32–37, 38, 43–55, 89–90, 127 Moretti, L. 10, 12, 20, 21, 22, 39, 42 Moss, S. 75, 76 Moyal-Sharrock, D. 127, 129 Nagel, T. 145, 146 Neta, R. 34, 39, 42

Nozick, R. 3, 58–60, 62, 70, 74, 75, 76, 103, 109, 112, 129, 144, 146, 148, 154, 155–156, 162, 165–166, 186, 189 Nuccetelli, S. 39, 42, 57, 111, 112, 189 Pedersen, N. 84, 94, 96, 111 Piazza, T. 10, 12, 20, 21, 22, 39, 42 Pollock, J. 56, 57, 145, 146, 167, 186, 189 Pritchard, D. H. 3–4, 5, 8, 20, 22, 42, 56, 57, 87, 94, 95, 96, 109, 111, 112, 126, 127, 128, 129, 130, 144, 145, 146, 148–149, 151, 152–156, 159, 160, 162, 164–165, 171, 183–185, 186, 187, 188, 189 Pryor, J. 25–26, 38, 39, 42, 43, 45–50, 54, 57, 94, 96, 109, 111, 112, 127, 129, 145, 146, 187, 189 Putnam, H. 96, 178–179, 187, 189 Ramsey, F. 64, 76 Rationalism 34 relevant alternatives 4, 28–33, 35, 37, 40, 115–116, 126, 142, 145, 157–158, 160 Roush, S. 70, 76 Russell, B. 188, 189 safety 3, 21, 58, 60–62, 71–74, 75 Schaffer, J. 39, 40, 42, 143, 146 Schiffer, S. 40, 41, 42 Schnurr, J. 57 Schonbaumsfeld, G. 4, 127, 129, 145, 146 semantic/content externalism 5, 79, 147, 173, 175–180 sensitivity 3, 30, 58–62, 69, 71–72, 74, 103–104 Shope, R. 74, 76 Silins, N. 10, 20, 21, 22, 34, 39, 42, 94, 96 Simion, M. 2, 50, 56, 57, 109, 112 Smith, M. 4, 22, 57, 157, 159, 161, 162 Sorensen, R. 109, 112 Sosa, E. 22, 41, 60, 76, 95, 96, 109, 112, 128, 129, 146, 188 Stalnaker, R. 58–62, 68, 71, 74, 75, 76 Steup, M. 22, 41, 76, 95, 96, 128, 129, 146, 188

Index  193 Stevenson, L. 127, 129 Strawson, P. F. 94, 96 Stroud, B. 94, 96, 127, 129, 145, 146, 148, 162 Thorpe, J. R. 187, 189 transmission 5, 6–7, 8–20, 21, 24–27, 30–34, 37–38, 39, 40, 44–45, 50, 56, 89–90, 94, 97–104, 107–109, 110, 111, 142, 145, 165–166, 169, 171, 173–176, 180–181, 183, 185, 186 Tucker, C. 10, 22, 39, 42 Turri, J. 21, 22, 76 underdetermination 4, 46, 92–93, 127, 148–154, 156, 159, 160 Unger, P. 126, 129 Vahid, H. 126, 130 Veber, M. 109, 112

Vogel, J. 24, 42, 79, 94, 96, 148, 155, 162 von Fintel, K. 75, 76 Weisberg, J. 94, 96 White, R. 94, 96, 145, 146 Williams, M. 127, 130, 146 Williamson, T. 2, 3, 8, 20, 23, 42, 43, 44, 54, 55, 57, 60, 64, 66, 67, 71, 75, 76, 94, 96, 109, 112, 126, 130, 164–165, 171, 189 Wittgenstein, L. 123, 127, 128, 129, 130, 185, 189 Wright, C. J. G. 42, 44, 57, 75, 76, 83–85, 86, 89, 94, 96, 104, 109, 112, 127, 130, 144, 145, 146, 157, 158, 162, 186, 187, 189 Yablo, S. 40, 42 Yalcin, U. 148, 162 Yamada, M. 94, 95 Ye, R. 109, 112