Current Issues in Theoretical Philosophy: Volume 2 The Externalist Challenge 9783110915273, 9783110183061

Citation preview

The Externalist Challenge

Current Issues in Theoretical Philosophy Edited by Richard Schantz

Vol. 2

w DE

G

Walter de Gruyter · Berlin · New York

The Externalist Challenge Edited by Richard Schantz

w DE

Cl

Walter de Gruyter · Berlin • New York

© Printed on acid-free paper which falls within the guidelines of the ANSI to ensure permanence and durability.

ISBN 3-11-018306-4 Bibliographic information published by Die Deutsche Bibliothek Die Deutsche Bibliothek lists this publication in the Deutsche Nationalbibliografie; detailed bibliographic data is available in the Internet at < http://dnb.ddb.de >.

© Copyright 2004 by Walter de Gruyter GmbH & Co. KG, D-10785 Berlin All rights reserved, including those of translation into foreign languages. No part of this book may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopy, recording or any information storage and retrieval system, without permission in writing from the publisher. Printed in Germany Data conversion: DTP Johanna Boy, Brennberg Printing and binding: Hubert & Co., Göttingen Cover design: Christopher Schneider, Berlin

Contents Introduction

1

I In Defence of Epistemic Externalism The "Challenge" of Externalism William P. Alston

37

Externalism and Skepticism John Greco

53

Counterexamples to Epistemic Externalism Revisited Thomas Grundmann

65

Conditions on Cognitive Sanity and the Death of Internalism Hilary Kornblith

77

Empiricism Externalized Richard Schantz

89

Circularity and Epistemic Priority Ernest Sosa

113

II Critiques of Epistemic Externalism The Chimerical Appeal of Epistemic Externalism Joe Cruz and John Pollock

125

In Search of Internalism and Externalism Richard Feldman

143

Inferential Internalism and the Presuppositions of Skeptical Arguments Richard Fumerton

157

Knowledge, Justification and the Cooperative World Keith Lehrer and David A. Truncellito

169

The Epistemological Promise of Externalism Barry Stroud

181

VI

Contents

Is Knowledge a Natural Phenomenon?

193

Michael Williams III In Defence of Content Externalism Social Cartesianism

213

John Haugeland Existence Proof for a Viable Externalism

227

Ruth Garrett Millikan Externalism, Epistemic Artefacts and the Extended Mind

239

Kim Sterelny Outing the Mind—A Teleopragmatic Perspective

255

Robert Van Gulick IV Critiques of Content Externalism Natural Intentionality

287

John Heil Phenomental Intentionality and the Brain in a Vat

297

Terence Horgan, John Tienson and George Graham On an Argument from Properties of Words to Broad Content

319

Frank Jackson Reference, Causal Powers, Externalist Intuitions and Unicorns Gabriel M. A. Segal

329

V An Exemplary Debate about Content Millikan's (Un?)Compromised Externalism

347

Georges Rey Comments on "Millikan's (Un?)Compromised Externalism"

Ruth Garrett Millikan

361

VII VI Self-Knowledge Believing that You Know and Knowing that You Believe Sven Bernecker

369

McKinsey Redux? Anthony Brueckner

377

Knowing What You Think vs. Knowing that You Think It Fred Dretske

389

Do We Know how We Know Our Own Minds yet? Pierre Jacob

401

What Was I Thinking? Social Externalism, Self-Knowledge, and Shifting Memory Targets Peter Ludlow

419

Anti-Individualism and Minimal Self-Knowledge: A Dissolution of Ebbs's Puzzle Brian P. McLaughlin

427

VII The Epistemic Significance of Perception Explaining Perceptual Entitlement Christopher Peacocke

441

Externalism and Disjunctivism James Van Cleve

481

VIII An Essay on Intentionality The Objects of Intentionality Colin McGinn

495

Index of Subjects

513

Index of Names

515

Contributors

517

Introduction This book is the second of a series of three volumes dedicated to central debates in contemporary theoretical philosophy. The title of volume I is What is Truth?, the title of volume III will be Prospects for Meaning. The debate between internalism and externalism has become a focal point of attention both in epistemology and in the philosophy of mind and language. Externalism challenges basic traditional internalist conceptions of the nature of knowledge, justification, thought and language. What is at stake, is the very form that theories in epistemology and the philosophy of mind ought to take. This volume is a collection of new essays reflecting the present state of the art concerning the exciting controversies between internalism and externalism.

The Debate between Internalism and Externalism in Epistemology Epistemic Internalism I will start with some comments on the internalism/externalism distinction as it arises in epistemology. Historically, from Plato to the present day, most theories of knowledge have been internalist in character. The traditional or standard analysis of knowledge holds that knowledge is justified true belief. Since, according to this analysis, justification is a necessary condition of knowledge, epistemologists have made every endeavour to develop theories of what epistemic justification or rationality consists in. The rationale for thinking that justification is necessary for knowledge is that merely accidentally true belief, beliefs held for bad reasons or no reasons at all, do not amount to knowledge. Justification has been the centre-piece of the theory of knowledge. And it is the central claim of internalism that typical epistemological questions such as "What is knowledge?", "What can we know?", "How can we be sure that our beliefs are justified?", "Can we be absolutely certain that we know anything at all?" and "Does knowing entail knowing that one knows?" arise from within a person's subjective cognitive perspective, and can only be adequately answered by referring to resources accessible to that person from that perspective. The modern era in philosophy, with its characteristic epistemological turn, was initiated by Descartes. The most conspicuous feature of his approach to epistemological problems was its radically internalist orientation. The aim of his famous "method of doubt" is to subject all our beliefs and principles to radical scrutiny, to test them to the limit, and to accept, finally, only those beliefs that, logically speaking, cannot be doubted. By applying a series of elaborate sceptical techniques, he intends to eliminate all dubious beliefs in order to find the absolutely certain ones, those that stand up to even the most extreme sceptical attack. According to Descartes, there are two kinds of

2

Introduction

beliefs that survive the process of systematic doubt, that are infallible, guaranteed to be true, and thus can count as knowledge: firstly, the belief that he himself exists as a thinking self, and, secondly, equally important, the beliefs he has about the specific contents of his own sensory experiences and other mental states. It is these latter beliefs that form the foundation, the empirical basis, of all other kinds of knowledge, including above all knowledge of the external world. So Descartes' foundational project is an ambitious attempt to reconstruct knowledge from the inside outwards—knowledge of our sensory experience is the starting point from which knowledge of the external world has to be inferred. His epistemological internalism comes to light in his principal assumption that only mental contents can function as the immediately accessible evidence that is requisite to provide a secure basis for knowledge. Justification is a matter of things purely internal to the subject's mind, a matter of things of which the subject can be justifiably aware of. Many philosophers still follow Descartes' lead. His resolute first-person perspective on the nature of epistemological problems has become part of our theoretical heritage. Here is a more modern statement of the internalist point of view by Roderick Chisholm, one of the most important epistemologists of the last century: The usual approach to the traditional questions of theory of knowledge is properly called "internal" or "internalistic". The internalist assumes that merely by reflecting upon his own conscious state, he can formulate a set of epistemic principles that will enable him to find out, with respect to any possible belief he has, whether he is justified in having that belief. The epistemic principles that he formulates are principles that one may come upon and apply merely by sitting in one's armchair, so to speak, and without calling for any outside assistance. In a word, one need consider only one's state of mind. 1

So what is required for a belief to be epistemically justified for a given person is for that person to have cognitive armchair access to evidence that supports the truth of the belief. What can serve for justification must be something like a belief or an experience, something of which the person has a cognitive grasp. The requirement that all justifying factors must be purely internally available can be interpreted in at least two ways: a strong position requires that the subject actually have the knowledge or justifiable belief that the justifier obtains, whereas a weaker version requires only that the subject be capable of acquiring this knowledge by appropriate focussing of attention, without any change of position, new information, and so on. The central idea is that by reflection or introspection alone we can determine whether a certain belief is justified or not. It is crucially important to distinguish between access to the justifier and access to the epistemic efficacy of the justifier. In order for a person to know or justifiably believe that p, it is not enough that she has access to other justified beliefs or other justifying factors, such as sensory experiences, providing her with adequate evidence for the belief that p. She must also know or justifiably believe that these other justifying factors are indeed adequate reasons or grounds for her belief that p. But it is far from obvious that the satisfaction of such severe requirements can be ascertained just on re-

1

Chisholm 1989, 76

Introduction

3

flection. This distinction opens up the possibility of advocating a mixed position, one demanding the cognitive accessibility of the ground on which a belief is based but not of its evidential adequacy. No doubt, internalism possesses a strong intuitive plausibility. How could something that does not fall within a person's ken, something to which perhaps she has no access at all, give her a reason for one of her beliefs? Internalists typically lay stress on having reasons for our beliefs, and this is typically interpreted as being able to use those reasons as premises in arguments, or to appeal to them when questioned. This is not something one can be unaware of. It has been argued, however, that part of internalism's intuitive appeal rests on a confusion between having reasons and giving reasons or, in other words, between the state or property of being justified in believing, and the activity of justifying the belief, of showing it to be justified. If one neglects this important distinction, one will be inclined to suppose that being justified entails the ability of showing that one is justified, that is to say, of producing a justifying argument. Now, of course, to successfully carry out the activity of justifying a belief one must appeal to other beliefs or other conceptually contentful states. But from the fact that we can justify a belief only by showing that it sustains appropriate relations to other beliefs, it does not follow that a belief can be justified only by its relations to other beliefs. We would have only few justified beliefs, if their justification depended on having engaged in the practice of establishing their credentials.2

Epistemic Deontologism The main motivation for the accessibility constraint of internalism derives from a deontological conception of justification, which has its origin in the works of Descartes and Locke. Descartes, in the fourth of his Meditations, explains error as the wrong use of the will in judging. He argues that the will is not limited by the perceptions of the understanding, so that we are free to choose, by a bare act of the will, whether or not to accept a proposition. The policy he puts forward is that we should only believe those propositions whose truth we clearly and distinctly perceive. Erroneous beliefs result when the limitless will exceeds the limits of the understanding, that is, when one attaches assent to a proposition ρ in the face of the recognition that one's evidence for p is less than adequate. We are fully responsible for these misuses of our intellectual freedom and may, therefore, justly be blamed. The virtuous course in these cases is to suspend judgement on p. Similarly, Locke, in his An Essay concerning Human Understanding, speaks of epistemic duties that we, as "rational creatures", have to meet and for the violation of which we are held accountable. In his view, our primary epistemic goal is to believe what is true and not believe what is false. This goal imposes on us the duty to seek the truth to the best of our intellectual faculties.

2

See Alston 1989

4

Introduction

In analogy to ethical deontologism which maintains that we have duties to act in certain ways and that we are responsible for our actions, epistemic deontologism maintains that we, as intellectual beings, have duties to believe on the basis of adequate evidence and that we are responsible for our beliefs. According to this conception, a person is justified in believing that p if and only if in holding the belief she satisfies, or does not violate, any intellectual duties or obligations. Epistemic justification is a normative notion, pertaining to what we should or should not believe. So it seems that the focus has to be on our epistemic responsibility, on whether or not a belief has been responsibly formed or is responsibly held. We are blameworthy if we do not live up to appropriate standards of epistemic conduct, if, for example, we base our judgment on misleading evidence and so acquire commitments that we should not hold. From the standpoint of epistemic deontologism, a justified belief is a belief which, by reference to our epistemic duties, it is permissible to hold; we cannot rightly be reproached for holding it. 3 But, so internalists argue, since a belief is permissible only if we have a justification for it, we must have access to the conditions that determine whether our beliefs are justified. Otherwise a person could do her epistemic best, could do as demanded of her, and hence be praiseworthy but still hold unjustified beliefs. Deontological conceptions of justification seem to presuppose a kind of doxastic voluntarism. As John Pollock pointedly puts it: "I have taken the fundamental problem of epistemology to be that of deciding what to believe." 4 And indeed, in order for talk of epistemic duties, responsibilities and permissions to make sense, we must have direct voluntary control over our beliefs. After all, I cannot have a positive duty to believe a proposition or a negative duty to refrain from believing it if it is not up to me to decide as to whether to believe it or not. We often hear that what a person ought to do, she can do. T h e dependence on doxastic voluntarism, however, brings to light a serious problem for epistemic deontologism. T h e trouble with doxastic voluntarism is that, as has been pointed out by William Alston and Alvin Plantinga above all, our beliefs are not generally under our direct voluntary control. 5 Although the relationship between the will and belief is a complex affair, I think it is fair to say that, at least for the most part, we do not choose our beliefs. We simply cannot believe or refrain from believing at will. Think of beliefs resulting from perception. When I see a cat on the sofa, and have no reason to suppose that the conditions of perception are abnormal, I am simply not free as to whether to believe that there is a cat on the sofa. I will form that belief, whether I like it or not. Deliberate choice does not play any role; a basic act of will is typically not involved. Rather, acquisitions of perceptual belief exhibit a characteristic involuntariness and passivity; they are events in which an external state of affairs forces itself on us. Or reflect on introspective beliefs. When I believe that I have a toothache, I cannot, by performing an act of the will, just choose to stop believing this. As long as I have the toothache, I will not succeed in getting rid of this belief, no matter how hard I try. Finally, a consideration of beliefs based

3 4 5

See Ginet 1975 Pollock 1986, 10 Alston 1989; Plantinga 1993a

Introduction

5

on reasoning also confirms the verdict that we cannot simply obtain beliefs through consciously willing to have them. If we accept the premises of a valid deductive argument, we must, are forced to, accept its conclusion too—regardless of whether we wish it to be true. None of these beliefs can be subject to epistemic duty. What has been criticized is the view that, generally, our beliefs are directly subject to effective voluntary control. This, however, is not to deny that we have some kind of indirect control over our beliefs or world-views. For we have voluntary control over many activities that influence their generation. Many of our beliefs are the results of explicit decisions as to how to conduct inquiry. It is up to us whether we investigate matters thoroughly, try to reason carefully, pay attention to evidence on both sides of an issue, seek the advice of other people, and so on. And we even have some kind of control over things that affect perceptual beliefs; even perception of the world involves an active element. We can try to make sure that the conditions of perception are optimal, direct our perceptual systems, focus on certain aspects, abstract from others, and the like. In these indirect, derivative senses, we are responsible for the beliefs we finally arrive at. But such an indirect control over our beliefs is not what the epistemic deontologist needs in order to support an internalist conception of justification.

Arguments against Epistemic Internalism The main difficulties of epistemic internalism can be traced back to the accessibility requirement. Its advocates seem to overestimate the cognitive powers of armchair reflection. We have seen that Chisholm claims that just by carefully thinking about it, we are capable of finding out what provides a justification for what. How questionable this claim is, becomes obvious when we look at the alternative non-deontological account of justification that is usually called a "truth-conductivity" conception. According to this widely held conception, justification is essentially or internally related to the cognitive aim of truth, to the aim of maximizing truth and minimizing falsity. Thus a person is justified in believing that ρ only if she is in a strong position to attain the truth, only if she has an adequate reason to believe that ρ is true. An adequate reason is, on this conception, a kind of truth indicator, something that makes it evidentially at least very probable that the belief that ρ be true. O n this account, a belief's being justified is a normative status, but the normativity in question has nothing to do with responsibility, duty, or permission. It is simply a matter of being appraised as good or bad, favorable or unfavorable, with respect to the basic epistemic end of seeking true beliefs and avoiding false ones. Here is how Lawrence Bonjour, interestingly an unwavering defender of internalism, brilliantly formulates and defends the truth-conductivity conception of justification: And, if our standards o f epistemic justification are appropriately chosen, bringing it about that our beliefs are epistemically justified will also tend to bring it about . . . that they are true. If epistemic justification were not conducive to truth in this way, if finding epistemically justified beliefs did not substantially increase the likelihood o f finding true ones, epistemic justification would be irrelevant to our main cognitive goal and o f dubious worth. It is only if we have some reason to think that epistemic justification constitutes a path to truth that

6

Introduction

we as cognitive human beings have any motive for preferring epistemically justified beliefs to epistemically unjustified ones. 6

If, however, conditions of justification have to be such that the fulfilment of those conditions guarantee that the belief in question is at least likely to be true, then the idea that we can determine by mere reflection whether certain conditions justify a given belief seems to lose its original intuitive appeal. According to familiar versions of coherentism, for example, epistemic justification is a holistic notion; an individual belief is justified only if it coheres, or harmonizes, in an appropriate way with a person's belief system, that is, with the totality of her beliefs. Since it is psychologically impossible for an agent exactly to survey the set of all of her beliefs, even coherentism seems to be unable to meet the accessibility constraint. The central problem for epistemic deontologism is that a belief might enjoy deontological justification without being based on a truth-conducive ground. In face of these difficulties it is small wonder that more modest versions of internalism have been proposed, ones which, while still holding that justification supervenes upon internal psychological conditions, drop the further requirement that these conditions must be reflectively accessible.7 Thus, modest internalism opens up the possibility that the facts on which the justification of a belief depends include, e.g., neural events proximate to the belief in question. But whether or not internal facts of these sorts obtain can hardly be determined by introspection alone. So this is indeed a modest or weak version of internalism—more defensible, surely, than accessibility internalism but without the latter's original intuitive appeal.

Epistemic External ism Pure externalist views deny that justification, at least in the customary sense, is a necessary condition for knowledge. Thus, they reject the traditional justified true belief account of knowledge. Advocates of externalist accounts of knowledge draw attention to features of the world other than the subject's reasons for belief. They urge that not all true beliefs need to be supported by reasons, classically conceived, in order to qualify as knowledge. Hence their view looks outward to the facts the beliefs are about, rather than inward to the reasons a subject has for holding her beliefs. David Armstrong, who introduced the term "externalism" as a terminus technicus in epistemology in his pioneering book Belief, Truth and Knowledge of 1973, characterizes this position in the following way: According to "Externalist" accounts of non-inferential knowledge, what makes a true noninferential belief a case of knowledge is some natural relation which holds between the beliefstate, Bap, and the situation which makes the belief true. It is a matter of a certain relation holding between the believer and the world. It is important to notice that [...] Externalist

6 7

Bonlour 1985, 8 Feldman and Conee 1985

Introduction

theories are regularly developed as theories of the nature of knowledge generally ply as theories of non-inferential knowledge. 8

7 and not sim-

The core of the non-justificational analysis of knowledge Armstrong here proposes is that knowledge merely requires some sort of natural relation between a person's belief and the fact which makes it true. A good example is Alvin Goldman's early causal theory of knowledge, developed primarily as a response to the Gettier problems, according to which a person's true belief that ρ counts as knowledge just in case it has the right sort of causal connection to the fact that p? The sorts that are right, or appropriate, comprise perception, memory, good reasoning and various other kinds of causal chains, and combinations of these. In Gettier examples, a person has a justified belief that is only accidentally true. The causal theory explains the lack of knowledge in these cases, because in them the basic requirement that there be an appropriate causal connection between the fact that ρ and the person's belief that ρ is not satisfied. Standard cases of perception provide a useful model of the right sort of causation. Fred Dretske, refining earlier accounts by explicating the causal relation in terms of the reception of information from a source, developed an influential information-theoretical analysis of perceptual knowledge on which a subject S knows that Λ: is F just in case her belief is caused or causally sustained by the information that χ is F received from the source χ by S.10 Goldman, originally, regarded the causal criterion as both a necessary and sufficient condition for knowledge. He did not just suggest supplementing the justification condition with his causal condition. For Goldman, then, the justification condition had to be replaced with an external standard—an appropriate causal relation to the external world. Later, Goldman came to advocate a related theory, the reliability theory of knowledge, which Armstrong had favoured from the beginning. 11 Remarkably, it was Frank Ramsey who first defended a reliability theory of knowledge. 12 Theories of this kind abandon the prerequisite of earlier, and simpler, causal theories that there be an actual causal chain linking the fact represented in the belief with the belief itself. Instead, proponents of the reliability account typically hold that a belief qualifies as knowledge just in case it is both true and empirically reliable, that is, just in case the belief is produced in a way that makes it objectively probable that the belief be true. There are two main types of reliabilism: reliable indicator theories and reliable process theories. According to theories of the first type, a belief counts as knowledge when it is a reliable indicator of its truth. According to theories of the second type, a belief counts as knowledge just in case it is produced, or is sustained, by a generally reliable process, that is, one that leads to a high proportion of true beliefs. 13 Armstrong's account belongs to the first type. He argues that the tight connection between a person's belief and the fact making it true which knowledge demands should be construed as a nomological connection. To illustrate his point, he draws an analogy

8

Armstrong 1973, 157

9 10 11 12

Goldman 1967 Dretske 1981 G o l d m a n 1979; Armstrong 1973 Ramsey 1931

13

G o l d m a n 1976, 1979; see also Talbott 1990

Introduction

8

between a thermometer which reliably indicates the temperature and a belief which reliably indicates the truth. His claim is that a belief about a perceived object x, that it is F, is non-inferential knowledge if and only if the belief is a completely reliable sign that χ is F. This depends on properties of the believer—e.g. that her senses and beliefforming mechanisms are functioning properly in a suitable context—such that lawlike connections of nature determine that, for any subject S and object x, if S has those properties and believes that χ is F, then χ is F. Similarly, Dretske maintains that the reasons we have for holding our beliefs should be nomologically connected to their truth, that is, it should be a law of nature that a person having such reasons for her belief will have a true belief. Since lawlike statements, in contrast to statements of mere uniformity, support subjunctive or counterfactual conditionals, the central idea of Dretskes account can also be cast in the subjunctive mode. Arguing that the possession of "conclusive reasons" to believe is both a necessary and a sufficient condition for knowledge, he goes on to characterize conclusive reasons by using a subjunctive construction: R is a conclusive reason for ρ if and only if R would not be the case unless p were the case. 14 A related proposal was made by Robert Nozick who claims that knowledge is true belief that "tracks the truth". Nozick also uses various subjunctive or counterfactual conditionals to explicate the specific sense in which knowledge is belief that co-varies with the facts and is thereby distinguished from accidentally true belief. He suggests, firstly, a "variation" condition: if ρ were not true, then S would not believe that p. But he argues that, secondly, a further clause, an "adherence" condition, must be added: if, in slightly different conditions, ρ were still true, then S would still believe that p. A variant of the subjunctive account of knowledge, the theory of relevant alternatives, holds that S knows that ρ only if there is no relevant alternative situation in which p is false but S would still believe that p. So in order to know that p, S must be able to discriminate the actual situation in which ρ is true from all relevant counterfactual situations in which ρ would be false. 15 Goldman, later, came to adopt a reliable process theory. In contrast to Armstrong's and other externalists' estimation of reliability theories, though, Goldman now introduced an important new perspective. Instead of rejecting the traditional view that knowledge demands justified true belief, he now argues for the alternative that we should think of reliability not as replacing justification but, rather, as providing a new analysis, a naturalistic reduction, of what justification amounts to. Since, for him, justification is by definition that which has to be added to true belief to convert it into knowledge, justified belief just is reliably produced belief. Similarly, Marshall Swain, presenting a version of reliabilism in terms of probability, holds that a belief is justified when it is based on reasons that are reliable indicators of the truth. 1 6 A crucial feature of reliabilism is that it does not require that the person have any sort of cognitive access to the fact that the belief-producing processes are reliable in

14 15 16

Dretske 1971 Goldman 1976 See Swain 1981

Introduction

9

order for her belief to be justified. All that matters is that the actual causal process by means of which the belief in question is generated is in fact reliable, whether or not the believer, or anyone else, has at the time, or indeed ever, any beliefs about the fact of reliability That is why reliabilism is a form—indeed the most influential form—of externalism: it makes the epistemic status of a belief depend on external factors, factors concerning the believer's environment, history or social context to which she may have no cognitive access at all. Against this background, externalists tend to warn of a level confusion, the confusion between knowing, or being justified in believing, that p—first-level knowledge or justification—and knowing, or being justified in believing, that one knows, or is justified in believing that ρ—second-level knowledge or justification. They insist that unreflective first-level knowledge does not require that the believer knows anything about her own reliability. She knows just in case her belief is de facto truth-reliable. O f course, if she has no beliefs on the reliability of her cognitive mechanism, she lacks something. And what she lacks, is second-level knowledge or justification: she does not know, or justifiably believe, that her belief is justified or amounts to knowledge. No wonder then that externalists generally dismiss the KK-thesis, the thesis saying that if a person knows that p, then she knows that she knows that p, which was tacitly or explicitly accepted by many philosophers since Plato and Aristotle. An offshoot of reliabilism is Alvin Plantinga's comprehensive theory of warrant and proper function. 17 Plantinga emphatically repudiates epistemic internalism and with it the whole idea of justification, replacing it with the concept of warrant which he too characterizes as the quality that when added to true belief yields knowledge. Purporting to go beyond reliabilism, this theory does not chiefly focus on reliable processes but rather on cognitive faculties that function according to a design plan in a suitable environment. A belief is warranted only if it is produced by a properly functioning faculty in a cognitive environment that is fitting for this kind of cognitive faculty— the environment must be the one or like the one for which it is designed. Either God or natural selection designed us to function successfully in our surroundings. On the other hand, Plantinga critically examines and rejects, sophisticated as they may be, the leading naturalist accounts of proper function that have been proposed. He argues that the notion of function demands the notion of design, and design in turn demands something non-natural. Hence, ultimately, his whole intricate theory calls for some theistic or supernatural underpinning. The design plan turns out to be the divine plan. Plantinga himself confesses to be a naturalist in epistemology, but a supernaturalist in anthropology.18 Yet another kind of reliabilism is presented by virtue epistemology and its attempt to make epistemic character traits, rather than epistemic principles or duties, the centre of epistemology. The leading figure of this approach to date is Ernest Sosa. The basic idea of his "virtue perspectivism" is that justified belief is belief that arises from the proper functioning of our intellectual virtues in an appropriate environment, and

17 18

Plantinga 1993a, 1993b Plantinga 1993b, 46

Introduction

10

not from intellectual vices. 19 An intellectual virtue is a competence or faculty or power of the mind to believe true propositions and to avoid believing false ones. So the virtues are truth-conducive, whilst the vices are inimical to truth. Human epistemic virtues are reliable cognitive mechanisms such as perception, memory, introspection, deduction and induction, whereas vices include wishful thinking, hallucination and clairvoyance. In contrast to generic reliabilism, virtue perspectivism restricts the class of reliable processes that are epistemically significant to those processes that have their source in an epistemic virtue, in a stable disposition for belief acquisition. Moreover, Sosa distinguishes between "animal knowledge" and "reflective knowledge". The latter higher sort of knowledge, the sort of knowledge traditionally desired, requires not just a virtuously functioning faculty, but, additionally, that the subject enjoys an appropriate epistemic perspective on her first-level beliefs, that she grasps the reliability of her faculties. Finally, virtue perspectivism differs from Plantinga's theory by eschewing the contentious notion of a design plan for our epistemic faculties.

Advantages of Epistemic Externalism Externalism holds many attractions for epistemologists. First of all, it offers a scientifically respectable account of knowledge. Many of its proponents contend that externalism is a form of epistemology naturalized and so follow Willard Van Quines advice to look upon epistemology as an enterprise within the framework of natural science, that is, to transform epistemology into a chapter of empirical psychology. The prime goal of naturalized epistemology is to answer the question of how we acquire our overall theory of the world. By adopting a genetic approach, Quine and his followers seek to provide a factual account of the link between sensory evidence and theory. In this undertaking, cognitive agents are viewed as natural phenomena, and so the focus of the scientific investigation of the acquisition of knowledge is on the causal relations between their sensory input and their theoretical output. Moreover, externalism is naturalistic in that it tries to reduce normative epistemic properties to natural properties and relations, the paradigm case being the attempted reduction of justification to reliability. And indeed the study of psychological processes and mechanisms of belief formation and alteration seems to be a descriptive and not a normative affair. Secondly, externalism seems to be capable of protecting knowledge against the threats of radical scepticism. Instead of setting extremely high standards for knowledge, externalism proposes more moderate constraints. True beliefs that arise via a reliable method are knowledge because of the way they arise. The subject need not, in addition, know that they have arisen in that way rather than in the ways radical scepticism suggests. It is a result of this modesty that we can quite plausibly attribute simple factual knowledge to young children and even to animals, as we indeed usually do, without having to assume that they are in possession of sophisticated knowledge or beliefs about the epistemic status of their beliefs.

19

Sosa 1991

Introduction

11

Arguments against Epistemic Externalism N o doubt externalism has many advantages. Nevertheless, there are also several serious problems confronting this position. The main objections to externalism are based on deeply rooted internalist intuitions. Many internalist critics are fond of pointing out that externalism marks a significant break from the modern epistemological tradition originating from Descartes. So we find Bonjour saying: W h e n viewed f r o m the general s t a n d p o i n t o f the western epistemological tradition, externalism represents a very radical departure. It seems safe to say that until very recent times, n o serious philosopher w o u l d have d r e a m e d o f s u g g e s t i n g that a person's beliefs m i g h t b e epistemically justified simply in virtue o f facts or relations that were external to his subjective conception. Descartes, for example, w o u l d surely have been quite u n i m p r e s s e d by the suggestion that his p r o b l e m a t i c beliefs a b o u t the external world were justified if only they were in fact reliably related to the w o r l d — w h e t h e r or not he h a d any reason for thinking this to be so. Clearly his c o n c e p t i o n , a n d that o f generations o f philosophers w h o followed, was that such a relation could play a justificatory role only if the believer possessed a d e q u a t e reason for thinking that it o b t a i n e d . 2 0

With this general point in mind, Bonjour and others have elaborated their criticisms of externalism by arguing that the satisfaction of externalist conditions is neither necessary nor sufficient for epistemic justification. The first objection challenges the necessity of the reliability requirement by considering demon worlds, worlds controlled by an evil omnipotent demon who produces, in the denizens of such a world, patterns of sensory experience exactly like our own, even though, in their case, the resulting perceptual beliefs are systematically false. 21 Despite the unreliability of their belief-generating processes, internalists quite plausibly claim that the believers in a demon world would be justified in their beliefs, just as much so as we are in ours. After all, they have the same reasons for belief as we do, and their whole epistemic situation may well be subjectively indistinguishable from our own. So the conclusion seems to be that reliability cannot be necessary for justification. Some externalists have responded by simply rebuffing the intuitions adduced by their internalist critics. Their hardline diagnosis is that these intuitions are misleading and in the end merely reflect the persistent internalist strand of the epistemological tradition. Externalists typically are supporters of the truth-conducivity view of justification. And if this view is correct, if justification indeed has some essential relation to the goal of truth, then the beliefs of the inhabitants of the demon world just are not justified—even though they may have done their best to bring their intellectual house in order. Other reliabilists, however, have responded by modifying and refining their theories. One kind of reliabilism, normal-worlds reliabilism, rejects a crucial assumption upon which the counterexample rests, the assumption namely, that the reliability of a cognitive process is to be assessed by its performance in the world that the believer in question inhabits. 22 Instead, so their alternative proposal, reliability should be assessed in normal

20 21 22

Bonjour 1985, 56 Foley 1987 Goldman 1986

Introduction

12

possible worlds, that is, in worlds consistent with our general beliefs about the actual world, about the kinds of objects and events that occur in it. Since the cognitive processes deployed by the victims of an evil demon apparently would be reliable in normal worlds, the reliabilist can now agree that the use of these same processes will generate justified beliefs even in a demon world. Thus we get the intuitively right result. The second, complementary objection to externalism tries to show that reliability is not sufficient for justification. Imagine a situation in which a person forms beliefs in a highly reliable way, that is, in which the requisite externalist conditions are satisfied, but in which she has no reason to regard her mode of belief formation as reliable and perhaps good reason to regard it as unreliable. Thus suppose that there is some person with a reliable clairvoyance capacity who forms beliefs about distant events on this basis, and these beliefs are usually correct. 23 Suppose further that this person has no evidence to think that she possesses such a cognitive capacity, or perhaps she even has strong evidence to the contrary. Internalists' intuitive claim is that, although the reliabilist conditions are clearly satisfied, this person is highly irrational or irresponsible and hence not epistemically justified in confidently accepting her clairvoyantly formed beliefs. While, again, several externalists have simply dismissed the intuition as misguided, others became convinced by this sort of case that a further strengthening of the basic reliabilist constraints is needed. So some externalists have introduced a "non-undermining" condition, a condition to the effect that, roughly, the subject neither believe nor have good reasons for believing that the generating cognitive process is unreliable. 24 So if the subject has adequate evidence against her possession of reliable clairvoyance powers, then the beliefs she arrives at on this basis would, according to this modified version of reliabilism, be unjustified. The new non-undermining condition would be violated. Yet another kind of problem facing reliabilism is the so-called "generality problem". 25 Reliabilism holds that a belief is justified if the cognitive process that causes it is reliable. But processes can be individuated in many ways. Whenever a particular belief is produced, the process token that induces it can be viewed as an instance of numerous process types of differing generality. The serious problem for reliabilism is that these different process types have different truth ratios, and hence different degrees of reliability. So which of these many types should be used in fixing reliability? Should the relevant types be construed broadly or narrowly? If the types are characterized too broadly, many clearly unjustified beliefs will have to be regarded as justified. Perceptual belief formation, for example, is usually deemed to be a reliable process, but it brings about unjustified beliefs as well as justified ones. Obviously, visual perceptions of objects at close range under optimal conditions are more reliable than visual perceptions of distant objects under substandard conditions. And even if the conditions of perception are optimal, some of the beliefs formed in such conditions, typically the more general

23 24 25

Bonjour 1985 Goldman 1986 Feldman 1985

Introduction

13

ones, are less likely to be mistaken than others. If, on the other hand, the types are characterized too narrowly, including detailed features of some specific situation, they may turn out to have only one actual instance, namely, the process token in question. In this "single case" scenario, the process type will be either 100 percent reliable or 100 percent unreliable, depending on whether the belief is true or false. Consequently, the belief is justified if and only if it is true, which is not a welcome result. The problem of the single case can be avoided, however, if general reliability is understood as a propensity, the propensity to produce a high truth ratio of beliefs, rather than as a strict frequency. Since, according to the propensity construal, actual track record is not sufficient, the additional problem is raised of adequately specifying the range of possible uses of the process. It seems reasonable, for various reasons, only to take into account the performance of the process in nearby possible worlds. With these plausible answers to the problem of the single case and to the related problem of the range of reliability, the way seems paved, finally, for a narrow principle of type individuation. For, plainly, it is narrow types that are needed to differentiate between those cognitive processes that do lead to knowledge and those that do not. Against this rather complex background, full of twists and turns, Goldman arrives at his well-considered solution of the generality problem. He advances the conjecture that the decisive psychological process type is the narrowest type that is causally operative in eliciting the belief token in question. 2 6

Externalism

versus Internalism

about

Content

Many of our mental states, processes and events, such as our beliefs, desires, perceptions and memories, are semantically evaluable. In contrast to stars, tables, or tomatoes, mental states have representational content or exhibit the feature of intentionality. When we believe or desire or perceive, we always believe or desire or perceive something. So what is characteristic of states with content is that they are directed at, or about, objects and states of affairs in the world. To be in a contentful mental state is to represent the world as being a certain way. In believing, e.g., we believe about certain things that they have certain properties or stand in certain relations. Therefore, beliefs are true or false depending on whether things are the way they are believed to be. O f course, mental states can fail to refer. Nevertheless, a non-referring mental state still has a satisfaction condition, a condition, specified by its content, which its purported referent, and the properties of its referent, would have to satisfy for the state to be correct. Representational states are what they are in virtue of their content, in virtue of how they represent things as being. Consequently, in our customary cognitive practice we find it appropriate to individuate such states, that is, to distinguish them from another, by their different contents.

26

Goldman 1986, 49-51

14

Introduction

But what is the nature o f content, the nature o f the things that we think or believe or desire? Answering this question involves taking sides in the fundamental debate between internalism, also often called "individualism", and externalism in the philosophy o f mind. Externalism and internalism proffer different conceptions o f the relations between mind and world. They are typically understood as theses about the individuation o f mental states, that is, as theses about the existence and identity conditions o f mental states. Externalism claims that the existence and identity o f entities outside o f the thinker herself determine the existence and identity o f her mental states. So, according to externalism, a belief, say, is intrinsically dependent upon the environment o f the believer. There is no way to characterize the beliefs a person has save by characterizing the disposition o f objects and properties in the external world that the belief is about. Thus the surrounding world is taken to be constitutive o f the very nature o f beliefs; after all, the world fixes what they are. There is no sharp metaphysical boundary separating mind and world.

Content Internalism Internalism offers a quite different picture o f the relationship between mind and world. Its defenders assert just what externalists dispute, namely that the contents o f the mind can be described in ways that do not require essential reference to how things are in the world outside. They can see no principal reason to embrace their opponents' view that mental content is essentially dependent on the physical or social environment in which we are situated. In sharp contrast to externalists, internalists maintain that the content o f mental states is determined only by the internal, non-relational properties o f the subject's mind or brain. For them, the mind is largely autonomous with respect to the world—not deeply penetrated by it. O f course, internalists, typically, do not deny that our beliefs are about things and states o f affairs in the natural world. They do not reject the very idea o f intentionality. Rather, what they insist upon, is that a belief could exist and have the very same content that it actually has even if its alleged object itself did not exist. So, according to internalism, mental content cannot have a relational nature. T h e traditional view o f content has been strongly internalist in character. Descartes' various sceptical arguments all rest on the assumption that our knowledge o f the content o f our sensory experiences, and o f the content o f the beliefs resulting from them, is logically independent o f any knowledge about the external world. Since there is no deductive bridge linking the way things appear to us and the way they really are, we cannot exclude the possibility that the world is very different from the way we take it to be. Hence, all our beliefs about the external world might be massively false. But, on the Cartesian picture, even if our beliefs were massively false in this way, even if a world outside our minds did not exist at all, we could still have the same sensory experiences and the same beliefs that we actually have. We could have exactly these experiences and exactly these beliefs even if we were brains in a vat—provided, o f course, something elicited in our brains the same internal states that we in fact enjoy. So, on the Cartesian view, the content o f an experience, or o f a belief, is an internal property

Introduction

15

of the subject. T h e individuation of mental states is completely independent of any external objects or states of affairs. Descartes is a metaphysical dualist; a sharp boundary separates the self-sufficient inner realm of the mind from the external realm of material substances. His metaphysical dualism of substances is accompanied by the epistemological claim that there is a fundamental asymmetry between our ways of knowing about our own minds and about the external world, including other persons' minds. We have a special direct cognitive access to what passes in our minds which contrasts sharply with our indirect ways of acquiring knowledge about the rest of the world. O n the Cartesian conception, we have, as it is often put, first-person authority with respect to the contents of our minds. Descartes' firm conviction that we are authorities about our own minds is grounded in his doctrine of self-intimation, the doctrine that our whole mental life is transparent to our introspective awareness. Moreover, he is convinced that introspective awareness is infallible. T h e Cartesian philosophy of mind, by setting the world dualistically over against our conceptual scheme, presupposes that mental individuation can indeed be explained without recourse to worldly conditions. It presupposes just the existence of thinkers and their thoughts. Descartes is the arch-internalist about the nature of mental content. After him, Locke, Leibniz, and Hume, despite the differences between their specific theories, shared most of the Cartesian internalist assumptions about the constitution of the content of mental representations. Gottlob Frege seems to be another internalist about content. He says, for example: Let us just imagine that we have convinced ourselves, contrary to our former o p i n i o n , that the n a m e O d y s s e u s , as it occurs in the O d y s s e y does designate a m a n after all. W o u l d this m e a n that sentences containing the n a m e " O d y s s e u s " expressed different thoughts? I think not. T h e thoughts w o u l d strictly remain the s a m e ; they w o u l d only be transposed f r o m the realm o f fiction to that o f truth. S o the object designated by a proper n a m e seems to be quite inessential to the thought-content o f a sentence which contains i t . 2 7

On Frege's semantic theory, there is a special Platonic realm of sense mediating between language and the realm of reference. A thought, as he uses the notion, is the sense expressed by a complete assertoric sentence. He characterizes sense as that property of an expression that has to be "grasped" in order to understand it. And though the sense of an expression is also described as the "mode of presentation" of its reference, his theory allows for the possibility that an expression may have a sense but lack any reference. Frege is asserting that our grasp of the sense of the name "Odysseus" is independent of whether or not it has a reference. Therefore understanding an expression, consisting as it does just in the grasping of entities of the realm of sense, occurs in splendid metaphysical isolation from the external physical world. Thinking, for Frege, is a mental activity in which we, in some unexplained manner, grasp, or apprehend, peculiar Platonic thoughts—objects which are mind-independent, objective and public, but nonetheless neither in space and time nor perceptible. Frege also believes in the trans-

27

Frege 1919, 191

16

Introduction

parency of sense. By grasping the sense of an expression we know everything there is to be known about its internal properties; a sense cannot have any properties that are undetectable by us. This is reflected in his criterion of identity, according to which a sentence S expresses a different thought than a sentence S" just in case it is possible for someone to understand both sentences while taking one to be true and the other to be false. We are capable of determining, by introspection alone, whether or not two sentences express the same or different thoughts. Concerning this criterion, the actual truth values of the sentences—their references, on Frege's view—are entirely irrelevant. In Frege's philosophy, a Cartesian view of the mind coalesces with a Platonic view of its contents. In contrast to Descartes' and Frege's dualism, many modern partisans of internalism rather tend to sympathize with a materialist metaphysics. If they believe at all that mental states and processes possess semantic properties, if they do not flirt with one of the various variants of eliminative materialism, then they are liable to believe that these properties supervene upon inner, non-relational features of the brain or body. John Searle, for example, says: If I were a brain in a vat I could have exactly the same mental states I have now; it's just that most of them would be false. . . . T h e operation of the brain is causally sufficient for intentionality. It is the operation o f the brain and not the impact of the outside world that matters for the content of our mental states. 2 8

Searle urges that the content of mental states is determined by operations of the brain. The psychology of an organism depends on internal physical facts; it is, in this sense, autonomous. Remarkably, in his pugnacious defense of internalism he goes so far as to claim that each of us is precisely a brain in a vat. The vat, he tells us, is a skull, and the messages from without reach us only by impacting on the brain and the rest of our central nervous system. 29

Arguments for Content Externalism The main arguments against content internalism and for its adversary, content externalism, derive from recent developments in the philosophy of language, associated primarily with the writings of Hilary Putnam, Saul Kripke, David Kaplan and Tyler Bürge. Presumably the most influential argument for externalism is Hilary Putnam's celebrated Twin-Earth thought experiment. Imagine that somewhere in the galaxy there is a planet, Twin Earth. Twin Earth is in almost all respects qualitative identical with Earth. Suppose that each Earthling has a doppelgänger on Twin Earth who is molecule for molecule identical with the Earthling. Moreover, our duplicates speak a language which is phonologically and syntactically the same as English. The only relevant difference between Earth and Twin Earth is that the liquid that plays the role of water

28

Searle 1 9 8 0 , 4 5 2

29

Searle 1 9 8 3 , 2 3 0

Introduction

17

on Twin Earth, though phenomenologically indistinguishable from water, is not really water, that is, H 2 0, but a different chemical composition, XYZ. Hence, when Oscar uses the term "water" on Earth, he refers to H 2 0, that is, to real water. When, however, his doppelgänger Twin-Oscar uses the term "water" on Twin Earth, he does not refer to H 2 0, but to XYZ. So what Oscar and Twin-Oscar refer to, and what they think about, is individuatively dependent on their actual environment—on the different natural kinds with which they are causally interacting. Suppose further that Oscar and Twin-Oscar are ignorant of the chemical composition of substances on their planets. Then, since they are exactly alike in every intrinsic physical respect—in the states of their brains and bodies, in their behavioural dispositions, and in their proximate patterns stimulations—this difference in reference, in extension, cannot be explained by any internal properties of them. The reference is different because the natural kind is different. And since reference is a component of meaning, Putnam concludes, memorably, that "'meanings' just ain't in the head". 30 They cannot be there because meaning depends on reference, and reference is a relation between a person and an external entity. Other philosophers soon pointed out that externalism in the philosophy of language has to be paralleled by externalism in the philosophy of mind. 3 1 Meanwhile Putnam himself is convinced that not only the meanings of words but the contents of the propositional attitudes as well are determined by facts about the physical environment and, moreover, by facts about the linguistic community to which the speakers belong. Mental content too is relational in nature. This externalist turn is what in the end motivated Putnam's rejection of his former functionalist theory of the mind. If propositional attitudes aren't inner states, the nature of which can be characterized independently of external factors, then the core thesis of functionalism, the thesis that mental states are algorithmic states in our heads, cannot be right. Functionalism ultimately fails because it embodies an internalist view of the mind. It fails, in other words, because the mind is not in the head either. Rather, Putnam claims, the mind is a "structured system of object-involving abilities". 32 Bürge conducts another thought experiment designed to show that external factors enter into the determination of the meanings of our words and the contents of our thoughts. 33 Imagine an English speaker who, although he has a large number of true beliefs about arthritis, does not know that arthritis is confined to the joints. He thinks falsely that this inflammation can occur in the bones as well as in the joints. Noticing an ailment in his thigh, which is symptomatically similar to the disease in his joints, he says to his doctor, "I have arthritis in my thigh". According to Bürge, he expresses, by this utterance, his false belief that he has arthritis in his thigh. When the doctor corrects his misconception, the patient relinquishes the belief. But let us now imagine a counterfactual situation in which the patient is neurophysiologically the same, but in which his social environment differs in just one respect: the word "arthritis" applies to

30 31 32 33

Putnam 1975, 227 McGinn 1977 Putnam 1992, 356 Bürge 1979

Introduction

18

inflammations of the bones as well as of the joints. Thus the patient's use of "arthritis" is correct in his counterfactual linguistic community. In this imagined situation, the patient would be expressing a true belief by the sentence "I have arthritis in my thigh", not a false one. So what he believes in the counterfactual case is different from what he actually believes. Yet there is no internal difference. The patient's entire physical and non-intentional mental histories remain constant. The difference in his mental contents stems from without—from differences in his social environment. It should be pointed out that, whereas Putnam's argument applies to natural kind terms, Burge's style of argument has a much wider application. Instead of "arthritis", various other sorts of words—artefact terms, adjectives, verbs or abstract nouns —could have been used to accentuate the distinctive role that one's linguistic community plays in fixing the meanings of one's words and the contents of one's thoughts. The relational or world-involving nature of representational content can also be brought to light by considering perceptual and indexical thoughts, thoughts whose objects are determined in a contextually sensitive way. Unlike general beliefs, genuine singular beliefs seem to require some psychologically relevant, often causal, relation between the believer and her objects; sometimes even a direct perceptual contact to items in the world, some face-to-face encounter with them, seems to be demanded. Hence these beliefs seem to essentially concern the particular individuals in the world they are about in that they could not exist unless their objects existed as well. In contrast to the class of de dicto beliefs, which relate believers to linguistic items, "dicta", perceptual and indexical beliefs are said to be irreducibly de re. Beliefs of these last two kinds involve relations between the believer and the relevant things, "res", in her vicinity as part of their identity. Suppose that, on seeing an animal before me, I say "That animal is a tiger", thereby expressing the demonstrative way of thinking that is made available by my seeing the tiger. What makes my thought a thought about that particular tiger? According to the traditional view, descended from Frege, the way, or how, something is represented determines what, which object, is represented. The idea is that the subject component of the thought, expressed here by the demonstrative "that animal", imposes a certain condition which an object has to uniquely satisfy in order to be the object of the thought. This view fails, however, because the way we represent an object is often not sufficiently detailed and precise to individuate it fully, that is, to single it out from all other objects. Moreover, sometimes we misrepresent things—what we, at first glance, regarded as a tiger might turn out to be a leopard or a hologram. Confronted by these sorts of problems, externalists have proposed an alternative picture of the mechanism of reference. In their view, what a singular thought is about is not solely determined by a Fregean sense, a conceptual ingredient of the thought which is sufficient by itself to fix its conditions of satisfaction, but also depends on features of the context in which the thought occurs. Bürge contends that a de re belief picks out its object in virtue of bearing an appropriate non-conceptual contextual relation to that object. 34 He urges that the content of such beliefs cannot be exhaustively

34

Bürge 1977

Introduction

19

characterized, or completely analysed, in terms of concepts, for the reason that these beliefs contain a crucial contextual element, an element that is independent of any concepts or other types of mental representations in the believer's repertoire. It is the context itself, and not any internal representation of the context, that determines the content of a singular thought. What particular object a persons thought latches onto depends on the circumstances in which the thought occurs and so cannot be derived from anything that happens within the boundaries of her body. It is a consequence of this conception of singular thoughts that they are neither true nor false absolutely, but only relative to their contexts of occurrence. So, in John Perry's instructive portrayal, such thoughts are "essentially indexical". 35 Their content is token-reflexive, contextually variable in the same sort of way in which the content of utterances of sentences containing indexical referring expressions, such as "I am happy today", is.

The Naturalization Project Since most externalists hold that intentionality is part of the natural order, they are engaged in the naturalization project, the project of reducing mental content, of showing how purely physical states are capable of representing and misrepresenting the world. The central question they have to answer is the question of what it is in virtue of which a mental state has the content it has. What, e.g., makes a particular belief

the belief that ρΐ

A natural suggestion is to explain content in terms of causal links between mental states and the world. This approach appeals to lawful covariation between mental states and certain external conditions. The core idea is to treat content as a kind of indication or natural meaning in Paul Grice's sense, the kind of meaning that obtains between the number of tree rings and the age in years of the tree, and between red spots and measles. Certain states are indicators of other states. There is a lawful dependency between the indicator and what it indicates. In Dretske's influential information-theoretic account, the point is put by saying that one event carries information about another event in virtue of their being a lawful regularity between them. This notion of carrying information is the basis for the attempt of Dretske and others to clarify how physical states can possess semantic content. The covariational account, however, is faced with the difficulty of accounting for the possibility of misrepresentation, that is, of error. There can be no misindication. Patently, however, there is misrepresentation. Now and then we form false beliefs. Suppose that beliefs about tigers have regularly been caused by the presence of tigers. But on some occasions beliefs of this type will be caused by things other than tigers, such as leopards, jaguars, holograms of tigers, and so on. Then, if the content of a belief is indeed fixed by what caused it, there is, apparently, no way to misrepresent a leopard as a tiger. A generalization of the problem of misrepresentation is what Fodor calls the "disjunction problem". Simple indicator theories seem to imply that the content of a

35

Perry 1993; see also Kaplan 1 9 8 9

20

Introduction

belief of a certain type is a disjunction of the indefinitely many sorts of causes that generate beliefs of that type. If, for example, beliefs of a certain type are caused at different times by tigers, leopards, jaguars and holograms of tigers, then what, on this account, is to prevent beliefs of that type from having the disjunctive content or truth condition: tiger-or-leopard-or jaguar-or-hologram of a tiger? One might try to meet the disjunction problem by specifying a privileged set of standard or ideal conditions for the formation of beliefs. Then one could identify the content of a belief with its cause under standard conditions, and take false beliefs to be beliefs that arise from other possible causes under non-standard conditions. It has proven extremely difficult, however, to specify such standard or ideal conditions in a non-circular way. In the end, we seem to have no other choice but to construe standard conditions as those in which we form true beliefs. But this gambit is question-begging, for it makes use of the semantic concept of truth. 3 6 Dretske tries to meet the disjunction problem in a different way. 37 H e equates the content of a belief of a certain type specifically with those conditions that cause tokens of that type during a learning period. In the process of learning to form them, beliefs acquire the semantic content that they have. Misrepresentation is then construed in terms of post-learning-period causes that differ from learning-period causes of beliefs of that type. An evident problem facing this approach is how to demarcate non-circularly the conditions that define learning periods. There seems to be no fixed point at which learning comes to an end. Moreover, a child, while still in the period of learning the concept of a tiger, will, presumably, also be exposed to leopards, jaguars or holograms of tigers, and hence might easily mistake one or the other of these various items for a tiger. So the correlation that learning establishes is between the concept of a tiger and tigers-or-leopards-or-jaguars-or-holograms of a tiger. The original problem returns. According to teleological theories of content, the disjunction problem can be solved by introducing the notion of biological function or purpose. 3 8 O n this account, beliefs are states with biological functions, and, accordingly, the attempt is made to analyse their contents in terms of those conditions they are biologically designed, or supposed, to covary with. It is these proper functions, designed by natural selection, that fix the content of a belief. The basic idea is that a belief has the content ρ just in case the belief-forming mechanism which produced it has the function of producing that belief only when it is the case that p. So tiger-beliefs represent tigers, not leopards, because it is their biological function to be held when tigers, but not leopards, are present. It is important that beliefs possess information-carrying functions, even if there are no conditions under which these functions can be carried out appropriately. In such cases our cognitive abilities fail. This is what is supposed to make misrepresentation possible. T h e beliefs formed in such cases have a false content; they fail to carry the information that, in virtue of the type they are tokens of, it is their natural function to carry. Content and truth fall apart.

36 37 38

See Cummins 1989; Fodor 1990, 31-49 Dretske 1981 Millikan 1984; Papineau 1987

Introduction

21

Fodor, however, has argued that the teleological account does not really solve the disjunction problem. 39 Suppose that frogs have evolved in environments in which the properties of being a fly and of being a small black dot lawfully covary. Then selectional processes do not show whether it is the biological function of the frog's relevant representation to represent flies or to represent small black dots, or both, or something else which covaries with both. After all, the frog's perceptual system is not malfunctioning when it responds to a small black dot rather than a fly. The determinacy of mental content seems to remain underdetermined by biological function and natural selection. Fodor himself proposes another solution to the disjunction problem. The essential idea of his "asymmetric dependency" theory is that falsehoods are semantically dependent on truths in a way that truths are not semantically dependent on falsehoods. The symbol "tiger" stands for tigers, and not for tigers-or-leopards, because tigers would still cause tokens of "tiger" even if leopards did not, but leopards would not cause tokens of "tiger" if tigers did not. In general, the causal relations between Fs and tokens of 'X' constitute the fact that "X' stands for Ks if for every Ζ the causal relations between Z s and tokenings of "X" are asymmetrically dependent upon it. Indubitably, the sufficient conditions for mental representations Fodor describes are very plausible. Nevertheless, his reductive account has not obtained universal acceptance. A general worry that is raised is whether the thesis of asymmetric dependency does not implicitly presuppose what it promised to be explaining. Does it more than to give expression to our natural conviction that error is parasitic on truth? Does it really provide us with an explanation of error?40

Arguments against Content Externalism Several renowned philosophers have objected that externalism about content cannot be reconciled with the causal or explanatory role that beliefs, desires and other prepositional attitudes are supposed to play in an agent's psychology, that is, with their role in mediating between perception and behaviour.41 The main argument is based on certain reasonable principles about causation, especially about mental causation. Mental events, so the argument goes, have their effects in virtue of their physical properties, not in virtue of their having content. The fact that the vernacular of folk psychology delineates neural events as takings of attitudes toward contents should not tempt us to think that contents really possess causal potential. In fact, content is causally impotent, screened off by the physical properties of neural events. So the problem that is raised, the problem of mental causation, is to find a causal role for contentful mental states in the aetiology of behaviour. According to externalism, mental content is constituted by relations to items in the non-mental world. And the charge is that these extrinsic relations to things outside of a person's body cannot themselves be involved in the causal

39 40 41

Fodor 1990 See Crane 1995 See Dennett 1982, Field 1978, Fodor 1981, Loar 1981, Pylyshyn 1984, Stich 1983

22

Introduction

transactions. Causation is local and intrinsic—everywhere. T h e workings of brains or minds are no exceptions. So the extrinsic relations of a mental state cannot contribute to its causal powers; they cannot be incorporated into the causal mechanisms. Since psychological explanations have to cite causally relevant features of events, and contents are not causes at all, there seems to be no theoretical role left for the notion of content. Consequently, Jerry Fodor proposes "methodological solipsism" as a "research strategy in cognitive psychology". He regards this as a corollary of his "computational theory of the mind", which is decisively influenced by Alan Turing's systematic conception of computation. Turing's idea of symbol manipulating machines for the first time opened up the possibility of defining any rational computation by simple operations on syntactically structured representations. According to the computational theory of the mind, mental states and processes are computational, that is, have their causal efficacy as a consequence of the formal properties of the representations over which they are defined. Fodor calls this the "formality condition", which psychological theories have to fulfil. Computational processes are formal in the sense that they apply to representations in virtue of their syntax. The crucial point is that being syntactic is a way of not being semantic. Syntactic properties are higher-level physical properties. Therefore, representations can, in virtue of their syntactic structures, causally interact with each other. T h e syntax of a mental symbol, its shape, is a determinant of its causal role, not its semantic properties. Zenon Pylyshyn takes the same line. He adopts the formality condition, and explicitly says: "cognition is a type of computation". 4 2 Stephen Stich is another philosopher of psychology who is convinced that externalist intentional attitude attributions are at odds with the requirements of a comprehensive scientific psychology. He maintains that the propositional attitudes, in virtue of their relational individuation, cannot be incorporated into scientific psychology. 43 According to him, the notion of intentional content is simply out of place in a genuine scientific theory about the mechanisms underlying behaviour. T h e alternative he offers to folk psychology is the "syntactic theory of the mind". Its basic idea is that causal relations between cognitive states mirror formal relations between syntactic objects. For Stich too, the laws of mental processes appeal to properties of representations that are only syntactic. Similarly, Daniel Dennett holds that the mind is a "syntactic engine", not a semantic engine. 4 4 A semantic engine is, he finds, a "mechanistic impossibility". Contents and meanings, according to him, are impotent, merely epiphenomenal with respect to the causal powers of a neural event. They are mere epiphenomena because they cannot influence the mechanistic or syntactic flow of local causation in the nervous system. 4 5 T h e predictions we arrive at in the "physical stance" have always hegemony over those of the "intentional stance". Dennett is, or used to be, an antirealist, a sort of instrumentalist, both about the entities purportedly described by folk psychology

42

Pylysyn 1984, XIII

43 44 45

Stich 1983 Dennett 1982 Dennett 1991b

Introduction

23

and about content. Meanwhile, however, he tends to concede a certain kind of reality to the intentional states. 46 So, according to methodological solipsism, cognitive psychologists, in trying to explain behaviour, must rely exclusively on what is in the head. Everything that happens outside an agent s head, or beyond the confines of her body, is irrelevant to psychological science. This is so because the causal profile of mental states depends only on their intrinsic syntactical properties. Naturally, causal links with sensory input and behavioural output should not be neglected—although, unfortunately, there is a general inclination in these circles to ignore them. Thus, the main thesis of methodological solipsism is that the systematic role mental states play in the organization of behaviour is just an internal matter of how they interact with each other, with proximal stimuli, and with subsequent bodily movements. Referential relations to states of affairs external to the agent do not figure in psychological explanations. Truth and reference, Fodor proclaims, aren't psychological categories. Psychology is, in this sense, an autonomous discipline. Methodological solipsism is often buttressed by arguing that molecule for molecule identical counterparts will have the same causal powers and, accordingly, will behave alike regardless of any differences in their environments. The world outside might have been different, or might change, without affecting a person's psychologically relevant properties, the ones that rationalize her behaviour, in the least. As far as psychological explanation is concerned, what counts is how the world is internally represented as being, not how the world really is. It is these modes of representations of things, shaped by a person's subjective point of view, that are constitutive of the causal-explanatory role of mental states. And so, according to the theoretical taxonomy of mental states demanded by methodological solipsism, mental states can be said to be the same or different according as their intra-individual causal role is the same or different.

T h e Case for Narrow Content Fodor argues that methodological solipsism does not entail the dismissal of commonsense or folk psychology. In contrast to Dennett's instrumentalism or his more recent "mild realism" and to the various versions o f eliminative materialism proposed by Stich, Paul Churchland and Patricia Churchland among others, 4 7 who all deny that the prepositional attitudes are accommodatable within a materialist metaphysics, Fodor insists upon full-fledged intentional realism—the view that there really are intentional attitudes that are causally involved in the production of behaviour. Moreover, he also insists upon scientific intentional realism, that is, he contends, further, that scientific psychology will still have to posit mental states, and will have to contain intentional laws subsuming them, including, in particular, laws that relate beliefs and desires to one another and to actions. For Fodor, mentalistic folk psychology is the breeding-ground of a serious physicalistically acceptable psychology.

46 47

Dennett 1991a See Stich 1983; Churchland, Patricia 1986; Churchland, Paul 1989

Introduction

24

Against this background, Fodor shows himself convinced that intentional psychology has to preserve some notion of content—narrow content, constituted by internal aspects of a thought, in contrast to wide or broad content, which comprises the referential relations a thought bears to external entities. Although Oscar's und Twin-Oscar's thoughts have different truth conditions, that is, different wide contents, there is also a sense in which they have the same thoughts, thoughts with the same narrow content. From the inside, their thoughts are semantically indistinguishable from each other. If Oscar were instantaneously and unknowingly transported to Twin Earth, his thoughts would, in a certain sense, remain unchanged. Moreover, it is in virtue of this commonality, this shared content, that they seem to behave in the same way. If, for example, both Oscar and Twin-Oscar are thirsty, then the notion of narrow content can be used to explain and to predict their reaching for a glass of H 2 0 or XYZ respectively. Would, so it is sometimes asked, an attempt to explain their identical behaviour as resulting from different intentional states not have to invoke a kind of miracle? It has been objected that for the claim that Oscar's and Twin-Oscar's behaviour is the same to be plausible, "behaviour" must not be construed as intentional behaviour, that is action, since actions themselves are normally individuated externally, in terms of the particular objects or substances they involve. 48 Thus, apparently, "behaviour" can only mean something like bodily motion physically described. However, there seem to be behaviours that are both intentional and relational. And, surely, it is the task of psychology to explain them. So perhaps "behaviour" means, more promising, the type of action both Oscar and Twin-Oscar perform. At any rate, many philosophers of psychology hold that a notion of narrow content, which, by not being answerable to the environment, is able to capture an intentional content common to the protagonists, is just what is needed for the purposes of rationalizing explanation. Without it, so the argument runs, important psychological laws and other generalizations could not be captured.

But what is Narrow Content? There is an intense debate about how the notion of narrow content should be characterized. Fodor himself takes the narrow content of a mental representations to be a function that maps an external context onto a wide content. 49 So mental representations have the same narrow content just in case they have the same wide content in all the same contexts. 50 It should be noted that, on Fodor's variant of the representational theory of the mind, mental representations are linguistic expressions within a special "language of thought", often referred to as "Mentalese", in which the computations constitutive of cognition occur. Roughly, to believe that p is to stand in a certain relation to a symbol that means that p. One problem facing this approach to narrow

48 See Peacocke 1981; Evans 1982 49 See also White 1982 50 Fodor 1990

Introduction

25

content is to define the operative concept of context. Furthermore, one would prefer a positive, informative characterization of narrow content, one that goes beyond mere abstraction from wide contents. A rival and more popular conception of narrow content, "conceptual role semantics" or "functional role semantics", has its roots in the philosophy of language. Sometimes even Wittgenstein, in virtue of his famous dictum that the meaning of an expression is its use, is claimed to be an advocate of this approach. 51 Functional role semantics has salient affinities both with structuralism in linguistics, and with what artificial intelligence researchers call "procedural semantics". Moreover, this approach is strongly influenced by functionalism in the general philosophy of mind, whose chief thesis is that what makes a mental state the type of state it is—what makes a wish a wish—is the functional relations it bears to other mental states, to sensory inputs, and to behavioural outputs. A further important source of inspiration for functional role semantics is the philosophy of Sellars. 52 Sellars argues that there are linguistic rules that license not only moves from linguistic stimuli to linguistic responses—"intralinguistic moves", but also moves from sensory episodes to linguistic responses—"language-entry transitions", and moves from linguistic episodes to extralinguistic bodily movements—"language-departure transitions". And he also accentuates that the network of inferential relations holding between sentences of a public language mirrors the inner network of functional relations holding between thoughts. 53 The various current versions of functional role semantics in the field hold that narrow content, or narrow meaning, is to be identified with functional or conceptual role. The central idea is that the meaning or content of symbols, public or mental, is fixed by the way they are related to one another. 54 But we must ask exactly what relations are supposed to constitute the content of a representation. The leading accounts attempt to characterize a representation's functional role in terms of its inferential relations, inductive or deductive, to other representation. Of course, most advocates of this approach hope that these inferential connections can be spelled out naturalistically, in terms of purely causal connections. It is worth mentioning that Fodor is among the sharpest critics of this alternative conception of narrow content. 55

Problems with Narrow Content The main problem this approach finds itself confronted with is to find a non-arbitrary way of specifying the relevant content-constituting connections, to show exactly how conceptual role engenders the determinate narrow content that mental states allegedly possess. What makes the problem so hard, is the holism of epistemic connections. According to the famous Duhem-Quine thesis, our beliefs about the world, or the sen-

51 52 53 54 55

Harman 1 9 8 2 Sellars 1 9 6 3 Loar 1981 Field 1977, 1978; Loar 1 9 8 1 , 1982; Harman 1987; Block 1986; Lycan 1 9 8 8 Fodor 1 9 8 7 ; Fodor & LePore 1992; see also Putnam 1 9 8 8

26

Introduction

fences expressing them, are not separately vulnerable to adverse observations, because it is only jointly as a theory that they imply their observable consequences. So once it is admitted that at least some epistemic connections of a belief determine its content, it is hard to avoid concluding that all of them do. Thus, if we follow conceptual role semantics, confirmation holism seems to lead to meaning holism, to the doctrine that the content of a belief is determined by the totality of its epistemic connections. If meaning holism is true, however, then every difference in conceptual role, every theoretical difference, generates a difference in meaning. As a result, it becomes virtually impossible for different people ever to share a belief, or ever to mean the same thing by what they say; it becomes even impossible for two time slices of the same person ever to mean the same thing by what they say. Hence hopes for a serious intentional psychology which, in order to achieve generality, has to quantify over all organisms that are in a specified mental state, would be in vain. Confronted by such unattractive, destructive consequences, one must try to distinguish a subset of the totality of a belief's epistemic connections as semantically relevant. Any such strategy, however, has to come to grips with the big problem that there seems to be no principled basis for distinguishing between those inferential relations of a representation that constitute its meaning and all of its other inferential relations, that is, for distinguishing between change of belief and change of meaning, between collateral information and determinants of content. In the end, any such attempt to isolate the semantically relevant inferences seems to entail a distinction between analytic statements and synthetic statements, a distinction which, in the face of Quine's devastating criticism of it, most philosophers believe cannot be enforced. Nobody has ever succeeded in clarifying precisely what this dichotomy comes to. This is not yet the end of the problems for conceptual role semantics. Many philosophers think that it is a desideratum on an adequate semantic theory that meaning determines reference and truth conditions. But, as Twin Earth cases reveal, functional role semantics does not satisfy this desideratum. T h e thoughts of Oscar and TwinOscar have the same functional role, but different reference and truth conditions. To cope with this difficulty, several philosophers have developed two-factor, or dual aspect, semantic theories. 56 T h e idea underlying this approach is that meaning, or content, consists of two relatively independent components. One component, the internal conceptual role component, gives an account of understanding and cognitive significance. T h e other component, typically a theory of truth, deals with the properly semantic relations between the representations in the head and their referents in the world. So two-factor theories make a conjunctive claim for each sentence or belief: what its conceptual role is, and what its truth conditions are. Consequently, the words we use in ascribing propositional attitudes have to make a dual contribution: they have to specify both functional role and truth conditions. 5 7

56

Block 1986; Field 1977; M c G i n n 1982; Loar 1982; Lycan 1984

57

Loar 1983; Lycan 1988

Introduction

27

Not all adherents of conceptual role semantics are convinced that such a supplementation of their theoretical framework is really requisite. Gilbert Harman thinks that a semantic theory with only the one factor of conceptual role will do. 5 8 In contrast to the prevailing solipsistic versions of conceptual role semantics, he advocates a non-solipsistic version of it, one, according to which his one factor does not stop at the skin, but, by taking into account perception and action, reaches out in the external world of referents. Block aptly portrays Harman's conceptual roles as "long-armed", in opposition to the "short-armed" conceptual roles of two-factor theory he himself favours. 5 9 Ultimately, Block thinks that the dispute is "merely verbal". Surely, no reasonable theory can entirely ignore the links between language or mind and the world. If we give up the concepts of reference and truth conditions, we risk giving up linguistic meaning and mental content as well. So it seems that the external relations of language have to be acknowledged anyway. Nonetheless, two-factor theories still have to struggle with the notorious holism problem, since the conceptual role component, the job of which it is to explicate the sense in which there can be considerable differences in meaning in cases where there are no extensional differences, remains a holistic affair. While reference and truth conditions are shared, conceptual role continues to vary idiosyncratically from person to person. Furthermore, these theories are also faced with a new problem, the problem of coordinating the two factors. What is the relationship between narrow content and wide content? In particular, these theories are committed to tell us which conceptual role meshes with which reference and truth conditions in order to constitute the total content of a belief. It is conceded that, in contrast to Fregean sense, conceptual role does not fix reference. So this approach needs, but apparently is not in possession of, some mechanism which keeps narrow content and wide content mutually coherent.

Externalist Replies The main argument for narrow content turns on the assumption that externalist attitude ascriptions are in conflict with the constraints of a scientifically adequate psychology. Bürge has countered that the intentional attitudes, in spite of their relational individuation, are well-suited for a mature psychology. H e argues that neither the theoretical commitments nor the language of cognitive psychology, as it is currently practised, conform to the strictures of internalism. Rather, psychological theories—theories of vision, of memory, of belief-formation, of reasoning etc.—deal partly with relations between persons, or animals, and their environments. The point of these theories is to explain in more depth how people see things, how they form beliefs about them, and how they remember them. The relations to the physical world, which are a vital part of these various theories, prompt externalist principles of individuation.

58

Harman 1987

59

Block 1986, 6 3 6 - 6 3 7

Introduction

28

Moreover, Bürge attacks the popular charge that externalism violates the allegedly eminently plausible supervenience principle that two persons' propositional attitudes, the determinants of their behaviour, could not be different unless what happens within the boundaries of their bodies were different. In his view, such a principle presupposes rather than establishes that the natural kinds of psychology should be internally individuated. It is not based on facts about the scientific methodology psychologists actually employ, or on a comparison to what counts as a good explanation in mature sciences. Rather, the internalist supervenience principle seems simply to beg the question, since it ignores the important role which the specific nature of a person's or animal's surrounding environment plays in psychological explanations. Attribution of mental states in psychology presuppose a normal background, a background which intentional explanations usually take for granted. The reason why molecule for molecule duplicates may nevertheless differ in respect of their beliefs is that they may have different backgrounds. Bürge emphasizes that even the natural sciences employ externalist styles of explanation. Geology and biology, for example, appeal to entities that are not supervenient on their internal material composition. In these sciences too we find a type of individuation of explanatory kinds that makes essential reference to the external relations in which instances of these kinds stand. 60 The crucial point is that whilst causation is local, individuation of natural kinds is not.

Externalism and Self-Knowledge Many philosophers claim that content externalism poses a serious threat to the traditional doctrine of privileged self-knowledge. There is indeed an asymmetry between our knowledge of our own mental life and our knowledge of the external world—including our knowledge of the mental life of others. We have, as it is often put, first-person authority with regard to the contents of our own mind. We generally know directly or immediately, without recourse to inference from evidence, what we ourselves think, desire, and hope, whereas others can get at the contents of our intentional states only indirectly—by relying on our sayings, actions and the context. The assumption, however, that we have this kind of privileged access to our own minds seems to come into conflict with content externalism's central thesis. If mental content is indeed partly determined by environmental and socio-environmental factors, then knowledge of content should depend on knowledge of these factors. But it cannot reasonably be supposed that these external circumstances will in general be accessible to us—at least not a priori, by simple reflection, without empirical investigation. Facts beyond our bodies are as accessible to others as they are to ourselves. Prima facie, it might seem that we have to give up either the thesis of privileged epistemic access or the tenet of content externalism. This is, no doubt, an unattractive choice for many philosophers. So it is small wonder that several philosophers have a

60

See also Davies 1986

Introduction

29

sneaking suspicion that the conflict is not genuine. Consequently, they endeavour to espouse a version of compatibilism, according to which content externalism can be reconciled with the doctrine of privileged access to the contents of our own thoughts. In support of such a reconciliation, it might be pointed out that a thinker's second-order beliefs are rationally responsive to her own conscious first-order beliefs, and hence will inherit certain environmental dependencies that shape her first-order beliefs. Paul Boghossian, Michael McKinsey and others, however, have countered that a compelling case can be made for incompatibilism. So Boghossian is convinced that the conflict between our ordinary conception of authoritative self-knowledge and content externalism is indeed genuine. 6 1 In particular, he attempts to show that compatibilism implies that we can have a priori knowledge of certain facts about the world, which everyone agrees can be known only on the basis of empirical investigation of the environment. His claim is that compatibilism leads to an absurd overextension in the realm of what can be known from the inside, by purely a priori reasoning. Other philosophers remain largely unimpressed. 6 2 Bürge has argued that a person's distinctive entitlement to self-knowledge does not conflict with content externalism, because that entitlement is neutral with respect to the issue of whether a person's mental states are individuation-dependent on environmental factors. Since externalism about mental content is a metaphysical thesis whilst privileged access is an epistemological one, there is no reason to be anxious that the two theses will encroach on each other's territory. In his eyes, a person's distinctive entitlement to self-knowledge has its source, first, in the role such knowledge plays in critical reasoning, and second, in the constitutive relation that holds between reflective thoughts and their contents. In cases of self-knowledge, the perspectives of the thinker as subject of the first-order belief and as subject of the second-order belief are unified, and it is this single perspective that forges a rational connection between reflective thoughts and their subject-matter. T h e entitlement to self-knowledge, Bürge contends, is independent of knowledge of the external factors that condition mental content.

Non-Conceptual Content Beliefs and other propositional attitudes have conceptual contents, contents normally specified by "that . . . " clauses. This means, roughly, that, in order to have a belief, a person has to possess the concepts that essentially characterize the content of that belief. Some philosophers assert, however, that not all content is conceptual, that some mental states have non-conceptual contents, contents that do not necessarily involve the possession and application of concepts. One prominent kind of mental state that is claimed to have such non-conceptual content is conscious perceptual experience. While the majority of philosophers still think that the content of perceptual experience is com-

61 62

Boghossian 1989, 1998; McKinsey 1991, 1994 See Brueckner 1992; McLaughlin & Tye 1998

30

Introduction

pletely conceptual, several philosophers such as Dretske, Evans and Peacocke argue that perceptual content is either completely or at least partly non-conceptual. 6 3 The case for non-conceptual perceptual content rests in part on the supposition that the character of experience is phenomenologically much richer and more fine-grained than can be formulated by using concepts at the disposal of an ordinary subject. T h e non-conceptualists contend that we do not possess, say, as many colour concepts as there are shades of colour we can sensorily discriminate. Our repertoire of colour concepts, expressible by words like "blue" or "red" is too coarse-grained to capture the finest detail, the specific character, of our colour experience. McDowell, however, points out that to regard the content of experience as non-conceptual is to commit oneself to what Sellars has branded as the "myth of the given"— the view that there is an unconceptualized given element in experience which plays a key role in the acquisition of empirical concepts and which, moreover, functions as an ultimate foundation of our beliefs about the world. 6 4 McDowell maintains that this picture cannot do justice to the intentionality of perception, and so in the end makes a mystery of the relation between mind and world. In his reply to the phenomenological case for non-conceptual content he concedes that not all qualitative aspects we are presented with in sensory experience can be captured by general concepts. In his countermove, however, he urges that these aspects can be expressed by special demonstrative concepts which are made available by the colours being given in a particular way in experience. We possess, he insists, conceptual capacities whose content is expressible with the help of demonstrative phrases like "that shade". These conceptual capacities enable the fine-grained sensuous detail which we find in experience, and which is supposed to escape the discriminative powers of our conceptual repertoire, to be embraced within that repertoire. And he is at pains to convince us that what is in play here is a genuine conceptual capacity, one which is not restricted in its exercise to a single occasion in which experience presents one with the respective sample. Rather, he maintains, the capacity is a "recognitional capacity", one which can persist into the future, if only for a short time, and which can, in the absence of the original sample, be exploited in thoughts based on memory. One can agree with McDowell that we do indeed possess such demonstrative conceptual skills which are made available to the subject by her perceptual state. Such skills permit us to express the content of experience, so that for each sensory feature there is a corresponding concept. But the question arises of whether that fact alone is capable of showing that the content of experience is conceptual through and through, that the very existence of experience depends on the deployment of concepts. One might come to think that the dependency is the other way around, that the demonstrative concepts themselves can only be elucidated in terms of the qualitative content of experience. But if this is so, then the content of experience is independent of the possession of the corresponding concepts.

63 64

Dretske 1981; Evans 1982; Peacocke 1992 McDowell 1994; Sellars 1963

31

Introduction

A detailed account of non-conceptual perceptual content has been given by Peacocke. 65 According to it, perceptual content is determined by a "scenario", a spatial type under which the region of space around the perceiver must fall if an experience with that representational content is to portray the environment correctly. So the correctness of perceptual content depends upon whether or not the volume of the real world around the perceiver actually instantiates the spatial type that gives the content. The scenario is individuated by a specification of surfaces and features in the environment, and their distances and directions from the perceiver's body as origin. But there is no requirement that the perceiver herself possesses the concepts that enter into characterizing the properties and relations that are constitutive of the scenario. The scenario, lacking any sentence-like structure, is not composed of concepts at all. Several people have contributed in various ways to the successful completion of the project. All of them are warmly thanked. I owe special thanks to Oliver Scholz whose constant encouragement and advice were a great help to me. Again, thanks are due to Gertrud Grünkorn, Sabine Vogt, Hans-Robert Cram and the editorial staff at Walter de Gruyter for their expert assistance. Finally, I would like to add a word of thanks to Ole Göbbels, who helped me with the technical realization of the project.

Bibliography Alston, William P., 1989, Epistemic Justification, Ithaca: Cornell University Press Armstrong, D.M., 1973, Belief, Truth and Knowledge, Cambridge: Cambridge University Press Block, Ned, 1986, "Advertisement for a Semantics for Psychology", in Midwest Studies in Philosophy, Vol. 10, Minneapolis: University of Minnesota Press, 6 1 5 - 7 8 Boghossian, Paul, 1989, "Content and Self-Knowledge", Philosophical Topics 17, 5 - 2 6 Boghossian, Paul, 1998, "What the Externalist Can Know A priori", in Wright, Crispin & Macdonald, Cynthia (eds.), Knowing our own Minds, Oxford: Clarendon Press, 2 7 1 - 8 4 Bonjour, Laurence, 1985, The Structure of Empirical Knowledge, Cambridge, MA: Harvard University Press Brueckner, Anthony, 1992, "What an Anti-Individualist Knows A Priori, Analysis 52, 1 1 1 - 1 8 Bürge, Tyler, 1977, "Belief De Rc", Journal of Philosophy 74, 3 3 8 - 6 2 Bürge, Tyler, 1979, "Individualism and the Mental", Midwest Studies in Philosophy, Vol. 4, Minneapolis: University of Minnesota Press, 7 3 - 1 2 1 Bürge, Tyler, 1986, "Individualism and Psychology", Philosophical Review 95, 3 - 4 5 Bürge, Tyler, 1988, "Individualism and Self-Knowledge", Journal of Philosophy 85, 6 4 9 - 6 3 Chisholm, Roderick, 1966; 2 n d ed. 1 9 7 7 ; 3 r d ed. 1989, Theory of Knowledge, Englewood Cliffs, N.J.: Prentice-Hall Churchland, Patricia, 1986, Neurophibsophy, Cambridge, MA: MIT Press

Churchland, Paul, 1989, A Neurocomputational

Perspective: The Nature of Mind and the Structure of

Cambridge, MA: MIT Press Crane, Tim, 1995, The Mechanical Mind, London: Penguin Cummins, Robert, 1989, Meaning and Mental Representation, Cambridge, MA: M I T Press Davies, Martin, 1986, "Externality, Psychological Explanation and Narrow Content", Proceedings

istotelian Society Supplementary

65

Peacocke 1 9 9 2

Volume 60, 263-83,

Science,

of the Ar-

32

Introduction

Dennett, Daniel, 1982, "Beyond Belief", in Woodfield, Α., (ed.), Thought and Object, Oxford: Clarendon Press, 1-95 Dennett, Daniel, 1987, The Intentional Stance, Cambridge, MA: M I T Press Dennett, Daniel, 1991a, "Real Patterns", Journal of Philosophy 88, 27-51 Dennett, Daniel, 1991b, "Granny's Campaigne for Safe Science", in Loewer, B. & Rey, G., (eds.), Meaning in Mind. Fodor and his Critics, Cambridge, MA: M I T Press, 87-94 Dretske, Fred, 1971, "Conclusive Reasons", Australasian Journal of Philosophy 49, 1-22 Dretske, Fred, 1981, Knowledge and the Flow of Information, Cambridge, MA: M I T Press Evans, Gareth, 1982, The Varieties of Reference, Oxford: Clarendon Press Feldman, Richard, 1985, "Reliability and Justification", Monist 68, 159-74 Feldman, Richard & Conee, Earl, 1985, "Evidentialism", Philosophical Studies, 48, 15-34 Field, Hartry, 1977, "Logic, Meaning and Conceptual Role", Journal of Philosophy 69, 379-408 Field, Hartry, 1978, "Mental Representation", Erkenntnis 13, 9-61 Fodor, Jerry, 1981, "Methodological Solipsism considered as a Research Strategy in Cognitive Science", in Representations, Hassocks: Harvester Press, 225-253 Fodor, Jerry, 1987, Psychosemantics, Cambridge, MA: M I T Press Fodor, Jerry, 1990, "A Theory of Content and Other Essays", Cambridge, MA: MIT Press Fodor, Jerry & LePore, Ernest, 1992, (eds.), Holism: A Shopper's Guide, Oxford: Basil Blackwell Foley, Richard, 1987, The Theory of Epistemic Rationality, Cambridge, MA: Harvard University Press Frege, Gottlob, 1919, Posthumous Writings, ed. Hermes, H., Kambartel, F., Kaulbach, F., tr. Long, P. & White, R„ Oxford: Basil Blackwell Ginet, Carl, 1975, Knowledge, Perception, and Memory, Dordrecht: Reidel Goldman, Alvin, 1967, "A Causal Theory of Knowing", Journal of Philosophy 64, 357-72 Goldman, Alvin, 1976, "Discrimination and Perceptual Knowledge", Journal of Philosophy 73, 771-91 Goldman, Alvin, 1979, "What is Justified Belief?", in Pappas, George., {cd.), Justification and Knowledge: New Studies in Epistemology, 1 -23 Goldman, Alvin, 1986, Epistemology and Cognition, Cambridge, MA: Harvard University Press Harman, Gilbert, 1982, "Conceptual Role Semantics", Notre Dame Journal of Formal Logic 23, 242-56 Harman, Gilbert, 1987, "(Non-solipsistic) Conceptual Role Semantics", in LePore, E. (ed.), New Directions in Semantics, London: Academic Press, 55-81 Kaplan, David, 1989, "Demonstratives", in Almog, J., Wettstein, Η. & Perry, J., (eds.), Themes from Kaplan, Oxford: Oxford University Press, 481 -563 Kripke, Saul, 1980, Naming and Necessity, Oxford: Basil Blackwell Loar, Brian, 1981, Mind and Meaning, Cambridge: Cambridge University Press Loar, Brian, 1982, "Conceptual Role and Truth-Conditions", Notre Dame Journal of Formal Logic 23, 27283 Loar, Brian, 1983, "Must Beliefs be Sentences?", in Asquith, P.D., & Nickles, T„ (eds.), PSA 1983, Vol. 2, East Lansing, MI: Philosophy of Science Association, 627-43 Lycan, William, 1984, Logical Form in Natural Language, Cambridge, MA: M I T Press Lycan, William, 1988, Judgement and Justification, Cambridge: Cambridge University Press McDowell, John, 1994, Mind and World, Cambridge, MA: Harvard University Press McGinn, Colin, 1977, "Charity, Interpretation and Belief", Journal of Philosophy 74, 521-35 McGinn, Colin, 1982, "The Structure of Content", in Woodfield, Α., (ed.), Thought and Object, Oxford: Clarendon, 207-58 McKinsey, Michael, 1991, "Anti-Individualism and Privileged Access", Analysis 51, 9-16 McKinsey, Michael, 1994, "Accepting the Consequences of Anti-Individualism", Analysis 54, 124-28 McLaughlin, Brian &Tye, Michael, 1998, "Externalism, Twin Earth, and Self-Knowledge", in Wright, Crispin & Macdonald, Cynthia (eds.), Knowing our own Minds, Oxford: Clarendon Press, 285-320 Millikan, Ruth, 1984, Language, Thought and Other Biological Categories, Cambridge, MA: M I T Press Nozick, Robert, 1981, Philosophical Explanations, Oxford: Oxford University Press Papineau, David, 1987, Reality and Representation, Oxford: Basil Blackwell Peacocke, Christopher, 1981, "Demonstrative Thought and Psychological Explanation", Synthese 49, 187217 Peacocke, Christopher, 1992, A Study of Concepts, Cambridge, MA: M I T Press Perry, John, 1993, "The Problem of the Essential Indexical", Nous 13, 3-21 Plantinga, Alvin, 1993a, Warrant: The Current Debate, Oxford: Oxford University Press Plantinga, Alvin, 1993b, Warrant and Proper Function, Oxford: Oxford University Press

Introduction

33

Pollock, John, 1986, Contemporary Theories of Knowledge, Totowa, N.J., Rowman & Littlefiedl Putnam, Hilary, 1975, "The Meaning of'Meaning'", in Mind. Language and Reality, Cambridge: Cambridge University Press, 215-271 Putnam, Hilary, 1988, Representation and Reality, Cambridge, MA: M I T Press Putnam, Hilary, 1992, "Replies", Philosophical Topics 20,347-408 Pylyshyn, Zenon, 1984, Computation and Cognition, Cambridge, MA: M I T Press Ramsey, Frank, 1931, "Knowledge" in Ramsey, The Foundations of Mathematics and Other Essays, ed. Braithwaite, R.B. New York: Harcourt Brace Schiffer, Stephen, 1977, "Naming and Knowing", in Midwest Studies in Philosophy, Vol. 2, Minneapolis: University of Minnesota Press, 28-41 Searle, John, 1980, "Minds, Brains, and Programs", Behavioral and Brain Sciences 3, 417-24 Searle, John, 1983, Intentionality, Cambridge: Cambridge University Press Sellars, Wilfrid, 1963, "Empiricism and the Philosophy of Mind", in Science, Perception and Reality, London: Routledge & Kegan Paul, 127-96 Sosa, Ernest, 1991, Knowledge in Perspective, Cambridge: Cambridge University Press Stich, Stephen, 1983, From Folk Psychology to Cognitive Science, Cambridge, MA: M I T Press Swain, Marshall, 1981, Reasons and Knowledge, Ithaca: Cornell University Press Talbott, William, 1990, The Reliability of the Cognitive Mechanism, New York: Garland Publishing White, Stephen, 1982, "Partial Character and the Language of Thought", Pacific Philosophical Quarterly 63, 347-65

I In Defence of Epistemic Externalism

The "Challenge" of Externalism W I L L I A M P. A L S T O N

I The title of this volume reflects a widespread assumption among philosophers that externalism, whether in epistemology or in philosophy of mind, is something radically new on the scene. This new entry might be viewed as either dangerous or hopeful, depending on the philosophical orientation. In this essay I will be concerned exclusively with epistemological externalism. I will argue that the sense of a radical departure from "the epistemological tradition" stems from (1) an oversimplified picture of that tradition and (2) a failure to realize the complexity of contemporary externalism. I will not go so far as to suggest that externalism poses no challenge. It is far from toothless. But the challenge is not exactly what it is often credited with being. I begin with a statement of the challenge from an internalist opponent. In Bonjour 1985 he writes. When viewed from the general standpoint of the Western epistemological tradition, externalism represents a quite substantial departure. It seems safe to say that until very recent times, no serious philosopher of knowledge would have dreamed of suggesting that a persons beliefs might be epistemically justified merely in virtue of facts or relations that are external to his subjective conception. Descartes, for example, would surely have been quite unimpressed by the suggestion that his problematic beliefs about the external world were justified if only they were in fact reliably caused, whether or not he had any reason for thinking this to be so. Clearly his conception, and that of generations of philosophers who followed, was that such a relation could play a justificatory role only if the believer himself possessed adequate reasons for thinking that the relation obtained. Thus the suggestion embodied in externalism would have been regarded as simply irrelevant to the main epistemological issue, so much so that the philosopher who suggested it would have been taken either to be hopelessly confused or to be simply changing the subject (...this may be what some externalists intend to be doing), (p. 37) And here is a statement by the dean of American epistemology in the second half of the 20th century, Roderick Chisholm. We must be on guard...in interpreting contemporary literature that professes to be about "internalism" or "externalism". Some of those authors who profess to view knowledge and epistemic justification "externally" are not concerned with traditional theory of knowledge. That is to say, they are not concerned with the Socratic questions, "What can I know?", "How can I be sure that my beliefs are justified?", and "How can I improve my present stock of beliefs?". Indeed many such philosophers are not concerned with the analysis of any ordinary concept of knowledge or of epistemic justification. Therefore their enterprise, whatever it may be, is not that of traditional theory of knowledge. (Chisholm 1989, pp. 75-76)

38

In Defence of Epistemic Externalism

To evaluate these claims we must consider what "the Western epistemological tradition" or "traditional theory of knowledge" has to say, or is assuming, about matters in dispute between contemporary internalism and externalism. But first we must settle on a way to construe these current positions.

II T h e terms 'internalism' and externalism' are used in epistemology, as elsewhere, in various ways. Thinking of them as applied to theories of epistemic justification, as is commonly done, the contrast that is most prominent in the literature has to do with whether it is a necessary condition of something's being a "justifier" of a belief, i.e., something that contributes to its justification, that the subject have some kind of high grade cognitive access to it. Internalism gives a positive answer, externalism a negative answer to this question. (A second contrast will be introduced in due course.) With respect to this first contrast, 'internalism' is the term that, as J. L. Austin put it in the old male chauvinist days, "wears the trousers". Internalism puts certain restrictions on what can contribute to justification, and, on this point, externalism is simply the denial of those restrictions. 1 Before moving on to the second contrast the first one needs to be sharpened up. First, there is a distinction between mere epistemic access and the successful exploitation thereof. T h a t is, we might think of internalism as holding that epistemizers are restricted to what the subject has some high grade of access to, or what the subject actually has knowledge or justified belief concerning. In Alston 1986 the former is called "access internalism" and the latter "perspectival internalism". Bonjour's phrase, "one's subjective conception" suggests the latter. (Only "suggests", because one might have a conception of a fact without knowing or being justified in believing that any such fact obtains. But I take it from what Bonjour has written elsewhere that he really intended the epistemically stronger version.) But the "access" version is more prominent in the literature. It would generally be thought that reasons for a belief would pass the internalist test provided the subject could fairly easily become aware of them, even if the subject had never actually incorporated them into his belief system. Moreover, the access version is not so obviously saddled with an infinite regress as the perspectival version. (If one can be justified in believing that ρ only if one is justified in believing that conditions for being so justified are satisfied, then that belief in turn is subject to a similar requirement, and so on ad infinitum. This implies that one cannot be justified in any belief without possessing an infinite hierarchy of justified beliefs.) Hence I will conduct this discussion in terms of access internalism. T h a t still leaves at least two alternatives dangling. First, how strong does the access have to be? If it is merely that the subject must be able, somehow, to ascertain that the justifiers obtain, it is doubtful that any epistemizing conditions favored by externalists,

1

Later we will see that something more positive is involved.

The "Challenge" of Externalism

39

e.g., reliability of belief formation, would be excluded. Much stronger versions of the access requirement have been proffered. For example, in Ginet 1975, Ch. Ill, it is required that any justifier be recognizable just by clear headedly asking oneself whether it does obtain. Rather than spend the entire essay on this issue, I will think in terms of a vague requirement of "relatively direct ascertainability". The other alternative is between the existence of the alleged justifier and its justificatory efficacy. One might well have strong access to the former and not the latter. Presumably the existence of a sensory experience is directly ascertainable by its possessor, but knowledge of its justificatory function vis-a-vis a perceptual belief is rather harder to come by. Much of the internalist literature suffers from ignoring this distinction. But since internalist concerns would not be fully accommodated by the right kind of access to the existence of the putative justifier while being unable to determine whether it deserves that title, I will think in terms of strong access to both being required by internalism. As previously noted, we may think of externalism, in the first instance, as the denial of this restriction. Justifiers need not be fairly directly ascertainable by the subject, either their existence or their justificatory efficacy. This does not imply that externalism holds that justifiers always or usually fail to satisfy the internalist restriction. It only denies that it is necessary for their justifying status that they do so. I mention this at the risk of overkill because misapprehension on this point is not uncommon. And this is particularly unfortunate, since externalists typically recognize justifiers that do satisfy an internalist restriction on access, e.g., sensory experiences and conscious reasons.

Ill Now I turn to the question of whether the "Western epistemological tradition" is uniformly internalist on the access issue. It would be a miracle if there were complete unanimity on this point, as well as on any other point, among all philosophers prior to the 20th century who have been concerned with what today we identify as epistemological issues. In order to keep the discussion within manageable bounds, I will restrict myself to the 17th and 18th centuries, which have largely set the tone for subsequent epistemology through the 20th century. In attempting to consider whether the major epistemologists of this period were internalists in anything like the way Bonjour and Chisholm are construing internalism, the first roadblock we run into is this. Both philosophers are typical of contemporary Anglo-American epistemologists in supposing that the "Western epistemological tradition" has at all times shared the twentieth century preoccupation with "justification" as a prime epistemic desideratum and as a necessary condition for a true belief's counting as knowledge. But there are a number of reasons why Descartes and Locke, for example, along with most other pre-twentieth century philosophers, are innocent of any such tendencies. The most important point is that so far from thinking that "justification" is required for turning a true belief into knowledge, they would deny that anything could do that trick. They took knowledge to be a psychological state of a radically different sort from belief (judgment, opinion ). Knowledge was usually thought of as some kind of immediate awareness of facts, rather than as a judgment that a certain

40

In Defence of Epistemic Externalism

fact obtains. One place in Descartes where this is particularly clear is Meditation IV, where he repeatedly refers to knowledge as a function of the understanding and to judgment as an exercise of the will. The main point he makes there is that it is incumbent on us, when judging, to follow the lead of the "clear and distinct perceptions" of the understanding in what we judge to be the case. Locke is much more explicit about meta-epistemology than Descartes. " K n o w l e d g e then seems to me to be nothing but the

perception of the connexion and agreement, or disagreement and repugnancy of any of our Ideas. In this alone it consists." (Essay, IV, 1, ii). 2 In chapter 14 of Bk. 4 he is at great pains to distinguish knowledge from judgment. "The Faculty, which God has given Man to supply the want of clear and certain Knowledge in Cases where that cannot be had, is Judgment: whereby the Mind takes its Ideas to agree, or disagree; or which is the same, any proposition to be true, or false, without perceiving a demonstrative Evidence in the Proofs." (IV, 14, iii). Hume in the chapter on Knowledge in the Treatise follows Locke in this. "It appears, therefore, that of these seven philosophical relations, there remain only four, which depending solely upon ideas, can be the objects of knowledge and certainty". (I, iii, 1) By "depending solely upon ideas" Hume means that those relations necessarily follow from the nature of the relata, and hence can be ascertained with certainty just by being aware of the ideas so related. What cannot be known in this way is, as in Locke, relegated to mere belief. By the time we reach the late 18th century we find both Kant and Reid treating knowledge as a judgment that satisfies certain conditions. Reid does us the favor of explicitly criticizing Locke's distinction and insisting that "there can be no knowledge without judgment, though there may be judgment without that certainty which we commonly call knowledge." (VI, 1.) But even though the treatment of knowledge as distinct from all judgment is not omnipresent in these centuries, that is not of much aid and comfort to those who would foist our 20th century concern with "epistemic justification" on their early modern predecessors. For we will not find them giving prominence to any concept very close to our concept of epistemic justification, in any of its many versions. I do not suggest that any important philosopher has been unaware that some of our beliefs (judgments) are better grounded, founded, supported, than others; and various terms are used for this. Locke, for example, speaks of degrees of probability. But even so, aside from Locke, we do not find prior to Kant and Reid a great deal of interest in cognitive achievements that fall short of knowledge. Such attention as they give to determinants of "positive epistemic status", to use as neutral a term as possible, is mostly focused on what makes for knowledge. How, then, can we evaluate Bonjour's claim without distorting either that claim or our 17th and 18th century forebears? So far as I can see, the closest we can come is

2

I am not concerned here with Locke's restriction of the object of knowledge to one's own ideas, something to which, in any case, he does not consistently adhere. My concern here is with the way in which he restricts knowledge to an immediate awareness of facts, distinguishing it from judgment (belief) in which some proposition is taken to be true without our being directly aware of a fact that makes it true.

The "Challenge" of Externalism

41

this. Let's coin the term 'epistemizer' for anything that engenders a positive epistemic status for any propositional attitude, whether knowledge or belief. We can then ask whether the 17th and 18th century epistemologists subscribed, at least in practice, to access internalism with respect to epistemizers. But even here we must tread cautiously. Where knowledge is concerned everyone, no matter how internalist, will recognize that knowledge that p requires that it is true that p. And it is not generally true that people have a relatively direct access to the truth of all the true proposition we can envisage. If that were the case, we would all know a great deal more than we do. Hence it would hardly be reasonable to use a direct access requirement for everything that contributes to knowledge (including truth) as a litmus test for access internalism. But, of course, since it is (almost) universally agreed that knowledge requires more than true belief (or true whatever might or might not be knowledge), we could get around this difficulty by restricting our litmus test to whatever contributes to the knowledge status other than truth. But there are problems with that also. Contemporary internalists who think that justification and truth are required for a belief's counting as knowledge have been sufficiently impressed by Gettier to recognize that they are not sufficient. In addition, something must be there that will obviate Gettier problems. And, again, internalists accept, for good reasons, that we lack direct access to the satisfaction of these anti-Gettier conditions, e.g., the absence of any fact that when added to the justifying conditions would result in the belief's not being sufficiently justified. Hence we can't simply say: an epistemologist is an access internalist about knowledge provided he holds that anything that contributes to knowledge status other than truth is something to which we have relatively direct access. And so we still lack an acceptable general formulation. Recall what is giving us this difficulty. The aspect of knowledge to which contemporary internalists require relatively direct access is "justification". But most 17th and 18th century epistemologists, especially those who take knowledge and belief to be disjoint, do not suppose that anything very close to 20th century justification to be a necessary condition of knowledge. Therefore we have fishing in vain for something connected to knowledge with respect to which they might or might not conform to the image those like Bonjour and Chisholm have of them as access internalists. But all is not lost. After all, these ancestors of ours do take there to be something that is required for knowledge other than truth. What prevented us from saying that they fit Bonjour's

image (as far as knowledge as concerned) i f f they took what turns a true whatever

into

knowledge to be directly accessible, was that for our contemporaries this also includes some condition to block Gettier problems. To be sure, we might rule that as irrelevant to the issues, since the 17th and 18 th century epistemologists had the misfortune of disappearing from the scene before Gettier's epoch making 1963 article and hence had no chance to take any position on cognitive accessibility to Gettier-problem-blocking conditions. But we can secure the same result in a less high-handed way. Let's say that they are accessibility internalists with respect to knowledge i f f they suppose that the subject has relatively direct access to anything constitutive of knowledge other than truth and what it takes to block Gettier problems. If they are innocent of any concern with the latter, we can still consider whether they are internalists with respect to other knowledge-making features.

42

In Defence of Epistemic Externalism

I'm afraid that is the best I can do to make Bonjour's claim into something that can be genuinely investigated. To sum up, the claim can be construed as follows. BJ: T h e epistemological tradition takes it that any contributor to positive epistemic status for beliefs ( j u d g m e n t s ) and any contributor to a k n o w l e d g e status, other than truth and what it takes to block Gettier problems, is relatively directly accessible to the subject.

Remember that the discussion is conducted on the fiction that "the epistemological tradition" is confined to the 17th and 18th centuries. Now that we have a question, what is the answer? Without trying to go into the matter in the way that would be required for a serious historical investigation, I would suppose that it is positive. For Descartes what makes for knowledge is "clear and distinct perception", and there is every reason to suppose that he took it that having such a perception carries with it a (perhaps infallible) capacity to spot it. When in Meditation III he comes to raise doubts about whether such perception yields knowledge, they are not based on any uncertainty as to when one has such perceptions, but rather on whether we are so constituted that they are a reliable guide to truth. Even one as determined to surmount every possible sceptical doubt as Descartes never worries about our ability to recognize clear and distinct perception. Locke, in parallel fashion, never suggests that one might fail to register a "perception of the agreement or disagreement of ideas", or might suppose one has such a perception when one doesn't. These are facts of a sort that are wholly open to our awareness. Nothing can get between me and facts like this that could make me miss them or hallucinate them. Something of the same sort could be said for Hume (where knowledge is concerned) and for other advocates of an intuitive conception of knowledge. The matter is less clear with respect to what makes for positive epistemic status of belief or judgment, mostly, I think, because this receives much less attention in the period. But my sense is that there too Bonjour's thesis holds up. I believe that a close reading of, e.g., Locke and Reid on this point would again support BJ. For Locke (Essay, IV, xvi, 4), the "grounds of probability" are "conformity of any thing with our own Knowledge, Observation, and Experience" and "the Testimony of Others, vouching their observation and Experience". Though the latter, involving sense perception, is obviously less directly accessible than the former, I believe that Locke would take it to be sufficiently directly and unproblematically available to satisfy Bonjour's supposition. And Reid, who took the forms of "evidence" that ground our judgments to be sense, memory, consciousness, or credit in testimony, seems for all the world to suppose (though I am unaware of his ever explicitly discussing the matter) that these "evidences" are unproblematically and relatively directly accessible to the subject of the judgment in question. I have already acknowledged the slapdash character of this survey. But since I am primarily concerned in this essay to contest the claim that "the epistemological tradition" is uniformly internalist, my excuse for superficiality is that since here I agree with BJ my position is one of acknowledging a point made by my opponent. Hence I am under no obligation of supporting the position as solidly as is required when I dissent from my interlocutor.

T h e "Challenge" o f Externalism

43

IV So if the denial of access internalism were all there is to externalism, I would have to agree with Bonjour and others that externalism makes a sharp break with the epistemological tradition. But externalists are generally as deeply committed to a more positive feature—the insistence that epistemizers be truth-conducive. (I will use 'TC' to abbreviate 'truth-conducivity'.). The view that epistemizers are essentially T C is the view that nothing can function as an epistemizer unless it contributes to making the propositional attitude in question at least likely to be true. Thus reliabilism, the most common externalist position, holds that a belief is justified (alternatively that it is has what it takes for knowledge other than truth and something to deal with Gettier problems), only if it has been formed in a reliable way, a way that can be relied on to produce mostly true beliefs in a wide spread of situations of the sort we ordinarily encounter. The same holds for such forms of externalism as Nozick's "tracking" requirement for knowledge, and Plantinga's view of "warrant" as belief acquisition by the proper functioning of cognitive faculties, though it would take more space than I have to show this. Moreover, T C is not just externally conjoined with the denial of strong access. They are intimately connected. It is because externalists hold that the point of having justified rather than unjustified beliefs is that we want our beliefs to at least be likely to be true, that they reject strong access requirements. They are convinced that much of what renders beliefs likely to be true is not directly accessible to reflection, introspection, or the like. Thus they take there to be a choice between a direct access requirement and TC, and they choose the latter. To be sure, the T C aspect of externalism does not present such a clear cut opposition to internalism as the denial of an access requirement. Some internalists, including Bonjour, are equally committed to TC. At another place in Bonjour 1985 he writes: If epistemic justification were not conducive to truth in this way, if finding epistemically justified beliefs did not substantially increase the likelihood of finding true ones, then epistemic justification would be irrelevant to our main cognitive goal and of dubious worth, (p. 8)

But other internalists, including Chisholm, Ginet, and Foley are equally insistent that there is no conceptual connection between justification (Foley says 'rationality') and the likelihood of truth. According to this traditional conception of "internal" epistemic justification, there is no logical connection between epistemic justification and truth. (Chisholm 1989, p. 76) 3 Epistemic rationality is distinguished from other kinds of rationality by its truth-directed goal. The goal is for one now to believe those propositions that are true and now not to believe those propositions that are false. However, to say that the goal that helps distinguish epistemic rationality from other kinds of rationality is a truth-directed goal is not to say that truth is a prerequisite of epistemic rationality. In particular, it is not to say that it is impossible for what is epistemically rational to be false; and likewise it is not even to say that it is impossible for most of what is epistemically rational to be false. (Foley 1987, p. 155)

3

Chisholm is not just making the uncontroversial point that justification does not entail truth, but the stronger claim that it does not entail even the likelihood o f truth.

44

In Defence of Epistemic Externalism

Because externalism includes this positive feature of truth-conducivity, as well as the negative rejection of internalist access requirements, in deciding whether externalism is a radical departure from the epistemological tradition, we must take both aspects into account. We have already seen that a good case can be made for its novelty with respect to the second, negative aspect. But what about the positive aspect? Here the situation is radically different. Go back to Descartes' Meditation III, where, as I indicated in the last section, Descartes' reason for his initial doubts as to whether clear and distinct perceptions provide knowledge is a doubt as to whether they are a reliable source of truth. It is worth while examining the passage in detail. First, Descartes goes over the things he is convinced that he knows with certainty and notes that in every case there is "a clear and distinct perception of what I am asserting". That leads him to "lay it down as a general rule that whatever I perceive very clearly and distinctly is true". But before he can allow himself to rest in this reassuring confidence a doubt insinuates itself, "...this would not be enough to make me certain of the truth of the matter if it could ever turn out that something which I perceived with such clarity and distinctness was false". But is there any chance that this should happen? Here the idea of a deceiving deity suggest itself. D i d I not see at least these things clearly e n o u g h to affirm their truth? Indeed, the only reason for m y later j u d g m e n t that they were o p e n to d o u b t was that it occurred to m e that perhaps s o m e G o d c o u l d have given m e a nature such that I was deceived even in those matters which seemed m o s t evident. A n d whenever m y preconceived belief in the s u p r e m e power o f G o d c o m e s to m i n d , I cannot but a d m i t that it w o u l d be easy for h i m , if he so desired, to bring it a b o u t that I g o w r o n g even in those matters which I think I see very clearly.. . A n d since I have n o cause to think that there is a deceiving G o d , and I d o n o t even know for sure yet whether there is a G o d at all, any reason for d o u b t which d e p e n d s simply on this s u p p o s i t i o n is a very slight and, so to speak, metaphysical one. B u t in order to remove even this slight reason for d o u b t , as s o o n as the o p p o r t u n i t y arises I m u s t e x a m i n e whether there is a G o d , a n d , if there is, whether he can be a deceiver. For if I d o not k n o w this, it seems that I can never be quite certain a b o u t anything else. 4

Descartes is then led into the famous "Cartesian circle", in which he seeks to remove the above doubt concerning the reliability of clear and distinct perceptions by showing that there is an omnipotent and perfectly good God and that we are endowed with our cognitive faculties by Him. In so doing he relies on clear and distinct perceptions to provide the premises of the argument, thus assuming their reliability in order to show that they are reliable. I have no concern here with that venerable problem. It suffices for present purposes to point out that Descartes will not recognize the genuineness of any putative source of knowledge unless he can show that it is maximally truth-conducive, i.e., infallible. This is to demand the highest degree of truth-conducivity of anything that will contribute to a particular cognition's counting as knowledge. And note that Descartes not only requires that clear and distinct perception be infallible, but that he know that they are, thus combining the externalist requirement of truth-conducivity and the internalist requirement of higher level knowledge of this.

4

The text is that in Cottingham, StoothofF, and Murdoch (trans.) 1984, Vol. II, p. 25.

The "Challenge" of Externalism

45

On the present point Locke is less explicit than Descartes, the reverse order having been noted on the relation of knowledge and belief. But I take it to be sufficiently clear that Locke holds that what makes a true cognition into a case of knowledge is such as to guarantee that truth, again the highest degree of truth conducivity. To wit. When we possess our selves with the utmost

security

(emphasis added) of the Demonstration,

that the three Angles of a Triangle are equal to two right ones, What do we more but perceive that Equality to two right ones, does necessarily agree to, and is inseparable from the three Angles of a Triangle?" (IV, i, 2) By this the Mind clearly and i n f a l l i b l y (emphasis added) perceives each Idea to to agree with itself and to be what it is, and all distinct Ideas to disagree, i.e., the one not to be the other. (IV, i, 4)

And speaking of knowledge of modes, So that we cannot but be infallibly (emphasis added) certain, that all the knowledge we attain concerning these Ideas is real, and reaches Things themselves. Because in all our Thoughts, Reasonings, and Discourses of this kind, we intend Things no farther, than as they are conformable to our Ideas. So that in these, we cannot miss of a certain undoubted reality. (IV, iv, 5) Wher-ever we perceive the Agreement or Disagreement of any of our Ideas there is certain Knowledge: and where-ever we are sure those Ideas agree with the reality of Things, there is certain real Knowledge. (IV, iv, 18)

It is clear from the larger context that 'certainty' in these passages denotes not feeling sure or confident, or entertaining no doubt, but rather what we might term epistemic certainty, being in such a position that it is impossible that one should be mistaken. I could also cite the same position, that what makes for knowledge endows the cognition with certainty in the above sense, for Hume, Kant, and others. Things are less clear cut, again, with views on the truth-conducivity of determinants of positive epistemic status for judgments in this period. But I believe that it is sufficiently clear that the leading figures were construing these determinants as exhibiting a built-in truth conducivity. Begin with Locke. His term of choice for what I am calling a determinant of "positive epistemic status" for judgments is probability. Though Locke does not say as much as we would like about how he thinks of probability, he says enough to indicate that he thinks a strong ground of probability to be a significantly reliable indication of truth. Here is the relevant portion of the first section of Ch. XV of Bk. IV, entitled O f Probability". As Demonstration is the shewing the Agreement, or Disagreement of two Ideas, by the intervention of one or more Proofs, which have a constant immutable and visible connection one with another; so Probability is nothing but the appearance of such an Agreement, or Disagreement, by the intervention of Proofs, whose connexion is not constant and immutable, or at least is not perceived to be so, but is, or appears for the most part to be so, and is enough to induce the Mind to judge the Proposition to be true, or false, rather than the contrary. (IV, xv, 1)

This is not lucidly expressed and requires some imaginative exegesis. The syntax would seem to indicate plainly what is spoken of as having (or not having) "an immutable and visible connection one with another" is the proofs. But the passage makes sense only if this phrase applies to the ideas in question, and so I will read it. Where we "perceive" an agreement or disagreement of ideas, there can be no question that such a relation

46

In Defence of Epistemic Externalism

obtains. But where the "proofs" show only a probability of such a connection, it "is, or appears for the most part to be so", and so the mind is induced to judge the proposition asserting such a connection to be true. Thus far Locke has not clearly committed himself to an objective truth-conducivity of the probability established by the proofs. The disjunction of'is' with appears' blocks any such implication. And the reference to what "induces the mind" suggests persuasive efficacy, rather than a reliable indication of truth. But in an example later in the section Locke is more explicit as to a conceptual connection between probability and likelihood of truth. He has just described a person who had demonstrated the theorem and continues with the contrasting case of one who makes a probable judgment. But another Man who never took the pains to observe the Demonstration, hearing a Mathematician, a Man of credit, affirm the three Angles of a Triangle, to be equal to two right ones assents to it; i.e., receives it for true. In which case, the foundation of his Assent is the Probability of the thing, the Proof being such, as for the most part carries Truth with it: The Man on whose Testimony he receives it, not being wont to affirm any thing contrary to, or besides his Knowledge, especially in matters of this kind. (IV, xv, 1)

This says as clearly as possible that a (sufficiently) probable judgment is one whose grounds constitute a reliable indication of truth. For here the proof is "such as for the most part carries Truth with it". And it does so because of the reliability of the testifier "in matters of this kind". Not only do we find Locke taking the determinants of positive epistemic status for judgments to be truth-conducive. His way of committing himself to this would seem to qualify him as a reliability theorist of judgment in a way that would be congenial to contemporary externalists. And at the beginning of section 3 of the chapter he nails this down by saying "Probability is likeliness to be true..." The other witness I will call for the prosecution is Thomas Reid. The most relevant material is found in his discussion of what he calls "first principles" in the Essays on the Intellectual Powers of Man, or at least such of these as can be construed as principles for epistemically evaluating judgments. They all have to do with the reliability of various grounds of judgments (sorts of "evidence", to use Reid's favorite term). Some of the principles take the form of affirming the real existence of what a given cognitive faculty seems to reveal. I shall take it for granted, that I think, that I remember, that I reason, and, in general, that I really perform all those operations of mind of which I am conscious. (I, 2, p. 34) 5 ...those things did really happen which I distinctly remember. (VI, 5, p. 622) ...those things do really exist which we distinctly perceive by our senses, and are what we perceive them to be. (VI, 5, p. 625) ...I shall also take for granted such facts as are attested to the conviction of all sober and reasonable men, either by our senses, by memory, or by human testimony. (I, 2, p. 40)

Sometimes the principle is in terms of the reliability of a source of knowledge. As by consciousness we know certainly the existence of our present thought and passions, so we know the past by remembrance. (I, 2, p. 34)

5

Page references are to the edition published by the M I T Press (1969).

The "Challenge" of Externalism

47

And sometimes Reid issues a blanket endorsement of our cognitive faculties. . . . t h e natural faculties, by which we distinguish truth f r o m error, are n o t fallacious. ( V I , 5, p. 6 3 0 )

Thus, as far as the positive, truth-conducivity aspect of externalism is concerned, Reid too cuts a respectable figure as an externalist.

V Thus the result of our investigation for the thesis that externalism represents a radical departure from the epistemological tradition is a mixed bag. Assuming that the samples from that tradition I have cited are representative, we get different verdicts for the two aspects of externalism I distinguished. As for the negative aspect, the denial of internalist access restrictions on epistemizers, the contentions of Bonjour and Chisholm seem to be borne out. But the positive aspect of externalism—the inherent truth-conducivity of epistemizers—presents a quite different picture. There the tradition seems to be unambiguously on the externalist side. It seems to have no truck with claims by internalists like Chisholm, Ginet, and Foley that there is no sort of conceptual connection between epistemization and truth. As already pointed out, this feature of externalism constitutes a less sharp contrast with internalism, since a significant number of those who identify themselves as internalists, on something like the access point, also take justification to necessarily be truth conducive. But it remains that a thorough assessment of the idea that externalism represents a radical departure from the tradition must take both dimensions into consideration. It will not have escaped the reader's notice that the main thesis of this paper is a historical one. It has to do with whether, or the extent to which, contemporary epistemologists have correctly represented past epistemology when they judge contemporary externalism to be a radical departure from that epistemology. But I am also concerned with what bearing a more accurate picture of the epistemological tradition could have for contemporary epistemology. To be sure, a discipline that is constantly interacting with its past, as philosophy is, needs to make sure that those interactions are with real thinkers and not with creatures of our present historical short-sightedness. Thus it is not necessary to search for some further importance of the historical inquiry to make it relevant to philosophy today. Nevertheless, it would not be amiss to identify some specific benefits that would accrue to contemporary epistemology from getting their predecessors straighten One fall out from the historical points made in this paper is how far afield is the frequent reference to the justified-true-belief conception of knowledge as the "traditional conception", and the associated assumption that everyone took this for granted until Gettier came along and opened our eyes. This distorts the facts in two respects. First, as pointed out above, many of our predecessors, so far from advocating or employing that conception of knowledge, did not take knowledge to be a kind of belief at all, with or without further conditions. Second, it is difficult to find anything very close to the 20th century concept(s) of epistemic justification prior to that century. Though I restricted

48

In Defence of Epistemic Externalism

myself here to early modern philosophy, I believe that a survey of the medieval period would reveal a similar scene. One suspects that those who take the "traditional JTB conception" line have read Plato's Theatetus where something like that view is discussed (though rejected by Plato) and then have just assumed that it is shared by everyone prior to 1963! I doubt that the JTB conception is clearly enunciated prior to Russell in the early 20th century. A recognition of this might well lead to a consideration of other ways of construing knowledge, e.g., the intuitive conception we found in Descartes, Locke, and others. It might even lead to a welcome slackening off of attempts to find a way of supplementing a JTB account so as to avoid Gettier problems. More directly relevant to the main concerns of this paper would be the effects of a recognition that opposition to access internalist requirements is not the only thing common to all forms of externalism, and that with respect to the positive common feature of truth-conducivity, contemporary externalism is very much in line with its major epistemological forebears. This would inhibit claims, such as we find in Chisholm, that externalists "are not concerned with traditional theory of knowledge ", and the resultant attempt to expel them from the club. (Chisholm 1989, p.75) And this, in turn, would or should result in a more discriminating sense of the way in which (many) internalists and externalists share some very important concerns and, to that extent, are engaged in a common enterprise, one with deep historical roots, even though they also disagree on some important epistemological issues.

VI To return to a concern for the history for its own sake, I would like to sketch the following scene and pose a question to which it gives rise. If I have accurately portrayed 17th and 18th century epistemology, what we have there is an intimate blend of a fairly radical intemalism with respect to the subject's access to epistemic facts and a wholehearted externalist emphasis on the necessity of truth-conducivity for epistemizers. We have seen this combination in Descartes, Locke, and and Reid, and I am convinced that some additional digging would unearth the same picture in the other leading epistemologists of the period. The question is this. What led to the falling apart of this combination? Why is it that by the time we reach the latter half of the 20th century, we have the leading English speaking epistemologists divided into warring camps, one of which espouses relatively direct first person access to epistemizers but rejects a requirement for epistemizers of truth-conducivity and the other of which opts for the opposite combination of acceptance and rejection? (Along with a few, like Bonjour, who stake out a middle position and strive to combine the two.) What happened between 1800 and 1950 that led to this result? I daresay a comprehensive answer to this question would be as multi-faceted and messy as we usually find in intellectual history. But I believe that we can discern several major lines of influence. First, under the impact of absolute idealism in the 19th century and pragmatism and verificationism in the 20th, philosophers' confidence in first-person immediate knowledge of anything fell out of favor. The Kantian "lesson" that all perception, all awareness of anything, is mediated by concepts was learned all

The "Challenge" of Externalism

49

too well, as some of us think. T h e supposition that at some points facts are simply presented, given to awareness runs contrary to that dictum. This wholesale retreat from immediate knowledge in the 19th century was intimately connected with the development of coherentist epistemologies that flowered most spectacularly in the British and American absolute idealists—notably the three B's, Bradley, Bosanquet, and Blanshard. To be sure, the early Russell and G . E. Moore spearheaded a realist reaction to all this. And pockets of direct realism are to be found throughout the 20th century. But one only need mention the names of Dewey, Wittgenstein, Sellars, and Q u i n e to make the point that the tide of opposition to immediate cognition of objective facts has been widespread and powerful. This has made life difficult for epistemologists like Chisholm and Ginet with strong access internalist intuitions. It has put pressure on epistemologists committed to a truth-conducivity component to epistemizers to either abandon access restrictions or seek to construe them in a way that does not commit them to a strong form of immediate access. And it has proved difficult to find a more moderate access requirement that will accommodate internalist intuitions. A second tendency has to do more specifically with the epistemic facts that were traditionally taken to be immediately known. Even if direct knowledge is not scrapped altogether, even if we preserve it for one's present conscious states, and even for middle sized perceived objects in one's immediate environment, it has come to seem increasingly implausible that epistemic facts can be known just on reflection, as Chisholm claims. After all the fact, where it is a fact, that a certain belief of mine rests on very strong truth-conducive grounds and lacks any sufficient overriding contrary evidence is an enormously more complex and abstract matter than the fact that an image of my high school graduation exercises is currently before my mind. It has come to seem to many that if some one thinks that such facts are directly given to awareness, they are confusing direct awareness of such a fact with something like a feeling of confidence in the belief or the absence of any doubt about it. This sceptical tendency is reinforced by the general rejection of a priori knowledge of substantive matters that is typical of, but by no means restricted to, logical positivism and various forms of "analytic philosophy". If a priori knowledge is restricted to the analytic, and if knowledge of epistemic statuses would have to be apriori, then the possibility of epistemic knowledge is thrown into doubt. At least such knowledge of epistemic facts as is available to us cannot rest on direct intuition of self-evident truths. And so anything like the kind of cognitive access to the epistemic status of propositional attitudes that internalism aspires to has come to seem to many to be a mere pipe dream. A third shift has also played a role, an increasing unwillingness to restrict knowledge in such a way as to deviate sharply from what we commonly recognize as such. Early modern epistemologists do not speak with one voice about this. This is partly because of differing optimism as to what can be validated as knowledge within internalist restrictions. But it is also partly because of difficulties of interpretation. (Does Leibniz take what we ordinarily think of as knowledge of the physical world to be such?) But there are a number of clear cases of philosophers drawing strongly counter-commensensical implications about the extent of knowledge from their requirements therefor. H u m e is the most obvious example, denying that we have knowledge strictly so called of anything other than ideas and impressions. Locke's explicit account of knowledge would

50

In Defence of Epistemic Externalism

have similar consequences had he consistently applied it, but his "robust sense of reality" led him to fudge those consequences at crucial points. Berkeley drew some hardto-swallow implications from his ideas about knowledge. Reid is a signal example in the 18th century of a protest against this, but nevertheless the general temper of the period was one of "let's bite the bullet, and admit that we lack knowledge where it is lacking (by our criteria for knowledge)". For whatever reason, in the 20th century, at least among English speaking epistemologists, the dominant tendency has been to take what is commonly regarded as knowledge as a touchstone for judging the adequacy of an epistemological theory. If the theory does not imply that what we ordinarily regard as knowledge (well grounded belief) has that status, it must be revised so that it will do so. This shift is another influence in the direction of modifying severe internalist restrictions on what will count as knowledge or as strong epistemizers of belief. These tendencies go a long way toward explaining the rise of externalist epistemologies that divorce truth-conducivity from direct cognitive access, saving the former at the expense of the latter. If we can't ascertain truth-conducive positive epistemic status of beliefs by any direct awareness of anything, we will have to find more indirect ways of spotting it or give up epistemology altogether. This attitude has been reinforced by a fourth intellectual development, the growth and prominence of an empirical scientific approach to our cognitive activities. In contrast to the twin faith of the early modern period in "man's glassy essence", the complete self-disclosure of the mind to introspection, and the correlative supposition that the mind wholly consists of conscious contents, perfectly suited to be open to introspection, 20th century psychologists have increasingly seen the mind, in particular the cognitive side thereof, as involving states and processes that are inaccessible to the subject's reflective self-awareness. Consciousness is increasingly seen as the tip the iceberg, an upshot of processes that are disclosed, if at all, only in the search for those hypotheses about underlying processes that will best account for public empirical data. From this perspective, if positive epistemic status of beliefs is a matter of their being formed in a truth-conducive, reliable way, then to get the best founded account of what ways of belief formation are best founded, we must turn to empirical investigations in cognitive psychology. Hence epistemic evaluation of beliefs must be founded on the best results of cognitive science, and we must give up the illusion that, as Chisholm likes to say, they are best conducted from one's armchair.

VII So much for a bird's eye view of the dissolution of the marriage of internalist access restrictions and truth-conducivity that was characteristic of early modern epistemology, and the resultant conflict of two parties, each of which staked its claim on one of the two previously congenial partners. I will close this essay by sketching a perspective from which to view the internalism-externalism opposition. First, since as noted above, internalists themselves are divided on the issue of how central truth-conducivity is, the clearest split is over internalist access requirements. And since internalists typically bring externalist conditions into their account of knowledge, the clearest conflict over those requirements has to do with justification. So I will confine

The "Challenge" of Externalism

51

myself to the debate over whether some strong first person cognitive access to justifiers (including their justificatory efficacy) is necessary for their justifying a belief. This debate presupposes that 'justified' denotes some unique objective positive epistemic status of belief, one that is of central importance in the epistemic evaluation of beliefs. Otherwise there would be no subject about the nature of which the debaters are disagreeing. In Alston 1993 I challenged this presupposition. Here is a brief sketch of the challenge. Upon considering various persistently controverted candidates for necessary conditions for the epistemic justification of beliefs, I concluded that the best diagnosis of the situation is that we are unable to identify any one objective epistemic status concerning which the various accounts of epistemic justification are differing. This being the case, we would do better to abandon the attempt to find the objectively necessary and sufficient conditions for a belief's being "justified". I cannot undertake here to repeat the arguments for this conclusion and can only refer the reader to the article in question. But clearly if that conclusion is sound, the externalist-internalist debate is misguided. No parties to the debate are disposed to deny that having a maximally direct cognitive access to what makes for positive epistemic status of beliefs is a good, valuable, desirable thing from the standpoint of our cognitive aim of forming true rather than false beliefs on matters of interest or importance to us. Hence unless there is some master epistemic desideratum that either does or does not require, e.g., direct cognitive access thereto or truth-conducivity, then there is nothing to argue about. Both are genuine cognitive desiderata. Each is such that we are better off with it than without. In 1. 2. 3.

Alston 1993 the following epistemic desiderata were listed. Having adequate reasons, grounds, evidence... for a belief. A beliefs being based on adequate evidence A subject's having some high grade cognitive access to the grounds...for the belief. 4. Possessing higher level knowledge, or adequately grounded belief, that a belief satisfies one or another lower level desideratum. 5. A system of beliefs being "coherent" in one or another sense of that term. 6. Satisfying, or not violating, intellectual obligations in holding a certain belief. If the epistemology of belief is not properly pursued by attempting to get the correct account of "justification", then how should it be done? My suggestion was a pluralist one. We should explore the whole range of epistemic desiderata, of which the above list is a sample, explicating them, assessing their viability, considering the contexts, interests, or purposes in terms of which one or the other is more or less important, and raising questions as to what is involved in determining when one or another of these desiderata has been realized. Since externalists and internalists would be left without a central issue on which to fight each other, they would be forced to cooperate in this multifaceted consideration of a variety of epistemic desiderata. But not to worry. In those investigations they would find plenty to disagree about. It is just that the disagreements would not be centered on one master issue.

52

In Defence of Epistemic Externalism

Bibliography Alston, William P. 1986. "Internalism and Externalism in Philosophy".Philosophical Topics, XIV, 1, pp. 179221. Reprinted in Alston 1989 Alston, William P. 1989. Epistemic Justification. Ithaca, NY: Cornell University Press. Alston, William P. 1993. "Epistemic Desiderata". Philosophy & Phenomenological Research, LII, 3, pp. 527551. Bonjour, Laurence 1985. The Structure of Empirical Knowledge. Cambridge, MA: Harvard University Press. Chisholm, Roderick 1989. Theory of Knowledge, 3rd edition. Englewood Cliffs, NJ: Prentice-Hall. Cottingham, John, Robert Stoothoff, and Dugald Murdoch, trs. 1984. The Philosophical Writings ofDescartes. Cambridge: Cambridge University Press. Foley, Richard 1987. The Theory of Epistemic Rationality. Cambridge, MA: Harvard University Press. Ginet, Carl 1975. Knowledge, Perception, and Memory. Dordrecht, Holland: D. Reidel Publishing Company. Reid, Thomas 1969. Essays on the Intellectual Powers of Man. Cambridge, MA: M I T Press.

Externalism and Skepticism JOHN GRECO

Externalism in the philosophy of mind is a thesis about the representational content of mental states. The idea is that content is not wholly determined by "what goes on inside the head." Put another way, the content of a person's mental states is partly determined by the person's external environment and her relations to that environment, as opposed to being wholly determined by internal objects and processes. Hence, Content Externalism (CE): The representational content of S's mental state Μ is (at least partly) a function of what is external to S, where something is "external to S" just in case it is part of S's environment. Externalism in epistemology is a thesis about the epistemic status of belief. The idea is that epistemically normative status is not wholly determined by what is internally accessible to the believer. Put another way, epistemically normative status is partly determined by factors that are not knowable by introspection alone. Hence, Epistemic Externalism (EE): The normative status of S's belief Β is (at least partly) a function of states of affairs external to S's perspective, where something is "external to S's perspective" just in case it is not something that S could know just by introspection. Reliabilism in epistemology is a version of epistemic externalism (or EE). The idea is that knowledge requires reliability, in the sense that knowledge must be arrived at by means that are truth conducive. Moreover, the reliability of one's means need not be apparent to the knower. That is, it need not be the case that one knows (or even believes) that one's belief was formed by a reliable means. Hence,

Reliabilism (R): Ra. S knows ρ only if S's believing ρ is the result of reliable cognitive processes. Rb. There is no requirement on knowledge that one knows (or even believes) that one's belief is the result of reliable cognitive processes.1 Another hallmark of reliabilism is that it does not require that knowledge be based on reasons. Thus while some reliable (and hence knowledge-producing) cognitive processes involve basing belief on reasons, not all reliable (and hence knowledge-producing) cognitive processes do. What is essential for positive epistemic status is adequate reliability, not adequate reasons. Finally, when knowledge is based on reasons, reliabilism does not

54

In Defence of Epistemic Externalism

require that the reasons be more than contingently reliable. That is, reliabilism does not require a logical or quasi-logical relation between reasons and belief. That is, it does not require a deductively valid relation, or even a relation of logical probability. Hence, Rc. Where S knows ρ on the basis of reasons R, there need be no necessary relation (logical or quasi-logical) between the truth of R and the truth of p: it is sufficient that the truth of R is a contingently reliable indicator of the truth of p. The standard view in epistemology is that EE is sufficient for rejecting skepticism. This is because, according to the standard way of thinking, the best skeptical arguments depend on internalist assumptions about knowledge and reasons. If externalism is true, then those assumptions are false, and hence the best skeptical arguments can be rejected as unsound. I can say that this is the "standard" view in epistemology because even internalists and skeptics tend to accept a version of it: i.e., they tend to accept that the best skeptical arguments require internalist assumptions, and that if externalism is true, then those assumptions are false.2 It is also part of the standard view in epistemology that content externalism (or CE) is not centrally relevant to the problem of skepticism. More precisely, the standard view is that the best skeptical arguments do not essentially depend on content internalism, and so cannot be rejected by adopting content externalism. 3 I would add that EE is not only sufficient for rejecting skepticism, but necessary as well. This is because, according to my way of thinking, the best skeptical arguments do not depend on further objectionable assumptions. That is, the only thing wrong with the best skeptical arguments is their (epistemically) internalist assumptions, and so if those assumptions are granted, the arguments must be sound. I would add, further, that a non-skeptical epistemology must accept reliabilism (or R): it is not enough to be an externalist to reject skepticism—you have to be a reliabilist as well. This, I will argue

1

2

3

The typical knower, one might think, does not typically have any second-order beliefs about the sources of her beliefs. (Here we must distinguish between currently having a (perhaps unconscious) second-order belief, and being disposed to form a second-order belief in response to a relevant query.) For example, see Laurence Bonjour, The Structure of Empirical Knowledge (Cambridge, MA: Harvard University Press, 1985); and Richard Fumerton, Metaepistemology and Skepticism (Lanham, M D : Rowman and Littlefield, 1995). See for example Keith DeRose, "How Can We Know that We're Not Brains in Vats?" The Southern Journal of Philosophy XXXVIII (2000) Supplement: 121-48; and Brian McLaughlin, "Skepticism, Externalism, and Self-Knowledge," The Aristotelian Society, Supplementary Vol. LXXIV, (2000): 93-118. See also DeRose's introduction to Skepticism: A Contemporary Reader, Keith DeRose and Ted Warfield, eds. (Oxford: Oxford University Press, 1999). A number of philosophers have argued that content externalism shows certain standard skeptical scenarios to be impossible; for example, any scenario entailing that no external things exists. But as McLaughlin notes, it is widely acknowledged (even among these philosophers) that skeptical arguments can be reconstructed so that they invoke no such scenarios. As examples, McLaughlin cites Ted Warfield, "A Priori Knowledge of the World: Knowing the World by Knowing our Minds," Philosophical Studies (1998), reprinted in DeRose and Warfield; and Collin McGinn, Mental Content (Oxford: Blackwell Press, 1992).

Externalism and Skepticism

55

below, is the lesson of Hume's problematic. In effect, Hume's skeptical arguments correctly assume that Ra is true, and show that Rb and Rc are true as well.4 John McDowell's recent work challenges the standard view in epistemology. In fact, McDowell's work might seem to turn the standard view on its head, claiming that CE is sufficient for rejecting skepticism and that EE is not particularly relevant. Moving to my additions to the standard view, McDowell clearly rejects R, and clearly rejects the thesis that any non-skeptical epistemology must accept R. In Part 1 of this essay I will try to clarify McDowell's position and its relationship to what I have called "the standard view in epistemology." Here I will argue that McDowell's position is indeed inconsistent with the standard view, since he thinks that his version of CE is sufficient for rejecting skepticism. Despite rhetorical appearances, however, McDowell also accepts EE. In fact, I will argue, EE plays an important role in McDowell's response to skepticism. In Part 2 I will argue that, contra McDowell, accepting R is necessary for rejecting skepticism, and accepting CE is not sufficient for rejecting skepticism. Of course these two theses are logically related: CE does not imply R, and so if R is necessary for rejecting skepticism, then CE cannot be sufficient. A corollary of Part 2, then, is that the standard view in epistemology is correct.

Part 1. McDowell's Response to Skepticism McDowell's response to skepticism is original, complex and somewhat difficult to figure out. 5 At times, McDowell emphasizes Sellars's idea that "knowledge must be in the space of reasons," and that therefore experience can ground knowledge only if it can function as a reason. 6 This theme, in turn, leads McDowell to place great emphasis on the idea that experience must be conceived as having conceptual content. It is only by virtue of having such content, he argues, that experience can function as a reason. This

4

5

6

I defend these theses at length in "Agent Reliabilism," Philosophical Perspectives, 13, Epistemology, James Tomberlin, ed. (Atascadero, CA: Ridgeview Press, 1999); and in Putting Skeptics in Their Place: The Nature of Skeptical Arguments and their Role in Philosophical Inquiry (New York: Cambridge University Press, 2000). Let me forestall an objection from the start. In various places McDowell claims that he is not trying to "answer" or "refute" the skeptic, and so one might think that his position is not supposed to respond to skepticism at all. But what McDowell means in such passages is that he is not trying to answer the skeptic on his (i.e. the skeptic's) own terms. Rather, McDowell's strategy is to identify assumptions in the skeptical reasoning that we need not accept, or might even have reason to reject. See for example "Knowledge and the Internal," Philosophy and Phenomenological Research 55 (1995), p. 888, note 19. As far as I can see, this is the only sane strategy for responding to skepticism that one could adopt. Since the skeptical arguments are typically valid, it is impossible that one might answer the argument on its own terms—not if this means rejecting the skeptical conclusion without challenging a skeptical assumption. And notice that this is not a comment on the strength of the skeptic's arguments—it is rather a simple point about the nature of logical validity. See Wilfred Sellars, "Empiricism and the Philosophy of Mind," in Science, Perception and Reality (London: Routledge, 1963).

56

In Defence of Epistemic Externalism

idea gets such great play in Mind and World that it can seem to be the governing idea in McDowell's response to skepticism. Here are some relevant passages. In the view I am urging, the conceptual contents that sit closest to the impact of external reality on one's sensibilities are not already, qua conceptual, some distance away from that i m p a c t . . . .the conceptual contents that are most basic in this sense are already possessed by impressions themselves, impingements by the world on our sensibility. 7 I have claimed that we are prone to fall into an intolerable oscillation: in one phase we are drawn to a coherentism that cannot make sense of the bearing of thought on objective reality, and in the other phase we recoil into an appeal to the Given, which turns out to be useless. I have urged that in order to escape this oscillation, we need a conception of experience as states or occurrences that are passive but reflect conceptual capacities. ... 8

But that cannot be the whole story regarding McDowell's response to skepticism, or even a very important part of the story. For the skeptical argument from illusion can run on a "thick" conception of experience as well as it does on a "thin" conception. Suppose that experiences are thick appearings: for example, suppose that S has a treetype experience only if something appears to S as a tree, only if S's experience is thick with the conceptual content that Here is a tree. The skeptic can still insist that illusion is possible, and that therefore perceptual knowledge requires something more than an appearance so conceived. What more? The most common way to continue the skeptical line of reasoning is to insist that there must be some good inference from appearance to reality—some good inference from something's appearing to be a tree to something's being a tree—and to insist that there is no good inference available. To be clear, I do not mean to defend the skeptical line of reasoning. My point, rather, is that the reasoning proceeds as well on a conception of experience as thick as it does on a conception of experience as "thin." And so insisting that experience is thick with conceptual content cannot give us resources for responding to skepticism. For all the play that the idea of contentful experience gets in Mind and World, however, it is not the governing idea in McDowell's response to skepticism. For that we must turn to earlier work, where he emphasizes a "disjunctive" account of experience over a "highest common factor" account. The highest common factor account of perceptual experience begins with the idea that perception and illusion can look the same from the point of view of conscious awareness. That is, non-deceptive episodes of perception and deceptive episodes of illusion can be subjectively indistinguishable. From this it is concluded that perception and illusion must share a highest common factor, and it is this highest common factor that functions as the basis of perceptual knowledge. McDowell's disjunctive account rejects this picture. On his account, a mere appearance in illusion is ontologically distinct from the actual appearance of an object in perception.

7 8

Mind and World ( C a m b r i d g e , M A : Harvard University Press, 1994), pp. 9-10. M i n d and World, p. 23.

Externalism and Skepticism

57

But suppose we say—not at all unnaturally—that an appearance that such-and-such is the case can be either a mere appearance or the fact that such-and-such is the case making itself perceptually manifest to someone. As before, the object of experience in the deceptive cases is a mere appearance. But we are not to accept that in the non-deceptive cases too the object of experience is a mere appearance, and hence something that falls short of the fact itself. On the contrary, we are to insist that the appearance that is presented to one in those cases is a matter of the fact itself being disclosed to the experiencer.9 Focus now on a sentence that is somewhat obscure in the context of the present passage: "As before, the object of experience in the deceptive cases is a mere appearance. But we are not to accept that in the non-deceptive cases too the object of experience is a mere appearance, and hence something that falls short of the fact itself." Here the claim seems to be that perceptual experience and illusory experience do not share a c o m m o n intentional object: that is, they are not about the same thing. Rather, the perceptual experience manages to be about the perceptual object (an object in the external world) in a way that mere illusion does not. Put another way, according to McDowell's disjunctive account of experience, the representational content of a perceptual experience is different from the representational content o f a mere illusion, in that the perception, but not the illusion, manages to be about an actual, external object. In other words, McDowell is arguing for a version o f content externalism about perceptual experience. This reading is put out of doubt by the following passage. If we lift Russell's restriction [on object-dependent propositions], we open the possibility that a subject may be in error about the contents of his own mind. ... One reason, then, to pursue a less restricted conception of object-dependent propositions is the interest of its radically anti-Cartesian implications. In a fully Cartesian picture, the inner life takes place in an autonomous realm, transparent to the introspective awareness of its subject; the access of subjectivity to the rest of the world becomes correspondingly problematic, in a way that has familiar manifestations in the mainstream of post-Cartesian epistemology. If we let there be quasi-Russellian singular propositions about, say, ordinary perceptual objects among the contents of inner space, we can no longer be regarding inner space as a locus of configurations which are self-standing, not beholden to external conditions; and there is no question of a gulf, which it might be the task of philosophy to try to bridge, or declare unbridgeable, between the realm of subjectivity and the world of ordinary objects. 10 T h e line of argument is clear: Traditional skeptical arguments must regard "inner space as a locus of configurations which are self-standing, not beholden to external conditions." If we reject this content internalism in favor of content externalism, traditional skeptical arguments do not have the resources to get off the ground. O r as McDowell puts it, "Without the 'highest c o m m o n factor' conception of experience . . . the traditional problems lapse." 1 1

9 10 11

"Criteria, Defeasibility, and Knowledge," Proceedings of the British Academy 6 8 ( 1 9 8 2 ) , p. 4 7 2 . "Singular T h o u g h t and the Extent o f Inner Space," in P. Pettit and J . M c D o w e l l , ed., Subject, Thought and Context (Oxford: Clarendon Press, 1986), pp 145-6. "Criteria, Defeasibility, and Knowledge," p. 4 7 9 .

58

In Defence of Epistemic Externalism

I have been arguing that the governing idea in McDowell's response to skepticism is his disjunctive account of experience, along with its attendant content externalism. But I claimed above that epistemic externalism plays an important role in McDowell's response to skepticism. How so? One way to see the point is to notice that McDowell's version of content externalism entails epistemic externalism. According to McDowell, genuine perception gives one "a satisfactory standing in the space or reasons," whereas mere illusion does not. In other words, genuine perception, but not illusion, gives rise to positive epistemic status. But on McDowell's view, the difference between genuine perception and mere illusion is not something that is knowable by introspection alone. Hence, McDowell's view entails EE. The role of EE in McDowell's response to skepticism becomes explicit elsewhere. In "Knowledge and the Internal," he argues against a "hybrid view" of knowledge, defined as one that makes positive epistemic status (or, in McDowell's terms, "satisfactory standing in the space of reasons") a function of what is knowable by reason alone. Although the motivation I have suggested for interiorizing the space of reasons is intelligible, it is surely something we ought to find suspect. The hybrid view's concession to luck, tagged on to a picture of reason as self-sufficient within its own proper province, comes too late; the very idea of reason as having a sphere of operation within which it is capable of ensuring, without being beholden to the world, that one's postures are all right ... has the look of a fantasy, something we spin to console ourselves for the palpable limits on our powers. To avoid fantasy, we would need to see our way to accepting that we cannot eliminate what the interiorized conception of reason conceives as a quite alien factor, the kindness of the world, as a contributor to our coming to occupy epistemically satisfactory positions in the space of reasons.12 The consideration I have offered suggest a way to respond to skepticism about, for instance, perceptual knowledge; the thing to do is not to answer the skeptics challenges, but to diagnose their seeming urgency as deriving from a misguided interiorization of reason.13 We have now canvassed three aspects of McDowell's response to skepticism: a) the idea that experience is thick with conceptual content, b) a disjunctive account of experience and its attendant content externalism, and c) epistemic externalism about perceptual knowledge. The relation among them, I take it, is this: McDowell's content externalism both entails his epistemic externalism about perceptual knowledge and supports his idea that perceptual experiences have conceptual content. Regarding the latter relation, he writes, Indeed, it is arguable that the 'highest common factor' model undermines the very idea of an appearance having as its content that things are thus and so in the world 'beyond' appearances (as we would have to put it). ... It would be wrong to suppose that the 'highest common factor' conception can capture, in its own terms, the intuition that I express when I say that the fact itself can be manifest to experience: doing so by saying that that is how

12 13

"Knowledge and the Internal," p. 886. "Knowledge and the Internal," p. 890. See also "Criteria, Defeasibility, and Knowledge," pp. 475-6.

Externalism and Skepticism

59

it is when, for instance, experiences as o f its raining are in a suitable way the u p s h o t o f the fact that it is raining. T h a t captures the intuition all right; b u t — w i t h 'experience as o f its r a i n i n g ' — n o t in terms available to s o m e o n e w h o starts by insisting that the object o f experience is the highest c o m m o n factor, a n d so falls short o f the fact itself. 1 4

This reading explains McDowell's claim that, "Without the 'highest common factor' conception of experience ... the traditional problems lapse." 15 McDowell's claim can now be cashed out as follows: Once we replace the 'highest common factor" account of experience (and its content internalism) with a disjunctive account (and its content externalism), this puts in place both epistemic externalism and an account of experience as conceptually thick and already representing the world. This, in turn, gives us resources for responding to various lines of skeptical reasoning. We can emphasize different aspects of the entire picture as needed.

Part 2. Reliabilism

and

Skepticism

I said that McDowell's response to skepticism is original and complex. Understood as above, it is also largely compelling. But it is not, I think, exactly right. In particular, I do not think that McDowell's view gives us all the resources that are needed to respond to skepticism. This is because it does not give us the resources to respond to Hume's skeptical reasoning—a "traditional problem" par excellence, but one to which McDowell gives little attention. I will turn to Hume's reasoning shortly. But first, I want to be more specific about the problem that (I claim) Hume's reasoning raises for McDowell's view. Put simply, Hume's skeptical reasoning shows that any non-skeptical epistemology must be reliabilist. McDowell's view entails epistemic externalism, but it does not entail reliabilism. Therefore, McDowell's view does not have the resources to respond adequately to Hume's skeptical reasoning. Put differently, content externalism does not entail reliabilism, and so adopting content externalism is not sufficient for responding to Hume's skeptical reasoning. 16 The problem I want to review in Hume is most commonly associated with skepticism about induction. The basic idea is that inductive reasoning cannot give rise to positive epistemic status and knowledge, because the reliability of induction cannot be

14 "Criteria, Defeasibility, and Knowledge," p. 474. See also "Knowledge and the Internal," pp. 890-1. 15 "Criteria, Defeasibility, and Knowledge," p. 479. 16 The problem is worse for McDowell, in fact. This is because McDowell's account of positive epistemic status is inconsistent with reliabilism, and therefore McDowell cannot respond to Hume. In other words, it is not just that McDowell must add something to respond to Hume—rather, what must be added is inconsistent with McDowell's account of the epistemically normative. Although this is a problem for McDowell's account of the epistemically normative, however, it is not a problem for content externalism in general. That is, content externalism need not be wedded to McDowell's additional views about epistemic normativity, which include his Sellarsian, anti-reliabilist commitments.

60

In Defence of Epistemic Externalism

vindicated in a non-circular way. A second skeptical argument in Hume, this one about perceptual knowledge, gets less attention. But the argument is structurally identical to the one about induction. Specifically, it is argued that perception cannot give rise to positive epistemic status and knowledge, because the reliability of perception cannot be vindicated in a non-circular way. Lets take a look at that argument first, and consider how McDowell would respond to it. Here is Hume: By what argument can it be proved, that the perceptions of the mind must be caused by external objects. . ? ... It is a question of fact, whether the perceptions of the sense be produced by external objects, resembling them: how shall this question be determined? By experience surely; as all other questions of a like nature. But here experience is, and must be entirely silent. The mind has never anything present to it but the perceptions, and cannot possibly reach any experience of their connexion with objects. The supposition of such a connexion is, therefore, without any foundation in reasoning. 1 7

It should be clear how McDowell would respond. On the one hand, he would reject Hume's assumption that "It is a question of fact, whether the perceptions of the sense be produced by external objects, resembling them." On McDowell's view, it is the very nature of perceptual experience that a) it is the appearing of an external object, and b) that it represents external objects as they actually are. And of course McDowell would reject Hume's second assumption as well: that "The mind has never anything present to it but the perceptions, and cannot possibly reach any experience of their connexion with objects." The point of McDowell's content externalism is that the mind does have external objects "present" to it, and that its "connection with objects" in perception is therefore direct, and not in need of any "foundation in reasoning." Hume's reasoning can be reconstructed more formally as follows. (HP) 1. All beliefs resulting from perception depend for their evidence on a) perceptual experience, and b) some assumption to the effect that perceptual experience is a reliable indication of external reality. Let us call this assumption the "Perceptual Reliability Principle" (PRP). 2. But PRP is itself a belief about the external world, and therefore can be known, if at all, only by perception. 3. Therefore, PRP depends for its evidence on itself. (1,2) 4. Circular reasoning cannot give rise to knowledge. 5. Therefore, PRP is not known. (3,4) 6. Therefore, there can be no perceptual knowledge. (1,5)

17 Enquiries concerning Human Understanding and concerning the Principles Selby-Bigge, ed., (Oxford: Clarendon Press, 1975), pp. 152-153.

of Morals, third edition, L. A.

Externalism and Skepticism

61

McDowell can deny both premise (1) and premise (2) of (HP): Perception does not depend for its evidence on PRP, and PRP is not known by perception. So far, so good. But how is this sort of response supposed to transfer to Hume's skeptical reasoning about induction? That is not at all clear. In fact, it seems clear that McDowell's response does not transfer. Put simply, on McDowell's view perception is by nature veridical, and it is therefore by nature reliable. But it is not at all plausible that inductive reasoning is by nature reliable. In fact, that is perhaps the most important lesson of Hume's skeptical reasoning about induction: that inductive reasoning is at most contingently reliable, as opposed to reliable by nature. According to Hume, induction is reliable only if the world is regular, but it is not a necessary truth that the world is regular: there are possible worlds where things are more chaotic, and where therefore past and present observations are not a good indication of unobserved events. Put another way, the reliability of induction depends on how the world actually is. 18 We can see Hume's point more clearly, and we can see more clearly that McDowell's position does not address it, if we reconstruct Hume's reasoning as follows. (HI) 1. All beliefs resulting from induction depend for their evidence on a) past and present observations, and b) some assumption to the effect that past and present observations are a reliable indication of unobserved events. Let us call this assumption the "Reliability of Induction Principle" (RIP). 2. But RIP is itself a belief about unobserved events, and therefore can be known, if at all, only by induction. 3. Therefore, RIP depends for its evidence on itself. (1,2) 4. Circular reasoning cannot give rise to knowledge. 5. Therefore, RIP is not known. (3,4) 6. Therefore, there can be no inductive knowledge. (1,5) How can McDowell respond to this sort of reasoning? The analogue to his response to (HP) is to deny premises (1) and (2); i.e. to deny that inductive reasoning depends on RIP, and to deny that RIP can only be known by inductive reasoning. But neither of these responses is open to McDowell. First, consider premise (1) of (HI). Hume thinks (1) is true because he thinks that having reliable evidence is not enough for knowledge—one must also be aware that one's reasons are reliable. But McDowell agrees with this, and therefore is committed to premise (1) of (HI). 19 What about premise (2) of (HI): that RIP is itself a belief about unobserved events, and therefore can be known, if at all, only by induction? Denying the analogous prem-

18

19

The present point does not depend on the claim that there could be worlds with no regularity at all. It is sufficient that the world need not be as regular as it is, and so induction need not be reliable enough to generate positive epistemic status and knowledge. This comes out clearly in McDowell's rejection of "full blown externalism," a defining characteristic of which is that it makes reliability an external constraint on knowledge. Cf. "Knowledge and the Internal,"

pp. 881-4.

62

In Defence of Epistemic Externalism

ise in (HP) was plausible, and it was made plausible by McDowell's disjunctive account of perceptual experience. According to that account, the reliability of perceptual experience is built into the very nature of perceptual experience itself, and so McDowell can plausibly deny that PRP is known only by perception. But how does that line of reasoning help with premise (2) of (HI)? First, it seems irrelevant to that premise. Moreover, the premise seems perfectly correct anyway: How else could one know such a thing as RIP except by inductive reasoning? Certainly it is not the kind of fact that can "make itself perceptually manifest to someone" or that can "be disclosed to the experiencer" through perception. But neither is it the sort of thing that can be known a priori, since Hume is right that induction is not reliable by nature. He is right that the reliability of induction depends on how the world actually is, and therefore cannot be known by a priori means. A corollary of this last point is that there is no necessary relation between inductive evidence and inductive conclusions. Put another way, if our inductive evidence is a reliable indication of the conclusions based on that evidence, then this is at most a contingent fact about us and about our relation to the world. And as a contingent fact, there is no knowing it a priori. O n this point, at least, Hume's reasoning seems to be sound. 2 0 If a response to Hume's reasoning is not available to McDowell, then how should one respond? It seems to me that the only adequate way to respond is to embrace reliabilism, and to thereby deny premise (1) of Hume's reasoning. Recall why premise (1) seems plausible to Hume and McDowell: both accept (1) because they think that having reliable evidence is not enough for knowledge—one must also be aware that one's

20

It would be remiss to ignore McDowell's explicit response to Hume's skeptical reasoning. In the passage that follows, McDowell is considering how his response to the Argument from Illusion can be extended to Hume's skepticism about induction. N o w my fourth Sellarsian idea can be put like this: there cannot be a predicament in which one is receiving testimony from one's senses but has not yet taken any inductive steps. To stay with the experience of color, whose simplicity presumably makes it maximally favorable to the contrary view: color experience's being testimony of the senses depends on the subject's already knowing a great deal about, for instance, the effect of different sorts of illumination on color appearances. . . . This makes for an easy extension to inductive skepticism of the epistemological move I have been recommending. . . . Nothing could be immediately present to one's senses unless one already had knowledge that goes beyond what is immediately present to the senses. So the supposed predicament of the inductive skeptic is a fiction. ("Knowledge and the Internal," p. 891) T h i s line of argument can be reconstructed as follows. 1.

"there cannot be a predicament in which one is receiving testimony from o n e s senses but has not yet taken any inductive steps." In other words, there could not be contentful observations unless one also has inductive knowledge, or knowledge about what is unobserved. 2. But there is contentful observation (The skeptic grants that much when he tries to set up the problem). Therefore, 3. There is inductive knowledge: the predicament in which one has contentful observation but no inductive knowledge is a fiction. (1,2) T h e problem with the above line of reasoning is this: even if it is exactly right, it does nothing to show where Hume's argument is wrong. And since it does not, it does nothing to show how McDowell's acc o u n t (contra H u m e ) makes inductive knowledge possible.

Externalism and Skepticism

63

reasons are reliable. But reliabilism denies this. According to Rb, there is no requirement on knowledge that one knows that one's beliefs are the result of reliable cognitive processes. Reliabilism also allows us to accept what is right about Hume's reasoning: that the reliability of induction depends on how the world actually is, and that therefore there is no necessary relation between inductive evidence and inductive conclusions. According to Rc, no such relation is needed for knowledge: Where S knows ρ on the basis of reasons R, there need be no necessary relation (logical or quasi-logical) between the truth of R and the truth of p: it is sufficient that the truth of R is a contingently reliable indicator of the truth of p. What about thesis Ra: that S knows ρ only if S's believing ρ is the result of reliable cognitive processes? McDowell is committed to Ra because he denies Rb, and the denial of Rb entails Ra. In any case, I will assume that there are independent reasons for accepting Ra, and that therefore Ra is an uncontroversial aspect of reliabilism. 21 The above considerations show that adopting reliabilism is sufficient for rejecting Hume's reasoning in (HI). I will now add that it seems to be necessary as well. As far as I can see, there is no other way to challenge Hume's reasoning. We may put the point this way: Ra is uncontroversial, and if we reject either Rb or Rc, then Hume has all the resources he needs for his skeptical reasoning to go through. 2 2

Conclusion The above considerations vindicate what I have called "the standard view in epistemology": that epistemic externalism is sufficient for rejecting skepticism while content externalism is not. To see this, we need only note that content externalism does not entail reliabilism, but that reliabilism is necessary for rejecting to skepticism. The above considerations also vindicate the additions that I suggested to the standard view: that epistemic externalism and reliabilism are necessary for rejecting to skepticism. To see this, we need only note that reliabilism is necessary, and that reliabilism entails epistemic externalism. 2 3

21 22

23

Even internalists should accept Ra. Internalists distance themselves from reliabilism by rejecting something like Rb and Rc. W h y does rejecting Rc give H u m e all he needs? Because there is a necessary relation between inductive evidence and belief only if something like RIP is part of that evidence. But if we need to know RIP as part of our evidence, then premise (1) of (HI) is true and Hume's reasoning goes through. I would like to thank Bill Brewer, Bill Jaworski, Duncan Pritchard, Richard Schantz and Ernest Sosa for their comments on earlier drafts.

Counterexamples to Epistemic Externalism Revisited T H O M A S GRUNDMANN

In contemporary analytic epistemology, there is a great deal of meta-epistemological discussion about the proper conception of justification. Epistemological externalists argue that a belief is justified if it has been produced by standards, methods or processes that are reliable and produce mostly true beliefs. 1 Epistemological internalists on the other hand impugn the justificatory power of such purely objective facts. For them, the first person perspective on facts is decisive for justification. Internalists typically claim that the objective reliability of the method in question is neither necessary nor sufficient for justification. 2 This discussion between externalists and internalists is not just an obscure matter for specialists; rather, it cuts to the very heart of epistemology. As long as we have no clear notion of what epistemological justification is, we will be unable to say what structure it has, what its sources are, or how extensive our justified beliefs are. Thus meta-epistemology has methodological primacy over substantial epistemology. 3 What is the nature of epistemological justification? Who's right—the externalist or the internalist? In answering questions like these, philosophers often avail themselves of conceptual analysis. Having appraised whether a given concept X is intuitively satisfied in a broad spectrum of imagined situations, one can make a substantive claim about the necessary and sufficient conditions for X. When we begin thinking about epistemological justification, we arrive rather quickly at a fundamental intuition: methods justify beliefs in an epistemological sense if they are efficacious means to attaining true beliefs. 4 A means is efficacious iff it brings about the desired end or at least makes the effectuation of that end probable. So an efficacious means for attaining truth is an objectively reliable belief-producing method, the results of which are usually true. A belief is therefore epistemologically justified if it has been formed by a method or process that generates mostly true beliefs. Clearly, this instrumentalist viewpoint directly supports externalism. But internalists have attacked this fundamental intuition with a number of notorious counterexamples. 5 These counterexamples are intended to demonstrate that objective reliability is neither necessary nor sufficient to satisfy our intuitive conception of justification. Surprisingly, a number of externalists have acknowledged the existence

1 2 3 4 5

See Goldman on externalism 1979 and more recently Sosa in Bonjour & Sosa 2003. See Bonjour on internalism 1985, Chap.3; Lehrer 1990; Foley 1993 and Bonjour in Bonjour & Sosa 2003. This point is emphasized by Alston 1989, p.82. See also Bonjour 1985, p.7f. See the summary found in Bonjour 2002, p.228-231.

66

In Defence of Epistemic Externalism

of internalist intuitions in these cases.6 How can this be reconciled with externalism? One faction of externalists—namely the revisionists—concedes that our concept of justification is internalist, but does not consider this concept particularly significant to the nature of justification.7 Its proponents suggest a scientific revision of the concept. From this perspective, externalism is recommendable for the simple reason that it makes justification into a purely descriptive and empirically investigable property. Another faction of externalists tries to marginalize the internalist intuitions, although its proponents do not go so far as to throw out conceptual analysis as a philosophical method. On this view, the conflict between externalist and internalist intuitions reflects a deep ambiguity at the heart of our pre-theoretic concept of justification.8 Since the concept is deeply ambiguous, externalist epistemology remains—like internalism—a respectable project. Neither of these externalist reactions seems very promising to me. The revisionist is right that the nature of a thing may not lurk in our concept of it. 9 But it does not by any means follow that the concept is completely irrelevant to understanding the thing. At the very least, it establishes what we are talking about; it fixes the reference. If we skip over the concept when we begin investigating something, there is no dismissing the suspicion that we have simply changed the subject. 10 Proponents of the ambiguity thesis run into serious problems as well. If our concept of justification really were ambiguous in the sense they suggest, the result would be a far-reaching bifurcation of our epistemological projects. And which of the projects—the externalist or the internalist—would have priority in case they come into conflict with each other? To avoid such problems, the assumption of a fundamental ambiguity in our concepts should be regarded as a last resort. We should first exhaust all the possibilities for dissolving the apparent conflict of our intuitions.11 So I would like to take a closer look at the so-called intuitive counterexamples against epistemological externalism. My thesis is that the truly internalist intuitions that crop up in certain counterfactual cases can be explained away, and that the rest turn out not to contradict externalism at all, since close inspection reveals that they rest on a tendentious description of the situations. Our evaluation of these cases can in fact be reconciled with a plausible version of externalism. First I examine cases like Bonjour's Norman that are intended to demonstrate the insufficiency of reliable belief-production for justification. Then I look at the demon-scenarios meant to show that reliability is also unnecessary for justification.

6

7 8 9 10 11

This even led one of the main proponents of externalism, Alvin Goldman in Goldman 1986, temporarily to abandon externalism. Acknowledgement of internalist intuitions by externalists can be found in Goldman 1988, Engel 1992, Plantinga 1993. This position is argued especially emphatically by Kornblith 1995, p.243f. See Bach 1985, Goldman 1988, Engel 1992, Foley 1993 and Alston 1993. Kripke's and Putnam's examples of natural kinds have demonstrated this especially clearly. See Grundmann 2 0 0 3 , p.34. For a thorough critique of the ambiguity thesis, see Grundmann 2 0 0 3 , p . 3 0 0 - 3 0 3 .

Counterexamples to Epistemic Externalism Revisited

67

I

Bonjour chooses the case of Norman to show that reliability is insufficient for justified belief. In The Structure of Empirical Knowledge he describes the case as follows: N o r m a n , under certain conditions which usually obtain, is a completely reliable clairvoyant with respect to certain kinds o f subject matter. H e possesses no evidence or reasons o f any kind for or against the general possibility of such a cognitive power or for or against the thesis that he possesses it. O n e day N o r m a n comes to believe that the President is in N e w York City, though he has no evidence either for or against this belief. In fact the belief is true and results f r o m his clairvoyant power under circumstances in which it is completely reliable. 1 2

So Norman formed the belief (p) that the president is in New York City by a process which we would consider unreliable, but which in this counter factual case is stipulated to be reliable. He has no evidence for the reliability of his clairvoyant power, neither is there any evidence that would undermine it (undercutting defeater). But he also lacks internal evidence for or against the belief that ρ (rebutting defeater). 13 In fact, from his perspective there is nothing to speak for or against the process that yielded p. According to Bonjour, we would not intuitively be inclined to classify Norman's belief (p) as justified. Can we accept this assessment? First let's consider some similar cases. Assume that Cora has reliable visual perception under normal conditions. She has however no evidence for or against the reliability of sight in general or in her individual case. She simply does not take much interest in such abstract questions. In a particular situation, she sees a car driving towards her and acquires the belief (q) that a car is driving towards her. She has no internal evidence that would speak against the truth of (q) and her visual perception is functioning reliably in this situation. I submit that we would deem Cora's belief justified. It is true that Cora's situation is not fully equivalent to Norman's, since she has internal evidence in the form of sense experience, whereas Norman lacks this ex hyothesis. But is this sufficient reason to evaluate the two cases differently? To take another example, let's imagine Simon has the reliably formed introspective belief (r) that he is currently having a particular thought. He does not have any evidence for or against the reliability of introspection, and he has no rebutting defeater against his belief that r. But since introspection is of course not an evidential process—one does not form the belief that one is having a particular thought by means of inner perception or something of the sort—Simon has no internal evidence for his belief either. Nevertheless, it seems to me we have the intuition that Simon's belief (r) is justified. And if this is our assessment of Simon's case, we should also take Norman's belief about the president to be justified since his situation is absolutely equivalent to Simon's.

12

B o n j o u r 1985, p . 4 l . T h e similar case o f Mr. Truetemp is described by Lehrer 1990, p. 163.

13

Bonjour's term "evidence" has to be referring to internal evidence, since the example might otherwise become inconsistent. Evidence is internal if it consists in introspectively accessible states that rationalize a belief. If "evidence" could also be understood in an externalist manner, then the reliability o f Norman's clairvoyance would present evidence for belief (p). T h i s however is excluded ex hypothesi.

68

In Defence of Epistemic Externalism

But Bonjour supplements the general description of the situation with a more subtle consideration to support the claim that Norman's belief is unjustified. 14 He asks what would happen if Norman were to reflect critically about why he has the belief that p. Either he would form the meta-belief that his clairvoyant power is reliable, or he would not. If he did form the meta-belief, it would be ex hypothesi unjustified, since Norman has no evidence for the reliability of clairvoyance. Hence the belief (p), which would be dependant on the unjustified meta-belief, would also be unjustified. If, on the other hand, Norman did not form this meta-belief, then the belief that ρ would be inexplicable and therefore unjustified. This addition that Bonjour makes to the general description does make his assessment plausible, but it is in my view tendentious. We have to distinguish two cases. In the first, Norman has often formed clairvoyant beliefs in the past 15 ; in the second, he forms his belief about the president by a novel process. 16 If Norman has often had clairvoyant beliefs and noticed it, then he would be justified in forming the meta-belief that he had come to believe (p) by a reliable clairvoyant process. But how can that be, if Norman ex hypothesi lacks all evidence for the reliability of clairvoyance? Well, Norman takes the propositions that he arrives at by clairvoyance to be true. So he believes on the basis of clairvoyance that what he discerns by clairvoyance is true. In any case, he recognizes by introspection that he has clairvoyant beliefs. If he notices that he often has true clairvoyant beliefs, then he can induce that he has reliable clairvoyant powers. The justification of this meta-belief occurs through a form of bootstrapping that is epistemically circular, but there is nothing objectionable about that if one has not already excluded externalism at the outset. Seen in this light, Bonjour s description of the case appears tendentious, since he does not consider the possibility that someone who forms beliefs by a reliable process is always able to reflectively justify the meta-belief that he forms these beliefs by a reliable process. In order to do this, the subject does not need any evidence that is independant of the process in question, as Bonjour seems to assume. But what about the second case, where the subject unknowingly acquires a completely novel epistemic faculty? Under these conditions, one can not make the kind of circular argument that invokes clairvoyant beliefs to support the reliability of clairvoyance, since induction presupposes a multitude of corroborating cases. So if Norman asks himself how he could possibly know where the president is at the moment, he will simply have nothing to say: "From his standpoint, there is apparently no way in which he could know the President's whereabouts." "there is no way, as far as he knows, for him to have obtained this information" 17

14 The following argument is from Bonjour 1985, p.42. 15 This corresponds to the characterization of the case given in Bonjour 2002, p.230f: "He (Norman, T.G.) frequently has spontaneous beliefs or hunches (...)." Lehrers description of Mr. Truetemp is similar. 16 This is how Bonjour originally described the case in Bonjour 1985. 17 Bonjour 1985, p.42

Counterexamples to Epistemic Externalism Revisited

69

Bonjour concludes that Norman ought to classify his belief as an "unfounded hunch." 18 But Bonjour's description distorts this case too. If Norman asks himself how he could know the president's whereabouts, he will have nothing to say. He has no knowledge about the reliability of the relevant belief-producing process. Realizing his ignorance, he might say something like: I don't know whether my belief (p) was produced reliably. From my perspective, I cannot be sure of the reliability of the belief-producing process in question, but neither can I exclude it. From this it does not follow that: My belief was not or cannot have been produced reliably. Which Bonjour seems to assume, when he says that (p) ought be classified as an "unfounded hunch" from Norman's perspective. Ignorance as to the reliability of the process simply does not imply its unreliability. If the epistemic status of the belief is an open question from Norman's perspective, it is far from obvious that he should distrust the belief-producing process he has yet to test and assume it could not produce justified beliefs. Indeed, why should the principle of presumed innocence not apply here as elsewhere? As far as I can tell, Bonjour's conclusion rests on a twofold distortion. First of all, we need nothing more than reliably formed clairvoyant beliefs in order to reflectively justify the meta-belief that we have reliable clairvoyant power. Secondly, it is simply false to suggest that a belief-producing process must be regarded as unreliable in the absence of a meta-justification of its reliability. Still, one does have the impression that Norman's clairvoyant belief really is unjustified. Where does this impression come from? My diagnosis is that Bonjour's description underdetermines Norman's situation. If we describe the situation more fully, then it becomes clear that there are, despite Bonjour's claim to the contrary, undercutting defeaters against p. And any plausible version of externalism would concede that the reliability of belief-producing processes is only enough to establish the prima facie justification of beliefs.19 Of course this prima facie justification can be overruled by undercutting defeaters. If a person acquires additional evidence that speaks against the truth or the reliability of the source of a reliably formed belief, then this belief indisputably loses its justification—even if the person continues to hold the belief. No externalist would or could deny that. Just how to understand the defeasibility of reasons within the framework of externalism is as ever a serious challenge to externalists, but I cannot say anything more about it here. 20

18 ibid. 19 See Goldman 1979, p. 18ff. Dretske 2000, p.595, calls reliabilism that denies the defeasibility of reasons "mad dog reliabilism." 20 In Grundmann 2003, p.334-339, I venture a few steps in the direction of an externalist answer to this question. The central problem for the externalist is as follows: if we accept that reliably formed beliefs are defeasible by contrary evidence, then we commit ourselves to the view that reliably produced beliefs can be epistemically inadequate, insofar as something that should occur may not. The reliable belief should have been abandoned in the face of the contrary evidence. Since it was not abandoned, it must be regarded as unjustified. That seems to contradict externalism in two ways. First, the very existence of defeaters appears to be a purely internalist matter of our mental perspective. Second, externalists typi-

70

In Defence o f Epistemic Externalism

Let's assume that externalism can do justice to the defeasibility of reasons. If it turned out that we do not attribute a justified belief to Norman because we implicitly assume the existence of defeaters, this intuition would not even speak against externalism if the belief in question was indeed formed by a reliable process, since externalism concedes that reliably produced beliefs are vulnerable to defeaters. In Normans case, as I shall attempt to demonstrate, several kinds of defeaters in fact undercut the reliability of the belief-production process in question. Assume that Norman's clairvoyance is accompanied by a characteristic phenomenology. When he acquires clairvoyant beliefs, he has a specific, unmistakable experience. Thus he can unambiguously identify tokens of this type of process. And when he asks himself whether clairvoyance is reliable, he has got to take his empirically supported background theories into account. But according to these background theories, clairvoyance is a mysterious epistemic faculty that does not fit our scientific view of the world. So if we attribute our background theories of the world to Norman, then he has an undercutting defeater against the clairvoyant belief—even if this belief was acquired by a reliable process. Maybe Bonjour would insist that his description explicitly rules out the possibility of Norman having any undercutting defeaters against the reliability of his clairvoyant power. If so, then we would have to attribute background knowledge to Norman that is fully different from the background knowledge we actually possess. This makes the example so abstract that we can hardly evaluate Norman intuitively at all. In any case, even if Norman has no reason to doubt the reliability of clairvoyance as such, there is nevertheless an undercutting defeater. Bonjour describes the situation such that there is no characteristic phenomenology of clairvoyance. From Norman's perspective then, clairvoyant beliefs belong to the general class of spontaneous, unmotivated beliefs. Norman is quite right to deny their reliability: spontaneous, unmotivated beliefs are not usually true. With this appropriately wide-meshed classification of the beliefs in fact produced by clairvoyance, Norman acquires an undercutting defeater against his clairvoyant beliefs—without having an undercutting defeater against clairvoyance as such. To sum up: my gloss on Bonjour's case of Norman (and Lehrers comparable case of Mr. Truetemp) is that the intuition that Normans clairvoyant belief is unjustified

cally evaluate the epistemic properties of a belief in light of its causal genesis and not on the basis of norms and obligations. But I do not think either of these objections is decisive. As for the first, we can evaluate the objective value of defeaters just as we can that of evidence. Defeaters usually initiate beliefinhibiting processes that are conditionally reliable. A belief-inhibiting process is conditionally reliable if it suppresses more false beliefs than true ones, assuming that its inputs are true. So taking defeaters into account avoids false beliefs. As for the second objection, there is also an externalist explanation of the norm that we should abandon beliefs impugned by defeaters. It is certainly plausible to assume that the reliability of a cognitive system would have been significantly improved during the evolution of the species or the learning history of an individual by the presence of defeaters initiating belief-inhibiting processes. This historical success of defeaters can explain why defeaters that actually exist in a cognitive system would have the function of initiating belief-inhibiting processes—even if they do not in fact initiate such processes. This is the externalist source of epistemic normativity. Hence the epistemic value and normative force of defeaters can also be given an externalist explanation. Bergman 2 0 0 0 agrees that the assumption of epistemic obligations and norms is not incompatible with externalism.

Counterexamples to Epistemic Externalism Revisited

71

either rests on a distorted description of the situation and can be explained away, or it is based on our implicit assumption that Norman's cognitive perspective contains some kind of defeater. In the latter case, the intuition is compatible with a plausible version of externalism. Bonjour's intuition is either manipulated or else it does not speak against externalism.

II Now let's have a look at the demon scenario, which internalists present as an argument that objective reliability is not even necessary for justification. Imagine two subjects, Jane and Mary, whose mental perspectives correspond exactly. They have the same beliefs and the same evidence on the basis of perception, memory and introspection. They have gathered the evidence for their beliefs with maximal care and perseverance. Their beliefs are optimally rationalized by the available empirical evidence. Their belief systems are consistent and coherent. What's more, they have at their disposal the same a priori argument for the reliability of empirical beliefs. The only difference between them is that Jane lives in an ideal world where her cognitive powers are reliable without exception and produce mostly true beliefs, whereas Mary lives in a demon world where she is permanently deceived without knowing it. If Jane slipped into Mary's skin and Mary into Jane's, neither would notice any difference. Obviously, Jane's beliefs are optimally epistemically justified. But internalists claim that we are intuitively inclined to evaluate Mary's beliefs just like Jane's. Hence Mary's beliefs would also be optimally epistemically justified, although they are ex hypothesi unreliably produced. Therefore reliability is not a necessary condition for epistemic justification. 21 Indeed, if we were to judge Jane's and Mary's beliefs epistemically equal, then we would have to accept that reliability is not a necessary condition for epistemic justification. But it does not seem at all clear to me that we would make that judgment. It is clear only that Jane and Mary themselves would judge the justificatory status of their beliefs to be equal, since they are unable to notice any difference between their

21

Arguments along these lines can be found in Lehrer & Cohen 1983, Cohen 1 9 8 4 , Foley 1985, LuperFoy 1988, Haack 1993, p . l 4 8 f „ Dretske 2 0 0 0 , Bonjour 2 0 0 2 , p.228f. O f course one could object that the demon scenario is not really conceivable and therefore not metaphysically possible. This seems to be implied by semantic externalism, which holds that the semantic content of beliefs and other mental states is dependant on the extra-mental world. If the ideal and demon extra-mental worlds are so radically different, then it would not be possible—according to semantic externalism—for the mental perspectives of their inhabitants to correspond. For semantic externalism as an anti-sceptical strategy, see Grundmann & Misselhorn 2 0 0 3 . But many philosophers doubt that semantic externalism is strict enough about the dependence of representational content on the extra-mental world to rule out radical deception and the unreliability of basic cognitive abilities. See Foley 1992, p. 139; Bürge 1986, p. 139f.—In any case, the objection does not undermine the internalists's demon argument, since it does not matter whether the demon scenario is really possible, only that we consider the case prima facie possible. For the analysis o f the concept of justification it is only relevant whether we would apply the concept in the demon world or not, assuming that the demon world is possible.

72

In Defence o f Epistemic Externalism

epistemic situations. 22 But it simply does not follow from their both believing that they are equally justified that they are in fact equally justified. If Mary were to evaluate the objective reliability of her own cognitive abilities, she would falsely arrive at a positive conclusion because she is permanently deceived about the outside world. Of course one can hardly fault Mary for believing what she does, since she cannot in principle realize that she is being deceived. But beliefs are not automatically justified because one cannot fault someone for holding them. 23 So it seems rather that we have exactly the opposite intuition. If we use a particular standard or method to justify our beliefs (whatever the standard or method may be) and want to show that it justifies our beliefs—in other words, if we want a metajustification of our belief-producing process—then we will try to demonstrate that it is truth-conducive or reliable. Even a resolute proponent of internalism like Bonjour allows that: "one can finally know that a given set o f standards for epistemic justification is correct or reasonable only by knowing that the standards in question are genuinely conducive to the cognitive goal o f truth. A n d this in turn apparently means that it is incumbent on the proponent o f such an epistemological theory to provide an argument or rationale of some sort to show that his proposed standards of justification are indeed truth-conducive (...). Providing such an argument is ultimately the only satisfactory way to defend such a theory against its rivals and against the sceptic (...). 2 4

If meta-justification of a standard of epistemological justification requires that we justify its reliability, then epistemic justification plainly implies reliability. Only because that is the case is the demon world hypothesis a challenge to our claim to justification. Otherwise the relevance of such a hypothesis would be inexplicable. In order to make this clear, let's assume for a moment that it would suffice for epistemological justification if a given belief appeared from our perspective to be probably true. If this were so, then meta-justification would only require us to demonstrate that we take our standard to be reliable, not that it really is reliable. That would make the task of meta-justification rather easy. We would simply have to introspectively inquire which method seemed reliable. We would not have to concern ourselves with the sceptical possibility that the real world might be completely different from what we think. So anyone who considers sceptical hypotheses a challenge to our claims to justification and demands that any meta-justification include proof of the reliability of the standard in question has already acknowledged that epistemic justification implies objective

22

T h i s point is emphasized by Foley 1985, p. 190: "Even if (...) our world is world w (the d e m o n world, T . G . ) it still can be epistemically rational for us to believe many o f the propositions we do, since the epistemic situation in world w is indistinguishable f r o m the epistemic situation in a world which has the characteristics w e take our world to have." See also B o n j o u r 2 0 0 2 , p.229.

23

Even when s o m e o n e has fulfilled all his epistemic obligations and behaves epistemically responsibly, his beliefs can still be epistemically unjustified. I develop this argument thoroughly in G r u n d m a n n 2 0 0 3 a , p . 2 5 1 - 2 5 5 . Internalists, too, argue that deontological and epistemic justification are not coextensive. See especially B o n j o u r in B o n j o u r & Sosa 2 0 0 3 , p. 175 ff. B o n j o u r 1985, p.9.

24

Counterexamples to Epistemic Externalism Revisited

73

reliability. Thus, it would be inconsistent for anyone who does so—as more than a few internalists do—to dispute that reliability is necessary at least for justification. Still, it cannot be denied that Mary's demon world beliefs have a certain epistemic merit. Clearly they are epistemically better off than, say, Catherine's. Catherine's basic evidence (i.e. perceptions) is subverted by an evil demon just like Mary's, but she also (i) holds beliefs not rationalized by her basic evidence, (ii) continuously draws false conclusions and overlooks inconsistencies, (iii) and has at her disposal no philosophical argument for the reliability of her basic cognitive abilities. Can externalism explain the epistemic difference between Mary and Catherine? 2 5 I believe it can, although it may not at first seem so. Something that separates Mary's beliefs from Catherine's is that they are produced by rational inferences. They arise by deductive, inductive or explanatory inferences from other beliefs (see ii) or from perceptual states with representational content (see i). 2 6 Inferential rationality of beliefproduction is however no uniquely internalist virtue; externalism can also evaluate valid inferences positively. From an externalist perspective, valid inference from states with content—though it does not automatically produce justified beliefs—can be conditionally reliable. A belief-producing process is conditionally reliable if it leads from a true input to a true output in most cases. 27 The decisive point is that all of our ordinary inferential methods are conditionally reliable in the demon world. That is obviously true for deductive inferences. They guarantee a transfer of truth-value in all possible worlds and are thus perfectly conditionally reliable. Inductive and explanatory inferences on the other hand are neither perfectly conditionally reliable (in our actual world they lead at best very often to true beliefs, given a true inductive basis and/or correct description of the explanandum), nor are they reliable in all possible worlds. Induction, for example, would be unreliable in a chaotic world that lacked the requisite uniformity of nature; explanatory inferences would be conditionally unreliable in a world without causal structure (if such a world is indeed possible). But the demon world is not that kind of world. It cannot be chaotic, since that would undermine the regularity of the deceptive appearances. And it cannot be without causal structure, since the demon hypothesis is intended precisely as a causal explanation of the deceptive evidence. In the demon world then, false belief results only from deception about basic evidence, not from conditional unreliability of our inferential methods. Hence Mary's beliefs about the world are epistemically meritorious. They may not be justified simpliciter, since they are mostly false, but at least they are conditionally reliable and therefore conditionally justified. Thus they are epistemically superior to Catherine's. And there is yet another way in which Mary's beliefs are from an externalist perspective epistemically preferable to Catherine's. Although Mary's empirically supported beliefs about the world are not justified, she—unlike Catherine—has at her disposal a philosophical meta-justification that her empirical beliefs about the world are justifed.

25 26 27

C o h e n makes a similar argument in C o h e n 1984, p.283f. See G r u n d m a n n 2 0 0 3 b on inferences from perceptual representations. Conditional reliability o f inferential processes is discussed in G o l d m a n 1979, p . 1 3

74

In Defence of Epistemic Externalism

She is justified in believing that her empirical beliefs are justified—even if in fact they are not. But can an externalist really claim that Mary has a second-order philosophical justification in the demon world? He would have to assume that the philosophical meta-belief had been reliably produced, which seems to contradict the deception scenario of the demon world. The suspicion is that a second-order justification in the demon world would have to implicitly appeal to higher-order internalist standards. So it seems an externalist could not consistently claim that Mary were justified at least in believing that her empirical beliefs about the world were justified. But let's take a closer look at the case. We started out with Jane, whose epistemic situation is ideal from an internalist as well as from an externalist perspective. Not only does she form her beliefs properly from an internalist standpoint, but the beliefproducing methods she employs are also in fact reliable. What's more, she has recourse to philosophical insight, which has the virtue of being able to support beliefs about metaphysical possibilities and necessities a priori, i.e. independently of experience. Jane's method of philosophical insight is ex hypothesi reliable. This does not automatically mean that philosophical knowledge about modal facts is infallible, as has traditionally been common to assume, but a priori justification of modal beliefs must at least make their truth probable. If Jane's a priori supported modal beliefs are reliable and Mary's beliefs ex hypothesi exactly correspond with Jane's, then Mary's modal demon world beliefs would only be unreliable if the evil demon could alter the metaphysically possible worlds as well as alter contingent facts. But this is not the case. The metaphysically possible worlds set a limit to the demon's freedom to act. This has a startling consequence: if Jane's aprioristic method for producing modal beliefs is reliable, then Mary's method in the demon world must be too. And if this method is reliable in the demon world and delivers a philosophical argument for the reliability of our empirical methods, then Mary is also justified from an externalist perspective in believing that her empirical beliefs are justified. Whatever the philosophical argument might look like, its conclusion would have to be something like: "Sense perception is reliable in all possible worlds." This conclusion would also be justified from an externalist perspective in the demon world, although it would happen to be false, since the reliability of sense perception falters in at least one possible world—namely the demon world. But the externalist justification of beliefs requires no guarantee of truth, merely the objective likelihood of truth. And this is not precluded by the falsity of a few results of the method. My objection to the internalists's demon world argument can be summarized as follows: If we compare the epistemic situations of two subjects whose mental perspectives exactly correspond, one of whom lives in an ideal world and the other in a demon world, then I do not think we have the intuition that there is no epistemic difference between the two. Rather, we must indeed have the intuition that there is such a difference, since otherwise we would not have to rule out demon scenarios when attempting to show that our beliefs are justified. It is just that we also have the intuition that the two subjects could not notice this epistemic difference. In any case, the epistemic difference does not imply that the beliefs of the inhabitant of the demon world have no merit. And externalism has no special problem explaining their epistemic merit, since the demon cannot completely undermine the reliability of all methods, although it may at first seem that he could. A method that is reliable in the ideal world may be unreliable

Counterexamples to Epistemic Externalism Revisited

75

in the demon world insofar as the facts that are decisive for truth value can be altered. But not all facts in the demon world are different from those in the ideal world: modal facts are the same in all possible worlds (as long as the proximity or distance of such worlds does not play a role). Structural uniformity and the principle of causality are constitutive of the demon world too. Thus not all of the methods that are reliable in the ideal world are unreliable in the demon world. In fact, the existence of reliable or conditionally reliable methods in the demon world is the deeper reason why we attribute a certain merit to Mary's beliefs. Hence externalism is able to explain the intuitions that arise when we evaluate the epistemic siuation in the demon world. QED.

References Alston, W. 1989: "Concepts of Epistemic Justification," in: Alston, Epistemic Justification, Ithaca/London, pp. 81-114. Alston, W. 1993: "Epistemic Desiderata," in: Philosophy and Phenomenological Research 53, pp. 527-551. Bach, K. 1985: "A Rationale for Reliabilism," in: The Monist 68, pp. 246-263. Bergmann, Μ. 2000: "Deontology and Defeat," in: Philosophy and Phenomenological Research 60, pp. 87102. Bonjour, L. & Sosa, E. 2003: Epistemic Justification. Internalism vs. Externalism, Foundations vs. Virtues, Oxford. Bonjour, L. 1985: The Structure of Empirical Knowledge, Cambridge (Ma.). Bonjour, L. 2002: Epistemology. Classic Problems and Contemporary Responses, Lanham. Bürge, Τ. 1986: "Cartesian Error and the Objectivity of Perception," in: P. Pettit et al. (eds.), Subject, Thought and Context, Oxford. Cohen, S. 1984: "Justification and Truth," in: Phibsophical Studies 46, pp. 279-295. Dretske, F. 2000: "Entitlement: Epistemic Rights Without Epistemic Duties?," in: Philosophy and Phenomenological Research 60, pp. 591 -606. Engel, Μ. 1992: "Personal and Doxastic Justification in Epistemology," in: Philosophical Studies 67, pp. 133-150. Foley, R. 1985: "What's Wrong with Reliabilism," in: The Monist 68, pp. 188-201. Foley, R. 1992: "Can Metaphysics Solve the Problem of Skepticism," in: Philosophical Issues 2, pp. 131147. Foley, R. 1993: Working without a Net. A Study in Egocentric Epistemology, Oxford. Goldman, A. 1979: "What is Justified Belief?," in: G. Pappas (ed.), Justification and Knowledge, Dordrecht, pp. 1-23. Goldman, A. 1986: Epistemology and Cognition, Cambridge (Ma.). Goldman, A. 1988: "Strong and Weak Justification," in: Philosophical Perspectives 2, pp. 51-70. Grundmann, Τ. & Misselhorn, C. 2003: "Transcendental Arguments and Realism," in: H.-J. Glock (ed.), Strawson and Kant, Oxford, pp. 205-218. Grundmann, Τ. 2003a: Der Wahrheit auf der Spur. Eine Verteidigung des erkenntnistheoretischen Externaltsmus, Paderborn. Grundmann, T. 2003 b: "Perceptual Representations as Basic Reasons," in: R. Schumacher (ed.), Perception and Reality, Paderborn. Haack, S. 1993: Evidence and Inquiry, Oxford/Cambridge. Kornblith, H. 1995: "Naturalistic Epistemology and Its Critics," in: Philosophical Topics 23, pp.237-255. Lehrer, Κ. & Cohen, S. 1983: "Justification, Truth, and Coherence," in: Synthese 55, pp. 191-207. Lehrer, Κ. 1990: Theory of Knowledge, Boulder/San Francisco. Luper-Foy, S. 1988: "The Knower: Inside and Out," in: Synthese 74, pp. 349-367. Plantinga, A. 1993: Warrant: The Current Debate, Oxford.

Conditions on Cognitive Sanity and the Death of Internalism HILARY KORNBLITH

Laurence Bonjour has presented the most sophisticated and the most detailed defense of internalism in epistemology. His first book, The Structure of Empirical Knowledge1, presented an extremely intuitive attack on externalism, combined with the most highly worked out version of the coherence theory to date. Bonjour has since abandoned the coherence theory in favor of foundationalism2, but his commitment to internalism remains unaltered. Thus, in his new book, In Defense of Pure Reason3, he speaks of externalist theories of justification as "merely wrong-headed and ultimately uninteresting evasions of the central epistemological issues." [1] On Bonjour's view, externalism "seems to simply change the subject without really speaking to the issues that an adequate epistemology must address." [96] The rejection of externalism is integral to Bonjour's development of his defense of a moderate rationalism. But it is thus especially interesting that, half way through the new book, Bonjour acknowledges that, on his account, both demonstrative and intuitive a priori justification turn out after all to have an externalist dimension... This is a stronger concession to externalism than I have heretofore been willing to make, but one that seems required by the facts of the situation. [128]

When the foremost defender of internalism in epistemology feels forced to make such a concession to externalism, we should all sit up and take notice for something important about the very tenability of internalism is at stake. In this paper, I examine the facts of the situation which force Bonjour to make this concession, a concession which he calls an externalist "condition of cognitive sanity." [128] I argue that this is no small concession on Bonjour's part, and that the real significance of this condition is that it signals the death of epistemological internalism.4

1 2

Harvard University Press, 1 9 8 5 . See, most explicitly, his "The Dialectic of Foundationalism and Coherentism," in J. Greco and E. Sosa, eds., The Blackwell Guide to Epistemology, Blackwell, 1 9 9 9 , 1 1 7 - 1 4 2 , where he claims that "coherentism is pretty obviously untenable, indeed hopeless." [139]

3

Cambridge University Press, 1 9 9 8 . Numbers in square brackets in the text hereafter will refer to pages of either this volume or the volume cited in note 1. I presented this argument quite briefly in "The Impurity of Reason," Pacific Philosophical Quarterly, 8 1 ( 2 0 0 0 ) , 6 7 - 8 9 ; see 8 1 - 8 2 ; I welcome the opportunity to expand on that overly brief presentation of the argument. I should say here that there is a form of internalism, quite different not only from Bonjour's, but also from any currently defended in the literature, which 1 believe is deserving of a hearing. This more defensible version of internalism is committed to a controversial empirical claim about the structure and capacity of the human mind, a claim which, in the end, I believe will prove false. It must, however, be admitted that the evidence currently available is not decisive. I present this version of internalism in "Epistemic Obligation and the Possibility of Internalism," in L. Zagzebski and A. Fairweather, eds., Virtue Epistemology: Essays on Epistemic Virtue and Responsibility, Oxford University Press, 2001, 231-248.

4

78

In Defence of Epistemic Externalism

I

The difference between internalism and externalism in epistemology turns on a question about the "cognitive accessibility" of putative justifying factors. Thus, in The Structure of Empirical Knowledge, Bonjour comments, According to a coherence theory of empirical justification, as so far characterized, the epistemic justification of an empirical belief derives entirely from its coherence with the believer's overall system of empirical beliefs and not at all from any sort of factor outside that system. What we must now ask is whether and how the fact that a belief coheres in this way is cognitively accessible to the believer himself, so that it can give him a reason for accepting the belief. [101]

The mere fact that a belief coheres with the rest of an agent's beliefs is not sufficient, on this view, for justification; an agent must have access to the fact of coherence. Indeed, on Bonjour's view, the agent must recognize that the belief coheres with his total body of beliefs, and hold the belief in virtue of his recognition of its coherence. Anything less than this does not give an agent a reason for accepting the belief. The agent must have "cognitive access" to the reasons for believing a given claim if those reasons are to count as reasons for him. It is this requirement of cognitive access which makes Bonjour an internalist. Note that the requirement of cognitive access can be applied just as easily to foundationalist theories as it can to coherentism. Indeed, this is Bonjour's current position. An agent may have some beliefs which are unjustified, on this view, because they do not meet appropriate foundationalist standards: there is no way to derive these beliefs from appropriate foundations using appropriate epistemic principles. But on an internalist version of foundationalism, there is another way in which an agent's belief may fail to be justified. Even when an appropriate foundational derivation for a claim exists, the derivation fails to provide a given agent with a reason to hold the belief unless the derivation is one to which the agent has cognitive access. Indeed, it is not just that the derivation must be accessible to the agent; the agent must be aware of the derivation and hold the belief in virtue of his recognition that it can be so derived. As Bonjour now puts it, . . . a belief's being epistemically justified requires that the believer in question have in his cognitive possession or be suitably aware o f . . . a reason or adequate basis for thinking that the belief is true. 5

All of this stands in sharp contrast to externalist theories of justification, such as reliabilism.6 A reliabilist holds that a belief is justified just in case it is produced by a reliable belief-forming process. An agent need not believe that the belief is so produced, nor need the agent have reason to believe that there is such a process at work. It is

5 6

"The Dialectic of Foundationalism and Coherentism," 118. See, especially, the work of Alvin Goldman: "What is Justified Belief?," in G. Pappas, ed., Justification and Knowledge: New Studies in Epistemology, Reidel, 1979, 1-23 and Epistemology and Cognition, Harvard University Press, 1986.

Conditions on Cognitive Sanity and the Death of Internalism

79

the fact that the belief is reliably produced which makes it justified, and nothing more than this is required. There need be no cognitive access to the reliability of the beliefforming process. Indeed, it is this difference over the requirement of cognitive access which forms the basis for Bonjour's well-known and highly intuitive objections to reliabilism in particular, and externalist theories in general. Bonjour presents the case of Norman the clairvoyant, a man who in fact has clairvoyant powers and who forms a belief on the basis of these powers; his belief is reliably produced. But, as Bonjour presents the case, Norman does not believe, nor does he, in any intuitive sense, have any reason to believe, that he is clairvoyant or that he has any reliable way of forming beliefs of the very sort he is currently forming. Intuitively, under these conditions, Norman does not form a justified belief even though his belief is, as a matter of fact, reliably produced. Bonjour draws the natural conclusion here: . . . t h e s e cases a n d the m o d i f i c a t i o n s m a d e in response to t h e m also suggest an i m p o r t a n t moral which leads to a basic intuitive objection to externalism: external or objective reliability is not e n o u g h to offset subjective irrationality. If the acceptance o f a belief is seriously unreasonable f r o m the believer's own standpoint, then the mere fact that u n b e k n o w n s t to h i m its existence in those circumstances lawfully guarantees its truth will not suffice to render the belief epistemically j u s t i f i e d . . . 7

The facts relevant to justification, on the internalist view, are all ones which are somehow registered "from the believer's own standpoint." Facts which are not registered by the believer—ones which the believer has not "cognitively accessed"—can not in any way be relevant to justification. More than this, the intuition here is that if two believers are identical "from the inside," then the justificatory status of their beliefs cannot be different. But whether a belief is formed by a reliable process need leave no internally accessible trace: one might have two agents, identical "from the inside," one of whom has a belief which is reliably produced, while the other has a belief which is produced unreliably. Reliabilism is thereby shown to be mistaken: it marks agents' beliefs as different in justificatory status even when their epistemic positions are wholly indistinguishable "from the inside." 8 T h e requirement that agents who are identical from the inside cannot differ in the justificatory status of their beliefs is thus the hallmark of internalism. 9 Even externalists

7 8

9

The Structure of Empirical Knowledge, 41. I have been using the phrase "from the inside" because I believe it best expresses the intuitive thought behind the internalist position. I have been putting the phrase in scare quotes because I believe that making good philosophical sense of it in a way which justifies, rather than merely gives expression to, the internalist position is not, in the end, possible. This point having been made, I will now simply leave out the scare quotes. A number of the issues about what this expression comes to are beside the point of the present paper; I attempt to address them in "How Internal Can You Get?," Synthese, 74(1988), 313-327. See, for example, Richard Foley's The Theory of Epistemic Rationality, Harvard University Press, 1987: Consider a pair of situations. In one, suppose that S has a good argument for ρ and that he believes p. So his belief ρ is epistemically rational. In the second situation, suppose that S believes, thinks, experiences, seems to remember, exactly what he does in the first situation, and moreover suppose that what he would believe on reflection is the same. The second situation, then, is such that S

80

In Defence of Epistemic Externalism

should acknowledge that there are powerful intuitions in favor of this requirement. If Jack's beliefs are justified, and, for all Mary can tell, she is in exactly the same epistemic situation as Jack—that is, from the inside, what it is like to be Mary is indistinguishable from what it is like to be Jack—then it seems both unfair and absurd to claim that Mary might be unjustified in her beliefs. When we claim that a person is unjustified in her beliefs, we are holding her responsible for a certain sort of mismanagement of her cognitive situation. If Jack has done nothing wrong—if his beliefs are fully justified—then Mary, whose epistemic situation is internally indistinguishable from Jack's, could not be blameworthy. When Bonjour objects that externalist theories are "merely wrong-headed and ultimately uninteresting evasions of the central epistemological issues," what he has in mind, I believe, is just this: theories of justified belief are ultimately theories of cognitive self-management, and insofar as they seek to provide an agent with epistemic guidance, they must provide a kind of epistemic advice that agents are capable of applying; but this applicability requirement 10 guarantees that agents who are, as best they can tell, from their own points of view, in the same epistemic situation, genuinely are in the same epistemic situation, for advice which turns on features of an agents epistemic situation which is invisible to him is thereby unusable. The internalist point of view may be put in a brief slogan, which I will call the Internalist Credo: no differences in justificatory status without internally accessible differences. It is just this Internalist Credo which Bonjour, reluctantly, now calls into question. And the reasons which force him to do this, I will argue, thoroughly undermine his internalism.

II Consider the account of a priori justification which Bonjour presents in his new book, In Defense of Pure Reason. Bonjour holds that an agent may be a priori justified in holding a belief as a result of a fallible process of apparent rational insight. Some beliefs, on this view, are directly justified by apparent rational insight; others are indirectly justified by way of demonstration from directly justified beliefs, where each step in the demonstration is certified by apparent rational insight. Bonjour distinguishes apparent rational insight from genuine rational insight: ...an apparent rational insight will count as genuine only if it actually involves the sort o f authentic grasp o f the necessary character of reality that ensures truth. [113]

As Bonjour points out, however, it would be a mistake to give a characterization of a priori justification in terms of genuine rational insight, and then claim that a priori justification is thereby infallible.

would not be able to distinguish any difference between it and the first situation were he somehow to be switched instantaneously from one to the other. Situations of this sort are epistemic twins, if S's belief ρ is epistemically rational in one situation, then it is epistemically rational in the other as well. [174-5] 10 I have discussed this applicability requirement in "The Psychological Turn," Australasian Journal of Philosophy, 60(1982), 238-253.

Conditions on Cognitive Sanity and the Death of Internalism

81

In the absence of a workable criterion of genuineness, this [move] safeguards the infallibility of rational insight only by refusing to call a state of mind a genuine rational insight if it turns out to be mistaken, no matter how subjectively or intuitively compelling it may have been. [113] Such a move clearly violates the Internalist Credo, and therefore must be rejected. Since the difference between apparent rational insight and genuine rational insight is not internally accessible, the distinction between the two is not one which internalists can appeal to as relevant to justification: either both kinds of insight carry justificatory weight or neither do. Bonjour holds that both do. Thus, I suggest, a moderate rationalism that abandons the indefensible claim of infallibility should hold instead that, subject to certain further conditions to be discussed below, it is apparent rational insight (and, correlatively, apparent self-evidence) that provides the basis for a priori epistemic justification. The further conditions which Bonjour enumerates are that apparent rational insight requires "a reasonable degree of care" in reflection, and must involve a genuine awareness by the person in question of the necessity or apparent necessity of the proposition in something like the strong logical or metaphysical sense, not merely a more generic belief that it is in some way obvious... [114] Once these conditions are added, however, a problem immediately arises, as Bonjour frankly admits. ...suppose that in a putative case of intuitive a priori justification, the person in question either believes mistakenly that he has carefully reflected on the claim in question or else has a clearly mistaken understanding of the relevant concept of necessity. On the present account, such a person does not in fact genuinely possess an a priori reason for thinking that the relevant claim is true ... because the conditions for access to such a reason are not satisfied. In [this] case, he fails to have even an apparent rational insight, his own subjective impression of the situation notwithstanding. [127] What should we then say about the person who fails to meet either the condition of careful reflection or the condition requiring an adequate grasp of the relevant notion of necessity: Though such a person seems to himself to be justified, his grasp of what is really going on is simply too defective to have any genuine epistemic force; in effect, he fails to pass what might be regarded as a condition of cognitive sanity. [128] But such conditions of cognitive sanity involve a blatant violation o f the Internalist Credo, for differences which are not internally accessible are now held to make a difference in justificatory status. Bonjour's commitment to internalism, as he himself points out [128], is thereby compromised.

82

In Defence of Epistemic Externalism

III Bonjour insists that a priori justification is fallible. Restricting a priori justification to those beliefs arrived at by way of genuine rational insight involves what Bonjour refers to as "the trivialization of the claim of infallibility." [113] It involves a trivialization because such a move would make it impossible to tell whether a given claim was justified in this way or not without knowing independently whether or not the claim of necessity was correct—thus making the appeal to rational insight entirely useless as an independent and self-contained basis for justification. [113]

But to say this is to say nothing more than that such a move would violate the Internalist Credo, a requirement which I acknowledge has considerable intuitive force, but which Bonjour himself ends up violating just a few pages later when he lays down his conditions on cognitive sanity. What would Bonjour's position look like if he abandoned the conditions on cognitive sanity and consistently stood by the Internalist Credo? O n e of the conditions on cognitive sanity is a requirement of a reasonable degree of care in reflection. This is not simply a requirement that the agent believe a reasonable degree of care has been exercised; such a belief, while necessary, is not sufficient for cognitive sanity. Now it is more than a mere logical possibility for an agent to believe sincerely that he has expended a reasonable degree of care in reflection without actually meeting, or coming close to meeting, standards of reasonable care. Agents who sincerely believe that they are reflecting with reasonable care, but are nevertheless very far from having done so, may engage in far less reflection, and use far less care, than agents who rightly recognize that they are not spending a good deal of time thinking an issue through. The condition on cognitive sanity which Bonjour insists on merely requires that the sincere but benighted agent be recognized as failing to meet minimum standards of care; the fact that such an agent believes he has met minimum standards, when he has not, should not be seen as redounding to the agents epistemic credit. The same is true of the agent who has "a clearly mistaken understanding of the relevant concept of necessity," when it is not clear to the agent that his understanding is in any way mistaken. It is, without a doubt, possible to believe that one understands a concept when, in fact, one does not. It is possible to have ideas which are, by any objective standard, hopelessly unclear and yet believe that one s ideas are perfectly clear. It is possible to believe that one's understanding is complete when, in fact, it is partial or worse. An agent may thus have a view of the relevant concepts here which is horribly defective by any objective standard and yet believe that his understanding of that concept is perfect. Such an agent is in a far worse epistemic position than the agent who rightly recognizes that he does not fully understand a given concept. The fact that the defective agent believes he has perfect understanding is, indeed, part of his problem; it is not something for which he should receive epistemic credit. Finally, consider Bonjour's view on the role of memory in demonstration. Following Bürge 11 , Bonjour does not believe that premises about the accuracy of one's memory are,

11 See "Content Preservation," Philosophical Review, 102( 1993), 457-488.

Conditions on Cognitive Sanity and the Death of Internalism

83

at least typically, essential parts of demonstrative reasoning; rather, that one's memory should be functioning properly is seen as "a background condition for the functioning of reason." [126] One is not even reasoning at all if one's memory is sufficiently defective, on this view. The belief that one's memory is properly functioning, however sincerely held, is not sufficient for genuine reasoning. Indeed, even a justified belief that one's memory is properly functioning is not sufficient for genuine reasoning. In particular, consider an individual who is suffering from Alzheimer's Disease and who believes that he is engaging in complex demonstrative reasoning; perhaps he believes that he is proving a theorem in geometry. Let us also suppose, however, that this agent is incapable of making accurate transitions from one line of a proof to the next because he cannot keep earlier steps clearly in mind. Nevertheless, he sincerely believes that he is engaging in demonstrative reasoning. Such an agent is not actually engaged in reasoning; he fails to meet a "background condition for the functioning of reason." He does not receive extra epistemic credit for believing that his memory is working properly. Quite the opposite; believing that one is properly functioning when one is so very far from competent is itself an additional epistemic defect. Bonjour's insistence that the conditions on cognitive sanity be evaluated by some objective standard rather than from the agents own point of view is surely reasonable. The alternative would require classifying agents who are epistemically defective as meeting the highest standards of justifiedness in cases where they mistakenly believe, as a result of those very epistemic defects, that they are functioning well. Note, in particular, the irony that would result from rejecting an objective requirement on cognitive sanity. Normal agents who, on occasion, fail to carefully reflect or fail to have an adequate level of understanding of relevant concepts, would thereby be ruled, by their own standards, as unjustified in their beliefs; but agents who are extraordinarily far from meeting objective standards of minimal competence would instead be counted fully justified, simply because they come to believe—precisely because they are so far from meeting objective standards—that they reason well. An epistemological theory which places those who fail to meet minimal objective conditions on cognitive sanity in the same class with those who meet the highest objective standards is surely focusing on properties which are irrelevant to proper epistemic evaluation. Bonjour is thus right to reject the agent's own standards in favor of an objective evaluation of the conditions of cognitive sanity. But to do this, of course, is to abandon the Internalist Credo.

IV Internalism is often motivated by appeal to evil demon cases.12 Consider two agents, one of whom is perfectly normal, while the other is being deceived by an evil demon. The normal agent is looking at a table in perfectly good lighting; he forms the belief that he is looking at a table, and uses the best sort of reasoning available. Surely the normal

12 See, for one important such example, Stewart Cohen, "Justification and Truth," Philosophical Studies, 46(1984), 279-295. But use of this kind of example to motivate internalism is quite widespread.

84

In Defence o f Epistemic Externalism

agent's belief is justified if anyone's belief ever is. T h e agent who is being deceived by the demon is not looking at a table; there is no table anywhere near him. Instead, the demon presents him with sense experience 13 as of looking at a table in normal conditions. He too comes to believe that he is looking at a table, and he comes to do so in the very same way, and for the very same reasons, as the normal agent. 1 4 What it is like to be the victim of such an evil demon is exactly what it is like to be a normal perceiver. Thus, if we judge the normal perceiver to be justified in his belief, it seems we must also judge the victim of the evil demon to be justified. T h e reasoning just given seems to presuppose the Internalist Credo: since there are no internally accessible differences between the normal agent and the victim of the evil demon, we may not hold that their beliefs differ in justificatory status. Thus, if the normal perceiver is justified in his belief, so is the victim of the evil demon. Bonjour's discussion of the conditions of cognitive sanity indicates, however, that we may also construct pairs of cases in which one agent is functioning normally while a second agent, who fails to satisfy conditions of cognitive sanity, is horribly impaired. From the inside, these agents are indistinguishable, and yet here Bonjour—and many others as well, I suspect—will be much less inclined to say that the beliefs of the two agents are alike in justificatory status. If the Internalist Credo were consistently followed, however, we would have to say that agents may fail to satisfy conditions of cognitive sanity and yet remain justified. Is there some consistent motivation available for following internalist intuitions in the evil demon cases, and yet insisting, with Bonjour, on some kind of objective reading of the conditions on cognitive sanity? One way to try to pull these pairs of cases apart is to appeal to the intuition that evaluation of an agent's belief as justified or unjustified is a matter of evaluating what an agent does with the evidence available to him. T h e agent who is deceived by an evil demon, or even an agent whose sense organs are in some way defective—so long as they are defective in ways which are not available to him—is still doing a fine job with available evidence; indeed, such an agent is, from the point of view of justification, doing exactly what he should. This is not to deny that such an agent is in an epistemically inferior position to the agent whose input systems are working properly. But—the intuition goes—how well one's input systems are working has nothing to do with justification. Justification is a matter of what one does with one's epistemic input, however that input was arrived at. O n this view, the agent who fails to meet conditions of cognitive sanity is in a fundamentally different position from the agent who is being deceived by an evil demon. When one fails to meet conditions of cognitive sanity, one is functioning in a way

13

14

T h i s may be regarded as a contentious way o f describing the example, for on s o m e externalist accounts o f what it is to have sense experience, the victim of the evil d e m o n is not having any sense experience at all. I will not worry about this issue in the text. Suffice it to say that what it is like to be a victim o f the evil d e m o n is exactly what it is like to be perceiving a table in normal conditions. Talk o f "the very s a m e way" and "the very same reasons" are subject to qualifications like those in note 13 above. I adopt the internalist m o d e o f describing these examples in the text both because I am trying to present the internalist motivation here, and also because I believe that these internalist descriptions are highly intuitive. T h i s is not to say that I endorse them.

Conditions on Cognitive Sanity and the Death of Internalism

85

which fails to meet minimum standards for justifiedness. Since justification is a matter o f what one does with one's epistemic input, and the agent who fails to meet conditions o f cognitive sanity is hopelessly botching things up with the input he receives, it is appropriate to judge such an agent unjustified even though the victim o f the evil demon is judged to be fully justified in his beliefs. Now although this is, I believe, a tempting line to take for the internalist, it is not an approach which will bear serious scrutiny. T h e distinction which the internalist wishes to make here between evil demon cases and cognitive sanity cases is one which would be well motivated if Descartes were right about our access to our own mental goings-on. 15 In particular, if the character o f an agent's reasoning were wholly transparent to him, then there would be good reason to distinguish between the victim o f the evil demon case and the agent who fails to satisfy conditions o f cognitive sanity. These cases would be judged to be different precisely because the victim o f the evil demon is in no position to tell that he is a victim; but, if the Cartesian view about mental transparency were right, then those who fail to meet conditions o f cognitive sanity would be able to tell, by introspection, that they fail. I f the character o f one's reasoning were perfectly transparent, then one could not fail to meet conditions o f cognitive sanity without being perfectly placed to recognize one's own shortcomings; such shortcomings would always be cognitively accessible to an agent. O n the Cartesian view, evil demon cases are thus different from cognitive sanity cases precisely because the Internalist Credo tells us to treat victims o f the evil demon differently from those who fail to meet conditions o f cognitive sanity. But this cannot be Bonjour's view, nor is it the view o f most other contemporary internalists. Bonjour himself points out, and most other internalists agree, that our understanding o f the relevant internal states and processes is fallible; this is precisely why the problem arises. But once one acknowledges that our understanding o f the relevant mental states and processes is fallible, it is no longer clear why the divide between factors which are relevant to justification and those which are not should be thought to line up neatly with the divide between internal states and processes and factors external to them. The Cartesian view o f our access to internal states and processes gives us a motivation for such a view, but, without it, the view seems terribly unmotivated.

V

Once Bonjour acknowledges that there are objective standards which an agent must meet if he is to satisfy the conditions on cognitive sanity, he thereby loses the right to wield the Internalist Credo against externalism; while the Internalist Credo rules out

15 1 have elsewhere defended the view that internalism presupposes a deeply Cartesian view about our access to our own mental states: see "How Internal Can You Get?," cited in note 8, above, and "The Unattainability of Coherence," in J. Bender, ed., The Current State of the Coherence Theory, Kluwer, 1989, 207-214.

86

In Defence of Epistemic Externalism

externalism, it rules out his own position as well. 16 Indeed, it arguably rules out all reasonable epistemic views. So much the worse, it seems, for the Internalist Credo. But Bonjour argues that his view, even with this concession, is fundamentally different from, and superior to, a thoroughgoing externalism. There are two reasons ... why such a view is still not externalist in the strong sense that pertains to, e.g., recent versions of reliabilism: First, where the external undermining factor or condition is in fact not present, the person can still have an adequate internal grasp of the reason why the proposition in question is likely to be true. His access to that reason depends on the undermining factor being absent, or rather on the contrary condition being present; but ... the absence of such a factor or condition (or the presence of the contrary condition) is still not part of the a priori reason. T h e present view is thus still quite different from those in which the external element is itself an essential part of the reason for thinking that the belief is true. Second, even though an apparent a priori reason may be undermined by a factor or condition of which the person is unaware, there is no general reason why he could not come to be aware of this factor or condition, even though it is possible that such an awareness may not in fact be achievable in a particular case. Neither of these things is true, mutatis mutandis, for standard versions of reliabilism. [128-9, n. 36]

Let us consider these two suggestions in turn. The first point Bonjour makes is that the person who in fact satisfies the conditions on cognitive sanity may, in virtue of that very fact, be in cognitive touch with the reasons which make his belief likely to be true. This is supposed to be different from what the thoroughgoing externalist claims. Now let us consider two different cases: a belief which is inferentially arrived at and a belief which is non-inferentially arrived at. In the inferential case, the reliabilist will say that an inferential belief is justified, like any other, if it is reliably produced. But the reliabilist certainly need not deny that the various beliefs which play a role in the inferential chain may themselves be thought of as reasons for the belief ultimately formed; indeed, it would be perverse to deny this. Now the reliabilist's view about what makes one belief a good reason for adopting another are different from Bonjour's: the reliabilist holds that nothing more than a reliable process connecting the two is required. But the reliabilist will certainly hold that agents who form beliefs as a result of inference may rightly be said to have reasons for their beliefs. Moreover, if one holds that one has cognitive access to one's beliefs, then one will hold that one has access to one's reasons in the inferential case. Bonjour's view does not differ from reliabilism on this point. What about the non-inferential case—assuming, for the sake of argument, that there are beliefs which are non-inferentially produced? Here Bonjour says that, on his view, the agent has access to his reason for belief, while the reliabilist must reject this claim. But again, the reliabilist need not deny that one has access to one's reasons; the reliabilist merely differs from Bonjour in what standards must be met if a reason is to be a good one. Thus, for example, take a standard perceptual case. Suppose one's view about perception is that when one looks at a table in good light and thereby forms the belief that there is a table in front of one, that belief is non-inferentially produced.

16 Michael Bergmann has also argued that internalist arguments against externalism are equally effective against internalism, although his reasons for making this claim are different from my own. See his "Externalism and Skepticism," Philosophical Review, 109 (2000), 159-194.

Conditions on Cognitive Sanity and the Death of Internalism

87

(The example is, of course, contentious, but the reader should substitute his or her own favorite example of a non-inferentially justified belief.) Now on Bonjours view, there is some kind of cognitive contact with experience that provides one with a reason for such a belief, even though the reason here does not do its justifying work by serving as the input to an inference. But the reliabilist need not deny this either. For the reliabilist, the connection between experience and belief which is required if that experience is to constitute good reason for holding the belief is nothing more than reliability: if the process by which the belief is formed tends to produce true beliefs, then one may well regard the experience as a good reason for holding the belief. And it is not unreasonable to hold as well, nor need the reliabilist deny, that the experience is something to which we have cognitive access. Bonjours first suggestion then, is that his view allows what reliabilism does not: namely, when an agent holds a justified belief, the agent has reasons for the belief which are within his cognitive grasp. But this suggestion turns out to be mistaken; reliabilists need not deny this. Reliabilists do not believe, of course, that agents need have cognitive access to all of the factors in virtue of which their beliefs are justified; but Bonjour, in the end, is in agreement on this point. What about Bonjours second suggestion? Bonjour recognizes that an agent's failure to satisfy the conditions on cognitive sanity may be something which the agent is not aware of and, in some cases, is not in a position to become aware of. It is for this reason that an agent may fail to be justified even though, as best he can tell from his own point of view, he is justified in his belief. Nevertheless, Bonjour holds that, ...there is no general reason why he could not come to be aware of this [undermining] factor of condition, even though it is possible that such an awareness may not in fact be achievable in a particular case. [129, n. 36]

And this is supposed to be different from reliabilism. Now I am not sure what Bonjour means when he says, on the one hand, that awareness of a particular condition "may not in fact be achievable in a particular case," and yet, at the same time, "there is no general reason why [the agent] could not come to be aware of this [condition]." Perhaps what Bonjour means is that, even though some agents may be in a position where they are incapable of achieving an understanding of their own shortcomings, other agents may sometimes be in a position to recognize their shortcomings. While this is true, it hardly marks a difference from reliabilism. Reliabilists will think that agents may sometimes be unjustified in their beliefs for reasons of which they are unaware. They will also think in some of these cases, the agent may have the ability to come to recognize the reasons for which he was unjustified, while in others, the agent will simply be unable to recognize the reasons for which he was unjustified. Reliabilists certainly do not hold that the factors which determine justifiedness—that is, the reliability of belief-forming processes—are fundamentally unknowable by human agents. Here too then, Bonjour has failed to show a respect in which his view is both different from and superior to reliabilism.

88

In Defence of Epistemic Externalism

VI When Bonjour is forced to give up his commitment to the Internalist Credo, he loses his most powerful weapon against externalism. More than this, the integrity of BonJour's view is thereby challenged, for it is no longer clear what consistent motivation there is for his way of viewing things. Bonjour's frank acknowledgment of the problems which his view encounters in light of the objectivity of the conditions on cognitive sanity spells the death of at least one form of internalism. Future internalists will need to show that there is some clear motivation for their view which does not, like the Internalist Credo, simultaneously undermine it. 1 7

17

Thanks to David Christensen for comments on a draft of this paper.

Empiricism Externalized RICHARD SCHANTZ

I I am an adherent of epistemological empiricism. I believe that the major part of our beliefs and our knowledge about the world depends, in a significant sense, on the evidence of the senses, on the information we gain directly in sensory experience. I am not, however, a proponent of semantic empiricism. I do not believe that the meaning of our words and the content of our thoughts can be made intelligible in terms of evidence, verification or justification. It is the concept of truth or, more exactly, of truth conditions, that has to be the key concept in viable theories of meaning and content. 1 Further, I am a proponent of externalism—both in the philosophy of mind and language, and in epistemology, that is, both with respect to meaning and content, and with respect to justification or the reasons for our believing. I hold that the content of our thoughts and the meaning of our utterances are partly determined by causal interactions with external objects and events, and so are not simply a function of what goes on inside our heads. And I am convinced that factors external to the subjects first-person cognitive perspective, factors that cannot be known by introspective reflection alone, that need not be directly, without further investigation, cognitively accessible from that perspective at all, can contribute to the justificatory status of a belief. In sum, I am a proponent of epistemological empiricism externalized. But what precisely does it take to be an epistemological empiricist? A first quite natural and tempting proposal is to characterize empiricism as the view that most, or even all, of our beliefs about the ways things are in the world around us are based on, or grounded in, sensory experience. This is not sufficient, however. Many philosophers there are who will unhesitatingly be prepared to accept this characterization without feeling themselves committed to any form of empiricism. They will accept it because they interpret the expressions "based on" and "grounded in" in a causal sense. The senses are causally relevant for the obvious reason that they mediate between objects and situations in the world and our beliefs and talk about them. What they will deny, however, is that this obvious fact, the fact that we do indeed acquire many of our beliefs about the world through the medium of the senses, is of crucial epistemological significance. When we are in the business of epistemology, of trying to justify our beliefs, of showing why it is rational to hold them, an appeal to sensory experience or sensory evidence will not help. It will not help, so it is often said, because experiences are purely natural

1

For m y defence of truth-conditional semantics see Schantz 1 9 9 6

In Defence of Episcemic Externalism

90

phenomena, brute results of impingements by the world on beings with appropriate sensory capacities. As products of causal connections in nature, experiences are nonconceptual, and so apparently also contentless. Hence they cannot also belong to the rational order, to the space of reasons, in which one can see something as justified or warranted in the light of something else. Or so it is argued. In this dialectical situation, it might seem that in order to still espouse a form of empiricism the experiences or appearances themselves must be brought within the space of reasons, and that the only promising way to do this is to imbue them with conceptual content. In my defense of an externalized version of empiricism, I will reject this recent tendency to conceptualize the dimension of sensory experience. Rather, I will defend the view that sensory experience carries a distinctive kind of nonconceptual representational content. And I will argue that experiences, qua bearers of nonconceptual content, are able to provide rational grounding for our beliefs about the world, and hence to play a significant epistemological role.

II Let us begin by casting a glance at the background of the current debate. Traditional empiricism held rather restrictive views on what the direct or immediate objects of awareness are when we perceive.2 According to it, what we are immediately aware of, or directly presented with in consciousness, is never a physical entity which exists independently of our perception of it. Rather, traditional empiricists maintain that the immediate objects of perception are a special kind of mental entities—often called "sense-data" or "appearances"—whose existence is dependent on our awareness of them. A sense-datum is supposed to be something that is given in experience, something that is simply received through the operation of sensibility, and whose apprehension does not involve the exercise of conceptual capacities. Furthermore, it is claimed that confrontations with sense-data, so conceived, provide the foundations of all empirical knowledge. Since beliefs about them, or judgments expressing them, are allegedly justified independently of all other beliefs, they are especially suitable to stop a well-known regress that is created by pointing out that beliefs justified only in relation to other beliefs are justified only if those others are also. It is important to note that if apprehensions of the given are to serve as the ultimate epistemic grounds for all our beliefs about the world they must play a justificatory role, and not merely a causal role. In this way, we are led to the classical foundationalist picture, according to which the system of all our beliefs can be divided into two parts: those which need support from others and those which can support others, but need no support from others. The latter are the foundations, the former the superstructure built on those foundations. The central thesis of foundationalism can be put by saying that all derived or mediately justified beliefs owe their justification ultimately to basic or immediately justified beliefs. Basic beliefs, in turn, are justified by being based on experiences of the given.

2

See Schantz 1990

Empiricism Externalized

91

III Traditional foundationalist empiricism has come under fire from many sides. A milestone is Willard Van Quine's attack on two dogmas by which he thinks empiricism was characterized: first, that there is a "fundamental cleavage" between analytic statements, statements true by virtue of meaning alone, and synthetic statements, statements whose truth is also dependent on the way the world is; and, second, the dogma of reductionism, according to which every meaningful statement can be reduced to statements that refer to immediate experience.3 What Quine puts in place of the second, more basic dogma is his holistic thesis that the sentences of our theory of the world are not separably vulnerable to adverse observations, because it is only jointly as one theory that such sentences imply their observable consequences. As he puts it: "My countersuggestion [...] is that our statements about the external world face the tribunal of sense experience not individually but only as a corporate body."4 Since the individual sentences of a theory do not generally have unique ranges of confirming or infirming evidence associated with them, we can adhere to any one of the sentences of a theory in the face of recalcitrant observations by revising the truth values of other sentences of the theory. An important reservation must be admitted, however. Observation sentences are exceptions. They enjoy a privileged status in the web of our beliefs, for they are separately susceptible to the test of observation.5 On Quine's view, then, the truth of our statements does depend on both, language and extralinguistic fact; it is just that this duality of factors cannot generally be traced back to the statements of our theory taken one by one. Despite his sharp criticism of traditional empiricism, Quine still thinks of his own epistemology as a version of empiricism, but as an enlightened, naturalized empiricism, one without dogmas. He aspires to retain the empiricist idea that our statements about the world have to face the "tribunal of experience" though they can only face it as whole system. The focal point of new, naturalized empiricism no less than of old, radical empiricism is the relationship between evidence and theory, between all the informations we attain by the senses and our overall theory about the world. Thus, the central question that Quine's epistemology tries to answer is the question of how we, physical inhabitants of a physical world, acquire our systematic theory about the world. It is this question, the question at bottom, of how knowledge of the external world is possible, that ultimately motivates his whole philosophy. Quite characteristically, his book The Roots of Reference begins with just this question: "Given only the evidence of our senses, how do we arrive at our theory of the world?".6 To answer this question Quine has to give an account of the relation between theoretical sentences and observation sentences, on the one hand, and of the relation between observation sentences and sensory evidence on the other. The explanation of the evidential or inferential relation between theoretical sentences and observation sentences is, in contrast to the complex causal aspects of

3 4

Quine 1 9 5 3 , 2 0 Quine 1 9 5 3 , 4 1

5 6

Quine 1 9 7 5 a , 3 1 4 - 3 1 5 Quine 1 9 7 4 , 1

92

In Defence of Epistemic Externalism

this relation, relatively unproblematic. To understand how theories are tested by predictions no appeal must be made to psychology, neurology, psycholinguistics or genetics. Observation sentences or, more exactly, their utterances, are true or false, and so it is easy to see how they can function as sources of evidence for theoretical sentences. Little more, Quine says, than "logical analysis" is required for that. 7 But what about the relation between sensory evidence and observation sentences? Of course, observation sentences, such sentences as "This is a rabbit", do not refer to sensory evidence but to intersubjectively accessible physical objects and events. What gives them the meanings that they have, their stimulus meanings, however, are not public objects or events. Rather, the meaning of an observation sentence consists of nothing but the sensory evidence that would verify or falsify it. For Quine, the good old verification theory of meaning is still beyond any shadow of a doubt.8 The concept of sensory evidence, as Quine employs it, proves to be a fairly problematic concept, however. While in his classic Two Dogmas of Empiricism the periphery of the fabric of sentences expressing our theory of the world was described, quite traditionally, as "experience", in later writings Quine went over to offer us a behavioristic substitute for the mentalistic concept of experience: he replaced talk about experience by talk about neural intake at the sensory receptors. The point of contact between theory and empirical world was now characterized as "surface irritations". Thus, the tribunal of experience turns out to be the activations of our nerve endings. 9 Quine literally identifies the sensory evidence, which our beliefs about the world are based upon, with these neurophysiological processes, when he says: "The stimulations of his sensory receptors is all the evidence anybody has had to go on, ultimately, in arriving at his picture of the world." 10 In marked contrast to his empiricist predecessors, Quine drops any awareness requirement concerning the evidential relation. Neural input is neither present to consciousness nor, normally, otherwise cognitively available to us. Few people, statistically speaking, know anything about the physiological processes at their sensory receptors.11 Thus, in order to answer the typical empiricist question of how evidence relates to the theory it supports, Quine directs the focus of his new, naturalized empiricism on the relation between a person's neural input and her verbal output, consisting, at the most basic level, of the utterance of observation sentences. Now, I certainly agree with Quine that each perception of physical objects in space and time is causally mediated by stimulations of our various sense organs. The pressing question, however, is of whether such stimulations of the various sensory modalities are capable of playing a central epistemological and semantic role within the framework of Quine's naturalistic philosophy. Are irritations of our bodily surfaces attractive candidates for the role of sensory evidence which, within Quine's new, enlightened empiri-

7 8 9 10 11

Quine Quine Quine Quine Quine

1990a, 1-2 1969, 81; 1974, 3 8 1981, 40 1969, 7 5 1981, 40

Empiricism Externalized

93

cism, are to be both the starting point of the acquisition of all empirical knowledge and the basis of all communication? In former publications, 12 I have tried to show that there is a deep tension in Quine's philosophy between his official theory of stimulus meanings on the one hand, and his thesis, often propagated, that language is a social art, and that, therefore, open behavior under publicly observable circumstances has to be the focus in the study of language on the other. The conflict is rather obvious. Since we do not share exteroreceptors with other speakers, stimulus meanings are by nature private, but linguistic meaning should be the same within a whole linguistic community. I argued that Quine should give up the concept of stimulus meaning as a central semantic notion. Instead he should give priority to the insights contained in his unofficial theory emphasizing the significance of public utterances about conspicuous external objects for the acquisition and use of language—external objects which the speakers perceive and concerning which they suppose that the persons they talk to perceive them as well. For example, he says: Linguistically, and hence conceptually, the things in sharpest focus are the things that are public enough to be talked of publicly, c o m m o n and conspicuous enough to be talked of often, and near enough to sense to be quickly identified and learned by name; it is to these that words apply first and foremost.13

According to this strand in Quine's thought, there cannot be more to linguistic meaning than what can be gathered from public behavior in socially observable conditions. Internal factors are irrelevant. Nothing is hidden. Meanings, so the saying goes, are not in the head. And neither are they in a platonic heaven. The problem with stimulus meanings is just that they cannot be gathered from observable data—at least not outside of the laboratory. Quine contends that stimulus meanings are to a certain degree intersubjectively controllable because in our practice we take account of the orientation of the speaker and of the relative positions of the objects. We tend to assume that similar objects and situations cause similar patterns of stimulations and that these again cause similar linguistic reactions. If, however, in our well-functioning linguistic practice we are primarily oriented towards shared external checkpoints, should an adequate theoretical representation of this practice then not attach more importance to the socially observable world of external objects and situations and, accordingly, shift the semantically relevant stimuli from the inner, private world into the common, public world? Already in Word and Object, he claimed that the similarity of sensory receptors and neural mechanism ought not to be of crucial importance in a viable theory of language: T h e uniformity that unites us in communication and belief is a uniformity of resultant patterns overlying a chaotic subjective diversity of connections between words and experience. Uniformity comes where it matters socially; hence rather in point of intersubjectively conspicuous circumstances of utterance than in point of privately conspicuous ones. 1 4

12 See Schantz 1996, 98-146; 1998 13 Quine 1969, 1 14 Quine 1960, 8

94

In Defence of Epistemic Externalism

The resultant uniform patterns of linguistic behavior are products of social practices in the midst of a common world. It is such social interactions, by means of which the basic connections between words and external objects are established, which ought to matter in semantics, may the patterns of stimulations or the internal wiring be as private or idiosyncratic as they like. Each speaker is free to internalize the common language in his own neural way as long as her linguistic behavior will tally with the standards of her community. 15 But to really take this insight seriously would mean to give up the idea that stimulus meanings, and neural intakes with them, can play a chief role within the philosophy of language. And whatever concessions Quine may have made meanwhile, 16 the essential character of his theory of language has not really changed. The happenings at our nerve endings, the neural intakes, remain the focal point of his semantic interests. The external objects and events that we usually talk about and towards which our attention is usually directed are still no part of his official behavioristic semantic explanations. The problem, however, as the following passage reveals, is that the seeming plausibility of his explanations in terms of sensory stimulation or neural intake is almost entirely dependent upon his other story—the one that locates the relevant stimuli in the external world: E m p a t h y dominates the learning o f language, both by child a n d by field linguist. In the child's case it is the parent's empathy. T h e parent assesses the appropriateness o f the child's observation sentence by noting the child's orientation a n d h o w the scene w o u l d l o o k f r o m there. [...] We all have an uncanny knack for empathizing another's perceptual situation, however ignorant o f the physiological or optical m e c h a n i s m o f his p e r c e p t i o n . 1 7

I whole-heartedly agree with everything Quine says here. Firstly, because he acknowledges the relevance of the outer perceptual situation. And secondly, because he describes our "uncanny knack" for empathizing another persons perceptual situation as empathizing the way the scene would look from her position, and not as empathizing the sensory stimulations she would receive from there. With respect to neurophysiological processes about which we know nothing the concept of empathy would be entirely misplaced. Therefore, he here prefers to talk about the ways things look to us. But for talk about things looking or appearing thus and so to us to even make sense a richer concept of sensory experience is requisite than the concept of sensory stimulation. Sensory experiences, however, were banished from his official theory of meaning and evidence and replaced by his behavioristic surrogate of the activations of nerve endings. Sensory stimulations, however, are no adequate successors to sensory experiences for both, what we perceive, and how we perceive it, the qualitative content of experience, remains underdetermined by the physical stimuli we receive at our exteroreceptors. Wittgenstein's duck/rabbit example helps to illustrate this kind of underdetermination. Although the lines on the paper and the pattern of stimulation at our retina

15 See also Quine 1990a, 44 16 See Quine 1993. 114, F n l ; 1995b, 349-350. I critically discuss Quine's recent concessions in Schantz 1998 17 Q u i n e 1990a, 42

Empiricism Externalized

95

remain identical, we are capable of seeing either a duck or a rabbit. Whether a speaker assents to or dissents from the query "gavagai?" does not depend solely on his stimulations but, rather, also on whether he sees what he sees as a duck or as a rabbit. Instead of being interested in the enormously complicated systematic correlations between patterns of stimulations and linguistic reactions, correlations which, whether one likes it or not, are mediated by many mental or intentional factors, Quine would have done better to rehabilitate the concept of sensory experience officially as a proper scientific concept, and to investigate, accordingly, the systematic correlations between experiences, the ways things appear to us, and our linguistic behavior. These sensory states with their characteristic qualitative or representational content cannot be reduced to Quine's quite hygienic behavioristic conceptual repertoire. A mere stimulation of our neuroceptors by an external object is clearly not sufficient for this object's appearing to us in a certain way, for its attracting our attention, and for our having a sensory consciousness of it. So I argue for a rehabilitation of a full-blown concept of sensory experience. There are such experiences, and they cannot be equated with a simple stimulus-response dialectic of observation sentences and triggered nerve endings, as Quine himself acknowledges. 18 Such a rehabilitation seems to be compatible with a naturalized form of epistemology. For empiricists, the central epistemological task is then, as it has always been, to scrutinize the ways in which, on the basis of our sensory experiences, we build up, control and revise our dynamic systems of belief about the world. Quine's unorthodox construal of sensory evidence leaves him with no resources to adequately take into account the important difference between a causal explanation of a belief, on the one hand, and an epistemic justification of it on the other. Most epistemologists take for granted that there is a tight connection between the concepts of justification and of evidence. A belief, it is typically said, is justified for a person just in case she has adequate evidence which supports it. Quine drops the notion of justification, but nonetheless wants to cling to the notion of evidence. For him, though, evidence is no longer what it was before—what a person can cite or put forward to justify or ground her beliefs about the world. Rather, sensory evidence is subjected to a rational reconstruction and, in accord with his genetic method, transformed into sensory causes. The various activations of the receptors are nothing but efficient causes for the scientific theories we construct about the world. It is no wonder then that many renowned philosophers have accused Quine of committing the genetic fallacy of conflating reasons and causes, justifications and explanations. This objection has been raised in one form or another by Richard Rorty, by Hilary Putnam, and by Barry Stroud. 19 And it has been raised by Donald Davidson as well.

18 19

Quine 1 9 8 1 , 185 See Rorty 1979, 2 1 3 - 2 5 6 ; Putnam 1 9 8 3 , 2 2 9 - 2 4 3 ; Stroud 1 9 8 4 , 2 0 9 - 2 5 4

96

In Defence of Epistemic Externalism

IV For a long time Davidson has urged Quine to give up his empiricism. He maintains that Quine might have taken this step without abandoning his naturalized epistemology at the same time. 20 After some initial resistance, Davidson meanwhile regards himself as an advocate of naturalized epistemology. What he rejects is a naturalized empiricism. Davidson claims that Quine himself has fallen victim to a dogma of empiricism, the third dogma, namely, the dualism of conceptual scheme, or language, and empirical content—"The third, and perhaps the last, for if we give it up it is not clear that there is anything distinctive left to call empiricism". 21 "Empirical content", as Davidson uses the term, includes such items as experiences, sensations, sense-data, and, of course, patterns of stimulations. 22 The idea underlying the scheme-content dichotomy is that there is an unconceptualized given element in experience which might be variously interpreted, or variously organized, by different conceptual schemes. The dogma of a dualism of scheme and unorganized content, so Davison argues forcefully, leads to conceptual relativism, to relativizing objective truth and objective reality to a conceptual scheme. And it leads to epistemological scepticism. Davidson attacks this dualism from both sides, from the side of the very idea of a conceptual scheme and, what chiefly interests me here, from the side of the idea of a given element untouched by our interpretative activities. He argues that empiricism must be abandoned because its central idea that meaning and knowledge rest on a sensory source of evidence is untenable. Certainly, he concedes, meaning and knowledge depend on perceptual experiences and sensory stimulations, but in a causal sense, not an evidential sense. Experiences, sensations and stimulations play an important causal role; they are causal intermediaries between external objects and events and our beliefs and talk about them. The main mistake of empiricism—in its enlightened no less than in its traditional form—, so runs Davidson's diagnosis, is to confuse the weighty causal role of the senses with an essential epistemological and semantic role, to confuse, in other words, causal with epistemic intermediaries. In a key passage he says: The relation between a sensation and a belief cannot be logical, since sensations are not beliefs or other propositional attitudes. W h a t then is the relation? The answer is, I think, obvious: the relation is causal. Sensations cause some beliefs and in this sense are the basis or ground of those beliefs. But a causal explanation of a belief does not show how or why a belief is justified. 23

To credit experiences—which, according to his anomalous monism, are token-identical with physical events—with an epistemic role is to conceive them as epistemic intermediaries, internal interfaces, between our beliefs and the objective world. And he is convinced that such a conception leads to scepticism because we cannot be sure that such intermediaries supply us with reliable information about the objective reality. In

20 21 22 23

Davidson Davidson Davidson Davidson

2001, 1984, 1984, 2001,

193-194 189 189 143

Empiricism Externalized

97

giving up epistemic intermediaries, the traditional veil of ideas, he attempts to make sure that we gain direct, or immediate, perceptual knowledge about external objects and events. So Davidson is, although he himself does not use the label, a proponent of "direct realism". He holds, that is, that we directly perceive external physical objects and events, not mind-dependent sense-data or appearances. His emphatic dismissal of epistemic intermediaries entices Davidson to further deny that what happens at our sense-organs has anything to do with justification or evidence at all. An experience cannot count as a reason for holding a belief, and hence has no influence on the question whether a belief is rationally warranted or not. Experiences can stand in causal relations to belief-states, but not in logical relations to the content of beliefs. For Davidson, justification is an exclusively logical affair. T h e conclusion he draws is that the dimension of sensory and perceptual experience is epistemically or evidentially completely irrelevant. According to Davidson, we have an epistemically direct access to external objects and events which via multifarious and complicated causal chains give rise to our beliefs about them, and these external objects and events are semantically and epistemologically of central importance; after all, it is they that determine the content of our beliefs and the meanings of our words. So Davidson is a resolute semantic externalist. Here, he thinks, and rightly so, he can ascertain a fundamental difference to Q u i n e whom, on the strength of his insistence on a private source of evidence, Davidson ranks as an obstinate internalist. Concerning the interpretation of observation sentences, Davidson says: T h e most i m p o r t a n t difference concerns the objects or events that d e t e r m i n e c o m m u n i c a b l e content. For Q u i n e this is the patterns o f nerve endings which p r o m p t assent to a sentence [...] M y a p p r o a c h is by contrast externalist: I suggest that interpretation d e p e n d s (in the simplest and m o s t basic situations) o n the external objects a n d events salient to both speaker a n d interpreter, the very objects a n d events the speaker's words are then taken by the interpreter to have as subject matter. 2 4

Davidson tried to persuade Q u i n e to relinquish his "proximal" theory of meaning and evidence once and for all, and to adopt instead Davidson's own "distal" theory. 25 Davidson too points out that there certainly are passages in Quine's writings in which he himself seems to favour a distal account, but that in his official doctrine the stimulations of his nerve endings gain the upper hand. Davidson's main objection to Quine's proximal theory is that, just like older empiricist or verificationist approaches, it leads to global scepticism: In spite o f abjuring sense data and the reduction o f theoretical terms to observational, Q u i n e ' s proximal account o f m e a n i n g and evidence leads to skepticism in m u c h the s a m e way as older theories did. T h e reason is that skepticism rests on the a s s u m p t i o n neither o f sense d a t a nor o f reductionism, but on the m u c h m o r e general idea that empirical k n o w l e d g e requires an epistemological step between the world as we conceive it a n d our c o n c e p t i o n o f it, a n d this idea is at the heart o f Q u i n e ' s proximal theory. 2 6

24 25 26

Davidson, 1990b, 321 See Davidson 1990a Davidson 1990a, 74

In Defence of Epistemic Externalism

98

Davidson goes so far as to say that proximal theories are "Cartesian in spirit and in consequence" 2 7 . So not even his unwavering physicalism can preserve Q u i n e from Cartesianism. Proximal theories are Cartesian insofar as they start from the perspective of the first person and take the realm of the subjective as the basis of all empirical knowledge. According to the Cartesian starting-point, the acquisition of knowledge about the world therefore requires an epistemic step from the realm of the subjective to the objective, mind-independent world. Davidson thinks that Q u i n e remains trapped in this old idea. And Davidson rejects this old idea as untenable, because nothing can enable us to bridge the gulf, once opened up, between the internal world and the external world. Davidson contends that Q u i n e cannot avoid sceptical consequences because the stimulations of a person might be just the way they are although the world outside might in general be quite different than she, on the basis of her stimulations, believes it to be. If, for example, we connect a person to a stimulation machine that induces the same sensory input in her as is usually induced by a rabbit scurrying by, then this person would, on the proximal theory, mean and believe the same by "rabbit" as we do. But she would be massively wrong, for her beliefs would not chime in with the facts in the world. If the proximal theory were true, then two speakers might be in accord concerning the meaning of all their sentences, and yet one and the same sentence might have different truth conditions. It might have different truth conditions since it is possible that different distal events cause similar patterns of stimulation at our sensory surfaces. Truth and meaning, belonging together so closely in Davidson's truth-theoretical semantics, seem to fall apart in Quine's theory of meaning. It is the concept of evidence, not the concept of truth, that plays the key role in Quine's verificationist theory of meaning. Davidson charges Q u i n e of never answering the decisive questions, firstly, of what exactly constitutes the sensory evidence on which our beliefs about the world are based, and, secondly, of what the connection is between sensory evidence and sensory stimulation. As we have seen, Q u i n e does answer these questions. It seems that Davidson cannot, and does not want to, admit that Quine, in many places, literally identifies sensory stimulation and evidence. For example, Q u i n e says in a passage, I have already quoted, and which Davidson himself quotes: T h e stimulations o f his sensory receptors is all the evidence a n y b o d y has h a d to g o on, ultimately, in arriving at his picture o f the w o r l d . 2 8

Davidson resists this identification because under normal circumstances we do not know anything about what happens at our nerve endings. Quine himself indicates that we have no awareness of such physiological processes, and he even regards it as a special benefit of his theory that it shuns questions about the focus of consciousness altogether. But if, under normal conditions, these processes are not cognitively accessible to us, how, then, Davidson asks, can they function as evidence? Here Davidson's epistemological

27 28

Davidson 1990a, 76 Quine 1969, 75

Empiricism Externalized

99

internalism comes to light. N o factor can be relevant to the justificatory status of a belief unless we are, or can become, aware of it. Semantically, with respect to content and meaning, Davidson is a robust externalist. But epistemologically he belongs to the party of internalists. Therefore, neither sensory stimulations nor perceptual experiences can be justifiers although beliefs about them can be. The only concept of evidence Davidson is prepared to accept is the concept of a logical relation between beliefs or between the sentences which express them. Consequently, he recommends a sort of coherentism, proclaiming that "[...] nothing can count as a reason for holding a belief except another belief". 2 9 Empiricism must be abandoned because the attempts to find an epistemic basis for knowledge and justification outside the scope of our beliefs, the attempts, in other words, to ground justification ultimately on the evidence of the senses are all in vain. Davidson likes to characterize such attempts as requiring a "confrontation" between our beliefs and objective reality, and he dismisses this idea as absurd. Since there is nothing other than our beliefs to which we can appeal in the justification of a belief, he rejects Quine's putative assumption that there is an epistemic relation between the holding true of observation sentences and sensory stimulations or neural intake as simply incoherent, as a kind of category mistake so to speak. In Davidsons eyes, sensory stimulations are not suited to serve as evidence on which we might base our beliefs about the world. But what assurance do we have that our beliefs, if coherent, are true? In contrast to Richard Rorty, for example, Davidson considers the question of whether coherence alone is a guarantee of truth, to be a legitimate one. T h e resources to meet the sceptical challenge, i.e., why could not our beliefs cohere in the right way and yet be systematically false about the real world, are to be found in his renowned account of the foundations of linguistic communication—his theory of radical interpretation—which leads to the conclusion that "[...] belief is in its nature veridical". 3 0 His famous principle of charity dictates that an interpreter must so interpret as to find her subjects mostly right about the external objects and events with which she can observe them causally interacting. That is the reason why Davidson thinks that scepticism cannot get off the ground. He reassuringly says: " T h e question, how do I know my beliefs are generally true? thus answers itself, simply because beliefs are by nature generally true". 3 1 In light of these anti-sceptical advantages, Davidson gave Q u i n e the well-meant advice to discard the allegedly superfluous concept of sensory evidence, and to join him in embracing coherentism. And indeed Q u i n e seemed to be willing to take Davidson's good advice to heart. In a short reply, he says: I agree that in m y theory o f evidence the term "evidence" gets n o explication a n d plays n o role. 3 2

29 30 31 32

Davidson 2001, 141 Davidson 2001, 146 Davidson 2001, 153 Quine 1990b

100

In Defence of Epistemic Externalism

And elsewhere he remarks that "evidence" and "observation" will no longer be technical terms for him. 3 3 Nonetheless, with regard to the crucial and controversial question of the location of the relevant stimulus Quine remains rather uncompromising. He is still firmly determined to prefer his own proximal theory to Davidson's distal theory: But I remain unswerved in locating stimulation at the neural input, for my interest is epistemological, however naturalized. I am interested in the flow of evidence from the triggering of the senses to the pronouncements of science. My naturalism does allow me free reference to nerve endings, rabbits, and other physical objects, but my epistemology permits the subject no such starting point. His reification of rabbits and the like is for me part of the plot, not to be passed over as part of the setting.34

Here Quine draws our attention to a fundamental difference between his own philosophical interests and those of Davidson. The question by which Quine is captivated is the question of the "flow of evidence", from sensory stimulations, the epistemological starting point, up to the claims that our scientific theories make about the world. It is the task of science to predict and control future sensory stimulations in the light of past stimulations. For him, our talk about external material objects is a conceptual apparatus that helps us to generate systematic correlations between our stimulations. He wants to explain the nature of the reification of bodies and their efficiency for integrating our system of the world. 35 According to Quine, the sphere of stimulations is basic; he calls, relative to this sphere, external objects "posits" and "hypotheses" 36 , and even "myths". 37 The hypothesis of ordinary physical objects is in its function and survival value similar to the hypothesis of the existence of molecules. Both hypotheses are underdetermined by the only evidence available to us. But, so Quine tries to reassure us, in spite of their merely hypothetical character, we have the best reasons to believe in the existence of ordinary physical objects. His naturalism demands to decide questions of existence immanently, within the system of the world we presently accept, within its ideology and standards. There is no higher tribunal, no first philosophy. The best thing we can do, therefore, is to take up the perspective of a historical theory and to respect its ontological commitments. And, for Quine, it is the theory of external physical objects that best succeeds in organizing the realm of our multifarious sensory stimulations. Quine reproaches Davidson for completely ignoring the important dimension of reification, the psychogenesis of reference; for Davidson, this dimension is, as Quine puts it above, not "part of the plot", but only "part of the setting". Davidson can counter, however, that this objection presupposes the untenable dualism of an invariant, uninterpreted content, which is simply given in experience, and our variant conceptual and theoretical contributions, our interpretations of the raw data. Davidson is convinced that this dualism inescapably leads to conceptual relativism and to epistemological

33 34 35 36 37

Quine Quine Quine Quine Quine

1990a, 2 1990a, 4 1 - 4 2 1990a, 2 3 - 2 5 , 2 9 - 3 1 I960, 2 1 - 2 5 1953, 4 4

Empiricism Externalized

101

scepticism. In opposition, Davidson claims that the existence of the familiar external objects is not a mere hypothesis for there is no ultimate source of evidence which could be identified and described without recourse to the external world. Whereas Quine's empiricist approach tears open a deep gap between the inner and the outer world, Davidson thinks that his own rejection of the third dogma of empiricism re-establishes our direct contact to the familiar external objects whose properties make our sentences and beliefs true or false. Davidson says: In giving up the d u a l i s m o f s c h e m e a n d world, we d o not give u p the world, but re-establish u n m e d i a t e d touch with the familiar objects w h o s e antics m a k e o u r sentences a n d o p i n i o n s true or false. 3 8

I share Davidson's view that it is the distal stimulus that matters to interpretation. It is external objects and events, salient for both the speaker and the interpreter, that give our words the meaning and our thoughts the content that they have. In a word, I share Davidson's semantic externalism that Q u i n e ought to have given a more prominent role in his semantics and his theory of knowledge. I share, that is, Davidsons anti-verificationist, truth-conditional semantics. It is truth that holds the key to meaning, not evidence. Semantic empiricism is unworkable. Moreover, I agree with Davidson that an adequate epistemology should not open the door to scepticism, but should rather ensure our direct cognitive access to the external world. Davidson thinks, however, that we can only stop the sceptic if we are prepared to abandon empiricism, which he sees as committed to epistemic intermediaries, once and for all. According to him, empiricism rests on a confusion of the causal role of the sensory systems with an epistemic role. Consequently, he argues for a separation of the study of perception from the philosophical theory of knowledge. 3 9 I think that Davidson's disregard for the dimension of perception and of sensory experience in the field of epistemology is a serious philosophical mistake. Internal coherence is not enough if our beliefs are to be genuinely empirical. With Quine, I hold on to the epistemic and evidential relevance of sensory experience—even though our opinions about the nature of experience differ considerably. Our responsible beliefs about the world must remain under the strict control of the deliverances of the senses. Epistemological empiricism must not be abandoned. O f course, traditional or radical empiricism is obsolete. A modern and moderate version of empiricism should not include the foundationalist idea that we directly perceive only sense-data, and on their basis somehow have to infer the existence and the nature of the external world. What we need instead is a version of empiricism that refuses to restrict direct perceptual knowledge to an inner realm of peculiar sensory entities, one allowing us to endorse the basic insight of direct realism, the insight that we directly perceive external objects and states of affairs. There is no screen of sense-data to hide them away from us. Indirect versions of realism are hopeless. Their main mistake is to hypostasize the sensory content of experience, thereby making a mystery of intentionality, of thoughts' ability

38 39

Davidson 1984, 198 Davidson 2001, 45-46

102

In Defence of Epistemic Externalism

to be about external objects at all. Moreover, a modern version of empiricism should be in harmony with the lessons of both semantic and epistemological externalism.

V We have seen that Quine's version o f naturalized empiricism has no internal resources to d o justice to its own juridical rhetoric of facing the tribunal of experience. So the crucial question that a tenable form o f empiricism must answer is just the question of what conditions experience must satisfy if it is to function as an authentic tribunal. In his subtle study Mind and World, J o h n McDowell has proposed an ingenious answer. His views on the intentionality o f perceptual experience will serve as a suitable foil to accentuate m y own point o f view. H e too urges that we should not abandon "minimal empiricism", which he describes in the following way: [...] in the sorts o f case that m u s t c o m e first for reflection o n the very idea o f directedness at the world, the world's verdict, to which thinking m u s t be answerable if it is to be thinking at all, is delivered by way o f a p r o n o u n c e m e n t f r o m (in Q u i n e ' s phrase) "the tribunal o f experience". 4 0

McDowell espouses both epistemological and semantic empiricism. 4 1 For him, both the justificatory status o f a belief and its content are only comprehensible in relation to sensory experience. O n l y if we can see how empirical thinking is responsible to the tribunal o f experience, will we be able to answer in a satisfactory way the fundamental question how empirical content—world-directedness—is as much as possible. McDowell thinks o f this "transcendental" question o f how empirical content, objective purport, is possible as the real source o f the typical epistemological problems that pervade modern philosophy—especially, o f course, o f the problem how empirical knowledge is possible. For this reason, he also characterizes his position as "transcendental empiricism". 4 2 McDowell argues that a pair o f opposite and equally untenable views about the relation between m i n d and world have dominated philosophy after the rise o f modern science. T h e poles o f this dilemma, or "intolerable oscillation", or "seesaw", are the Myth o f the Given, on the one hand, and Coherentism on the other. McDowell attempts to impale Q u i n e on the former horn and Davidson on the latter, and then to force his own way between the horns o f this dilemma. McDowell objects that coherentism, by acknowledging only causal relations between our beliefs and experiences, threatens to disconnect empirical thinking from independent reality. 43 Internal coherence alone, with no experiential grounding, will not do. D a v i d s o n s picture renounces, as McDowell likes to put it, "rational constraint" from the external world on the dynamic web of our beliefs. 4 4 Moreover, he argues that the

40 41 42 43 44

McDowell 1995, 231-232 See Brandom 2002 McDowell 2002, 287 McDowell 1994 McDowell 1994, 25

Empiricism Externalized

103

versions of coherentism advocated by Richard Rorty and Wilfrid Sellars go wrong for essentially the same reason. T h e activity of justifying empirical beliefs is not a self-contained game but must be rationally responsive to the world itself which is disclosed to us through the deliverances of the senses. McDowell insists that if we are to understand how exercises of concepts can constitute justified empirical judgments about external reality, how our thinking can be genuinely empirical, we must recognize rational relations between experiences and beliefs; we must, in other words, make room for the idea that our experiences can be related to our beliefs as grounds or as reasons for them. McDowell maintains that Davidson's denial of the epistemic significance of experience will lead philosophers to recoil back into the Myth of the Given. This is, as McDowell, following Wilfrid Sellars, describes it, the view that the space of reasons extends more widely than the space of concepts, in the sense that epistemic justifications have an ultimate foundation in bare, nonconceptual presences—something which is simply received in experience independently of acquired conceptual capacities, and to which we can only point. T h e idea is that by appealing to a given element in experience the required rational constraint from outside the sphere of thought and judgment can be ensured. McDowell shares with Davidson the conviction that the relations in virtue of which a belief is justified—relations such as implication and probabilification—can only be conceived as relations which hold between conceptually organized items. For them, the attempt to extend the space of reasons so that it can include nonconceptual impingements from the world is hopeless. If experiences are conceived as nonconceptual, they cannot be a rational basis, a source of justification, for our beliefs or worldview. So far McDowell agrees with Davidson. Nevertheless, he cannot accept Davidson's unconstrained coherentism. In order to escape the intolerable oscillation between the coherentist threat that empirical thinking is, as McDowell puts it, "frictionless spinning in a void" 4 5 on the one side and the Myth of the sensuous given on the other, McDowell endeavours to develop a more suitable conception of experience. He aims to rehabilitate the Kantian thought that empirical knowledge results from the interplay between sensibility and understanding, between receptivity and spontaneity—sensibility being the source of intuitions, the experiences of outer and inner sense, while from the understanding arise concepts. Against this Kantian background, McDowell expresses the central idea of his conception of experience in the following way: "receptivity does not make an even notionally separable contribution to the co-operation." 4 6 In other words, the operations of sensibility must engage those very capacities which Kant attributes to the understanding. O n McDowell's view, experiences are states which are passive, products of receptivity, and yet they are already possessed of conceptual content. Conceptual capacities are not exercised on a nonconceptual given but are already drawn on in receptivity. 47 It is not that concepts first come into play

45 46 47

McDowell 1994, 11 McDowell 1994, 9 McDowell 1994, 9-13

104

In Defence of Epistemic Externalism

in beliefs or judgments which are based on experiences; they are already at work in the experiences themselves. Our experiences represent things to be thus and so. And when we are not misled, we take in how things are. That things are thus and so is the conceptual content of an experience. A perceptual judgment simply endorses the conceptual content already possessed by the experience. Experience, being characterized as "openness to the layout of reality" 48 , is supposed to enable independent reality itself to exert a rational influence on the evolving web of our beliefs. The postures we adopt, our empirical judgments that things are thus and so, are answerable to the world through being answerable to experience. In this way McDowell hopes to make intelligible how experiences, impressions made by the external world on our sensory equipment, can stand in rational, not merely in causal, relations to beliefs, how an experience can be someone's reason for a belief. Although he holds that experiences are conceptually articulated McDowell rejects the view, developed in the first instance by David Armstrong and George Pitcher 49 , that they are beliefs or inclinations to beliefs. 50 How things appear to us is not under our control, but McDowell urges that it is up to us to decide whether or not to believe that things are as the appearances represent them to be. Not that he assumes that belief is always a matter of choice, that all beliefs are under direct voluntary control. It can hardly be denied that some beliefs we simply find ourselves with. Nevertheless, he wants to reserve the title of belief and judgment for a kind of cognitive state which essentially belongs within the scope of a subject's actively making up her mind about something, and with respect to which the question of justification can always be raised. Thus, experiences, according to McDowell, are through and through conceptually informed, and even always propositionally contentful, and yet prejudgmental and nondoxastic.

VI By acknowledging the epistemic significance of experience, McDowell's picture gains an advantage over Davidson's coherentism. In contrast to what he hopes, though, I think his position is not the right place for our thinking to come to rest. I want to defend, rather, a version of the view McDowell brands as a myth, namely, that the space of justificatory relations extends further than the space of concepts. Firstly, I will defend the claim that there is a given element in experience which is independent of thought and which is possessed of a distinctive nonconceptual content. In this regard, I sympathize with C.I. Lewis who remarked: "There is in all experience that element which we are aware we do not create by thinking and cannot, in general, displace or alter." 51 And secondly, I will argue that this given element, properly naturalized, is indeed fit to play an important evidential role in the justification of beliefs about physical objects—though not, of course, the gigantic role classical foundationalism credited it with.

48 49 50 51

McDowell 1994, 26 See Armstrong 1961; Pitcher 1971 McDowell 1994, 60-63, 140 Lewis 1956, 48

Empiricism Externalized

105

I agree with McDowell that perceptual experience is not itself a form of belief. No beliefs are essential to perceptual awareness. All that is necessary for my perceiving something is that it phenomenally appears a certain way to me. Where I differ from him, is in holding that these appearings do not necessarily involve the possession and application of concepts. Perceptual experience is not itself a form of recognizing, identifying, or classifying the things around us. No doubt, to recognize something as a crocodile, or to see that it is a crocodile, is to form the belief that it is a crocodile, and, like all beliefs, this requires applying concepts. But not all perceiving is perceiving as or perceiving that,52 It seems obvious that we can see a crocodile, even though we do not recognize it as such, and even though we do not possess the concept of a crocodile. Seeing a crocodile, for example, is constituted by certain visual experiences, certain ways the crocodile looks to us, and these experiences require no conceptualization, no understanding of what sort of thing a crocodile is. Lack of concepts does not blind us to the entities we phenomenally experience; it only prevents us from identifying them as the entities they in fact are. We must not confuse describing an experience, or its phenomenal content, with the having of an experience. Describing an experience involves the deployment of concepts, having an experience does not. To forestall possible objections—my claim is that sensory experiences are conceptually independent of beliefs or thoughts. I do not want to go so far as to claim that experiences are always causally completely independent of doxastic states or conceptual capacities. Rather, I admit that our beliefs, hopes, and expectations do causally influence—in certain cases, and to a certain degree—the qualitative features of our experience. The crucial point, however, is that there are limits to the extent to which the content of propositional attitudes can affect perceptual structure. The Miiller-Lyer illusion nicely illustrates that experiential content is not wholly penetrated by what we know and believe. Our knowledge that the two lines are the same length does not make it phenomenally appear to us as if they are the same length. As Jerry Fodor, in an important contribution to perceptual psychology, has persuasively argued, perceptual content is modular, informationally encapsulated from our knowledge and beliefs. 53 How we believe things to be does, in an important sense, not affect how they appear to us. Experiential content is, to a significant degree, immune to high-level cognitive influences. Perception is not as plastic, as malleable by background assumptions, as Thomas Kuhn, Paul Feyerabend, Paul Churchland and company try to make us believe. I have argued that there is such a phenomenon as immediate, nonconceptual awareness of a given element in experience. It should be clear by now that this given element, as I understand it, is different from mere sensory stimulation. The purely physical stimulation of one's sensory receptors by an object is clearly not sufficient for one's being sensorily presented with, and thus aware of, that object. And the sensuous given also need not be regarded as a peculiar nonwordly item—a sense-datum as it was traditionally conceived. The given is not a sensory object but the nonconceptual content of

52 53

See Dretske 1969, 1981; Schantz 1990 See Fodor 1983, 1984

In Defence o f Epistemic Externalism

106

a nonrelational sensory state. As a defender o f the adverbial analysis o f sensory experience, I hold that when undergoing a sensory illusion we do not sense any object, but sense in a particular objectless way. 5 4 In my eyes, this move is the best way to block the famous Argument from Illusion, the most powerful method o f introducing sensedata from antiquity to the present. My motto is to save appearances without returning to classical foundationalism. T h e case for nonconceptual representational content rests in part on the supposition that experience can have a finer-grained content than can be formulated by using concepts at the disposal o f an ordinary subject. T h e nonconceptualists claim that we do not possess, say, as many colour concepts as there are shades o f colour we can sensorily discriminate. Our repertoire o f colour concepts, expressible by words like "blue" or "red" or "scarlet" is too coarse-grained to capture the finest detail, the specific character, o f our colour experiences. O f course, McDowell has to dismiss the thesis that the content o f perceptual experience is completely, or even partly, nonconceptual. And indeed that is what he does. 5 5 In a discussion with McDowell, I have argued that his reply to the phenomenological case for nonconceptual content is not really satisfactory. 56 His response to my criticism still leaves me largely unconvinced. 5 7 Be that as it may, McDowell is deeply persuaded that states with nonconceptual content cannot intelligibly constitute a subjects reason for believing something. Against Christopher Peacocke's assertion that nonconceptual experiences can be justifiers for our beliefs, McDowell objects that he "has to sever the tie between reasons for which a subject thinks as she does and reasons she can give for thinking that way. Reasons that the subject can give, in so far as they are articulable, must be within the space o f concepts." 5 8 He points out that an unsurprising answer by an ordinary subject to the question why she holds some observational belief about a perceived object might be: "Because it looks that way". 5 9 This reason too, he claims, is articulate, at least "minimally", and then he continues "so it must be no less conceptual than what it is a reason for". 6 0 In my aforementioned paper, I have argued that, even if we grant the requirement o f minimal articulability, McDowell's suggestion fails. W h e n a subject says "Because it looks that way", she expresses a demonstrative thought, and o f course this thought is a conceptually contentful intentional state. But that only shows that the articulation o f the reason involves the use o f the demonstrative concept "that way". It does not show, what McDowell needs to show, that the reason itself, the experience the demonstrative refers to, has conceptual content. From the conceptual character o f the demonstrative "that way" it simply does not follow that the way demonstratively referred to itself has

54

See Schantz 1 9 9 0

55 56 57 58 59 60

McDowell 1994, See Schantz 2 0 0 1 McDowell 2 0 0 1 McDowell 1994, McDowell 1994, McDowell 1994,

44-65

165 165 166

Empiricism Externalized

107

a conceptual character. As far as McDowell's argument from articulability is concerned, the way might well have a nonconceptual character. Actually, I think, however, that we need not and should not acknowledge that there is any requirement of articulability. In this requirement McDowell's internalism about justification comes to the fore. The factors that determine the epistemic status of a belief must be cognitively available to the subject's mind; they must themselves be known or at least knowable just by carefully reflecting on the matter. Interestingly, McDowell combines his form of epistemological internalism with a rather strong version of content externalism. His reflections on the relationship between mind and world are centred around the assumption that the representational content of a subject Ss mental state is determined by factors that are external in the sense of being part of 5s environment. Having a veridical perceptual experience is, under standard conditions, having an environmental fact made manifest to us. These sensory episodes are individuated by the very external facts they disclose to us, not by how it is with one subjectively. So, for McDowell, the existence of perceptual experiences depends on the obtaining of the corresponding facts. Against the persistent content internalism of the Cartesian picture, he argues convincingly that cognitive space, the space of our cognitive operations, is not a realm whose structure is entirely independent of external reality. Rather, in his view, mental content in general requires directedness towards the external world. The mental is not internal, not in the head, as the saying goes. Since object-dependence, world-involvingness, is a feature of the intentional nature of a thought, there is no gap between intentionality and its objects. 61 No interface needs to mediate between subject and object. So content externalism paves the way for McDowell's direct realism. One might even think that McDowell's content externalism entails a version of epistemic externalism, a version of the view that factors external to our perspective, factors that cannot be known by introspective reflection alone, can contribute to the justificatory status of a belief. According to his "disjunctive" account of experience, which he prefers to traditional "highest common factor" accounts, veridical perceptual experiences and illusory experiences do not have a common intentional object—the former is about a real external object, whereas the object of the latter is a mere appearance. 62 And he holds that the ontological difference between illusory and non-illusory sensory episodes need not be ascertainable simply through introspection. So it seems that whether or not a given episode gives rise to positive epistemic standing is not something that is accessible just on reflection. On the other hand, however, McDowell makes it clear that he wants to hold on to the "root idea" of internalism, which he characterizes in the following way: The root idea is that one's epistemic standing on some question cannot intelligibly be constituted, even in part, by matters blankly external to how it is with one subjectively. For how could such matters be other than beyond one's ken? And how could matters beyond one's ken make any difference to one's epistemic standing? 63

61 62 63

See McDowell 1998, 2 5 5 See McDowell 1998, 3 8 6 - 3 8 7 McDowell 1998, 3 9 0

108

In Defence of Epistemic Externalism

For McDowell, the factors that determine the epistemic status of a belief must not be "blankly external"; whether or not they obtain must make a difference to one's total epistemic situation. Hence, finally, his position seems to be that though worldly facts are external, in the sense of not being reducible to mental facts, they are not "blankly external", not beyond one's ken, since they are not outside the reach of our rational powers. After all, we can directly perceive physical facts, we are confronted with them, and thereby acquire non-inferential knowledge about them. Thus, worldly states of affairs can be internal to a person's first-person cognitive perspective, in the sense of being accessible to that person from that perspective. So McDowell is indeed an advocate of a moderate version of epistemological internalism. The requirement of articulability has to be understood in this light. It seems to me, however, that this requirement ultimately rests on a confusion between having reasons and giving reasons or, in other words, between the state of being justified and the activity of justifying the belief, of showing it to be justified. If one neglects this important distinction, one will be inclined to suppose that being justified entails the ability of showing that one is justified, that is to say, of producing a justifying argument. Now, of course, to successfully carry out the activity of justifying a belief one must appeal to other beliefs or other conceptually contentful states. But from the fact that we can justify a belief only by showing that it sustains appropriate relations to other conceptually contentful states, it does not follow that a belief can be justified only by its relations to other conceptually contentful states. We would have only few justified beliefs, if their justification depended on having engaged in the practice of establishing their credentials. 64 For a belief to be justified it does not even seem to be necessary to be capable of engaging in such activities. Many subjects lack the intellectual or linguistic capacities requisite to articulate their reasons. A subject may have, in the particular way an animal appears to her, an adequate reason to believe that it is a crocodile, without being able to describe or conceptually characterize the way the animal appears to her, and hence without being able to provide a justifying argument. Our incapacity to specify the significant features of our experience of an object does not prevent us from having, in the way it appears to us, a reason to believe that it is thus and so. Note that I do not claim, as C.I. Lewis and many of his followers did, that the given is ineffable. Ordinary people are capable, at least in principle, of characterizing in conceptual terms the large-scale aspects of their experiences, and some artists are even capable of giving fairly detailed phenomenological descriptions of their visual fields. My point is, rather, that the real evidential force of experience can hardly be fully captured by, and is not dependent on, such conceptual formulations as accurate as they may be. The evidential force, the reason-constituting status of sensory content, is reflected in the fact that the truth of corresponding beliefs about the physical world provides the best explanation for our experiences having the specific features that they have. McDowell, on the other hand, seems to think of all empirical knowledge as higher level in character. For him, epistemic justification in general consists of the reflective

64

See Alston 1989

Empiricism Externalized

109

activity of giving and asking for reasons, of continually scrutinizing and criticizing the rational warrant for our judgments. It is not enough for our perceptual beliefs to have epistemic authority—for example, in virtue of being the result of reliable belief-forming processes, that is, processes that are likely to generate true beliefs. In the wake of Sellars' epistemological point of view, McDowell insists that the authority of perceptual beliefs must be recognized as such by the person whose belief it is, and this requires that its rational credentials are available for critical scrutiny. But why should we suppose that the acquisition of basic empirical knowledge essentially involves knowledge about the rational status of our beliefs? N o doubt, to possess such meta-knowledge is, in certain contexts, an extremely useful thing—for example, when we undertake to answer the sceptic. But we should not allow the critical evaluation of our beliefs, the demanding task of reflectively taking charge of the shape of our active thinking, which may indeed be necessary in special contexts to be build into the very idea of experience or of empirical content. It is a consequence of McDowell's brand of normative epistemology that experience of the world is restricted to users of language. Following Kant, he even proclaims that the application of concepts requires self-consciousness, the potential presence of the "I think" of apperception. In contrast, my position allows us to preserve the strong intuition that animals and prelinguistic infants can enjoy sensory episodes of the very same kinds as users of language. We should acknowledge that the sensory and conceptual capacities of animals and humans lie on a continuum. According to my alternative picture, experiences are subjective representational states with objective nonconceptual content, and yet they can serve as justifiers, evidential probability-makers, for our beliefs. The epistemic significance of experience does not depend on its conceptualization. If, under normal circumstances, an object appears redly to me, then, it is quite plausible to say, I am more justified in believing that it is red rather than that it is blue or yellow. In such a situation I have, in the way the object appears to me, a reason to believe that it is red. Thus, the positive epistemic status of a perceptual belief depends upon being appeared to in appropriate ways. By saying that a subject has, in the way an object appears to her, a reason to believe that it is thus and so, I do not mean to suggest that the subject must have arrived at her belief by any process of reasoning or inferring. So my appearings are, in McDowell's terminology, "transparent", not "opaque" 6 5 ; they themselves disclose features of the environment to us. The subject does not have to use the way the object appears to her as a reason, as a premise for concluding, that it is thus and so. N o conscious discursive process needs to mediate the objects appearing thus and so to the subject and the subject's consequent belief that it is thus and so. It is important to note that for an ordinary perceptual belief to be justified, one does not have to believe that one is appeared to in these ways. It is the experiences themselves, the ways of being appeared to, not our beliefs about them, that are required for justification. In the ordinary course of things, we rarely have beliefs about appearances. Our perceptual beliefs are usually about external objects and events—not about, as they

65

McDowell 1994, 145

110

In Defence of Epistemic Externalism

say, qualia. T h e position I defend is, therefore, a type o f direct realism. W e acquire by the senses, normally, direct knowledge o f physical objects and events. W h a t makes this knowledge direct is that it is not based o n other knowledge or beliefs.

References Alston, William: 1989, Epistemic Justification. Essays in the Theory of Knowledge, Cornell University Press, Ithaca & London Armstrong, David: 1961, Perception and the Physical World, Routledge & Kegan Paul, London Brandom, Robert: 1995, "Perception and Rational Constraint: McDowell's Mind and World', Philosophical Issues, 7 Perception, 241-259 Brandom, Robert: 2002, "Non-inferential knowledge, perceptual experience, and secondary qualities: placing McDowell's empiricism", in Smith, Nicholas (ed.), 2002, 92-105 Davidson, Donald: 1984, Inquiries into Truth and Interpretation, Clarendon Press, Oxford Davidson, Donald: 1990a, "Meaning, Truth and Evidence", in: Barrett, Robert & Gibson, Roger (eds.), Perspectives on Quine, Blackwell, Oxford, 68-79 Davidson, Donald: 1990b, "The Structure and Content of Truth", Journal of Phibsophy 87, 279-328 Davidson, Donald: 2001, Subjective, Intersubjective, Objective, Clarendon Press, Oxford Dretske, Fred: 1969, Seeing and Knowing, Routledge & Kegan Paul, London Dretske, Fred: 1981, Knowledge and the Flow of Information, Basil Blackwell, Oxford Evans, Gareth: 1982, The Varieties of Reference, Clarendon Press, Oxford Fodor, Jerry: 1983, The Modularity of Mind, M.I.T. Press, Cambridge/MA Fodor, Jerry: 1984, "Observation Reconsidered", Philosophy of Science 51, 23-43 Lewis, C.I.: 1956, Mind and the World Order, Dover Reprint, New York McDowell, John: 1994, Mind and World, Harvard University Press, Cambridge/MA McDowell, John: 1995, "Precis of Mind and World", Philosophical Issues, 7 Perception, 231-239 McDowell, John: 1998, Meaning, Knowledge, and Reality, Harvard University Press, Cambridge/MA McDowell, John: 2001, "Comment on Richard Schantz, "The Given Regained"", Philosophy and Phenomenologtcal Research 62, 181-184 McDowell, John: 2002, "Responses", in: Smith, Nicholas (ed.), 2002, 269-305 Peacocke, Christopher: 1992, A Study of Concepts, MIT Press, Cambridge/MA Peacocke, Christopher: 1998, "Nonconceptual Content Defended", Philosophy and Phenomenological Research 58, 381-388 Pitcher, George: 1971, A Theory of Perception, Princeton University Press, Princeton Putnam, Hilary 1983, Realism and Reason, Philosophical Papers, volume 3, Cambridge University Press, Cambridge, 229-247 Quine, Willard Van: 1953, From a Logical Point of View, Harvard University Press, Cambridge/MA Quine, Willard Van: 1960, Word and Object, MIT Press, Cambridge/MA Quine, Willard Van: 1969, Ontological Relativity and Other Essays, Columbia University Press, New York Quine, Willard Van: 1974, The Roots of Reference, Open Court, La Salle Quine, Willard Van: 1975, "On Empirically Equivalent Systems of the World", Erkenntnis 9, 313-328 Quine, Willard Van: 1981, Theories and Things, Harvard University Press, Cambridge/MA Quine, Willard Van: 1990a, Pursuit of Truth, Harvard University Press, Cambridge/MA Quine, Willard Van: 1990b, "Comment on Davidson", in: Barrett, Robert & Gibson, Roger (eds.), Perspectives on Quine, Blackwell, Oxford, 80 Quine, Willard Van: 1993, "In Praise of Observation Sentences", Journal of Philosophy 90, 107-116 Quine, Willard Van: 1995a, From Stimulus to Science, Harvard University Press, Cambridge/MA Quine, Willard Van: 1995b, "Reactions", in: Leonardi, P. & Santambrogio, M. (eds.), On Quine. New Essays, Cambridge University Press, Cambridge, 347-361 Rorty, Richard: 1979, Philosophy and the Mirror of Nature, Princeton University Press, Princeton, 213-256 Schantz, Richard: 1990, Der sinnliche Gehalt der Wahrnehmung, Philosophia, Munich Schantz, Richard: 1996, Wahrheit, Referenz und Realismus, de Gruyter, Berlin & New York Schantz, Richard: 1998, "Was ist sinnliche Evidenz?" Logos 5, 175-214

Empiricism Externalized

111

Schantz, Richard: 2001, "The Given Regained, Reflections on the Sensuous Content of Experience", Philosophy and Phenomenological Research 61,167-180 Sellars, Wilfrid: 1963, Science, Perception and Reality, Routledge, London Smith, Nicholas (ed.): 2002, Reading McDowell, Routledge, London Stroud, Barry: 1984, The Significance of Philosophical Scepticism, Clarendon Press, Oxford

Circularity and Epistemic Priority ERNEST SOSA

A. Circularity and a

World-Connection

In attaining knowledge that ρ one may depend on some prior knowledge that , by believing

on the epistemic basis of that prior knowledge, which one might have enjoyed without ever reaching the further conclusion that p. One might thus depend for one's knowledge of

on one's knowledge that q, if one concludes that ρ only by reasoning through . One might perhaps have attained knowledge of

in some other way, but, as things in fact stand, one attains it only by concluding it from (aided perhaps by auxiliary premises). If that is how things stand, then it would seem vicious to base one's belief of only on one's belief of

. This would seem circular, viciously so. But why exactly is the circularity here vicious? The full story on vicious circularity is yet to be told, and threatens to be winding and difficult. Here are some initial thoughts. When and why is a subject's intellectual state viciously circular? If one holds that it is sprinkling (pi), that car wipers are running (p2), that circles are forming in the puddles (p3), that there is a pitter-patter on the window panes (p4), that the streets are getting wet (p5), that drops are falling on one's bare arms (p6), etc., then from any five of these one can reason (aided by auxiliary premises) to the sixth as conclusion. There is then circularity in one's concurrent use of all these arguments, as in each case one bases acceptance of the conclusion on belief in the premises. What is vicious about this? In our intellectual lives we want in any case not just the splendid integration displayed by the combined arguments in 1; we want also to get at the truth and avoid falsehood. And we know that this can happen only through appropriate contact with how things stand, which cannot be just a matter of stories fitting together, however richly and tightly. Apart from the value of rational coherence, there is therefore an additional, externalist, value to be satisfied: namely, that our beliefs be somehow reflective, or indicative, however indirectly, of how things actually stand beyond our brains and our skins, and beyond our beliefs themselves. Part of the mechanism connecting a belief with how things stand in the external world may involve reasoning. But it cannot always consist entirely of reasoning. In particular, perception would seem required at some stage. Recall the example of propositions pi—p6 above. No matter how well reasoning from any five may help to make one's belief in the sixth epistemically virtuous, no such argument by itself, nor even the six jointly, will serve adequately to connect any of the six beliefs with relevant facts. It is only by further appeal to what we see, hear, and feel about the drops, the circles in the puddles, the pitter patter, etc., that we may be able to make the required connections. If for each of the six propositions we depend sim-

114

In Defence of Epistemic Externalism

ply on reasoning from the other five, this leaves it open that the whole story be just a nicely coherent fiction. To the extent that a bit of reasoning is circular, then, in providing no support for an essential component in need of support except by invoking that very component essentially, to that extent is such reasoning precluded from connecting its conclusion with how matters stand beyond the subjects skin (for conclusions that pertain to such matters). Reasoning in a circle fails to make such connection unless in some other stretch, through some supplement, it surpasses the circle. Thus, when the combination of the six subarguments for p i — ρ 6 is supplemented with further premises concerning one's various ways of perceiving the circles in the puddles, the drops on ones arms, the pitter patter on the window panes, etc., then, thus aided, the combined subarguments may after all render one's beliefs epistemically virtuous not only in the respect of integration, but also in the equally important respect of connection to the world beyond the subject's skin. Importantly, the further premises that specify one's perceptual access to the relevant facts should be viewed not only as revealing one's beließ that one perceives thus, but also as specifying that one does perceive thus or at least that one then enjoys the sort of perceptual sensory experience that would help constitute such perception. Appropriate connection with the world in that case requires "reasoning" that includes such perceptual uptake, such sensitivity to one's (truth-indicative) perceptual experience through appropriate beliefs. Reasons for one's having those beliefs may thus be found in such experience. In a stretched sense of "reasoning", one's perceptual uptake does then involve reasoning, belief acquisition or retention based on other concurrent (or nearly concurrent) elements in one's psychology: namely, one's (truth-indicative) perceptual experience. The reason why a broad circle can be less frustratingly inadequate than a narrow one is that it is likely to be more richly coherent, providing fuller integration among diverse beliefs. Accordingly, insofar as such richness and tightness of explanatory and other integration is among our intellectual values, broad circles are likely to be better than narrow ones, other things being equal. Suppose one rests with a mere circle, however, whether narrow or broad, reasoning to some component belief Β based only on such reasons as derive from other members of the circle that in turn depend on B. If so, then one will fail to connect appropriately with the world beyond, with how things stand beyond one's skin (when the conclusion pertains to such matters), or at least beyond one's mind. 1

1

By way of clarification, what unsuits a stretch of reasoning for securing knowledge, whether animal or reflective, and especially for securing what is most distinctive of animal knowledge (namely, the external connection) is not just that it circles back on itself, but that it is "unsupplemented," detached from appropriate "external input," whether this takes the form of sensory input or of simple a priori truths rationally intuited. Incidentally, the term "reasoning" should here be stretched to include also ones responses to sensory experience through introspective and perceptual beliefs, and one's responses to simple a priori truths through rational intuition.

Circularity and Epistemic Priority

115

B. Can Knowledge Depend on a Fortunate Context? What Potential Mishaps Need Ruling out with Logical or Epistemic Priority? Reasoning is related epistemically to knowledge in two ways, corresponding to the two sorts of knowledge, the reflective and the unreflective. The latter, animal knowledge, is concerned with the acquisition and sustainment of apt, reliable belief, whereas the former requires the belief to be placed also in a perspective within which it may be seen as apt. Animal knowledge repels circularity for reasons that do not concern reflective knowledge. Beliefs about matters beyond the subjects skin, in particular, must be acquired through mechanisms connecting that belief with how it is out in the world. Essential circularity unsuits a stretch of reasoning for making that connection unaided. Such circularity is therefore vicious as provider of animal knowledge. When reasoning circles back on itself essentially, it fails to connect with the world as it must in order to render its conclusion apt. What is distinctive of reflective knowledge is unbothered by such detachment, however, since reflective knowledge, while building on animal knowledge, goes beyond it precisely in the respect of integrating one's beliefs into a more coherent framework. This it does especially through attaining an epistemic perspective within which the object-level animal beliefs may be seen as reliably based, and thus transmuted into reflective knowledge. A belief that ρ by a given subject in given circumstances is safe if and only if not easily would that subject then hold that belief without being right. The belief is sensitive, by contrast, if and only if had the belief been false, the subject would not have held it. These are logically independent modal statuses that a belief might have, since subjunctive conditionals do not contrapose. Which then, if either, should be required for knowledge, as a sort of "tracking" requirement? This question may now be seen to pertain most directly to the acquisition of apt belief and animal knowledge, to the question, for example, of what is required if one is to see that it is a canary before one, or a zebra. Must one be in a situation wherein one s belief is sensitive, or is it enough that one's belief be safe? The requirement of safety rather than sensitivity has much to be said for it, although it implies that, so long as one's context rules out the possibility that the seen bird be a canary, one might know it to be a goldfinch even if the aspects of its appearance that guide one's belief do not distinguish goldfinches from canaries. Similarly, if one's context rules out the possibility that the seen equine be a striped mule, then, according to the safety theory, one might know it to be a zebra, even if the aspects of its appearance that guide one's belief do not distinguish zebras from striped mules. In each case the creature's appearance enables one knowledgeably to classify it, even though one's visual experience fails to privilege that classification over its rivals. Rival alternatives are ruled out rather by the very context in which one frames ones belief. If the habit of trusting such appearances is apt, correct, and justified, therefore, it enables one to know that the creature is a zebra. What if one asks whether the creature is a striped mule? Can one knowledgeably say no simply through knowing the correct, incompatible, classification? Relevant here, surely, is the epistemic status of the habit of thought, cognitive process, or mechanism that leads to that classifying belief. How should we think of such habits as that of

116

In Defence of Epistemic Externalism

trusting one's ostensible perceptions at face value, and those enabling us to classify the small, yellow bird, for example, or the striped equine? We can usefully think of those habits as implicit beliefs invoked silently (through universal instantiation and modus ponens) when we trust our experiences at face value, or when we classify through our perceptual appearances. Such trusting or classifying beliefs will then constitute knowledge only if they are safe (would be held in the context only if true) and also apt and justified, and only if the underlying habits (implicit beliefs) on which they rest are also apt, justified, and true. Consider then the relevant implicit belief that an animal would (in the context) look a certain way only if it were a zebra. What makes that belief apt and justified? (It is of course by hypothesis a true belief.) This belief must itself have been sustained right up to the present moment in a way that renders it correct not just as a matter of arbitrary choice, or in any other way just accidentally. Given this, one cannot be just lucking into the right belief that it is a zebra rather than a striped mule. There must be some underlying rational or reliable way of acquiring beliefs that yields one's implicit belief, that animals with such an appearance would be zebras. Based on such a belief, with adequate epistemic status, one should then be able knowledgeably to classify. Several are the ways to acquire such implicit general beliefs: through training or schooling, for example, or by noticing correlations then buttressed through deliberate testing; or one can just pick them up through life experience, or through appropriate testimony. Some may even turn out to be innate mechanisms triggered in the way language acquisition is triggered while vastly underdetermined by available evidence. Our experience-trusting habits of thought, where in normal circumstances we take our shape and color appearances at face value, may be of this sort. And here we encounter a limitation of our model: Are such habits really plausibly viewed as beliefs in the corresponding generalized conditionals, implicit beliefs that kick in with appropriate input (through universal instantiation and modus ponens). This seems an artificially intellectualized way to think of our ingrained tendency to take our experiences at face value. It seems more natural to view it rather as just the tendency, among others, to think one sees something white and round when one has a visual experience as if one does so. Noncircular and prior ruling out of certain alternatives is required especially for the epistemic status proper to unreflective ("animal") knowledge. When one sees a goldfinch, where too easily it might have been a canary, or sees a zebra, where too easily it might have been a striped mule—where those incompatible alternatives are hence relevant—one needs to reason via a lemma that the actual situation is not one of those, that the creature is not one of those indistinguishable ones. This lemma must be known, moreover, and known with priority to the conclusion that it supports. The reasoning involved must reach back to perceptual or other sensory input, moreover, if it hopes to provide the needed connection with the relevant facts, a requirement we must impose on any stretch of reasoning that aims to give knowledge of its conclusion. Such reasoning must not circle back on itself essentially; it must rather reach out and touch the world. Reasoning that provides animal knowledge is required to hold up with truth, aptness, and justification at every lemma on which it relies essentially for the sake of connecting its conclusion back eventually with the relevant portions of external reality. This is

Circularity and Epistcmic Priority

117

bound to happen through some sort of perception, aided perhaps by memory. At this level circularity is plausibly prohibited, since circles of reasoning will do little to connect any conclusion of such reasoning with the facts beyond the circle. Reasoning that provides animal knowledge must connect its conclusion with the relevant externalia. If one is to know unreflectively, such reasoning must "rule out" certain alternatives with priority through lemmas essential to that reasoning. What makes certain alternatives relevant in the quest for animal knowledge that ρ is hence that they (or their like) require ruling out knowledgeably and with priority if one's reasoning is to make the connection required for nonaccidental grasp of the fact that p. Thus, in canary-infested surroundings one's belief that it's a goldfinch is too much at risk to constitute animal knowledge just based on the visual appearances. In that case, its being a canary must be ruled out with priority if one's reasoning is to connect those appearances with the fact of its being a goldfinch and not a canary. The same goes for the belief that one sees a zebra, where striped mules abound. The requirement of safety gives the right result here with no need to distinguish relevant from irrelevant alternatives, since it makes that distinction automatically and with no need for a separate clause. In each case (goldfinch, zebra, sphere) one will know only if one's belief is safe. And the safety of one's belief will depend on whether in the circumstances it would be too easy to go wrong in believing as one does, which in turn might depend on how common canaries (or striped mules, or sphere illusions) are in the circumstances. If too common, that makes it too easy to go wrong in believing one sees a goldfinch. That is, however, compatible with one's having no need to rule out any such alternatives with priority when the objective is reflective and not just animal knowledge. Having grasped how it is around us through connections involving the perception/ memory/reasoning required for animal knowledge, further reasoning on that basis may enhance our conscious integration and explanatory coherence, lending epistemic virtue to our beliefs. Again, there is no more vice in this circular procedure than in a case of visually-apparent-sprinkling/circles-in-puddles/pitter-patter-on-the-window-panes/carwipers-wiping/umbrellas-up/felt-drops-on-one's-bare-arms/recalled-forecast, etc. There is nothing wrong with accepting various subarguments in such a case concurrently, believing the conclusion of each partly on the basis of the other beliefs used as premises. It would of course be bad to hold those subarguments concurrently absent any connection with the relevant externalia. Even when one holds the lot of them concurrently interlocked, however, this does not imply that one holds them so detached. (And a special, rational form of viciousness spoils attempts to reach conclusions about the contingent world around us through reasoning detached not only from the world beyond but even from those states and beliefs required as peripheral intermediaries for the desired connection with the world. Prominent among these states are the experiential states whose job it is precisely to mediate in that way between our contingent beliefs about the world around us and the world that they are about. Much circular reasoning would be defective in just that way.) When in that rain example one believes the ultimate conclusion that conjoins the six propositions p i — p 6 , there is a brief argument for each conjunct based on what one sees or hears or feels. Each such argument proceeds directly, moreover, from what one perceives or seems to perceive in the way pertinent to that particular conjunct:

118

In Defence of Epistemic Externalism

one from sight, one from hearing, etc. But the circularity-avoiding combination of these arguments does not exhaust the rational support available for that six-term conjunction. For the six members can also attain explanatory integration, aided perhaps by auxiliary premises, even if the arguments displaying such integration cannot avoid circularity, inasmuch as one will be arguing from some members of that conjunction to some other member as conclusion, and then eventually to the six-term conjunction as ultimate conclusion. True, each of the six arguments about the rain and the puddles adds some further connection to the world beyond, whereas an epistemic perspective that coheres with one's first-level "animal" beliefs need not secure any further connections with externalia not already secured at the lower level, through the mechanisms of belief acquisition that operate on the relevant subject matter to yield the relevant first-level beliefs. So the coherence induced by the epistemic perspective does not have this benefit, of securing further connections with the relevant externalia or other subject matter. But does this mean that it can confer no other benefit? Is there no difference between having and lacking an "affirming" perspective on oneself, on the surrounding world, and on the relations between the two, a perspective on the basis of which one can coherently endorse one's own beliefs? Is such a coherent picture of things of no value simply because it cannot possibly be elaborated while avoiding all circularity? This should seem the opposite of obvious to a philosopher, since it discounts a main perennial objective of philosophical reflection: namely, the elaboration of such a coherent, "affirming" view of oneself and one's place in the scheme of things. What is more, the member beliefs of such a system are enhanced precisely through being part of the integrated view and not just loose from other beliefs in a less integrated mind. So these member beliefs attain a higher epistemic grade, a higher grade of knowledge, reflective knowledge, than they would attain otherwise, other things being equal

C. Deep Externalism and Epistemic Justification 1 Intuitively it seems to many that epistemic justification is an internal matter that ought not depend on your contingent emplacement. This is redolent of the Kantian view that moral quality derives from how it is in the subject's mind and not from contingent external features that determine the causal consequences of his policies and decisions, and thus also how well these turn out in the actual world. There is in epistemology a kind of mirror image of such Kantianism: epistemic worth should derive from the qualities of the subject's mind and not from contingent external features that will determine (a) the causal origins of the subject's beliefs and inference patterns, and (b) how reliably truth-conducive such causal determination of beliefs and inference patterns turns out to be. Suppose epistemic justification should indeed depend only on such internal matters independent of the (autonomous) subject's contingent emplacement, so as to ensure that the subject's relevant behavior or conduct

Circularity and Epistemic Priority

119

is truly his own, truly attributable to him and not, for example, to the "luck" of the circumstances. How, more positively and fully, should we then conceive of such justification? Some have tried to understand it as a matter of being blameless, of avoiding any wrongful violation of epistemic norms required for right belief formation. But this way of understanding epistemic justification takes us only so far, and comes up short: if you are brainwashed or brought up epistemically in a superstitious community, you may bear no blame for beliefs and inference patterns that still fall short epistemically. You are not then really "wrongfully violating" any norms, since how you proceed is in a way not really "up to you". It derives rather from aspects of your psychology before which you are then helpless. Internal epistemic justification hence seems inexplicable just in terms of epistemic blamelessness. How then is it to be understood? Before taking up this question, we face an objection as follows. Suppose we insist on a hardline reliabilist analysis of epistemic justification (Jl), according to which, S s belief that ρ is epistemically justified in world w iff S s belief that ρ is the product of a belief-forming process that is reliable in w. As we have seen, J l then faces the New Evil Demon problem. Here now is the hardline response: "Yes, the inhabitants of the demon world are unjustified in most of their beliefs." If you think this response is outrageous, consider the following. Some pre-theoretic epistemic intuitions are of course obdurate, or at least much harder to push aside: thus, perhaps, the intuition that it is possible for one to know that one sees a (real, flesh-and-blood) hand. How obdurate is the intuition that the victims of a demon world are justified in their beliefs? Is it clear that any intuition about how matters stand epistemically in such a world will be strong enough to make one want to abandon reliabilism and join the opposition? A hardline reliabilist might also add that, while unjustified, demon-world victims are blameless for believing as they do. Does the intuition that the demon-world victims are justified in their beliefs remain even once we distinguish this deontological conception of justification from the conception of epistemic justification as the remainder when you subtract true belief from knowledge (Gettier factors perhaps aside)? In response it must first be granted that the hardline response does hold some attraction, especially if we conceive of the evil demon world as one in which our usual processes would lead us astray sufficiently often. Thus, we are to suppose a world where in the normal situation one would go astray in taking one's experience at face value. In this bad sort of demon world, you plausibly lack epistemic justification even when using mechanisms of belief formation that are adequately reliable in our actual world, such as taking experience at face value. This notion of epistemic justification is presumably the one invoked when we countenance as possible some highly evolved beings who do gain knowledge in their possible world, by properly and reliably forming ("apt" or "warranted") beliefs, despite the fact that their processes, mechanisms, etc., of belief formation, while successful in their world, would be miserably inadequate in ours. The demon's victim would still be justified, surely, on some notion of epistemic justification that is still intuitively gripping (however it may relate to knowledge, and however it may relate to whatever else is necessarily contained in knowledge. This is the notion that the classical foundationalist explicates in terms of (i) a foundation where one takes the given and (ii) a superstructure that one builds on that foundation through

120

In Defence of Epistemic Externalism

valid reasoning. According to internalists, traditional epistemology is focused on such rational justification, and our Gettier-related study of "knowledge" and reliability spoils that focus. Traditionally the project has been to answer the skeptic, largely by showing how, despite what the skeptic has to say, we still remain rationally "justified" in certain of our important beliefs about the world around us, our friends and their thoughts, etc. It is this kind of justification that, according to internalists, you might still have even if your animal mechanisms were unreliable in certain unlucky circumstances. That being so, the project for the reliabilist is to show how, within a broad reliabilism, within truth-connected epistemology, we can understand such a notion of justification. Take notions like "not being subject to blame", or take the deontological conception(s). Such notions have been used prominently in attempts to show that through them reliabilists (broadly enough conceived) can define a notion of justification in accordance with internalist intuitions. But these concepts are really too weak for what the internalist has in mind. What the internalist cares about is, at bottom, the idea of being properly responsive to one's experience and reasoning properly therefrom, the idea of adjusting ones believing properly to one's "evidence". But in this regard one can fail even if one is not subject to blame, even if one has violated no epistemic duties (since one had no choice, having been brainwashed or the like). Internal epistemic justification is more positively a matter of taking proper account of one's data and reasoning properly on their basis. So it is a matter of intellectually flawless or at least relevantly virtuous procedure. One is justified when one's experiences and rational intuitions are properly reflected in one's beliefs, and when one's further beliefs (those that do not so directly reflect one's experiences and intuitions) are arrived at through good enough reasoning. This is why it seems that the victim of the demon might be no less justified than we are. After all, the victim could be taking quite proper account of his experiences and intuitions and could be reasoning beautifully from these. How well his experiences in turn reflect the state of his contingent surroundings seems irrelevant to the evaluation of his intramental proceedings. 'Epistemic justification' is perhaps a context-dependent expression, in the sort of way indexicals are context-dependent. The following is thus more than just accidentally true: X is here i f f X is where I am (is co-located with me). Yet, with respect to a hypothetical situation, we would not determine whether an object would be here by determining whether it would in that situation be (there) co-located with me. Consider now this principle (J2): A belief Β is justified-a in world W = Df in world W, Β is formed through a process that is in fact virtuous enough in our actual world, i.e. one whose employment in our actual world would yield a high enough preponderance of truth over falsity. According to J2, even if one were a demon victim one would still be justified (one would be justified-a) in taking one's experiences at face value. That is to say, in a world W in which one was a demon victim, one's beliefs acquired through our normal perceptual faculties would come out of processes that in our actual world are reliable, and hence those beliefs of ours in that world would be justified-a.

Circularity and Epistemic Priority

121

2 The skeptic still has a response: Are we still justified even if the actual world is itself a demon world? Consider this conditional: (D) If the actual world is a demon world, then our beliefs acquired through our normal perceptual processes are justified. Is this a true conditional? Well, either it is an indicative, material conditional or it is a disguised subjunctive (or in any case modally strong) conditional. If it is an indicative, material conditional, then it is true vacuously, since its antecedent is false. And if it is rather a subjunctive conditional, then we are back to our earlier reasoning. Conditional D thus seems true regardless of interpretation. (Of course, also vacuously true is the opposing material conditional that shares D s antecedent but denies its consequent.) Are we here begging the question against the skeptic? Aren't we just assuming that the actual world is not a demon world? Perhaps we are, but the most formidable skeptic does not want to take it as a premise that the actual world is a demon world. Rather, he takes it as a premise that the actual world might be a demon world. And his premise is not that for all we know the actual world is a demon world. No, this is rather his conclusion, or close to that. His premise is only that metaphysically (or at least logically) the actual world might be a demon world. And we can of course agree with that premise, while rejecting the skeptical conclusion nonetheless. A less formidable skeptic is answered more easily, if he needs to be answered at all. According to this lesser skeptic we might metaphysically be in a demon world, and cannot just assume or presuppose that we are not; after all, it is part of his skepticism not just to argue that we do not know what we take ourselves to know, but also actually to put in question all of our supposed knowledge of the contingent, external world around us. Mark well, this means not just to raise the question as to whether we are wrong in the likes of the following single hypothesis: that most of our beließ are true. No, also put in question, along with that hypothesis, is each of our contingent beliefs, particular or general, about the external world. So it would not do for the dogmatist to just take for granted certain contingent conditionals. The skeptic is precisely not granting any of that. On the contrary, he is putting all such conditionals in question, along with every other contingent truth about the external world. So it would beg the question against him to take for granted an answer to that which is in question. True enough, if we allow the skeptic to put all of that in question, in one fell swoop, then there is no defeating him without begging the question. Well, perhaps there is, actually, since we can press against the skeptic a charge of arbitrariness.2 Why is he putting in question all propositions about the external world, and not those about the realm of the necessary or about his own states of mind? After all, it is not as though we are infallible about everything in these realms. It is not as though here we could never make a mistake. And even if we take ourselves to be infallible and incorrigible about certain of our beliefs at least in these realms, how would we know which of them are

2

A famous Reidian charge, to which Descartes also seems quite alive (ironically, as Reid's charge is m a d e as an objection to Descartes), as suggested by the true depth o f the skepticism that he takes up, which is in fact Pyrrhonian and not just Cartesian.

122

In Defence of Epistemic Externalism

truly so protected against error? Would we not need to employ, in order to demonstrate our prowess, the very faculties that a skeptic does or might put in question? H o w then could we defeat such a skeptic without begging the question? If we are at least right about these restricted areas in which we are assured of success if careful enough, it might be replied, we can then see how we enjoy, with regard to these restricted domains, something truly special. We are there exempt from dependence on the luck o f our emplacement, on the luck o f a clear channel, of a favorable m e d i u m . With regard to these special domains, there is no such dependence on channel or m e d i u m . So our control is at a m a x i m u m , our need for luck at a m i n i m u m . N o wonder the knowledge that we enjoy in such cases has been so exalted in our tradition, and has even been raised to the level of the only true knowledge. N o wonder so many have been tempted to be "skeptical" about any other knowledge. All other knowledge would then reasonably fall below the exalted level, since all other knowledge would be dependent on matters beyond our true control, and subject to the luck of channel or m e d i u m , or other external circumstances. Even if we go along quite far with such reflections, however, should we not equally wonder about the luck of proper teaching, o f proper upbringing, of proper constitution. A n d if these matters are to a substantial extent also beyond our ability to control, if how smart we are, how persistent, how patiently and deeply reflective, if these are all, to an impressive degree, matters of constitutional luck, as with raw I Q , then how well we do even in such matters will depend, for a clear example, on whether one is R a m a n u j a n or a mental defective. A n d that would seem largely a matter of luck, something outside one's control. S o why exactly is the luck o f our emplacement so detrimental to true knowledge, whereas the luck o f our constitution is o f no concern? W h a t if it might more easily turn out that our constitution be altered in a knowledge-affecting way, than that our emplacement be altered with such a consequence? WTiat if the quality o f our relevant external channels a n d media are quite stable modally, whereas our relevant respects of constitution are somewhat stable modally, but much less so? Are beliefs dependent on the stable channels still to be assessed as lesser knowledge by comparison with beliefs dependent on our less stable relevant constitution?

3 In conclusion, I claim that: (a) our broad coherence is necessary for the kind o f reflective knowledge traditionally desired; and (b) such broadly coherent knowledge is desirable because of the integration it imports and also because in our actual world it helps us approach the truth and avoid error. T h i s is not to deny that there is a kind of "knowledge", properly so called, that falls short in respect o f broad coherence—"animal knowledge", as we might call it. It is rather only to affirm that beyond "animal knowledge" we humans, especially the philosophical or at least reflective, aspire to a higher knowledge. T h i s higher knowledge we might call "reflective", since by definition it requires s o m e appropriate degree of broad coherence including o n e s ability to place one's first-level knowledge in epistemic perspective.

II Critiques of Epistemic Externalism

The Chimerical Appeal of Epistemic Externalism J O E C R U Z AND J O H N POLLOCK

1. Introduction Internalism in epistemology is the view that all the factors relevant to the justification of a belief are importantly internal to the believer, while externalism is the view that at least some of those factors are external. This extremely modest first approximation cries out for refinement (which we undertake below), but is enough to orient us in the right direction, namely that the debate between internalism and externalism is bound up with the controversy over the correct account of the distinction between justified beliefs and unjustified beliefs.1 Understanding that distinction has occasionally been obscured by attention to the analysis of knowledge and to the Gettier problem, but our view is that these problems, while interesting, should not completely seduce philosophers away from central questions about epistemic justification. A plausible starting point in the discussion of justification is that the distinction between justified beliefs and unjustified beliefs is not the same as the distinction between true beliefs and false beliefs. This follows from the mundane observation that it is possible to rationally believe claims that are false. Still, most epistemologists have claimed that truth and falsity play a crucial role in distinguishing between justified and unjustified belief because they maintain that believing truths is the ultimate objective of human rational cognition. Much of the explanation for the continued attraction of externalist theories of epistemic justification derives from a specific perspective on the conviction that truth is the sole or primary evaluative standard against which rationality is judged. By this view, externalism has a strong prima facie attraction, as properties of epistemic agents that are truth-aimed—like the actual reliability of their cognitive processes—are (in at least an intuitive sense) external to the cognitive agent. Thus, externalism appears to hold out the most promise of properly respecting the connection between belief formation and truth as the evaluative standard of rationality. The appeal of externalism, we maintain, is illusory. Our strategy is to consider the motivations behind externalism and to show that each is misguided or can be accommodated in an acceptable internalist theory. This, combined with the difficulties faced by externalist theories, seems to us to constitute a decisive case in favor of internalism. Contemporary externalist accounts have germinated in light of three sources that constitute refinements of the generic truth motivation, above, even if these sources

1

In this paper we will treat 'justified' and its cognates as s y n o n y m o u s with 'rational' and its cognates. Thus, we ignore the fact that they are sometimes distinguished in the epistemological literature in light of what some take to be the overly deontological connotation o f justification and the greater theoretical neutrality o f rationality. See, e.g., Plantinga (1993a).

126

Critiques of Epistemic Externalism

have not always been articulated explicitly or particularly clearly. Firstly, since truth has been viewed as the ultimate epistemic aim, externalists have attempted to answer to that aim directly by making it a non-negotiable constraint on a theory of justification that whatever justification is will co-vary in some way with truth. Secondly, analyses of justification have often emerged in close affiliation with analyses of knowledge. The connection to knowledge recommends a true-belief oriented approach to justification, and thus externalism has been tempting because it appears uniquely able to build in this orientation. This second source of attraction to externalism is closely related to the first, but is worth considering separately because—in our view—it suffers from an additional burden in that knowledge is a poorly behaved concept. Finally, it has appeared to many epistemologists that the only possibility for a naturalistically acceptable account of rationality is some version of externalism. We reject all three of these avenues to externalism. Against the first, we maintain that the ways externalists attempt to incorporate truth as the evaluative standard of rationality either conflates rational beliefs with true beliefs or avoids the conflation in a wholly ad hoc way. Neither of these is acceptable for a theory of rational belief. Our concern with the second motivation is that there is no stable concept of knowledge that constitutes the target of epistemic inquiry in the way that the externalist envisions. If this is right, strong scruples with respect to forging a connection between justification and knowledge are misguided. In response to the last motivation, we sketch an internalist theory that is recognizably naturalistic, and thereby show that externalism is not uniquely amenable to naturalism. We turn now to laying out the relevant terrain in order to clarify just what the externalist proposes.

2. Externalism

contra

Internalism

We can begin to understand epistemic externalism by trying to make sense of what it denies, namely, epistemic internalism. One species of internalism claims that reflective, careful agents are able to make assessments of their reasons for a belief in order to determine whether that belief is justified. The view is that we can determine, for any given belief, whether it is sustained by reasons that are good enough to make it justified. Chisholm offers a representative statement of this sort of internalism: ...Epistemologists p r e s u p p o s e that they are rational beings. T h i s means, in part, that they have certain properties which are such that, if they ask themselves, with respect to any o n e o f these properties, whether or not they have that property, then it will be evident to them that they have it. It m e a n s further that they are able to k n o w what they are thinking a n d believe a n d that they can recognize inconsistencies ( 1 9 8 9 , p. 5).

This and related Chisholmian formulations are notoriously vague, as they do not distinguish between the claims that epistemic agents have reflective access to the content of their beliefs, to the reasons for their beliefs, to the reasonableness of the connection between their reasons and their beliefs, all of these together, or yet some other quantity. T h e one feature of Chisholm's internalism that is clear is that it proposes a fairly dramatic ability on the part of the epistemic agent to find out through reflection what

T h e Chimerical Appeal of Epistemic Externalism

127

the epistemic credentials of her belief are. The sense of 'internal' operating here, then, is that the crucial elements of justification are such that an epistemic agent can locate them by reflection alone. This can only be true of elements that are inside the agent, and, more specifically, within her reflective purview. Many epistemologists have followed Chisholm's lead in understanding internalism as committed to a strong and encompassing version of internal reflective access to properties that make a belief rational 2 , but this is somewhat misleading. There are different kinds of access to reasons and beliefs that the internalist may propose, and not all kinds demand the thoroughgoing introspective insight that Chisholm does. Insight that is more limited in scope seems to also allow for an internalist view. For instance, one of us has defended a view where epistemic agents are able to determine through introspection whether they are conforming to their rational norms in individual cases, even if the nature of the rational connection between their internal states or the content of the relevant states is not always transparent (Pollock 1974; 1979; 1986; for the latest statement of this view, see Pollock and Cruz 1999). This is nonetheless an internalist theory in the sense that epistemic agents have introspective access to whether their cognitions are rational. A view like this is also internalist in a weaker but—in our view—ultimately preferable sense. We maintain that the fundamental idea behind internalism is that the justifiability of a belief is exclusively a function of internal states, where internal means simply inside an agent's cognitive apparatus. What that means is that we can vary everything about a situation other than an epistemic agent's internal states without affecting which of her beliefs are justifiable. In particular, varying contingent properties such as the reliability of the cognitive processes that produce or sustain a belief will not affect whether that belief is justified. We have called this cognitive essentialism (Pollock 1986; Pollock and Cruz 1999). According to cognitive essentialism, the epistemic correctness of a cognitive process is an essential and inherent feature of that process. It is not necessary to reference some external quantity such as reliability in order to certify the correctness of the cognitive process. We aim to make this plausible later in this essay, but at this point it should be clear that cognitive essentialism does not implicate or emphasize access. We think that this is a virtue, in part because we think that the genuine details of cognitive access are to be revealed by cognitive psychology, not by the a priori fiat of a theory of rationality. Be that as it may, we will continue to talk about internalism in terms of access issues. This is useful because it accords so well with the way most internalists themselves conceive of internalism. It cannot be denied that 20th century epistemology has been primarily internalist (Plantinga 1990). Internalist considerations have been accorded a special place in epistemology because of several related concerns: First, one of the projects that sometimes rides piggyback on an assessment of rationality has been to illuminate how one might improve the quality of one's beliefs. If improvement is to be possible, however, it needs to be possible to determine which belief among many candidate beliefs is most justified. Internalists have been driven, then, by what we might think of as rational solip-

2

See (Moser 1 9 8 5 ) or (Foley 1 9 8 7 ; 1 9 9 3 ) .

128

Critiques of Epistemic Externalism

sism since, in their view, we are alone in the task of refining our beliefs and the only resources we have to work with are the ones that can be brought to bear internally. Second, epistemic concepts have been viewed as deontological in character. That is, it has seemed to many epistemologists that intellectual achievement is at least partly a matter of duty-fulfillment. Fulfilling a duty, however, seems to require that one be able to do the things that duty requires and to determine what one's duties are. In order to secure the means to an intellectual duty, an epistemic agent will need to be able to reflect on her condition and on the resources she has available. Thus, a reflective requirement has been built into epistemic theorizing. Third, intellectual achievements have had some historical association with defeating the skeptic, and it has sometimes been suggested that the ability to answer the skeptic is a crucial component of epistemology. The only answers that the epistemic agent can give in reply to a skeptic's claims, though, are the ones that are available to her reflective capacities. In principal, externalism seems to gain a foothold when at least some of these motivations for internalism are denied. That they can be denied should not be a complete surprise. Introspective access to the properties of a belief that make it rational or even a more weakly internalist cognitive essentialism do not seem demanded by the goals of epistemology as such. There is the ubiquitous goal of epistemology where the epistemologist takes it upon herself to say what justification or rationality is. There is a second current where the epistemologist maintains that the believer must be able, in general, to determine whether or not her beliefs are rational. In principle, an epistemologist may pursue the first kind of theory without pursuing the second. It is open to the epistemologist to reject the claim that the right philosophical account of belief quality must enable actual regulative, duty-oriented, or skeptic-answering evaluations to take place. This is just the opening that the externalist needs. What makes things unfortunately murky here is that both the deontological heritage of epistemology and the need to answer the skeptic have been denied by some internalist epistemologists. It is our view that these sources of internalism are best thought of as closer to optional than the first, regulative source. It is the rejection of the regulative character of justification that provide a more powerful point from which to reject internalism, and the development of the literature on externalism bears this out. For example, Goldman (1980) specifically repudiates the regulative character of epistemology, where the regulation must take place in a first-person way. In his positive view, he claims—to a first approximation—that what makes a belief justified is that it is produced by a psychological process that usually produces true belief (Goldman 1979; Goldman 1986). The process must in some sense instantiate a set of reasons, but they need not be reasons that are accessible to the rational agent. Plantinga (1993a; 1993b) is also associated with rejecting the first personal regulative conception of rationality. His theory of proper functions claims that a belief is justified in case it is probably true and is the product of a properly functioning cognitive process working in an environment for which it is appropriate (Plantinga 1993b, pp. 4-20).

T h e Chimerical Appeal o f Epistemic Externalism

129

3. Externalism as a Proposal about Justified Belief Formation Understanding externalism as the denial of the regulative strand of cognitive access is a plausible way of generating a taxonomy of theories of justification, but it should be emphasized that this is only a heuristic device. The actual motivations behind externalist theories are not, of course, principally or solely concerned with rejecting internalism. At the beginning of this essay we identified the putative truth-aimedness of justification as a source of motivation for externalism. Since truth is thought to be the ultimate normative aim of belief formation, the externalist has found it inviting to simply build into her theory some covariation between justified belief and true belief. This way of putting the externalist's ambition highlights an important distinction that must be made when considering externalism. Externalism can be understood as providing answers to two quite different questions. On the one hand, the externalist might describe the correct epistemic norms—where norms dictate how to come to a belief 3 —with an eye toward emphasizing some external feature of those norms. An example of such a proposed norm might be "It is permissible to hold a belief if it is generated by a reliable cognitive process." On this view, the norms for belief themselves appeal to reliability. On the other hand, the externalist might aim solely to tell us what makes epistemic norms correct and appeal to external properties only with respect to that meta-justificational question. So, it might be urged that from the set of possible epistemic norms a cognizer can have, the norms that are aimed at the truth are the correct ones. We have labeled these two kinds of externalism belief externalism and norm externalism (for further discussion, see Pollock and Cruz 1999, pp. 130-42). What would make someone adopt belief externalism? In "What is Justified Belief?" Goldman writes: C o n s i d e r s o m e f a u l t y p r o c e s s o f b e l i e f - f o r m a t i o n , i.e., p r o c e s s e s w h o s e b e l i e f - o u t p u t s w o u l d b e c l a s s e d as u n j u s t i f i e d . H e r e a r e s o m e e x a m p l e s : c o n f u s e d r e a s o n i n g , w i s h f u l t h i n k i n g , reliance o n e m o t i o n a l attachment, mere h u n c h or guesswork, a n d hasty generalization. W h a t d o t h e s e f a u l t y p r o c e s s e s h a v e in c o m m o n ? T h e y s h a r e t h e f e a t u r e o f u n r e l i a b i l i t y : T h e y t e n d to p r o d u c e error a large p r o p o r t i o n o f the time. B y contrast, w h i c h species o f belieff o r m i n g (or belief-sustaining) processes are intuitively j u s t i f i c a t i o n - c o n f e r r i n g ? T h e y i n c l u d e standard perceptual processes, r e m e m b e r i n g , g o o d reasoning, a n d introspection. W h a t these

3

In a reasonably non-committal sense we take epistemic norms to be recipes that capture a peculiarly epistemic propriety between cognitive states. S o m e epistemologists, for instance, have proposed that there is an epistemic norm regarding the g o o d inferential relationship between perceptual states a n d beliefs about those perceptual states. O n e way of putting this norm is, " I f a person has a clear sensory impression that χ is F (or o f x's being F) and on that basis believes that χ is F, then this belief is prima facie justified" (Audi 2 0 0 1 , p. 4 3 ) . Another way is: " I f S believes that he perceives something to have a certain property F, then the proposition that he does perceive something to be F, as well as the proposition that there is something that is F, is one that is reasonable for S " ( C h i s h o l m 1966, p. 4 7 ) . A third is: " H a v i n g a percept at time t with the content Ρ is a defeasible reason for the cognizer to believe P-at-t" (Pollock and C r u z 1999, p. 2 0 1 ) . Other formulations of this norm vary (cf., B o n j o u r 2 0 0 0 , p. 3 0 ; Field 2 0 0 0 , p. 134; Plantinga 1993, p. 9 9 ; Sosa 1980, p. 2 2 ) , but they all seem to be after the same thing, namely the claim that sensing that ρ is a fallible but g o o d basis for an inference to the belief that p. For a longer treatment o f epistemic norms, see (Pollock and C r u z 1 9 9 9 , chapter 7) and (Cruz, unpublished manuscript).

Critiques of Epistemic Externalism

130

processes s e e m to have in c o m m o n is reliability: the beliefs they p r o d u c e are generally reliable ( 1 9 7 9 , pp. 9 - 1 0 ) .

Goldman is claiming that rational beliefs have something in common and that that common thing, reliability, is constitutive of justified belief. Goldman thus proposes an early version of his process reliabilist theory. The basic idea behind process reliabilism is that a belief is justified if and only if a reliable cognitive process produces it. For example, the reliabilist explains why perceptual beliefs are justified by pointing to the fact that, in the actual world, perception is a reliable cognitive process. Similarly, deduction is a reliable cognitive process, so beliefs deduced from other justified beliefs will be justified. O n the other hand, wishful thinking is not a reliable cognitive process, so beliefs produced by wishful thinking are not justified. The reliability of a cognitive process is the indefinite probability of beliefs produced by it being true. It makes the justifiedness of a belief depend not only on the processes that produced it, but also on whether those processes happen to be reliable in the actual world. Reliability is neither something that the epistemic agent will have access to nor is it cognitively essential in our sense (i.e., justification may vary even if properties internal to an agent are kept constant). This proposal is belief externalist because Goldman is claiming that epistemic norms must themselves reference reliability. He signals as much when he writes, early in the piece, that he assumes that "a justified belief gets its status of being justified from some processes or properties that make it justified" (p. 2). This seems to be saying that there is some property occurring at the level of the epistemic norms that confers justification on beliefs. Goldman is driven to belief externalism by his sense that he has identified reliability as the unifying property of epistemic norms. 4 Belief externalism of this sort has been thoroughly criticized in the literature. The simplest objection is that, intuitively, it is not the extrinsic properties of the processes that generate a belief that is responsible for its justifiedness. This intuition has been the source of repeated attempts by internalists to generate cases that show that, for instance, reliability is neither necessary nor sufficient for justification. The best known of these is Bonjour s case of Norman who has reliable clairvoyance and who believes that he has clairvoyance, although he lacks any justification for that belief (1980, p. 62; 1985, p. 41). Suppose Norman comes to believe that the President is in New York City on the basis of his clairvoyance. Bonjour claims that our intuitive response is that, in the absence of any justification for his belief in his own clairvoyance, Norman is irrational to accept his belief that the President is in New York. Since reliabilism claims that Norman is justified in that belief, it gives the intuitively wrong results In response, reliabilists either deny the intuition or generate ancillary principles to address it. For instance, Goldman claims that the Norman case cannot be coherently described in such a way that it does not violate a plausible no-undermining—or, as we prefer to put it, a no defeater—clause of reliabilism. This no defeater clause maintains that a belief is justified if it is the product of a reliable cognitive process and if there

4

Indeed, the simplicity and unity of the underlying justification-conferring property is a feature of externalist theories that externalists have sometimes been keen to emphasize, since it seems to contrast so starkly with the theoretically heterogeneous collection of norms proposed by internalists.

The Chimerical Appeal of Epistemic Externalism

131

is no more reliable cognitive process that, if available and employed, would defeat the belief. Goldman says that Norman ought to reason in such a way as to verify his clairvoyant powers. If he does not, he is failing to use a process available to him that would render his belief unjustified (1986, p. 112). We do not need to rely solely on intuitions about cases in order to criticize reliabilism. Belief externalism has been criticized in light of a very fundamental difficulty it faces. There seems to be no non-ad hoc way to circumscribe a cognitive process in order to assess its truth-aimedness. In the specific case of reliability, this has come to be known as the generality problem (Pollock 1984; Feldman 1986; Conee and Feldman 1998). 5 There are really two problems that fall under the generality heading. The first has to do with individuating process types. There seems to be no non ad hoc way to isolate a cognitive process in order to assess its reliability. Suppose someone forms a belief about an object by using her normal vision in good light. Which process should be assessed for its reliability? Should we individuate the process as vision right at this moment, vision over the last ten years, the whole vision/cognitive system, or by some other individuating scheme? The difficulty with specifying a principled answer to this challenge has dogged this brand of reliabilism nearly from the start. The other problem has to do with the environment. Which elements of the environment are relevant to assessing reliability? It might first be proposed that in judging the reliability of a cognitive process in a particular instance we should take account of everything about the circumstances in which it is used. If it makes any sense at all to talk about the reliability of the cognitive process "under the present circumstances" (in all their specificity), it seems that it must be the (indefinite) probability of producing a true belief, conditional on everything true of the present circumstances. But the present circumstances are infinitely specific and include, among other things, the truth-value of the belief being produced by the cognitive process and the fact that that is the belief being produced. Consequently, this indefinite probability must go the same way as objective definite probabilities and be either 1 or 0 depending upon whether the belief in question is true or false. Thus this reliabilist criterion entails the consequence that in order for a belief to be justified it must be true, and this is a denial of the starting point of a theory of rationality. Goldman has tried to evade these challenges by claiming that the processes and environmental conditions are individuated intuitively (1979, p. 12). He claims that, if there is vagueness involved this only reflects the vagueness of our concept of justification. This response, however, is not particularly persuasive (cf. Alston 1995). We think that the failure of this prominent kind of externalism is instructive, but we prefer not to build our case against externalism around the missteps of a particular view. Notice, though, that there is a much more general point to be made here about belief externalism. It appears that the strategy of attempting to build a truth-aimed component into norms of belief formation will always founder on a tension between belief formation processes that fall short of truth and ones that if employed, would guarantee

5

The label for the problem comes from Goldman's own recognition of it in his (1979).

132

Critiques o f Epistemic Externalism

truth. T h e lesson appears to be that if truth-aimedness is treated as the critical value that an epistemic norm must have, then there is no principled way to resist the slide to epistemic norms that yield only true beliefs. T h e alternative strategy is to attempt to articulate epistemic norms that can fall short of truth, but the degree of reliability demanded will always seem ad hoc. 6 This is just the generality problem writ large for belief externalist proposals that respond to the truth-aim motivation by trying to build the external component into the belief formation processes themselves.

4. Externalism about the Correctness of Epistemic Norms We take the above considerations as showing that belief externalism is ill-advised, and we take it as some confirmation of this that Goldman, for one, has since admitted that he can and does accept internalist rules for belief formation (1994, p. 304). But this is to say nothing about another sense of externalism, norm externalism. Norm externalism acknowledges that we must employ wholly internalist norms in our reasoning, but proposes that alternative sets of internalist norms should be evaluated in terms of external considerations. The premiere external consideration is, again, truth-aimedness. T h e truth motivation appears again in a slightly different form and at a higher level. For the norm externalist, it is the entire framework of belief formation processes that must answer to truth as the fundamental normative aim of epistemology. Why would someone be a norm externalist? As a first pass, norm externalism is attractive because it seems as if norms should be criticizable on some grounds apart from the norms themselves. Apparently, the internalist must assume that our actual ultimate epistemic norms are immune from criticism because whatever norms she expects to employ in her self-criticism will be norms that are internal. This is surely odd, and the norm externalist's thought is that there must be a higher standard against which the norms are evaluated. Consider two people who employ different epistemic norms. They could then hold the same belief under the same circumstances and on the basis of the same evidence and yet the first could be conforming to her norms and the second not conforming to hers. If a person's epistemic norms were always beyond criticism, it would follow that the first person is justified in her beliefs and the second is not, despite the fact that their identical beliefs are based upon the same evidence. Because this case is possible, it seems as if we ought to be able to determine which of those norms are correct. This makes a strong prima facie case for norm externalism. These are just the sorts of motivations that lie behind current versions of norm externalism. For example, in "An Internalist Externalism," William Alston takes the simple course of proposing that a belief is justified if it enjoys the sorts of access properties that

6

Another way to put this point would be in terms of defeat. "When the belief externalist attempts to build in a no defeater condition, she is moved toward perfect reliability. T h i s conflates rationality and truth. There is pressure, however, to include a no defeater condition because any epistemic norm will be trutha i m e d s o m e o f the time and will fail to be truth-aimed at other times, depending on how circumstances are described. N o defeater conditions are introduced to handle this possibility.

T h e Chimerical Appeal of Epistemic Externalism

133

are typical of an internalist view and is also probably true. Both features are necessary for a belief to be justified, and this goes to show that Alston is concerned to maintain an external evaluative standard for beliefs, i.e., truth. 7 The drive to norm externalism is operative in Goldman's later work on reliabilism as well. In Epistemology and Cognition, Goldman first defends a rule framework for epistemology, where the conditions for having a justified belief can be captured as a set of justificatory rules (J-rules). He acknowledges that the content of the J-rules might turn out to appeal only to internal considerations, while the criterion of Tightness for the J-rules might be objective (p. 72). This is Goldman's way of making the distinction between belief externalism and norm externalism. At the level of questions of the correctness of the norms, Goldman favors a truth-aimed, or verific, theory. He gets there by defending a consequentialist rather than deontological account of the criterion of correctness. The argument is that no purely deontological account satisfies the aims of epistemology. Once Goldman has concluded that a consequentialist view of the criterion of correctness is necessary, he assesses various possible consequentialist conceptions including non-verific consequentialist accounts. According to Goldman, the failure of non-verific consequentialist theories is that they would allow beliefs to be justified even if they were not true. This is the point of what he calls 'fantasizer counterexamples.' These are alleged counterexamples to theories of justification where the epistemic agent is guilty of fantasizing to such a degree that wildly false beliefs turn out justified because they are sanctioned by his norms. He writes, "...if these fanciful constructions were consistently false, I would feel little inclination to say that this sort of process confers justifiedness. Certainly it would not yield the kind of justifiedness that brings us close to 'knowledge'" (p. 101). T h e external property of truth serves as insurance against this kind of massive fantasy We think that Goldman is fundamentally confused on this point. We insist that a theory of justification needs to respect the constraint that justified beliefs are not the same as true beliefs (though, of course, it is possible for them to be coextensive). This applies to belief sets as well as belief tokens. A set of beliefs can be justified while being false. Goldman's complaint about theories of rationality that are predicated on non-verific consequentialist theories of the criterion of correctness is that they can yield beliefs that are not true. But being not true is not in any way a criticism of a belief that is otherwise justified. Are we begging the question against verific consequentialist theories here? We do not think so. The kind of judgments we would make of intellectual achievement if we insisted that justified beliefs must be mostly true beliefs would be completely bizarre. We would end up insisting that most of what was believed about the natural world for the last 6000 years was not only false but also irrational. Worse, we would be in the position of thinking that our own scientific beliefs about the world are irrational, since we have some inductive reason to think that they are probably false. That seems completely wrong (see also Cohen 1984).

7

O n e problem with Alston's view that we d o not pursue here is that it relies o n a theory o f objective probability without actually offering one. W e have persistent doubts that a theory o f objective probability that is appropriate for epistemic justification can be constructed. See (Pollock 1989).

134

Critiques o f Epistemic Externalism

The same problem is in stark evidence in Plantinga's work. Plantinga is also a norm externalist. He proposes that: A belief has warrant for me if (1) it has been produced in me by cognitive faculties that are working properly...in a cognitive environment that is appropriate for my kinds of cognitive faculties, (2) the segment of the design plan governing the production of that belief is aimed at the production of true beliefs, and (3) there is a high statistical probability that a belief produced under those conditions will be true (1993b, pp. 46-7).

(1) is a distinctive feature of Plantinga's view that has attracted a great deal of attention. Plantinga himself heroically attempts to fend off some of the most pressing difficulties faced by the notion of proper function (1993b, chapter 2). (3) is an attempt to address generality problems faced by reliabilism. Plantinga aims to meet these challenges by proposing an account of objective probability (1993b, chapters 8 and 9). Though we have flagged our concern about objective probability and while Plantinga's defense of proper functions is ripe for continued careful scrutiny, we do not wish to challenge (1) or (3) here. (2) is the most relevant for our project. Apparently, Plantinga thinks that an orientation toward truth is simply demanded by having warrant as one's subject matter. Unfortunately, there is a perennial problem in dealing with Plantinga's theory in a discussion of internalism and externalism. He is at pains to emphasize that he seeks a theory of warrant and not a theory of justification, where warrant is whatever yields knowledge when added to true belief. That makes it so that warrant has to itself incorporate truth in order to allow knowledge to meet Gettier counterexamples in a way that is reminiscent of some now quite old proposals on knowledge (e.g., Lehrer and Paxon 1969). More importantly, this project makes warranted' not at all the same thing as 'rational.' Rationality, in our view, is not about helping knowledge to meet the Gettier problem. Perhaps all of this is just as Plantinga intends. That is, perhaps Plantinga does not mean to be talking about rationality at all. One might think that this is completely evident on the basis of his rejection of justification, but Plantinga rejects justification because he thinks that it has an ineliminable deontological cast. We do not think that that is the crucial thing about justification, even if talking about justification in terms of duty is a useful way to make connections to traditional epistemology, and a useful way to capture the notion of epistemic permissibility (as we do in Pollock and Cruz, p. 123). Our view is that justification is consequentialist while at the same time being distinct from warrant. 8 Insofar as Plantinga might be thought to be discussing the topic we are concerned about, he sometimes marshals counterexamples to non-verific accounts of the correctness of epistemic norms. These counterexamples are really no different than Goldman's fantasizer counterexamples. For example, Plantinga offers Paul, who "suffers from a brain lesion induced by radioactive fallout from a Soviet missile test. He now reasons differently from the rest of us; when appeared to in the church-bell fashion, he forms the belief that something is appearing to him in that fashion, and that it is orange"

8

Furthermore, if Plantinga's warrant has nothing at all to d o with rationality, then he is not entitled to criticize Pollock's view as not providing a theory of warrant (as he does in 1993a, chapter 8). After all, Pollock never meant to be talking about warrant.

The Chimerical Appeal of Epistemic Externalism

135

(p. 170). Paul is reasoning according to his norms, but Plantinga is worried that Paul's norms are incorrect. Plantinga, also following Goldman, sees fit to point out here that Paul's reasonings according to his pathological norms would not, when added to true belief, yield knowledge. This leads Plantinga to say that a non-verific account of the criterion of correctness will not do. If Plantinga means to be giving an account of rationality, it appears that Plantinga, like Goldman, is conflating rational beliefs with beliefs that are true or are somehow very closely connected with truth. Goldman's and Plantinga's appeal to knowledge is spurious, and it is precisely here where the truth-aimed motivation for externalism shades into a mistaken reverence for the analysis of knowledge. Traditionally, knowledge requires justified true belief plus some measure to guard against Gettier counterexamples. 9 Thus, on the face of it, truth and justification are independent elements of knowledge. It is inappropriate to demand that a theory of justification needs to bring a belief closer to knowledge by incorporating truth. The role of a theory of justification in knowledge is to distinguish between beliefs that are true but only accidentally so, versus beliefs that are true because the believer has some positive reason to think so (Unger 1975). Goldman's defense of verific accounts of the criterion of correctness therefore seems in grave danger of conflating rational beliefs with true beliefs. This is just the charge we made against belief externalism above. Though it is not always easy to see, externalists appear to be repeatedly making the same error. Norm externalists think that being aimed at the truth is the criterion of correctness for epistemic norms. It appears that the reason that they think this is that they are driven by a concept of knowledge that requires something more than true belief, but where what must be added must still be intrinsically truth-aimed. We doubt that any such concept of knowledge can be made sense of. There are at least two concepts of knowledge that have drawn the attention of epistemologists. O n e is a kind of connection between the knower and the fact known where truth is the sole important element. Thus, true belief is sufficient for knowledge of this sort. 10 The second concept of knowledge relevant to epistemology—and the concept that has received more attention—brings together true belief with justifying reasons. It is important to see how this second concept is supposed to operate. The justifying reasons are seen as the mechanism by which the genuine knower can be distinguished from the person who believes the truth, but does so accidentally or spuriously. Having reasons will ensure that the knower is brought along to the belief according to standards that are rationally defensible as procedures for belief formation. Unfortunately, coming to a true conclusion through rationally defensible procedures for belief formation can sometimes still fail to yield knowledge. This is the force of Gettier counterexamples (1963). The problem

9

Here we are restricting our attention to 'justificationist' strategies for analyzing knowledge. There are at least three other prominent ways with knowledge, however, namely causal theories, tracking theories and contextualist theories. See, e.g., Goldman (1970), Nozick (1981) and DeRose (1991) and Lewis (1996) respectively. 10 Lehrer (1990) discusses this concept of knowledge, and it can be found in Goldman's relatively recent work on social epistemology (1999).

136

Critiques of Epistemic Externalism

is that the truth believed and the rational procedures used may well be disconnected in the sense that the rational procedure accidentally arrives at truth. For example, in the classic Brown in Barcelona case, Smith reasons impeccably from his justified belief that Jones owns a Ford to arrive at the belief that either Jones owns a Ford or Brown is in Barcelona. Gettier points out that it can turn out that it is the second disjunct that makes the belief true, while the first (plus the innocent disjunction introduction) is what makes the belief justified. That the two quantities are present in the same belief is an accident, and so the specter of spurious epistemic achievement which initially motivated the second concept of knowledge looms again. Solving the Gettier problem amounts to forging a connection between the justifier and whatever makes for the truth of the belief. In our view, the Gettier problem is a curiosity, and the vast literature that it has inspired has shown that the second concept of knowledge is not at all well behaved. The point here, however, is that the externalist treats knowledge as being composed of true belief plus a notion of justification that also has truth-aimedness built in. It is far from obvious how building in truth aimedness is going to help, though, since it looks as if Gettier problems are going to be replicated at the level of justification itself unless justification is identical to the truth. This is simply and inevitably a consequence of justification and truth being different. Thanks to Gettier, the difficulty is now obvious in the case of the second concept of knowledge but it should be clear in the case of externalism's concept of knowledge, too. Externalism grasps for a concept of knowledge where the justification component is sufficiently separate from truth to maintain a distinction between mere true belief and knowledge, but where justification is at the same time sufficiently attached to truth so that having it will make knowledge likely and attainable. We are skeptical that there is any sense to be made of this blended sense of justification and the concept of knowledge that drives the externalist to hold out for it. Success in articulating this sense of justification appears just as likely as success in solving the Gettier problem. In summary, one can be an externalist by being either a belief externalist or a norm externalist. The belief externalist tries to formulate epistemic norms directly in terms of externalist considerations, but it is impossible to construct non ad-hoc procedural norms in this way. The norm externalist proposes instead to recommend evaluations of procedural norms on the basis of considerations of reliability. Norm externalism initially appeared compelling because it provided a way to preserve the internal and procedural nature of epistemic norms while still allowing for external assessment of those norms. Combined with internalism's apparent inability to make sense of the comparative evaluation of norms, it seemed that norm externalism was very promising. Unfortunately, norm externalism conflates justification and truth or relies on a concept of knowledge that is suspect. So, norm externalism must be rejected. As far as we can see, externalism has nothing to contribute to the solution to traditional epistemological problems. 11

11

Might the externalist be stalking a completely different epistemic achievement that has accidentally been called 'justification'? This is one way of reading the claim in, e.g., Goldman (1988) where he distin-

The Chimerical Appeal of Epistemic Externalism

137

5- A Naturalistic Internalism We have seen that externalism is motivated by the view that rationality must have a close association with truth. The externalist has tried to forge that close bond by attempting—either at the level of the norms or at the level of the criterion of correctness for the norms—to make belief-forming processes answer to truth 'directly'. That is, externalists about epistemic justification have tried to build into their theories a metaphysical connection to the truth that is much stronger than an epistemic agent's internal norms with regard to what it takes to be reliable or true. In rejecting this on behalf of internalism, it may seem that we are rejecting truth as the normative aim of epistemology. In a way, this is correct. Our conclusion is that truth is not a very interesting part of our best understanding of rationality. This is out of step with a very common view on the part of epistemologists, both internalists and externalists alike. The most common response that we have encountered to the rejection of a close tie between rationality and truth is the question, "what else is there for rationality to answer to?" The worry is that, if we do not think that the goal of rationality is to secure truths, then we have somehow missed the only imaginable target of rationality or that we have opened the door to a completely subjective or arbitrary conception of rationality. Moreover, the prima facie case for truth-aimed epistemology is clear. When we visit the doctor, we hope that she will form beliefs that are true. Even if her diagnoses sometimes fall short of truth, truth is what we expect her goal to be. In more mundane cases of belief formation we may find ourselves with extremely little evidence on which to base our beliefs. The evidence that we finally go with, though, appears to have a little something extra in its favor, and it seems plausible to think that that extra something is higher likelihood of pointing to the truth. Describing belief formation in this way is intelligible if the unit of analysis is the person. In a fairly straightforward sense, people have aims, and sometimes, perhaps often, people have the aim of reaching the truth. Epistemologists, however, typically do not focus on the personal level in their discussions of belief formation. 12 If the commitment is to be taken seriously with respect to belief formation, the 'aims' talk will ultimately need to be replaced by a specification of how belief formation aims at the truth when the relevant level is the subpersonal. Our understanding of the way that epistemologists have talked about subpersonal belief formation is through epistemic norms. When we take account of the role of epistemic norms in belief formation, it becomes much more difficult to sustain the claim that the aim of belief formation is truth because the norms themselves identify a proximal endpoint for the formation of beliefs. Aims, goals, and targets are appropriate for agent-level discussions of explicit decisions, as in

12

guishes weak and strong justification. There was a time when we were inclined to accept that internalist and externalist epistemologists were simply after two different things. It is now not so clear what the point of externalist theories of justification might be, except to accommodate the externalists problematic intuitions about the concept of knowledge. At any rate, we would be happier with externalism if externalists themselves regarded their enterprise as simply different from traditional theorizing about rationality. Our experience is that they rarely embrace this view. It seems that some participants in the debate have been mislead into thinking otherwise again by concentrating on the case of knowledge, where the level of analysis is often enough the personal level.

138

Critiques of Epistemic Externalism

the case of the doctor. We hope that she has the goal of achieving truth, as opposed to merely operating in a way that avoids malpractice suits. T h e metaphorical extension of this goal to her epistemic norms is strained, however. Beliefs formed in accordance with epistemic norms obey the constraints and demands of the norms themselves. Thus, there is an important and compelling alternative for the overall normative aim of belief, namely that the formation of beliefs follows the directives of the epistemic norms. That this is so should not be a surprise. W h e n epistemic norms are introduced as an intermediary between the raw materials for belief formation and the truth, there is inevitably the possibility of treating the norms themselves as the proximal endpoints of justified beliefs. Showing that truth is the distal aim of justified belief—with norms as the vehicle—would then require a further argument on the part of the externalist. Arguments of that sort were shown above to be inadequate. T h e framework tells against another kind of attempted vindication that has been advanced for truth-aimed epistemology. It might be thought that a commitment to veritism offers a way to mark off the distinctive nature of epistemic reasons (Harman 1997). T h e thought seems to be that, if belief formation is not aimed at truth, then there is no way to distinguish epistemic reasons from non-epistemic reasons. T h e reply to this claim is now straightforward, however. T h e argument only goes through if it can be shown that there is no adequate way of characterizing epistemic reasons apart from their truth-aimedness. But that is patently not the case given the nature of epistemic norms. Epistemic reasons will be just those contentful states that figure in epistemic norms. T h e norms, in turn, are identified by their output of epistemically laudable belief. We conclude from this that a truth-aimed externalist epistemology is not entitled to carry the day solely on the allegation that it is intrinsic to beliefs or the processes of belief formation that they must aim at truth. Instead, the right picture in our view involves truth being a person's aim (at least some of the time), while beliefs are formed through epistemic norms that only answer to themselves as evaluative standards. At the start of this essay, we claimed that a final motivation for externalism is its apparent affinity with naturalistic approaches in epistemology (Goldman 1994; 1999; Kitcher 1992). T h e reason that externalism has been viewed to have an advantage in this regard is partly an accident of the literature. As a matter of recent historical fact, it has primarily been externalists that have proposed self-consciously naturalized epistemologies. Another reason naturalism and externalism are sometimes associated is that the best know naturalized epistemology—that of Goldman—is also the best-known externalist epistemology. As we indicated above, his view articulates a crucial role for cognitive belief-forming processes in a theory of justification, and he has proposed that it is the job of cognitive science to understand those processes. Since cognitive science is a naturalistic inquiry, Goldman's process reliabilism has been seen to be especially amenable to naturalistic motivations. There is room here to worry about the specific proposals of naturalistically inclined externalists. We propose, however, simply to show that externalism does not have exclusive rights to naturalism. Thus we end with a sketch of an internalist theory whose contours reflect the anti-externalist considerations above and is at the same time plausibly naturalistic. (For the full discussion, see Pollock 1986; 1995; Pollock and Cruz 1999).

The Chimerical Appeal of Epistemic Externalism

139

We take it that theories of human rationality describe contingent psychological features of the human cognitive architecture, though it is not entirely an accident that humans are built the way they are. Environmental pressures have led to our evolving in particular ways. We represent one solution to various engineering problems that were solved by evolution. There is no reason to think that these problems always have a single, or even a single best, solution. Certain engineering problems call for arbitrary choices between solutions that work equally well. On the other hand, many of the more general features of our cognitive architecture may reflect the only, or one of a small number of, solutions to general problems of cognitive engineering. For example, it can be argued that very general logical and computational constraints on cognition dictate that any sophisticated cognitive agent will engage in defeasible reasoning, will reason defeasibly from perceptual input, will reason inductively, and will engage in certain kinds of planning behavior (Pollock 1995). When that is true, it provides an informative explanation for why human cognition works as it does. A simplistic view of evolution has it that evolutionary pressures select for traits that enhance survivability of an organism. In describing a naturalistic alternative to externalism, we do not need to go beyond this simplistic view, but it should be kept in mind that a serious appraisal of evolution will reveal a much more subtle theory than we are relying on here. We might regard the design problem for rationality to be that of creating an agent that can survive in a hostile world by virtue of its cognitive capabilities. This, however, presupposes a prior understanding of what cognition is and how it might contribute to survivability. We take it as characteristic of rational cognition that a cognitive agent has doxastic states ("beliefs", broadly construed) reflecting the state of its environment, and conative states evaluating the environment as represented by the agents beliefs. It is also equipped with cognitive mechanisms whereby it uses it beliefs and conations to select actions aimed at making the world more to its liking. A rational agent has beliefs reflecting the state of its environment, and it likes or dislikes its situation. WTien it finds the world not entirely to its liking, it tries to change that. Its cognitive architecture is the mechanism whereby it chooses courses of action aimed at making the world more to its liking. Within this "doxastic-conative" loop, epistemic cognition is, in an important sense, subservient to practical cognition. The principal function of cognition is to direct activity (practical cognition), and the role of epistemic cognition in rationality is to provide the factual background required for practical cognition. We can evaluate how well rational agents implement the doxastic-conative loop. A judgment of how well an implementation performs must always be relative to a set of design goals that one implementation may achieve better than another. As remarked above, a natural design goal is to construct an agent capable of using its cognitive capabilities to survive in an uncooperative environment. This is motivated by considerations of evolution and natural selection. There is probably no privileged design goal in terms of which to evaluate cognitive architectures, and truth certainly does not recommend itself as the only consideration. The key point is that cognition is evaluated relative to practical goals. Thus, our view is a consequentialist view. We can accept the arguments for a consequentialist conception of epistemic norms offered by externalists like Goldman and Plantinga.

140

Critiques o f Epistemic Externalism

Where Goldman and Plantinga go wrong is in their insistence that attaining truths is the only sensible goal of rational cognition. The overall human cognitive architecture is what it is, and human rationality is defined relative to it. We can evaluate cognitive architectures in terms of how well they achieve their design goal, and this is reminiscent of norm externalisms evaluation of epistemic norms in terms of reliability. But one big difference is that practical and epistemic cognition are evaluated as a package. The ultimate objective is not truth, but practical success through the operation of epistemic norms. Our diagnosis of the attraction of externalism (simpliciter) is that it accommodates the widely held conviction that rationality is extremely closely connected with the attainment of truths. This has led externalists to propose a connection between truth and rationality that threatens to blur the distinction between the two. We think that it is misguided to forge the connection so closely, but without a connection to truth, there is an apparent lacuna in the theory of rationality, namely the puzzle about what rationality answers to. We propose that this lacuna is filled with an account of the relationship between practical and theoretical cognition sketched above. This broadly naturalistic framework seems most faithful to the goal of distinguishing rational beliefs from irrational beliefs within internalist framework. 13

References Alston, W. ( 1 9 8 6 ) . Internalism and Externalism in Epistemology. Philosophical Topics, 14(1). Alston, W. ( 1 9 8 8 ) . An Internalist Externalism. Synthese, 74, pp. 2 6 5 - 2 8 3 . Alston, W. ( 1 9 9 5 ) . H o w to T h i n k A b o u t Reliability. Philosophical Topics, 23, pp. 1-29. Audi, R. ( 2 0 0 1 ) The Architecture of Reason. N e w York: O x f o r d UP. Bonjour, L. ( 1 9 8 0 ) . Externalist Theories o f Empirical Knowledge. In P. A. French, Τ. E. Uehling, & Η . K. Wettstein (eds.), Midwest Studies of Epistemology V: Studies in Epistemology . Minneapolis, M N : University o f Minnesota Press. Bonjour, L. ( 1 9 8 5 ) . The Structure of Empirical Knowledge. C a m b r i d g e : Harvard UP. Bonjour, L. ( 1 9 9 9 ) T h e Dialectic of Foundationalism and Coherentism. In J . Greco a n d E. Sosa (eds.), The Blackwell Guide to Epistemology, Maiden: Blackwell. C h i s h o l m , R. ( 1 9 8 9 ) . Theory of Knowledge. (3 r d ed.), Englewood Cliffs, N J : Prentice-Hall. C o h e n , S. (1984). Justification and Truth. Philosophical Studies, 46. C o n e e , E „ & Feldman, R. (1998). T h e Generality Problem for Reliabilism. Philosophical Studies, 8 9 ( 1 ) , pp. 1-29. Cruz, J . (unpublished manuscript). Epistemic Norms and the Sellarsian Dilemma for Foundationalism. D e R o s e , Κ. ( 1 9 9 5 ) . Solving the Skeptical Problem. The Philosophical Review, 104, pp. 1-52. Feldman, R. ( 1 9 8 5 ) . Reliability and Justification. The Monist, 6 8 ( 2 ) , pp. 159-174. Field, H . (2000). Apriority as an Evaluative N o t i o n . In Boghossian, P. and Peacocke, C . (eds.), New Essays on the A Priori, N e w York: O x f o r d UP. Foley, R. ( 1 9 8 7 ) . The Theory of Epistemic Rationality. C a m b r i d g e , M A : Harvard UP. Foley, R. ( 1 9 9 3 ) . Working Without a Net: A Study of Egocentric Epistemology. O x f o r d : O x f o r d UP. Gettier, E. ( 1 9 6 3 ) . Is Justified True Belief Knowledge? Analysis, 2 3 , pp. 121-23.

13

T h a n k s are owed to J i m Pryor, T o m Kelly, and Melissa Barry for conversation and c o m m e n t s on the themes o f this essay.

The Chimerical Appeal of Epistemic Externalism

141

Goldman, A. (1979). What is Justified Belief? In G. Pappas {ed.), Justification and Knowledge. Boston, MA: D. Reidel. Goldman, A. (1980). The Internalist Conception of Justification. In P. A. French, Τ. E. Uehling, & Η. K. Wettstein (eds.), Midwest Studies in Philosophy V, pp. 27-51, Minneapolis: University of Minnesota Press. Goldman, A. (1985). The relation between epistemology and psychology. Synthese, 64, pp. 29-68. Goldman, A. (1986). Epistemology and Cognition. Cambridge: Harvard UP. Goldman, A. (1994). Naturalistic Epistemology and Reliabilism. In P. A. French, Τ. E. Uehling, & Η. K. Wettstein (eds.), Midwest Studies in Philosophy XIX: Philosophical Naturalism, pp. 301-320, Notre Dame, IN: University of Notre Dame Press. Goldman, A. (1999). Internalism Exposed. The Journal of Philosophy, 46(6), pp. 271-93 Goldman, A. (1999b). Knowledge in a Social World. Oxford: Oxford UP. Harman, G. (1997). Pragmatism and Reasons for Belief. Reprinted in Reasoning, Meaning, and Mind, pp. 93-116, Oxford: Clarendon Press. Kitcher, P. (1992). The Naturalists Return. The Philosophical Review, 101(1), pp. 53-114. Lehrer, Κ. & Paxon, Τ. (1969). Knowledge: Undefeated justified true belief. Journal of Philosophy, 66, pp. 225-37. Lewis, D. (1996). Elusive Knowledge. Australasian Journal of Philosophy, 74, pp. 549-67. Moser, P. (1985). Empirical Justification. Dordrecht, Holland: D. Reidel. Nozick, R. (1981). Philosophical Explanations. Cambridge, MA: Harvard UP. Plantinga, A. (1988). Positive Epistemic Status and Proper Function. In J. E. Tomberlin (ed.), Philosophical Perspectives, 2, Epistemology, 1988, pp. 1-50, Atascadero, CA: Ridgeview. Plantinga, A. (1990). Justification in the 20th Century. Philosophy and Phenomenological Research /(supplement), pp. 45-71. Plantinga, A. (1993a). Warrant: The Current Debate. New York: Oxford University Press. Plantinga, A. (1993b). Warrant and Proper Function. New York: Oxford University Press. Pollock, J. (1974). Knowledge and Justification. Princeton: Princeton University Press. Pollock, J. (1984). Reliability and Justified Belief. Canadian Journal of Philosophy 14, pp. 103-114. Pollock, J. (1986). Contemporary Theories of Knowledge. Savage, MD: Rowman and Littlefield Pub. Pollock, J. (1990). Nomic Probability and the Foundations of Induction. New York, NY: Oxford University Press. Pollock, J. (1995). Cognitive Carpentry. Cambridge, MA: MIT Press. Pollock, J. & Cruz, J. (1999). Contemporary Theories of Knowledge, 2 nd ed., Lanham, MD: Rowman & Littlefield Pub. Sosa, E. (1980) The Raft and the Pyramid. In Midwest Studies in Philosophy Vol. V: Epistemology Unger, P. (1975). Ignorance: A Case for Skepticism. Oxford: Clarendon Press.

In Search of Internalism and Externalism RICHARD FELDMAN

The recent debate about internalism and externalism in epistemology is hindered by the lack of clarity in the terms "internalism" and "externalism" and by a significant lack of uniformity among participants to the debate over the commitments of the two kinds of view. In this paper I will attempt to provide one way to sort things out, using the origins the debate as the basis for classifying theories.

I. Origins of the Debate The words "internalism" and "externalism" were originally used as labels for categories of theories that philosophers wanted to criticize or defend. Hilary Kornblith traces the current general usage of "externalism" to David Armstrong's 1973 book, Belief, Truth and Knowledge.1 Armstrong wrote According to "Externalist" accounts of non-inferential knowledge, what makes a true noninferential belief a case of knowledge is some natural relation which holds between the beliefstate, Bap, and the situation which makes the belief true. It is a matter of a certain relation holding between the believer and the world. 2

The natural relation Armstrong had in mind was "a law-like connection between the state of affairs [of a subject's believing that p] and the state of affairs that makes '/>' true such that, given the state of affairs of [the subject's believing that p], it must be the case that One idea that we can extract from this early usage is that theories of knowledge highlighting this sort of law-like connection are externalist theories. Armstrong makes clear that he is contrasting this view with "Cartesian" views and with theories that make certain beliefs justified because they are "initially credible." According to these other theories, beliefs are justified because they stand in the right relation to other beliefs or because they have initial credibility. Presumably, Armstrong took these to be "unnatural" relations or states. Such theories count as internalist theories. A few years before Armstrong wrote the passage just quoted, AJvin Goldman defended a causal theory of knowledge. 4 His theory required for knowledge that one

1

Hilary Kornblith, "Internalism and Externalism: A Brief Historical Introduction," in Hilary Kornblith,

ed. Epistemology: Internalism and Externalism (Maiden, Ma: Blackwell, 2001). See p. 1. 2 3 4

David Armstrong, Belief, Truth and Knowledge (London: Cambridge University Press, 1973), p. 157. p. 166. Alvin Goldman, "A Causal Theory of Knowing," Journal of Philosophy 64 (1967): 3 5 7 - 7 2 .

144

Critiques of Epistemic Externalism

have a true belief that is connected by an appropriate type of causal chain to the fact that the belief is about. The causal connections between beliefs and states in the world that Goldman's theory highlights surely count as natural relations in Armstrong's sense. Goldman also contrasted his own theory with Cartesian style alternatives. He concludes this paper by writing that his view flies in the face of a well-established tradition in epistemology, the view that epistemological questions are questions of logic or justification, not causal or genetic questions. 5

Although Goldman does not use the words "internalism" and "externalism" here, it is clear that, were he to employ Armstrong's terminology, he would describe this theory as externalist and characterize the Cartesian views as internalist. Some years later, Goldman did discuss internalism and externalism explicitly. He says that internalism "takes its inspiration from ... Descartes." 6 Among the Cartesian themes he discusses are infallibility and the first person perspective. He ends up defending a reliabilist theory, a version of externalism. In the same volume as the one in which Goldman's paper appeared, Laurence Bonjour published an influential critique of externalism. Bonjour notes that the extraordinary thing about externalism is that in implies that "a person's beliefs might be epistemically justified in virtue of facts or relations that were external to his subjective conception." 7 The causal and reliabilist facts and relations that Armstrong and Goldman describe are prime examples of the external things to which Bonjour refers. From these passages we can draw a few ideas about some of the things that are associated with internalism and externalism. Internalism has its roots in Descartes, it emphasizes a first person perspective, and it disallows facts such as reliability or causal connectedness from contributing to epistemic justification (when the believer does not know about them). Externalism, in contrast, makes epistemic status partly dependent upon such facts. One way to spell the distinction between internalism and externalism is in terms of the kinds of things upon which epistemic facts supervene. Internalism, on this view, is a thesis holding that certain epistemic matters supervene on certain internal states. 8 The internal states in question are experiences, apparent memories, other beliefs, and the like. In general, they are mental states of the believer. Internalism implies that necessarily if two people are internally alike, then they are epistemically alike. If there are epistemic differences, then there must be internal differences. The kind of supervenience in question is strong supervenience, which applies across worlds, so that believers in different worlds who are internally alike are epistemically alike. Varieties of internalism can be constructed by varying the epistemic states that are alleged to supervene on these internal states. A general characterization of internalism is thus

5 6 7 8

p. 372. Alvin Goldman, "The Internalist Conception of Justification," Midwest Studies in Philosophy 5 (1980): 27-51. The quotation is from p. 27. Laurence Bonjour, "Externalist Theories of Empirical Knowledge," Midwest Studies in Philosophy 5 (1980): 53-73. The quotation is from p. 56. For discussion of this view, see Earl Conee and Richard Feldman, "Internalism Defended," in Hilary Kornblith, ed., Epistemology: Internalism and Externalism (Maiden, MA: Blackwell, 2001), pp. 231-260.

Search of Internalism and Externalism

145

I N T l . An epistemological theory is an internalist theory iff it implies that epistemic status strongly supervenes on mental states. Externalism denies this supervenience thesis, holding that epistemic status supervenes in part on external facts and relations of the sort mentioned. It allows that people who are internally alike may be epistemically different in virtue of differences in external facts, such as facts about causation or reliability. Some philosophers would argue that internalism implies more than just the supervenience thesis stated in ( I N T l ) . I will consider some alternative accounts of internalism in Section III. In Section II I will focus on what exactly is supposed to supervene on that internal base.

II. Epistemic Status In this section I will discuss just what epistemic state internalists take to supervene on internal states. 9 I will consider first the idea that it is knowledge that supervenes on internal states. I will then turn to other epistemic states.

A. Knowledge If one ties internalism very closely to Cartesian epistemology and one makes some particular assumptions about just what Cartesian epistemology amounts to, then one can end up thinking that it is knowledge that is supposed to supervene on internal states. T h e assumptions in question are that the clear and distinct perception of the truth of a proposition is necessary (and sufficient) for knowledge and that whether one clearly and distinctly perceives the truth of a proposition is an internal, or mental, matter. I will not discuss the merits of these theses as interpretations of Descartes. O n e can combine this kind of Cartesian view with ( I N T l ) , yielding the thesis that knowledge supervenes on internal (mental states). Tying internalism this closely to Cartesianism is something of an embarrassment to nearly all participants in the recent debate about internalism and externalism. It is an embarrassment to internalists since it almost surely entails a rather pervasive skepticism that almost all internalists wish to deny. Given what clarity and distinctness is supposed to be, it is plain that we clearly and distinctly perceive little. Hence, given that it is a necessary condition on knowledge, it follows that we know little. But few contemporary internalists are willing to accept this consequence. Externalists who take internalism to imply that knowledge supervenes on internal states suffer from a different kind of embarrassment. They could quickly reject inter-

9

Some of the material in this section is a revised version of parts of my paper "Internalist Epistemic Evaluations," in Matthias Steup, ed., Contemporary Debates in Epistemology, forthcoming.

146

Critiques of Epistemic Excernalism

nalism on the basis of the skeptical implications just mentioned. There is no need to resort to more elaborate claims and arguments, some of which will be discussed below. Furthermore, if internalism is limited to the kind of Cartesianism sketched here, and if externalism is just the denial of internalism, then theories dramatically different from reliabilism and causal theories would count as externalist. Any fallibilist theory, including fallibilism as advocated by seemingly paradigmatic internalists, would count as externalist. This point deserves brief elaboration. A fallibilist theory is a theory holding that one can have knowledge when one has evidence that is not logically decisive. Given that a body of evidence is not logically decisive, there are possible cases in which a person has that evidence and the belief in question is false. It follows that it is possible for two people who have the same evidence, and who are otherwise internally alike, to be such that one has knowledge and the other does not. Given the possibility of deceptive demons and brains in vats, any non-skeptic must accept fallibilism. Few, if any, internalist targets of externalist criticism reject fallibilism. Thus, if externalists are to classify theories in ways that fit their arguments, they must reject the idea that internalism is committed to Cartesian infallibilism. Internalism is not the view that knowledge supervenes on internal states.

B. Warrant According to epistemological folklore, prior to 1963 the accepted view was that knowledge is justified true belief. The publication of Edmund Gettier's "Is Justified True Belief Knowledge?" changed everything. 10 Prior to Gettier, one could view the internalism/ externalism debate as a debate about the nature of the third condition on knowledge. That is, it could be taken as a debate between those who think that the third condition is an internal one—having adequate evidence or good reasons—and those who think it requires the appropriate kind of causal or reliable connection. One can formulate the debate about the third condition on knowledge in either of two ways. One is to identify the possession of reasons with justification, and then attribute to the externalists the view that knowledge does not require justification. The quotation from Goldman in Section I supports this, since he contrasts his theory with theories that make knowledge a matter of justification. On this alternative, internalists hold that knowledge requires justification and externalists deny it, holding that knowledge requires causal connections or reliability instead. The other way to formulate the debate identifies justification with the "third" element of the analysis of knowledge, whatever it is. Internalists and externalists agree that knowledge requires justification, but they differ about the nature of justification. Although not much turns on it, I'll adopt the first way of formulating the issue. Adopting the way of framing the debate just described enables us to continue to say that internalists and externalists differ over whether justification is necessary for knowledge. But now formulating the internalist position clearly requires a bit more

10

Edmund Gettier, "Is Justified True Belief Knowledge? Analysis

2 3 (1963): 2 1 - 3 .

Search of Internalism and Externalism

147

care. Ever since Gettier, internalists have denied that what is needed for knowledge beyond true belief is just something internal. They have thought there is some condition needed to deal with the Gettier problem, and this condition, whatever it is, will not be a purely internal matter. That is, two people who are internally alike could differ in that one is in a Gettier case and the other is not. Thus, internalists think that justification is an internal matter and is necessary for knowledge, but they deny that only internal factors are needed to differentiate true beliefs that are knowledge from true beliefs that are not. The consequence of this is that we cannot describe internalism as the thesis that whatever it is that distinguishes true beliefs that are knowledge from true beliefs that are not knowledge supervenes on internal factors. Alvin Plantinga has called whatever makes this distinction "warrant". 11 Ever since Gettier, internalists have realized that warrant does not supervene on internal states. Internalism is not the thesis that warrant supervenes on internal states.

C. Well-foundedness There is another way true beliefs can fail to be knowledge. This yields another epistemic status that cannot plausibly be regarded as the subject matter of the central internalist thesis. A person can believe a proposition for which he or she has good reasons. But if the belief is not based on those good reasons, and it results from wishful thinking or a mistaken inference from different information, then the person lacks knowledge. In other words, the person lacks knowledge when the belief is not based on the justifying reasons. Such beliefs are ill-founded. Beliefs that are based on good reasons are well-founded}1 Exactly what it is for a belief to based on something is a matter of some controversy.13 It is not, or at least not obviously, an internal matter of the sort internalists have in mind. Presumably, a person can be ignorant of the bases of his belief, and no amount of introspection need reveal this to him. Perhaps this makes it implausible to suppose that well-foundedness supervenes on purely internal states. Two believers might be internally alike, yet one's belief might result from different causes than the others. If the proximate causes of beliefs are always other mental factors, then this would always make for a mental difference, though not always an accessible difference. But it is not clear that beliefs must always have a proximate cause that is mental. It is possible for a belief to be directly caused by external factors. If this makes it the case that the belief is based on those factors, then what a belief is based on need not be internal by any standards. Of course, one might say that such external causes are not bases of belief. A

11 Alvin Plantinga, Warrant: The Current Debate (Oxford: Oxford University Press, 1993), p. 3. 12 One source for this terminology is Richard Feldman and Earl Conee "Evidentialism," Philosophical Studies, 48 (1985): 15-34. 13 For a recent discussion, see Daniel Mittag, "On the Causal-Doxastic Theory of the Basic Relation," Canadian Journal of Philosophy 32 (2002): 543-60.

148

Critiques of Epistemic Externalism

basis is something that a believer must use in some more intentional way in forming the belief. If that is what basing is, then perhaps well-foundedness can more plausibly be regarded as an internal matter. However, I will not argue for this here. Instead, I will grant that well-foundedness does not supervene on internal states. Internalism is not the thesis that well-foundedness supervenes on internal states.

D . Justification T h u s far, I have argued that internalism is best not taken to be the view that knowledge, warrant, or well-foundedness supervene on internal states. Another possibility is that it is justification that is supposed to supervene on internal states. Given the choice made earlier about the use o f the word "justified", this thesis may appear obviously true. Since a person is justified in believing a proposition just in case the person has good reasons for it, justification is an internal matter provided the possession of good reasons is an internal matter. It seems clear that this is an internal matter, and thus it seems clear that this version of internalism is correct. I will examine later one reason to think otherwise, but for the present section I will assume that whether one possesses good reasons to believe something is an internal matter. O n the present construal, internalism is the relatively uncontroversial thesis that whether one is justified in believing something—whether one has good reasons to believe that thing—supervenes on one's internal states. Externalists are not apt to deny this thesis. What, then, is externalism? A plausible answer to this question is that externalism is the thesis that justification, so construed, is not necessary for knowledge. Instead, what is needed is causal connectedness or reliability or some such external factor. This way of formulating things makes sense of the origins of the debate. T h e causal theorists and reliabilists do deny that justification is necessary for knowledge, and thus they turn out to be externalists. Proponents of more traditional theories claim that justification is necessary, and they turn out to be internalists. This gets the classifications right, at least in that it classifies the paradigm cases properly. H a d we used the word "justified" differently, externalists would have turned out to be those who thought that justification is not an internal matter. But this way o f formulating things would have its liabilities, since it is somewhat difficult to see how to formulate some externalist theories as theories of justification. Consider, for example, the causal theory. A person can be justified in believing a proposition that the person does not actually believe. It is difficult to see what a causal theorist's account of justification might be. It won't do to say that a person is justified in believing a proposition, p, just in case there is some potential link between the state of affairs making ρ true and the person's potential belief in p. If ρ is true, then there is always some potential causal connection of the sort specified. Another difficulty in construing the causal theory as a theory o f justification follows from the fact that a belief can be justified without being true. Since there needn't be anything that makes the belief true, there need not be any causal connection of any sort between any such fact and the actual or potential belief in p. T h i s is, perhaps, why causal theories are often formulated as theories of knowledge rather than as theories of justification.

Search of Internalism and Externalism

149

Sticking with the decision to use "justified" in a manner more congenial to internalism, then, requires making internalism be the thesis that justification supervenes on internal states and that justification is necessary for knowledge. Of course, as noted earlier, it is possible for internalists to require more than the supervenience thesis discussed in this section. That is, they might assert not just ( I N T l ) , but rather something stronger. In the next section I will examine these more demanding accounts of internalism. The conclusion of the discussion so far is that internalism can be expressed by the following claim: INT2. An epistemological theory is an internalist theory iff it implies that (i) epistemic justification strongly supervenes on mental states and (ii) epistemic justification is necessary for knowledge.

E. A Complication Externalist theories about content complicate matters in a manner that deserves brief attention. If externalism about content is correct, then two people who are internally alike may be believing different propositions. The different propositions are determined by their external circumstances. The differing contents of their beliefs may make it the case they are justified in believing different things. This undermines the letter, though not the spirit, of internalism as formulated in (INT2). I think that the simplest response for internalists is to concede that if externalism about content is true, then it is not the case that people who are internally alike are justified in believing the same propositions. However, something very close to this may nevertheless be true. If externalism about content is correct, then people who are internally alike have beliefs that correspond in certain way. Their external circumstances will determine which propositions their internal states relate them to. Epistemological internalism will then be the thesis that if two people are internally alike, then if one is justified in believing a proposition, then the other is justified in believing the corresponding proposition. Thus, if content externalism is true, then epistemological internalism is best taken as the thesis that internal states determine which beliefs are justified, but external factors partially determine what the contents of those beliefs are. I will ignore this complication for the remainder of this paper.

III. Internalism A significant aspect of the debate about internalism and externalism has turned on the supposition that internalism requires more than the conditions specified in (INT2). 1 4 In

14

Some of the material in this section is a significantly modified version of portions of my paper, "Chisholm's Internalism and Its Consequences," Metaphilosophy 34 (2003): 603-620

150

Critiques of Epistemic Externalism

this section I will argue against this supposition. In thinking about this, it is important to keep distinct two questions: what's the best version of internalism? what counts as an internalist theory? One might think that something more is required for a good theory of justification. For example, one might think that justification requires not merely having good reasons but also knowing that one's reasons are good reasons. This, however is no reason to insist that internalism itself imposes this more demanding requirement. It is also worth emphasizing at the outset of this discussion that the words "internalism" and "externalism" have not been used uniformly and any decision about their usage here is to some extent stipulated. Still, I think that the origins of the debate provide some basis for rejecting the more restrictive accounts of what internalism is. One possibility for a strengthened account of internalism turns on the idea that there are internal (or mental) states to which believers lack access and that it is contrary to the spirit of internalism to allow epistemic status to supervene on such states. Thus, one might propose a more restrictive thesis: I N T 3 . An epistemological theory is an internalist theory iff it implies that (i) epistemic justification strongly supervenes on accessible mental states, and (ii) epistemic justification is necessary for knowledge. It is unclear just what additional constraints the access requirement places on justification. Perhaps there are non-conscious mental states. Perhaps they count as inaccessible when the only way one can learn about them is through extensive psychotherapy. Perhaps, on the other hand, the fact that one can learn about them through extensive psychotherapy shows that they are, after all, accessible. I find the issue too obscure to resolve, and I will remain neutral on it here. The difference between (INT2) and (INT3) will not be crucial for the discussion that follows. Roderick Chisholm is often cited as a leading figure among recent internalists. We might take Chisholm's views to represent paradigmatic internalist claims. Some of his remarks imply that internalism requires more than even ( I N T 3 ) demands. In characterizing his internalist view, Chisholm wrote: We can k n o w what it is, on any occasion, that constitutes our g r o u n d s , or reason, or evidence for thinking that we k n o w . 1 5

It is worth calling attention to an ambiguity in this passage. O n one reading, it requires only that the things which are our reasons are such that we can know them. Assuming that grounds are mental things, this is an endorsement of something like (INT3). O n a stronger reading, it requires that, for each justified belief, we know what our reasons for that belief are. A second quotation suggests that Chisholm had in mind the stronger reading. T h e internalist a s s u m e s that, merely by reflecting u p o n his own conscious state, he can formulate a set o f epistemic principles that will enable h i m to find out, with respect to any possible belief he has, whether he is justified in having that belief. T h e epistemic principles

15

Theory of Knowledge, 2nd ed., (Englewood Cliffs, NJ: Prentice Hall, 1977), p. 17.

Search of Internalism and Externalism

151

that he formulates are principles that one may come upon and apply merely by sitting in one's armchair, so to speak, and without calling for any outside assistance. In a word, one need only consider one's own state of mind. 1 6

This imposes what we might call a requirement for knowledge of a "linking principle." It implies that for a person to be justified in believing one proposition on the basis of some grounds or reasons, then the person must also know (or be justified in believing) that those grounds support that proposition. Or, more exactly, it implies that the person must be able to formulate the principles linking the grounds to the belief. 17 It is possible to detach these two parts of Chisholm's view. That is, it is possible to construct a theory according to which a belief is justified if one has good reasons for it, whether or not the believer can formulate principles linking the evidence to the belief. And it is possible to insist that the believer be able to do this, as Chisholm seems to do here. For convenience, I will call theories that reject the linking principle requirement simple good reasons theories (SGR theories) and theories that endorse some version of the requirement complex good reasons theories (CGR theories). A CGR theory can require only that justified believers can figure out or can have access to the linking principles or it can require that they actually have beliefs about the matter. If one holds that the former view is essential to internalism, then one might describe internalism as follows: INT4. An epistemological theory is an internalist theory iff it implies that (i) epistemic justification strongly supervenes on internal states, (ii) epistemically justified believers always can know which internal states contribute to the justification of any particular belief, and (iii) epistemic justification is necessary for knowledge. As stated, (INT4) combines a supervenience thesis with an additional requirement which is not formulated in terms of supervenience. Still, (INT4) is consistent with the internalist thesis stated above, according to which believers who are internally alike are also epistemically alike. This rests on the plausible assumption that if there is a difference in what two people can know about which accessible states contribute to the justification of particular beliefs, then there is an internal difference between them. I will return to (INT4) after briefly looking at some even stronger conditions on what counts as internalism. There are even stronger theses that one might associate with internalism. A less than careful reading of the passages from Chisholm might lead one to take his position to be that internalists hold not just that people can know what justifies their beliefs, but that in fact they always do know this. Thus, one might propose something like:

16

T h e o r y of Knowledge, 3rd ed., p. 7 6 .

17

There is a difference between being able to formulate the general epistemic principles about justification and knowing in a particular case that one's evidence supports o n e s belief. Requiring either amounts to requiring some sort o f access to linking principles. I will not be careful about this distinction in what follows.

152

Critiques of Epistemic Externalism

I N T 5 . An epistemological theory is an internalist theory iff it implies that (i) epistemic justification strongly supervenes on accessible internal states, (ii) epistemically justified believers always know which accessible internal states contribute to the justification o f any particular belief, and (iii) epistemic justification is necessary for knowledge. Finally, one might propose that not only do people always know these facts about justification, but that they never make any mistakes about the matter. This yields a principle such as: I N T 6 . An epistemological theory is an internalist theory iff it implies that (i) epistemic justification strongly supervenes on accessible internal states, (ii) epistemically justified believers always know which accessible internal states contribute to the justification o f any particular belief and they never have false beliefs about what contributes to the justification o f any particular belief, and (iii) epistemic justification is necessary for knowledge. I think that we can reject theories that meet the conditions specified in ( I N T 5 ) and ( I N T 6 ) with little argument. People are often justified in believing things and they often have knowledge o f the world around them. Few have given much thought to epistemic justification and fewer still know much about what justifies their beliefs. Even epistemologists who study these matters may fail to know what justifies their beliefs, and they may make mistakes about the matter. These theories require far too much for justification. I f any C G R theory is correct, it is the less demanding sort specified by ( I N T 4 ) . O f course, the fact that theories conforming to ( I N T 5 ) and ( I N T 6 ) fail does not show that these principles do not capture what internalism is. It may even be that some philosophers who are labeled as internalists have said things that suggest that justification requires the sort o f meta-knowledge clause (ii) o f these principles imposes. 18 Internalism may be a view about justification that has the implication that few beliefs are justified. However, there is no reason to saddle internalism with such a strong thesis. Rejecting this meta-knowledge requirement does not make one a causal theorist or a reliabilist. O n e can still be in the tradition that highlights the need for reasons or evidence. Furthermore, even an allegedly paradigm internalist such as Chisholm was not committed to anything so strong. He said only that by reflecting one can formulate epistemic principles. He did not say that all justified believers already know them. He did not even say that one can easily figure them out, or that a moments reflection would reveal them. He just said that, according to internalism, one can figure out these principles from one's armchair. Presumably, he thought this is what epistemologists do, and that anyone can (in some sense) do epistemology. I conclude that ( I N T 5 ) and ( I N T 6 ) can be rejected as accounts o f internalism. Chisholm himself seemed to be committed to something like ( I N T 4 ) as a statement o f what internalists are committed to. Exactly what the implications o f internalist theo-

18

One could read the passage from Bonjour quoted in Section I in a way that suggests this.

Search of Internalism and Externalism

153

ries are by this standard is difficult to settle since the "can" in ( I N T 4 ) is a potentially important hedge. Richard Fumerton has written that one o f the key issues that separates internalists and externalists concerns "the question o f whether one takes access to inferential connections to be a necessary condition for inferential justification." 1 9 As I understand Fumerton, he is asserting that to be an internalist, one must accept a C G R theory. Again, the notion o f "access" is less than fully clear. I will discuss ( I N T 4 ) below. But first, it will be worth considering briefly the merits o f C G R theories themselves. Hilary Kornblith has called attention to puzzling examples in which people are significantly confused about what leads them to certain o f their beliefs. 20 In one such example, a person looks at a pair o f photographs and tries to determine which photo shows the happier person. People are able to do this well, but when asked how they do it, they give incorrect answers. Typically they say that they judge how happy people are by their mouths—happier people have the corners o f their mouths upturned. In fact, however, the people in the photograph do not differ with respect to the shapes o f their mouths. T h e difference that actually causes the people to judge as they do is the size o f their pupils. 21 Although Kornblith does not use this example to argue against C G R theories, it will be instructive to consider what the theories imply about it. Imagine a person who is the subject o f the photo sorting experiment. She looks at a photo and forms the belief the person in the photo is happy. 22 We can make the most trouble for C G R theories if we assume that the photo sorter does in fact know, and have a justified belief, that the person in a particular photo is happy. Given a C G R theory conforming to the dictates o f (INT4), it follows that the photo-sorter can know what constitutes her grounds or reasons or evidence for thinking that she knows that the person in the photo is happy. Against this, one might argue that the photo sorter does not know what her grounds are. Her grounds are actually that the person has large pupils. She mistakenly thinks that her grounds are that the person has an upturned mouth. Two things makes examples like this less than decisive. For one thing, as already has been emphasized, Chisholm's principle says that people can know what their grounds are. The example shows, perhaps, that people sometimes do not know what their grounds are. It would take more argument to show that they cannot know them. Perhaps, with suitable reflection, the photo sorter could figure out her grounds. It must granted, however, people are unlikely to figure this out on their own, and it is difficult to see

19 20 21

22

Richard Fumerton, Metaepistemology and Skepticism (Lanham, M D : Rowman and Littlefield, 1995), p. 159. Hilary Kornblith, "Roderick Chisholm And The Shaping of American Epistemology," Metaphilosophy 3 4 (2003). As Kornblith acknowledges in a footnote, this experimental setup does not fit well with the claim that people are accurate in their judgments about the relative happiness o f the people in the photos. Since the experimental photos are manipulated versions of the same photo, the people in the photos are equally happy. Observers who rate them as unequally happy are therefore always making a mistake. Still, we can grant that people are typically able to make these sorts o f judgments accurately. For simplicity I will assume that the person's belief is that one person is happy and the other is not, rather than the comparative belief that one is happier than the other. Nothing turns on this simplification.

154

Critiques of Epistemic Externalism

why one would hold a theory requiring that actual justification turns on this rather remote ability. Second, it may be that the cause of the photo sorter's belief is the shape of the mouth. But it is not so clear just what her grounds are. There is a question about how precisely to describe the grounds that justify the belief in the example. Just as people are able to identify other people on sight without being able to articulate their distinguishing features, the research Kornblith describes suggests that people are able to assess others for happiness without being able to tell just what it is that distinguishes the people who appear happy. The happy people did "look happy" to the subjects, even if they are unable to say what looking happy consists in. This more generic description—looking happy—is presumably one that the subjects could rather easily identify as their reason for thinking that the person pictured is happy. Thus, this example yields a good objection to C G R theories only if the actual grounds are the more specific appearance of the mouth, and not the more general "looking happy". There's little reason to saddle the theories with the account of grounds that undermines the theory. This example, I conclude, is less than decisive as an objection to C G R theories. Whatever the implications of examples such as the ones to which Kornblith calls attention, C G R theories impose the demanding requirement that justified believers always be able to figure out that their evidence supports their beliefs. In Fumerton's terms, they always have access to the linking principles. Even though the example just discussed does not decisively refute such theories, this requirement strikes me as excessive. I have no conclusive argument for this position. It seems to me that unsophisticated people can have justified beliefs as a result of perception or as the result of inductive inference. I don't grasp the sense in which they must have access to principles governing such matters. Unless "access" is understood in a way that allows that people have access to a lot more than they are able to think of or articulate, the access requirement is too strong. Even considerable reflection might not yield the desired principles. Perhaps it is consistent with this that they still "can" formulate the principles and that they still have "access" to them. If so, then the knowledge and access would only materialize under fairly remote circumstances. I do not see why actual knowledge or justification depends upon this factor. This leads me to think the requirements of CGR theories are too demanding. The considerations of the past few paragraphs address the merits of C G R theories, not the contention in (INT4) that internalism is committed to the view that justified believers have some kind of access to linking propositions. I turn next to this issue. Must one endorse a linking belief requirement of some sort, and thus endorse a CGR theory, in order to be counted as an internalist? Is an SGR theory a version of internalism? It may be that the words "internalism" and "externalism" simply do not have clear enough meanings to enable resolution of the question of whether an SGR theory is a version of internalism. However, it is worth noting some good reason to count SGR theories as internalist. For one thing, they do make justification supervene on internal factors. More significantly, they do not make justification dependent upon contingent "natural" relations of the sort Armstrong had in mind when he coined the term "externalism." That is, they make justification dependent upon one's having good reasons, not

Search of Internalism and Externalism

155

upon contingent facts about causal relations or reliability. This, it seems to me, is a sufficiently telling point to warrant excluding such a theory as a version of externalism. Perhaps an SGR theory is best seen as a kind of hybrid. It makes one's reasons suitably internal, but it permits the connections between one's reasons and one's beliefs to be "external." They are external in that the justified believer need not have any grasp of them or access to them. But I think that it is worth resisting that classification as well. To see why, it will be useful to contrast SGR theories with what William Alston has defended as a hybrid theory, his "internalist externalism." 23 Alston's theory requires that the grounds that justify beliefs be suitably internal, just as SGR theories do. This is the internalist part of his theory. Alston also rejects the requirement that justified believers have access to any linking principles. Again, SGR theories concur. This is what Alston regards as the externalist feature of his theory. But further, and this is the crucial point, Alston holds that what justifies a belief must have some significant connection with the truth of that belief. As a result, he must reject the strong supervenience thesis that, on my view, is at the heart of internalism. Demon world examples make the point vivid. A person in the normal world can have a mental twin in a demon world. As I see it, internalism implies, quite correctly, that the person and the twin are justified in believing the same propositions. Alston's insistence on his version of the truth connection requires him to reject this plausible assumption. 24 This, I think, and not the rejection of the linking belief requirement, is what really makes his theory externalist. In contrast to Alston's theory, SGR theories (as I am understanding them) are committed to the strong supervenience thesis. Alston seemed to find such a theory to be a peculiar kind of theory because he took internalism to be the view that all justifiers are internal things and he took the linking facts to be justifiers. Thus, if my feeling of warmth contributes to my being justified in believing that I am warm, then there are two justifiers present: the feeling and the fact that the feeling justifies the belief. Alston took internalists to be committed to the view that justified believers have access to all justifiers. He rejected the requirement that one have access to the linking fact. But he also took the linking fact to be a contingent fact. This is the truly externalist feature of his theory. His theory implies that two believers can be internally alike, in that they have the same reasons to believe the same thing. Yet one is justified in believing that proposition and the other is not. And the reason for this is the fact that one of the people is in a situation in which that reason is regularly connected to the truth of the belief and the other is not. In contrast, SGR theories hold that what's a good reason for one is a good reason for the other. Alston's requirement for a truth connection imposes something closely resembling the contingent natural relations that the paradigm externalist theories—the causal theory and reliabilism—imposed. In my view, it is this fact, and not the fact that the believer need not have access to the linking fact, that makes his theory a kind of externalism.

23 William Alston, "Internalism and Externalism in Epistemology" and "An Internalist Externalism," reprinted in William Alston, Epistemic Justification: Essays in the Theory of Knowledge (Ithaca: Cornell University Press, 1989), pp. 185-245. 24 See "Internalism and Externalism in Epistemology," p. 223.

156

Critiques of Epistemic Externalism

SGR theories differ in this respect. This is why it is best not to see SGR theories as hybrids. Instead, it is best to see them as genuine versions of internalism. I conclude that (INT2), or perhaps (INT3), best expresses what internalism is. Externalism is the denial of these theses.

IV. Conclusion My goal in this paper has been to provide some basis for drawing a distinction between internalist and externalist theories in epistemology. Using the origins of the internalism/externalism debate as my guide, I have argued that internalism is best seen as the thesis that epistemic justification is a necessary condition for knowledge and that it strongly supervenes on internal mental states. Internalism so construed is consistent with the fact that people can be justified in believing things without having access to principles linking their evidence to their beliefs. Thus, for example, a person can have strong inductive evidence for a conclusion, and be justified in believing that conclusion, without having access to principles about inductive reasoning. I have not attempted to defend internalism in this paper, though I do believe that it is correct. 25 One crucial element of any such defense is defense of the claim that justification is necessary for knowledge. I believe that it is, and I think that any inclination to the contrary rests on mistakenly taking literally uses of the word "knows" as it appears in sentences such as "The trees know that spring has arrived." 26 Another central part of my defense of internalism relies on the idea mentioned near the end of Section III, in connection with Alston's views. It is that the conviction that believers who have the same evidence are alike with respect to justification. I see no plausible basis for differentially evaluating, with respect to justification, the beliefs of a person in a normal world and his demon world twin. I assume that it would be mad (and unjustified) for any of us in a normal world to believe that our experiences are the result of the efforts of a demon or a diabolical neuroscientist. It would be equally mad (and unjustified) for the victims of a demon or a diabolical neuroscientist to have those beliefs. They are justified in many of their ordinary beliefs about the world, just as we are. Or, if skeptical arguments are more successful than I think, they are not justified in their ordinary beliefs, but neither are we. Of course, we can differ with respect to knowledge, warrant, and well-foundedness. But internalism does not claim that these are internal matters. It is a thesis about justification.

25

I do say more in defense of internalism in "Internalist Epistemic Evaluations" and in "Internalism Defended" (written with Earl Conee). 26 Earl Conee and I discuss this point in "Making Sense of Skepticism," in Evidentialism (Oxford: Oxford University Press, 2004).

Inferential Internalism and the Presuppositions of Skeptical Arguments RICHARD FUMERTON

The key to winning the war against skepticism is negotiating favorable rules of engagement. Frantic jockeying for position is not new, of course. The attempt to saddle one's opponent with burden of proof is as old as philosophical dialectic. The epistemic conservative, for example, assigns a prima facie probability to a proposition merely because it is believed.1 Given such a view the fact that we believe propositions under skeptical attack comes in more than a little handy when we try to refute the skeptic. In the last few decades, the focus has been on metaepistemology in an attempt to gain the high ground in the battle over skepticism. Specifically, naturalistic and externalist analyses of epistemic concepts provide ways of undercutting classical skeptical arguments by undercutting the implicit metaepistemological presuppositions on which they relied. In this paper I will focus on a view that I call inferential externalism. I will begin by defining the controversy between inferential internalists and externalists. After explaining why the issue is so important when it comes to evaluating traditional skeptical arguments, I examine arguments for and against inferential internalism, and conclude by briefly discussing the resources available to the inferential internalist in combating skepticism.

The Inferential Internalism!Externalism

Controversy

The internalism/externalism controversy is notoriously difficult to define. I have argued elsewhere that it is in the end pointless to suppose that there is some one best way to understand the debate.2 The metaepistemological task is surely only to distinguish and evaluate as many interesting controversies as we can. At least some philosophers seem to understand the internalism/externalism debate as essentially a debate over what kind of access one must have to epistemically relevant conditions before those conditions can constitute justification. Thus in his well-known critique of traditional foundationalism, Bonjour seemed to argue that some set of conditions X can justify S in believing Ρ only if S is aware that X obtains and that X makes likely the truth of P.3 One must be care-

1

2 3

Chisholm defends a version of this view in his third edition of Theory of Knowledge, (Englewood Cliffs: Prentice-Hall, 1989), pp. 63-63. For a convincing refutation of epistemic conservatism see, Richard Foley's The Theory of Epistemic Rationality (Cambridge, Mass.: Harvard University Press, 1989), Chapter 7. See my Metaepistemology and Skepticism (Lanham: Rowman and Littlefield, 1995), Chapter 3. Laurence Bonjour, The Structure of Empirical Knowledge (Cambridge: Harvard University Press, 1985), pp. 30-33.

158

Critiques o f Epistemic Externalism

ful lest the statement of such a view becomes literally unintelligible. If one maintains that for any set of conditions X that one proposes as constitutive of S's justification for Ρ those conditions, to be successful, must always be fortified with some other set of conditions describing S's access to X, then the view is hopeless. Call the satisfaction of access conditions to X, A. Will X together with A constitute justification for S to believe P. Of course not. Our strong access requirements require access (call it A') to the new proposed sufficient conditions for justification X and A. But the conjunction of X, A, and A' will still not constitute S's justification for believing Ρ as the view requires access to these three conditions, and so on ad infinitum. To avoid this problem, the strong access internalist must distinguish carefully a view about what is constitutive of justification from a view about what is necessary for justification. If the view is to be intelligible the access internalist must argue that when some set of conditions X constitute S's justification for believing Ρ those conditions will be such that they entail that S has access to them. The access, however, need not be part of what constitutes the justification. An analogy might be helpful. Ρ cannot be true unless it is true that Ρ is true—P's truth entails (in some sense of entails) that it is true that Ρ is true. But it would be a serious mistake to argue that P's truth is constituted by its being true that Ρ is true. The correct analysis of what it is for Ρ to be true should not make reference to metatruths about P's truth even if the correct analysis of P's being true must reveal why P's being true entails that Ρ is true. The access in strong access internalism can, of course, be understood in many different ways. The strongest view requires actual introspective, or at least noninferential, access. More modest access internalisms require only potential access (though still usually potential introspective or noninferential access). The potentiality in question can itself be understood in many different ways ranging from logical, to lawful, to causalrelative-to-circumstances. Now even if access internalism involves no conceptual regress, the view strikes me as highly problematic. Actual access internalism would require one to have an infinite number of ever more complex beliefs in order to justifiably believe anything. Potential access internalism would require the possibility of acquiring an infinite number of ever more complex beliefs in order to justifiably believe anything. If the potentiality is logical possibility, that might not present an insuperable problem, but by the same token access internalism would lose all of its philosophical bite. I can't even think of an externalist view of justified belief that would deny that it is always logically possible to access the conditions that constitute justification. But if the potentiality is some species of nomological possibility, it seems to me wildly implausible to suppose that we have the capacities that would be required. Let us call the kind of actual and potential access internalism described above, comprehensive access internalism. As I indicated in the introduction, I am interested in examining a more restricted thesis of access concerning justification. The thesis is restricted to inferential justification and asserts that in order to be inferentially justified in believing one proposition Ρ on the basis of another Ε one must be justified in believing that Ε confirms or makes probable P. Most externalists will allow that to be justified in believing Ρ on the basis of Ε one must be justified in believing E. Even hard-core reliabilists, for example, tend to embrace this claim—epistemic garbage in,

Inferential Internalism and the Presuppositions of Skeptical Arguments

159

epistemic garbage out, even if the process generating the output is itself impeccable with respect to taking one from truths to truths. Requiring access to the legitimacy of the inferential connection (the reliability of the process in the reliabilist's lingo), however, is almost universally rejected by externalists and naturalists in epistemology. Reliabilists explicitly deny that one needs access to the reliability of the belief-producing process even when the input takes the form of other justified beliefs. Epistemologists who understand knowledge or justified belief in terms of beliefs that tracks facts, even when the tracking requires the use of other beliefs, deny that one needs any epistemic access to the existence of the tracking relations.

Inferential Internalism and Traditional Skeptical Arguments It seems obvious that inferential internalism was explicitly or implicitly taken for granted by most of those in the history of philosophy who either argued for skepticism or took skepticism seriously. Consider Hume's famous argument for skepticism with respect to the senses. After rejecting epistemological direct realism, Hume argued that to be justified in believing anything about the external world we would need to justifiably infer the existence of enduring, perception-independent objects from the occurrence of fleeting, subjective perceptions. He then went on to argue that since we can't acquire any justification for believing that perceptions are reliable indicators of external objects, we have no justification for believing propositions asserting the existence of physical objects: But as no beings are ever present to the mind but perceptions it follows that we may observe a conjunction or a relation of cause and effect between different perceptions, but can never observe it between perceptions and objects. 'Tis impossible, therefore, that from the existence or any of the qualities of the former, we can ever form any conclusion concerning the existence of the latter, or ever satisfy our reason in this particular. 4

It is clear that Hume simply took it to be obvious that in the absence of evidence indicating a connection between perceptions and objects reason cannot sanction inferences from the former to the latter. It is not much of an overstatement to suggest that a good part of the history of epistemology in the last few centuries has been an attempt to respond to the kind of argument Hume presented on Hume's terms. Descartes sought proof of a benevolent and omnipotent God to license the inference from mental representations to objects represented. Berkeley, Mill, and twenty-century phenomenalists tried to reduce physical object talk to talk about perceptions in an effort to make the epistemic transition easier. Locke and Russell (in at least one of his philosophical incarnations) sought to bridge the inferential gap through reasoning to the best explanation. But they all agreed with the legitimacy of Hume's challenge. Find a way of justifying

4

David Hume, A Treatise of Human Nature, ed. Selby-Bigge (London: Oxford University Press, 1888), p. 212.

160

Critiques of Epistemic Externalism

belief in the legitimacy of problematic perceptual inferences or abandon any claim to rational belief in conclusions that can only be justified through those inferences. Now I'm not primarily concerned with the question of whether one can thwart skepticism after accepting inferential internalism (although I will make a few brief concluding remarks which bear on the question). It would obviously be easier, however, to have inferentially justified belief if one embraced inferential externalism. In fact, unless there is a fairly wide assortment of necessarily true epistemic principles, knowable a priori, I suspect that inferential internalism will win the day for skepticism. If one is a foundationalist, it depends in part on how broad one can get one's stock of noninferentially justified beliefs, something that obviously depends on one's analysis of noninferential justification. Consider, as an example, however, just one potential problem, the problem of justified belief about the past. If beliefs about the recent past are inferentially justified by present "memory" states, how exactly would one begin to justify belief in the reliability of memory? Virtually every hypothesis one could even entertain would itself inevitably rely for its proof on our remembering something, but inferential internalism implies that without access to the reliability of our memory inferences, no beliefs based on memory are justified. Without an epistemic principle sanctioning memory that is knowable a priori, the inferential internalist seems to be stuck in an all too fleeting present.

Arguments for and Against Inferential

Internalism

But why accept inferential internalism, especially in light of the difficulties it will present in meeting the skeptical challenge. Indeed, the ease with which one can develop skepticism given the view might be taken to constitute a reductio of the position. Those who took inferential internalism seriously have labored long and hard to solve the problem of perception, the problem of induction and the problem of memory, and with a singular lack of success. The philosophically unsophisticated will look at you as if you have gone mad when you even try to explain the justification you seek in support of the legitimacy of such fundamental reasoning. Given that such people seem not the least bit inclined to abandon their claim to have justified beliefs about the external world, the future, and the past, despite the blank stares they offer in response to a request for evidence indicating the legitimacy of their inferences, should we not at least suspect that the ordinary understanding of justification requires nothing as strong as what is proposed by the inferential internalist? Maybe so. But there is also prima facie evidence to indicate widespread acknowledgment of the legitimacy of the inferential internalist's standards for inferential justification. Consider the astrologer who predicts the future based on beliefs about the positions of planets relative to one another. It seems obvious to me that a sufficient condition for rejecting astrological inferences as rational would be that the astrologer has no reason to believe that the positions of planets relative to one another has any connection to the truth of his predictions. To justifiably infer that a solution is acid from the fact that litmus paper turned red in the solution one must surely be justified in believing

Inferential Internalism and the Presuppositions of Skeptical Arguments

161

that litmus paper's turning red in a solution is a reliable indicator that the solution is acidic. If a detective infers from someone's appearance that the person is guilty of a crime, one will surely demand evidence for supposing that appearances of this kind are correlated with guilt before one will concede the rationality of the conclusion. These commonplace examples and indefinitely many others like them surely indicate that we do in fact embrace the inferential internalist's account of what is necessary for inferential justification. We may pick and choose when we decide to make an issue of someone's lacking reason to believe in an inferential connection. There is at least a grain of truth, a harmless grain, in the idea that justification is contextual. We do take for granted the justification of certain beliefs and the legitimacy of certain inferences in contexts in which we are concerned with the justification of other beliefs or inferential connections. Two theologians engaged in combat over some esoteric question concerning the motive of God in testing Abraham will probably take as unproblematic the evidence of statements in the New and Old Testament. When David Hume is trying to figure out the subject matter of ethical judgments he seems to take for granted our understanding and knowledge of the distinction between internal and external reality. In explaining his moral theory in the Treatise he even seems to suppose that we know that some qualities are primary while others are secondary.5 And he does all this despite the fact that in other contexts where he is primarily concerned with this putative knowledge he rejects even the intelligibility of the distinctions he uses elsewhere. Perhaps, then, we give ourselves the legitimacy of our commonsense inferences in precisely the way in which theologians give themselves the legitimacy of the bible as a source of truth. Once we actually turn our attention to the philosophical task of justifying belief in what we take for granted we must apply the same standards that we insist on applying to astrologers and fortune tellers, just as once the atheist raises questions about the veracity of the bible literally interpreted, the theologian has no choice but to accept the criteria for rational belief partially captured by inferential internalism. How might the inferential externalist respond to all this? Often when I discuss with externalists the example of the astrologer's inferences or the scientist who relies on the color of litmus paper to reach conclusions about solutions containing it, they claim that I am confusing conditions necessary for justification with conditions necessary for metajustification. To be justified in believing or asserting that one has an inferentially justified belief in Ρ on the basis of E, they will concede, one must be justified in believing that Ε makes likely P, but awareness of inferential connections is not required to be justified in believing Ρ on the basis of E. The distinction between being justified in believing Ρ on the basis of Ε and being justified in believing that one is justified in believing Ρ on the basis of Ε is, of course, legitimate. I argued earlier that even the comprehensive internalists ought to recognize that second-level justification is not constitutive of first-level justification, Moreover, I have expressed strong reservations about the claim that first-level justification even entails actual or potential second-level justification (in any interesting mode of potentiality). Nevertheless, the response to the above criticism is in a way simple. I am not

5

Hume, op. cit., p. 469.

162

Critiques of Epistemic Externalism

confusing being inferentially justified in believing Ρ on the basis of Ε with being justified in believing that one is inferentially justified in believing Ρ on the basis of E! The thought experiments employed were mine and I know what I wanted you to consider in evaluating the thought experiments. I meant just what I said when I suggested that you would reject as irrational the astrologer's predictions about the future based on the position of planets for the reason that the astrologer has no justification for supposing that there is a connection between the two states of affairs. My claim was that the absence of justification for believing that there is a probabilistic connection is a sufficient condition for the astrologer's belief being unjustified. Even if one were to discover at some future date that there is some amazing connection between planets, birth dates, and the lives of men and women, we would still never concede that the astrologer's inferences were rational unless we became convinced that the astrologer had some reason to believe that those connections held. Now I am, of course, making claims about what you would say upon considered reflection and I might be wrong. If you think that the rationality of the astrologer's inferences depends on nothing more than the mere existence of regularities between planets and lives of people then you should reject the above appeals as having any significance for the initial plausibility of inferential internalism. If, on the contrary, you feel the force of the thought experiments you must look elsewhere to escape acceptance of the view. A more plausible move for the inferential externalist in dealing with the above examples would be to distinguish importantly different kinds of evidence. As we use the term "evidence" we certainly do characterize litmus paper's turning red in a solution as evidence that the solution is acidic. The approach of very dark clouds is evidence of an imminent storm. Footprints on a beach is evidence that someone has walked on the beach recently. But it is surely obvious upon reflection that one's evidence for believing that the solution is acidic is not the color of the litmus paper by itself. One does not infer that there is about to be a storm from the knowledge in isolation that there are dark clouds. And without the appropriate background knowledge one would not infer that people walked a beach recently after observing footprints on the beach. The "inferences" in question, or more accurately, the underlying arguments supporting the beliefs in question, are enthymematic, and are almost certainly enormously complex. The color of the litmus paper, the dark clouds, and the footprints are only "evidence" for the respective conclusions in a highly derivative, potentially misleading sense. There are two different ways in which one might try to make this point clearer. One approach is to distinguish derivative from fundamental epistemic principles. Derivative epistemic principles (e.g. the principle that one can reasonably infer that a solution is acidic from the fact that litmus paper in the solution turns red) are derivative precisely in the sense that to justifiably use them to reach a conclusion, one must have inferential justification for supposing that the principle is true. We must be able to derive the reliability of the litmus test before it can be used to generate rational belief. It also seems obvious that these epistemic principles I am calling derivative are straightforwardly contingent, empirical truths, although as we shall see later this feature of derivative epistemic principles distinguishes them from non-derivative epistemic principles only given certain metaepistemological views.

Inferential Internalism and the Presuppositions of Skeptical Arguments

163

If an epistemic principle is derivative in virtue of the fact that it can be rationally employed only when one has inferential justification for accepting it, then the most natural move would be to define fundamental, non-derivative, epistemic principles as those one can legitimately employ, whose employment can yield justified beliefs, even if one can appeal to no evidence from which one can infer the truth of the principle.6 Rather than distinguish two sorts of epistemic principles, derivative and fundamental, it might in the end be clearer if we simply refuse to apply the label "epistemic principle" to such rules of evidence as the litmus test for acid, fingerprint tests to resolve questions concerning identity of persons, observation of dark clouds to predict storms and the like. Instead, one might reserve the term "epistemic principle" exclusively for what the first approach calls a fundamental epistemic principle. We don't so much derive, the proponent of this approach might argue, secondary epistemic principles which we then use in lieu of fundamental epistemic principles. Rather the so-called secondary epistemic principle is just a premise to be used against a complex of background information, which premise would need to be justifiably believed if we are to use it to gain inferential justification. Our justification for believing that the solution was acidic really involved an implicit inference from the entire complex body of evidence from which one could legitimately infer (according to genuine epistemic principles) the conclusion in question. In the end the two approaches may just differ terminologically. Whether one dignifies the assertion of a probabilistic connection between red litmus paper and acid with the term "epistemic principle", quickly adding that the principle must itself be inferentially justified, or one views the proposition describing a probabilistic connection as simply another premise which one must be inferentially justified in accepting before one can use it to gain justified belief in some conclusion, one is recognizing that the epistemologist must recognize epistemic principles which are in some sense more fundamental than the derivative principles. In any event the point of making the distinction between fundamental and derivative epistemic principles, or if you prefer between genuine and spurious epistemic principles, was to offer the inferential externalists a way of avoiding arguments designed to establish the prima facie plausibility of inferential internalism. The inferential internalist, you will recall, pointed out that we would never allow for the possibility that our astrologer had inferentially justified astrological predictions unless we were convinced that the astrologer had reason to believe that there were connections between positions of planets and the affairs of humans. True enough, the inferential externalist could now concede, but irrelevant to the truth of inferential internalism. All the example shows is that no-one would operate on the supposition that the astrological inference in ques-

6

It's obvious that I am presupposing that epistemic principles have can have a truth-value. If one takes seriously certain claims about the normativity of epistemic concepts and one also takes seriously noncognitive views about the normative, one might, I suppose, insist that epistemic principles take the form of imperatives, imperatives which are neither true nor false, but the following of which define the acceptability of moves in the epistemic game (just as, for example, the imperatives governing the movement of chess pieces define acceptable moves in the game of chess).

164

Critiques of Epistemic Externalism

tion was non-en thymematic. Even the astrologer, we are supposing, would regard the principles of astrological prediction as either those derivative epistemic principles we discussed above, or, more plausibly, as not being genuine epistemic principles at all. Propositions describing connections between planets and the lives of people would be viewed merely as premises to be used in inferring conclusions, and as we noted above, even inferential externalists typically allow that beliefs taken as input must be justified if the output they generate is to be justified. To plausibly argue for inferential internalism we should be looking at examples of fundamental epistemic principles, principles which are not inferentially justified, which cannot be viewed merely as conclusions which we accept as a result of some other mode of reasoning.7 At this point, however, the inferential externalist might owe us some kind of account of how one would recognize a fundamental epistemic principle. One can trot out examples, of course, beginning with everyone's favorites, the rules of deductive logic. But deductive reasoning by itself is never going to save us from skepticism with respect to the vast majority of what we believe. How shall we recognize fundamental non-deductive principles of reasoning? The difficulty is that for most externalists and naturalists, epistemic principles, principles licensing an inference from beliefs of one sort to beliefs of another are almost certainly going to be contingent empirical truths. The whole point of a reliabilist understanding of justified belief is that we are supposed to secure a connection between justified belief and true belief. Each individual justified belief might be false, but most justified beliefs will be true, and that because we will understand justified belief as belief resulting from processes that usually result in true beliefs (provided that the relevant input beliefs are true). 8 But it will always be a contingent feature of the world that, relative to an environment X, true beliefs of kind A taken as input and processed by the relevant software of the brain result in true beliefs of kind B. The connection between input beliefs and output beliefs sanctioned by an inferential rule is essentially a nomological connection. As such it looks suspiciously like the connection between litmus paper turning red in a solution and the solution's being acidic. As a matter of contingent fact, the occurrence of the one state of affairs might typically signal the occurrence of the other. But this would be only a causal connection which, one might suppose, we need to discover a posteriori before we would be entitled to employ the litmus test in arriving at conclusions.. By the same token, probabilistic connections as understood by externalists and naturalists are contingent nomological connections which it seems we should also need to discover a posteriori before we could be viewed as rational in implicitly following the relevant rule prescribing the inference in question. Again, how are we supposed to tell, according to the externalist

7

8

M y thanks to Mike Huemer for raising this suggestion initially in conversation and correspondence with me. H e subsequently published some of these ideas in Skepticism and the Veil of Perception (Lanham, M D : Rowman and Littlefield, 2001). This is, of course, a crude caricature of the view. More sophisticated versions of the theory will often relativize reliability of an environment. Furthermore, the analysis of reliability inevitably gets complicated and usually turns to counterfactuals. Once we arrive at a counterfactual analysis of reliability, however, it is hard to tell what logical connection there remains between beliefs being reliably produced and beliefs being generally true.

Inferential Internalism and the Presuppositions of Skeptical Arguments

165

or the naturalist when a contingently reliable inference is genuinely fundamental in a way which obviates any need for awareness of the reliability of the inference? Notice that the inferential internalist is likely to have a principled answer to the question of how to distinguish fundamental (or genuine) epistemic principles from derivative (or spurious) epistemic principles. The inferential internalist is likely to suppose that fundamental epistemic principles, deductive and non-deductive, are necessary truths knowable a priori. At least such a view is the inferential internalists best hope of ending a potentially vicious regress when it comes to inferential justification. There must be some probabilistic connections, according to the inferential internalist, which one knows without inference. Otherwise every inferentially justified belief would lead one to a regress of inferences designed to justify acceptance of inferential connections. Setting aside the above problem, however, let's look at an obvious candidate for a fundamental inference, a deductive inference. In the case of someone S who infers Ρ from Ε where Ε logically entails Ρ is the inferential internalist right in maintaining that in order for S to justifiably believe Ρ on the basis of E, S must be aware of the fact that (or have at least a justified belief that) Ε entails Ρ (the inference in question is legitimate). Frankly, the answer still seems to me obviously yes. We can easily imagine someone who is caused to believe Ρ as a result of believing Ε where Ε does in fact entail P, but where the entailment is far too complicated for S to understand. Unless S "sees" that Ρ follows from E, would we allow that the inference in question could generate justified belief? Now to be sure a great deal hinges on what is involved in someone's believing one proposition on the basis of another. Certainly, if S believes Ρ after he believes Ε because he read in the tea leaves that if Ε is true Ρ is true, then we don't really have a case of S's inferring Ρ from Ε at all. S is inferring Ρ via a chain of inferences—one from propositions describing tea leaves to a proposition asserting the conditional, if Ε then P; the other, inferring Ρ from Ε and if Ε then P. The inferential externalist can treat this situation just as he treated the astrological inference or the litmus paper inference. The inferential internalist, the argument goes, has misidentified the inference in question, and in so doing has misidentified the source of the intuition that a genuine inference from Ε to Ρ (where Ε entails P) requires awareness of the connection between Ε and P. But even if the inferential internalist must be careful in describing the relevant hypothetical situation, it still seems possible to describe a case in which S is caused by his justified belief in E, and Ε alone, to believe P, where Ε does entail P, but where S is completely unaware of the entailment. And I would still suggest that in any philosophically interesting sense of inferential justification, S would not be justified in believing Ρ on the basis of E. The inferential externalist has another move. He can argue that believing Ρ on the basis of Ε requires more than simply being caused to believe Ρ by a belief that E. The hypothetical situation described above might still fail to constitute a situation in which S genuinely infers Ρ from Ε where Ε entails Ρ but S has no justification for believing Ρ on the basis of E. There is, I think, some plausibility to the claim that inference involves more than mere causal connection between beliefs. But the source of this intuition will, I think, provide little solace for the inferential externalist. In fact, I suspect that we may not concede genuine inference unless there is veridical or nonveridical

166

Critiques of Epistemic Externalism

perception of a connection between the proposition from which Ρ is inferred and P. But this perception will be just what the inferential internalist claims is the awareness of the connection that is necessary for inferentially justified belief.9

Resources Available to the Inferential Internalist I have no doubt not persuaded many confirmed externalists that inferential internalism is true, but I wanted to close with a few thoughts about the resources available to the inferential internalist when it comes to justifying belief in fundamental logical and probabilistic connections. With respect to logical entailment, at least a great many philosophers have convinced themselves that one can have foundational, a priori knowledge that entailment exists. Potential regresses of justification in the case of deductive inference are cauterized by a priori knowledge of entailment. As I suggested earlier, it may be plausible to argue that there are probabilistic connections that can be known a priori in precisely the way that entailment is known a priori. If Keynes's view of the epistemic relation of making probable is correct, if one takes making probable to be a quasi-logical relation holding between propositions, then one can plausibly argue that one can know a priori that certain propositions make probable others.10 But that's a big "if". If fundamental non-deductive epistemic principles are contingent truths knowable only a posteriori, is their any hope of satisfying the skeptic from within the framework of inferential internalism? Epistemological externalists and naturalists will have no difficult allowing that one can use a mode of inference to justify the legitimacy of that very mode of inference. If the externalist seeks justification for believing the perceptual inferences are reliable, there is no reason, given the view, for disallowing the use of perception to discover its reliability. If a method of reasoning requires inferential justification, then the utter absurdity of supposing that one can legitimately use that method of reasoning to justify relying on it has always seemed to me a reductio of externalism and naturalism.11 In any event such an attempt is ruled out by the inferential internalist who will correctly point out that given his view we will have simply begun an endless and ultimately futile regress. One can, of course, attempt to use one method of reasoning to reach conclusions about the legitimacy of another method of reasoning. Thus philosophers have heroically tried to find a non-question begging, deductively valid proof for the legitimacy of deductive reasoning, and others have tried to use reasoning to the best explanation to justify reliance on inductive reasoning. Bayeseans try

9 I leave open the possibility, here, that a non-veridical perception can yield justified belief in a probabilistic or logical connection (just as hallucinatory experience can yield justified belief about the external world). Again, my thought on this matter is heavily influenced by correspondence with Mike Huemer. 10 See John Keynes, A Treatise on Probability (London: Macmillan, 1921) 11 This is one of the fundamental claims I argue in Metaepistemology and Skepticism, though I now see that I didn't always put the point carefully enough. Notice, that here I did qualify the claim attributing it only to methods of reasoning that require inferential justification.

Inferential Internalism and the Presuppositions o f Skeptical Arguments

167

to use Bayesean reasoning to justify virtually every kind of non-deductive inference we employ. But again, the inferential internalist will argue that we are only delaying the inevitable. We will ultimately get to some principle of reasoning which we must regard as knowable a priori, or we will face a vicious regress. So are we inferential internalists stuck with a choice between allowing only deductive inference (and the inevitable skepticism that will entail) and a Keynesian conception of probability? There is one, somewhat pathetic, route to at least continuing a conversation with the skeptic. I said that one can't use a method of reasoning to generate justified reliance on inferences of that sort. But one can use a method of reasoning to show that the method of reasoning is illegitimate. Consider the following example which I have appealed to in a number of different contexts. When I was a kid my dad bought me one of those magic eight balls. You ask the eight ball a question, shake the ball, and an answer floats to a transparent opening in the surface. Now imagine a society as devoted to eight-ball reasoning as we are to inductive reasoning, or perceptual reasoning. Imagine too that eight-ball reasoning is used only to discover the truth about matters to which we have no independent access, questions, for example, about the existence of God or the existence of an afterlife. Our hypothetical eight-ball reasoning society is, however, plagued by a few philosophical skeptics who keep insisting that eight-ball reasoning without some justification for accepting its reliability, yields no knowledge or justified belief. Unfazed, our eight-ball reasoners stare at their eight ball, ask whether eight-ball reasoning is legitimate, shake the ball, and get the answer "Yes." Well the skeptic won't and shouldn't buy this for a moment as a way of settling the issue. But the skeptic should, I think, concede that the eight-ball reasoner is at least better off than had the answer "No" bubbled to the surface. At least eight-ball reasoning isn't having internal difficulties. There are other ways in which a kind of reasoning can get itself in trouble. Suppose the eight-ball reasoners are trying to discover whether or not there is a God. They shake the ball and get the answer "No". They shake it again and get the answer "Yes". They shake it yet again, and get the answer "Maybe." As they continue to ask the question and shake the ball they get a more or less random string of "Yes"'s, "No"'s, and "Maybe"'s. Again, eightball reasoning has fallen apart internally. Now at least some of our fundamental belief-producing processes seem to fare relatively well when it comes to internal consistency. I seem to remember putting my keys on the mantelpiece now. A few moments from now, I also seem to remember seeming to remember putting the keys on the mantelpiece and I still seem to remember putting the keys there. The process can be indefinitely repeated. Am I using memory to justify relying on memory? I don't think so, but at least I'm not using memory to throw a monkey wrench into my conviction that memory is reliable, and in the end perhaps that is all we can do in the face of an ultimately successful skeptical attack. 12

12

Although I've already acknowledged his contribution to specific points discussed in the paper, I'd like to again express in a general way appreciation to Mike Huemer. M a n y o f the ideas discussed in this paper were developed in the course of correspondence with him, particularly the sections of the paper which explore objections to the second clause o f the principle o f inferential justification.

Knowledge, Justification and the Cooperative World KEITH LEHRER AND DAVID A . TRUNCELLITO

/ . Introduction Traditionally, accounts of epistemic justification (hereafter, simply 'justification') have fallen into one of two camps. Internalist theories have maintained that all of the factors relevant to justification are factors which are internal to the epistemic agent, such as her other beliefs, her perceptual states, or her reasons. Externalist theories, on the other hand, have maintained that at least some of the relevant factors are external to the agent, such as her causal history, her environment, or certain features of her cognitive apparatus. 1 Here, we shall argue that reflection upon the notion of epistemic justification suggest a different kind of theory. Justification is, in some broad sense of the term, a normative or evaluative notion. What is distinctive about extant accounts is that they restrict their evaluations to the agent. We want to suggest a dynamical interactive model of justification, one which evaluates not only the agent but the world in which she operates. The goal of our epistemic practices is often taken to be the obtaining of true belief (as well as, perhaps, the avoidance of false belief) 2 . This can be construed as the attempt to generate a "fit" between what one believes and the way the world is. And this, in turn, suggests that justification is a measure of whether this fit is sought in the right way. But whether a fit between the agent and the world is sought in the right way depends upon two things: the agent and the world. Therefore, the correct account of justification must evaluate both the agent and the world; it must be a "Janus-faced" theory, which looks with one face to the agent and with the other face to the world, evaluating both vis-ä-vis their role in the pursuit of knowledge. Knowledge, then, is a cooperative enterprise between epistemic agents and their epistemic environment; only if both work properly will knowledge obtain. While this is a rather evocative way of putting the point, it may be somewhat misleading, in that it seems to attribute agency to the world. So, let us put the point as it was put in Lehrer (1990): "To obtain knowledge we need the right mix of internal and external factors." 3

1

2 3

If you like, you can envision all such accounts as falling along a spectrum, with pure internalism at one end and pure externalism (the view that all factors relevant to justification are external to the agent) at the other end. This may be somewhat over-simplistic, and it may not completely capture our epistemic goals, but surely the formation of true belief is at least a primary goal of our epistemic practices. p. 151.

170

Critiques of Epistemic Externalism

II. The Concept of Epistemic Justification Let us begin, then, by turning to the very concept of epistemic justification. Rather than offering an account of the conditions under which a belief is justified, we want to understand what it is for a belief to be justified. It seems that there are three components to the concept of epistemic justification: goal-directedness, truth, and normativity. The correct analysis of the concept, then, will capture the relationship between these three components.

II.i Goal-Directedness We take it that any concept of justification—e.g. epistemic justification, moral justification, economic justification, pragmatic justification—essentially depends upon goaldirectedness. That is, an action (or belief, desire, etc.) can only be justified relative to some goal or other. To determine whether an action is justified, we must determine to what end that action was performed; then we can proceed to evaluate that action vis-ä-vis that end. For instance, if someone asks whether she would be justified in robbing a bank, we must ask what end she is aiming to attain. If her goal was to attain a sum of money, then the action is justified; if her goal was to do what is morally right, then her action is unjustified. To put the point another way, we can say that her action is economically justified but morally unjustified; we can evaluate the justificatory status of an action along several dimensions, insofar as the actor is working towards several ends. The same can be said of belief. A belief might be justified in one respect but unjustified in another. For instance, the mother of a son who has not returned from war might be epistemically justified in believing that he has died—the evidence in support of that belief is substantial—but prudentially justified in believing that he is alive— the hope, however slim, that he will someday return would bring her some comfort, and coping with the realization that he has been killed would be far too difficult for her. Again we evaluate justification along several dimensions, depending on the ends towards which the belief in question might be adopted. However, we are concerned here with the question of whether a belief is epistemically justified, that is, justified for the purposes of obtaining knowledge, without consideration of its justificatory status along other dimensions.

II.ii Truth What, then, is the relevant goal when we speak of epistemic justification? Truth. When we evaluate a belief to determine whether or not it is epistemically justified, the relevant goal towards which this belief is directed is truth. As we have suggested above, beliefs might be adopted as a means towards satisfaction of other ends—e.g. economic gain, happiness, well-being—but these are irrelevant to the epistemic justificatory status of those beliefs.

Knowledge, Justification and the Cooperative World

171

Let us begin by simply noting that most epistemologists will take this point to be relatively uncontroversial, even if they agree about little else with regard to epistemic justification. To consider a few examples, here is Lehrer, whose account of (personal) justification is based on the internal state of acceptance: "There is a special kind of acceptance requisite to knowledge. It is accepting something for the purpose of attaining truth and avoiding error with respect to the very thing one accepts. More precisely, the purpose is to accept that ρ if and only if p." 4 And Alvin Goldman, reliabilist and externalist, tells us something quite similar: "true belief is a prime determinant of intellectual value, and in particular, a critical value for justifiedness"; indeed, Goldman tells us that this claim is "unremarkable".5 As one final testament, William Alston, in his "Concepts of Epistemic Justification", defines the "epistemic point of view" as "the standpoint of the aim of maximizing truth and minimizing falsity."6 And these authors are not alone: there is widespread agreement that truth is the relevant goal when it comes to determining whether a belief is epistemically justified, even though this agreement is accompanied by broad disagreement as to the exact nature of the relationship between justification and truth. One might simply argue that there are, as we have noted above, a number of ends towards the satisfaction of which a belief might be adopted, and that picking out epistemic justification is simply a way of indicating that knowledge is the intended goal: "That's what makes it epistemic justification, after all." If epistemic justification is conceptually related to knowledge—and, it is widely agreed that justification is a necessary condition of knowledge—then truth, being the other necessary condition besides belief, seems to be the relevant end of justification. If knowledge is justified true belief plus a qualification7, then the relevant goal in determining whether a belief is epistemically justified should be justification, truth, belief, or some combination. But justification cannot be the relevant goal: to say that a belief is justified if and only if it is justified, is to do no more than to utter a tautology. Nor can belief be the relevant goal: to say that a belief is justified if and only if it is a belief, is not only misguided but false: there are some beliefs which are unjustified. This leaves us with truth as the component of knowledge which is relevant to epistemic justification. However, it would be incorrect to maintain that a belief is justified if and only if it is true. This would ignore what is an important fact about justification and truth, namely, that a person may fail to be justified in believing something, even if it is something true, because he believes it for the wrong reasons. The relation between truth and justification cannot be the identity relation. However, justification must be analyzed in terms of truth to sustain the connection with knowledge. Other concepts—goal-directedness toward truth seems a promising candidate—must be invoked to explicate the relation between justification and truth, of course; but the fore-

4

Lehrer (1990), p. 11.

5 6 7

Goldman ( 1 9 8 6 ) , p. 98. See pp. 84-5. Gettier-type cases suggest the need for a fourth condition. The right kind o f match between internal factors and external factors may provide us with a prophylactic against Gettier cases, but it may be imperfect.

172

Critiques of Epistemic Externalism

going reflections persuade us of the necessity of maintaining a conceptual connection between justification and truth in our analysis.

Il.iii Normativity So, epistemic justification is partly a matter of aiming at the goal of truth. But it is more than that—one can aim at truth without being justified in what one believes, if for instance one's way of aiming at the truth employs the wrong reasons and is not an acceptably right way of pursuing truth. This suggests that there is something more to justification than merely aiming at truth; one must aim at truth in the right way. That is, justification includes a normative component. Justification is goal-directed, and consequently we must consider both the goal or end and the means taken towards that end. In the case of epistemic justification, the goal is truth and the means is the way we form and evaluate our beliefs. So, our analysis of epistemic justification should make reference to both the goal of truth and the way in which we conduct ourselves in pursuit of that goal. To put the point another way, truth is the component of knowledge which ensures that we get the right result; justification is the component of knowledge which ensures that we get the right result in the right way. Rather than focusing solely on the goal of truth, then, let us consider the method of seeking truth. An initial, though inadequate, account based on the psychology of the subject is the following: J l : S's belief that ρ is justified iff S formed or evaluated the belief that ρ by aiming at truth or, J l ' : S's belief that ρ is justified iff S's belief that ρ is truth-directed. We have seen that the concept of justification must invoke some concept other than the mere truth of S's belief, and we have suggested truth-directedness as the link between justification and truth. This formulation has the advantage of emphasizing that justification requires that a subject aim at truth without adding any additional conceptual constraints. However, this formulation seems insufficient as an account of justification. For, S might aim at the truth in an illegitimate, improper or unreasonable way, and if she does, then she is not epistemically justified even if she succeeds in obtaining a true belief by her faulty practices. We are concerned to ensure that justification is not the result of luck or otherwise faulty practices. Our initial formulation above insufficiently constraints the way in which S aims at truth. Merely aiming at truth will not suffice; if justification is to be a positive term of appraisal, then it must only apply to beliefs which are formed or evaluated in legitimate and proper ways in order to attain the goal of true belief. Not all beliefs aimed at truth are justified, even if they hit the target.

Knowledge, Justification and the Cooperative World

173

We want to argue that any purely psychological formulation—i.e., any formulation which fails to consider the right way to aim at truth—will fail to capture the concept of epistemic justification. We do want to acknowledge truth-directedness as a necessary condition on justification, even though it is not sufficient. That is, while not all beliefs aimed at truth will be justified, it is certainly true that no beliefs which fail to be aimed at truth will be justified. So, we have at least a partial analysis; we must now supplement it in the right way. The relevant further condition concerns the way in which we strive towards the goal of truth. Roughly, then, something very close to the following must be right: J2: S's belief that ρ is justified iff S arrived at the belief that ρ by seeking truth in the right way We have suggested that the connection between justification and truth (or, properly speaking, truth-directedness) must be in terms of approaching the goal of truth in the right way. Many accounts of justification have been offered as attempts to explicate this notion of "in the right way", and some try to reduce it to naturalistic terms. Indeed, naturalized epistemology has been in vogue for the last quarter-century or so. It might turn out that an account of justification can be offered in strictly naturalistic terms; what is important, though, is that it respect the correct conceptual analysis of justification. And that analysis, we shall argue, essentially invokes a notion of epistemic value. 8 Here is the case for inclusion of a normative component in our analysis of epistemic justification. The goal of justification is truth, as many have claimed, but justification is achieved only if the goal is of truth is pursued in the right way. So, we must have the combination of the goal of truth with a way of achieving it; and the two must interact. Each is essential to justification. Justification must be aimed at the right goal in the right way. Normativity is coiled at the very heart of justification. The naturalistic move, then, is half the story. The right goal is truth, and, assuming a naturalistic reduction of truth, that part of justification is naturalistic and externalistic. But the right way of achieving a goal is irreducibly normative. At the level of conceptual analysis, we must respect the observation that justification is a term of approbation, and that approbation accrues to beliefs which are arrived at or evaluated in the right way. We need not say, for the sake of conceptual analysis, which processes, practices or methods are the right ones. This is paradigmatically a task at the material level. We insist, however, that there is a distinction between right and wrong processes, practices and methods, and that this distinction is essential to the concept of justification.

8

Actually, as we shall see below, it involves two references to epistemic value.

174

Critiques of Epistemic Externalism

III. Alstons "Concepts of Epistemic

Justification"

J2 is not yet our final formulation. In order to see how it should be supplemented, it will be instructive to consider the analysis offered by Alston in his "Concepts of Epistemic Justification", as it bears certain important similarities to our own and yet contains a crucial difference. Alston considers a wide variety of possible analyses of justification, discarding them one by one. He eventually offers the following as a plausible candidate: S is J in believing that ρ iff S's believing that p, as S does, is a good thing from the epistemic point of view.9 A few notes: the subscript Je is meant to indicate that this is an evaluative notion of justification; Alston contrasts this with a deontological notion. The epistemic point of view, according to Alston, "is defined by the aim of maximizing truth and minimizing falsity in a large body of beliefs."10 So, while his formulation is not identical with that which we have offered (in particular, he needs to tell us more about what he means by "is a good thing"), there are affinities: both make reference to the goal of truth, and both include a normative component. However, Alston notes the following worry: his formulation renders justification without a link to truth. "What reason is there for taking J to be independent of truth? I think the answer to this has to be in terms of the 'internalist' character of justification. When we ask whether S is justified in believing that p, we are, as we have repeatedly been insisting, asking a question from the standpoint of an aim at truth; but we are not asking whether things are in fact as S believes."11 This is quite illuminating. To apply Alston's observation to our partial formulation J2, it could be the case that an agent is aiming at truth in the right way but that the world isn't the way she takes it to be. If this is the case, though, there is no reason to think that her justified beliefs (according to this analysis of justification) are in any better epistemic stead than her unjustified beliefs—she might be a brain in a vat, for instance. Alston, rather than attempting to solve this problem directly, appeals further to the internalistic feature of justification he has noted; indeed, Alston tells us that "I don't see any way of remedying these defects other than by introducing an 'internalist' constraint." 12 However, the problem he has raised has a more obvious solution.

9 10 11 12

p. 70. p. 59. p. 71. Introduction to his (1989), p. 4. H e discusses this further in the passages following the above-quoted text from his (1985).

Knowledge, Justification and the Cooperative World

IV. A Second Normative

175

Component

The worry that Alston has brought to our attention is that things might not be the way an agent takes them to be, and that this is relevant to justification. We agree with Alston: if justification is meant to mark those beliefs which stand in better epistemic stead than others, and if it is also an evaluation of the way in which we aim at truth, then there must be some reason to think that the way we aim at truth—the processes, practices and methods we use to form and evaluate our beliefs—are usually successful in obtaining truth. To put the point directly, then, in order for a belief to be justified, we must form and evaluate our beliefs in the right way, and the external world must cooperate. This is not to say that the world must be exactly as we think it is, or that the world need be the way we desire it to be. Rather, the point is that, if we proceed in the right way in forming and evaluating our world view, and the world adequately matches our description of it 13 , we can arrive at justified beliefs; failing either of these conditions, though, our beliefs will not be justified in a way sufficient for knowledge. Our beliefs are most likely to be true, and are most worthy of positive epistemic appraisal, if we do what we should and the world is the way it should be. This suggests the following formulation: J3: S's belief that ρ is justified iff S arrived at the belief that ρ by seeking truth in the right way, and the world is the way it should be. Further, we wish to argue that this additional clause introduces another dimension of normativity into our analysis. We should say more, then, about what it means for the world to be as it should be. Consider Goldman's notion of normal worlds, which plays an important role in the account of justification offered in his Epistemology and CognitionThis notion is akin to the notion of the world being as it should be, which we have been discussing here. We should note that, of course, Goldman takes his account, as do many others, to be paradigmatically naturalistic and non-normative. We, on the other hand, will employ a normative variant of Goldman's notion of normal worlds — call our notion that of an epistemically adequate world. By way of introducing our notion, then, let us turn, then, to Goldman's notion of normal worlds. In the presentation of his account, Goldman introduces normal worlds to address the matter of which cognitive processes are to be counted as reliable, and thus as justification-conferring. His answer is that the relevant processes are those which are reliable in normal worlds, which he defines as "worlds consistent with our general beliefs about the actual world." 15 So, for instance, we take it that deduction is a reliable process but

13 T h e match need not be exact. T h e match must suffice to prevent the defeat or refutation of the justification by errors in what the person believes. See Lehrer (1990, 1997 and 2000) for an account for how a justification may be defeated or refuted. 14 It should be noted that Goldman had abandoned the notion of normal worlds by the publication of his (1992). 15 Goldman (1986), p. 107.

176

Critiques o f Epistemic Externalism

that clairvoyance is not; it is our beliefs about the reliability of these processes which is relevant to Goldman's account of justification. And these beliefs are in turn based upon more general beliefs about the world; roughly, we take it that perception is an accurate means by which to gain knowledge about the world, that the world follows certain natural laws, and that, for the most part, things are the way they seem. Thomas Reid discusses such a set of his desiderata under the heading "First Principles of Contingent Truths", in his Essays on the Intellectual Powers1^·, for instance, that memory is reliable, that things are as we perceive them, that we have some freedom over our actions, that reason is not fallacious, that other minds exist and can be known, that nature is uniform and thus induction is reliable. While we need not espouse each of these principles, the general point is made clear: a normal world is one in which things are mostly as they seem to be to a person of common sense and sound faculties. In normal worlds, then, we will be justified if we form and evaluate our beliefs in right way. This formulation incorporates a normative interpretation of the notion of a normal world which we consider essential. 17 We interpret a normal world in a normative way, calling it an epistemically adequate world, when we obtain true beliefs by forming our beliefs in the right way. (Hereafter we use 'adequate world' as short for epistemically adequate world'.) An adequate world is a world that is what we expect it to be when we form our beliefs according to the right processes, practices and methods. An adequate world is one in which, for instance, skeptical hypotheses concerning the demon or the brain in a vat are false, and our relation to an adequate world is such that we are able to have a great deal of success in our appropriate efforts to reach the goal of true belief. Adequate worlds are just those worlds in which things are the way we think they should be (epistemically speaking) when we form our beliefs in the way we should. This normative interpretation of an adequate world implies that there are some ways the world could be which are better than others for the pursuit of truth. There are epistemically better and worse ways for the world to be. This might explain, for instance, the distaste epistemologists and others share for skepticism: skepticism can be construed as the denial that the world is as it should be. Skepticism amounts to the claim that the world lacks a certain kind of epistemic value. When we reject skepticism, we must argue that the world in fact does have this value, and that is why we are able to attain justified beliefs (and knowledge). The notion of the world being as it should be epistemically—in other words, the notion of an adequate world—is an essential component of the concept of justification.

16 17

pp. 2 6 6 f f . G o l d m a n does not fill out the notion of normal worlds. H e gives a few examples of normal and nonnormal worlds, and tells us "Obviously, the notion o f normal worlds is quite vague." ( G o l d m a n 1986, p. 107). So, we do not mean to suggest that our notion o f normal worlds is the same as Goldman's; indeed, as a naturalist about epistemology, he would likely find our notion objectionable. But our goal is not tofillout Goldman's notion, but rather to take his notion as a starting point for introducing our own related notion.

Knowledge, Justification and the Cooperative World

177

V. The Janus-Faced Nature ofEpistemic Justification We have offered an analysis which contains two distinct normative components: one adhering to epistemic agents, and one to the external world. We might say, then, that epistemic justification is "Janus-faced"; there are two distinct but related sides to the concept. Each refers essentially to a kind of epistemic value, but in different ways. 1 8 O n e face looks inward at the practices of the agent, while the other face looks outward at the world. This suggests a new approach to the problem of justification. T h e debate between internalists and externalists has led most theorists towards one camp or the other. However, by recognizing the Janus-faced nature of justification, we can transcend those battle lines, and acknowledge that both internal and external factors are relevant to justification, albeit in distinct ways. We cannot ignore the fact that we must proceed in the right way, forming and evaluating our beliefs by the right processes, practices and methods. Here is an internal normative component to the concept of justification. N o r can we ignore the requirement that the world cooperate when we proceed in the right way yielding us success in obtaining truth. This provides an external normative component to justification. A useful way to clarify the character of the theory of justification we are proposing is to consider objections arising from conflicting claims in the literature. T h e first objection opposes our internal constraint that the person pursues truth in the right way in order to be justified in what she believes. Some authors 1 9 have suggested that if someone cannot help but believe something, it is permissible for the person to believe it, and, consequently, the person is justified in believing what she does. O n this account a person who is invincibly incompetent or irrational in believing something is, as a result of their incompetence or irrationality, justified in believing it. Our account implies that this view is wrong. Notice that if a person can't help but believe something irrationally, we still consider her beliefs irrational (though we might be less likely to blame her for them). Similarly, if a person does not form her beliefs in the right way, even if she cannot help her deficiency, she is not epistemically justified in believing what she does, that is, she is not justified in believing what she does in the way required to obtain knowledge. We might consider the person morally blameless in believing what she does in the wrong way because she cannot help herself, and some wish to say that she is morally justified in believing what she does because of her disability. But when the question arises of whether believing in this way will lead to knowledge and is justified for the purpose, epistemically justified, the answer must be negative. T h e second objection opposes our external constraint on justification that the world must be as it should be in order for a person proceeding in the right way in forming his belief to be justified in what he believes. What do we say, then, if the world is not as it

18

We shall leave open for now the question of whether these are the same kind of value or whether they are different in kind.

19

This view has been attributed to H u m e and (perhaps in error) to Reid, but it is clearly advanced by John Pollock.

178

Critiques of Epistemic Externalism

should be? Can we draw no distinction between justified and unjustified beliefs? Lehrer and Cohen have argued 20 that an agent who proceeds in the right way in forming her beliefs, but who is invincibly deceived by an evil demon, is justified in believing what she does. Even in a demon world, we can distinguish between those who are justified in what they believe and those who are not even though, due to the machinations o f the demon, they will both fall victim to error. Evil demon worlds are paradigmatic cases o f the world's failing to cooperate, and we propose that in such an instance, the person fails to be epistemically justified in what she believes. However, we can readily accommodate Lehrers and Cohen's intuition, and thus can also make sense o f their distinction between justified and unjustified beliefs in the demon world. Lehrer and Cohen presented their argument as an objection against reliabilism, and, as such, they are keen to emphasize the internal factors o f justification. They were, consequently, interested in emphasizing that there is a difference, as we would put the matter, between forming beliefs in the right way and forming them in the wrong way even in a world which does not cooperate with our efforts. Their intuition, that the distinction between justified and unjustified belief is tenable even in the demon world, is widely accepted. 21 We have no substantive disagreement with the point that Lehrer and Cohen made, and agree that there is a linguistic intuition that sustains their use o f the word 'justified'. There is a sense in which beliefs in the demon world may be said to be justified when they are formed in the right way. However, it has been central to the view that Lehrer himself defends that the kind of justification required to obtain knowledge must be a special kind of justification, undefeated justification, which, as we noted, rests on a match between internal and external factors. Beliefs in the demon world, whatever sort of justification they may possess, will fall short o f the mark o f knowledge because o f the missing match. Using the Janus-faced analysis o f justification, we can clarify the matter by drawing a distinction between two kinds o f justification, internal justification and what we have called, epistemic justification. Those subjects who aim at truth in the right way are internally justified in what they believe, but this is only one face of justification. If the world is not cooperating, then the second face is missing and the justification is insufficient. If, on the other hand, the two faces o f Janus are in agreement—that is, the subject has proceeded in right the way and the world is as it should be—then the belief is epistemically justified in the way required for knowledge. The belief constitutes knowledge in this case, and there is no room for the skeptic or any informed critic to refute the claim to knowledge. We might say, then, that the belief is irrefutably justified11.

20 21

See their (1983), as well as Cohen's ( 1 9 8 4 ) . We want to emphasize, though, that Lehrers own account of justification, though often misconstrued as such, is not an internalist account. Rather, Lehrer combines internal and external elements in theory of undefeated justification which is required for knowledge. Lehrer ( 1 9 9 0 , 2 0 0 0 ) .

22

Note that Lehrer uses this term to refer to an independently developed, albeit related, notion of justification in his ( 2 0 0 0 ) . However, there are strong affinities between the "match" theory presented there and the Janus-faced analysis developed in the current paper.

Knowledge, Justification and the Cooperative World

179

Justification, then, comes in two stages, one corresponding to each o f Janus' two faces. Even a brain in a vat can be internally justified, and thus we can distinguish between brains-in-vats which conduct themselves as they should—that is, in a way which would lead to knowledge in an epistemically adequate world—and those who do not. But to have knowledge, one must have irrefutable justification, and this is only possible when the world cooperates. This is why skepticism is an important problem—a skeptical hypothesis is one in which the world is not as it should be, and thus one in which at most internal justification is possible. 23

VI. Conclusion Strictly speaking, then, our Janus-faced theory o f justification matches external and internal factors to obtain a notion o f justification sufficient for knowledge. It maintains that some o f the factors relevant to justification are external to the agent. However, it is not a purely externalist theory, as it essentially incorporates an internalist desideratum requiring that the subject proceed in the right way. We prefer to think o f our theory as undermining the internalist-externalist dichotomy. We do not give priority to the internalist component nor to the externalist component. Each component is essential to epistemic and irrefutable justification. Like Janus, justification has two faces, one looking outward and one looking inward. T h e one face evaluates the practices o f the agent; the other face evaluates the world. It is only when the two faces see eye-to-eye that we have irrefutable justification and, thus, knowledge.

References Alston, William P., 1985. "Concepts of Epistemic Justification." Monist, 6 8 , pp. 57-89. Reprinted in Alston (1989). , 1988. "An Internalist Externalism." Synthese, 74, pp. 2 6 5 - 2 8 3 . Reprinted in Alston (1989). , 1989. Epistemic Justification: Essays in the Theory of Knowledge. Ithaca, NY: Cornell University Press. Cohen, Stewart, 1984. "Justification and Truth." Philosophical Studies, 4 6 , pp. 2 7 9 - 2 9 6 . Goldman, Alvin I., 1986. Epistemology and Cognition. Cambridge: Harvard University Press. , 1992. "Epistemic Folkways and Scientific Epistemology." In Liaisons, Cambridge: M I T Press. Hume, David, 1739. A Treatise on Human Nature. Oxford: Oxford University Press.

23

It is worth noting in passing that the problem of skepticism and the Gettier problem are distinct. Again, we can see this by attending to the second face of Janus. A skeptical hypothesis is one in which the world is not as it should, i.e. is not an epistemically adequate world. In such a world, irrefutable justification and knowledge are unattainable. Gettier problems, on the other hand, can arise in epistemically adequate worlds. We might put the point by saying that epistemically adequate worlds are nonetheless not epistemically perfect worlds—they will have local irregularities and glitches which will cause us to occasionally fall short of knowledge even though we have formed and evaluated our belief in the right way.

180

Critiques of Epistemic Externalism

Lehrer, Keith, 2000. Theory of Knowledge (second edition). Boulder, CO: Westview. , 1997.Self-Trust. Oxford: Oxford University Press. , 1990. Theory of Knowledge. Boulder, CO: Westview Lehrer, Keith and Cohen, Stewart, 1983. „Justification, Truth, and Coherence." Synthese, 55, pp. 191-207. Pollock, John, 1986. Contemporary Theories of Meaning. Savage, MD: Rowman & Littlefield. Reid, Thomas, 1983. Inquiry and Essays. R. Beanblossom and K. Lehrer, eds., Indianapolis: Hackett. Sartwell, Crispin, 1991. "Knowledge is Merely True Belief." American Philosophical Quarterly, 28, pp. 157165. , 1992. "Why Knowledge is Merely True Belief." Journal of Philosophy, 89, pp. 167-180. Truncellito, David Α., 1999. The Concept of Epistemic Justification. Doctoral dissertation, University of Arizona.

The Epistemological Promise of Externalism BARRY STROUD

There have been two distinct but related tendencies in the philosophical theory of knowledge. One, going back to Socrates as represented by Plato, is to ask what knowledge is: to seek to define it and to distinguish it from truth, belief, certainty, reasonable belief, and other closely related notions. T h e other approach focusses on human beings as they find themselves in the natural world and asks how they come to know the sorts of things they know. Questions of this second kind tend to become central to philosophy at times of upheaval, innovation, or even revolution within human knowledge itself. This was especially true of Europe in the seventeenth century. T h e rise of the 'new science' eventually overturned deeply entrenched ways of thinking of the world and our relation to it. Philosophical concerns with knowledge in our own day are still dominated by descendants of problems and theories introduced at that time. A central problem for philosophy since that time has been to understand the role of sense-experience in human knowledge, and to see whether or how we can know what we do about the world on the basis of what we perceive to be so. When that question is understood as completely general—about any perceptual knowledge of the world at all—it has proved difficult to avoid the conclusion that we perceive much less than we might unreflectively think we do. A familiar line of reflection seems to imply that, strictly speaking, sense perception alone is no reliable guide to the way things are in the world that produces it. Unaided by knowledge from other sources, it seems incapable of giving us access to an independent public world at all. This raises the question how we know any of the things we think we know about the world. This question is potentially disturbing or challenging only to those who have found the reach of sense perception to be severely limited compared to the richness of what they think we know on that basis. If the world around us were available to perception alone—if we could sometimes perceive the objects around us and perceive that many of the things we believe are true of them—there would be no general problem of explaining our knowledge. We would know what much of the world is like simply by perceiving it. So if an urgent challenge to our unreflective confidence about the world in general can be seen to arise at all, it is because of the force of those philosophical reflections that seem to shrink what we get through the senses alone to some meagre and apparently inadequate basis. T h e threatening reflections about sense perception start from the undeniable fact of perceptual illusions or mistakes. It sometimes happens that a person perceives something that is not there, or misperceives what is there, and so on the basis of perception is wrong about what is so. This obvious fact of life in itself implies no general scepticism. If there are also occasions on which we succeed in seeing or otherwise perceiving how things are in the independent world, then we do often know how things are, and

182

Critiques of Epistemic Externalism

we know it by perception. But if reflection on the nature of perception leads to the conclusion that what we perceive even when we happen to get true beliefs about the world extends no further than what we also perceive in illusions or when our beliefs about the world are false, then perceptual knowledge of the world in general is threatened. 1 What is true of perceptual illusions comes to be taken as representative of what perception even at its best can provide us with. That generalizing move from 'some' to all' is what leads to the conclusion that we never strictly speaking perceive what is so in the independent world. And that in turn seems to make it impossible to understand how we get knowledge of the world by perception. The philosophical study of knowledge in the early and middle years of the twentieth century was dominated by just such a conception of sense perception. It did not yield satisfactory answers to the question of how knowledge of the world is possible on that basis. One strategy was to try to bring the world we believe in closer to what we can perceive; every meaningful and potentially knowable proposition about the world was to be expressed exclusively in terms of possible sense experiences. 2 No plausible equivalences were found. Another idea was that everything we believe about the world beyond the strict limits of possible perception is a 'posit' we introduce or a huge 'theory' we construct to make the best overall sense of the limited sensory data we receive.3 But indefinitely many different 'posits' or 'theories' can be constructed, all consistent with the same perceptual data but incompatible with one another, so perception alone could provide no distinctive reassurance about the particular 'theory' we actually hold about the world. Taking fully seriously the idea of strict limits to what human beings perceive makes it seem impossible to account for the knowledge of the world that we all think we have. Scepticism seems the only strictly correct answer. The project of defining knowledge resurfaced in that same period in the twentieth century, partly under the influence of the idea that the proper or perhaps only method of philosophy is 'analysis'—the definition or clarification of those fundamental concepts we use to make sense of the world in the ways we do. Not only were relentless efforts made to define the concept of knowledge, 4 with more and more elaborate conditions being added to eliminate yet another clever counter-example to the necessity or sufficiency of what had been offered so far. It was also widely believed that success in that task would overcome the unsatisfactory sceptical plight that had haunted epistemology. The idea was that the restricted deliverances of sense perception seem insufficient for knowledge or reasonable belief about the world only because of a mistaken or distorted

1

2

3

4

For influential examples of reflection along these lines see, e.g., G. E. Moore, "Sense-Data", ch. 2 of his Some Main Problems of Philosophy (Allen & Unwin, London, 1953); Η. H. Price, Perception (Methuen, London, 1932) ch. 1; C. I. Lewis, Analysis of Knowledge and Valuation (Open Court, LaSalle, Illinois) ch. 7. See, e.g., C. I. Lewis, Analysis of Knowledge and Valuation ch. 8; A. J. Ayer, "Phenomenalism", in his Philosophical Essays (Macmillan, London, 1959). For a clear demonstration that the sought-for equivalences do not hold see, e.g., R. Chisholm "The Problem of Empiricism", The Journal of Philosophy 1948. See, e.g., the last section o f W . V. Quine, "Two Dogmas of Empiricism" in his From a Logical Point of View (Harvard University Press, Cambridge, Mass., 1953) or his Word and Object (MIT Press, Cambridge, Mass., 1960) ch. 1. The response to the challenge of E. Gettier, "Is Justified True Belief Knowledge?", Analysis 1963 eventually amounted to hundreds of articles or books up to at least the middle 1980s.

T h e Epistemological Promise of Externalism

183

conception of knowledge. The promise was that by getting the definition of knowledge right we would see that the apparent threat of scepticism vanishes. This strategy had much to recommend it. Anyone who feels forced to the conclusion that what is available in sense perception could never explain how knowledge of the world is possible must be thinking of knowledge in some way or other. If the conception of knowledge at work in such thoughts could be shown to be faulty, there would be no good reason to draw that conclusion. And if on a different and more accurate conception of knowledge strict limitations on sense perception represent no threat to knowledge of the world, the definitional or 'analytic' project would be vindicated as providing the solution to the philosophical problem of knowledge. This strategy seemed to gain further encouragement from the line of reasoning by which many philosophers explained the source of the traditional epistemological quandary. Restricted as we are to limited perceptual resources, it was argued, we could have reason to believe something about the world beyond them only if we could discover some connection between our limited sensory data and what is so in that wider world. No such connection could be established a priori, independently of experience; it could be discovered only on the basis of experience. But no appeal to sense experience alone could ever take us beyond the limited sensory data. So nothing available within sense experience alone could give us any reason to believe anything about what is so beyond it. That is why sense experience was said to be incapable of supporting any belief about a wider world. This reasoning appears to assume that in order to know or have reason to believe something I must know or have reason to believe that I know or have reason to believe it. Sensory data were held to give us no reason to believe things about the wider world because we could discover nothing within those data that would give us reason to believe in any connection they might bear to something outside them. Given only our restricted perceptual data, we could find nothing we could recognize to be such a reason. An externalist definition or conception of knowledge seems to promise a way out of this quandary. It implies that we do not have to meet that requirement in order to know or have good reason to believe something. According to externalism, the conditions of knowledge or reasonable belief are not conditions we must know or even believe to be fulfilled in order to fulfill them. To know or reasonably believe what is so it is sufficient if the conditions of knowledge or reasonable belief are in fact fulfilled. The view is externalist in holding that some of the conditions of knowledge or reasonable belief can be fulfilled independently of, or outside' or 'externally' to, what the subject believes about the fulfillment of those conditions. Just as one can know something without knowing that one knows it—even while believing that one does not know it—so one can have reason to believe something without knowing or reasonably believing that one does. If that is correct, and if the traditional problem arises only from insisting on stronger conditions for knowledge or reasonable belief, then the problem arises only from demanding more than we actually require for knowledge. An epistemological challenge raised on that basis would be empty; it would rest on a misunderstanding or distortion of the meaning of 'know' or 'reasonably believe' as they are actually used. This is the substance of an epistemological externalist diagnosis of the traditional problem.

184

Critiques o f Epistemic Externalism

To make good on this claim and on the consequent dissolution of the traditional sceptical threat, a correct definition of knowledge must be found, and it must be externalist in the required sense. Many philosophers have been working on that definitional project. T h e task has not been completed to everyone's satisfaction. Some insist on defining knowledge without using epistemic' terms like 'justified' or 'reason to believe'. Knowledge is accordingly to be understood as something like a true belief that is acquired or sustained in a certain way. One 'reliabilist' form of externalism says that the true belief must be acquired or sustained by a properly-functioning, reliable belief-forming mechanism, where reliability is defined in terms of something like the frequency with which it produces true beliefs. 5 Another externalist definition says that a knows that p if and only if a knows something r that is a conclusive reason for ρ in the sense that it would not have been true unless it was true that ρ , 6 Another version, related to this, says that a knows that ρ if and only if as belief that p 'tracks' the truth. 7 Less ambitious externalist definitions of knowledge do not require the elimination of all 'epistemic' terms. But a correct definition even containing such terms must remain externalist in the required sense in order to undercut the sceptical quandary in this way. I do not want to take up the question of the precise formulation, and so the definitional adequacy, of epistemological externalism. Rather I would like to assess the role that any such definition might play in the diagnosis and dissolution of the traditional epistemological problem. T h e quandary I have described arises from the realization that the most we can perceive is severely restricted compared to the richness of what we claim to know on the basis of it. Epistemological externalism promises to remove the difficulty by supplying a correct understanding of the idea of knowledge or reasonable belief. I believe that no one who sees our perceptual position as restricted in the way the traditional problem implies could get from an externalist definition of knowledge the understanding and reassurance he seeks about knowledge of the world. If all the limited perceptual information we receive is in fact reliably produced by just those unperceived states of affairs which we all believe to hold in the wider world, then according to the externalist definition we know that the world is as we believe it to be. But that would not give a reflecting philosopher who arrives at that restrictive view of perception a satisfactory understanding of how knowledge of the world is based on sense perception. T h e deficiency I have in mind would not be immediately discernible to a philosopher who reflects only on other people's sense perception and knowledge, not his own. Even if he believes that other people's perceptions alone give them no information as to what is so in the world, he might still try to explain their beliefs by appealing to what he takes himself to know about the world. He might find that the beliefs those people get on the basis of their restricted perceptions are in fact true and are produced by the world by processes that are reliable. That would be to find that the conditions

5 6

See Alvin G o l d m a n , " W h a t is Justified Belief?", in G . Pappas (ed.), Justification Dordrecht, 1979). F. Dretske, "Conclusive Reasons", Australasian Journal of Philosophy 1971

7

R . Nozick, Philosophical

Explanations

and Knowledge

(Reidel,

(Harvard University Press, Cambridge, Mass., 1981) ch. 3.

The Epistemological Promise of Externalism

185

expressed in an externalist definition of knowledge are fulfilled by those people. He could conclude that therefore they know what they claim to know. He would not have to regard what he takes to be the limited scope of their strictly perceptual data as a threat to their knowing what they think they know or to his own understanding of their knowledge. But someone who reflects on the role of sense perception in human knowledge in general must also acknowledge that, being human, he is one of the people to whom his conclusions apply. If he holds as a general thesis that perception even at its best gives no information about the independent world, then he must concede that his own sensory data are restricted in that same way. But that is to concede that whatever is so beyond what he perceives could be any one of indefinitely many different possible ways compatibly with the limited sensory data he receives. Nothing he could ever strictly speaking perceive to be so could discriminate among those possibilities. Of course, he might believe that the world is in fact only one of those possible ways, and he might think he knows which way that is. Perhaps he also believes, and claims to know, that the world's being that particular way is precisely what produces in him the limited sensory data he receives. He might also believe that it does so by processes that produce mostly true beliefs. If this last belief of his is in fact true, then he fulfills the conditions expressed in a reliabilist externalist definition of knowledge. If that definition is correct, it follows that he knows what he thinks he knows about the world, despite the poverty of his purely perceptual information. The question is not whether that is something the person knows; an externalist definition of knowledge implies that it is. The question now is whether a person's fulfilling the conditions of that definition of knowledge would put him in a satisfactory position for understanding himself as knowing by means of perception the things he thinks he knows about the world. I think it would not, as long as he holds that perception alone gives no information about what is so in the independent world. 8 The philosopher who reflects along those familiar lines could insist that if the world as he believes it to be does produce his perceptions and beliefs by processes that are reliable, then (according to externalism) he knows what he thinks he knows about the world. But he would also have to acknowledge that if his perceptions and beliefs are produced by a world that is different from the way he believes it to be, then he does not know what he thinks he knows, despite his believing it. And he would have to concede that all the strictly perceptual information he could ever get is compatible with each of those different possibilities. Nothing he could ever perceive to be so could distinguish between them. On the assumption that knowledge of the world is possible at all only on the basis of what is available in perception, this would leave the theorist unable to understand himself as knowing or having any reason to believe anything about what is so beyond

8

I argued this in some detail in "Understanding Human Knowledge in General" in M. Clay & K. Lehrer (ed.), Knowledge and Skepticism (Westview, Boulder, 1989). I tried again in "Scepticism, 'Externalism', and the Goal of Epistemology", Proceedings of the Aristotelian Society: Supplementary Volume LXVIII 1994. Both essays are available in my Understanding Human Knowledge (Oxford University Press, Oxford, 2000).

186

Critiques of Epistemic Externalism

what he perceives to be so. He might believe that he knows or has reason to believe such things, but he would see that he cannot legitimately draw any stronger conclusion on the basis of anything he could ever perceive. In his attempt to understand his own knowledge, he would find himself in the quandary that severely restricted perceptual data have always seemed to lead to: scepticism about any matters of fact beyond the limited sensory data. If this is right, then the diagnosis of the traditional sceptical reasoning offered by proponents of an externalist definition of knowledge was off the mark. It is not acquiescence in a faulty non-externalist conception of knowledge that leads to the epistemological quandary. The severely restricted scope of perception is what creates the threat. A philosophical theorist who accepts strict limitations on perceptual access to the world can never achieve a satisfactory understanding of any knowledge he thinks he has of what lies beyond those limits, even if he accepts an externalist definition of knowledge. And as long as such a restricted view of the scope of perception seems inevitable, or even plausible, we will not have a philosophically satisfying explanation of how perceptual knowledge of the world is possible. I think the greatest promise of epistemological advance in the face of this quandary lies not in an externalist definition of knowledge or even in a positive externalist answer to the epistemological question. Promise of more secure progress lies rather in externalism's potential for exposing and thereby undermining the assumptions needed for anyone to be faced with that apparently insoluble problem in the first place. What generates the completely general problem is the idea that no one ever perceives what is so in the independent world. The problem then is to explain how any beliefs whose contents go beyond that restricted domain of perception can gain any support from what is perceived. To be faced with that problem we must understand ourselves as never perceiving what is so in the world beyond or independent of what is perceived while nonetheless holding a great many determinate beliefs about that world. But could anyone coherently understand himself and others in that way? What must be true of a person who has perceptions with determinate contents, and beliefs about an independent world, but who never perceives what is so in the world that he believes in? Is that even possible? Could anyone consistently find that that is the position human beings are in? These are questions about the conditions of being in certain psychological states, and of understanding oneself and others as being in such states. To explore the promise of externalism on these matters would be to move away from a focus on knowledge in particular towards an understanding of psychological states and their contents in general. There is a conception of the mind that is broadly-speaking externalist in holding that the contents of a person's perceptions, thoughts, beliefs, and other attitudes are fixed in part by what is true in the world in which the person has those attitudes. This is not simply the truism that people get into the psychological states they do because of what happens in the world around them. Rather the idea is that the very contents of a persons perceptions or thoughts or beliefs on a particular occasion—what they are perceptions of or thoughts or beliefs about—are determined in part by what is or has been so in the world in which the person acquires those attitudes. Attitudes have those contents the possession of which best explains a person's acting and feeling and

The Epistemological Promise of Externalism

187

responding as he does. This is the basic externalist idea that seems to me to have the richest epistemological potential. Taken in one way, it appears to have the consequence that our perceptions and beliefs about the world are in general correct. 9 That is because whatever states of the world are responsible for our having the perceptions and beliefs we do are what, in general, those perceptions or beliefs are about, or are directed towards. For any particular perception or belief, or even for a sizeable number of them, it is possible for a person to be wrong, even wildly wrong, in what he perceives or thinks. But the externalist idea is that perceptions and beliefs in general must in large part represent the world as it in fact is; they would not be the very perceptions and beliefs they are if they did not. That is the sense in which what is so in the world in which people hold certain psychological attitudes is what determines the very contents of those attitudes. For the whole body of perceptions and beliefs taken all together, then, most of them, or most of the most fundamental of them, must be true. This has a very reassuring ring, and it might be taken as enough in itself to solve the problem of knowledge of the world once and for all. It means that the beliefs and other propositional attitudes that any philosopher would attribute to himself and others in the course of investigating human knowledge would be largely true. We could not turn out to be completely or mostly wrong about the world. If such an externalist conception of thought and experience could be established, it looks as if we would have a general guarantee of the truth of most of the beliefs anyone could ever have. Even if such an externalist view of the contents of beliefs could be established, I do not think it would be enough to give us a satisfactory explanation of how human knowledge of the world is possible. For one thing, true belief is not the same as knowledge or true reasonable belief, so an epistemological question can remain open even if we are assured that beliefs are mostly true. What needs to be accounted for is not just the truth of our beliefs, but human access to that truth. How are our beliefs known to be true, or how do we have good reason to believe them? These questions are not answered by the truth of our beliefs alone. An externalist conception of the mind and its contents in general would seem to provide answers to these questions since it implies that what we perceive to be so must also be true, at least for the most part. And if the things we perceive to be so are included among the very things we also believe to be so in the independent world, then not only the truth but also our knowledge of the truth of many beliefs about the world would be guaranteed along externalist lines. Anyone who understands perception in this externalist way could see that what we perceive gives us good reason to believe many of the things we believe about the world. There can be no better reason for believing something, and so no better way of coming to know it, than seeing or otherwise perceiving that it is true. Such an externalist view of perception would therefore account not only for the truth of our beliefs, but also for our access to those truths, and so for our knowledge of the world.

9

For an emphasis on this implication of externalist views see, e.g., Donald Davidson, "A Coherence Theory of Truth and Knowledge", now together with other relevant essays in his Subjective, Irttersubjective, Objective (Oxford University Press, Oxford, 2001).

188

Critiques of Epistemic Excernalism

This understanding of perception would provide a satisfactory account of our knowledge precisely because it accords us unproblematic perceptual access to the way things are in the independent world. We would not have to infer how things are from something we perceive; we would often see or otherwise perceive that things are a certain way. That would give us knowledge of many things that are so in the world, and that knowledge would give us good reason to believe many other things about the world that we do not perceive to be so. But if further reflection on the nature of perception led persuasively to the conclusion that we actually have no such perceptual access to the independent world, the satisfactory understanding of our position promised by an externalist conception of the mind and its contents would no longer be available. In yielding to those threatening reflections about sense perception we could no longer understand ourselves as having the reasons to believe things that we thought we had. This apparent threat to our understanding of human knowledge would be overcome once and for all if it could be shown that perception simply cannot be understood in that way. One way to do that would be to show that the restrictive view of perception is wrong and that an externalist conception of perception and its contents is correct. That would require establishing the truth of the externalist view that what we perceive to be so is included among the very things we believe to be so in the independent world, and that what we perceive and believe to be so is, for the most part, in fact true. But there is a question whether that view can be established in a way that provides the reassurance it seems to promise. The view cannot be shown to be true by first identifying all of a person's or a community's perceptions and beliefs and then discovering independently that all or most of them are true. That could indeed give us great reassurance, but the idea behind the externalist view is that the contents of perceptions and beliefs cannot be identified independently of identifying what is so in the world in which they are held. That means that anyone who attributes perceptions or beliefs to anyone must hold some determinate beliefs himself about the world. But fulfilling that condition of the successful attribution of perceptions and beliefs does not give an attributor any greater reassurance of the truth of the beliefs he attributes than the assurance he has of the truth of the beliefs he relies on in attributing those beliefs to others. An attributor will inevitably regard as true most of the beliefs he attributes. But if his grounds for his own beliefs could be brought into question by reflections on the nature of perception, he could be faced with a general problem of knowledge even if he began by accepting an externalist view of mental contents. Support for the externalist view comes from the evident impossibility of anyone's identifying or attributing beliefs or perceptions or other attitudes without holding some beliefs himself about what is so in the world. There appears to be no way for one person to come to any verdict about what another person perceives or believes except by finding some intelligible connection between that person's responses and what is so in the world around him. That requires beliefs about the world on the part of the attributor, and beliefs about one thing's being reason to believe another. He must appeal to the truth of such beliefs in making sense of what the other person does. That means that in general, or in the normal case, what one person comes to believe another person perceives or believes will be something that the attributor himself also perceives or

The Epistemological Promise of Externalism

189

believes to be so. Very general agreement between them is a consequence of fulfilling the conditions of successful attribution of any perceptions or beliefs; disagreement is intelligible only against a larger background of mostly shared attitudes. The externalist view in its reassuring-sounding form says that the contents of the perceptions and beliefs that people attribute to one another are for the most part true. But I think it must be conceded that the truth, in general, of the perceptions and beliefs that people all agree in attributing to one another does not follow from their agreeing in attributing them. Widespread agreement in perceptions and beliefs—even inevitable agreement—does not imply that what everyone agrees in perceiving or believing to be so is in fact true, even for the most part. 10 So it does not imply that the externalist view of perception and belief in its reassuring-sounding form is correct. And it therefore does not imply that a restrictive view of the scope of perception is wrong. Even if this failure of implication is conceded, I think it represents no obstacle to externalism's potential for showing that perception cannot be understood in the restrictive way that generates the apparently insoluble problem of knowledge of the world. To avoid the consequences of that view of perception it is not necessary to establish the positive doctrine that what we perceive to be so is in fact, for the most part, just the way things are in the world. It would be enough to show that on that restrictive view of perception no one could find himself presented with the completely general epistemological problem in the first place. 11 That would not amount to a reassuring positive answer to the epistemological problem. It would not show that the restrictive view of perception is false, and that we do perceive and so know how things are in the independent world. But it would completely eliminate any danger of arriving at a negative sceptical answer to the epistemological problem from general reflections about what human beings perceive. One way of working towards such an ambitious result would be to show that if the restrictive view of the scope of perception were true, no one could consistently attribute any perceptions or beliefs to anyone. That would mean that the general epistemological problem could not then arise. Someone who does not attribute perceptions or beliefs to anyone cannot be faced with the problem of explaining how what people perceive gives them reason to believe what they believe about the world. Contrapositively, anyone who found himself faced with such a problem concerning perceptions and beliefs he thinks people actually have could not consistently hold that the contents of all those perceptions reach no further than what the restrictive view of perception implies. And if some of the perceptions he attributes are conceded to be perceptions of what is so in the independent world, once again the general epistemological problem does not arise. Either way, no one could be faced with a pressing general problem of how what is perceived can give us reason to believe what we believe about the world. 10 I have argued in more detail for this in "Radical Interpretation and Philosophical Scepticism" in L. E. Hahn (ed.), The Philosophy of Donald Davidson (Library of Living Philosophers, Open Court, LaSalle, Illinois, 1999), now also in my Understanding Human Knowledge, and in "Anti-Individualism and Scepticism" in M. Hahn & Β. Ramberg (ed.), Reflections and Replies: Essays on the Philosophy of Tyler Bürge (MIT Press, Cambridge, Mass., 2002). 11 This suggestion is made but not developed very far in the last section of my "Radical Interpretation and Philosophical Scepticism".

190

Critiques of Epistemic Externalism

I think the ideas behind an externalist understanding of the mind and its contents strongly encourage pursuit of such an outcome. It is not a project that anyone can be said to have explicitly carried out in convincing detail. Nor is it a goal to be reached overnight. It would require a thorough investigation in an externalist spirit of what a person must believe about the world in order to recognize the presence in it of people with perceptions and beliefs with determinate contents that he understands. Without being able to think of human beings in that way, no one could be presented with a pressing epistemological problem. And on an externalist understanding of the conditions for the attribution of such perceptions and beliefs there is good reason to think that an attributor must take himself and others sometimes to perceive what is so in the independent world. To ascribe beliefs about the world to anyone we must find some way to connect something we take to be so in the world with the believer's believing what we think he believes. It is not enough simply to find something that we believe to be so in the person's vicinity, for instance, or something of which he has no inkling. Ascribing beliefs requires finding some reason the believer actually has to believe what we take him to believe: either something that is a reason or at least something he takes to be such a reason. People's reasons for believing what they do typically lie in other things they believe. An assumption essential to the traditional epistemological problem is that our reasons for believing what we do about the world ultimately depend at least in part on what we perceive to be so. That means that anyone who attributes beliefs must eventually attribute to the believer perceptions of something which is, or which is taken to be, reason to believe what he is thought to believe. The contents of the beliefs we attribute are to be identified somehow in terms of what the perceptions we attribute are thought to give the believer reason to believe. This applies both to attributions of beliefs and perceptions to ourselves in the first person and to the attribution of such attitudes to others. As things are, we typically ascribe perceptions with determinate contents by finding that the perceiver perceives something that we perceive or know or believe to be so in the world we all inhabit. On the view that no one ever strictly speaking perceives what is so in the independent world, we could not attribute perceptions to ourselves or others in this familiar way. Even if we had many beliefs about the way the world is, and we invoked the truth of those beliefs in attributing beliefs to ourselves or others, we could not find that anyone ever perceives the truth of any of those things we believe to be so. To do so would be to grant that some people do perceive what is so in the world after all, and that is what the restrictive view of perception denies. So the only perceptions we could ascribe to anyone on that view would be perceptions whose contents imply nothing about how things are in the independent world. But we would nonetheless have to regard perceptions of that kind as providing reasons to believe things about the independent world if we are to attribute any such beliefs at all. The epistemological promise I see in an externalist conception of the mind and its contents in general lies in its potential for showing how and why no one could do that. Focus on the conditions of the attribution of perceptions and beliefs raises the question of what contents, if any, we could ascribe to any perceptions at all if they had to be understood only in that restrictive way. And even if we could recognize some such

The Epistemological Promise of Externalism

191

perceptions, that raises in turn the further question o f how and whether we could regard them as giving anyone reason to believe anything that goes beyond those restricted contents. I f we could not, we could attribute no beliefs about the world to anyone. Pursuing these questions would lead to greater appreciation o f what I think externalism shows to be a necessary connection between what a person can be understood to think or believe to be so and what he can intelligibly be thought to perceive to be so. 1 2 T h e restrictive view o f perception, and so the traditional problem o f perceptual knowledge o f the world, thrive on denying or ignoring the connection. Exploration o f that important link is what I think promises the richest epistemological rewards.

12

I begin to explore this connection and its anti-subjectivist implications as applied to the particular case of perceptions o f and beliefs about the colours of things in my The Quest for Reality: Subjectivism and the Metaphysics of Colour (Oxford University Press, New York, 2 0 0 0 ) chs. 5-7.

Is Knowledge a Natural Phenomenon? MICHAEL WILLIAMS

1. Two Approaches to Knowledge In Knowledge and its Place in Nature, Hilary Kornblith divides recent philosophical accounts of knowledge into two main types: "those that require nothing more than a certain responsiveness to features of the environment, such as reliability accounts, tracking accounts, and the like" and "those that require a high degree of sophisticated meta-cognitive processing".1 Kornblith favours an account of the first type. In his view, knowledge is true belief resulting from the exercise of a capacity that reliably produces true beliefs, in the circumstances in which it was exercised. Accounts of knowledge of Kornblith's first type represent a revolutionary alternative to the long tradition of identifying knowledge with some form of justified true belief. First, such an account is purely reliabilist, in the sense that it denies that knowledge has any essential connection with the ability to justify or give reasons for one's beliefs. In this way, such an account is radically non-justificationist. It is also radically externalist. According to radical externalism, a person may know that Ρ in virtue of fulfilling some condition C without knowing, or even having the ability to come to know, the he or she fulfills that condition. A purely reliabilist account of knowledge requires only that knowers be blessed with appropriately reliable cognitive capacities. There is no need for any awareness of how reliable one's capacities are. No account of the second type is purely reliabilist. However they may differ—and they can differ in fundamental ways—all theories of the second type insist that one cannot be a knower unless one has some capacity for assessing—hence for justifying— one's beliefs. And no account of the second type can be radically externalist either. In many cases, one assesses the credibility of one's beliefs in the light of one's knowledge of one's reliability in gathering and processing information. Thus to link knowledge with a capacity for self-conscious critical scrutiny of one's epistemic performances is to make knowledge essentially dependent on what Kornblith calls "sophisticated metacognitive processing". Kornblith favours pure reliabilism because, he thinks, it allows knowledge to be understood in a thoroughly naturalistic way. Kornblith takes talk of understanding knowledge very seriously. Physicists do not aim to understand our concepts of the physical world, but the physical world itself. In the same spirit, what we philosophers should aim to understand is not our concept of knowledge but knowledge itself. Knowledge is a "natural phenomenon", to be studied in the way that we study any feature of the 1

Hilary Kornblith, Knowledge and Its Place in Nature (Oxford: Oxford University Press 2002), p. 72. Subsequent references given by "KPN" and page number.

194

Critiques of Epistemic Externalism

natural world. Knowledge and its Place in Nature is an important systematic defense of this uncompromisingly naturalistic view. An issue that brings the differences between the two approaches to a head is the question of how to understand attributions of knowledge to animals and small children. Accounts of knowledge that stress the need for sophisticated meta-cognition strongly suggest that knowledge properly so-called is confined to human adults. For many philosophers who favour such accounts, talk of "animal knowledge" stretches the concept of knowledge, possibly past its breaking point. At a minimum, if they recognise animal knowledge at all, such philosophers are apt to treat it as not just very different from human knowledge but as decidedly second-rate.2 Kornblith thinks that such an attitude is completely misguided. Attributions of knowledge to animals are not confined to sentimental pet owners. Cognitive ethologists, who study animal cognition in a systematic way, find such attributions indispensable, and their talk of animal knowledge is no mere fagon de parier. In cognitive ethology, appeals to animal knowledge do serious explanatory work. Animal knowledge is a legitimate object of empirical study, with a "good deal of theoretical integrity to it". 3 Nor is there anything second-rate about the sort of knowledge that animals possess. Rather, knowledge is essentially the same in humans and other animals: it is true belief, reliably generated. Knowledge is essentially the same in humans and other animals. Kornblith is well aware that his espousal of pure reliabilism puts him at odds with philosophers, such as Brandom, Davidson and myself, who think that, when it comes to understanding knowledge, there is no ignoring justification. In repudiating what he takes to be the anti-naturalist strain in Brandom, Davidson and myself, Kornblith focusses on what he takes to be a common element in our views: the idea that to be a knower is to be a participant in a social practice. In taking this tack, I will argue, Kornblith misses the main point, which has to do with whether knowledge of the environment is essentially bound up with what he calls "meta-cognitive processing." Brandom, Davidson and I all think that it is. If we are right in this, pure reliabilism is seriously inadequate. Knowledge is not a natural phenomenon.

2. Animal

Beließ

Animals have to forage for food, attract mates, deal with predators, protect and nurture their young, and so on. To meet such biological needs, they have to recognise features of their environments that promise sustenance, pose threats. They need to get where they need to be: back to the den or south for the winter. There is no doubt, then, that animals (or their nervous systems) are in constantly changing informationbearing states. These states may be in some ways analogous to beliefs. But are they

2 3

As exemplars of this attitude, Kornblith mentions Laurence Bonjour, Keith Lehrer, Colin McGinn, and Ernest Sosa. Se K P N , pp. 71-2 and 104-6. K P N , p. 28.

Is Knowledge a Natural Phenomenon?

195

more than that? And even if animals have genuine beliefs, do their beliefs amount to knowledge? These questions suggest two ways of arguing for an essential discontinuity between animal "knowledge" and human. A critic might argue that animals may have beliefs, but definitely lack knowledge. Or more radically, he might deny that animals even have beliefs. I shall first consider the radical view, which Kornblith finds in Robert Brandom and Donald Davidson. Both Brandom and Davidson associate belief with speaking a language; and both see speaking a language as an essentially social undertaking. But according to Kornblith, Brandom sets the standards for being a believer even higher, arguing that having beliefs requires participation in a highly specific social practice. I shall look first at Kornblith's reasons for repudiating Brandom's supposedly more exacting standards. In later sections, I shall have something to say about Davidson. We shall find that, in the matter of animal beliefs, the views of Davidson and Brandom are not really so different. Brandom is not unqualifiedly hostile to reliabilism, in that he does not deny that there may be cases of knowledge that answer to pure reliabilist standards. However, he claims that such cases are necessarily exceptional. The key to his argument is the Sellarsian claim that genuinely cognitive beings must understand what they say (or think). Moreover, understanding a claim requires a grasp of how it might serve as a reason for other claims, or how other claims might serve as reasons for it. Thus: A n assertion, even if true, is not taken to express knowledge unless the one making it understands the claim being made. A practical grasp of the significance of making the claim is inseparable from an appreciation of its role as a possible reason for other claims, and something that reasons can in turn be offered for. It is being caught up in this way in the game of giving and asking for reasons that makes a performance the undertaking of a commitment (the making of a claim) in the first place. 4

Thus while we might allow cases of knowledge in which a believer cannot himself give reasons for a claim (though we can, on his behalf as it were), the existence of such cases cannot provide a reason for supposing that an appeal to brute reliability could simply replace the appeal to reasons in explicating the nature of knowledge. But since Brandom doubts that animals are involved with reasons in the way that genuine believing demands, he declines to credit animals with beliefs. Brandom says that beliefs are caught up in "the game of giving and asking for reasons". Kornblith takes this talk of a "game" very seriously, taking Brandom to hold that being a believer requires participation in a social practice of giving and asking for reasons. Since animals don't play this game, they don't count as believers. How much participation is required? This question is pressing, Kornblith thinks, because there is wide variation, perhaps in part culturally determined, in the extent to which humans give or ask for reasons. Some people are highly "Socratic": they are

4

Robert Brandom, Making It Explicit (Cambridge MA: Harvard University Press 1994), p. 214. Quoted by Kornblith, KPN, p. 75. Subsequent references to Making It Explicit given by "MIE" and page number.

196

Critiques of Epistemic Externalism

always pressing people to justify what they say or think. Others are laconic: men of a few words who show little interest in either justifying their own beliefs and actions or in challenging those of others. Perhaps some people see the whole practice as rude or boorish. (Nietzsche remarks that a nobleman doesn't give an argument: he gives his word) We can even imagine a tribe of Superyankees, a community of individuals who never give or ask for reasons Would we have to see such hyper-laconic people lacking beliefs? Surely not. Kornblith identifies two factors that may have led Brandom astray. The first is a tendency to over-emphasize the importance of communication (hence to over-emphasize the importance of language). As evidence of this, Kornblith notes Brandom's remark that "The possibility of extracting information from the remarks of others is one of the main points of the practice of assertion, and of attributing beliefs to others".5 To be sure, our need to rely on other people for information gives us an interest in their trustworthiness as informants, and this may lead us to ask, from time to time, whether something we have been told can be relied on. But the most important reason for attributing beliefs and desires to other people is that, unless we do so, we can make no sense of their behaviour, linguistic or otherwise. Even when dealing with Superyankees, we would need to credit them with beliefs and desires, on pain of being unable to make sense of them at all. These points suggest to Kornblith that Brandom gets the relation between having beliefs and being involved with a social practice of reason-giving exactly backwards. Extensive involvement in such a practice is not a precondition of having beliefs: rather, how extensively a person participates in such a practice is something that we will appeal to his beliefs and desires to explain. (He avoids asking for reasons because he finds it impolite.) The relation between having beliefs and participating in a social practice of reason-giving is causal, not constitutive. A defender of Brandom might say that Kornblith reads too much into Brandom's talk of the game of giving and asking for reasons as a social practice. True, for various reasons, Brandom thinks that sophisticated conceptual abilities depend on learning a language, and that languages are socially developed and transmitted. But with respect to animal "beliefs", the key issue is the dependence of conceptuality on sensitivity to reasons. Whether such sensitivity requires participation in any particular social practices is a further question. Kornblith agrees that Brandom sometimes explains the Rationality Requirement in a way that plays down the social angle, making being a believer depend on sensitivity to reasons. But the trouble with this more modest requirement, he thinks, is that animals easily meet it. In explaining animal behaviour, recourse to vocabulary is unavoidable. But to characterize an animal's behaviour in intentional terms—as warning against a predator, or distracting a rival—is eo ipso to credit the animal with sensitivity to reasons.

5

Robert Brandom, "Insights and Blindspots of Reliabilism". Monist, 81 (1998); reprinted in Brandom, Articulating Reasons (Cambridge MA: Harvard University Press 2000), pp. 97-122. Quotation p. 120. Subsequent references given by "AR" and page number. Passage quoted by Kornblith, KPN, p. 79.

Is Knowledge a Natural Phenomenon?

197

This brings us to the second factor that induces Brandom to deny that animals have beliefs. According to Kornblith, Brandon plays fast and loose with important distinctions and, by so doing, facilitates a slide from the modest (but useless) requirement to the stronger requirement just rejected. Kornblith quotes Brandom: Practitioners who can produce and consume assertions are linguistic beings. Practitioners who can produce and consume reasons are rational beings. Practitioners who can produce and consume knowledge-claims are cognitive beings [T]hese are three ways of talking about the same practices and the same capacities. 6

But while to explain actions by appeal to beliefs and desires is to treat the actor as capable of being moved by reasons—in Brandom's terms, as a "consumer of reasons"— the capacity to be moved by reasons is not the same as the capacity to produce reasons. However, by running them together, Brandom makes plausible his suggestion that having reasons depends on the ability to speak a language. This does not yet get us to social practice, since having the capacity to produce reasons is not the same as being a participant in a practice that calls on that capacity. (Remember the Superyankees.) But Brandom compounds his error by failing to distinguish capacities from practices. This allows him to identify being in a position to articulate one's reasons with being an active participant in a social practice of giving and asking for reasons. By virtue of two mistaken and unargued assimilations, Brandom is able to slide from a creature's being a suitable subject of intentional explanation to its being a player in a social game. Correct for this, Kornblith claims, and Brandom's argument evaporates.

3. Discrimination and

Understanding

While Kornblith mounts a powerful assault on Brandom's Sellarsian scepticism about animal knowledge, I think that the Sellars-Brandom outlook can be defended. What clouds the picture, I think, are certain features of Brandom's way of stating his position that invite criticism from naturalists like Kornblith. To see how damaging such criticisms really are, we need to get the central issues more clearly in view. In support of his view that Brandom's case against animal belief rests on a very demanding idea of participation in a social practice, Kornblith cites the following: A community precluded from giving reasons for beliefs cannot so much as have the concept of reliability—nor, accordingly, (by anyone's lights) of knowledge. Its members can serve as measuring instruments—that is reliable indicators—both of perceptible environing states, and each other's responses. But they cannot treat themselves or each other as doing that. For they do not discriminate between reliable indication and unreliable indication. 7

6 7

MIE, p. 203. Q u o t e d by Kornblith, K P N , p. 81. AR, p. 107. Q u o t e d by Kornblith, K P N , p. 82.

Critiques of Epistemic Externalism

198

Kornblith comments that it is entirely unclear why those who are not part of a community that gives and asks for reasons cannot have the concept of reliability. Superyankees, while hyper-laconic, need not by supposed to trust everyone equally: they just keep their doubts to themselves. They have the concept of a reason and they surely make use of it, for they "discriminate between reliable and unreliable indicators". Indeed, many non-human animals do this too, for they "demonstrably learn". Pigeons pecking at colored disks are able to discriminate between reliable a n d unreliable indicators o f f o o d . T h e y don't self-consciously bring to bear a concept o f reasons, but that isn't required for discriminating between reliable an unreliable indication. 8

But although Kornblith takes himself to be giving a single argument here, the Superyankees and the animals are very different. Unlike the pigeons, the Superyankees have the concept of a reason, and make self-conscious use of it in deciding whom to trust. An unfortunate feature of Kornblith's treatment of Brandom's scepticism with respect to animal beliefs is that it fails to link Brandom's conception of beliefs, as essentially involved with sensitivity to reasons, with the question of reflective knowledge, preferring instead to focus on a supposed (an obscurely motivated) insistence that reasons-sensitivity presupposed participation in a social practice of argumentation. I think that this causes him to miss the deepest sources of Brandom's reservations about pure externalist reliabilism. As we saw, Brandom's starting point is the claim that believers must understand what they are up to. So from Brandom's standpoint, the first question to ask is whether mere differential responsiveness to environmental conditions—even acquired or selected for responsiveness—is sufficient for understanding. T h e answer is "No". As Brandom points out, the world is "full of reliable indicators": pieces of iron rust in wet environments but not in dry ones, land mines explode when stepped on with a force exceeding a certain threshold, bulls charge red flags, and so on. But according to Brandom, such "reliable dispositions to respond differentially to stimuli, and thereby to sort the stimuli into kinds, do not qualify as cognitive, because the responses that are reliably elicited are not applications of concepts". 9 Since they are not applications of concepts, they do not involve the formation of beliefs. Why is this? What is the difference between a parrot trained to squawk "Red" in the presence of red things and a person who can make a genuine, noninferential observation report? Brandom's answer is that a genuinely conceptual response is "inferentially articulated". Thus: B e l i e f s . . . a r e essentially things that can serve as premises a n d conclusions o f inferences. T h e subject o f g e n u i n e perceptual beliefs is, as the parrot is not, r e s p o n d i n g to the visible presence o f red things by m a k i n g a potential move in a g a m e o f giving a n d asking for reasons: a p p l y i n g a concept. T h e believer is a d o p t i n g a stance that involves further consequential c o m m i t m e n t s (for instance, to the object perceived b e i n g colored) that is [sic] incompatible with other c o m m i t m e n t s (for instance, to the object perceived being green), and that o n e can show one's entitlement to in terms o f other c o m m i t m e n t s (for instance, to the object per-

8 9

KPN, p. 82. AR, p. 108.

Is Knowledge a Natural Phenomenon?

199

ceived being scarlet). N o response that is not a n o d e in a n e t w o r k o f such broadly inferential involvements is recognizable as the application o f concepts. A n d if not, it is n o t recognizable as a belief, or the expression o f a belief either. 1 0

Brandom calls this failure to take on board the demanding requirements imposed by genuine conceptual activity reliabilism's "Conceptual Blindspot." 11 I agree with Brandom's claim that propositional content requires inferential articulation. But his way of stating his argument, in the passage just quoted, might appear to offer an opening for a defender of Kornblith's views about animal belief. Brandom has a fondness for turns of phrase that, incautiously read, might seem to offer openings to his naturalist opponents, by suggesting that his position depends on an outmoded approach to animal psychology, as well as by blurring a distinction between two quite different ways of gathering information. Let me begin with the point about outmoded psychology. In the argument just given, Brandom appears to set up a stark opposition between mere reliable differential responsiveness and the sort of responsiveness to reasons that depends on the subject's possessing the concept of reliability. With this opposition in place, he can argue, in effect, that if animals lack the capacity for epistemic reflection, then they are no more genuinely cognitive beings than are thermometers, or for that matter, iron bars. But why does Brandom feel free to blithely assimilate animal perceptual capacities to the sort of "differential responsiveness" shown by bars of iron. If we posed this question to Kornblith, his answer would surely be to point to Brandom's predilection (shared with Sellars) for the language of stimulus-response psychology. In Sellars's case, this predilection might be excusable in that many of his most influential writings belong to a period when classical behaviourist learning theory was still widely accepted. But no such indulgence can be granted Brandom, since Kornblith's case for animal beliefs turns on the inadequacy of psychology of this type. That Brandom's argument misses the mark, we might argue, can be seen from the fact that Kornblith himself holds that, even when coupled with biological needs, mere differential responsiveness is an insufficient basis for crediting an organism with beliefs. Plants are a case in point. While plants need sunlight for photosynthesis and turn towards the sun to get it, no one would explain this in terms of the desire for light and the belief that more can be obtained by turning in a certain direction. This is because there is a lower level (physico-chemical) explanation for plant phototropy. By contrast, in the case of much animal behaviour, no such lower level explanation is available.12 In making this claim, Kornblith is inspired by cognitive ethology: the study of animal cognition as displayed by behaviour in an animal's natural environment. Cognitive ethologists have recorded highly complex patterns of behaviour—some involving interactions with other animals. For example, ravens have been observed to co-operate in stealing an egg from a nesting hawk, one raven distracting the hawk while the

10 AR, p. 108-9. 11 AR, p. 109. 12 KPN, p. 38f.

200

Critiques of Epistemic Externalism

other snatched the egg. Behaviour like this resists explanation in stimulus-response terms. It can only be explained in terms of mental representations, individuated by their informational content. Stimulus-response psychology, of the kind Brandom gestures towards, tried to explain behaviour as bodily movements produced in response to physical conditions. But as ethologists argued even during behaviourism's heyday, if we think of behaviour as bodily movement, we cannot even capture the behaviour we want to explain. The behaviour of the ravens we naturally describe as "distracting another bird" cannot be reduced to bodily movements, even if we allow bodily movements to be described in such general terms as "wing flapping". Birds flap their wings for many reasons. Avoiding all reference to reasons, and insisting on characterizing animal behaviour as mere movements, we lose the ability to "recognize subsequent repetitions of the same behaviour." 13 To understand an animal's behaviour, we need to know what the animal believes about the environment and what it is doing about it: e.g. that it is fleeing a predator, having detected a rustling in the grass. To sum up, animals aren't plants, much less iron bars. In predicting plant motion, "reference to stimulus conditions is sufficient." By contrast, [t]he elaborate behaviour o f ravens in distracting a hawk so as to steal her eggs is not a simple response triggered by s o m e environmental condition. While the behaviour is straightforwardly explained by appealing to beliefs a n d desires, n o o n e has ever offered an explanation o f such c o m p l e x behaviours in terms that obviate the need for representational states. 1 4

But if explaining animal behaviour demands reference to mental representations, it will be possible—indeed often necessary—to see animal representations as inferentially articulated. The articulation may be less fine-grained than would be the case if we were dealing with human perceptual beliefs, but so what? Animal concepts need not be identical with human concepts. In arguing that animal behaviour cannot be explained in the lower-level terms appropriate to phenomena like plant phototropy, Kornblith relies on Fodor's claim that psychological kinds need not be related in any simple way to physical kinds because psychological states are functional states. Beliefs are like camshafts. Camshafts, while physical objects, do not constitute a physical kind: they come in different shapes and sizes, may be made of different materials, and so on. But they constitute an important functional kind, in virtue of the common role they play in automobile engines. Beliefs, too, may be differently realized in different animals (within and across species). But this does not make them illegitimate kinds in psychology. The legitimacy of talk about beliefs, individuated by content, is "secured by showing that this allows for a psychological theory that succeeds in prediction and explanation." 1 5 But as we have seen, we need belief-desire talk to capture the relevant behavioural similarities (which cannot be captured by movement-talk) just as we need camshaft-talk to capture common problems in physically heterogeneous engines.

13 14 15

KPN, p. 34. KPN, p. 42. KPN, p. 40.

Is Knowledge a Natural Phenomenon?

201

However, while we can accept this defence o f the importance o f psychological kinds, the claim that psychological states are functional states does not itself legitimate the appeal to animal beliefs. As Kornblith himself recognizes, the case for animal knowledge presupposes animal beliefs, where beliefs are more than just information-bearing states. But functional states are information-bearing, and information-bearing states can be characterized in functional terms. So while it is true that, following Sellars, B r a n d o m sometimes writes as if animal psychology were a matter o f "responses" being triggered by "stimuli", with an implicit suggestion that "stimuli" can be characterized in physical terms, it is not clear that this matters. B r a n d o m can agree that we need to see animals as being in information-bearing states, functionally characterized. W h a t he will deny is that beliefs are mere functional states. T h e position o f the camshaft carries all sorts o f information about the states o f other engine-components and about what the engine needs to continue running. But this doesn't give engines a belief-desire psychology. W h a t we need to be shown, then, is that describing an animal as "warning against a predator" commits us to treating the animal as displaying genuine conceptual activity. We need to be shown that such descriptions are more than functional and quasi-intentional. N o w I d o not mean to suggest that Kornblith thinks that psychological states are mere functional states. After all, he himself speaks o f beliefs as "individuated by content". But in the light of the argument just given, the question that needs answering is: what can Kornblith, consistent with his naturalism, say about how content is fixed. WTiat makes animals' psychological states genuinely conceptual, rather than merely information-bearing? O n e answer that Kornblith cannot possibly admit is Brandom's. B r a n d o m is not a functionalist, if functionalism is meant to be a form o f naturalism. To be sure, there is a certain analogy between functionalism and Brandom's inferentialist conception o f meaning, in that both involve a kind o f holism. A part is a camshaft in virtue o f the way it interacts with other engine components. Similarly, for an inferentialist like Brand o m , a beliefs content depends on its relations to other beliefs, and in some cases to circumstances. But B r a n d o m departs from naturalist functionalism in holding that we cannot capture content-determining inferential relations without recourse to normative vocabulary. Reasons must be intelligible as good reasons. Moreover, those to w h o m we attribute sensitivity to reasons must be such that we can see them as taking normative attitudes—treating some reasons as good, others as not so g o o d — f o r this is what, in practical terms, sensitivity to reasons comes to. Whether animals meet these conditions is much less clear. But notice again, this concern is not obviously dependent on a dogmatic assumption that sensitivity to reasons involves participation in a social practice. T h e fundamental issue is whether being a believer requires a measure o f rational self-awareness: the capacity to assess beliefs for their epistemic credentials. As we saw, Kornblith accuses B r a n d o m o f conflating the capacity to be moved by reasons with the capacity to produce reasons. However, B r a n d o m is guilty o f no such conflation. His view is rather that any belief (in the right circumstances) can serve as either a justifier or a to-be-justified. Following Sellars, he takes this to be true even of beliefs expressed by spontaneous observation reports. Observation-reporting is based on trained (hence reliable) discriminative capacities: to that extent, the epistemology o f observation-reporting has an externalist component. Further, because observation reporting is based

202

Critiques of Epistemic Externalism

on reliable reporting dispositions, reports issued by a competent speaker are typically default-credible. But they are defeasible: they can be assessed (in the light of our knowledge o f when they can and cannot be trusted), even if typically they do not need to be. Brandom is not confusing responding to reasons with producing them. His point is that rather all beliefs are fully within the space o f reasons: even those that might seem prima facie to be the most promising candidates for a fully externalist treatment. 16 What Kornblith needs, then, to justify the move from information-bearing states to beliefs, is an account o f belief-content that swings free o f Brandom's demand for a measure o f rational self-awareness, Without such an account, he has no real response to Brandom's claim that pure reliabilism suffers from a conceptual blindspot.

4. Function and Content We can now bring Davidson into the discussion. As Kornblith notes, one reason why Davidson is inclined to associate belief with language-mastery is that he is doubtful that non-linguistic evidence can fix belief-contents in any very precise way. Sometimes Davidson makes the claim that non-linguistic evidence underdetermines thought-contents, an argument that fails to impress Kornblith since all theories whatsoever are underdetermined by evidence. But Davidson is also inclined to make a stronger claim: that attributions o f thoughts to animals lack clear sense. Attributing thoughts, we can explain a person's actions. However, .. .without speech we cannot make the fine distinctions between thoughts that are essential to the explanations we sometimes confidently supply...The dog, we say, knows that its master is at home. But does it know that Mr. Smith (who is his master), or that the president o f the bank (who is that master) is at home? W e have no idea how to settle, or make sense of, these questions. 17

Clearly, in this argument Davidson anticipates Brandom's insistence that propositional content is bound up with inferential articulation. We cannot "make sense" of attributing particular thoughts to a dog because we have no idea as to what other thoughts which would have to be supposed to be recognized by the dog follow from the thoughts attributed. However, Kornblith thinks that this is just wrong. We do know how to make sense of and settle such questions. Dogs don't know about bank presidents: it is easy to show that they cannot make such discriminations. And while it may be difficult to express canine thoughts in English, it can be difficult to express the thoughts of foreigners in English too. 16

17

Here Brandom closely follows Wilfrid Sellars's treatment of observational knowledge in "Empiricism and the Philosophy of Mind," first published in Herbert Feigl and Michael Scriven eds., Minnesota Studies in the Philosophy of Science, vol. 1 (Minneapolis: University of Minnesota Press 1956). Reprinted with an Introduction by Richard Rorty and Study Guide by Robert Brandom (Cambridge MA: Harvard University Press). Also reprinted in Sellars, Science, Perception and Reality (London: Routledge 1963). References are to this version. For Sellars's account of observation reporting, see section VIII. Donald Davidson, "Thought and Talk" in Davidson, Inquiries into Truth and Interpretation (Oxford: Oxford University Press, 2nd edition 2001), p. 163. Quoted by Kornblith, KPN, p. 86.

Is Knowledge a Natural Phenomenon?

203

I think that it is significant that, in responding to Davidson here, Kornblith falls back on the idea of discrimination (elsewhere recognized as insufficient for belief-attribution), finessing the question of inferential articulation. But let that go for the moment. If not the sort of inferential articulation that goes with language-mastery, what does fix the contents of animal thoughts? Kornblith has surprisingly little to say about this crucial question. But it is not hard to see which direction he would be likely to take. His whole defence of the idea that knowledge is a natural phenomenon is conducted from the perspective of cognitive ethology, a perspective that commits him to the following claims: 1. An animal's cognitive capacities are seen as the product of natural selection. The best explanation of the animal's cognitive capacities is that they were selected for. 2. Behaviour that contributes to fitness makes certain informational demands on the animal, and the animal's cognitive capacities were selected for their ability to play this role. It is a reasonable hypothesis that, in characterizing animal knowledge, Kornblith would respect these commitments across the board. This would put him with those who, like Ruth Millikan, think that the contents of animal representations are determined by the selected-for biological function of the cognitive capacity with which those representations are involved. 18 There is a notorious problem with this "natural teleogy" approach to content: the disjunction problem. A frog responds to a fly's buzzing across its field of vision by flicking out its tongue. This is a clear example of an informational demand's being made on an animal by a biological need: in this case, the need to eat. But what is the content of the frog's mental representation: "Here's food", "Here's a fly" (since frogs will also flick out their tongues at BBs) or "Here's a small, rapidly moving dark thing"? The teleogical theorist's answer to this question is, as Fodor says, that there is a state S of the frog's nervous system such that: (i) 5 is reliably caused by flies in Normal circumstances; (ii) S is the Normal cause of an ecologically appropriate, fly-directed response; (iii) Evolution bestowed S on frogs because (i) and (ii) are true of it. 19 There is strong reason to think that Kornblith would be sympathetic to this idea, for he takes a very similar line in connection with a (related) problem that Brandom raises with respect to pure reliabilist accounts of knowledge. Brandom notes that the reliability of a process of belief-acquisition is always assessed relative to an appropriate reference class of environments. But we can construct cases in which our judgment that a method of belief-formation is reliable fluctuates with how we describe its theatre of operation. Brandom concludes that "no naturalistically statable facts" pick out a particular way

18

For a number of influential essays on this theme, see Ruth Garrett Millikan, White Queen Psychology and Other Essays for Alice (Cambridge MA: M.I.T. Press 1995). 19 Jerry Fodor, "A Theory of Content I" in A Theory of Content and Other Essays (Cambridge MA; M.I.T. Press 1992), p. 70.

204

Critiques of Epistemic Externalism

of fixing the reference class as "uniquely privileged or correct."20 Kornblith responds, reasonably enough, that in ethology we judge the reliability of an animal's cognitive capacity with respect to its natural range. Viewing animal cognition in a Darwinian perspective, an animal's natural range is the "privileged" environment, and one that is perfectly well described in naturalistic terms. That animals can be fooled (in the laboratory) is neither here nor there. Whatever the merits of this response to Brandom's epistemological point, the analogous strategy offers no response to the disjunction problem concerning content. The thought would be that frogs have fly-detectors, because, in their natural environment, flies are what frogs are after. But as Fodor says, this claim simply begs the question. We could equally well say that the frog is after little ambient black things, which in the natural environment happen to be flies. Fodor's point is that it doesn't matter what we say. Nothing in the evolutionary story requires us to characterize the content of the frogs representations in one way rather than another. Or as Fodor wittily makes the point: "Darwin cares how many flies you eat, but not what description you eat them under." 21 This point is implicit, it seems to me, in Davidson's point about canine beliefs. Davidson's point has less to do with underdetermination than with indeterminacy. And it isn't just a question of how fine-grained our content ascriptions can reasonably be. For while of course, Kornblith is right—dogs can discriminate some things and not others—what we do not need to do, in explaining canine behaviour, is worry about distinguishing what dogs "think about" from how they think about it. Indeed, if we stick with an evolutionary-cum-pure-reliabilist approach to animal cognition, there will be no way of drawing such a distinction for, as Fodor says, this approach will never discriminate between reliably equivalent content-ascriptions. 22 But if we don't have to

20

AR, p. 117. In making this point, Brandom draws on Goldman's barn-fa$ade thought-experiment. In Goldman's example, a normal perceiver, in good conditions of perception, comes to believe that there is a barn in front of him by seeing one. However, unbeknownst to him, he is in Barn Fa9ade County, where the locals have built lots of extremely realistic barn fa$ades to make the country side look more picturesque. He is thus looking at one of the very few real barns around. So does his belief that there is a barn in front of him amount to knowledge? According to Brandom, this question is an embarrassment for purely naturalistic reliabilism. Reliabilists say that, to qualify as knowledge, a belief must be formed by a reliable method. But whether a method is reliable seems to depend on how its area of operation is described. In Barn Fagade County, taking a look from the roadside is not a reliable method for determining that there is a barn in the field. In Goldman's example, that the perceiver has found himself in front of one of the few real barns is just a matter of luck, so that the truth of his belief is a matter of luck also. But if Barn Fagade County is the only county, out of a hundred in the State, that goes in for barn fakery, then maybe the perceiver's way of forming his belief is not so unreliable after all. But what if the surrounding states have followed Barn Fafade County's lead in landscape enhancement? Maybe just looking is really unreliable after all. See Alvin Goldman, "Discrimination and Perceptual Knowledge," Journal of Philosophy 73, no. 20 (1976), pp. 771-791.

21

"A Theory of Content I", p. 73. Brandom notes in passing that naturalistic programmes in semantics "are at their weakest when addressing the question of what distinguishes representations that deserve to be called beliefs form other sorts of indicating states". Brandom mentions Millikan's theory, but also Fodor's non-teleogical semantic naturalism. AR, p. 211, n.8. "A Theory of Content I," p. 73.

22

Is Knowledge a Natural Phenomenon?

205

distinguish between what dogs "represent" and how they represent it, then talk of semantic representation is simply out of place. For all that Kornblith has argued, Davidson and Brandom are right: to the extent that animal cognition is intelligible in pure externalist-reliabilist terms, we don't have to think of animals as having fully-fledged beliefs with propositional content, reflecting conceptual activity. Information-bearing states will do. A final note before leaving this topic. Davidson expresses his scepticism about canine belief-contents by choosing examples of semantically distinct, co-referring expressions ("Mr. Smith," the bank president"). This is not surprising, since his point is to suggest that there is no clear sense to the question "How does the dog conceive of his master?" Co-referring expressions also play a prominent role in Brandom's account of the inferential articulation necessary for propositional content. Substitution-inferences are very important for Brandom, for they play a central role in his account of the objectively representational aspect of language and thought. Because of differences in what they know (or believe), people think of the same things in different ways. You know, but I don't, that Franklin was the first U.S. postmaster. So if I tell you that Franklin invented bifocals, you learn (via a substitution inference) that the first U.S. postmaster invented bifocals. In learning from me in this way, you treat the content of my belief de re: you see me as believing of the first postmaster that he invented bifocals, De re attitude ascriptions—in which we identify what a person is talking about, whether or not he thinks of it in the way that we do—are the key to communication between people who do not share all their commitments. Or as Brandom puts the point, "Identifying what is being talked about allows me to extract information across a doxastic gap." 2 3 Brandom takes his account of the significance of de re attitude ascriptions to reveal the way in which the objectively representational dimension of thought and language is essentially connected with the social-perspectival character of representation. This is not the "I-We sociality" of the players of a communal game, but the "I-Thou" sociality required by the recognition that genuine believers, in virtue of their different beliefs, can entertain different conception of the same things. It is not my purpose here to defend Brandom's views of these matters, or even to explain them in detail. What I do want to suggest, however, is that Kornblith gets Brandom's view backwards. H e does not argue from the social character of the game of giving and asking for reasons to the exclusion of animal beliefs. Rather, he argues from what is involved in genuine beliefascriptions (as opposed to ascriptions of other kinds of information-bearing states) to the social dimension of inferential articulation. 24 At the same time, however, I think that Brandom's choice of words is unfortunate, having some tendency to blur the differences between himself and a naturalist like Kornblith. If we can "extract information across doxastic gaps" by using de re attitude ascriptions, why can't we bypass the problem of determining (or stating in English) the content of animal beliefs? Why can't we say of the dog that he believes oflsAr. Smith that he has just entered the room? T h e dog may not think of the person he recognizes

23 24

AR, p. 181. See especially AR, ch. 5.

206

Critiques of Epistemic Extemalism

as Mr. Smith, and he may not think of his current surroundings as exactly a room. But this just means that the doxastic gaps between people and dogs are typically larger than those between people and other people. But we can extract information form the dog's recognition of his master as surely as we can extract, from people's utterances, information that the speakers are not themselves in a position to articulate. What is wrong with using de re ascriptions across the board? What Brandom has to do here is to join Davidson that differences of belief are discernible only against a background of massive commonality. Where no such commonality exists, the possibility of interpretation lapses, and with it the possibility of seeing genuine belief. Brandom muddies the waters here—unnecessarily—by loose talk of "extracting information". This invites us to think of animal communication as much more like human communication than Brandom has any business supposing it to be. We can extract information from any mechanism, organism or process that we can recognize as being in information-bearing states. But only other speakers can literally inform us.

5. Responsibility and

Meta-Knowledge

Kornblith's defence of animal knowledge goes through two stages: first he argues that animals have beliefs, where beliefs are more than informational states; and then he argues that those beliefs (because reliably formed) amount to knowledge. So far, I have concentrated on the first stage. Now it is time to look at the second. However, the two issues are not really separate. Where Kornblith endorses pure externalist reliabilism, I myself favour a "hybrid" account of knowledge. While knowledge is a kind of justified true belief, justification has two aspects. There is first of all a procedural aspect: to be justified, a belief must be formed (or retained) in an epistemically responsible way. However, it is quite possible to conduct oneself in an epistemically responsible way and still employ a method which is, in the circumstances, unreliable. A belief formed in this way will not amount to knowledge, even if it happens to be true. So justification has a reliabilist aspect too. Knowledge requires responsibility and reliability. Such a view of knowledge meshes smoothly with the Sellarsian ideas that also inform Brandom's thinking. Kornblith agrees that the responsibility/reliability distinction is an important distinction, and used to argue for an account of knowledge along these lines.25 However, it seems clear that, in the case of animal cognition, the notion of epistemic responsibility gets no purchase. Kornblith's proposal is to drop the demand for epistemic responsibility. But I think that this is easier said than done. On Kornblith's own showing, drop-

25

Hilary Kornblith, "Justified Belief and Epistemically Responsible Action," Philosophical Review, 92 (1983), pp. 33-48.

Is Knowledge a Natural Phenomenon?

207

ping the demand for responsibility opens a yawning gulf between animal cognition and human knowledge. Kornblith takes me to hold that epistemic responsibility requires showing due respect for norms governing epistemic procedures, and that such norms are instituted by social practices. So like Brandom, I disqualify animals from the status o f knowers on the grounds that they are insufficiently (or no appropriately) social. Kornblith does not deny that epistemic norms are often socially generated and enforced. But he claims that this social grounding is not theoretically fundamental: reliability is. In fact, reliability is theoretically more fundamental than any kind o f justification (distinct from reliability itself). Epistemic norms are explained by reliability. Obviously, we have an interest in having true beliefs, for having true beliefs is important for achieving our purposes, whatever they happen to be. But why do we have an interest in knowledge, if knowledge involves more than true belief? At least part o f the answer to this question is that beliefs are typically acquired. So if we want our beliefs to be (so far as possible) true, we need our ways o f acquiring them to be reliable. In some cases, the most reliable method at our disposal will involve the careful marshalling o f evidence. But not in all. For example, knowledge is often acquired by the unselfconscious exercise o f basic cognitive capacities, such as perception or memory. Indeed, without such basic information-gathering capacities, more sophisticated strategies would get no purchase. So while justification by reasons, or through procedural correctness, can be a route to knowledge, it is not essential for knowledge. Indeed, it is important only in so far as it reflects something more fundamental: truth-reliability. Recognising this allows us to understand animal and human knowledge from a unified theoretical perspective. In fact, I don't disagree with this. I think that our ideas about justification change and should change in accordance with our interest in improving our reliability. T h e hybrid theory naturally accommodates this point. But what I really want to argue is that, as in Brandom's case, Kornblith's focus on the social practice aspect o f my view diverts attention from the really crucial issues. It is a curious feature o f Kornblith's treatment o f Brandom, Davidson and myself that he does not associate any o f us very closely with the idea that knowledge requires meta-cognition, though he does note that Davidson appears to hold such a view. Davidson holds that one cannot have beliefs unless one has the concept o f a belief. This is because a person "cannot have a belief unless he understands the possibility o f being mistaken, and this requires grasping the contrast between truth an error." 2 6 Believers recognise that their beliefs are sometimes false, and then they change their minds. This is what it is to be sensitive to reasons, thus to be a believer rather than a functionally characterized information-processor. T h e demands o f epistemic responsibility point in the same direction. Even beliefs formed by the unselfconscious exercise o f perceptual capacities are fully within the space o f reasons and fully subject to considerations o f epistemic responsibility. But we must

26

"Thought and Talk", p. 170. Quoted by Kornblith, K P N , p. 88.

208

Critiques o f Epistemic Externalism

be careful not to read too much into talk of "meta-cognition". The Sellarsian account of observation, which I share with Brandom, does not require constant self-monitoring. But it does require the capacity to allow for observational errors, and a capacity to rethink the significance of putative observational evidence, should the need arise (e.g. if we find reason to think that the conditions of observation are not standard). Thus the responsible handling of observational beliefs requires not just the concepts of truth and reliability, but an extensive knowledge of our observational capacities, particularly the errors to which they are subject, Kornblith goes wrong because he assimilates the demand for meta-cognition with the demand that knowledge be "reflective." If knowledge is to be reflective, he thinks, it must depend on constant self-monitoring: "sophisticated meta-cognition" through which we run (continual or continuous?) introspective checks on the state of our belief-system.27 While there are philosophers who seem to incline to this view, neither Brandom, Davidson nor myself should be numbered among them. What we argue, in our different ways, is that sensitivity to reasons or epistemic responsibility requires the capacity for assessing the credibility of one's commitments, revising them if need be. This capacity requires extensive knowledge of one's epistemic strengths and weaknesses: for example, the sorts of things that one is or is not good at recognising. This is the only way in which knowledge depends on meta-cognition. But it is enough to separate human knowledge from animal. I think that this is a point that Kornblith should concede. His objection to philosophers who want to make knowledge everywhere dependent on reflection (= introspection) is that, as recent work in cognitive science has shown, introspection can be wildly unreliable. Indeed, people have all kinds of inaccurate beliefs about their ways of forming beliefs.28 Cognitive science can help us to allow for our deficiencies. Thus cognitive science, not introspection or intuition, is the road to "epistemic self-improvement".29 This might seem to count against me too. But it doesn't. The sort of reliabilityknowledge that I take to be involved in the capacity for rational belief-revision is itself empirical and open to revision. Our knowledge of our epistemic capacities cannot be wildly off-base, across the board. But it can be improved, in just the ways that Kornblith suggests. However, Kornblith's own views about epistemic self-improvement show how different animal "knowledge" is from human. Consideration of the demands of epistemic responsibility helps clarify the ways in which knowledge is and is not a social undertaking. To become sensitive to reasons is to learn what can and cannot reasonably be questioned, and why. Thus acquiring sensitivity to reasons requires a lot of know-how, as well as extensive meta-knowledge. To become a knower is to become acculturated. This is not something that one could, even in principle, do for oneself. At the same time, epistemic responsibility makes demands on individuals. The aim of acculturation is to make me responsible for my beliefs: to

27 28 29

K P N , ch. 4. K P N , p. 11 If. K P N , p. 1 2 0 f .

Is Knowledge a Natural Phenomenon?

209

teach me enough about myself and the world for me to be able to take responsibility for the commitments I acquire, revising them when appropriate. On Kornblith's own showing, animal "knowledge" is nothing like this. Indeed, from Kornblith's ethological standpoint, knowledge is much less an individual matter than it is for social theorists like Brandom, Davidson and myself. O n Kornblith's view, knowledge "first enters our theoretical picture at the level of understanding of the species". This is because, for Kornblith, animal knowledge (as opposed to belief) is to be understood in terms of capacities which, in virtue of their informational reliability, have been selected for. We might say, animals don't need the capacity for epistemic assessment because they don't test hypotheses: they test themselves. But this is why they are not truly sensitive to reasons. They cannot really change their minds, though the information-acquiring and processing capacities of the species can change over time. Pure reliabilism attracts philosophers of Kornblith's temper because it seems to offer a way of naturalizing. But this is not a point in its favour, since epistemology cannot and does not need to be naturalized. Animal cognitive capacities change with changes in the natural environment. But our views about how inquiry is best conducted—in common life and in various special disciplines—change along with our understanding of the world itself. Such changes cannot be anticipated or reduced to rules. So to claim that knowledge need not be naturalized does not mean that it should be understood in some "anti-naturalist" (let alone supernaturalist) way. It is to claim, rather, that although we can illuminate the concept of knowledge, and explain why it matters to us to have a concept like that, 30 knowledge itself is not a proper object of theory, scientific or otherwise.

30 This approach to knowledge is pursued in a very illuminating way by Edward Craig in Knowledge and the State of Nature (Oxford: Oxford University Press 1990).

III In Defence of Content Externalism

Social Cartesianism J O H N HAUGELAND

I begin with brief, critical overviews of three well-known arguments in the philosophy of language, due respectively to Nelson Goodman, W. V. O. Quine, and some amalgam of the later Wittgenstein and Saul Kripke. I then maintain that the failings in these three arguments have a common origin. And I conclude by suggesting that this common origin is a modified legacy of Cartesianism—a legacy that Heidegger, by contrast, managed to overcome

1. The New Riddle of Induction In Fact, Fiction, and Forecast, Goodman considers two problems concerning induction. The first is the traditional logical/epistemological problem of justifying our formal canons of good inductive inference. This problem he "dissolves" by treating it as a matter of defining a general term—namely, 'good inductive inference'—by a process of accommodation between theoretical considerations and intuitively clear positive and negative instances. (§§ 3.2 and 3.3) Given that approach, the issue reduces to "the constructive task of confirmation theory", which can be addressed with standard syntactical tools, with reasonable expectation of progress. The second problem, which Goodman dubs "the new riddle of induction" (§§ 3.4 and 3.5), has to do not with the legitimacy of formal inference rules, but rather with the legitimacy of the inductive hypotheses and predicates to which those rules are applied. Some hypotheses and predicates are, as he puts it, "lawlike" and "projectible" respectively, whereas others aren't; and only the former are suitable for use in inductive inferences. For example, the observation that one piece of copper conducts electricity does support the inductive hypothesis that other pieces of copper also conduct electricity; but the observation that one member of a lecture audience is someone's third child does not support the general hypothesis that the other members of that audience are also third children. The former hypothesis is lawlike, and its predicates projectible·, but the latter are not. So far, however, these are still just names. The difference between the two examples may be clear enough at an intuitive level; but providing a general philosophical account of it is another matter. That's the new riddle of induction. We can see why this riddle is new—that is, not just a variant of the traditional problem·—by turning to Goodman's most famous illustration: the choice between the familiar predicates 'blue and 'green, on the one hand, and the contrived alternatives 'grue and 'bleen on the other. The latter are defined in terms of the former and some specified time, t (no earlier than the present), as follows:

214

In Defence of Content Externalism

Something is grue just in case, either it has already been examined by t and is green, or it has not yet been examined by t and is blue; (and, contrariwise, for 'bleeri). It's essential to be clear about how these alternative predicates work. To project the predicate 'grue'—by adopting, for instance, the hypothesis that all emeralds are and will be grue—is not to predict that the emeralds we already know and love will turn blue at t. No, for them to be (and stay) grue just is for them to be (and stay) green—because it will always be the case that they were first observed before t. The 'grue' prediction diverges only for "new" emeralds, ones that have never been seen before t. To predict that they, too, will all be grue-—"just like" their predecessors—is to predict that they will all turn out to be blue (and presumably were blue all along, even before they were observed). That's a bony prediction, of course. Nobody denies that. But, the philosophical question is: Whyi What, exactly, is loony about it? The question arises because, given the definitions, it would seem that, at or before t, every confirming instance of "All emeralds are green" is equally a confirming instance of "All emeralds are grue". So induction itself is neutral between them. To be sure, the 'grue' hypothesis is not really confirmed by old green emeralds—even though, by definition, those old emeralds are also all grue—because the predicate 'grue' is not projectible, and the hypothesis is therefore not lawlike. But that's merely a fancier formulation of the same issue. In this context, 'non-projectible' is just a big Harvard word for loony; and that's what we want to understand. Now, what makes this a new difficulty is that—unlike the difference between good and bad inference rules—the difference between these two hypotheses, and their putative relations to instances, cannot be drawn syntactically. Their logical forms are all exactly the same. In other words, what's new about the "new riddle" is that syntactical methods are powerless against it. Accordingly, we need some non-syntactical criteria for projectibility (and hence, lawlikeness). Goodmans own proposal is subtle and ingenious. What distinguishes legitimate (projectible) inductive predicates, he says, lies in the history of actual projectings— that is, actual adoptions of hypotheses containing the predicates in question (or synonyms of them). Thus, to the question of what rule will "make the proper choice" between "All emeralds are green" and "All emeralds are grue", he replies that: ... we must consult the record of past projections of the two predicates. Plainly green', as a veteran of earlier and many more projections than grue', has the more impressive biography. The predicate 'green', we may say, is much better entrenched than the predicate 'grue'. We are able to draw this distinction only because we start from the record of past actual projections. (94)

As this suggests, and as becomes explicit in the following paragraphs, the idea is that a predicate is more projectible to the extent that it is better entrenched; and, to a first approximation, a predicate is better entrenched to the extent that there have been more projections in the past of hypotheses containing it. This first approximation, however, needs—and receives—considerable refinement. Predicates do not, in fact, garner entrenchment from all prior projections containing them, but only from those that are supported (by some positive instances), not violated (by any negative instances), not exhausted (by having all their instances checked), and

Social Cartesianism

215

not overridden (by incompatible hypotheses projecting still better entrenched predicates). What's more, predicates can inherit entrenchment from other well-entrenched predicates by being related to them in various logical and theoretical ways; and so on. The upshot is that the degree of entrenchment of a predicate reflects and encodes a good deal of actual prior experience—in the social history of empirical practice. But it is important to appreciate how this social/empirical pedigree works. The fact that a given predicate has been successfully used in the past (along with other pedigreed predicates) is not taken as evidence that it is legitimate. For that would amount to establishing its legitimacy inductively, which would be out of order for at least two reasons. In the first place, there would clearly be something fishy about using induction in an account of inductive legitimacy; but second, and more to the point, the aim is not to ascertain or establish the legitimacy of inductive predicates, but rather to explain what such legitimacy consists in. Entrenchment does not show or confirm the legitimacy of a predicate, but rather constitutes it. That Goodman understands entrenchment as constitutive of legitimacy, and not just as a sign or manifestation of it, is made explicit in his answer to a pair of rhetorical questions a few pages later. Must we not explain why ... the really projectible predicate happens to have been the earlier and more often projected? A n d in fact wasn't it projected so often because its projection was so obviously legitimate, so that our proposal begs the question? I think not. ... T h e reason why only the right predicates happen so luckily to have become well entrenched is just that the well entrenched predicates have thereby become the right ones. (98)

Distressingly, however, this passage runs together two quite different issues, implying that the same response addresses both. One question is why only the right predicates get entrenched; and the constitutive answer that entrenchment is what makes predicates "right" is at least to the point. But the other question is why some predicates "happen" to become entrenched, while others don't; and the reply doesn't speak to that issue at all. This is not a minor quibble: for, were it to emerge that som e further factor systematically explained which predicates get entrenched and which don't, then this further factor might well offer a deeper account of predicate legitimacy itself, even granting the constitutive point. To see how such a story might go, return to the opening gambit with 'grue' and 'bleen'. A natural objection is that they—unlike 'blue' and 'green'— are defined in terms of a specific, arbitrary time. But Goodman counters that the situation is perfectly symmetrical: 'green and 'blue' can be formally introduced and explained in terms of 'grue', 'bleen', and that same arbitrary time just as easily as the other way around. (74, 79f) The trouble is: that reply doesn't work. The symmetry Goodman invokes is merely formal; the situation is not really symmetrical at all. For it is simply not true that—in the real world—'green' and 'blue' could ever be introduced or explained in terms of'grue' and 'bleen. No one could ever first come to understand the words 'blue' and 'green' on the basis of a prior understanding of'grue' and 'bleen' (plus some formal definitions), for the obvious reason that no one could ever come to understand the words grue' and 'bleen' (in the intended senses) except by already understanding 'blue' and 'green', and then being given those definitions.

216

In Defence of C o n t e n t Externalism

Though I take that to be self-evident, sketching an argument for it may still be useful, as a step toward seeing what underlies it. So imagine trying to teach the words 'grue' and 'bleen' to a child old enough to learn color words, but not old enough to understand the concept of when something was or will be first observed. If you proceed in the ordinary way—How else?—by showing her various examples, and coaching her on which ones to call what, then, if it's before t, she'll certainly learn the words; but she'll just take them to mean green and blue. (They're the same lessons, after all.) After t, on the other hand, the task is hopeless. If you start with 'grue', and show her both old and newly discovered samples, she may take the word to mean a blue-green range of colors, or perhaps a disjunction. But when you introduce the bleen contrast set (also with old and new samples), everything falls apart. One emerald is grue, but another—that looks just the same to her—is not; likewise for sapphires; and so on. (In fact, unless they're distinguishable on other grounds, you can't tell the grue from the bleen stones either.) An essentially similar point can be made about measuring instruments. There are many ways to build a machine that can reliably distinguish green things from blue ones, without any information about what time it is, or when the samples were first observed. But there is no way to build a machine that, after t, could reliably distinguish grue from bleen, subject to the same constraints; for it can no more tell an old stone from a newly discovered one (without extra information) than can you or your young pupil. It seems to me that these simple facts about learning and instrumentation are not only a likely explanation of why predicates like 'green' and 'blue' (as opposed to 'grue' and 'bleen') are entrenched, but also the key to developing a more basic insight into predicate legitimacy itself.

2. The Indeterminacy of Translation In sections 12 to 16 of Word and Object, Quine argues for an ineliminable empirical indeterminacy in the translation of what he calls "terms of divided reference" (91—95) and the concomitant "apparatus of objective reference" (53, 61, 70). The former, now often called "sortals", are words like 'rabbit', 'chair', 'emerald', and 'person' that (as we ordinarily understand them) denote distinct, enduring individuals—entities that we can keep track of, count, distinguish one from another, and reidentify later (in the sense of: "the very same one"). Correlatively, the apparatus of objective reference comprises the pronouns and quantifiers, along with various devices for expressing unity, plurality, number, numerical identity, distinctness, and the like. The alleged indeterminacy can be introduced as follows. Suppose we have a way of translating some language (call it the "away" language) into our own ("home") language that is fully satisfactory by all empirical criteria; and suppose further that our translations often use sortal nouns and the apparatus of objective reference. Then there are also other ways to translate that language into ours—systematically related to the first, but intuitively quite different—that are equally satisfactory by all empirical criteria. But if they're all equally satisfactory, then there can be no empirical grounds for choosing

Social Cartesianism

217

among them. Quine concludes that, although the translations are different, there is no fact as to which o f them is the "right" one, because there is no "objective matter to be right or wrong about" (73); and that's the indeterminacy. Before considering the argument for this conclusion, it is important to be clear about when and where it is supposed to apply. T h e indeterminacy o f translation has nothing to do with how "different" or "alien" the away speakers or their language might be (as compared to us and our own). On the contrary, "home" and "away" can be as close as Minneapolis and St. Paul, the "strangers" as familiar as the people next door. Keeping that firmly in mind, then, here's the general argument strategy. Insofar as we have an empirically satisfactory translation scheme that uses sortal nouns and the apparatus o f objective reference, we can generate an intuitively different scheme by systematically altering how those nouns and that apparatus are used, but in such a way that the alterations "cancel each other out"—at least as far as empirical adequacy is concerned. Thus we get alternative translations that are intuitively different but empirically equivalent. So, adopting the famous example, suppose a rabbit scurries by, and (while pointing at it) the away speaker says "Gavagai." According to one translation method (which, by hypothesis, is systematically satisfactory across the board) this means "Lo, a rabbit." But, as Quine notes, no one can point at (or see, touch, stuff in a bag) a rabbit without also pointing at (seeing, touching, stuffing in a bag) a temporal stage o f a rabbit, an undetached part of a rabbit, a portion o f the spatio-temporal fusion of all rabbits, and a concrete manifestation of the abstract universal rabbithood—and likewise, of course, for apples and emeralds, ducks and people, (cf 51 if) Thus, it seems, we could adopt an alternative—equally good—translation method according to which our sample utterance means "Lo, a temporal-rabbit-stage" or "Lo, an undetached-rabbit-part". Those substitutions alone, however, wouldn't quite work. If Inge goes to market and says something that means (according to our first method) "Please give me two rabbits", this could not be rendered equally well as "Please give me two temporal-rabbit-stages" or " . . . two undetached-rabbit-parts". For a single rabbit could satisfy either of those requests—yet we can be confident that Inge herself would not be satisfied. Still worse, asking for two rabbit fusions or two rabbithoods wouldn't even make good sense. But these only show that our alternative methods have to be a little fancier. In particular, if we use a different type o f noun, we must also use (in compensation, as it were) a correspondingly different apparatus o f objective reference. Thus, if 'rabbit' is to be replaced with either 'temporal-rabbit-stage' or 'undetached-rabbit-part', then 'two ...s', must be replaced with something like 'two standardly-coherent-concrete- units-of ...s'. On that method, then, Inge's request becomes: "Please give me two standardly-coherent-concrete-units-of temporal-rabbit-stages" or " . . . undetached-rabbit- parts"—which does, at least, convey what she wants. Finally, if the replacement noun were 'the rabbit-fusion' or 'rabbithood', the required compensation would again be the same, but without the final's'. It is immediately striking that, in all four of Quine's alternative translations, the needed "compensatory adjustments" are essentially the same. On reflection, however, it's not hard to see why this should be so. What Quine's alternative nouns are alternatives

218

In Defence of Content Externalism

to is sortals ("terms of divided reference"). But these are just those nouns the denotata of which come in something like "standardly coherent concrete units"— which is to say, distinct, durable individuals. When Inge says (something equivalent to) "two rabbits", she means, of course, two of these standard rabbity individuals. In English, that intention is implicit in the sortal noun 'rabbit' (especially when used with a numeral and plural ending). But Quine is surely right that languages don't all have to express this intention in the same way. What is common to his four alternative renditions of'gavagai' is that, unlike 'rabbit', they do not have any standard mode of individuation built into their senses as nouns. Accordingly, whenever such individuation is important to what's being communicated—such as in a request for two rabbits—it would have to be indicated in some other way. That's what an auxiliary construction like "standardly-coherent-concrete- units-of" (or, more likely, some equivalent particle or inflection) would provide. And that provision is what is meant by a compensatory adjustment in the apparatus of objective reference. What are we to make of this? It is well known that, although all natural languages have ways of indicating tense, case, aspect, and so on, the surface structures of these indications can differ quite widely. What Quine has shown, in effect, is that the surface structures for indicating reference to standard wholes or units—concrete individuals— can differ as well. What's particularly noteworthy, however, is that both the argument and its conclusion take for granted that all the languages in question do have some way or other for indicating reference to individuals. Thus, insofar as this new variability in surface structure is like the others, we can also say that affording reference to individuals is itself a linguistic universal—at least for human languages. The discovery of this deeper universal, which Quine evidently presupposes, would seem to be at least as important philosophically as the surface variability in how it can be manifested.

3. Private

Language

In Wittgenstein on Rules and Private Language, Kripke interprets Wittgenstein's Philosophical Investigations as posing a "sceptical paradox" and then proposing a "sceptical solution" to that paradox. A sceptical paradox is an argument to the effect that we do not (or cannot) have something without which we could not live as we do. (It's paradoxical because we obviously do live as we do.) There are two possible ways to resolve such a paradox: what Kripke calls a straight solution refutes the argument that we don't have the something in question; a sceptical solution, on the other hand, accepts that we don't have it, but then shows how we can perfectly well live without it. According to Kripke, what Wittgenstein argues we can't have, but ultimately don't need, is privately determinate meanings, or (what comes to the same thing) privately determinate rules for the correct use of our words. I will not be concerned with whether Kripke's interpretation of Wittgenstein is satisfactory; nor will I pay much attention to the argument that leads to the sceptical conclusion. Rather, I will focus mostly on the proposed sceptical solution to the paradox—and that, indeed, only in connection with the example that Kripke discusses most

Social Cartesianism

219

fully—namely, arithmetic terms. Even within these limits, however, it is necessary to be clear about what the paradox is supposed to be. Though Kripke doesn't introduce it quite this way, the difficulty can be seen as arising from the following plausible line of thought. A person can use his words in any way he wants to. If he decides to use the word 'duck'—in his own private way—to mean bucket, he is free to do so, and no one can stop him. ('Private', here, does not mean secret or covert; it just means belonging to a single person—like "private property".) If that much is right, however, then, even in the ordinary case, an individual's meanings are ultimately up to him or her. Thus, if I use the word 'duck' to mean what everybody else means by it, that just shows that I have chosen—tacitly and unconsciously, perhaps—to go along with the crowd. Since I could have decided to mean something else by that word, my not doing so still amounts to a choice—by default, so to speak. Therefore, all meanings must rest, in the end, on individuals' decisions, deliberate or otherwise, and are, in that sense, private. The sceptical argument, however, shows that this is impossible. The meaning of a word does not merely determine how it is used, or likely to be used, but how it ought to be used. If 'duck' means duck, then calling a bucket a duck is not just odd, it's wrong—a »muse of the word 'duck'. In other words, meanings, by their very nature, are normative rules. And the essential problem is that individuals cannot impose norms on themselves. For that would be like taking a dictator, with absolute legal authority, to be bound by her own law. But she can't really be bound by her own law since, given her authority, if she changes her mind and does something different, that just changes the law—which is equivalent to saying that the law didn't bind her in the first place. Similarly, but on a smaller scale, an individual cannot, on his own authority, bind himself by his own private norm. In sum: if meanings must be normative, but individuals can't impose norms on themselves, then private, individual meanings are impossible. That's the sceptical conclusion. But if, as the preceding "line of thought" seemed to show, all meanings must rest, in the end, on private decisions, then the two arguments together imply that there can be no meanings at all—which is paradoxical. According to Kripke, Wittgenstein accepts the sceptical conclusion, but denies the paradox, because meanings and linguistic norms are not private at all. Rather, they are both essentially social. Kripke illustrates the point with an extended discussion of arithmetic terminology—in particular, the word 'plus'. He invites us to imagine that he (Kripke), an experienced and competent adder, happens never to have added any integer larger than 56. (Certainly, some integer is the largest he's ever added.) So, by hypothesis, if he is now given the problem "68 plus 57 equals ...", this will be completely new to him, in the sense that he's never actually computed it before. And the question we are to address is: how ought he to respond to this new problem, given what he himself has hitherto always meant by the word 'plus'? The question is not about arithmetic itself, of course, but about meaning. More specifically, it's about how Kripke's own private mental history (understandings, intentions, decisions, etc.) can determine how he ought to respond now, in order to remain consistent in what he means by 'plus'. We did allow, of course, that he is an experienced adder; and we can allow further that, in consequence of that experience, he has quite definite dispositions, such that

220

In Defence of Content Externalism

he will reliably and confidently respond to the above problem with "125". But how he will and would respond to various problems is a different matter from how he should respond—and it's the latter that we're asking about. Obviously, that depends on what 'plus' has meant to him; but that's just where we came in. To make the issue vivid, Kripke introduces a "bizarre sceptic" who suggests that maybe his response to the new problem ought to be "5". To facilitate discussion, he also introduces a new integer function, called quus, which has the same value as 'plus' for all argument pairs both members of which are less than 57, but has the value 5 for all other pairs. So, by our hypothesis, every time Kripke has computed the plus function so far, he has also computed the quus function—because, for all the problems he has actually solved, they coincide. Given that, the sceptic's suggestion can also be put this way: maybe what Kripke has meant all along was not plus but quus. Certainly the mere fact that he used the word 'plus' cannot, by itself, settle what he meant by it. Well, then, what does settle it? The short answer (after a long argument) is that nothing about Kripke himself—mental or dispositional, past or present—suffices to settle what the words he uses mean. That is, nothing about him as an individual— nothing private in that sense—determines how he ought to use or understand words. Rather, as has already been mentioned, the norms that govern correct usage are, and can only be, social. Hence, in particular, the norm governing the correct use of 'plus', such that the correct evaluation o f ' 6 8 plus 57' is 125, is social. This doesn't mean, of course, that the values of integer functions are determined by society—mob rule in arithmetic, as one might put it—but rather that, by determining what our words mean, social norms determine which functions we have words for. This account, however, cannot be satisfactory as it stands. For the quus example raises not only the issue of how meanings can be determinate but also the issue of how they can be non-arbitrary. The social-norms account may well address the former; but it doesn't speak to the latter at all. Thus, for all Kripke's Wittgenstein says, there could perfectly well be a community in which there is a norm-governed word meaning quus and none meaning plus—an implication Kripke even seems to endorse: T h e set of responses in which we agree, a n d the way they interweave with our activities, is o u r form of life. Beings w h o agreed in consistently giving bizarre quus-like responses would share in another f o r m of life. By definition, such another form of life w o u l d be bizarre and incomprehensible to us. ... However, if we can imagine the abstract possibility of another f o r m of life (and no a priori a r g u m e n t would seem to exclude it), the members of a c o m m u n i t y sharing such a quus-like f o r m of life could play the game of attributing rules and concepts to each other as we do. [p. 96]

I don't know what to say about abstract possibilities for unspecified "beings"; but I don't believe we can imagine, concretely, any "quus-like" human civilization here on Earth. And I am encouraged in this negative opinion by the historical fact that, although the numerous known civilizations have differed widely in many ways, they nevertheless all had systems of arithmetic, and they all agreed exactly on every single sum, difference, product, and quotient that they had numerals to express. This, I suggest, cannot be mere coincidence.

Social Cartesianism

221

4. What's Missing in these Pictures? I now propose to show that the three arguments just reviewed are all deficient in the same way: in discussing language and meaning, they all leave out of account the same essential factor. That factor is the world itself—the real world that we all live in. Consider first the predicates 'grue' and 'bleen'. No one ever has or ever will sincerely project those predicates. Why is that? Its not beside the point that—apart from the odd philosopher—the very idea would never occur to anyone. But why wouldn't it? The above remarks about teachability are surely relevant: it's simply not possible to learn the predicates 'grue' and 'bleen' except by first learning 'green and 'blue', and then grasping the switchover at t. But notice: this is not a claim about how the mind works, nor about language or semantics. (The companion point about machines shows that much—since they don't have any minds or language.) The indistinguishability, after t, of old grue emeralds and newly discovered bleen ones, is not a fact about the mind or language; it's a fact about the world. It may seem that this begs some question, by taking it for granted that 'green' is projectible and 'grue' is not. I do take that for granted, of course—we all do—but it doesn't beg any question. The issue was never whether 'green' is a legitimate inductive predicate, as opposed to 'grue', or even whether we can tell, but rather why. Goodman was not asking how we know that the one is projectible and the other not, but rather what makes that the case. This is the only question that the constitutive, entrenchment account could so much as address. What's wanted is not a vindication but an explanation. And the real answer to that explanatory "Why?" question is that no genuine kind of thing will be consistently grue (for any given t), whereas many kinds of thing—emeralds, chlorophyll, copper chloride—are and will be consistently green. It doesn't matter for the explanation whether we know this general fact about genuine kinds; it only matters that it's true. It's this fact—whether we know it or not—that makes 'green' a legitimate inductive predicate, and 'grue' a loony one. (Or, if it's not a fact, then some other predicate is the legitimate one, and we'll have to find out the hard way.) The reason that there are projectible predicates, the reason that induction works at all, is that worldly entities—at least many of them—instantiate genuine kinds. To a first approximation, a genuine kind is a consistent and reliable collocation of structures and properties, and/or combinations thereof. Emeralds are a genuine kind because a certain crystal structure, composition, color, hardness, heat capacity, and so on are reliably collocated in them. Our own perceptual organs, and the sorts of measuring instrument I mentioned above, also depend on genuine kinds in their own internal components. (The molecules in our retinal cones that enable them to respond selectively to different colors are instances of genuine kinds.) Finally, given that there are such kinds, there can be consistent and reliable kinds of interaction—including, for instance, those that make perception, measurement, and science possible. All of this has to do with the actual things themselves. It does not depend at all on the social history of language use. Accordingly, and contra Goodman, predicates are not legitimate because they have become entrenched; at best, they become entrenched

222

In Defence of C o n t e n t Externalism

because they are legitimate. And they are legitimate—to the extent that they are— because they track the real world. The indeterminacy of translation, in the (1960) version considered above, turns out to be a less startling thesis than it might first have seemed. It can be seen as a new variation on the familiar theme that unrelated languages can serve the same ends—provide the same communicative facilities—in notably different ways. And, given that way of seeing the point, the underlying assumption that the end or function in question is a linguistic universal can begin to look like the more important discovery. Quine himself doesn't actually discuss the matter from this point of view; but we can. Why is it that all natural human languages afford ways of referring to and talking about concrete individuals? No sooner is the question asked, I think, than the answer is obvious. Individuals are everywhere, and they are important to all human forms of life. In the first place, individuating people is prerequisite to the social training (and norms) that make public language possible. (Word and Object begins with the sentence: "Language is a social art.") But further, the example of Inge and the two rabbits is clearly representative of countless transactions, commercial and otherwise, that we all engage in every day. And, as we have seen, this presupposes a standard individuation of those rabbits that is both mutually understood and unambiguously invoked in the verbal request. Inasmuch as such social training and transactions are ubiquitous in human life, every natural human language must facilitate them—in, as Quine points out, one way or another. It is important to realize that this significant prevalence of individuals is neither a consequence of individuative language, nor a necessary truth. If, for instance, we were slimes, living out our amorphous lives in some semi-liquid goo, with no sharp boundaries or stable forms, we would have neither use nor opportunity for any terms of divided reference. Even if, somehow, factual communication were possible and worthwhile under such conditions, it could be nothing like any actual human language—and precisely because the relevant facts are so different. By the same token, the real reason that divided reference is both feasible and valuable in all human languages is ultimately not a fact about language as such at all. Rather, it's a contingent fact about the world we live in, and what matters to us in it. It's not hard to see why there's never been a civilization that didn't do arithmetic in basically the way that we do—at least as far as they got. You can't build or maintain a city without a good deal of arithmetic calculation; and it won't work if you don't get the right answers. What we need to consider, therefore, is what makes those answers the right ones. Arithmetic calculation is initially important in connection with counting and measurement. Only three simple points need to be made about counting. First, what are counted, in the most primitive and fundamental case, are the same distinct, durable individuals that we refer to with terms of divided reference. Second, in order for counting and arithmetic to make sense, you've got to be able—in practice—to delineate sets of individuals, and also a few relations like disjointness, subsets, and unions. And, third, once you've got those things, along with counting itself, the addition function—the one that's

Social Cartesianism

223

worth having—is already determined. If you've got a basket of 68 apples, and you add 57 more, you're just going to get 125 apples—no matter what anybody's norms are. Measurement is similar, but more interesting. First, of course, once measurement of lengths, weights, volumes, and the like is well defined (in terms of standard, iterated units), measurements of relevant combinations are just as predetermined by those of their components as are the counts of the apples. 68 inches end-to-end with 57 just is 125 inches—again regardless of anybody's norms. What makes this more interesting than counting is that the additivity of measurements depends on what's being measured. Not all measurements—not even all those expressed numerically—are additive. Α pH value of 8, for instance, is not twice as acidic as a pH value of 4; nor is a hardness of 8 on the Mohs scale twice as hard as 4. But the most revealing case is temperature: 8 degrees Celsius is not twice as hot as 4 degrees, in any useful sense. But, with Lord Kelvin's discovery of absolute zero, it was also discovered that there are scales on which temperature is additive: 8 degrees Kelvin is indeed twice as hot as 4 degrees. What this relatively recent discovery reveals—and precisely because it is not lost to history—is that the possibility of additive measurement scales is a contingent fact, depending on what is being measured. Ancient peoples must have discovered this fact pretty early about distance and perhaps time; and they would need to have known it also for weight, volume, and maybe area, before they could build and manage cities. Two conclusions matter here. First, like the prevalence of durable individuals, the additivity (given the right scale) of some measurable phenomena (but not others) is not a fact about society or language but a fact about the world. And, second, once the relevant facts were discovered, it was—as history attests—clear to everyone that there is only one right way to add. It remains only to say what 'right' could mean in that conclusion. I certainly do not deny that social norms are prerequisite for all human languages, nor that these norms support and explain an essential right/wrong distinction in regard to usage. My point is rather that this does not go far enough. In addition to the issue of what is right or wrong according to a given body of social norms, there is an issue of the right or wrong norms for a society to have. In connection with arithmetic, at least—and quite possibly other things—having any norms other than the ones we in fact have would quite clearly be a mistake. And this, I have argued, follows from the determinate, contingent character of the world we actually live in.

5. Diagnosis and Conclusion When Descartes more or less launched modern philosophy in 1642, his most fateful move was not the evil demon hypothesis and hyperbolic doubt, nor the ontological proof and the argument that God is not a deceiver, nor even the cogito and the criterion of clarity and distinctness. No, the fateful move was an unstated assumption on which all of these depend. That assumption is that, if our ideas exist at all, they are fully determinate in and of themselves, quite apart from anything else that there may or may not be. The determinacy that matters here concerns not the formal reality of

224

In Defence of Content Externalism

those ideas (roughly their ontological character as modes of thinking substance), but rather their objective reality (roughly their intentional content as representations). Taking the latter to be "internally" determinate in this way is the defining assumption of classical internalism. No one could accuse any of Goodman, Quine, or Kripke's Wittgenstein of being internalists in this classical sense. Quite the opposite: they are among its most incisive and effective critics. Important differences notwithstanding, they are alike in arguing that society and public language are prerequisite to the possibility of any distinctively human meaningfulness or intentional content at all. And society is certainly not internal in the way that Descartes meant. Insofar as there is an element of scepticism in these later positions—only Kripke frames the issue with that term—it is not traditional scepticism about the possibility of knowledge, but a new-fangled scepticism about the possibility of legitimate, determinate meanings for individual people considered in isolation. Thus, it is scepticism about the very assumption that made Cartesian scepticism possible. In other words, the common conclusion that meanings are legitimate and determinate—to whatever extent they are—only in a historical, communal setting, makes Cartesian scepticism in its original form not just false but unintelligible. That's why none of these authors is a classical internalist. Yet, if I am right, there is a peculiar irony in what actually results. For I have argued that each of them still neglects an essential element in the determinacy-cum-legitimacy of these admittedly public meanings. That further essential element is the world itself. The irony is that, inasmuch as they systematically deny the world a role, they do not so much demolish the Cartesian barrier as merely shift it "outward" a notch. That expanded inner/outer perimeter now encloses not only private thoughts and feelings, but also public language and social norms; but it still excludes the real things and happenings that those thoughts and conversations are primarily about. Commitment to remaining within this enlarged cordon sanitaire, when considering questions about meaning, is what I mean by "social Cartesianism". As I have argued in the preceding sections, its effects are still philosophically distorting and pernicious. It still amounts to a kind of immanence/transcendence distinction, privileging a new social-linguistic version of "subjectivity". In other words, "social" though it may be, its still Cartesianism—and the epithet is still pejorative. In a famous footnote (KrV: Β XXXIX), Kant calls it "a scandal of philosophy and human reason in general" that there is no proof of "the existence of things outside us". Referring to this passage, Heidegger says: The "scandal of philosophy" is not that this proof is still missing, but that such proofs are expected and attempted again and again. ... Dasein, properly understood, defies such proofs, because, in its being, it in each case already is what subsequent proofs hold must first be demonstrated for it. (SZ 205)

Although it is now the legitimacy and determinacy of meaning that are held questionable, the scandal is still with us, and still essentially the same. For what dasein in each case already is, in its being, is being-in-the-ivorld as a unitary phenomenon. That means that individual people, everyday social living (including talking), and the everyday world

Social Cartesianism

225

are first intelligible only as a unity—that is, as an integrated whole. Only on the basis of that prior whole, can those three respective moments be singled out for even relatively focussed attention. Since there is no possibility of adequately understanding any of them apart from both of the others, there is no more prospect of a philosophically defensible cordon around two than around just one. Acknowledgments: I am grateful for discussions of this paper at Michigan State University and Bates College, and for detailed critical comments from Michael Kremer, Jason Bridges, and, above all, Joan Wellman.

References Goodman, Nelson. Fact, Fiction, and Forecast, third edition. Hackett Publishing Company, 1979. Heidegger, Martin. Sein und Zeit, eleventh edition. Max Niemeyer Verlag, 1967. Kripke, Saul A. Wittgenstein on Rules and Private Language. Harvard University Press, 1982 Quine, Willard Van Orman. Word and Object. The M.I.T. Press, 1960.

Existence Proof for a Viable Externalism R U T H G A R R E T T MILLIKAN

Externalism, as I am understanding the term, is a thesis about the nature of thoughts, as distinguished from language. For example, Kripke's suggestion that the referent of a public language proper name is determined by its history is not, just as such, an externalist thesis. On the other hand, Putnam, in "The Meaning of 'Meaning'," did seem pretty clearly to be talking about the nature of our thoughts of such things as water, beech trees and aluminum, perhaps as well as about language. And his essay is often taken as the original and also the paradigm defense of externalism. Putnam argued that a person's internal psychological state does not determine the referent or extension of that person's thought. Hence, if what a person means or intends were determined solely by that person's psychological state, what a person means or intends would not determine the referent or extension of his thought. Denying the consequent, Putnam concluded that what a person means or intends with a thought is determined by more than that person's internal psychological state. As he put it, "Meaning just ain't in the head!" If we explain the externalist idea in this crude way, however, it becomes hard to see how anyone could deny it. If the question were, merely, how are the referents or extensions of thoughts determined, it seems patently obvious that nothing inside someone's head could, by itself, determine that anything in particular existed outside the head. Referents and extensions are existent things and existent sets. What happens to exist or not exist, when and where, outside one's head is surely a contingent matter. How could what is inside a head determine that anything at all had to exist outside that head? But if not, how could it determine, all by itself, that its thoughts had referents and what these referents were? Something has been stated wrong. Externalism should not be so obviously true. One's first thought here may be that Putnam has denied the wrong premise. What seems obvious is that meanings, taken alone, without adding the world, do not determine extensions. But how has it been shown that what is in one's the head does not determine what one means? Let us take a second look. What exactly did Putnam's arguments show? What he seems actually to have argued is that what is in the head, when combined with what is in the world, does not determine the reference of a thought in the way that was classically supposed. He argued this, specifically, for the case of thoughts of natural kinds. Classically, it was supposed that what was in the head picked out a set of general properties, and that the world determined what, if anything, had these properties. If anything in the world did have these properties, it was part of the extension of the natural kind thought. Putnam argued that natural kind thoughts were, instead, "indexical." By this he apparently meant two things. First was that the extension was determined by

228

In Defence of C o n t e n t Externalism

some sort of concrete existential relation between the thought and (parts of) its extension, such as a causal, temporal and/or spatial relation. (Water is the so-looking and so-feeling stuff around here.) Second was that the relation between the thought and its extension that detemined this extension was not determined by being represented in the mind. This relation was not determined by being thought of. Now many have felt that there is nothing in Putnam's arguments to support the second of these theses. See, for an especially clear statement of this, (Fumerton 1989). Be that as it may, surely there is nothing in Putnam's arguments, at least, to show that there is nothing at all within the head that determines which relation the extensiondetermining relation shall be. Putnam's slogan "Meaning just ain't in the head," if we take "meaning" to be whatever, taken along with the world outside the head, determines extension, does not seem to be supported by his arguments. Hence a very common reaction to Putnam's arguments has been, exactly, to claim that what is in the head determines what kind of relation must exist between a certain thought and its referent or extension for that to be its referent or extension—though it does not necessarily determine it by representing that relation. Perhaps that relation is the relation of exemplifying certain properties or relations whose identity is determined by what is in the head. That was the classical view. Or perhaps it is the relation of being the cause of the thought. Or perhaps it is the relation of covarying with the thought in a counterfactual supporting way, and so forth. Indeed, perhaps the reference-determining relation can vary from one kind of thought to another, say, from thoughts of natural kinds to thoughts of individuals, and so forth. Perhaps the best known example of this position is Block (1986). Taking the meaning (as distinguished from the reference or extension) of the thought to correspond to this reference-determining relation or to whatever determines it, Putnam's observations about natural kinds may be true, yet fail to show that meaning is not in the head. Now I have no wish to argue over how the terms "internalism" and "externalism" should be used. Nor do I think that generalized arguments for or against abstract philosophical positions—internalism/externalism, realism/antirealism, individualism/antiindividualism and so forth—are ever of much, if any, value. What may be of value, however, is to lay out a well-articulated position on how people's heads actually do interact with the world they are in to create the phenomena of meaning and reference in thought, and to show how this concrete position explains various phenomena that we are interested in. If at the same time we can show how to avoid certain problems that we wish to avoid, problems that have concerned those interested in the internalism/externalism debates, we will have made progress. I have attempted to articulate such a position over the years in various books and papers. The position happens to be, in what seems a very strong sense, externalist. For it implies not only that basic reference or extension is always determined by a concrete existential relation between the thought and its referent or extension. It also implies that what determines this relation to be the relation that determines reference or extension is not merely a matter of what is currently in the head. This claim might reasonably be expressed, using Putnam's phrase, by saying that "meaning is not in the head." But it is a more extreme position than the one Putnam actually gives us arguments for in "The Meaning of 'Meaning'."

Existence Proof for a Viable Externalism

229

I cannot, of course, re-present the whole of a developed theory of thought reference and thought meaning in a few pages. But what I can do is to outline the theory, supply references for where more complete exposition and defense can be found, and then concentrate on showing how this particular theory addresses basic concerns about externalism that have been voiced by internalists. The theory takes thoughts to be inner representations. It is peculiar in what it takes inner representations to be. One way to understand it is as a modification of classical functionalism. It modifies classical functionalism in two ways. First, it defines inner representations by the way they function, not just in the head, but as parts of much larger systems that include portions of the environment. Second, the functions by reference to which inner representations are defined are not mere dispositions of the representations within their inner and outer environments, but what I call their "proper functions." These are dispositions they were selected to have by natural selection, or dispositions that a Normal development of the biological system has produced by means of the organism's interacting with its environment in a Normal way. I capitalize "Normal" to mark off a special sense of that word. What is biologically Normal, as I use that term, is not what is common or average. It is the way examples of a lineage of biological systems have functioned in the past on those occasions that accounted for their selection. Better, the way they functioned that accounted for the members of this lineage not having been selected against in situations where members of the lineage not functioning this way were or would have been selected against.1 It is this reference to a certain kind of history of selection and/or development that adds the radically externalist twist to this theory of mental representation. What a thing was designed to do is not always evident just from its inner structure, or from its inner structure plus the structure of it's current environment. Accordingly, whether an inner structure is a representation is not merely a matter of its form or current dispositions. Inner representations are defined by reference to the way representations Normally function in a wider system that includes the organism's Normal environment. But they cannot be defined merely as items that are in fact functioning in a Normal way. Rather, they are defined as items produced by systems—genetic systems, perceptual or cognitive learning systems—that, if functioning Normally, would produce representations that were capable of functioning Normally. These producing systems would produce properly functioning representations, given their Normal operation, given Normal operation of the rest of the inner cognitive systems, and given Normal conditions in the outer environment within which these systems were operating. That is, the outer environmental conditions would have to be of the same relevant kind as those in which the representation-producing systems operated historically when they were selected for. But these representation-producing systems may not in fact have been functioning Normally or functioning in an environment Normal for them. In that case, they will undoubtedly

1

More careful definition and discussion of the notion "proper function" and "Normal" may be found in (Millikan 1984, Chapters 1 and 2; Millikan 1993 Chapters 1 and 2; Millikan 2002).

230

In Defence of C o n t e n t Externalism

produce something that cannot operate like a Normal representation does. We say that what they produced are indeed inner representations, but that they are not Normal ones. Primary ways for representations to fail to be Normal are being false, or empty, or equivocal. I will come to the full story on that a bit later. At the moment, a point I need to emphasize is that when representation producers do not function Normally, this is not, in general, because there is anything wrong with these producers. Usually it is because the environment they are operating in is not Normal, relative to their particular historically Normal method of functioning properly. This point needs to be leaned on, since failure to understand it has produced well-known but mistaken criticisms of the position I wish to advocate (Dretske 1986, 1988; Neander 1995). Let me illustrate the point with an analogy.2 There are such things as ice cream-making machines. Their function is to make ice cream. But this is a function they cannot perform, no matter how well they are built, unless they are in an environment that supplies certain materials and conditions. The right ingredients have to be placed by someone or something into the right input containers. And, in the usual case, the machine has to be right side up and plugged in to a source supplying the right kind of electrical power. And the machine has to be turned on. So much is trivial. But the obvious conclusion has often been overlooked. If a device is failing to perform its function, this does not necessarily imply that the device itself is malfunctioning. It does not imply that the device is sick or broken. Mostly, when devices are not performing their functions, it is because the environments they are in are wrong, or the necessary inputs are not being put in, or, for the most trivial example, because they are not turned on. Similarly, eyes may fail to see because they are in the dark or because they are closed. Eyes that are seeing double (not something, presumably, that they were selected for doing) may do so because the optometrist has interposed an instrument between them and what they are seeing that constitutes an abNormal environment relative to their historically Normal way of functioning properly. Similarly, eyes that see mirages are trying to see under atmospheric conditions not historically Normal for distance vision. It is similar for systems whose proper functions are to produce non-empty, nonequivocal, true representations. They only work properly and Normally when placed in the right sorts of environments and given the right sorts of inputs. I have said that the theory of thought that I advocate defines inner representations by the way they function, not just in the head, but as parts of much larger systems that include portions of the the environment. Like systems that use inner representations, ice cream machines require to be in the right sort of environment and to have the right sort of input in order to work properly. But they are not defined by reference to the way they function, but only by reference to the output they are designed to produce. In what sort of way\ then, must something be designed to function in order to be an inner representation?

2

Indeed, it is not merely an analogy if the position I have taken on the proper functions of h u m a n intentions hence of h u m a n artifacts is correct. See, perhaps best, (Millikan 2004, Chapter 1).

Existence Proof for a Viable Externalism

231

I use the term "representation" in much the same way as the zoologist, C. R. GaJlistel: 3 I use the term representation in its mathematical sense. T h e brain is said to represent an aspect o f the e n v i r o n m e n t w h e n there is a f u n c t i o n i n g i s o m o r p h i s m b e t w e e n an aspect o f t h e e n v i r o n m e n t a n d a brain process that adapts the animal's behavior t o it. (Gallistel 1 9 9 0 , p. 3)

Unpacking this a bit, the idea is that it is, in part, because the animal responds to its environment with behavior-governing brain states that have aspects that run isomorphic to certain aspects of that environment that the animal's behavior manages to vary as a function of the environment so as to be adapted to that environment. Behavior needs to vary as a function of the environment. Inner representations supply a mechanism by which that is accomplished. It is important that this description of inner representations requires them to have behavior-governing functions. For example, inner representations are not representations merely because they covary with environmental causes. T h e representation must, Normally, be used by the biological system as a representation. T h e isomorphism between representations and environmental aspects has to be one that helps to explain how the animal's behaviors become adapted to its environment so as to serve their biological functions or purposes. Unpacking this idea even more, a Normally operating inner representation is part of a total system that includes (1) a representation producing mechanism, (2) some aspects of an organism's environment, and (3) a representation using mechanism. The producing mechanism produces representations that run isomorphic to the environmental aspects as defined by some rule of correspondence. T h e representation-using part of the system relies on this particular rule of correspondence in producing behaviors that will be effective given that the environment actually displays the aspects represented. 4 This use of the term "inner representation" is very broad indeed. Myriads of things inside the body other than thoughts are representations in this sense. But the internalism/externalism issue primarily concerns thoughts, beliefs and so forth. H o w then, exactly, do thoughts and beliefs fit in? Although I have argued elsewhere that sentences can be a very misleading model for beliefs, 5 in the current context that model will serve us well enough. We can think of beliefs as like mental sentences, containing words that are rearrangeable into other mental sentences that mean different things. We can think of beliefs as being compositional, and think of the extensions of the true ones—their truth makers—as determined by some sort ofTarskian mapping. Certain definite kinds of rearrangements of sentence parts correspond to certain definite kinds of possible transformations on the

3

4 5

T h e exposition I will give concerns, in fact, only one of three basic types of inner representation, namely "descriptive" or "indicative" representations. For more complete discussions see (Millikan 1984) especially chapter 6; (Millikan 1993) especially chapters 3 to six; (Millikan 2004) especially chapter 6 and chapter 13ff. For more detailed formulations of this theory of inner representations, see (Millikan 1984), especially Chapter 6; (Millikan 1993), especially Chapters 3 to 6; (Millikan 2004), especially Chapter 6. Millikan 2000, chapter 8 ff.

232

In Defence of Content Externalism

aspects of the world to which the sentences must correspond if they are to contribute to the Normal functioning of their users. Certain substitutions of sentence parts correspond to substitutions in aspects of world affairs mapped and certain additions correspond to additions, and so forth. That is, there is a functional isomorphism between the whole set of possible mental sentences a certain person could affirm and the set of world affairs to which these sentences would correspond, were that person's cognition to proceed entirely Normally. But cognition proceeds perfectly Normally only in the case of a person all of whose beliefs are true. (This last underlines the importance of distinguishing what is Normal from what is common or average.) In the case of beliefs, presumably Normal function involves, at least, interaction with other beliefs to create new beliefs, interaction with desires to produce more desires, and so forth, a final result being decision-making and productive action. The semantic rules that define the Normal correspondences between a person's belief states and aspects of their environment are determined mainly by the history of that person's individual cognitive systems. Let me explain. Mechanisms whose proper function is to produce inner representations that will correspond to the world in accordance with certain definite semantic rules have to have a systematic method or methods of achieving this goal. This does not mean that the methods always work, of course. It doesn't mean that no mistakes are ever made. It means only that there must be conditions under which the method or methods do work, these conditions having been present in the past on the occasions when mechanisms worked well enough to be retained or selected for. Some mechanisms have been selected for by natural selection during evolutionary history. Others are selected for or tuned for their jobs through processes of learning. Mechanisms whose proper functions are to turn out beliefs that correspond, in accordance with definite semantic rules, to the world, have probably invariably been selected for or tuned by learning. What kind of learning is this, and how does it take place in a Normally developing cognitive system? To supply details here is, in the end, a job for developmental psychologists and neurologists. But some part of the story may be discernable even at the distance of the philosopher. In order to learn to make beliefs that correspond in systematic ways to the world, one must learn how to identify, through the senses, the various objects, kinds, properties and so forth to which the elements of these beliefs are to correspond. That is, one must develop concepts of each of these various objects, kinds, properties and so forth. Developing adequate concepts of these kinds of things involves, paradigmatically, learning to reidentify them via their perceptual manifestations in a variety of ways, under a variety of conditions, given a variety of intervening media, and so forth. The proximal stimuli that may manifest the same distal object or property are extremely numerous and diverse, for they depend on diverse outer conditions and intervening media. For this reason the business of producing new beliefs that correspond by uniform rules to the world with any regularity would be completely impossible unless abilities to recognize each of these various distal objects and properties in a variety of ways, under many different mediating conditions, were very highly developed. The representationmaking mechanisms must be very versatile in recognizing, or better, reidentifying, the

Existence Proof for a Viable Externalism

233

same object, the same kind, the same property or relation, and so forth as the same one again under lots and lots of different external circumstances. How can I have concepts of things I have only heard about and can't recognize by normal perceptual means? I will get to that below. But for a full exposition and defense of this view of what having empirical concepts entails, see (Millikan 1984) Chapter 15 and (Millikan 2000) chapters 1 through 5. The question how the developing intellect learns to make these reidentifications correctly, concerns what can be called the natural epistemology of concepts, as distinguished from the natural epistemology of judgment or belief. It concerns the developmental processes by which children (and perhaps some animals) Normally acquire, through experience, an inner representational belief system in which the concepts or mental terms are non-equivocal, non-redundant, and non-empty—and a system which the belief-using devices have, at the same time, learned how or been tuned to use. That story is told in (Millikan 2000) chapter 7. It is a story that requires realist ontological underpinnings. These are provided in (Millikan 2000), chapter 2. But that there is a Normal way for children to acquire concepts or "mental terms" that are non-equivocal (no failures to recognize that Tweedledee and Tweedledum are two persons, not one), non-redundant (no Oedipal failures to identify Jocasta with Mom) and non-empty (no phlogiston concepts) does not have the least tendency to imply that no such faulty concepts are in fact employed by the average adult. Nor does it imply that the employment of faulty concepts shows that something was wrong with the minds of the persons that developed these faulty concepts. O n this, compare (Millikan 2000) chapter 14. What do we say, then, about mental terms that are defective? Well, if they are empty, if they do not exemplify any real ability to reidentify anything real, we say that they don't refer (counting real properties, real kinds and so forth as "referents" for ease of exposition). We don't say, for example, that they refer to "possiblilities" any more than we say that a can opener that isn't working correctly is opening possible cans. If the defective terms are equivocal, we can just say that they refer equivocally. If they are merely redundant, they refer and refer univocally, but they are not in accord with what might be called an important "regulative ideal" for the cognitive systems. For the very first job of the cognitive systems is to manage to learn to reidentify what is referentially the same thing as the same again. Should we say that defective mental terms have "meanings," or that persons who harbor them mean something when they use sentences that express these terms? They usually feel the same to the mind as mental terms that are not defective. They were developed by systems whose jobs were to develop (Normally functioning) mental terms. But there isn't any kind of entity with a rightful place in our ontology that corresponds to them: no possible individuals, no possible kinds or possible properties, no Fregean senses, no Carnapian intensions, no Quinean meanings-under-glass-in-the-museum. Quine was right about all that. And Gareth Evans was right when he pointed out that the idea that there might be modes of presentation that didn't present anything made no sense at all. But what do we say about, not faulty mental terms, but, granted there are such things, empty mental descriptions composed of healthy parts—say, a thought of the present king of France*. It certainly doesn't correspond to any peculiar entity such as a

234

In Defence of Content Externalism

possible person. Nor does it correspond, of course, to an actual entity, to a referent. Each of its parts and its mental syntax are all perfectly fine. But if this is a real thought, and not just a pretend thought (such as the "thought of Santa Clause" for us adults), healthy parts or no, it is defective as a whole. It can, however, be expressed using healthy English words and healthy syntax. For this reason, the English phrase that expresses it can be translated. In that sense the phrase expressing it means something. (Was there something more I was supposed to say?) Now, at last, I am in a position to try to mollify some internalist anxieties for the case of one kind of externalism—the kind that I have just outlined. First, a useful analogy may help to make this kind of externalism more intuitive. No one doubts that whether the person you are apparently remembering at the moment is Great Aunt Nelly or Great Aunt Stella is a matter of the history of that particular memory. It is a matter of which aunt (if either) caused the apparent memory in that sort of way that memories are caused when they are Normal. If no one caused your memory in that way, it isn't actually a memory. Similarly, my proposed kind of externalism claims that basic thoughts of things in the empirical world, such as thoughts of particular people and thoughts of water or of dogkind or of the color red, are rather like memories. What they are thoughts of depends on what in the outside world they were developed, in a Normal way, as a certain kind of cognitive response to. That is, what they are thoughts of, indeed, whether they are genuine thoughts of anything, rather than, say, just sorts of inner itches or tickles, depends on their causal histories. In a way, this claim is just a broadened empiricism. The classical empiricists claimed that all thoughts have to have a certain kind of history involving the senses, and that this was not just an accidental fact about human psychology, about what happens to cause the having of ideas in humans, but a fact that followed from the very nature of what thought is. If a thought is a copy of an impression, for example, then it is a thought, in part, by virtue of its causal history. What I have changed here is that the relevant causal history is not the history merely of a person's phenomenology or sensory input, but the history of the wider system in which the individual human head has been operating—a system that has included aspects of the human's environment. Now to this sort of externalist position, one standard internalist objection is that if historical events involving things outside your head determine what you are thinking of, so that what is presently inside your head leaves the matter indeterminate, it appears that you have no way of knowing what it is you are thinking of. To this objection, the first proper response is a question. What does the internalist suppose constitutes your knowing what you are thinking of? Would you have to think of your thought, then think of its object, and then judge that the one represented the other? How would you think of the object in order to make this judgment? Would it just appear there before your mind, bare, the way Russell supposed sense data do? Or would it have some kind of essence that defined its real nature, which essence could appear directly before your mind? Or would you have to think it through another thought-representation of it? Notice that the last option would seem to make the truth of your judgment depend on the fact, only, that two of your thoughts represented the same, rather than that either represented some particular thing outside. So long as the thought that you think of and the thought that you think with, in making the last

Existence Proof for a Viable Externalism

235

sort of judgment, represented the same object, you would be judging correctly what you were thinking of. Now what I have proposed is the following. There is no such thing as having an external object, kind or property directly before your mind. And I haven't the slightest idea what it would be to have something's nature before my mind, certainly without the thing, the kind, the property, or whatever, whose nature it was, being dragged along. True, an incredulous stare, as David Lewis was fond of pointing out, is not an argument. But I claim, positively, that something akin to the third alternative above is both readily understandable and plausible. To know what you are thinking of just IS

to recognize when you are thinking of the same thing again. Nor need this recognition consist in making a judgment. To recognize when your thought is of the same thing again is usually effected just by joining these thoughts, or being prepared to join them, in mediate practical or theoretical inference processes that use them as a middle term. These are processes that will perform in a Normal way to produce true conclusions or helpful desires or intentions only if these terms do in fact represent the same. For details on this proposal, see (Millikan 2000), Part Two. That you know what you are thinking of when you think of an object, kind, or empirical property follows directly from the theory of empirical concepts sketched above. For on that theory, having a concept consists, in part, of the ability to reidentify the object of thought despite the variety of its manifestations to the senses, hence, of course, its possible diverse effects at the level of conceptual representation. A corollary is that one can know what one is thinking of more clearly or less clearly, depending on how versatile one is at reidentifying what one thinks of in perception and thought. Some of one's concepts are more adequate than others; some come closer to the ideal than others. Nor does this conflict with common sense. We often know of a person under a certain description, such as "the president of the ACLU," yet are ready to say that we don't know who that person is. To know of a person under a certain description is, in general, to know how to reidentify them in at least one way, namely should one encounter them again as fitting the same description. Indeed, usually a definite description affords one a variety of different ways to identify its referent, because one generally knows a variety of different ways to reidentify the correlate of each term in the description. But despite having a definite description in mind, we may still say that we don't know who that person is. I don't, for example, know who the current president of the ACLU is. So what would be required in order really to know who that person is? On this question, I strongly recommend Böer and Lycan's book Knowing Who (1986), and I suggest that the lesson generalizes to all kinds of knowing what something is. One can know who one is referring to or thinking of in any of a variety of ways, the more the better, I suggest. The more ways one knows, the closer one is to the unobtainable ideal that would define perfect functioning of one's concept-forming and using capacities. The externalist theory I have presented parts company with common sense here, however: According to common sense, you know just by reflection whether you are thinking of something and if so of what. On the theory of empirical concepts I have proposed, whether you are thinking of something is tested by you over time empiri-

236

In Defence of Content Externalism

cally. That is what the naturalized epistemology of concepts referred to above is about (Millikan 2000, chapter Seven). A second standard internalist's objection to externalist theories is that on such theories it is not possible to know with certainty what you believe. If what you believe depends on the history of your thought, not just on what is inside your head, how can you be sure what you believe? This may at first appear to be the same as the first internalist's objection just discussed. But the overlap is only partial. Knowing what you are thinking of, as I have just described it, is not the same as having beliefs about your beließ. Again, the first response to this objection should be a question, indeed, two questions. First, why is the internalist so sure that we always do know, with certainty, what we believe? Surely Freud rocked the complacency of even the ordinary person on that point long ago. True, people generally do know in a lot more detail about what they themselves believe than about what anyone else believes. But people are also likely to know more about the details of what is currently in their mouths than anyone else knows. This is because they have a different way of finding out what is currently in their m o u t h than other people do. Similarly, it seems clear that we have a different way of finding out what we currently believe than other people do, and it seems to be a pretty accurate way. This brings up the second question we should ask the internalist. By what sort of mechanism does the internalist suppose that we find out about our own beliefs? This ability would not be explained, of course, just by the fact that the beliefs were in our heads. O u r neurotransmitters are in our heads too, but that doesn't make it any easier for us to know about them. It was, of course, traditional philosophical doctrine that the mind is somehow transparent to itself, that it does, somehow, know all about itself. All that is necessary to know about minds is to be a mind. But then why have we been disputing about the nature of thought and of mind for 2500 years? And why does it take small children three or four years to learn to employ concepts of mental states, such as the concept of belief, appropriately? The internalist demands a story from the externalist about how we manage correctly to recognize our own beliefs. But where has he given his own story on this? Be that as it may, there was an interesting story on this told by the very first of the radical externalists, namely, Wilfrid Sellars. Sellars suggested that having acquired the ability to identify the presence of certain kinds of beliefs in other persons, one might later learn how to identify beliefs in oneself, roughly by the following means. O n e might catch oneself on the verge of candidly asserting something, and preface this asserting with either a spoken or a thought "I believe that...." To fill out the Sellarsian story, however, we would need to have a pretty definite view of the relation of thought to language. Sellars's own view was that the extensions of thoughts were determined by their inferential roles, including responses to external objects and properties with thoughts, and that inferential roles were internalized linguistic roles. That is very far from the position on thought that I have presented here. So to fill out Sellars's suggestion about how you find out what your beliefs are, I will need to fill in a different view of the relation of language to thought. Earlier I mentioned thoughts of things you have heard about but would be unable to recognize in an ordinary way, such as Socrates and, if you are like me, molybde-

Existence Proof for a Viable Externalism

237

num. How can you have concepts of these things on the theory of concepts I have suggested here? The proposal I have offered is that, Normally, language carries manifestations of distal affairs in the world in a way exactly analogous to the way light waves, sound waves, odors, tastes, gravitational fields and so forth do. Thus, in the Normal case (notice the capital 'N'), believing what you are told is in relevant respects exactly like believing what you see. Understanding and believing the messages carried by fact-stating language is merely another form of perception. O n e way of reidentifying an object or kind or property, then, is by knowing a word for it in the language spoken by the community you are in. And one way of having a concept, though not yet a very filled out or adequate one, may involve knowing just a word for its referent, rather than knowing what it looks like and/or feels like and so forth. Compare here the very slender concept I have of the current president of the ACLU, since I have not the slightest idea who he or she is. Obviously this position on the relation of language to thought requires to be much clarified and to be argued for at length. The clarification and arguments are in (Millikan 2000 Chapter 7; 2004 Chapter 9; 2005 Chapter 10). But if the position is correct, learning a public word for a thought referent is a way of coming more fully to understand what you are thinking of when you think of that thing. If I learn the name of the current president of the ACLU, I will know better who I am thinking of when I think of him or her. Conversely, misunderstanding or using a public language word wrongly is way of failing to know exactly what you are thinking of. Return now to the Sellarsian idea that you can come to know what you believe by catching your disposition to utter an informative sentence candidly, and prefacing the sentence with an overt or covert "I believe that....". Granted that you correctly use and understand the public language that you (propose to) speak, and that you are able to recognize your candid moods (this not to be taken for granted), you will express in language or understand in thought exactly what it is that you believe. No one else can come to know what you believe in this way. And we usually take it that this is a pretty accurate way of coming to know what we believe. An interesting result follows. It follows that you could not know what you believed from inside if you had no language. That seems to me an entirely plausible result. Others may disagree. But in any event, if the internalist thinks there is some problem that the externalist has but that he doesn't have concerning how a person comes to know what they believe, he must begin by giving us his own story on how this magic comes about.

References Böer, S.E. and W.G. Lycan 1986. Knowing Who (Cambridge MA: MIT Press). Block, N. 1986. "Advertisement for a semantics for psychology." In Studies in the philosophy of mind, Midwest studies in philosophy X, ed. P. French, T. Euhling and H. Wettstein. (Minneapolis: University of Minnesota Press). Dretske, F. 1986. "Misrepresentation." In Belief: Form, content, and function, ed. Radu Bogdan, 17-36. New York: Oxford.

238

In Defence of Content Externalism

Dretske, F. 1988. Explaining Behavior. Bradford Books/MIT Press. Fumerton, R. 1989. "Russelling causal theories." In C.W. Savage and C.A. Anderson (eds.), Rereading Russell: Essays in Bertrand Russell's Metaphysics and Epistemology, Minnesota Studies in the Philosophy of Science, vol.12, 108-118. Minneapolis: University of Minnesota Press. Gallistel, C. R. 1990. The Organization of Learning (Cambridge MA: The M I T Press) Millikan, R. G. 1984 Language, Thought and Other Biological Categories. (Cambridge MA: MIT Press). Millikan, R. G. 1983 White Queen Paychology and Other Essays for Alice. (Cambridge MA: M I T Press). Millikan, R. G. 2000 On Clear and Confused Ideas; An Essay on Substance Concepts. (Cambridge: Cambridge University Press.) Millikan, R.G. 2002. "Biofunctions: Two paradigms." In Cummins, A. Ariew and M. Perlman, eds., Functions: New Readings in the Philosophy of Psychology and Biology, (Oxford: Oxford University Press). Millikan, R.G. 2004, Varieties of Meaning. (Cambridge MA: M I T Press). Millikan, R.G. 2005. Language: A Biological Model, (Oxford: Oxford University Press). Neander, K. 1995. "Misrepresenting and malfunctioning." Philosophical Studies 79, 109-41. Putnam, H. 1975. "The meaning of'meaning'." In Minnesota Studies in the Philosophy of Science, vol. VII, ed. Keith Gunderson. Minneapolis: University of Minnesota Press.

Externalism, Epistemic Artefacts and the Extended Mind K J M STERELNY

I Niche Construction and Epistemic

Engineering

A common picture of evolution by natural selection sees it as a process through which organisms change so that they become better adapted to their environment. However, agents do not merely respond to the challenges their environments pose. They modify their environments, filtering and transforming the action of the environment on their bodies A beaver, in making a dam, engineers a stream, increasing both the size of its safe refuge and reducing its seasonal variability. Beavers, like many other animals, are ecological engineers. They act to modify the physical challenges posed by their environment. Nests, burrows and other shelters reduce the impacts of adverse weather and of other agents. Animal also modify their exposure to biological risks. Hygienic behaviour reduces the impact of disease. Intensive grooming; moving to new roosts; using a "latrine burrow"; disposing excrement in faecal sacs; these all improve an animals prospects of avoiding disease. So many organisms are like the beaver; they partially construct their own niches. They are ecological engineers, and, as John Odling-Smee and Kevin Laland have shown, niche construction is often of great evolutionary significance, transforming the effects of natural selection on both the ecological engineers and their descendants. 1 Ecological engineering is visible to selection, for such alterations often have fitness effects that are stable across generations. So niche-constructing behaviour itself evolves. Yet until fairly recently, the evolutionary analysis of behaviour has tended to underplay the environment-altering aspects of agent action by treating treat environments as a fixed set of constraints on action. The great exception to this approach has been the social environment. Frequency dependent models of social behaviour, game theory, and sexual selection theory all treat the environment as an ensemble of other agents also making choices, and hence the social environment is not a fixed background to action. But with this exception, niche construction continues to be underplayed. For example, (Krebs and Davies 1997) is the field-defining anthology of behavioural ecology, now in its fourth edition. Yet this collection contains no paper primarily on physical or biological niche construction. This practice of treating the environment as fixed background is particularly a feature of optimality modelling. Optimality models share a set of fundamental elements. They specify the set of strategies available to an agent in a given situation; they include

1

See (Odling-Smee, Laland et al. 1 9 9 6 ; Laland and Odling-Smee 2 0 0 0 ; Laland, Odling-Smee et al. 2 0 0 0 ; Odling-Smee, Laland et al. 2 0 0 3 ) , and also (Jones, Lawton et al. 1 9 9 7 ) .

240

In Defence of Content Externalism

a measure of the costs and benefits that attach to each strategy; and they specify the features of the environment that determine those costs and benefits. Costs, benefits and their environmental determinants are fixed. T h e agent can only choose the right strategy given these fixed factors, the one that best trades benefit against cost. How, for example, should a bird feeding its young trade the number of trips back to the nest against the load carried on each trip? W h a t will maximise the rate of food delivery to its chicks; an important consideration since better fed chicks have better survival prospects. Starling foraging has been analysed with this question in mind. Since the journey from foraging site to nest takes time and energy, the optimum strategy would seem to be that of bringing back the heaviest load possible, thus cutting down on travel costs. But starlings must keep what they have caught in their beak and so their efficiency declines as they forage. They must trade declining efficiency against transit costs. T h e optimum behaviour for starlings depends both on the distance between nest and foraging site, and the rate at which their efficiency declines. W h e n travel costs are low because the bird is foraging close to its nest, a lighter load is more efficient. W h e n travel costs are high, it is worth paying the cost of foraging less efficiently and carrying more per load (Kacelnik 1984). T h e cost of information is quite often neglected in such models. In the case of fairly simple decision problems like that of the starling, this simplification seems reasonable. For the starling decision model has a low cognitive load. T h e information required for correct strategy choice is not extensive and may well be acquired as a by-product of their routine activity. Starlings maximise their fitness by maximising the rate at which they deliver food to their chicks, so they need to know the distance from foraging site to nest and the rate at which their efficiency declines. That information is available (though not necessarily registered) as a by-product of feeding. In other circumstances, birds must take into account both risks to themselves in foraging and in transit, and risks to the nest occasioned by their visits. For such birds, the cognitive load might be heavy. Agents often need information that is not ready to hand, not generated as an automatic by-product of their ecological activity. So they are epistemic as well as ecological agents, acting to change the informational character of their environment. Many animals are epistemic engineers. Low-tech epistemic agency is ubiquitous. We see it in such routine action as a raptor choosing a hidden lookout post which nonetheless offers good views of its hunting ground. However, epistemic action is often more extensive and systematic than this. Many birds, once they have built their nests "decorate" them with moss and other greenery. T h e nests of small passerines, in particular, are often very hard to find, both by interested humans and (presumably) the birds' predators. Such nests are almost certainly cryptic by design. Their unobtrusiveness is no mere side-effect of the methods and materials from which they are constructed. These birds are engaged in epistemic counter-measures against their enemies, attempting to render their predators' informational environment opaque. Social interaction can have the opposite effect. Many species of bird give "contact calls" advertising their location to their mates while foraging, and hence making this information available for free. This paper considers epistemic action in the human lineage, and in particular the ideas of Andy Clark, Dan Dennett and Stephen Mithen, who all take the invention

Externalism, Epistemic Artefacts and the Extended M i n d

241

and use of epistemic artefacts—tools for thinking—to be central to the explanation of human intelligence and human culture. I agree with this line of thought. But I shall argue that Clark, in particular, in concentrating on epistemic tools used only by a single agent, has somewhat mis-characterised our relation to our tools, and hence has mischaracterised the relationship between this program and more traditional representational theories of mind. In contrast to Clark's view, I take this relationship to be one of mutual support. For our use of epistemic artefacts explains the elaboration of mental representation in our lineage, and this elaboration explains our ability to use epistemic artefacts. In discussing our abilities to represent aspects of our environment, Andy Clark sometimes talks of "portable" representational resources: resources that an agent can deploy in responding to a range of challenges. I shall argue that the effective use of epistemic tools often depends on such portable resources. Elsewhere, I have called these states "decoupled representations". For though they track features of the external world they are not locked into driving specific actions (Sterelny 2003). When decoupled representations track the world accurately they are, in Godfrey-Smith's apposite phrase, "fuels for success" (Godfrey-Smith 1996). For whatever an agent's plans, accurate decoupled representation of the world improves that agent's prospects of success. In sections III and IV I shall argue that our use of epistemic tools depends on our ability to use a rich store of decoupled representations.

II Human Epistemic Agency For hundreds of thousands of years we have engineered our physical and biological environments. But epistemic agency has been an equally pervasive feature of hominid life styles. Indeed, Stephen Mithen has argued that tools are often both material and epistemic artefacts. A fish trap traps fish, and hence it plays a direct economic role in the life of its user. But it can also serve as a guide for the construction and location of further fish traps. The products of material culture—tools, shelter, vehicles, weapons, clothes—serve both to rework our physical and biological circumstances, and as templates, props and prompts for their own reproduction. As Mithen sees it, this is both a deep and a crucial feature of human history: perhaps 100,000 thousand years deep (Mithen 2000). Co-opting physical tools for epistemic purposes long predates using specialised epistemic artefacts. But these too have a deep history. With the invention and elaboration of pictorial representation, humans came to be makers of specialised epistemic artefacts. It is very difficult to date the first appearance of specialised epistemic artefacts, but unmistakable, superbly executed paintings are over 30,000 years old (Mithen 1998a). In Mithen's view, the use and elaboration of epistemic artefacts explains the extraordinary acceleration in both the richness and the variability of human cultures over the last 50,000 years or so. He thinks our archaeological record shows the marks of a cognitive breakthrough. Yet in all probability human cognitive powers have expanded without any significant change in human genomes. Instead, their expansion is a result of the proliferation and transmission of epistemic artefacts; of tools

242

In Defence of Content Externalism

for thinking. As he sees it, we have got smarter, even though our naked brains have not 2 . In their recent work, both Andy Clark 3 and Dan Dennett 4 have also stressed the role of these cognitive tools in the explanation of human intelligence. Human intelligence needs explanation, for our minds are qualitatively different from those of our closest kin; even (if Mithen is right) our extinct kin with similar-sized brains. O n the Dennett-Clark-Mithen view, we are not distinctively intelligent in virtue of being innately endowed with a superbly powerful, autonomous internal computational engine. Our special cognitive powers derive from our ability to extend our minds' capacities through interacting with our environment. They do not doubt that there was some internal cognitive difference between those humans who formed the base of the cognitive explosion of our lineage, and other primates with similar sized brains. But the difference was not that of having fully human intelligence. Rather, it was a small change that acted as a trigger (a "seed", in Clark's terminology) of a runaway coevolutionary process between humans and their artefacts. I think this idea is right, and shall sketch some of these ability-enhancing interactions before turning to the implications of this picture. 1. Most obviously, we alter our environment to ease memory burdens. We do so by storing information in the environment, by recoding it, and through social organisation. Environmental storage is most obvious of these techniques. Our contemporary environment is full of purpose-built tools for easing burdens on memory. These include diaries, notebooks and other "organisers". Indeed, Clark and Chalmers rely heavily on the functional similarity of internal and external memory in arguing that human cognitive systems literally include (some of) these external tools. (On this, more later). Filofaxes are new tools, but purpose-built aids to memory are certainly ancient. Not only is pictorial representation over 30,000 years old; there are cryptic objects in the archaeological record of similar age—incised bones—which may well be systems for

2

I have two major reservations about this hypothesis. First, I have been convinced that the idea of an explosive jump in human culture from about 50,000 years ago is much oversold (see McBrearty, S. and A. Brooks (2000). "The Revolution That Wasn't: A New Interpretation of The Origin of Modern Human BehaviorJournal of Human Evolution 39(5): 453-563. And I think that though human genomes are unlikely to have changed much, human brains almost certainly have, as a consequence of changes in our developmental environment. But these reservations do not undercut Mithen's basic point that epistemic artefacts massively enhance our cognitive capacities.

3

See especially Clark, A. (1997). Being There: Putting Brain, Body, and World Together Again. Cambridge, MIT Press. Clark, A. and D . Chambers (1998). "The Extended Mind." Analysis 58(1): 7-19. Clark, A. (1999). "An Embodied Cognitive Science?" Trends in Cognitive Science 3(9): 345-350. Clark, A. (2001). "Reasons, Robots and the Extended Mind." Mind and Language 16(2): 121-145. Clark, A. (2002). Minds, Brains and Tools. Philosophy of Mental Representation. H. Clapin. Oxford, Oxford University Press: 66-90. Clark, A. (2001). Mindware: An Introduction to the Philosophy of Cognitive Science. Oxford, Oxford University Press.. See especially Dennett, D. (1993). "Learning and Labeling." Mind and Language 8(4): 540-547. Dennett, D. (1995). Darwin's Dangerous Idea. New York, Simon and Shuster. Dennett, D. (1996). Kinds of Minds. New York, Basic Books. Dennett, D. (2000). Making Tools for Thinking. Metarepresentation: A Multidisciplinary Perspective. D. Sperber. Oxford, Oxford University Press: 17-29.

4

Externalism, Epistemic Artefacts and the Extended Mind

243

the long-term storage of information (Mithen 1998b). Furthermore, and deeper still in the past, ecological tools have informational side-efifects. Trail marking—making easilynoticed alterations to the bush while walking through it—may be older still, as there is some evidence that bonobos mark trails in this way. Written records are obviously wonderful memory aids. Writing is a recent invention, but humans have long re-coded information in public language to make it easier to recall. In songs, stories and rhyme, the organisation of the information enables some elements to prime others. Such recoding enables us to partially substitute recognition for recall. Finally, as human societies became technically and socially more complex, the division of cognitive labour eased memory burdens; for then no member of the group had to know everything the group as a whole needed to know. 2. We transform difficult cognitive problems into easier perceptual problems Andy Clark has discussed a nifty example of this transformation. It turns out that even those with a good visual imagination find it hard to extract information from an image which is not made salient in its initial description. For example, it is much harder to notice the second interpretation of an ambiguous figure in imagination that it is when confronted with the physical image. Sketches reveal unanticipated ways of seeing the image. This, he suggests, is why initial sketching is so important for artists; why they do not just construct the finished work in imagination and then execute it. (Clark 2001),pp 132-133). Likewise, even the strongest chess players prefer to analyse positions with a real set and pieces. But though Clark's example is apt, it is important to emphasise that turning cognitive tasks into perceptual tasks is not recondite, it is routine. We do it (for example) when we re-present quantitative information as a pictorial pattern, in pie-charts, graphs, maps. 3. We transform difficult perceptual problems into easier ones. For example, in shaping wood with a chisel and hammer, it is useful to mark the spot which you wish to strike, making it easier to focus attention on the exact working surface. Epistemic action of this kind could easily be ancient. Handaxes need to be made by reductive techniques, flaking the tool out of a larger matrix. This process would certainly be eased by scratching the point on the rock face you needed to strike. Contemporary human environments are often transformed in this way. Pedestrian crossings are painted in bright contrasting colours so they stand out against a background; likewise other warning signs. Students highlight text to help focus their attention and to aid search for later study. I have argued (Sterelny 2003) that in learning contexts, language is a very powerful tool for making small perceptual differences salient. Different linguistic labels for similar-looking birds alert individuals to the existence of a diagnostic perceptual difference between them. 4. We transform difficult learning problems into easier ones. For we alter the informational environment of the next generation. Non-human animals mostly alter the epistemic environment of their young as a passive byproduct of their own activities. We too alter our children's learning environment as a byproduct of our own economic and ecological activity. The development of a rich vocabulary that matches the biological, physical, artefactual and social kinds of the local environment is very important for effective action in those domains. But those taxonomies also make the informational environment of the next generation more transparent. Unlike other animals, we also

244

In Defence of Content Externalism

invest heavily in teaching. We deliberately provide information to the next generation: "that is a horsefield's bronze-cuckoo; you can tell if from a shining bronze-cuckoo by its more prominent white eyebrow and the fact that its chest bands are incomplete". But we do not just provide information verbally: learning is scaffolded in many other ways. Skills are demonstrated in a form suited for learning. Completed and partially completed artefacts are used as teaching props. Practice is supervised and corrected. The decomposition of a skill into its components is made obvious; subtle elements will often be exaggerated, slowed down or repeated. Moreover, skills are often taught in an optimal sequence, so that one forms a platform for the next. 5. We engineer workspaces so that frequent tasks can be completed more rapidly and reliably. Kitchens, for example, are organised so that the tools are handy, and so that there are safe and suitable workspaces (chopping boards and the like). Cooks often organise their ingredients before they begin cooking so that they are ready to hand when needed, and so that their physical sequence and grouping corresponds to the recipe being used (Kirsh 1995; Kirsh 1996). Clark too describes a cute example of workspace engineering. Skilled bartenders use the distinctive shapes of glasses and their sequence to cue recall for customers' orders and to code the order in which they will be served. Their ability to respond accurately to multiple simultaneous orders plummets if they are forced to use identically shaped glasses ((Clark 2001) chapter 8). Organised workspaces are now a ubiquitous feature of human life. Cognitive tools, too, are simplified and standardised to enhance performance on repeated tasks. Improvements in notation systems—the switch from imperial to decimal currency and measurement—makes many routine calculations easier, faster, and less error-prone.

Ill Dumb Brains, Smart Minds? Clark and Dennett both emphasise the importance of epistemic agency in explaining human intelligence, though with an important difference in emphasis. Dennett stresses the developmental role of tools for thinking. The symbol-soaked, cognitively engineered environments in which we now develop have profound implications for our internal cognitive systems. In particular, the experience of learning a public language and using sentential representations makes thought somewhat language-like. In contrast, Clark emphasises mind-artefact linkages. He has been deeply impressed by "embodied cognitive science"; intelligent agency is not explained via the computational transformation of sentence-like mental representations. Rather, embodied cognitive science favours models of adaptive behaviour that emphasise the use of information freely available in the world 5 and which emphasise the use of heuristics with low information demands. However, despite his sympathy for this line of thought, Clark suspects that some human cognitive traits remain "representation-hungry". So his best guess is that a full theory of human intelligence cannot be given by incremental improvements on these

5

In slogan form: "the world is its own best model".

Externalism, Epistemic Artefacts and the Extended M i n d

245

models. Nonetheless, he thinks the profound change in human cognitive evolution is the development of effective epistemic artefacts 6 . He is thus inclined to back the following horse: "to depict much of advanced cognition as rooted in the operation of the same basic kinds of capacity used for on-line, adaptive response, but tuned and applied to the special domain of external and/or artificial cognitive aids—the domain, as I shall say, of wideware or cognitive technology." {{Clark forthcoming) chapter 8)

To a significant extent we are serial processing inference engines. But we are so only because of our reliable, designed, adjusted coupling with a staggering array of cognitive artefacts, including linguistic and quantitative systems of serial representation. Our extended mind uses and processes linguiform representations even though (quite likely) naked human brains do not. Furthermore, some of these artefacts can literally be parts of an agent's cognitive system. In The Extended Mind, Andy Clark and David Chalmers develop a thought-experiment about an Alzheimers sufferer (MrT, as I shall call him). Mr Τ cannot unaided remember the location of an exhibit he wants to visit. But he manages such problems by writing down in a notebook crucial information for his daily plans notebook. He then acts by consulting this book. Clark and Chalmers argue that the information in the notebook plays the same functional role for Mr Τ that an ordinary (non-occurrent) belief plays for ordinary human agents. They conclude we should count the notebook as part of the patient's mind, and the location of the exhibit as one of Mr T's beliefs (Clark and Chambers 1998). Clark is careful not to trivialise this extension of the boundaries of the mind. He insists that agents' minds include only those external tools to which they have regular, unfettered access: "the props and aids which can count as part of my mental machinery ... are at the very least, reliably available when needed and used or accessed pretty much as automatically as biological processing and memory" (2001, ρ 139). While agreeing with Clark on the fundamental role of epistemic agency in explaining human rationality, I have reservations about this picture. Even when there is a reliable link between user and tool, there are important differences between internal and external cognitive resources. The external storage of information is very important, but the psychological and evolutionary dynamics of mind /filofax relations are critically different from those of mind/memory interactions. So I do not think it is helpful to think of epistemic artefacts as literally parts of the minds of the agents that regularly use them. Moreover and more importantly, Clark underplays the importance of nonexclusive use of epistemic artefacts. Many of our most important cognitive tools are common-use tools, not parts of coupled systems. Let me begin with reliable mind-artefact linkages and the differences between internal and external storage. The effective use of tools for the external storage of infor-

6

It is important not to exaggerate Clark's views here. H e explicitly acknowledges that humans have coevolved with their epistemic artefacts, and this changes human brains. Nonetheless, he does not emphasise the role of these internal changes in his explanation of the qualitative differences between humans and other primates.

246

In Defence of Content Externalism

mation depends on the agent's perception of them. Yet there are important differences between perception and internal monitoring. Organisms are communities of co-operative and co-adapted parts. Within the agent, there is no war of deception and counterdeception. Internal routes of information flow—from memory; feedback loops from action—will be under selection for reliability. Consider the internal signals generated as an agent tries to grasp and lift an object. Over time, we would expect these signals to become cleaner and less noisy; and internal devices to become more efficient in picking them up and using them to drive appropriate responses. Internal signals keep information available on-line. Moreover, as there is no danger of deception, they do not have to be epistemically vetted. All else equal, over evolutionary time the internal informational environment of an agent will become more transparent. None of this is true of perception. An agent's perceptual systems, like their introspective systems, are under selection to deliver reliable and veridical judgements about the environment. However perception operates in an environment of active sabotage by other agents. So despite the superb design of some of our sensory systems, perception of the external environment often delivers signals that are noisy, somewhat unreliable, and functionally ambiguous (was that really an "s", we ask ourselves, puzzling at a notice or even our own handwriting). In sum, the world we perceive is a world made in part by other agents who are not in the job of making our life as easy as possible. Moreover, perceptual tasks are not as predictable as introspective tasks. T h e features of the external environment we need to notice have changed, sometimes rapidly. We live in an unstable world. An agent's access to externally stored information is neither as reliable nor as uncontaminated as access to internally stored information. Thus Mr T's external memory is less reliable after dark; when he forgets his glasses; when his pen leaks or his pencil breaks; when it rains and his book gets wet. And we have not yet considered the issue of epistemic sabotage by other agents. For Mr Τ is at risk of thought insertion and deletion to the extent that others have access to his notebook. He may be able to guard against thought insertion by recognising his own handwriting and disregarding anything in his notebook not in his own hand. But that does not solve the problem of having his beliefs stolen. This set of problem simply does not arise for such of Mr T's information that he still codes internally. In assessing the parallel between internal and external memory, we cannot disregard the various threats to which external memory is especially subject, even though in Mr T's particular circumstances some of these are probably unlikely. Unless there is something we have not been told, he is probably not a likely target for active sabotage. But even more important is the fact that external epistemic artefacts are used in shared and sometimes contested space. Indeed, our most important cognitive tools—language, well designed notation systems and the like—are the multi-generational product of many minds. Clark's favoured examples of the use of tools to extend our cognitive abilities tend to be of solitary activities: working through a large multiplication task; an artist using initial sketches in producing a painting; an academic writing a paper by revising drafts; cutting, pasting and annotating his way from one version to the next. Cognition, for him, remains paradigmatically a solitary vice though one prosthenically enhanced by wideware.

Externalism, Epistemic Artefacts and the Extended M i n d

247

How typical are such cases of the role of external tools in extending our cognitive powers? Think instead of conversations, discussions, brainstorming. In such activities the public symbols you use as props and prompts are not your own, and yours too are taken up (sometimes half way through) by others. Likewise, scientific labs are shared spaces, and the tools are often shared tools. Notebooks, experiments, programs and papers are more often than not the result of many hands and minds. The same is true of decision and action in many commercial and administrative organisations. Files, for example, are often joint products. Teaching too is interactive, and the props used in teaching will not the exclusive cognitive product or resource of one individual. External tools are in the public domain, and that has implications for the cognitive resources needed to make effective use of those tools. Jointly used epistemic artefacts are often less than optimal for any of their users: they need to be individualised at each use. When my department is making an appointment and I am considering a candidate's file, I need to check the file to make sure everything is there and to see what others have added. I may need to re-order the material or extract what I need. Moreover, though human interactions are often co-operative, they are not exclusively so. Deception and the hidden agendas of others are issues that arise in the use of such tools. These complications matter. To parody Clark's picture, he thinks reliable access to dovetailed tools explains how agents with dumb brains can nonetheless be smart. It is his Alzheimer's example writ large: external tools finesse the grave limits of the naked brain. The use of epistemic tools in a common and often contested space increases the cognitive demands on the naked brain. Epistemic artefacts are typically (i) jointly produced, (ii) they exist in a common and contested space, (iii) and they often have more than one user. As a consequence, the use of epistemic tools is often cognitively demanding. Agents using common tools cannot afford to be dumb. This fact forges a link between the Clark-Dennett-Mithen picture of minds enhanced by epistemic artefacts and representational externalism. I take that externalism to be the idea that internal states are representations because they are tracking states: their representational properties are constituted by relations to the agent's environment 7 . An externalism worth defending also claims that these representational properties explain the behaviour, or the behavioural capacities, of the representing agent. I shall argue that such representations explain our ability to use of epistemic artefacts successfully.

7

T h e different versions of naturalistic semantics (for example, the different versions of indication theory) give differing accounts of this tracking relation. I shall not attempt to adjudicate between these versions, largely because I think it is unlikely that any single tracking relation corresponds cleanly to a folk psychological notion of content. Nor is it likely that only one tracking relationship is explanatorily important.

248

In Defence of Content Externalism

IV Two Kinds of Externalism In their 1998, Clark and Chalmers contrast their extended mind with representational externalism: "In these cases, the human organism is linked with an external entity in a two-way interaction, creating a coupled system that can be seen as a cognitive system in its own right. All the components in the system play an active causal role, and they jointly govern behavior in the same sort of way that cognition usually does ... Our thesis is that this sort of coupled process counts equally well as a cognitive process, whether or not it is wholly in the head. This externalism differs greatly from standard variety... When I believe that water is wet and my twin believes that twin water is wet, the external features responsible for the difference in our beliefs are distal and historical, at the other end of a lengthy causal chain. Features of the present are not relevant: if I happen to be surrounded by XYZ right now (maybe I have teleported to Twin Earth), my beliefs still concern standard water, because of my history. In these cases, the relevant external features are passive. Because of their distal nature, they play no role in driving the cognitive process in the here-and-now. This is reflected by the fact that the actions performed by me and my twin are physically indistinguishable, despite our external differences. In the cases we describe, by contrast, the relevant external features are active, playing a crucial role in the here-and-now. Because they are coupled with the human organism, they have a direct impact on the organism and on its behavior. In these cases, the relevant parts of the world are in the loop, not dangling at the other end of a long causal chain. Concentrating on this sort of coupling leads us to an active externalism, ... " (1998 pi 1)

I do not think this is the right picture of the relationship of representational externalism to the extended mind. In rejecting this picture I emphasise two themes. One is the importance of portable information for the use of epistemic artefacts, The second is the importance of tracking relations in explaining the successful use of these artefacts. As a consequence of these two themes, in my view our ability to use epistemic tools effectively is not just enhanced by decoupled representations, it depends on our ability to form and use accurate decoupled representations. Embodied cognitive science often explains adaptive action by models in which an agent's representational resources are tightly fused to specific acts and problems. Webb's robot cricket recognises and finds an appropriate mate by being able to follow a sound signal to its source. It does so through having ears structured to channel sound but only of the right frequency, and by having response mechanisms dedicated to driving the cricket to the source of the sound. The system represents (or at least tracks) the source of the sound, but not in any way that is a general fuel for success ((Clark 2001) pp 127-128). The use of epistemic artefacts cannot, I claim, be supported by information tightly coupled to a particular task. Rather, the evolution of multi-purpose representation is tied to the evolution of our capacity to use our epistemic artefacts. I shall return to the role of multi-purpose information shortly; let me now turn to the explanatory salience of tracking. Externalists take representational properties to be some species of systematic, natural relationship between an internal state of the agent and some aspect of that agents' environment. Representational properties are relational properties, and in particular, they are relational properties that explain the existence of cognitive states by explaining

Externalism, Epistemic Artefacts and the Extended M i n d

249

our dispositions to behave adaptively. For the representational properties of thoughts explain the success and failure of agents as they try to achieve their goals on the basis of their representations of their environment. It may be the case, as Clark and Chalmers note, that we can explain the execution of a particular act without appealing to these representational properties of the agent's thoughts. Likewise, it may even be true that in explaining the success of any particular act, we do not need to appeal to the representational properties of thought. For example, in explaining Michael's success in finding the wine, it might be enough to explain his orientation at, and movement towards a particular cupboard. The execution of that action sequence, together with the fact that wine was in the cupboard, explains why Michael found the wine. But a disposition to succeed cannot be explained in the same way. For such dispositions are open-ended and counterfactually robust. If (within limits) Michael would have found the wine wherever it was, this can be explained only on the basis of a capacity to accurately track the wine's location (Sterelny 1990; Godfrey-Smith 1996). Epistemic tools support open-ended and counterfactually robust dispositions to succeed. Thus one distinguishing feature of good field naturalists is their identifications skills about the plants and animals of their local patch. But another distinguishing feature is their ability to use key guides, floras, handbooks, field guides—in short, relevant epistemic artefacts—in unfamiliar patches to identify unfamiliar plant and animals. They use these tools successfully because they know which characters are typically diagnostic; because they know how these epistemic tools are organised; because they can identify habitat types and types of behaviour. Without understanding how these tools are internally structured, and without accurate representations of such biological phenomena, a flora is just dead weight. Field biologists have to learn to use these tools. The standard field guide to the birds of Venezuela illustrates well over a thousand species (Hilty 2002), and describes many more forms of these species (for example, immature birds) using a set of standard terms for describing the parts of a bird's plumage (scapulars, primaries, secondaries, vent etc). A user who does not understand the significance, organisation and presentation of this information would be overwhelmed by its quantity. But once these tools are mastered, they effect a large and open-ended expansion of their user's skill base. Thus a skilled field biologists has a disposition to succeed in identifying plants and animals in a wide variety of circumstances. We cannot explain this disposition without appealing to the accuracy of her ecological and taxonomic representations and her grasp of the organisation of biological data bases. Field biologists thus have an open-ended disposition to succeed in a wide variety of biological projects. Moreover, those dispositions depend on information that is not fused to specific acts in the manner of the cricket's information about the nature and location of a potential mate's call. The biologist's information is multi-purpose. Agents acquiring it typically do not know when, whether, or how this information will be used. A field botanist learning the taxonomically-informative varieties of leaf shape will not know the circumstances in which that information will become relevant to her actions. Agents who live in unstable environments face challenges whose demands cannot be predicted in advance. They are disposed to succeed only if their inner cognitive states both accurately track their environment (often enough) and support a variety of plans.

250

In Defence of Content ExternaJism

I have argued that the use of epistemic tools depends on such rich representational resources, and that the disposition to use these tools successfully depends on the existence of appropriate tracking relations between inner resources and environment. But this argument vindicates no particular version of externalism: no particular hypothesis about the relation between internal state and external environment which constitutes tracking of the state of the environment 8 . Indeed, several different kinds of tracking relationships may have played important roles in the evolution of rich representational capacities (Godfrey-Smith 2002; Godfrey-Smith forthcoming; Sterelny 2003). Let me now connect these ideas about the explanatory importance of tracking to the fact that the information needed to use epistemic tools is often multi-purpose. The cricket's information about the direction and identity of a sound is tied to a very specific purpose. This is not true of the information needed to use biological tools and in this respect they are typical of epistemic tools. Agents with no information about botanically diagnostic characters; agents with misinformation about such characters and agents able to recognise the plants of her own patch only through unanalysed gestalts will all struggle to use a flora. Only informed agents can use a flora. Moreover, agents typically acquire this information piecemeal, without knowing its functional significance. This information is not learned in a form which ties it to specific uses, routines or skills. It is not embedded in executing specific tasks. While this is typical of the informational demands on the use of epistemic artefacts, it is not typical of the informational demands on nonhuman animals 9 . Decoupled representation has coevolved with the use of epistemic artefacts because agents need a rich information base to effectively use epistemic artefacts, and because they acquire that information piecemeal, without knowing its functional significance. If this co-evolutionary hypothesis is right, we could not use epistemic tools without internal states that both accurately track features of our world and which can contribute to an open-ended array of action. And if we did not use such tools, we would have relatively little need of such states. Let me sketch the case for thinking that biological epistemic tools are typical rather than exceptional. First, public representations and other epistemic tools have to be interpreted. Perhaps photos, drawings, maps and the like can be used by treating them as natural signs of the environmental structures they map. But even with such symbols, interpreters need to master the regularities of their fit to the world. A map, for example,

8

However, it does rule some options out. As Peter Godfrey-Smith notes, we can explain an agents successful action on the basis of its accurate representation of its environment, but only if "accurate tracking" is defined independently of success. The facts which constitute tracking the world must be independent of the behavioural triumphs of the mind-owning organism. Godfrey-Smith argues that Millikans ambitious version of teleosemantics does not define what an agents thoughts are about independently of the agent's dispositions to succeed in his or her behavioural plans; Godfrey-Smith, P. (1996). Complexity and the Function of Mind in Nature. Cambridge, Cambridge University Press. If so, the correspondence of those thoughts with the world cannot explain the agent's successes.

9

I doubt that decoupled representation is uniquely hominid, but if this line of thought is right, the extensive use of information not tied to fairly specific tasks may be uniquely hominid This conjecture is supported by the fact that It is surprisingly difficult to find unequivocal evidence of the use of spatial and social maps, or evidence of broad-banded understanding of the causal properties of physical objects in nonhuman animals.

Externalism, Epistemic Artefacts and the Extended Mind

251

maps some aspects of its domain. Thus maps of an underground system typically represent the order of the stations and the connections between the various lines, but they do not map the distance between stops. When distances are mapped, an agent needs to learn the scale of the map, and the arbitrary conventions maps use; for example, using colour to code hight above sea level. Moreover, these features of the use of maps and similar representations must be genuinely learnt. They are variable and contingent, so (in contrast, perhaps, to internal symbols) they cannot simply be implicit in the automatic routines for the use of a representation. Yet, equally, there are regularities: having learnt to use a map of the Sydney suburban network, using that for London is less intractable. Map reading depends quite extensively on portable information. For maps are read with different purposes, and this makes different aspects of a map important. The same map of a chunk of New Zealand's fiordland will be used very differently by hikers, hunters and fishermen. Models and templates, likewise, require interpretation. A fish-trap carries information about how and where to make other fish-traps. But the template cannot be blindly copied, even by an agent who could commit every detail to memory. For the length, depth and shape of the trap must be sensitive to the size and depth of the inlet the trap closes at low tide, and to the strength of the tidal water that will run over it. When another agent makes an artefact for his own purposes, it is not automatically a perfect template for my use: the other agent may be larger or shorter; weaker or stronger; a left-hander. I shall need to modify as well as copy his production. Our informationintensive ways of life often relies on reverse-engineering the tools and skills of others. But reverse engineering depends on a rich prior background that enables an agent to select a model, filter noise from signal (not everything the model does while demonstrating a skill is part of the skill), and adjust for their differences with the model. So the reverse-engineering agent needs a rich, relevant and largely accurate system of decoupled representation. Imitation is sometimes discussed as if it were a cognitively cheap form of learning, but the very limited imitative capacity of the great apes is a signal that it is cognitively demanding, though the exact nature of that demand remains controversial. Symbol systems are now amongst our most important epistemic artefacts. These tools make possible forms of human reasoning that would otherwise be impossible. Without positional notation and without algorithms which decompose large arithmetic operation into elementary ones, accurate quantitative reasoning would be impossible. In using these notation systems we do indeed "store information in the world". Yet the appropriate use of these symbol structures is cognitively demanding. The innumerate are not rare even in western societies making serious attempts to make numeracy skills universal. The genuinely arbitrary symbol systems of language impose greater demands still. In particular, there are persuasive argument for accepting the Grician view that to understand an utterance, we have to represent its speaker meaning (Origgi and Sperber 2000). And while analyses of speaker meaning vary widely, they all involve representations of thoughts; beliefs about beliefs and/or intentions. Perhaps an agent can acquire the concept of a tiger from a tiger-photo, without representing that photo as a representation. I do not see how an agent could acquire the concept of a tiger from tiger-conversations without representing "tiger" as a representation.

252

In Defence of Content Externalism

Furthermore, if an agent must guard against deception or manipulation, then the cognitive load of the use of shared epistemic tools is greater still. For an agent then needs not just bare metarepresentational capacities but assessment tools. Dan Sperber has argued, very plausibly that the meta-tools of folk logic—the notions of truth and evidence, of sound and unsound reasoning—evolved as a response to this vetting problem (Sperber 2001). Counter-deception is a problem whose informatinal load is both heavy and unpredictable: there is no telling in advance what you will need to know in order to know that another is a liar. As I argue in (Sterelny forthcoming), the vetting problem is particularly pressing for linguistically coded information. The arbitrariness of linguistic encoding and the fact that language is not stimulus bound are two of the features which make language such a powerful system. But they also make it a deception-subject-system. Other informational vehicles—especially templates and models, but to some extent diagrams and the like—have intrinsic marks of reliability. It is much harder to fake a tool than a story. The use of epistemic tools in a public space involves quite complex problems of co-ordination. A recipe is a fairly standard example of an epistemic artefact. So consider a group of friends jointly producing a meal by following a recipe. Each agent must (a) monitor what others are doing; (b) negotiate a division of tasks; (c) negotiate a division of shared space and shared work surfaces; (d) negotiate a division of shared tools—who gets to use which chopper when. There will often be reasonably generous margins of error, but they must form and follow a co-ordinated plan. None of this can be done by reflex. It will all depend on what they know of one another, their materials and their tools. W h o gets flustered easily? W h o thinks that if some is good, more is better? Without a rich stock of background information, many joint activities will unravel. Of course, if there are repeated interactions of the same kind, as in a commercial restaurant, then these co-ordination decisions will be standardised and will not be re-litigated case by case. But when friends get together to cook a joint feast, all this must be done on the fly, and usually is, without any great drama. However, the fact that we usually solve such problems without great stress does not show they have a light cognitive load. A lesson of all these examples is that human problem solving makes intensive demands on memory. Very often, when we solve problems, the information we use is not available on line at the site of action. Each agent in the cooking co-ordination problem brings to the kitchen a good deal of knowledge of food and cooking, and (very likely) of the quirks and idiosyncrasies of their partners in gastronomy. The apprentice maker of fish-spears is likely to have acquired information about fish and the water, and about the properties of wood and twine, by trial and error exploration of his local environment. We soak up a good deal of social, technical, psychological and ecological information in contexts which do not signal the specific relevance of that information for later action. Sometimes we acquire information for specific functional purposes. We learn the characteristics of a program like Endnote only because we want to use it in composing documents. But much of the information we acquire and then use in the general exploration of our world is not labelled for its relevance for action. At the time we find out that George thinks there is no such thing as too much garlic, we do not know how or whether this quirk will be relevant to our later actions.

Externalism, Epistemic Artefacts and the Extended Mind

253

So the use of epistemic tools often requires the recall of information not initially acquired to facilitate their use. Moreover, the use of epistemic tools—especially jointly used epistemic tools—requires agents to integrate information across informational domains. A particularly clear example of this integrative demand is the use of artefacts as templates or models. Imagine a boy learning how to make a fish spear from an adult, using a completed example to guide his own construction. To acquire this skill in this way, the boy will need to grasp (a) ecological information: he needs to understand why the fish spear works, and to do this, he will need to understand something about fish, how to spot them, and how to make allowances for misleading perceptual appearances; (b) he will need to be able to interpret the adult's behaviour; for example, to understand that a demonstration is slowed down or exaggerated; (c) he will need some understanding of the physical properties of the materials with which he is working. Learning to make and use artefacts from artefacts will typically require an agent to integrate information about the physical properties of the tool; information about its uses; and information about the intentions of the model. In short, the effective use of epistemic artefacts depends on informational resources internal to the agent. Moreover, for the most part these resources are portable. They are not tied to particular functions Time to sum up the state of play. Dennett, Clark, Mithen (and of course others) have argued that we cannot understand the evolution, operation or development of human intelligence without understanding the multitude of ways in which we act as epistemic agents, transforming the informational load on our own decision making and that of others, and the ways we act as epistemic tool makers, constructing devices that help us meet those transformed informational challenges. These ideas are true and important. In this paper, my own twist has been to emphasise the social rather than the individual aspects of the use of epistemic tools, and (as consequence) the informational load on their effective use. A consequence of this shift in emphasis (it is no more) is that I see the extended mind perspective on human cognition as more nearly complementary with some of the research agendas associated with externalism and the representational theory of mind. This too is only a change in emphasis. Even if I am right in my assessment of the informational load on the use of wideware, that would not vindicate a picture of the mind as an autonomous, sentence-crunching computational engine. But it would vindicate the idea that portable cognitive resources (Godfrey-Smith's "fuels for success"; my "decoupled representations") play a central role in human cognitive life, probably a much more central role than they play in the cognitive life of other agents. Moreover, the importance of portable resources cannot be explained without taking into account their tracking profiles. Portable cognitive resources are fuels for success only because they are both world-tracking and action-driving.

References Clark, A. (1997). Being There: Putting Brain, Body, and World Together Again. Cambridge, M I T Press. Clark, A. (1999). "An Embodied Cognitive Science?" Trends in Cognitive Science 3(9): 345-350. Clark, A. (2001). "Reasons, Robots and the Extended Mind." Mind and Language 16(2): 121-145. Clark, A. (2002). Minds, Brains and Tools. Philosophy of Mental Representation. H. Clapin. Oxford, Oxford University Press: 66-90.

254

In Defence of Content Externalism

Clark, A. (2001). Mindware: An Introduction to the Philosophy of Cognitive Science. Oxford, Oxford University Press. Clark, A. and D. Chambers (1998). "The Extended Mind." Analysis 58(1): 7-19. Dennett, D. (1993). "Learning and Labeling." Mind and Language 5(4): 540-547. Dennett, D. (1996). Kinds of Minds. New York, Basic Books. Dennett, D. (2000). Making Tools for Thinking. Metarepresentation: A Multidisciplinary Perspective. D. Sperber. Oxford, Oxford University Press: 17-29. Dennett, D. (1995). Darwin's Dangerous Idea. New York, Simon and Shuster. Godfrey-Smith, Ρ (1996). Complexity and the Function of Mind in Nature. Cambridge, Cambridge University Press. Godfrey-Smith, P. (2002). "On the Evolution of Representational and Interpretive Capacities." The Monist 85{\)·. 50-69. Godfrey-Smith, P. (forthcoming). On Folk Psychology and Mental Representation. Mental Representation. H. Clapin, P. Staines and P. Slezak, Greenwood Press. Hilty, S. L. (2002). Birds of Venezuela. Princeton, Princeton University Press. Jones, C., J. Lawton, et al. (1997). "Positive and Negative Effects of Organisms as Physical Ecosystems Engineers." Ecology 78·. 1946-1957. Kacelnik, A. (1984). "Central Place Foraging in Starlings (Sturnus vulgaris). I Patch Residence Time." Journal of Animal Ecology 53. 283-299. Kirsh, D. (1995). "The Intelligent Use of Space." Artificial Intelligence 73: 31-68. Kirsh, D. (1996). "Adapting the Environment instead of Oneself." Adaptive Behavior 4{3/4): 415-452. Krebs, J. and N. Davies, Eds. (1997). Behavioural Ecology: An Evolutionary Approach. Oxford, Blackwell. Laland, Κ. N. and F. J. Odling-Smee (2000). Niche Construction, Ecological Inheritance and Cycles of Contingency. Cycles of Contingency. R. Gray, P. Griffiths and S. Oyama. Cambridge, MIT Press: 117-126. Laland, Κ. N., J. Odling-Smee, et al. (2000). "Niche Construction, Biological Evolution and Cultural Change." Behavioral and Brain Sciences 23: 131-175. McBrearty, S. and A. Brooks (2000). "The Revolution That Wasn't: A New Interpretation of The Origin of Modern Human Behavior." Journal of Human Evolution 3.9(5): 453-563. Mithen, S. (1998a). The Supernatural Beings of Prehistory and The External Storage of Religious Ideas. Cognition and Material Culture: The Archaeology of Symbolic Storage. C. Renfrew and C. Scarre. Cambridge, McDonald Institute: 97-106. Mithen, S. (1998b). A Creative Explosion? Theory of the Mind, Language and The Disembodied Mind of the Upper Palaeolithic. Creativity in Human Evolution and Prehistory. S. Mithen. New York, Routledge: 165-186. Mithen, S. (2000). Mind, Brain and Material Culture: An Archaeological Perspective. Evolution and the Human Mind: Modularity, Language and Metacognition. P. Carruthers and A. Chamberlain. Cambridge, Cambridge University Press: 207-217. Odling-Smee, F. J., Κ. N. Laland, et al. (1996). "Niche Construction." American Naturalist 147: 641-648. Odling-Smee, J., K. Laland, et al. (2003). Niche Construction: The Neglected Process in Evolution. Princeton, Princeton University Press. Origgi, G. and D. Sperber (2000). Evolution, Communication and The Proper Function of Language. Evolution and The Human Mind. P. Carruthers and A. Chamberlain. Cambridge, Cambridge University Press: 140-169. Sperber, D. (2001). "An Evolutionary Perspective on Testimony and Argumentation." Philosophical Topics 19, 401-413. Sterelny, K. (1990). The Representational Theory of Mind: An Introduction. Oxford, Blackwell. Sterelny, K. (forthcoming). Folk Logic and Animal Rationality. Rational Animals? S. Hurley and M. Matthew Nudds. Oxford, Oxford University Press. Sterelny, K. (2003). Thought in a Hostile World. New York, Blackwell.

Outing the M i n d — A Teleopragmatic Perspective ROBERT VAN GULICK

I The Space of Ins and Outs Spatial metaphors shape our talk and thought of minds and worlds and how each stands to the other. We talk and think at least implicitly in terms of borders, boundaries and containers. Mental acts occur within the "inner world" of mind, but outer things are also said to be "in mind" when we want, sense or think of them. The things we see with outward looking eyes do not of course themselves get taken in like foods we eat, yet in a sense they cross within when once perceived, or so at least the inner-outer metaphor would have it. Action and perception serve as the dual interface where what occurs in one linked ordered realm can reach across and have effect upon the other. The inner-outer metaphor has long had critics (Dewey 1925, Ryle 1949, Wittgenstein 1953) and has been much challenged in recent years, both by those who hold that the nature of mind depends more intimately on supposedly external factors (Putnam 1975a, Bürge 1979, 1982, Wilson 1995) and by those who yet more radically call the very division between inner and outer itself into question (Alston 1999, Noe and O'Regan 2001). My present aim is modest: merely to offer a view of the issues from one particular perspective, that of the teleopragmatic theory of mind and the sort of nonreductive pluralism that naturally follows from it. (Van Gulick 1980, 1992, Lycan 1987) At the outset, we need to distinguish three distinct, though not wholly unrelated, sorts of questions, each of which is really a family of questions that subdivides and clusters along a number of further dimensions. • The Type-Individuation Issue—To what degree and in what ways, if any, are (or should) mental properties or kinds (be) individuated on the basis of external factors? • The Causal Issue—In what ways, if any, are externally individuated mental properties or kinds causally relevant? • The Ontological/Constitution Issue—In what ways, if any, are mental items (states or processes) constituted in part by external factors? I will address each of these three families of questions in turn, but first I need to sketch the basic teleopragmatic view that provides the perspective for my survey.

256

In D e f e n c e of C o n t e n t F.xternalism

II The Teleopragmatic

View

The teleopragmatic theory is a version of functionalism (Block 1980), according to which something counts as a mental state, or as a state of a particular mental type, because of how it functions within the organized system of which it is a part. Mental kinds are functional kinds. A given state counts as a belief that snow is white, a desire for a cup of coffee or a pain in one's left knee in virtue of the functional role it plays. Similarly a process is one of reasoning, remembering or perceiving in virtue of how it functions within its containing system. Mental items differ from non-mental ones not in their underlying substantial bases, but in their roles and functions. That is the core of functionalism, but immediate questions arise: What counts as a relevant functional role? When are roles equivalent? And in general how are such roles to be categorized and individuated? Much of the criticism of functionalism is in fact directed at specific answers to these questions rather than at the basic theory itself. For example, objections have been raised against particular versions of the view that aim to define the relevant functions and roles purely in terms of abstract computational relations, simple causal links of production and inhibition, or purely individualistic internal roles (Searle 1980, Baker 1986, Block 1980). One could reject all these specific proposals without rejecting functionalism per se. However, to give the general doctrine any real import one must spell out the key notions of functional role and functional equivalence. The teleopragmatic view aims to do just that. It treats the relevant notion of function as a teleological one, i.e. as one that defines roles at least partly in terms of how they contribute to the probable realization of the ends or goals of their containing system (Dennett 1971, 1978, Van Gulick 1980, Sober 1985, Lycan 1987). The biological analogy is automatic; hearts do many things (e.g. they make noises) but only a few concern their function (to pump the blood). The teleopragmatist views minds as fundamentally biological, at least in the original and paradigm cases, so it is only natural to use a biologically based notion of function in describing the relevant roles (Millikan 1984). The roles associated with memories, perceptions and intentions are all concerned with their teleological functions, i.e. with what they are supposed to do in order to contribute to the successful operation of the mind and minded organism of which they are parts. The pragmatic aspect of the view falls out as an immediate consequence. Mental states and processes function to guide the organism to success in dealing with its environment and world. Thus the fulfillment of a role is always a matter of contributing, whether directly or indirectly, to some practice and the end or goal at which it aims. The effectiveness or adaptivity of a given item thus depends in part upon what goal or end is aimed at. (Putnam 1972, 1975, Van Gulick 1980). The two defining aspects of the teleopragmatic view are clearly manifest in the way it treats the psychologically central notion of information. The idea of minds as information processors has been a commonplace for forty years or more (Lindsay and Norman 1972), but what is meant by "information" is often less than clear (Dennett 1971, 1987, Dretske 1981, Van Gulick 1980). An adequate information processing model of mind will need to use many different notions of information. Moreover, at least some of them will have to involve the active possession of information by the organism in a way

O u t i n g the M i n d — A Teleopragmatic Perspective

257

that allows (at least potentially) for its adaptive application to behavior. Purely passive notions of information transmission or storage that concern mere reliable covariance, such as those associated with information theory, will not suffice for all the respects in which we need to think of minds and organisms as information processors. For example, the sense in which information about the visual scene is present in a frog's cortex is not the same as that in which it is present in the light array that impinges on its retina. Both sorts involve nonrandom covariance, but the frog's cortically realized information is also available to direct the frog's behavior in ways likely to enhance the realization of its goals, whether that be catching flies or avoiding predators. It is only because the frog can apply the information in such ways that it can be said to possess or embody the information in a sense that entails understanding it. Being informed or possessing information in that sense is thus essentially teleological and pragmatic in nature. Having distinguished these importantly different senses of information, we should nonetheless be cautious about how we draw their boundaries in real contexts. As we will see below, there may not always be an objective answer to the question of where one crosses over from information storage in the merely passive sense to that which involves more active possession. T h e teleopragmatic view also has important methodological implications. It offers a perspective not only on the mind, but on our attempts to model and understand it as well. It regards all understanding as having an essentially pragmatic aspect, and this quite naturally commits it to a form of theoretical and conceptual pluralism. Moreover, that commitment applies just as much when it is minds themselves that are the objects of our understanding. W h a t counts as the right or best way to explain or describe mental states or processes will depend on our specific cognitive situation: what it is about them that we wish to understand, and relative to what sorts of interests, goals and contexts of engagement we aim to do so. T h e modes of representation or conceptualization best suited to one such context may be quite different from those optimal relative to another. Nor should we expect there to be some neutral or universal perspective or context of engagement that simply models or represents the object of our understanding just as it is in itself. Teleopragmatism thus naturally combines with the sort of nonreductivism and commitment to the autonomy of the special sciences that has been widespread in the philosophy of mind since at least the mid nineteen seventies (Fodor 1974, Putnam 1975b, Boyd 1980). According to the nonreductivist, we need a plurality of conceptual and theoretical schemes to cognitively engage our world, which shows order and regularity at many different levels and provides modes of epistemic and interactive access along many different parameters. T h e nonreductivist is a physicalist in accepting that everything real is physically realized, but he denies that the physical sciences supply all the concepts and frameworks we need to understand the world. I would need to say a great deal more to do full justice to the teleopragmatic view, but for present purposes I hope it will suffice to stress the following key features. • It situates minds in a biological context. • It appeals to a teleological notion of function that thus involves goals, ends and interests.

258 •

•

In Defence of Content Externalism

It treats cognition and understanding as having an essential pragmatic aspect that thus depends upon the modes of access and engagement between knower and object. It is committed to theoretical and conceptual pluralism.

Ill Out of What? Having identified our teleopragmatic vantage point, we can now look from it at our first family of questions, those concerning type-individuation: To what degree and in what ways, if any, are (or should) mental properties or kinds (be) individuated on the basis of external factors? Two issues require immediate clarification: Which mental kinds or properties is one asking about? And what is meant here by an "external factor", i.e. external to what and external in what sense? The relevant mental kinds might be either among those referred to by common sense ("folk") psychology or those invoked by some present or future scientific theories. The answer is likely quite heterogeneous. Whether and to what degree we should type-individuate in externally sensitive ways will depend on which particular mental states or processes we are considering, as well as on the explanatory context in which we are appealing to them. For example, it might be best to individuate memory states in birds externally if we are concerned with explaining their success in seed storage and retrieval, but more apt to individuate them internally if we were concerned to explain their neural basis and temporal transformation. This is just what the teleopragmatist would expect, given her commitment to conceptual pluralism and contextual explanation. Nonetheless, one would like to say more than, "It all depends." Thus in sections IV and V below I will discuss what teleopragmatism implies about which sorts of mental states and contexts most lend themselves to external modes of individuation and which do not. However, first we need to clarify the issues raised by our other initial queries: What is meant here by "external"? External to what? And external in what sense? "External" can be taken quite literally to mean "spatially outside the physical body of the organism." In that sense the distal features of the organisms environment all count as external, and perhaps even proximal stimuli might qualify. The chair that I am looking at across the room, the chemical composition of the liquid that fills the rivers on my planet (Putnam 1975a) and the linguistic practices of the American English speaking group to which I belong (Bürge 1978) are all external in this extra-bodily regard. On this interpretation, the boundaries of my skin and other surface membranes mark the border between internal and external factors. That is surely one legitimate and potentially useful way to distinguish inner from outer, and more or less the one invoked by those individualists who take molecular duplicates to share all their non-external mental states (Stich 1978). However, drawing the boundary at the skin may not work in every case. In particular what is outside the mind would not seem to always coincide with what is outside the body. When I feel a pain in my sore right Achilles tendon or touch my left index

Outing the M i n d — A Teleopragmatic Perspective

259

finger to my nose with eyes closed in a physical exam, are the bodily states that are the objects and causes of my experiences to be considered inner or outer? They are internal to my body and shared with my molecular duplicates, but relative to the mind/nonmind boundary they seem not unlike the bookcase I see to my left or the wind I hear blowing in the trees outside my window. In such cases it seems more apt to draw the inner/outer border between the mind and body, or between that part of the body that is the organism's mind (which may or may not coincide with its brain) and the rest of its body. However, the existence of any such border may be dismissed as an illusion by those who regard mind as a fundamentally embodied phenomenon or who believe that we should think in terms of minded organisms and not reify minds themselves as things (Ryle 1949, Dennett 1987). If minds are not parts of bodies as stomachs are, then asking which part of the organism is its mind might be more like asking which part of it is its life (or is alive) than like asking which part is its heart or digestive system. Though the option of drawing an intra-body line between mind and non-mind regions is appealing, it remains controversial. Moreover, even if such a line can be validly drawn, it may not serve in every case to mark the inner/outer border relevant to the question of externalism. Just as body parts, such as tendons and fingers, may be aptly counted as external in the relevant cases of proprioception or bodily sensation, so too may various mental states when they are the objects of our self-awareness. If one distinguishes between the state of awareness and object of awareness and asks how the content of the state depends upon its object, then it may be apt to regard many of our introspected mental states as external in the relevant sense. Relative to introspection, we seem to need an inner/outer border that separates domains within a single mind. Consistency may oblige the externalist to make such an intra-mental division, but it is far from obvious how it might be drawn. What counts as relevantly "external" to a given state cannot be a matter of simply isolating it from all other mental states. That would count too much. The content and properties of a mental state might depend upon other mental states in many ways that seem orthogonal to the issue of externalism. For example, it is a functionalist platitude that a state can be a belief that ρ only if it exists within the context of many other contentful states including many other beliefs (Dennett 1978, 1987, Block 1980). Thus one must find a more subtle way of drawing an intra-mental border appropriate for asking the externalist question in the introspective and memory cases. The goal would be to mirror that which divides outwardly directed mental states from the sorts of environmental factors on which externalists think their contents and properties may depend. For example, according to the teleopragmatist, informational relations including those involving nomic covariance between neural states and environmental features play an important role in determining the contents of perceptual states (Dennett 1970, Van Gulick 1980, Dretske 1981). To the extent that introspection is a matter of inner-directed perception, one would expect informational relations to play an analogous role. The contents of meta-mental states would depend in that sense upon the features of the mental states that were their objects and which anchored the originating end of the relevant informational connection.

260

In Defence of Content Externalism

Thus one might perhaps appeal to informational links and their termini as factors in drawing the inner/outer border in the introspective case. The way in which the content of a meta-mental state depends informationally upon its object seems more relevant to the externalism question than that in which the content of a belief depends upon its associated inferential network. Thus there may be principled ways of applying the inner/outer distinction in the intra-mental context without needing to fence off mental states from all the mental relations on which their contents and other mental properties holistically depend. Some intra-mental relations might be regarded as external in the relevant sense though others were not. One final caveat is needed. If there are self-reflexive mental states—which I take to be more than an idle speculation—then it may be impossible to draw a line that divides meta-mental states from their objects, since any such reflexive state of self-awareness would be its own object. If self-reflexive states turn out to play an essential role in the nature of consciousness (Van Gulick 2001, 2004), then relevant inner/outer lines may need to be drawn not merely within a single mind but even within a single mental state. Having found the question, "External to what?" more complicated than at first it may have seemed, we need to consider its companion question, "External in what sense?" Given a pluralistic outlook, we should probably not expect there to be only one relevant sense of external dependence. Heterogeneity again is likely. At a minimum, we need to distinguish two quite different sorts of dependence, which we can call supervenient dependence and descriptive dependence. The former is primarily an ontological matter, but the latter turns more on practical and methodological issues. X's having property F (or being of type T) superveniently depends upon properties of some Y just if changes in such properties of Y can alter the fact of whether or not X is or is not Ε Put another way, in cases in which Y is distinct from X, the base of facts on which X's being F (or not F) depends includes properties of items other than X itself. Any such factors are thus among those that constitute the fact of its being F. Strongly relational properties, such as being a father or a sibling, depend on outside factors in this supervenient sense. Esau became a sibling when his brother Jacob was born, and his being a sibling continued to depend superveniently on the fact of Esau's existence. By contrast in a case of descriptive dependence what is at issue in not the fact of X's being F, but our ability to specify or describe the relevant type or property. As a practical matter we may have no way of adequately referring to the relevant property other than via various external relations. Roughly, we might say that X's being F descriptively depends on Y's being G just if we have no adequate way to describe or refer to the fact of X's being F other than by appealing to facts about Y's being G. Thus whether or not X's being F descriptively depends upon various facts about other items will vary contextually with our state of knowledge and our theoretical and descriptive powers. It is in part a practical and pragmatic matter which should probably include some additional parameters, so that we should say that X's being F descriptively depends on Y's being G, with respect to cognitive agent A at time t (or epistemic context E). Externalist theories about the individuation of mental types might involve either supervenient or descriptive dependence. The externalist might make either a claim of ontological dependence or merely one about our practical need to appeal to external factors in our theorizing. There seem to be cases of both sorts.

O u t i n g the M i n d — A Teleopragmatic Perspective

261

On one hand, folk psychological states of knowledge (and other so called "factive" states) superveniently depend upon external factors. I can not be in a state of knowing that the US has attacked Iraq unless the US has indeed attacked Iraq. In that sense the mental property of "knowing that P" depends upon external factors in the supervenient ontological sense. The fact of my knowing that the US has attacked Iraq is constituted in part by extra-mental events in the Middle East . By contrast, the folk psychological concept of belief does not seem in any obvious way to involve a similar external ontological dependence. My beliefs are not limited to what is true. Nor can events or changes in the Middle East by themselves alter the fact of whether or not I believe the US has attacked Iraq. Closer examination may also reveal such ontological dependence in the belief case, as we will see below, but it is not obvious as in the knowledge case. However, even if there is no ontological dependence, we may still be unable in practice to specify or describe the relevant mental property without appeal to external factors, in particular to those that must hold for the belief to be true. As a practical matter, folk psychology provides us no real means of specifying most beliefs independently of their satisfaction conditions (and the same holds for most wants, intentions, wishes, or thoughts). Thus the folk psychological property or type of "believing that P" may descriptively depend in that sense on external factors, whether or not it superveniently depends on them.

IV Type-Individuation

and Success

Having clarified the meaning of "external", what can the teleopragmatic view show us about its application? As theoretical pluralists, we should expect a comprehensive understanding of the mind to require many modes of individuating mental types, but within that heterogeneity there may be contexts in which externalist modes are more apt or less so. Can we say anything about which sorts of contexts incline one way or the other? Given the diversity of contexts in which mental types and properties are invoked, we can not survey the full range of relevant contexts. Instead I will focus on two specific issues relevant to the larger question. For the balance of this section (IV) my concern will be with the notion of success, and in the next section (V) with that

of causation. Many mental and psychological properties and kinds are "success-loaded", including many of those invoked by folk psychology (Ryle 1949, Dennett 1971, Bürge 1986b). The case of "knowledge" is simply the most familiar. Many other folk psychological kinds also embody a success element and thus entail and depend upon external factors. The paradigm senses in which one can be said to see, remember, hear, learn, perceive or understand all involve an assumption of accuracy or correctness. Though one can misperceive, misremember or misunderstand, the default implication in the un-negated case is one of getting it right. Nor can one see what is not there, although one may seem to see things that are in fact absent. Thus all these common folk notions are success-loaded in the relevant sense, and the fact of one's being in any such state depends

262

In Defence of Content Externalism

or supervenes upon external facts about their accurately represented objects. The fact of my seeing the cobalt blue coffee mug on my desk supervenes not only on my inner experiential state but also on the mug, the desk and the other relevant features of my immediate visual environment. Were those latter items not present, my mental state would not be one of seeing a cobalt blue mug on my desk but at most one of seeming to see one. Success-loaded notions are widespread in scientific psychology as well. The information processing model of mind relies throughout on types and kinds that embody a veridical notion of information (Lindsay and Norman 1972). An organism can on occasion be misinformed and get things wrong, but in so far as it is receiving, storing and applying information it is informed about its world in a sense that entails its largely "getting it right" (Dennett 1971, Van Gulick 1980, Dretske 1981). I am not claiming informational notions are used always and only in the success-loaded sense—indeed information-based theories must accommodate the notion of error and misrepresentation—but they are ubiquitous in that sense and the whole information processing model of mind could not really get off the ground without them. Success notions are used in both folk and scientific psychology to describe the capacities and powers of organisms and agents. When we want to specify an organism's mental abilities or achievements—whether to perceive, retrieve or act upon information—we typically describe it in terms of its successful environmental engagement. Before we can address the question of how its capacities are realized, we need a clear account of what they are, a specification of what it knows or what it has the ability to come to know about its world. Thus one obvious task for which we need success-loaded mental types is that of describing many of the psychological capacities we aim to explain. Capacity descriptions are of interest as well in comparing the relative abilities and lacks of different organisms and agents. Organisms differ widely in what they can perceive, remember or understand, and we need an adequate system of description for categorizing their respective powers as well as their similarities and differences. Interesting generalizations often emerge at the level of environmentally characterized powers. For example, animals occupying similar ecological niches, even if biologically quite distant by measures of evolutionary separation, are often found to have convergently similar powers and capacities for detection and sensory processing of the environmental features that are relevant to their similar ecological contexts. Perhaps more importantly, the fact of what capacities an organism or agent has often depends upon external environmental factors. An organism's detection capacities or its power to acquire reliable information through its sensory channels will typically depend upon the special structure of the range of environments in which it operates. Detection abilities for example depend upon the range of possible confines the organism is likely to encounter. As the epistemological literature on reliability has long noted, features that reliably correlate with and indicate a property in one set of contexts may fail to do so in another. Many output powers are similarly contextual. An organism's ability to adapt its behavior in response to input from its sensory channels will often depend upon the context in which it is embedded. As noted above, in section II, the teleopragmatist regards an organism's capacity to respond in appropriate ways as an essential aspect of its being

Outing the M i n d — A Teleopragmatic Perspective

263

informed or possessing information in the active sense. In so far as such powers to use information are contextually dependent, so too is the degree to which the organism can be said to be informed in that key sense. Thus it seems clear that for the task of describing mental powers and capacities there will often be good reason to individuate mental types in contextually and externally sensitive success-loaded ways. Still someone might argue that when we turn from describing capacities or achievements to explaining capacities or successful behavior, we must shift to noncontextual and non success-loaded modes of typing (Fodor 1980). Appealing to success-loaded notions to explain capacities or achievements might seem to pose a threat of circularity, of making a "virtus dormativa" appeal that explains successful F-ing by appeal to a capacity to F. Though some appeals to success-loaded capacities might indeed reduce to triviality, that surely is not so across the board (Dennett 1978). We need only be sure that the success-loaded capacities to which we appeal are not too closely defined in terms of the very achievement or ability in need of explanation. Explaining an organism's ability to detect food toxins by appeal to a toxin detector, would not provide much of any explanation, but explaining its ability by appeal to the spectral sensitivity of its visual system and the reliable correlation between toxicity and spectral profile of foods in its natural habitat could provide a genuine if partial explanation. We often explain complex capacities as the result of interactions among simpler capacities, and the components invoked in the explanandum can themselves be success-loaded in externally sensitive ways. The relevant component abilities may be at the organism-level (personal level) as in the toxin detection case or abilities at the suborganism (sub-personal) level. Visual perception supplies many examples of the latter sort. For example visual object recognition relies upon the ability to compute object shape from retinal stimuli, and shape detection is in turn achieved through a variety of processes many of which are contextually based. For example, the "structure from motion" process (Marr 1976) computes the three dimensional shape of moving distal objects from their dynamically changing retinal projections. However, those computations reliably recover information about three dimensional shapes only in contexts in which moving objects are rigid and not deformed by motion. Thus the subpersonal processes carry information about object shape only on the assumption that the organism is in fact operating in an environment in which objects satisfy the rigidity constraint. We are able to explain the organism-level ability to recognize object types only by appeal to a contextually sensitive information processing ability to compute shape from motion at the sub-organism level. As such examples make clear, contextual modes of individuating mental states are not restricted to the task of describing whole organism abilities. They also figure importantly in explanations of capacities and processes at the sub-organism level. The contextually sensitive nature of information processing at the sub-personal level should also make us skeptical, or at least cautious, about the common strategy of trying to decompose success-loaded mental types into a purely internal mental component plus a non-mental external component. Internalists might argue that factive states, such as knowledge and perception, that entail the existence of what is known or perceived, should be treated as hybrids whose mental nature is restricted to its inter-

264

In Defence of Content Externalism

nal aspect, perhaps to a state of believing or seeming to see. T h e external world must conform to the way it is represented by that state in order for it to count as a case of knowing or perceiving, but the fact of how it represents the world might be regarded as itself independent of any external factors and determined solely by internal factors. Such an approach may seem initially to provide an attractive way for the internalist to cordon off the domain of the genuinely mental from dependence on external factors, but its plausibility fades in the face of examples such as the "structure from motion process", which make plain how deeply inward contextual dependence penetrates and the degree to which what, if anything, is represented often depends upon the specific structure of constraints in the organism's environment. There is a convergence at this point between the basic teleopragmatic view and the general and widespread appeal made to the principle of charity in the intentional interpretation of behavior. Both offer reason to doubt the basic strategy behind the hybrid view, that of attempting to isolate the external aspect of success-loaded states in a separable non-mental component. T h e hybrid view treats representational states, such as belief, as purely internal states and then adds a separate success component to generate knowledge. It regards belief per se as primitive and true belief as simply a special and derivative case. However, for the teleofunctionalist, external aspects of success come in at the ground floor. T h e central and correlative teleofunctional notions of active information possession and goal directedness are both defined in terms of the organism's successful engagement of its world. As a biologically oriented view, its focus is on the explanation of adaptive behavior rather on behavior simpliciter. Obviously it needs to accommodate false belief, misperception and other forms of cognitive error, but it treats them as derivative cases within a framework that regards true belief and accurate perception as the default values and defining cases of their type (Van Gulick 1980, Bürge 1986b). A similar presumption lies behind most appeals to the principle of charity (Quine 1960, Davidson 1970, Dennett 1971, 1987) which instructs us to interpret intentional agents so that most of their beliefs are true. The principle is motivated not by considerations of tolerance or civility but by a recognition that the very notion of a beliefforming system or a perceptual system that regularly and systematically produces mostly error is at base incoherent, or so at least it is claimed. Such views are widely held but not uncontroversial, and I cannot here prove them right. For the purpose of this present survey, I hope it will suffice to note that if one accepts such a view of the principle of charity, as so many do, then even non-factive intentional states such as belief turn out to depend upon external factors far more than the hybrid model would suggest. Though individual beliefs may carry no specific entailments about the truth of the external conditions they represent, they are nonetheless possible only within the general context of a larger system of beliefs that on the whole tracks the relevant portions of the world. Thus the fact of whether or not some system counts as a believer essentially depends upon external facts about its world and its relation to it. An internalist might concede that our folk notions of belief and perception are contextual and externally dependent in such respects, but still argue that it is possible and necessary for many purposes to individuate mental types in ways that are narrow and not similarly sensitive to external factors. Narrowly individuated types might seem

Outing the M i n d — A Teleopragmatic Perspective

265

particularly apt for some scientific purposes, and it would seem we should be able to isolate an internal component of our psychological engagement with the world that exhibits systematic regularities and lends itself to description in insightful explanatory terms (Fodor 1980). Given my teleopragmatic commitment to theoretical and conceptual pluralism, I would certainly not want to preclude the development of such internal types nor deny their possible utility. One can be a robust externalist about the mind and still allow that for some purposes it may be apt to individuate mental types internally, and one should remain open to that option. However, one should be cautious about assuming that there must be an isolable internal mental component that exhibits interesting regularities and order when considered and described apart from its context. O f course, there must be an internal component, but whether and to what degree it in itself exhibits any systematic mental regularity or order is less obvious. It might fail to do so either because it showed no systematic regularities at all or because it showed none that were mental. It might for example exhibit only computational or physical regularities of a sort that could not be regarded in any interesting sense as mental. The threat seems more than a mere idle possibility, and thus the existence o f a purely internal yet systematically regular mental component should not be taken for granted. From a teleopragmatic perspective, understanding is often a matter of reciprocal cognitive engagement. Mind and world interact and track each other in mutual two-way causal loops. Models that focus solely on internal factors are unlikely to capture the dynamic order of the mind if that order results in large part from causal arcs that pass out into the world, effecting changes there that then loop back to bring about successor mental states. The flow from mental state to mental state often takes the regular form it does only because it passes through such inner-outer-inner causal loops. Consider three such examples involving visual perception, sensory motor coordination, and social relations. Philosophers sometimes treat vision as a passive process in which externally generated stimuli act upon our sensory organs to initiate an inputdriven linear process that results in an inner representation o f the world. In fact, vision is a far more active process o f expectation, exploration and engagement (Gibson 1979). Retinal inputs are o f course very important, but how they get interpreted downstream depends upon a wealth o f more internal factors. Prior set and perceptual readiness can affect which stimuli get attended to and processed, what they are taken to indicate, and which of various competing interpretations if any comes to dominate the output of the system. Those internal factors are of course themselves in many cases the results o f prior environmental inputs. Moreover, we should not think of the retinal stimuli as static inputs. In normal real world cases the stimulus array constantly changes as we explore our visual environment. Even when our bodies are at rest, our eyes are almost always in directed purposive movement. We saccade many times a second, and as we do the retinal input changes in ways that reflect the world-eye dynamic coupling. However, as we well know, the visual world does not jump or slide around as we saccade our gaze, and that is because the visual processing system "anticipates"' or "expects" and "compensates" for the retinal image changes that result from those shifts. Indeed the system's prior representation o f the visual scene consists in large part o f its being disposed to receive just such a changing flow o f inputs in response to its shift o f gaze. Ecologically based theories, like

266

In Defence of C o n t e n t Externalism

those in the tradition of J.J. Gibson (1979), sometimes speak of harmonic or resonant couplings between the organism and its environment. The reciprocal nature of this link is all the more important when the organism's perceptual systems are coupled with its actions. As the organism alters not only its perspective on the environment but the environment itself, it generates a continuous sequence of reafference that feeds back into a coordinated process of matching, expectation, and accommodation, as the system's inputs and outputs mutually entrain each other in successful purposive action. The simple act of reaching out to lift my coffee mug and tip it back enough to sip a last bit of the still warm and pleasantly bitter liquid that lies at its bottom involves a complex interplay of flowing input-output streams that engage each other in systematic and regular coordination. It is easy to be seduced by the picture theory of perception and by the model of seeing as the having of an image in the mind. But most perception is in the service of action or the potentiality of action, and the cortical representations it produces must carry information in a way that is able to coordinate that sensory-motor engagement in a smooth and regular way. We do it so effortlessly in most cases, that we easily lose sight of the difficulty of the task and the nature of the processes that underlie it. However, their complexity becomes readily apparent once we address the link explicitly, as when we try to design robotic systems capable of even limited purposive behavior. We must also resist the temptation to push those sensory motor representations down to the unconscious or nonphenomenal level. Again the picture theory threatens to seduce us into restricting the phenomenal aspect of perception to the having of an inner image and demoting all the processes of sensory-motor coordination to some lower level. From a teleopragmatic perspective, understanding is always a practical matter of the capacity for appropriate response and interaction, and that is no less true of conscious understanding. The states produced in me when I look around my room count as my awareness of it in large part because of how they would (or could) subserve my actions toward it. The harmonic feedback of input and output—of afferent, efferent and reafferent flow—is an essential component of the phenomenology of acting in the world and thus of our experience of the world itself. Thus even at the phenomenal level the reciprocal nature of mind-world engagement is manifestly expressed. Social interactions provide some of the most compelling cases of mutual cognitive engagement, not only in humans and higher mammals but even in far simpler social organisms. Consider for example the mating dances performed by jumping spiders. The male, who is typically much smaller than the female, must approach his prospective mate with just the right gestures, despite having had no prior opportunity to observe or learn what they might be. If his initial signals are correct, they elicit a specific response from the female which in turn provokes a counter response from the male and so on in a continuing and innately programmed pas de deux which ultimately results in successful copulation, though also sometimes in the female's making a post coital meal of her suitor. The male spiders stock of innate information thus embodies a good deal of understanding about the mating rituals of its species, all of which is embedded in a control system that functions only within the specific context of its reciprocal social engagement with conspecific females. It is only relative to that specific external context that the innately determined internal organization of the male can be seen to exhibit

Outing the M i n d — A Teleopragmatic Perspective

267

any order or regularity. It is a sort of dynamic key that takes its shape only in its active mutual interplay with the lock to which it matches. What is true of jumping spiders applies as well to far more sophisticated social interactions, including many of those we humans participate in all the time. Our ability to play our respective parts in many situations relies upon the implicit social understanding which is built into the complex systems of response and counter-response through which we mutually and all but automatically generate the coordinated flow of our interpersonal relations. T h e dynamic lock-and-key model again seems apt. If we were to consider just one side of any such engagement in isolation from its reciprocal social counterpart, it would exhibit little if any coherence or regularity. That order emerges only in the context of mutual interaction. T h e implicit understanding that underlies our social relations thus provides a third example of reciprocal cognitive engagement. Like visual perception and sensory motor coordination, it too involves forms of cognitive organization that exhibit regularity and order only in contexts of mutual dynamic engagement. Such examples should make us cautious about assuming that there must always be a well behaved mental component that exists solely within the cognitive agent considered in isolation from its world. At least in some cases, the absence of any such component must be regarded as more than a mere skeptical possibility. We might summarize the findings of this section in four main claims: 1. For many purposes it is appropriate to individuate mental-types in ways that are descriptively and/or superveniently sensitive to external factors. 2. Externally sensitive types are relevant in many folk psychological and scientific contexts, at both the whole organism (or personal) level and at lower levels as well. 3. Externally sensitive types are often apt when we aim to describe or explain capacities, as well as when we rely on mental notions that are less obviously but still implicitly success-loaded, such as information, representation and even belief. 4. We should be cautious about assuming that in every mind-world situation there must be a coherent isolable purely internal mental component.

V Causal

Relevance

What then of our second family of questions? Are externally individuated mental types causally relevant? As we have just seen, they can be theoretically useful in many contexts, but not all explanatory or descriptive uses require causally relevant kinds. So their usefulness does not by itself resolve the issue of their causal relevance. However, before we address that further question, we need to clarify the key notion of causal relevance, which might be read in at least two quite different ways, one stronger than the other. In the weaker or more inclusive sense, we might count as causally relevant any kind that "can be usefully invoked in the context of a causal explanation or generalization." Alternatively we might restrict causal relevance to those kinds which pick out causally potent properties, i.e. properties that actually "make things happen." All kinds that are causally potent in that latter sense would seem to be causally relevant in the weaker sense, but not vice versa.

268

In Defence of Content Externalism

There are at least two important sorts of cases in which we invoke kinds or properties in causal explanations despite their not being causally potent in the strictest sense. They are importantly different, and the properties invoked in the two sorts of cases fall short of strict causal potency in dissimilar ways. In some cases, the properties are not causally potent at all, at least not with respect to the relevant effects, while those invoked in the other sort of case are aptly regarded as potent but perhaps not in the strictest or ultimate sense, in as much as their powers derive from those of more basic properties that underlie or realize them. Causal explanations of the first sort are those in which we usefully appeal to generalizations that invoke kinds or properties that are reliable nomic indicators or markers of strict causal potency though themselves not potent, or at least not so with respect to the effects being explained. We may for example know that items that have Ρ1 reliably also instantiate some other properties whose specific identity we do not know but which we have good reason to believe endow such items with the power to produce C* effects in some range of situations Si,...., Sn. Thus generalizations about PI type items causing C* effects in Si,..., Sn situations might be quite useful in causal reasoning and explanation. Why did X act in way C*? Because X is a PI and was in a Sn situation. For example, we might answer the question of why Jones did not get sick despite his close contact with his daughter during her bout of chicken pox, by noting that he had been exposed to the disease earlier in his life. In this case the biographical or historical property of "having been previously exposed" serves as our PI, but its causal relevance depends upon its reliable correlation with the present property of "having residual circulating antibodies to the herpes zoster virus." It is the latter property that is causally potent, yet given the generally reliable correlation between past exposure and current circulating antibodies, we can usefully invoke the former property in causal generalizations and inferences about who is likely to become ill in the relevant situations. Moreover, one could do so even if one did not know what the correlated directly causal properties were, as long as one had good reason to believe that there were some such properties. We know for example that being of African American descent substantially increases one's likelihood of developing hypertension, but we do not know what the more specific factors are that underlie the link. They are likely at least in part genetic, but social and dietary aspects may play some role as well. Despite our ignorance of the specific pathways, it is still relevant for a physician to accord ethnic background some degree of causal relevance in making clinical predictions or deciding whether and how to treat a borderline case. O n e might want to draw further distinctions among different sorts of markers depending on the type of link that underlies the correlation between marker and potent property. A pharmaceutical company might produce its various medications with distinctive shapes and colors to avoid confusion. However, even if those correlations were quite reliable, they would remain more or less arbitrary and conventional. By contrast, the link is more intimate with other markers that track their potent correlates more naturally. Being male may be only a marker for midlife risk of hair loss or cardiovascular disease, with the actual causally potent properties being hormonal factors that correlate with gender. However, given the natural if less than invariant biological links between

Outing the M i n d — A Teleopragmatic Perspective

269

gender and hormone levels, the correlation between marker and potent property seems less accidental and more a matter of overlapping or naturally clustered causal factors than in the colored pill case. Though we may thus wish to distinguish among different sorts of marker cases, they all share the basic separation between the indicator appealed to by the explanation and the actually potent property that it tracks. Explanations of the second sort invoke dispositional and functional properties which serve as more than mere markers or correlates of causally potent kinds. To have the relevant disposition is to really have the associated causal power, though some might nonetheless deny that dispositions and functional kinds are causal in the strictest sense since their powers derive from those of the underlying properties in their specific realization bases. We can predict the behavior of a circuit in response to a current if we know its electrical resistance or estimate the rate of a reaction in the presence of a catalytic enzyme, but the specific causal interactions that produce the net outcomes in those situations do not rely upon those functional properties per se. It might seem to be the realizing properties rather than the realized functional ones that do the actual causal work. Two substances that are equally malleable (or fragile) may predictably respond in the same gross ways to equivalent forces, but the underlying explanations in the specific cases may be quite different at the level of micro-realizations, which is arguably where the real "pushing and pulling" gets done by properties that are causally potent in the "make it happen" sense. Nonetheless, I think we should be cautious about demoting dispositional properties to inferior causal status. It is not obvious that strict causal potency resides only at the realization level. The fact that the causal powers of the functional property derive from those of its realizers need not deprive the functional property of real causality. An organic solvent does really have the power to dissolve the carbohydrate sample the chemist places in it, a radio amplifier does in fact have the power to increase the energy of the signals that it transmits, and the analgesic the doctor gives her patient does indeed have the power to reduce his pain. In each case, there is a more specific causal story to be told at the realizer level, but that does not entail that the attribution of causal power to the realized property is false, as opposed to being simply less than ultimate. Just as one can draw further distinctions within the marker category, one might also draw further distinctions among functional and dispositional cases, based on how strictly one wished to interpret the notion of causal potency. If nothing less than ultimate basic micro-level links count as strictly causal, then most functional and dispositional properties are not strictly causal. However, if one follows ordinary practice and construes causal potency more broadly to include macro-properties among the realm of those that make things happen, then many functional and dispositional properties qualify as having real causal powers. Functional and dispositional properties are thus linked more tightly with strict potency than mere markers are. A marker could in principle occur without its normally correlated potent property, but a dispositional or functional property cannot occur without some realization base of underlying causal properties. As a matter of ontology, something can have the relevant functional or dispositional properties only by having some micro-properties or other that realize them, but as a matter of explanation we can and often do invoke those functional properties in causal explanations without need-

270

In Defence of C o n t e n t Externalism

ing to know anything about the details of their specific realizations. One typically does not need to be able to explain how or why the wire has the resistance that it does or how the enzyme facilitates the reaction, but only that it does so and with what magnitude of effect. Having distinguished among various ways in which a property or kind might have or lack causal relevance, we can return to our core question and ask in what respect, if any, externally individuated mental types are causally relevant? In particular are there any good reasons for believing that externally individuated mental kinds are not causally relevant in ways that internally individuated mental kinds are? Mental properties of the two sorts seem on a par with respect to what we have called weak causal relevance. Both externally and internally individuated types get invoked regularly in the context of causal explanations. Thus if there is any significant difference in their causal status, it must concern their causal relevance in the stronger sense that requires causal potency of the "make it happen" sort. Are externalist kinds more likely to lack causal potency in the strict sense for either of the two reasons we considered above: either because they are mere markers for potent properties or because they are potent in only a derivative sense and depend upon underlying realizers for the causal powers? Neither option seems likely to selectively disadvantage externalist mental properties. Though external kinds may serve as mere indicators in some causal explanations, they are not typically limited to such a role. As noted above, we often appeal to contextual or externally individuated causal powers to explain an organism's behavior, especially its successful behavior: the fish fled because it acoustically detected the ecologically reliable waveform signature of an approaching predator, the bees maximized their foraging results by coordinating their memories of past nectar availability with the contextually based timing signal provided by solar position across the sky, or Tom got to the meeting on his own by retracing the route he had traveled to the conference last year. Indeed the most prominent recent candidates for epiphenomenal status as mere markers have not been externalist kinds but experiential qualia (Jackson 1982, Chalmers 1996, Kim 1999), and they are most commonly treated as internal kinds. The causal status of qualia is not an issue I can address here (nor is their internality), but the existence of controversy about alleged epiphenomenal qualia only underscores the fact that there seems to be no reason to judge externalist mental properties as at a comparative causal disadvantage as mere markers. Nor do externalist mental kinds seem obviously any worse off in the dispositionalfunctional regard. Though teleopragmatism regards most externalist mental states as functional, it regards internalist states the same. Whether looked at narrowly or widely, most mental states of either sort appear to be essentially functional in nature. Desiring, remembering, intending, meaning, and sensing all get understood as functional processes from both perspectives. The differences between the wide and narrow points of view all turn on which functional relations and networks get included or reflected in one's model, and not on whether functional relations and networks are essential. T h e externalist extends the essential base of functional links and relations further out into the world and requires his model to reflect differences within that extended sphere as factors in how it functionally characterizes or individuates more internal ele-

Outing the Mind—A Teleopragmatic Perspective

271

ments of the network. T h e internalist, by contrast, may decline to include those more distal features in her scheme, but she will still need a model of functional relations to capture her alternative narrower network. Whether one models widely or narrowly, it is functional relations that are crucial. Thus from the teleopragmatic point of view, internalist and externalist mental kinds are equally functional in nature, and neither should be any more vulnerable than the other to challenges of causal potency that turn on general claims about the status of functional properties per se. Arguments such as those that aim to restrict causal power in the strictest sense to the level of ultimate realizers (Kim 1990) would work as well (or not) to deprive internalist mental kinds of such status as they would against external kinds. They would either work against both or against neither, and thus they give us no ground to prefer one over the other as having superior causal status. T h u s if the anti-externalist critic is to find any reasons for regarding externalist properties as less causally relevant in the strict potency sense, they will have to turn on more specific differences between internalist and externalist kinds. T h e local nature of causation would seem to provide the prime candidate for demoting externalist properties to inferior causal status. T h o u g h direct action at a distance involves no contradiction and remains a logical possibility, there seems ample evidence for believing that is not how things work in our world. As a matter of natural fact, real world causal transactions appear to depend essentially upon local interactions. W h a t happens at a distant time or place can affect what happens at a local point, but only via a chain of intervening links that bridge the separation through a sequence of more immediate cause and effect connections. T h u s it might be argued that the sorts of distinctions among mental states drawn by the externalist can not affect their causal powers in the strict sense. T h e externalist distinctions turn on solely non-local differences, which in themselves can have no impact on local causal powers. T h u s if strict causation is always local, mental states that differ in only external respects cannot differ in any of their strict causal powers. We might formalize that line of reasoning as follows: Local Causation Argument (LCA) Ρ

1.

1 Ρ 2,3,

2. 3. 4.

Ρ

5.

4,5

6.

6

7.

Direct (unmediated) causation is always local (i.e. involves causes and effects that are spatio-temporally contiguous.) Direct causal powers supervene solely on local factors. Externalist distinctions involve non-local differences. Mental items that differ only in external respects can not differ in their direct causal powers. An item's direct causal powers exhaust its strict causal powers. Mental items that differ only in external respects can not differ in their strict causal powers. External factors are not relevant to a mental item's strict causal powers.

272

In Defence of Content Externalism

In so far as causal relevance is read in the strong sense of strict causal potency, the Local Causation Argument (LCA) gives a negative answer to what we have called the causal question. According to the LCA, externally individuated mental kinds or properties are not causally relevant in the strict sense of being potent in the "make it happen" way. That at least is what the LCA purports to show, but is the argument sound? Though it seems initially plausible, at least two important objections might be raised against it. The radical option would be to dispute the local nature of causation that underlies its first two steps. The more moderate, and more promising, alternative would be to challenge step 5 which restricts strict causal powers to direct causal powers. The notion of a "strict causal power" and the intuitive idea of "making it happen" that lies behind it have has no settled meaning, but as we will see below there are good theoretical reasons in this context for not restricting it to direct causal powers. As noted above, there is no logical bar to causal links that are non-local but direct and unmediated by intervening cause and effect transactions. Direct action at a distance is a logical possibility, but according to the LCA it is not a real world nomic possibility. Science rather than logic rules it out, and premise 1 is thus put forward on empirical rather than a priori grounds. But does the scientific evidence in fact support such a claim? There are at least some reasons for caution. Though Newton famously declined to offer any explanations for gravitational attraction, his classical mechanics seemed to treat it as involving action at a distance and was thus decried by others as appealing to suspect occult forces. More contemporary physical theory deals with continuous fields of force, and general relativity recasts gravity in terms of the continuous curvature induced in the structure of space itself by matter. Thus as currently conceived gravity may not pose much of a threat to the localist view of causation. However, other aspects of current physics may be more problematic. Quantum mechanics, in particular the phenomena of so called "entangled particles", might seem to require a more holistic model of causation, where what happens in one part of a large system is dependent upon what happens simultaneously in other spatially distant regions. I lack the scientific expertise to adjudicate such claims, so I will leave it to others to challenge premise 1 (Silberstein and McGeever 1999). My own concerns focus more on step 5 and its restriction of strict causal powers to those that are direct. Since the notion of a "strict" causal power has no agreed upon meaning, it is difficult to say whether or not it extends to non-direct powers. What really matters for present purposes, is whether or not contextual causal powers are significantly inferior in causal status to direct powers. Does the local nature of direct causation imply that contextual causal powers are less robustly real or potent? Step 5 of the LCA assumes it does, but there are good reasons to think otherwise. All parties to the dispute should nonetheless agree that certain counterfactual supporting conditionals are true. If we let χ be the item whose causal powers are at issue and C be the relevant context, then are there a number of conditionals about χ that are true in C but not otherwise. For example it may be true in C that if χ does F then R will result and if χ does G then S will result. "We can assume that none of these conditionals are true in any relevant alternative non-C contexts, nor true of χ considered noncontextually. However, there will be other slightly more complex conditionals that are true of χ noncontextually, namely those that follow our list above but add the con-

O u t i n g the M i n d — A Teleopragmatic Perspective

273

ditions of C to their antecedents: If C obtains and χ does F then R will result, and if C obtains and χ does G then S will result. Those latter conditionals are true of χ independent of its actual context. Which conditionals specify x's strict causal powers? D o those in the latter set alone do so? O r do both sets of conditionals describe equally real and potent powers? Again the internalist critic eager to demote externalist properties to inferior causal status would opt for a positive response to the first (Fodor 1990), but the externalist would likely affirm the second (Bürge 1986b, Van Gulick 1989). Contrary intuitions can be generated. The internalist can argue that whether x's doing F results in R does not turn on any changes in χ itself. The different cases all involve changes in the causal structure and potential of the larger context, while χ retains the same causal profile throughout, namely that given by the more complex conditionals that specify what results it would produce by various actions in a diversity of possible contexts. Suppose for example two identical electrical switches in two lamps that differ only in that one contains a functioning light bulb and the other a burned out one. Turning the first switch will light its lamp, but turning the second will not. Yet it may seem odd to say that this is because the first switch has a causal power the second lacks. Does it not seem more reasonable to regard them as having exactly the same causal powers, and to explain the difference in outcome as the result of causal differences in the non-switch part of the lamp? The internalist must take care not to beg the question by simply asserting that contextual powers are not real because they do not supervene on x's intrinsic structure, since that is just what is at issue. Yet the internalist can try to sway our intuitions on that point by appeal to examples such as that of the lamp switch, or that of two keyless men in identical rooms, one of which is locked and the other not. In all such cases, the differences in outcome seem due entirely to differences in the causal powers of the context not in the causal powers of the relevant x, whether switch or man. The switches themselves do the same thing in both lamps, and the differences in outcome all concern what happens further down the causal chain. However, the externalist can reply in kind and try to pull our intuitions in the other direction. Contextual powers are not mere markers or indicators of causal potency, but concern the very causal chains that produce the actual results in the relevant contexts. And it is those wider contextual powers that often rightly concern us. I want the brakes in my car to have the power to stop its movement without need of further conditions. Few drivers would regard two sets of brakes, one of which is properly connected to the wheels and one of which is not, as having the same causal powers because the unconnected set would produce the desired stopping in other contexts in which it was connected. O u r practical concern as motorists is rightly with the wide or contextual power, and as the brake example shows such powers do not seem in any significant sense less real or of inferior causal status. Or so the externalist claims in hope to tilt our intuitions in her favor. How should one respond to this conflict of competing intuitions? Which sorts of cases are most relevant for assessing the causal powers of externalist mental properties or kinds? We can individuate causal powers widely as well as narrowly, but should we do so and why? The teleopragmatic perspective is of value in answering these questions and perhaps not surprisingly provides at least four reasons for taking wide causal powers seriously.

274

In Defence of Content Externalism

First in so far as we are explaining in actual cases how organisms or mental agents make things happen, our concern is often with explaining external environmental results, either particular external effects or perhaps more typically with explaining regular patterns of such results or the capacities to produce them. Consider the case of camouflage (Van Gulick 1989). An organisms power of camouflage and its ability to avoid detection does not supervene on its intrinsic properties alone but depends as well upon the visual properties of its surroundings and the perceptual discriminatory powers of the predators in that environment. Two lizards might be molecule-for-molecule duplicates and alike in all their intrinsic properties yet differ greatly in their powers of camouflage if their natural environments differ in their background appearance or in the discriminatory powers of their local predators. It is these contextually based causal powers that matter to the biologist interested in explaining or modelling population changes or evolutionary developments. From that perspective it would be more appropriate to group together two species that were intrinsically quite different but alike in camouflage ability vis ä vis their respective environments than it would be to group together intrinsically similar species that differed significantly in their respective contextual powers to avoid detection. The same often applies in the case of mental properties, as the allusion to predator detection powers already suggests. As noted above in section IV, many psychological explanations aim to explain successes, and the mental powers and abilities to which they appeal are contextually based. Our powers to perceive, detect, communicate, carry out our intentions and engage the world successfully in so many other ways typically depend upon aspects of our causal profile that are potent or effective only in a restricted range of contexts that resembles those we actually inhabit. Thus it is to such wide contextual powers that we naturally appeal for such purposes. This leads us to our second point: the relevant contextual aspects to which we appeal in individuating the wide causal powers of mental properties are not accidental, but typically involve a tight adaptive match between mental and environmental causal structures. The contextual dependence of causal powers in the mental domain is not random as in the imagined case of the two identical men in their respectively locked and unlocked rooms. As the teleopragmatist would lead us to expect, the causal profiles of our mental states and processes have been shaped through selection to reflect the causal and informational structures of the particular environments in which they operate. That they have the wide powers that they do within that range of environments is not a mere coincidence, but rather the result of a long process of environmentally based selection and adaptation. Third, as a practical matter the wide causal powers of our mental states are often the only ones we are able to specify or describe. Though we typically know what wide powers an organism or agent has and what wide results it is able to achieve, we often know neither the causal profile of the narrow powers involved nor which features of the context crucially interact with that narrow profile to produce the wider powers. We may know a great deal about the organism's in situ ability to detect predators but little about the intrinsic causal structure of its perceptual mechanism or the environmental aspects to which they are tuned. Thus as a practical matter there may often be no option for specifying narrow causal powers instead of wide ones.

O u t i n g the M i n d — A Teleopragmatic Perspective

275

Moreover, our inability to specify the narrow causal powers in such cases may result from their nature as well as from mere ignorance, which brings us to our fourth

and final point. The narrow components that underlie some of our wide causal powers may not exhibit much regularity or order if considered apart from their contextual embedding. In that respect examples like the lamp switch may mislead us. We should not assume that if χ has a specifiable wide power in a given environment, we can factor it neatly into two tidy components: a regular and coherent narrow causal power plus an equally describable but distinct environmental embedding. Organism-environment relations, including mind-world relations, often involve a dynamic resonance whereby each side of the causal encounter shapes and reshapes the other in an unfolding lockand-key process. The two sides of the encounter may be so tuned and adapted to the other that neither exhibits much systematic order or regularity except in the context of their mutual dynamic engagement as in the spider mating case discussed above in section IV. This is not to deny that in such cases there is a narrow component, but it reminds us not to assume that it must lend itself to regular or coherent description. It may exhibit such order only in its wider context. In such cases we have ample reason for taking the contextual powers seriously and not regarding them as inferior in causal status. Indeed it is the narrow causal powers that seem derivative. Their order and coherence depends upon contextual factors, and they have the natures that they do only because they exhibit systematic regularity in just those sorts of contexts. These four reasons together give us good grounds for taking wide causal powers seriously and for not being persuaded by examples like the light switch case that aim to tilt our intuitions against regarding them as real and strictly potent. Thus contrary to step 5 of the LCA, an item's strict causal powers are not necessarily exhausted by its direct causal powers, and the local nature of direct causation in itself provides no basis for failing to accord wide powers real causal status. With regard to our second key question, we can thus conclude that externalist mental properties and kinds are as causally relevant as internalist properties, not merely in the weak sense of being usefully invoked in causal explanations, but in the strong sense of being causally potent as well.

VI Outside Parts and

Constitution

We come finally to our third family of questions, which concern the Ontological Constitution Issue: In what ways, if any, are mental items constituted in part by external factors? This final set of questions is perhaps the most radical. One might give externalist-leaning answers to our first two questions but still balk at doing so for the third. One might agree that for at least some purposes it is apt to type-individuate mental states in ways that are sensitive to external factors, and even acknowledge that such kinds can be causally relevant in both the strong and weak sense, but still deny that the relevant mental items are themselves composed or constituted of anything non-internal. How one might wonder could my thoughts, my experiences or my desires themselves have actual parts that are not internal to me?

276

In Defence of Content Externalism

Some may object that I have made the question of external constitution sound more radical than it is, since there are some fairly modest ways of giving it a positive answer. According to the relational theory of propositional attitudes, intentional mental states involve a relation between the agent of the attitude and its object which is typically a proposition. Georges believing that Saddam is alive is a relational complex that includes both George and Saddam among its constituents, at least in so far as one takes an objective view of propositions as themselves being complexes having individuals, properties and relations as constituents. If Saddam is a constituent of the proposition that Saddam is alive, then the fact of George's believing that Saddam is alive involves an overall complex having external items such as Saddam among its constituents. A similar conclusion may appear to follow from the sort of externally supervenient dependence, discussed above in section IV. If Otto's knowing or seeing that the gun is smoking supervenes in part upon the smoking gun, then should we not regard that gun as a constituent of that epistemic or perceptual fact. If so, then we seem to have come by external constitution "on the cheap", i.e. to have found a way to derive an affirmative answer to the external constitution question from fairly familiar and not especially controversial views. What should one make of such suggestions? How radical is the external constitution thesis? That depends upon how one unpacks the relevant notion of mental items having external constituents and on which sorts of mental items one is asking about, whether mental facts or mental states. One might concede that the fact of Otto's seeing the smoking gun or the fact of George's believing Saddam is alive include external items—the gun and Saddam—among their constituents but still deny that external items are constituents of Otto's perceptual state or of George's belief. One might subdivide the supervenience base on which such mental facts depend into mental and non-mental spheres. A fact of perceiving would be a complex that consists partly of an internal mental state of an agent plus various non-mental contextual items to which that mental state was related. There are obvious analogies with other sorts of relational facts. The fact of Esau's being a sibling or a brother supervenes in part on factors external to Esau—e.g. on Jacob, but it does not follow that either of the brothers is himself even partly constituted by external elements. The fact of their being brothers depends on factors external to each of them, but the things that are the brothers are just Esau and Jacob. Similarly one might argue that the gun is not a constituent of Otto's mental state, even though that mental state counts as a seeing of the gun in virtue of relations that hold between them. We can thus restate the Constitution Question as follows: Given that mental facts supervene upon complex sets of external and internal factors, are there good reasons to subdivide those complexes into mental and non-mental spheres in ways that restrict the mental components to the purely internal domain? One might reject such an option either because one thought there could be no principled division of the relevant complexes into mental and non-mental spheres, or because one thought that such a division, though possible, would not restrict the mental sphere within the internal domain. If one frames the issue in terms of functional networks as the teleopragmatist would, then the question becomes that of whether one should divide the relevant networks into mental and non-mental portions, and if so whether the mental portions of the network are all

Outing the M i n d — Λ Tclcopragmatic Perspective

277

internal in the relevant respect (e.g. external to the organisms body or brain)? The teleopragmatist should again opt for a pluralist approach. There are many legitimate ways to subdivide the relevant networks, and though some may restrict the mental to the internal domain, others will treat it as reaching much further out into the network. Any such partition should be made on principled grounds, for example on the basis of the similarities or dissimilarities between the types of relations that obtain in different regions of the network or the kinds of dependencies or interdependencies among them. What the teleopragmatist should reject is the anti-externalist claim that there is no legitimate way of parsing the networks that includes external items among the constituents of mental states. Indeed the teleopragmatist can offer at least three sorts of reasons for sometimes including external portions of the network within the mental sphere. They deal respectively with the functional informational nature of the networks, the reciprocal causal links within them, and their social dimensions. Let me say just a bit about each. Informational links play a key role in binding together the elements of the networks on which the mental supervenes. Some connect organisms and agents to their environments, but others occur solely inside the organisms or wholly within the external environment. As information flows through the environment across the sensory interface and through the nervous system, it preserves aspects of its order and content as it is repeatedly transformed and modified in medium and format. Information is carried not just across inner-outer links but outer-outer and inner-inner ones as well. In all cases, the local transitions and links are similar in important ways. Order and information in the covariational sense passes from one element or region to the next, although its physical basis typically changes as does its format and the accessibility of the relevant content in many cases. Of course as noted above in Section II, there is an important difference between being informed or possessing information in the active sense and merely doing so passively. Active possession requires the ability to use the information in some adaptive way; one must have at least the capacity to act or respond in ways that are shaped by the relevant information. However, the relevant difference may in practice often be more a matter of degree than kind, and when we look at the details of actual networks, the similarities between internal and external informational links may be more salient than their differences. Many inner-to-inner links may primarily involve the transmission or transformation of passive information in ways not significantly unlike those found in outer-to-outer links. Imagine for example a functional causal arc that passes from environment to sense organ, to sensory nerves, to sensory motor cortex, to motor nerves, to muscles and out to environmentally focused motion. Ordered patterns and the information that they carry may be preserved across every link in that sensory motor arc. Those patterns get transferred to different media or substrates as they move through the sequence, and they get transformed in other ways as well, especially in their format and mode of encoding the relevant content, but if one asks at which stage the transformation from passive to active information possession occurs, one may be hard pressed to find any such boundary. The organism does indeed make use of the information that comes in its sensory channels to adaptively guide its movement, but it does so by being organized in such

278

In Defence of Content Externalism

a way that the sequence of passive-to-passive informational transitions that occur are just the ones that aptly match the pattern of its response to the pattern that it picks up from the environment. O r to put the point in more explicitly functionalist terms, the system's active possession of the information is a matter of its being functionally organized to produce the appropriately adaptive match between the order and pattern of its output and that of its environmental input. And that functional organization in turn is realized by the requisite sequence of passive-to-passive transformational links. T h e point of making these observations is not to deny that there are interesting differences between the informational processes internal to organisms and those out in the environment. T h e teleopragmatist does not deny that. T h e aim is to stress that there are also lots of important similarities between internal and external informational links, perhaps enough to justify some externalist modes of dividing up the overall networks. Reasons of our second sort concern the reciprocal relations that hold in some parts of the relevant networks and the many ways in which inner and outer portions of the network depend upon each other. As we noted above in section V, mental relations often involve a dynamic engagement between the organism and its environment. T h e capacity for successful interaction that constitutes the organism's understanding of its environment often relies upon causal loops that pass back and forth between inner and outer domains in a sort of resonant reciprocation. In such cases it may be difficult to isolate coherent purely internal processes that one might identify as the mental components of the relation. It may be more apt to see the relevant understanding as realized in the whole process of inner-outer engagement rather than as restricted to its inner aspects alone. T h e point is highly abstract, and some specific examples may help to make it more intelligible. Consider two such cases, the first involving visual perception and the second a kind of memory. O n the sort of Cartesian inner observer model that might seem plausible from a strongly internalist perspective, visual stimuli initiate a causal chain that results in a distinct and separate internal state of visual awareness and understanding. T h a t state may in turn produce behavioral responses that affect the environment, but both the internal and external components of the overall causal interaction are well defined in distinct and independent ways. However, the empirical facts about visual perception may suggest a quite different model, one that treats visual experience as a form of engaged exploration (Gibson 1966, 1979, Noe and O'Regan 2001). Though it may seem that at any moment we have a detailed internal representation of our overall visual environment present to us in experience, in fact we are able to generate such detailed information at any given time for only a small portion of the visual field with the rest monitored and represented at a much lower level of resolution if at all. However, we can readily shift our gaze and visual attention to whatever features or regions concern us and gain access to detailed information about them. Thus as long as the complete information is in fact available from the environment, the organisms active visual engagement gives it virtually instant access to it. Looked at in that way, the organism's visual experience of its surroundings is not solely a process produced internal to the organism but equally a matter of their ongoing interactive engagement. T h e organism's experience of that portion of its world involves its being visually in contact or coupled with it in ways that provide multiple

Outing the Mind—A Teleopragmatic Perspective

279

shifting avenues of access and coordination. Such models offer a more externalist view of the constituents of visual experience, one which perhaps includes external objects themselves as constituents. They view experience not as something that happens solely in the head but as a mental process that involves both the organism and its world as well as their engagement (Alston 1999, Noe and O'Regan 2001). As a second example, consider the sort of memory process that might be involved in retracing a route one has followed in the past, perhaps through a forest or an unfamiliar city. Internal traces play a role in generating the appropriate behavior but so too do contextual features and cues from the environment. O n e might be able to follow the route by actually re-walking it and taking advantage of the interaction between current stimuli and the low levels of residual activation or sensitivity left in one's inner neural networks by one's prior perceptions along the path. Those inner components might be too weak and incomplete in themselves to enable one to produce any explicit account of the route, but nonetheless in context they might enable one to navigate successfully even without explicit recall of the path. T h e past structure of the route is preserved partly in the neural traces left by one's past encounter but also partly in the current external structure of the route and the items found along it. Were those external features altered to any great degree the inner traces by themselves could no longer function to guide one in making the right moves. Thus not only is ones memory contextually dependent, but one could say some of the constituents of the effective memory are ontologically external. T h e structure of information that is needed to accomplish one's goal is distributed broadly across external as well as internal regions of the relevant functional network. Like the visual experience case, the path memory example does not compel us to parse the relevant networks in ways that include external elements as constituents of the relevant mental processes, but it does give us further reason for regarding that as a plausible option. Let us turn then to our third and final set of reasons, those that appeal to the social aspect of the functional networks on which many mental states and processes supervene. Social holism has played an important role in the development of contemporary externalism, but it has been invoked more typically in regard to type-individuation questions (Bürge 1979). However, I believe it can provide support as well for externalist views on the constitution issue. Three social aspects of the mental are especially relevant: the social division of epistemic labor, the social embodiment of rational constraint, and the socio-cultural nature of the sphere of purposive action. From a teleopragmatic perspective, each of the three offers a plausible basis for delineating the sphere of the mental in ways that include external items among its constituents. Hilary Putnam's famous phrase "the division of epistemic labor" (1975a) reminds us that cognitive achievements often depend upon the collaborative activity of multiple social agents and their diverse epistemic contributions. For the teleopragmatist, knowing or understanding some aspect of reality requires the ability to engage it, interact with it, and respond in ways that adaptively reflect that understanding. O n e has to have some mode of access or contact with it—indirect and distant though it may be—as well as some capacity to act in ways adaptively shaped by that contact, though those links as well may be quite indirect. T h e causal functional networks that tie a given cognitive agent to the objects of her understanding will often include links that pass through

280

In Defence of C o n t e n t Externalism

other cognitive agents and depend upon the further causal functional links that they in turn connect to within the larger network. My ability to know or even succeed in thinking about the distribution of matter and energy in the early universe depends essentially upon my social links to the enterprise of current astrophysics and its elaborate structure of observational systems, theoretical models, computer programs and specialist practitioners who make their specific contributions to the group activity by which they (and indirectly we) engage those distant regions of space and time. It is only because I am embedded in such a larger network that any words or images I may use internally can have the meanings or references they do and by which they enable me to think about those long ago and far off parts of spacetime. In this cooperative aspect the outer portions of the network are very much like those that lie within. The networks within the organism's brain or nervous system that underlie its mental processes typically involve the cooperative interaction of many special purpose processors that carry out more limited mental or epistemic tasks. Each is limited in its capacity, but when they are organized effectively they endow the containing organism with mental powers far greater than those they individually possess. The parallels between inner and outer collaboration are captured vividly in the AI theorist Marvin Minskys (1986) notion of the "mind as a society of selves." Viewed from a teleopragmatic perspective, the sorts of epistemic cooperation that exist within the inner and outer portions of the overall functional network appear as perhaps differing more in degree than in kind. In both cases, collaborative interactions between parts of the relevant network produce a greater degree of understanding through the more far reaching and effective engagement with the world that they make possible. When one looks at the overall network as a web of cooperative epistemic and teleofunctional connections, the borders marked by the physical boundaries of particular animal bodies may not seem nearly so salient or significant. The social embodiment of rational constraint provides a second phenomenon that may justify giving an externalist answer to the constitution question. It is an axiom of the functionalist model of mind that a state can be a belief or other intentional attitude only if exists within a larger network of states that exhibit a significant degree of rational coherence relative to their intentional contents (Dennett 1978, Van Gulick 1980). O n a functionalist model, exhibiting a wealth of such inferential and evidential connections is essential to realizing the role associated with the relevant intentional state. Nor is it just functionalists who endorse such a linkage. Similar views of rationality constraints as constitutive features of propositional attitudes have been advanced by philosophers who are not explicitly functionalists such as Donald Davidson (1970) or even some like John Searle (1983) who take themselves to be anti-functionalists. Thus the thesis of such a link is widely, if not universally, accepted. The rationality constraints must be implemented by underlying causal processes in particular systems. They might for example be realized internally by syntactic rules governing formal processes that operate on representations in an inner language of thought. It has often been suggested that by mirroring semantic relations syntactic rules might thus bridge the gap between rational norms defined over intentional contents and purely causal regularities realized in physical structures (Fodor 1981, Haugeland

Outing the M i n d — A Teleopragmatic Perspective

281

1985). No doubt internal regularities, whether syntactic or otherwise, play an important role in implementing the required constraints. What is not always recognized is that external social processes often also play a significant role. Many of the critical and corrective processes that keep our beliefs consistent, coherent and in correspondence with the available evidence operate within the realm of social interchange and mutual discussion. The need to articulate and defend our beliefs in conversation with others often forces us to recognize inferential connections we might otherwise ignore. Contradictions and unsupported beliefs that might persist in the isolation of a single mind will more readily be detected and removed if exposed to the rational pressures of social dialogue. Like mutually interacting metering systems, cognitive agents can keep each others' belief more accurately anchored to the evidence. Nor are social processes limited to the removal of error and the reduction of cognitive dissonance. They have a positive rational aspect as well. Through communication and cooperative intellectual activity we often extend the inferential reach of our beliefs and more fully draw out their positive implications and overall coherence. If the rationality constraints are indeed essential to the propositional attitudes, then in so far as those constraints are implemented in external as well as internal processes, it is appropriate to view those attitudes as at least partly constituted by external aspects of the overall functional network. The cultural context of purposive human action provides a third social reason for taking constitutional externalism seriously. Much of our behavior counts as meaningful and goal-directed only because of the social cultural context in which it takes place. Only relative to that external structure of artefacts, practices, institutions and enabling technology does the relevant behavior exhibit systematic regularities of the kind that meaning and goal-directedness require. The dependence of meaning on social cultural context is almost too obvious and familiar to mention. Yet it is worth asking how it looks from a teleopragmatic perspective with its biological orientation and its focus on functional networks. When we think ecologically about biological systems, we often find it useful to define the domain of the biological in a way that includes more than just organisms. Though only organisms may be strictly speaking alive, much else might be considered as part of the biological domain. One might ask how a specific system, such as a lake or swamp has been altered by the presence of life within it. What dynamic relationships hold within it because of the organisms it includes? The answers to such questions will typically appeal to objects (nests, burrows and beaver dams), processes (nitrogen fixation, fermentation or greenhouse effects) and relational structures (food webs, eutrophying lakes or rain forest eco-climate systems) that would not exist at all, or at least not in anything like their present form, in the absence of life. I think it is fair to count them all as parts of the biological domain. By analogy, one might also count the cultural and social structures within which human action takes place as part of the mental world. From an ecological perspective, the mental is not something that exists solely within individual agents but throughout the larger network of organized structures that provides the ongoing evolving context for meaningful purposive action. Just as we might count corral reefs and carbon cycles as constituents of the biological world, so too we might count languages, law courts,

282

In Defence of Content Externalism

and baseball leagues as constituents o f the mental, at least o f the particular human forms it takes in our present context. None o f these social phenomena—the division o f epistemic labor, the social embodiment o f rationality constraints and the cultural context o f purposive action—compels us to count external items as constituents o f the mental, but each gives us further good reason to include externalist models among our options on the constitution issue.

VII Retrospective Having examined each o f the three main questions distinguished at the outset, we can ask what our teleopragmatic survey has revealed about the terrain o f the externalism debate. Four main results stand out. • Given its commitment to theoretical and conceptual pluralism, teleopragmatism entails that we should not expect any o f our major questions to have only a single right answer. • Each o f the three main issues turned out to encompass a wealth o f more particular questions that merited careful articulation and distinction. • With respect to each more particular question, we found there were good reasons for taking an externalist answer seriously, if not as the only plausible answer at least as one o f the pluralistic options to which we should remain open. • T h e externalist options in many such cases owed their appeal in part to features that are central and paradigmatic o f the teleopragmatic perspective such as its view o f understanding and cognition as processes o f reciprocal engagement between organisms and their worlds, the essential dependence o f mental states on their roles within large functional networks, and the importance o f externally anchored notions o f success, adaptation and goal achievement. Thus I hope our survey has demonstrated both the value o f the teleopragmatic perspective and the interest and importance o f the externalist options it reveals.

References Adams, F. ( 1 9 9 3 ) . Fodor's modal argument. Philosophical Psychology, 6, pp. 4 1 - 5 6 . Alston, William ( 1 9 9 9 ) . Back to the theory of appearing. In J. Tomberlin (ed.) Philosophical Perspectives 13, Epistemology. Oxford: Blackwell Publishing. Baker, L. ( 1 9 8 5 ) . A farewell to functionalist«. Philosophical Studies, 48, pp. 1-14. Baker, L. R. ( 1 9 9 4 ) . Content and context. Philosophical Perspectives, 8, pp. 17-32. Block, N. ( 1 9 8 0 ) . Troubles with functionalism. In N. Block (ed.) Readings in Philosophy of Psychology Volume 1. Cambridge, MA: Harvard University Press. Boyd, R. ( 1 9 8 0 ) . Materialism with reductionism: what physicalism does not entail, In N. Block Readings in Philosophy of Psychology Volume 1. Cambridge, MA: Harvard University Press. Braun, D. ( 1 9 9 1 ) . Content, causation, and cognitive science. Australasian Journal of Philosophy, 6 9 , pp. 375-89.

Outing the Mind—A Teleopragmatic Perspective

283

Bürge, Τ. (1979). Individualism and the mental. Midwest Studies in Philosophy, 4, pp. 73-122. Bürge, Τ. (1982). Other bodies. In A. Woodfield (ed.) Thought and Object. Oxford: Oxford University Press. Bürge, Τ. (1986). Intellectual norms and foundations of mind. Journal of Philosophy, 83, pp. 697-72. Bürge, Τ. (1986b). Individualism and psychology. Philosophical Review, 95, pp. 3-45. Bürge, Τ. (1989). Individuation and causation in psychology. Pacific Philosophical Quarterly, 707, pp. 30322. Bürge, T. (1995). Intentional properties and causation. In C. Macdonald & G. Macdonald (eds.) Philosophy of Psychology: Debates about Psychological Explanation. Oxford: Blackwell. Chalmers, D. (1996). The Conscious Mind. Oxford: Oxford University Press. Clark, A. & Chalmers, D. (1998). The extended mind. Analysis, 58, pp. 7-19. Davidson, D. (1970). Mental events. In L. Foster and J. Swanson (eds.) Experience and Theory. Amherst. MA: University of Massachusetts Press Dennett, D. (1971). Intentional Systems. Journal of Philosophy, pp. 87-106 Dennett, D. (1978). Brainstorms. Cambridge, MA: Bradford Books/MIT Press. Dennett, D. (1987). The Intentional Stance. Cambridge, MA: M I T Press. Devitt, M. (1990). Meanings jusc ain't in the head. In G. Boolos (ed.) Meaning and Method: Essays in Honor of Hilary Putnam. Cambridge: Cambridge University Press. Dewey, J. (1925). Experience and Nature. Chicago: Open Court. Dretske, F. (1981). Knowledge and the Flow of Information. Cambridge, MA: M I T Press. Dretske, F. (1992). What isn't wrong with folk psychology. Metaphilosophy, 23, pp. 1-13. Egan, F. (1991). Must psychology be individualistic? Philosophical Review 100: 179-203. Egan, F. (1992). Individualism, computation, and perceptual content. Mind, 101, pp. 443-59. Egan, F. (1996). Intentionality and the theory of vision. In K. Aikins (ed.) Perception. Oxford: Oxford University Press. Francescotti, R. M. (1991). Externalism and the Marr theory of vision. British Journal for the Philosophy of Science, 42, pp. 227-38. Fodor. J. (1980). Methodological solipsism as a research strategy in cognitive psychology. Behavioral and Brain Sciences, 3, pp. 63-109. Fodor, J. (1981). Representations. Cambridge, MA: M I T Press. Fodor, J. A. (1991). A modal argument for narrow content. Journal of Philosophy, 88, pp. 5-26 Gibson, J. (1966). The Senses Considered as Perceptual Systems. Boston: Houghton Mifflin. Gibson, J. (1979). The Ecological Approach to Visual Perception. Boston: Houghton Mifflin. Haugeland, J. (1985). AI The Very Idea. Cambridge, MA: M I T Press. Heil, J. & Meie, A. (1991). Mental causes. American Philosophical Quarterly, 28, pp. 61-71. Jackson, F. (1982). 'Epiphenomenal qualia', Philosophical Quarterly, 32, pp. 127-36 Jacob, P. (1992). Externalism and mental causation. Proceedings of the Aristotelian Society, 66, pp. 203-19. Jacob, P. (1993). Externalism and the explanatory relevance of broad content. Mind and Language, 8, p. 131. Jackson, F. (1982). Epiphenomenal qualia. Philosophical Quarterly, 32, pp. 127-36. Kim, J. (1989). The myth of nonreductive physicalism. Proceedings and Addresses of the American Philosophical Association, 63, pp. 31-47. Reprinted in Kim, J. 1993. Kim, J. (1990). Explanatory exclusion, and the problem of mental causation. In E. Villanueva (ed.) Information, Semantics and Epistemology, Oxford: Basil Blackwell Kim, J. (1992). Downward causation and emergence. In A. Beckermann, Η. Flohr and J. Kim (eds.) Emergence or Reduction? Essays on the Prospects for Nonreductive Physicalism, Berlin: DeGruyter. Kim, J. (1993). Supervenience and Mind. Cambridge: Cambridge University Press. Kim, J. (1999). Mind in a Physical World. Cambridge, MA: M I T Press. Kitcher, P. S. (1984). Narrow taxonomy and wide functionalism. Philosophy of Science, 52, pp. 78-97. Kitcher, P. S. (1988). Marr's computational theory of vision. Philosophy of Science, 55, pp. 1-24. Lindsay, Ρ and Norman, D. (1972). Human Information Processing. New York: Academic Press. Lycan, W. (1987). Consciousness. Cambridge, MA: M I T Press. Marr, D. (1982). Vision. San Fransicso: W. H. Freeman McClamrock, R. (1995). Existential Cognition: Computational Minds in the World. Chicago: University of Chicago Press. McGinn, C. (1991). Conceptual causation. Mind, 100, pp. 525-46. Millikan, R. (1984). Language, Mind and Other Biological Categories. Cambridge, MA: M I T Press. Minsky, M. (1986). The Society of Mind. New York: Simon and Schuster.

284

In Defence of Content Externalism

Oppenheim, P. and Putnam, H. (1958). Unity of Science as a working hypothesis. In H. Feigl, M. Scriven, and G. Maxwell Minnesota Studies in the Philosophy of Science Volume II. Minneapolis: University of Minnesota Press O'Regan, Κ and Noe, A. (2001). A sensorimotor account of vision and visual consciousness. Behavioral and Brain Sciences, 24. Owens, J. (1983). Functionalism and the propositional attitudes. Nous, 17, pp. 529-49. Peacocke, C. (1993). Externalist explanation. Proceedings of the Aristotelian Society, 67, pp. 203-30. Peacocke, C. (1995). Content, computation, and externalism. In E. Villanueva (ed.) Content. Atascadero, CA: Ridgeview Publishing. Pettit, Ρ (1986). Broad-minded explanation and psychology. In P. Pettit & J. McDowell (eds.) Subject, Thought and Context. Oxford: Oxford University Press. Perry, J. (1979). The problem of the essential indexical. Nous, 13, pp. 3-21. Putnam, H. (1975a). The meaning of'meaning'. Minnesota Studies in the Philosophy of Science, 7, pp. 131193. Reprinted in H. Putnam, Mind, Language, and Reality. Cambridge: Cambridge University Press, 1975. Putnam, H. (1975b). Philosophy and our mental life. In H. Putnam Mind, Language and Reality, Philosophical Papers Volume 2. London: Cambridge University Press. Putnam, H. (1978). Meaning and the Moral Sciences. London: Routledge and Kegan Paul. Putnam, H. (1987). Meaning, other people, and the world. In Representation and Reality. Cambridge, MA: MIT Press. Rowlands, M. 1995. Against methodological solipsism: The ecological approach. Philosophical Psychology, 8, pp. 5-24. Quine, W. (1960). Word and Object. Cambridge MA: MIT Press. Ryle, G. (1949). The Concept of Mind. New York: Barnes and Noble. Searle, J. (1980). Minds, brains and programs. Behavioral and Brains Sciences, 3, pp. 417-24. Searle, J. R. (1983). Intentionality. Cambridge: Cambridge University Press. Shapiro, L. A. (1993). Content, kinds, and individualism in Marr's theory of vision. Silberstein, Μ. and McGeever, J. (1999). The search for ontological emergence. The Philosophical Quarterly, 49, pp. 182-200. Sober, E. (1985). Panglossian functionalism and the philosophy of mind. Synthese, 64. Sosa, E. (1991). Between internalism and externalism. In E. Villanueva (ed.) Consciousness. Atascadero, CA: Ridgeview Publishing. Stalnaker, R. (1993). Twin earth revisited. Proceedings of the Aristotelian Society, 63, pp. 297-311. Stich, S. P. (1978). Autonomous psychology and the belief/desire thesis. Monist, 61, pp. 573-91. Van Gulick, R. (1980). Functionalism, information and content. Nature and System, 1, pp. 139-62. Reprinted in W. Lycan (ed.) Mind and Cognition. Oxford: Blackwell, pp. 107-129, 1990. Van Gulick, R. (1989). Metaphysical arguments for internalism and why they don't work. In S. Silvers (ed.) ReRepresentation. Dordrecht: Kluwer. Van Gulick, R. (1992). Nonreductive materialism and intertheoretical constraint. In A. Beckermann, Η. Flohr and J. Kim Emergence or Reduction? Essays on the Prospects for Nonreductive Physicalism. Berlin: DeGruyter. Van Gulick, R. (1993a). Understanding the phenomenal mind: Are we all just armadillos? In M. Davies and G Humphreys (eds.) Consciousness. Oxford: Blackwell. Van Gulick, R. (1993b). Who's in charge here and who's doing all the work? In J. Heil and A. Meie (eds.) Mental Causation. Oxford: Clarendon Press. Van Gulick, Robert (2001). Inward and upward: reflection, introspection and self-awareness. Philosophical Topics, 28-2, pp. 275-305. Van Gulick, R. (2004). Higher-Order Global States (HOGS): An Alternative Higher-Order Model of Consciousness. In R. Gennaro Higher-Order Theories of Consciousness. Amsterdam and New York: John Benjamins, 2004. Wilson, R. A, (1992). Individualism, causal powers, and explanation. Philosophical Studies, 68, pp. 10339. Wilson, R. A. (1994). Causal depth, theoretical appropriateness, and individualism in psychology. Philosophy of Science, 61, pp. 55-75. Wilson, R. A. (1995). Cartesian Psychology and Physical Minds: Individualism and the Sciences of the Mind. Cambridge: Cambridge University Press. Wittgenstein, L. (1953). Philosophical Investigations. New York: Macmillan.

IV Critiques of Content Externalism

Natural Intentionality* JOHN HEIL

1

Preamble

Human beings, and doubtless other creatures as well, have a capacity to represent their surroundings. What is the basis of this capacity? When you entertain the thought of a beetroot, what is it about you in virtue of which your thought concerns the beetroot—and not, for instance, a tomato or a visually indistinguishable wax beetroot? These are contentious matters. It would be foolhardy to attempt a definitive account in the space of a few pages. Instead, I shall merely sketch a view designed to make sense of intentionality in light of a plausible ontological scheme. Many philosophers would regard this as putting the cart before the horse. Surely, they would argue, it is a mistake to hold intentionality hostage to ontology. Ontology is a sorely vexed domain. If accounts of intentionality must await agreement among the ontologists, we risk having to wait forever. On the contrary. The prospects of a naturalistic grounding for intentionality can only be appreciated if we have some sense of what the natural world might afford. I shall freely appeal to a conception of properties (defended elsewhere) that holds a key to the basis of our understanding of intentionality (and, although this is not a topic I shall pursue here, consciousness). (See Martin 1997; Martin and Heil 1999; Heil 1998, chap. 6; Heil 2003.)

2 Internalism and Externalism Nowadays philosophers divide into two camps. One camp, the internalists, epitomize traditional approaches to the character of thought: thoughts owe their significance to intrinsic features of thinkers. This is the kind of view Descartes, for instance, presupposes in discussing the evil demon. The demon has the power to alter (or even eliminate) the world around us without affecting the contents of our thoughts. This is the source of Descartes's central epistemological conundrum: what assurance can we have that our thoughts "match" a mind independent "external" reality? A second camp, the externalists (or "anti-individualists") contend that thoughts owe their character to contextual factors: what your thoughts concern depends on causal

*

The central thesis defended here stems from the work of C. B. Martin; see Martin and Pfeiffer ( 1 9 8 6 ) ; and Martin and Heil (1998).

288

Critiques of Content Externalism

relations you bear to your surroundings, perhaps, or on relations in which you stand to your linguistic community. Your twin in the demon world might intrinsically resemble you in every detail, in particular both you and your twin could have qualitatively indiscernible mental lives, yet the contents of your thoughts would differ wildly from those of your twin. The idea can be illustrated by means of a simplified example. Suppose (as many philosophers really do suppose) your thoughts of trees concerned trees because they were caused by trees. In that case, your twin in the demon world might have thoughts qualitatively indistinguishable from your tree thoughts, yet your twin's thoughts would not concern trees. How so? We are assuming that the contents of an agent's thoughts depend on what caused them. In the demon world, your twin's "tree" thoughts would concern whatever caused them. Whatever that is, it would not be a tree. (Your twin's "tree" thoughts are brought about by a particular incantation on the part of the demon.) One consequence of a view of this kind is that occupants of the demon world are not, pace Descartes, deceived! When an agent in the demon world entertains a thought he might express by "uttering" the sentence "That's a tree," the agent is not entertaining a false thought about trees, but a true thought about something else—a particular demon incantation perhaps.1 Externalist conceptions of mind have been inspired by "Twin Earth" thought experiments. Water is H 2 0 . Imagine a remote planet exactly like Earth in every obvious respect but one: the clear, colorless liquid that fills lakes rivers, oceans and ice trays on Twin Earth is not H 2 0 , but XYZ, a liquid with a superficial resemblance to water, but with an utterly different chemical composition. Inhabitants ofTwin Earth (at least those who speak a language indistinguishable from English) call this liquid "water." Water is H 2 0 , however, so when inhabitants ofTwin Earth speak of "water" or entertain thoughts they would express with utterances in which (the word) "water" figures, they would not be speaking or thinking of water but of (as we might call it) twin water. The moral often drawn from imagined cases of this kind is that the significance of what we think and say depends on contextual factors, in particular it depends on causal relations we bear to our surroundings.

3 The Dart-Tossing Model What exactly do Twin Earth thought experiments show? Do they, for instance, establish that internalism is false? That depends on what internalism amounts to. Suppose the intentional character of states of mind were tied to their dispositionality: what makes

1

Many readers will recognize this as Putnam's Brain in a Vat argument (1981, chap. 1). For an account of that argument see Heil (1987). A more sustained attempt to motivate externalism can be found in Heil (1992), chap. 2, which draws on Baker (1987); Bürge (1979, 1986); Davidson (1987); Dretske (1988); Millikan (1984, 1989). T h e roots of externalist accounts of mental content lie in Wittgenstein (1953), but Hilary Putnam's " T h e Meaning o f ' M e a n i n g " ' (1975a) is the inspiration for much subsequent work on the topic.

Natural Intentionality

289

a thought about a tree a thought about a tree is the difference it makes to the dispositionalities of the thinker. What kind of difference? The thought might be an internal manifestation of a disposition to interact with trees in characteristic ways, to describe trees as trees, to utter sentences in which "tree" figures, to assent to queries, "Is this a tree?" and so on. 2 The thought's dispositionality takes advantage of the built-in projective character of dispositions.3 A salt crystal is soluble in water. The salt possess a disposition for dissolving in water (and water possesses the reciprocal disposition for dissolving salt). Intelligent agents are, whatever else they are, complex dispositional systems; thinkers are agents with appropriately focused dispositional makeups. Dispositionality underlies the projective character of thought. What of the agent's environment? Return to Twin Earth. An inhabitant of Twin Earth who entertains thoughts he would express by uttering sentences containing "water" is entertaining thoughts about XYZ, not thoughts about water. In this the externalists are right. But what is it in virtue of which inhabitants of Twin Earth are thus characterizable? On a dispositional account of intentionality, the projective character of thought—its ofness, or forness, or aboutness—stems from its dispositional nature. What the thought concerns, however, can depend on what context in something like the way in which what "here" or "now" designates depends on the location of the speaker or the time of utterance. Wittgenstein was right: Thoughts do reach out to the world; what they find there depends on the world. 4 On a view of this kind, the externalist is right in supposing that what a thought designates often depends on what there is to be designated. The designation need not turn on incoming causal connections, however. Thoughts entertained by inhabitants of Twin Earth concern XYZ, twin water, not water, but not because the thoughts are caused by XYZ—although they could very well be. The thoughts "project to" twin water, not water, because twin water, not water, is on the scene. If you wanted a model for this, think of a dart thrower. A dart thrower is responsible for the direction taken by a dart. What happens to the dart once it is released, however, what the dart hits, depends on what is "out there" to be hit. Does this imply that the "water" thoughts of an inhabitant of Twin Earth instantaneously transported to Earth would suddenly become thoughts of water? Not obviously. Insofar as the Twin Earth native's "water" thoughts are bound up with endless other thoughts—including the thought that this watery stuff is no different from the watery stuff encountered yesterday—we need not suppose his "water" thoughts become water thoughts merely by virtue of a shift in locale. Compare a case in which a person is transported to another room while asleep and who, when questioned on awakening, thinks: "I am still here."

2 3 4

This is a caricature. An agent's dispositional makeup is interconnected, focused, and undoubtedly complex. George Molnar called this "physical intentionality." Cf. Wittgenstein (1922, § 2.1511, 2.1515).

290

Critiques of Content Externalism

4

Swampman

In a much-discussed thought experiment, Donald Davidson imagines that, while wandering through a swamp, he is struck by a bolt of lightning and vaporized (Davidson 1987). Simultaneously, another bolt strikes a nearby tree stump rearranging the particles that make it up to produce, wholly by chance, a moleculefor-molecule duplicate of Davidson: Swampman. How are we to describe Swampman's states of mind? It seems natural to say that Swampman has Davidson's thoughts, beliefs and preferences. Swampman differs from Davidson, not in what he wants and believes, but in the truth value of many of his beliefs. Swampman falsely believes he was a student of Quine's, that he has published papers in Dialectica, and that he once visited Terra del Fuego. I think this natural interpretation is the right one, but philosophers of an externalist persuasion disagree. Swampman, they contend, lacks appropriate causal relations to objects and events in the world. 5 Until these are established, Swampman's thoughts lack content. They resemble marks in the sand traced by a foraging ant that happen to spell out the English word, "HELLO" (Putnam 1981, chap. 1). The marks are intrinsically meaningless. They might come to acquire significance were they embedded in an appropriate context. (The marks might be photographed and used as a greeting card, for instance, or on someone's home page.) We can agree that marks in the sand (or, for that matter, images, utterances, or signs generally) lack intrinsic significance. Such items take on significance when they acquire an expressive use. An ant does not use marks it traces in the sand to mean anything. Are Swampman's thoughts similarly intentionally inert? Swampman is dispositionally indistinguishable from Davidson. This, I suggest, is enough to endow Swampman's thoughts with significance. Their significance lies, not in the thoughts' intrinsic qualitative nature, but in Swampman's intrinsic dispositional makeup. I like to think of Swampman as a counter-example to externalism: if, on externalist grounds, we would be obliged to deny that Swampman has endless thoughts, externalism is mistaken. Swampman's thoughts are anchored in his dispositional condition. In virtue of this condition, Swampman is disposed to interact with the world in particular ways. What ways? The very ways Davidson himself would have interacted with the world. Like Davidson, Swampman would call trees trees, water water, spades spades.

5 Causally Loaded States of Mind You may find the view being bruited here incredible. Before dismissing it outright, however, you should be certain you are doing so for the right reasons. A view of the kind under consideration does not deny that certain states of mind are "causally loaded" in

5

Externalism comes in different flavors. Some externalists focus on causal-historical connections, others emphasize social matters (especially thinkers' linguistic communities), still others focus on biological function. I use a simple causal theory for purposes of illustration, but the points I make here apply, mutatis mutandis, to other strains of externalism.

Natural Intentionality

291

the sense that their character depends in part on causal relations involving the agent. Memory and perception are like this. It is apparently built into our concept of what it is for an agent to remember some object or event that the agent stand in an appropriate causal relation to the remembered object or event. Suppose you have a vivid image of a childhood experience. The image counts as a memory of the experience only if there is a causal link of the right sort between the experience and the subsequent image. Similarly, your perceiving this tree depends on your standing in the right kind of causal relation to the tree. 6 In this way we distinguish memories and perceptions from mere imagery or hallucination. None of this affects the idea that intentionality works from the "inside out": states of mind owe their projective character to intrinsic features of agents. Consider referring. Does your successfully entertaining thoughts of a particular object require a causal chain connecting you to that object? This is not obvious. Imagine thinking a thought you would express as follows: "There is an object at exactly two million kilometers from me in the direction of Alpha Centauri." If there happens to be an object so located, you have succeeded in referring to it!

6 A Worry from Kripke7 In an intriguing study of Wittgensteins "private language argument," Saul Kripke suggests a reason for rejecting dispositional accounts of intentionality (Kripke 1982). The argument is straightforward, but requires stage setting. Suppose that speaking a language is a matter of following certain rules. The rules govern utterances in two senses. First, a rule underlies agents' actions insofar as these actions are based on the rule. 8 Second, rules serve a normative function: they ground distinctions between correct and incorrect actions. A chess move is incorrect only so long as it violates a rule of chess and the agent who makes the move intends to be playing chess (and so undertakes to be governed by the rules of chess). Kripke's famous example involves addition. Mastering arithmetic involves mastering a rule for summing numbers. The rule guides arithmetical judgments and provides a norm against which such judgments are taken to be correct or incorrect. We judge that 2 + 3 = 5 and 68 + 57 = 125 are correct, that is, they accord with this rule, and that 2 + 3 = 4 and 68 + 57 = 5 fail to accord with the rule, and so are incorrect. Similarly, our use of words is guided by

6

Suppose you see what appears to be a ghost gum ten meters in front of you. Suppose, further, that there is a ghost gum ten meters in front of you but that your visual experience is caused by an image of an identical tree off to the right reflected in a camouflaged mirror (if the mirror were removed you would have an indistinguishable visual experience, but it would be caused by the tree in front of you, not the reflected tree). With the mirror in place, the tree you see is the reflected tree, not the tree in front of you. This is apparently due to the fact that the cause of your tree experience is the reflected tree. See Martin (1959, chap. 6); Grice (1961).

7

The argument discussed in this section and the next are examined in rather more detain in Martin and Heil (1998). Talk of a rule's underlying an action is shorthand for an agent's being motivated by his acceptance or inculcation of the rule.

8

292

Critiques o f C o n t e n t Externalism

rules, and to the extent that words and concepts words express figure in our thoughts, thoughts themselves are rule governed. The question pursued by Kripke's Wittgenstein concerns what it is to be guided by a rule; what makes it the case that you are following a particular rule? It will not do to suppose that, if you set out to follow the addition rule, your mathematical judgments must always accord with that rule. Error is always possible. Suppose, on a particular occasion, you judge that 68 + 57 = 5 (in the course of balancing your checkbook, say). You have judged incorrectly; your judgment fails to conform to addition rule, a rule you intended to following judging as you did. Any finite device, including a mechanical calculator, could err in this way.9 Note further that the rule you are following ranges over sequences you will never have occasion to sum. Indeed, the rule pertains to an infinite number of sequences, thereby outrunning the capacities of any finite intelligence. Now imagine an agent, Dwayne, who grasps, not the rule for addition, but a different rule. This rule yields results that coincide with the addition rule up to a particular point, then deviate. (Following Kripke, we can call this rule the quaddition rule.) According to the rule for quaddition, 2 + 3 = 5; 2 quadded to 3 yields 5. Suppose Dwayne subsequently judges that 68 + 57 = 5? As it happens, 68 quadded to 57 does yield 5: this is a perfectly correct application of the quaddition rule! The idea is that addition and quaddition coincide up to a particular point, then diverge. Call this point the divergence point. Now, imagine that Wayne has mastered the addition rule, while Dwayne embraces the rule for quaddition. Imagine, as well, that neither of these agents has ever considered (or will ever consider) numbers lying beyond the divergence point. Their judgments largely coincide. On rare occasions when their judgments fail to coincide, they agree that one of them must be wrong, and make corrections accordingly. What, asks Kripke's Wittgenstein, makes it the case that Wayne and Dwayne are following different rules? What is it about Wayne and Dwayne in virtue of which it is true that Wayne has grasped the addition rule, and Dwayne has grasped the rule for quaddition? Kripke argues that Wittgenstein's own response to this query is skeptical and deflationary: there is no fact of the matter here, or at least no fact about the constitution of Wayne and Dwayne, respectively, in virtue of which it is true that they are following the rules we have supposed they are following. What rules an agent grasps is largely a matter of the attitudes prevailing among members of the agent's community. The conclusion is supported by the contention that it is hard to see how any "fact about" Wayne or Dwayne could serve as a basis for a rule with infinite application. There could be no state of Wayne or Dwayne that constituted their grasp of the rule for addition or the rule for quaddition, for instance.

9

In the case o f a simple mechanical system, talk o f intention is misplaced. T h e point however, is that a device we devise—intentionally—to operate in accord with a particular rule can fail to do so owing to what we would describe as a mechanical breakdown. In one sense, a malfunctioning calculator behaves exactly as it ought to behave: it does whatever laws o f nature require o f it. T h e sense in which a malfunctioning calculator malfunctions is tied to intentions o f its maker. T h e challenge is to see how intentionality could have a wholly dispositional basis.

Natural Intentionality

293

What of the idea that an agent's acceptance of a rule (the addition rule, for instance) is grounded in the agent's dispositional makeup? This might at first seem promising. Trouble arises, however, when we consider cases of error, cases in which an agent "violates" a rule he accepts. Suppose, for instance, Wayne and Dwayne both enter "125" after writing 68 + 57. Dwayne, but not Wayne, is mistaken (68 quadded to 57 is 5). Both Wayne and Dwayne are, on this occasion at least, disposed to answer "125," however. This will be so whenever anyone answers any question at all in any way whatever. How then could Dwayne's dispositional makeup be such that "125" fails to express the rule he intends to follow, if Wayne's (apparently identical) dispositional makeup makes it the case that he is correctly following the addition rule?

7 Individuating Dispositions A sensible answer to this question requires distinguishing an agent's or a system's overall dispositional makeup from its dispositional components. Consider a salt crystal. The crystal is disposed to dissolve in water. Now consider a salt crystal encased in Lucite. Although the encased crystal does not lose the disposition to dissolve, the crystal-encased-in-Lucite is not so disposed. The addition of a Lucite coating blocks the manifestation of the crystal's disposition to dissolve in water, but does not eliminate it. Compare this with a case in which the salt is chemically modified so as to lose its disposition to dissolve in water. In general, we shall want to distinguish (A) (Β)

object ο possesses disposition D at t, and, at ί,, ο loses D; ο possesses D at t, and at tx acquires a disposition, D', that blocks one kind of manifestation of D.

Suppose now that an agent who has mastered a rule—the rule for addition, say—has acquired a particular disposition, D . D manifests itself in various ways depending on its reciprocal disposition partners (Martin 1997; Martin and Heil 1999). Thus, if D a is the disposition associated with the addition rule, D a might manifest itself as an utterance of "125" in response to the question "What is the sum of 68 and 57? "10 This manifestation might be blocked or modified, however, by the presence of blocking or distorting disposition partners. Given the query "What is the sum of 68 and 57?" the system comprising Da together with some other disposition or dispositions, D\ might yield an utterance of "five." This might be due, not to the absence of Da, but to the presence of D a together with D/(see Bird 1998, 2000).

10 Like any disposition, Da would manifest itself differently in concert with different reciprocal disposition partners. D might yield an utterance, an inscription, a thought, or nothing at all, depending on these reciprocal partners.

294

Critiques of Content Externalism

Although Wayne and Dwayne might be similarly disposed to answer "125'' to "What is the sum of 68 and 57?" it need not follow that they are dispositionally indistinguishable. Wayne, but not Dwayne, possesses D . Dwayne, possesses some other disposition, D^ that grounds his mastery of the quaddition rule. This is perfectly consistent with the possibility that, on occasion, the "correct" manifestation of Da or D could be blocked or distorted by the presence of assorted other factors. We need not embrace "rule skepticism" of the sort associated with Kripke's Wittgenstein.

8 Infinite Use of Finite Means A loose end remains. In mastering the rule for addition, Wayne apparently acquires a disposition for an infinity of responses. Wayne, however, is finite, as is his dispositional makeup. How could a finite agent house a disposition for infinite manifestations? In one obvious sense, no finite thing could yield an infinite number of outputs. Nevertheless a disposition could be for an infinite number of different kinds of response, even if the disposition itself would not survive the production of more than a handful of these. Imagine a chip in a simple calculator. The chip can sum endless numbers. If the calculator's memory is limited, it—the calculator—will reach a point where numbers it attempts to sum are too large to be contained in its memory. The chip has the capacity to sum these numbers, but this capacity requires additional memory if it is to be exercised. We can fix the problem by adding more memory, but eventually the problem will reoccur; at some point the system will simply disintegrate. Here is a simpler example. Imagine a chip capable of summing every number upto 1,000. Now, imagine an identically engineered chip made of a material that would last only long enough for the chip to perform a handful of calculations. Both chips are disposed to sum numbers up to 1,000; but the second chip could never complete the task owing to its frailty. In this case, it would be excessive to deny that the chips were mathematically equivalent, although possessing importantly different physical limitations on the manifestation of their respective competencies. In general, then, it would seem to be impossible to construct a physical mechanism capable of performing an infinite number of distinct operations. The mechanism would eventually deteriorate, run out of space, or succumb to cosmic catastrophe. Such contingencies, however, need not affect the nature of a component engineered to sum numbers. The operation of the component depends on the presence of mechanisms that, like the component itself, lack infinite staying power. This, however, need be no bar to the possibility that the component itself has an infinite capacity, a capacity perfectly capable of grounding the mastery of an infinite rule.

Natural Intentionality

9 Intentionality

and

295

Dispositionality

Dispositionality provides a natural basis for intentionality. Projectability—of-ness, forness, about-ness—is built into dispositions. Dispositions are of or for particular manifestations with particular kinds of reciprocal disposition partner. A disposition can be for a non-existent manifestation: a salt crystal drifting in a universe that lacked water would still be water-soluble. Unicorns do not exist, but a soap bubble would burst were it punctured by a unicorn horn. Dispositionality, then, includes the "mark of intentionality": a capacity to project to the non-existent. This might lead to new worries.11 Are we now saddled with Meinongian nonexistent entities, possibilia, and the like (non-actual, merely possible unicorns, for instance)? I doubt it. We characterize dispositions by reference to their actual or possible manifestations. Dispositions are not relations to actual or possible manifestations, however. Objects possess dispositions by virtue of possessing particular intrinsic properties. The nature of these properties insures that they will yield manifestations of particular sorts with reciprocal disposition partners of particular sorts. In this regard, the dispositions "point toward" non-actual, merely possible manifestations. The "pointing" is grounded in the disposition, however, not in a relation the disposition bears to anything else. Consider a simple example inspired by Locke. Locke defends a broadly dispositional conception of colors: a color is a power possessed by an object to cause experiences of certain sorts in conscious observers. Imagine a universe containing a single ripe tomato. Is the tomato red? I believe Locke's view is that the tomato is red. Locke characterizes colors by reference to certain of their manifestations, but the colors exist even when the manifestations do not, even when the manifestations are, owing to the absence of suitable reciprocal disposition partners, impossible.

10

Conclusion

Many philosophers today accept without question arguments purporting to establish externalist accounts of intentionality. By and large, these arguments are founded on appeals to intuitions about "Twin Earth" cases, those in which intrinsically similar agents entertain thoughts with dissimilar contents. We can agree—up to a point—with the intuitions without thereby embracing externalism, however. In any case, we have available a resource ideally suited to account for the kind of projection associated with intentionality: dispositionality. Dispositions are ofor for particular kinds of manifestation with particular kinds of disposition partner. Dispositions preserve the mark of intentionality in being of or for particular kinds of manifestation with particular kinds of non-existent—possible, but nonactual—objects. This is not mysterious or spooky; it is a feature of dispositions possessed by rocks, or blades of grass, or quarks.

11

I owe this observation to David Armstrong who credits it to J . J . C . Smart.

296

Critiques of Content Externalism

My suggestion is that we make use of the "natural intentionality" afforded by dispositions in making sense of the kinds of intentionality we find in the minds of intelligent agents. In so doing, we return to a long tradition that takes the contents of thoughts to be fixed by intrinsic features of thinkers.

References Baker, Lynne Rudder. (1987) Saving Belief: A Critique of Physicalism. Princeton: Princeton University Press. Bird, Alexander. (1998) "Dispositions and Antidotes." The Philosophical Quarterly 48: 227-34. Bird, Alexander. (2000) "Further Antidotes: A Response to Gundersen." The Philosophical Quarterly 50: 229-33. Bürge, Tyler. (1979) "Individualism and the Mental." Midwest Studies in Philosophy 4: 73-121. Bürge, Τ. (1986) Individualism and Psychology. Philosophical Review, 45: 3-45. Davidson, D. (1987) Knowing One's Own Mind. Proceedings and Address of the American Philosophical Association, 60: 441-458. Dretske, F. (1988) Explaining Behavior: Reasons in a World of Causes. Cambridge: M.I.T. Press. Grice, Η. Ρ (1961) "The Causal Theory of Perception." Aristotelian Society Proceedings (suppl.) 35: 121-52. Gunderson, Keith, ed. (1975) Language, Mind, and Knowledge (Minnesota Studies in the Philosophy of Science 7). Minneapolis: University of Minnesota Press. Heil, John (1998) Philosophy of Mind: A Contemporary Introduction. London: Routledge. Heil, John. (2003) From and Ontological Point of View. Oxford: Clarendon press. Kripke, Saul. (1982) Wittgenstein on Rules and Private Language: An Elementary Exposition (Cambridge: Harvard University Press). Martin, C. B. (1959) Religious Belief Ithaca: Cornell University Press. Martin, C. B. (1997) "On the Need for Properties: The Road to Pythagoreanism and Back." Synthese 112: 193-231. Martin, C. B., and John Heil (1998) "Rules and Powers." Philosophical Perspectives 12: 283-312. Martin, C. B., and John Heil (1999) "The Ontological Turn." Midwest Studies in Philosophy 23: 34-60. Martin, C. B. and K. Pfeifer. 1986 "Intentionality and the Non-Psychological." Philosophy and Phenomenological Research 46: 531-554. Millikan, R. G. (1984) Language, Thought, and Other Biological Categories. Cambridge: M.I.T. Press. Millikan, R. G. (1989) "Biosemantics." Journal of Philosophy, 86: 281-297. Putnam, Hilary. (1975a) "The Meaning of'Meaning'." in Gunderson 1975: 131-93; reprinted in Putnam 1975b: 215-71. Putnam, Hilary. (1975b) Philosophical Papers, vol. 2. Cambridge: Cambridge University Press. Putnam, H. 1981. Reason, Truth, and History. Cambridge: Cambridge University Press. Wittgenstein, Ludwig. (1922/1961) Tractatus Logico-Philosophicus, trans. D. F. Pears and B. F. McGuinness, London: Routledge and Kegan Paul. Wittgenstein, Ludwig. (1953/1968) Philosophical Investigations, trans. G. Ε. M. Anscombe, Oxford: Basil Blackwell.

Phenomental Intentionality and the Brain in a Vat T E R E N C E HORGAN, J O H N TIENSON AND G E O R G E GRAHAM

In the First Meditation, Descartes envisioned the possibility that he was the victim of an enormously powerful, enormously clever deceiver, that his experiences were radically nonveridical, and that his beliefs were massively false. Versions of this scenario have figured centrally in epistemology ever since, especially in discussions of radical Cartesian skepticism. In recent philosophy, often the scenario is posed in a high-tech guise: as some variant of the "brain-in-a-vat" scenario. Suppose, for instance, that there are some intelligent creatures elsewhere in the cosmos, who are very different from humans in physical composition and physical appearance, and whose surrounding environment is quite different from our own. They are in no causal contact with earth, and know nothing of it or its inhabitants. In the course of their scientific investigations, they synthesize a structure out of organic molecules that happens to be an exact physical duplicate of your own brain; they hook it to a computer in such a way that its ongoing brain-activity happens to exactly match your own, throughout its existence. Also, there is substantial counterfactual exact physical similarity between this artificial brain and yours; various counterfactuals about what would happen to and within your brain if certain physical events were to occur to it or within it—for instance, counterfactuals about what would happen physically if the brain were to receive certain potential physical inputs to its sensory-input neurons-—are also true of the synthesized brain. Hereafter when we speak of the Brain in a Vat (for short, the BIV), usually we will mean this particular synthesized brain in this particular setup—a specific version of the generic brain-in-vat scenario. When one first contemplates the generic scenario, certain pre-theoretic intuitive judgments about it tend to arise strongly and spontaneously, given the knowledge that physical processes underlying human mentality occur in the human brain. One intuitively judges that the BIV's mental life exactly matches ones own, in a very strong way; this means, inter alia, that the BIV has numerous beließ, both perceptual and nonperceptual, that exactly match one's own beliefs. Also, one intuitively judges that many of these matching beliefs—-including perceptual beliefs in particular—are veridical in one's own case but nonveridical in the BIV's case. We will refer to these strong, stubborn intuitions about the BIV scenario as Cartesian intuitions. As we said, the generic idea of the B I V is still commonly employed in epistemology, especially in discussions of Cartesian skepticism. In epistemological contexts, Cartesian intuitions about the BIV are routinely taken for granted, and are routinely presumed to be correct. A deep schizophrenia has arisen in contemporary philosophy, however, because meanwhile these same Cartesian intuitions have come into question in philosophy of mind.

298

Critiques of Content Externalism

A number of currently influential theories of mental intentionality are versions of what we will call strong externalism·. roughly and generically, the view that (1) all intentionality is grounded in certain past or present causal connections between states of the cognitive system and states of the external world, and (2) there can be no mental intentionality without some suitable kind of actual connection between what is going on "in the head" and the wider environment. Strong externalist theories include (i) causal theories of content that find the necessary connection in the causal antecedents of the state, (ii) covariational theories that find the connection in certain kinds of systematic correlations between occurrences of an internal state and occurrences of an external state of affairs, (iii) teleosemantic theories that look to environmentally situated proper functions that certain internal states possess in virtue of evolutionary design, and (iv) learning-based theories that invoke internal adaptational changes in the creature's own history. (See Stich and Warfield 1994 for a representative sample of such theories.) Current versions of strong externalism evidently are in direct conflict with Cartesian intuitions about the BIV scenario. Advocates of various strong externalist views typically respond to this situation by "biting the bullet" and saying what their theory requires them to say about the BIV. Sometimes it is claimed, especially by advocates of causal and covariational approaches, that the BIV s beliefs and other intentional mental states refer to various internal states or processes in the computer to which the brain has input and output connections—and to various properties of, and relations among, those states and processes in the computer (cf. Putnam, 1981). This stance denies the Cartesian intuitions in two important ways. First, it thoroughly rejects the putative match in intentionality that those intuitions say exists between your own mental life and the mental life of your BIV duplicate. Second, it asserts that the BIV's beliefs are no less true than your own beliefs, rather than being systematically nonveridical in the way the Cartesian intuitions say they are. Sometimes it is claimed, especially by advocates of teleosemantic and learning-based versions of strong externalism, that the kind of BIV described above—viz., one that has not come into existence as a result of evolution and natural selection—completely lacks mental intentionality. This stance obviously denies Cartesian intuitions altogether: it asserts that your BIV has no beliefs at all—let alone beliefs that match your own and are systematically nonveridical. But Cartesian intuitions stubbornly tend to persist, despite the recent philosophical popularity of strong externalism. Our own view is that this is because Cartesian intuitions are correct, whereas strong externalism is deeply mistaken. We do not deny, however, that external factors play some important roles with respect to the some aspects of mental reference and mental intentionality: important lessons about such matters have emerged from the work of Kripke, Putnam, Bürge, and other philosophers. 1 An

1

McGinn (1989) distinguishes two kinds of externalism about mental content: strong and weak. He argues against strong externalism, while embracing a teleosemantic version of weak externalism. McGinn (1991) argues, as we do below, that intentional content is internal to phenomenology; he proposes wedding this position to weak teleosemantic externalism. Together, these two texts suggest a position something like the one we will set forth—with the added ingredient of a teleosemantic approach to the externalistic aspects of mental reference.

Phenomenal Intentionality and the Brain in a Vat

299

adequate overall philosophical position about mental intentionality and mental reference, we claim, should accommodate these lessons while still thoroughly accommodating—and vindicating—Cartesian intuitions about the BIV. In this paper we outline a position that meets these constraints, and we argue for the position largely by appeal to the way it both vindicates Cartesian intuitions and explains why they are correct. The same general position is also described and defended in Horgan and Tienson (2002), and the present paper is complementary to that one. Here we develop the position beyond what was said there, by further elaborating the treatment of narrow mental intentionality, wide mental intentionality, and mental reference. We also argue for the position in a way that complements the argumentation in Horgan and Tienson (2002); here we place much heavier emphasis on the BIV scenario and its vindication. Other philosophers who have defended positions similar in various ways to ours include McGinn (1989, 1991), Flanagan (1992), Searle (1992), Strawson (1994), Siewert (1998), Loar (forthcoming a, forthcoming b), and Pitt (2004).

1. The Position

Sketched

We begin by sketching our position in broad brushstrokes, leaving various important details to be filled in later in the paper. Central in the position is the role of phenomenology or phenomenal consciousness, by which we mean those aspects of ones mental life such that there is "something it is like" to undergo them. Briefly, the position goes as follows. Phenomenology is narrow, it is not constitutively dependent upon anything "outside the head" (or outside the brain) of the experiencing subject. Indeed, it is not constitutively dependent upon anything outside of phenomenal consciousness itself; in this sense, it is intrinsic. Your phenomenology, being narrow and intrinsic, supervenes nomically upon physical events and processes within your brain. Hence, your phenomenology is shared in common with a BIV physical duplicate of your own brain. Phenomenology is also richly and pervasively intentional·, there is a kind of intentionality that is entirely constituted phenomenologically (we call it phenomenal intentionality), and it pervades our mental lives. Among the different aspects of phenomenal intentionality are the following. First, there is the phenomenology of perceptual experience: the enormously rich and complex what-it's-like of being perceptually presented with a world of apparent objects, apparently instantiating a rich range of properties and relations—including one's own apparent body, apparently interacting with other apparent objects which apparently occupy various apparent spatial relations as apparently perceived from one's own apparent-body centered perceptual point of view. Second, there is the phenomenology of agency: the what-it's-like of apparently voluntarily controlling ones apparent body as it apparently moves around in, and apparently interacts with, apparent objects in its apparent environment. Third, there is conative and cognitive phenomenology: the what-it's-like of consciously (as opposed to unconsciously) undergoing various occurrent propositional attitudes, including conative attitudes like occurrent wishes and cognitive attitudes like occurrent thoughts. There are phenom-

300

Critiques of C o n t e n t Externalism

enologically discernible aspects of conative and cognitive phenomenology, notably (i) the phenomenology of attitude type and (ii) the phenomenology of content. The former is illustrated by the phenomenological difference between, for instance, occurrently hoping that Bush will fail to be re-elected and occurrently wondering whether Bush will fail to be re-elected—where the attitude-content remains the same while the attitude-type varies. The phenomenology of content is illustrated by the phenomenological difference between occurrently thinking that Bush will fail to be re-elected and occurrently thinking that Bush will succeed in being re-elected—where the attitude-type remains the same while the attitude-type varies. Since phenomenal intentionality is entirely constituted phenomenologically, and since phenomenology is narrow, phenomenal intentionality is narrow too. Hence, there is exact match of phenomenal intentionality between yourself and your BIV physical duplicate. This exactly matching, narrow, intentional content involves exactly matching, phenomenally constituted, narrow truth conditions. But whereas the narrow truth conditions of your own beliefs are largely satisfied, those of your BIV physical duplicates matching beliefs largely fail to be satisfied; thus, the BIV's belief system is systematically nonveridical. So Cartesian intuitions about the BIV are correct. O n the other hand, exact match in narrow content between your own intentional mental states and the corresponding states in your BIV physical duplicate does not require or involve exact match in referents (if any) of all the various matching, putatively referring, thought-constituents. For instance, certain of your own occurrent thoughts that you would express linguistically using certain proper names—say, the thought that Bush is not a genius—involve singular thought-constituents whose referents (if any) are determined partly in virtue of certain external relations that obtain between you and those referents. Thus, your occurrent thought that Bush is not a genius involves a singular thought-constituent that purports to refer to a particular specific person (viz., Bush); its actually referring, and its referring to the specific individual to whom it does refer, depends upon there being certain suitable external relations linking you to a unique eligible referent (viz., Bush). A Twin-Earthly physical duplicate of yourself, in a Twin-Earthly duplicate local environment, would refer to a different individual (viz., Twin-Bush) via the corresponding singular thought-constituent of the corresponding occurrent thought. And in the case of your BIV physical duplicate, the matching singular thought-constituent fails to refer at all, because the BIV does not bear suitable externalistic relations to any suitably reference-eligible individual in its own actual environment. (Parallel remarks apply to thought-constituents that purport to refer to natural kinds, such as the thought-constituent that you yourself would express linguistically with the word 'water'.) For mental states involving thought-constituents for which reference depends upon externalistic factors, there are two kinds of intentionality, each involving its own truth conditions. First is the kind of intentionality already mentioned above: phenomenal intentionality, with truth conditions that are phenomenally constituted and narrow. Second is externalistic intentionality, with wide truth conditions that incorporate the actual referents (if any) of the relevant thought-constituents. Your own thought that Bush is not a genius, and the corresponding thoughts of your BIV physical duplicate and your Twin Earth physical duplicate, have matching phenomenal intentionality, with

Phenomenal Intentionality and the Brain in a Vat

301

matching narrow truth conditions. (These truth conditions are satisfied in your case and in the case of your Twin Earth duplicate, but not in the case of your BIV duplicate.) O n the other hand, your own thought that Bush is not a genius and your Twin Earth duplicate's corresponding thought do not have matching externalistic intentionality, because the wide truth conditions of these respective thoughts do not match: the truth value of your own thought depends upon the intelligence level of Bush, whereas the truth value your Twin Earth duplicate's corresponding thought depends upon the intelligence level of an entirely different individual, viz., Twin-Bush. (Each thought's wide truth conditions are indeed satisfied.) As for your BIV duplicate's thought, it lacks externalistic intentionality and wide truth conditions, because its singular thought-constituent purporting to refer to a person called 'Bush' does not actually refer at all. If the picture we have sketched can be satisfactorily elaborated, then the resulting position will meet the two constraints we mentioned in the introduction. O n one hand, the position will vindicate Cartesian intuitions: you and your BIV physical duplicate have exactly matching conscious occurrent mental states with exactly matching narrow intentional content, and for vastly many of the BIV's conscious occurrent beliefs, the narrow truth conditions are not satisfied. O n the other hand, the position also will accommodate externalistic aspects that govern the reference of certain thought-constituents, such as thought-constituents that purport to refer to individuals or to natural kinds. Your own conscious occurrent mental states and the corresponding ones of your BIV duplicate do not have matching externalistic intentionality. O n the contrary, the BIV's mental states lack externalistic intentionality altogether.

2. Methodological Remarks Before we elaborate the position just sketched, let us comment briefly about the structure of our argument for this position, and about how this argument fits into the wider dialectical landscape of philosophical views and arguments concerning mental intentionality. To begin with, we believe that a strong case can be made for phenomenal intentionality—and for its narrowness, its richness, and its pervasiveness in human mental life—without reliance on Cartesian intuitions about the BIV. O n e important line of argumentation involves focusing the reader's introspective attention on certain specific actual or potential experiences, as a way of prompting appreciation that such experiences reveal the presence of one or another specific kind of phenomenal intentionality—in some cases involving forms of non-sensory phenomenology that are more subtle, less starkly vivid, and thus more easily overlooked than is sensory phenomenology. For instance, Galen Strawson (1994) discusses a form of non-sensory, inherently intentional, phenomenology that he calls "understanding experience." He argues that understanding and other related kinds of occurrent mental states and processes are very commonly, if not always, laden with distinctive phenomenal character that is non-sensory and inherently involves the intentional content of what is understood. He points out, for example, the phenomenological difference between hearing speech in a language that one does not understand and hearing speech in a language that one does understand.

302

Critiques of Content Externalism

Imagine two people side by side hearing the same spoken sequence of sounds, with one of them understanding the language and the other one not. At a certain relatively raw sensory level, their auditory experience is phenomenologically the same; the sounds are the same, and in some cases may be experienced in much the same way qua sounds. Yet it is obvious introspectively that there is something phenomenologically very different about what it is like for each of them: one person is having understanding experience with the distinctive phenomenology of understanding the sentence to mean just what it does, and the other is not. Such examples are readily multiplied. (See, for instance, the works cited above at the end of the introductory section.) But even though there is already a strong case to be made for phenomenal intentionality without appeal to the BIV, nonetheless the fact that Cartesian intuitions about the BIV are so common and so persistent is itself dialectically important; it provides additional, reinforcing, evidence. Here is why. The persistence and widespread occurrence of Cartesian intuitions is a datum that calls out for explanation. The initial, albeit defeasible, presumption about these intuitions is that their persistence and their ubiquity reflect their correctness—although it remains possible that in the end, the best explanation will be a "debunking" explanation that treats them as erroneous. So if we provisionally assume that the Cartesian intuitions are correct, and we then seek out a plausible non-debunking explanation for their occurrence, by far the most plausible source for such an explanation is phenomenology. The relevant explanatory hypothesis asserts, about somone with Cartesian intuitions, (i) that one is judging that a BIV that is a physical duplicate of oneself would also be a phenomenal duplicate of oneself, and (ii) that one's Cartesian intuitions reflect one's visceral appreciation that a phenomenal duplicate of oneself would, by virtue of having a mental life phenomenologically exactly like one's own, thereby have a mental life that exactly matches one's own intentionally. (The visceral appreciation mentioned in (ii) need not necessarily rise to the level of explicit belief) This explanatory hypothesis provides the most plausible—indeed, really the only plausible—non-debunking explanation of the Cartesian intuitions. Thus, the persistence and widespread occurrence of these intuitions provides significant evidential support for the explanatory hypothesis itself, via "inference to the only plausible non-debunking explanation." Of course, the ultimate strength of this evidential support depends upon whether or not the position sketched in section 1 can be developed in a coherent and credible way—in some way that, inter alia, smoothly and naturally accounts for the role of externalistic factors vis-ä-vis mental reference and mental intentionality. Addressing this task will be the business of the next section. But before turning to that, we have two final dialectical observations. First, we realize that if there were a good case in favor of strong externalism about mental intentionality, such a case would count against our own position—and would also constitute grounds for believing that the best explanation of lingering Cartesian intuitions about the BIV must be some kind of debunking explanation. But our own view is that strong externalism is seriously lacking in solid epistemic support—its widespread recent popularity notwithstanding. Although the kinds of considerations often cited as grounds for strong externalism do indeed show that external factors are constitutively involved in the fixation of reference for some reference-purporting thought-constituents,

Phenomenal Intentionality and the Brain in a Vat

303

we believe that such considerations fall far short of warranting strong externalism itself. The discussion to follow in section 3 should make clear why this is so. Second, it bears emphasis that even among those philosophers who maintain (as we do) that there is a narrow kind of mental intentionality that is wholly constituted by phenomenology, there are important theoretical differences. To some extent, these differences are reflected in the extent to which the various alternative approaches to phenomenal intentionality are able to accommodate Cartesian intuitions about the BIV. We will return to this matter briefly at the end of the paper.

3. The Position

Elaborated

We will now develop further the position sketched above. In doing so we will use Cartesian intuitions about the BIV as a guide and motivator: we will elaborate the position in ways that seem needed in order to accommodate Cartesian intuitions—and that also seem independently plausible, both phenomenologically and theoretically. What we say here will develop the initial sketch into a somewhat more complete picture, but will still leave many important matters open for yet further exploration. It will be useful to articulate the position by means of a device briefly employed already in section 1, viz., by discussing two different kinds of physically possible physical duplicates of yourself: not only a BIV physical duplicate (of the sort described in the second paragraph of this paper), but also a Twin Earth physical duplicate. 2 We assume, here and below, what was said in section 1; our present purpose is to elaborate that discussion without repeating it. In particular, we assume that a physical duplicate of your own brain, undergoing exactly the same physical inputs and exactly the same internal physical processes, would also be a phenomenal duplicate: it would undergo phenomenology that exactly matches your own.

3.1 Matching Phenomenal Intentionality: Self-Indexical Aspects and Phenomenally Constituted Attribute-Reference In considering the extent of mental match between yourself and your BIV phenomenal duplicate, it is useful to begin by focusing on matching perceptual experience, and on matching beliefs whose intentional contents are the contents of one's current perceptual experience. Phenomenologically, perceptual experience is richly presentational: it presents, to the experiencing subject, a richly textured apparent world of apparent concrete objects apparently instantiating numerous properties and relations. For instance, experi-

2

You do not really have an exact physical duplicate on Twin Earth, however, because on Earth people sometimes have the occurrent thought that water is H20, whereas on Twin Earth they have instead the thought that water is XYZ—differences that would manifest themselves in somewhat different behavior (e.g., differences in linguistic behavior). But we will use the useful term "Twin Earth phenomenal duplicate" for a person who is as much like you as is consistent with such differences.

304

Critiques of Content Externalism

ence presents various apparent objects apparently instantiating various j/w/^-properties, i/z^-properties, relative-position relations, and relative-orientation relations. Among the experientially presented apparent objects is one's own apparent body, with its various apparent component parts—a body that is apparently under one's voluntary control. Some of the relations apparently instantiated by the various apparent objects—for instance, relative-position relations and relative-orientation relations—are experientially presented within a self-oriented reference frame. Thus, apparent objects are experientially presented as lying at various distances from oneself (as well as from one another, as presented from one's own visual point of view), and as apparently oriented various ways vis-a-vis oneself (as well as vis-ä-vis one another, as presented from one's own visual point of view). This point of view is what Husserl called the "zero-point" in this selforiented reference frame. Up and down in this reference frame normally depend heavily upon kinesthetic/tactile aspects of phenomenology, in combination with visual aspects: roughly, down is the direction that one's own apparent body apparently tends to move of its own accord, and the direction of a surface to which one's apparent body apparently tends to stay attached when apparently not moving relative to that surface. Two points about such experientially presented properties and relations are of key importance. First, often their presentation in experience includes a self-indexical aspect, involving the self-oriented nature of the experiencing subject's perceptual reference frame. Because of this indexical element, if two different experiencing subjects have exactly similar phenomenology, the relevant kind of intentional match between them will not be literal identity of intentional content. Rather, two distinct selves are involved, and hence two distinct self-oriented experiential reference-frames. The intentional match between them consists in the fact that the corresponding intentional contents of the two distinct subjects' corresponding perceptual-experiential states are, so to speak, "identical apart from differences in the identity of the experiencer who is the self in the self-oriented perceptual reference frame." This theme surely deserves further exploration and elucidation, but that is a task we will not pursue here. 3 The second key point is this. When experience presents various apparent objects as apparently instantiating properties and relations such as shape-properties and relative-position relations, experience thereby acquaints the experiencing subject with such properties and relations, and this mental acquaintance-relation grounds mental reference to these properties and relations. Such mental reference is wholly constituted phenomenologically.4 It makes no difference, so far as this phenomenally constituted and refer-

3

4

Two important issues here are (1) the way(s) that the self-indexical aspect figures in the intentionality of presentational experience, and (2) the way(s) that this affects content-individuation. Does the selfindexical dimension of presentational experience inevitably involve the explicit presentation of the self in experience, or is it typically something different and more subtle? We are inclined toward the latter. D o the mental states of two distinct experiencers have literally identical intentional contents when these respective states differ only in the identity of the experiencer who is the self in the given state's self-oriented reference frame, or is this difference sufficient to constitute non-identity of content? Again we are inclined toward the latter. This does not mean, of course, that actual instantiation of the relevant properties and relations is wholly constituted phenomenologically. On the contrary, typically these properties and relations are externally constituted, in the sense that their being instantiated (if and when they are) is a matter of how things are

Phenomenal Intentionality and the Brain in a Vat

305

ence-grounding form of acquaintance is concerned, whether or not the relevant experiential presentations are veridical. In the case of your BIV phenomenal duplicate, for instance, the perceptual-experiential presentations are radically illusory: there are no real objects that are really perceived by that experiencing subject and that really instantiate the relevant properties and relations. But no matter: your BIV's duplicate's perceptual experience acquaints the BIV with shape-properties and relative-position properties just as much as your own perceptual experience does, even though this acquaintance occurs via radically nonveridical experiences of merely apparent instantiations of these properties and relations by merely apparent objects. And for the BIV, such experiential acquaintance with the properties and relations grounds mental reference to them—just as it does for you. Experientially presented apparent instantiation of the properties and relations suffices to acquaint the experiencing subject with them, and thus suffices to ground mental reference to them, whether or not the experiencing subject is ever experientially presented with actual instantiations of them. 5 In addition to phenomenally constituted reference to certain properties and relations that is based on direct experiential acquaintance, there are further kinds of phenomenally constituted reference to properties and relations too. The experiencing subject conceives these properties and relations by way of aspects that go beyond what is directly experientially presented—for instance, dispositional aspects and causal-role aspects. But the conceiving of such properties and relations, with such aspects, is itself phenomenally constituted (and hence narrow), and largely rests on the experiencer's capacity for "conceptual bootstrapping" on the basis of phenomenally constituted mental reference to those properties and relations (including causation) that are most directly presented in experience. It is an important philosophical question which kinds of properties and relations are ones to which creatures with human-like phenomenology bear phenomenally constituted, reference-grounding, experiential-acquaintance relations. Although we cannot pursue this large topic at any length here, we take it that the range of such properties and relations is very extensive. It appears to include, inter alia, temporal relations,

causal relations, properties like being a temporally persisting object, being an animal, being an agent, and being a person, numerous artifactual kinds like being a container and being a table, and numerous social relations and properties like being friend o f , being a boss o f , and being a politician.6

It also appears to include numerous language-involving

externally to the experiencing subject. But on our view it is a serious non sequitur to infer, from the fact that a given property or relation is externally constituted (in the sense just explained), to the conclusion that mental reference to that property or relation must involve an externalistic constitutive aspect. 5

Although for expository simplicity we will talk freely of properties and relations and o f reference to them, we mean to remain officially neutral regarding certain metaphysical issues about such entities. We remain neutral, for instance, about the extent (if any) to which there are, over and above precise and determinate properties (e.g., various precise shapes all o f which are somewhat similar to the shape o f Texas), certain determinable properties (e.g., being Texas-shaped) that have a (possibly vague) range o f precise determinant properties falling under them.

6

Objection: "But surely properties like being

a boss of and being

a politician

are not wholly constituted

phenomenologically! Their instantiation depends constitutively u p o n how things are external to the experiencer." Reply: O u r claim is not that these properties themselves are wholly constituted p h e n o m enologically, but rather that mental reference to them is thus constituted; cf. note 4.

306

Critiques of Content Externalism

properties and relations, such as uttering a meaningful statement and speaking a language I understand. Many of these properties and relations involve aspects that are not fully and directly presented in experience (as mentioned two paragraphs ago). For instance, being a container is partly a matter of dispositional capacities to contain something, and paradigmatically (if not inevitably) is partly a matter of something's having its causal origin as an artifact that was deliberately created in order to contain things. Although such properties and relations typically are not presented in perceptual experience as immediately and straightforwardly as are features like position and shape, this certainly does not mean that they figure in phenomenology in a "merely theoretical" or inferential way, as opposed to an "observational" way. On the contrary, the presentational aspects of perceptual experience are rife with potentialities. You experience apparent enduring objects as having back sides, even though those sides are not directly presented; you experience an apparent cup as a cup, even though its being a cup includes certain dispositional features and causal-origin features that are not directly experientially presented; you experience other apparent bodies that suitably resemble your own apparent body as persons who are acting for intelligible reasons, even though many key features that make for personhood are not directly experientially presented, and even though the presumptive mental lives of others are not directly experientially accessible to you; and so on. Because of the extensive range of properties and relations to which the experiencing subject bears the relation of phenomenally constituted mental reference, there are numerous potential thoughts that have only narrow, phenomenal, intentionality. (Thoughts that have both phenomenal intentionality and externalistic intentionality will be discussed below.) Roughly, these are thoughts that are expressible linguistically using only (i) logical vocabulary, (ii) predicates expressing properties and relations to which the experiencer can mentally refer in a phenomenally constituted way, and (iii) certain first-person indexical expressions. So for instance, you might have a thought you could express linguistically by saying "A picture is hanging crooked on a wall directly in front of me." Your BIV phenomenal duplicate and your Twin Earth phenomenal duplicate would have corresponding occurrent thoughts with exactly matching, phenomenally constituted (and hence narrowly constituted), truth conditions. In terms of logical form, these matching thoughts have matching, doubly existential, contents involving the respective experiencing subjects as the respective referents of the first-person indexical thought-constituent: there is an χ and there is a y such that χ is a picture, y is a wall directly in front of me, and χ is hanging crooked on y (relative to the up-down axis of my self-oriented visual/kinesthetic reference frame). 7

7

This formulation may make the self-indexical aspect more explicit than it is in the thought itself; cf. note 3.

P h e n o m e n a l Intentionality and the Brain in a Vat

307

3.2 Mental Reference to Concrete Particulars: Grounding Presuppositions and Externalistic Factors Although some kinds o f mental reference are fully constituted by phenomenological factors alone (and hence purely narrowly), other kinds are constituted in a way that involves not only phenomenology but also certain externalistic factors. Singular mental reference to concrete particulars generally (perhaps always) works this way. Suppose, for example, that you have an occurrent thought that you could express linguistically by saying "That picture is hanging crooked," where the singular thought-constituent expressible linguistically by 'that picture' purports to refer to a picture on the wall directly in front of you. This thought-content involves certain phenomenally constituted presuppositions, which we call grounding presuppositions, that must be satisfied in order for the singular thought-constituent to refer: roughly, there must be an object at a certain location relative to oneself (a location that one could designate linguistically by a specific use o f the place-indexical 'there'), this object must be a picture, and there must not be any other picture at that location that is an equally eligible potential referent of 'that picture'. If these grounding presuppositions are satisfied by some specific concrete particular in your ambient environment—some particular object that is a picture and is uniquely suitably located—then your singular thought constituent thereby refers to that very object. Which object your thought-constituent refers to, if any, thus depends jointly upon two factors, one phenomenally constituted and one externalistic: on one hand, the phenomenally constituted grounding presuppositions, and on the other hand, the unique actual object in your ambient environment that satisfies those presuppositions. Compare the corresponding singular thought of your Twin Earth phenomenal duplicate. T h e corresponding singular thought-constituent also refers to something, because on Twin Earth too there is a unique object that satisfies the matching, phenomenally constituted, grounding presuppositions. However, that thought-constituent refers to a different concrete particular than yours does, viz., the specific picture that is suitably located in front o f your Twin Earth phenomenal duplicate rather than in front of yourself. T h e referents are different, for you and for your Twin Earth duplicate, because in the respective cases two different objects respectively satisfy the matching, self-indexical-involving, grounding presuppositions. Relative to you, the presuppositions are satisfied by the picture in front o f you; relative to your duplicate, they are satisfied by the picture in front o f her/him. Next, compare your B I V phenomenal duplicate's singular thought corresponding to your own singular thought, and the BIV's singular thought-constituent corresponding to your own thought-constituent that you could express linguistically by 'that picture'. For the BIV duplicate too, this thought-constituent has phenomenally constituted grounding presuppositions that match those o f yours. However, nothing in the BIV's ambient environment satisfies those presuppositions. Hence, in the case o f your B I V phenomenal duplicate, that thought-constituent fails to refer to anything. According to the conception o f mental intentionality we are here setting forth, these observations are pervasively generalizable. In general, singular thought-constituents have

308

Critiques o f C o n t e n t Externalism

phenomenally constituted grounding presuppositions, and the referent (if any) o f such a thought-constituent is the unique object (if any) in the experiencer's own environment that satisfies the applicable, phenomenally constituted and self-indexical involving, grounding presuppositions. Needless to say, the nature o f grounding presuppositions is an important philosophical topic worthy o f further investigation. We will return to this topic briefly in section 3.6.

3.3. Mental Reference to Natural Kinds. Section 3.2 focused on singular thought-constituents that purport to refer to concrete individuals. These thought-constituents normally would be expressed in language via singular terms, such as names or definite descriptions. Let us return now to predicative thought-constituents, those that purport to refer to properties, relations, or kinds. These include both general thought-constituents (roughly, ones that normally would be expressed in natural language via general terms such as verbs or adjectives, and in formal language as predicates) and certain singular thought-constituents that purport to refer to kinds, properties, or relations (thought-constituents expressible in natural language, for instance, by certain mass terms like 'water'). Although many predicative thought-constituents refer in a way that is wholly constituted phenomenally, not all o f them do. Among those that do not, as is widely acknowledged in light of the writings of Kripke and Putnam, are certain thought-constituents that purport to refer to natural kinds—for instance, thought-constituents expressible in language by terms like 'water' or 'gold'. The key points in the preceding section carry over to these kinds o f predicative thought-constituents, mutatis mutandis. Here too, the relevant thought-constituents have certain phenomenally constituted, reference-fixing, grounding presuppositions; and such a thought-constituent refers to the unique kind, property, or relation (if any) that satisfies those presuppositions. Here too, it is an externalistic matter whether or not the presuppositions happen to be satisfied at all; and it is an externalistic matter what (if anything) satisfies them. Here on Earth, the phenomenally constituted grounding presuppositions governing the thought-constituent expressible linguistically by the word 'water' are satisfied by H 2 0 . O n Putnam's Twin Earth, the matching phenomenally constituted grounding presuppositions are instead satisfied by a different natural kind: X Y Z . So for human experiencers on Earth, the relevant thought-constituent refers to H 2 0 , whereas for twin-human experiencers on Twin Earth, the corresponding thought-constituent refers instead to X Y Z . And for your B I V phenomenal duplicate, the corresponding thought-constituent does not refer at all, because nothing satisfies those grounding presuppositions. Delineating the range o f predicative thought-constituents whose reference depends in part upon externalistic connections between the experiencer and the experiencers environment, and also the range of predicative thought-constituents whose reference is instead constituted in a wholly phenomenal way, is an important philosophical task. Some cases are especially vexing, and various competing positions are possible. For instance, do predicative thought-constituents that purport to refer to colors refer in a

Phenomenal Intentionality and the Brain in a Vat

309

phenomenally constituted way, or instead in an externalistic way? Or do they, perhaps, fail to refer to any genuine properties at all? 8 The approach to mental intentionality we are describing here is officially neutral about predicative thought-constituents that purport to refer to colors and to other so-called "secondary qualities." But again, two key tenets of the approach are these. First, regardless of the exact range of predicative thought-constituents whose reference is partly externalistic, there are vastly many predicative thought-constituents whose reference is instead constituted wholly phenomenally. Second, when the reference of a predicative thought-constituent is partly externalistic, this is a matter of there being some unique kind, property, or relation in (or instantiated in) the experiencer's ambient environment that satisfies certain phenomenally constituted, reference-fixing, grounding presuppositions.

3.4. Mental Reference and Social Deference Perceptual experience presents to the experiencer an apparent world that is richly social: a world full of apparent persons apparently interacting with each other and with the experiencer in multifarious ways. Experience presents many of these apparent persons as apparently interacting with oneself linguistically, one experiences oneself as belonging to a linguistic community. Occasionally one experiences oneself as apparently employing language in a deferential way—as employing terminology that purports to refer to individuals, kinds, properties, or relations that one does not know much about oneself but that one believes are known about, and identifiable by, others in one's apparent linguistic community who have suitable expertise. Thus, the apparent use of language in an apparently socially-deferential way, within an apparent linguistic community that apparently engages in the "division of linguistic labor," are all aspects of the experiencing subject's overall phenomenology. So these aspects are all present in your BIV duplicate's experience as well as in your own—even though, in the case of the BIV, such experience is systematically nonveridical. How do phenomenally constituted grounding presuppositions work, in the case of apparently language-mediated constitutents of thought that are employed in an apparently socially deferential way? Suppose, for instance, that you have an occurrent thought that you could express linguistically by saying "Elm trees grow around here," and that you have no ability yourself to distinguish elm trees from various other kinds of trees. Consider your Twin Earth phenomenal duplicate's matching thought, and your BIV phenomenal duplicate's matching thought, each with a matching thought-constituent corresponding to your own 'elm tree' thought-constituent. How do the phenomenally

8

You, your Twin Earth phenomenal duplicate, and your BIV phenomenal duplicate all have phenomenally matching color-experience: apparent objects are experientially presented as having the same color-appearances. But this leaves open (i) what properties colors themselves are, (ii) whether there are such properties at all, and (iii) whether mental reference to color-properties (if there are any) is constituted purely phenomenally, or instead has an externalistic aspect.

310

Critiques of Content Externalism

constituted grounding presuppositions work, ones that you yourself share in common these phenomenal duplicates? Three interrelated factors come into play here. First are phenomenally constituted grounding presuppositions governing the thought-constituent you would express with the word 'tree', and also governing the corresponding thought-constituent of your Twin Earth phenomenal duplicate and your BIV phenomenal duplicate. These presuppositions already put substantial constraints on what could count as a referent of the thoughtconstituent purporting to refer to elm trees. (Dogs or automobiles, for instance, couldn't be elm trees.) Likewise, mutatis mutandis, for the corresponding thought-constituent of your Twin Earth phenomenal duplicate and your BIV phenomenal duplicate. Second are phenomenally constituted grounding presuppositions about the existence of actual experts in one's ambient environment, about these experts belonging to a genuine linguistic community that includes you too, and about their using elm tree' in some specific, referentially determinate, way. When these presuppositions are satisfied, your own thought-constituent that purports to refer to elm trees has its referents determinately fixed by the actual usage of these actual experts in your environment. This happens for you, and (with different referents) for your Twin Earth phenomenal duplicate. But it does not happen for your BIV phenomenal duplicate, because for it those grounding presuppositions are not satisfied; it is not really a member of a genuine linguistic community at all. Third are certain counterfactual facts about your phenomenology: facts about experiences that you would have if you were to have various kinds of investigation-experiences and/or expert-consultation experiences with respect to what counts as an elm tree. Such facts too reflect specific constraints on what counts as an elm tree, insofar as they provide evidence about how experts use the term elm tree' that is not at odds with how experts actually use that term. So in effect, these counterfactual-phenomenal facts underwrite assumptions that defeasibly have the status of grounding presuppositions: assumptions that have this status provided that they are not in conflict with the actual usage of actual experts in the experiencer's actual environment. For you yourself and for your Twin Earth phenomenal duplicate, these defeasibility conditions are indeed met, because actual expert usage conforms to what the relevant counterfactual phenomenology would present such usage to be. So, counterfactual phenomenology reflects specific, linguistically deferential, constraints on the reference-class of your own elm-tree' thought-constituent and on the reference-class of your Twin Earth duplicate's corresponding thought-constituent. What about your BIV phenomenal duplicate? It here becomes important that you yourself and your BIV duplicate are exactly similar not only with respect to your actual ongoing physical brain-processes and phenomenology, but also are robustly counterfactually similar. (See the second paragraph of the present paper.) Thus, the following is true of both you yourself and your BIV duplicate: if you (i.e., either one of you) were to have the experience of apparently consulting an expert about the apparent objects you apparently call "trees," then you would have the experience of apparently being shown or told some key distinguishing features of the apparent objects that you experience the apparent experts as apparently calling "elm trees." Now, in the case of your BIV phenomenal duplicate, there is not a discrepancy between (i) such counterfactual-phenomenology facts and (ii) actual linguistic practice by

Phenomenal Intentionality and the Brain in a Vat

311

actual experts in the BIV phenomenal duplicate's actual linguistic community. There is no such discrepancy because there does not exist a real linguistic community of which your BIV phenomenal duplicate belongs. So the defeasibility conditions that govern the counterfactual-phenomenology based, defeasibly grounding, presuppositions are not violated in the case of the BIV; thus, those assumptions do indeed have the status of grounding presuppositions. Counterfactual-phenomenological facts about the BIV thereby provide additional grounding presuppositions governing the BIV s elm tree' thought-constituent, over and above those provided by facts about the BIV's actual phenomenology. These further presuppositions involve the ways that apparent experts in the BIV's apparent linguistic community apparently would employ the apparent words 'elm tree'. The BIV's counterfactual phenomenology—the phenomenology that would ensue were the BIV to apparently seek out information about expert linguistic usage—places real constraints on what is required in order for something to be a referent of the BIV's 'elm tree' thought-constituent; it does so even though the apparent linguistic community is not itself real. The key point is that the operative, phenomenally constituted, grounding presuppositions rest upon more than the BIV's actual experience; they also depend partly upon certain counterfactual aspects of its experience.9

3.5 Mental Reference to Theoretical Entities Apparent ongoing scientific inquiry, together with the apparently linguistically mediated dissemination of the apparent results of such apparent inquiry, are aspects of the apparent world that is experientially presented to the experiencing subject. As a result of such experiences—including experiences of apparently studying science at a university, of apparently reading about esoteric theories with names like "Loop Quantum Gravity" in an apparent magazine apparently called Scientific American, of apparently conversing with apparent persons who apparently speak of themselves as "physicists," and the like—your BIV duplicate has a rich stock of thought constituents that purport to refer to various quite esoteric objects, processes, properties, and relations, many of which are not experienced observationally. Putative reference to such esoteric entities does not require any actual, externalistic, reference-constituting connections to them, any more than such externalistic connections are required for putative reference to tables, chairs, or other persons. Instead it arises via the experiencer's capacity for conceptual bootstrapping. Such bootstrapping begins with phenomenally constituted grounding presuppositions 9

But suppose that aliens elsewhere in the universe synthesize a brain, hook it to a computer in such a way that the brain's actual internal physical processes (and thus its actual phenomenology) happen to exactly match yours throughout its life, but do not construct the brain-computer interface in a way that provides robust counterfactual similarity to your own counterfactual brain activity and phenomenology. Perhaps, for instance, this counterfactual is true: if this BIV were to seem to itself to be seeking out information about how experts use 'elm tree', then the computer would cease to provide the BIV with inputs or life-support, and the BIV would die. For such a BIV, presumably there would be no real further constraints on what it takes to be a referent of the BIV's 'elm tree' thought-constituent, apart from whatever constraints are already constituted by the BIV's actual phenomenology. Apparent social deference generates real further constraints only when the appropriate counterfactual phenomenology is in place.

312

Critiques of C o n t e n t Externalism

governing thought-constituents purporting to refer to concrete objects and natural kinds that experience presents as directly observable. From there the bootstrapping proceeds up—often via apparent social-linguistic mediation, and often in a way that rests in part upon the fact that the experiencer can refer mentally to the relation of causation purely by virtue of experiential presentations of apparent instantiations of it—to phenomenally constituted grounding presuppositions governing thought-constituents that purport to refer to various kinds of non-observable "theoretical" entities and properties. Your BIV phenomenal duplicate has a full range of phenomenally constituted occurrent beliefs that exactly match your own, including scientific beliefs. For you and your BIV phenomenal duplicate, these matching beliefs have exactly matching, phenomenally constituted, grounding presuppositions.

3.6. Phenomenal Intentionality, Externalistic Intentionality, and Two Kinds of Truth Conditions As we said in section 3.1, some potential thoughts have only the narrow, phenomenally constituted, kind of intentionality. Roughly, these are thoughts none of whose reference-purporting constituents are such that their reference (if any) depends in part upon externalistic factors. But in addition, there are numerous potential thoughts employing reference-purporting constituents whose reference (if any) does partially depend upon such factors. (We will call these externalistic reference-purporting thought-constituents.) Such thoughts are capable of two kinds of intentionality: phenomenal intentionality and externalistic intentionality. 10 Phenomenal intentionality is narrow: there is a match in phenomenal intentionality between your own occurrent thought (say, a thought you could express linguistically by saying "That picture is hanging crooked") and the corresponding occurrent thoughts of your Twin Earth phenomenal duplicate and your B I V phenomenal duplicate. Externalistic intentionality, on the other hand, is wide: it incorporates the specific referents (if any) of thought-constituents whose reference is fixed in a way that involves external factors. Your picture-thought and your Twin Earth duplicate's corresponding picture-thought are about different pictures, and thus these corresponding thoughts have non-matching externalistic intentionality. Your BIV phenomenal duplicate's corresponding picture-thought fails to refer to any picture at all, and hence does not have externalistic intentionality. Thoughts with externalistic reference-purporting constituents are capable of two kinds of truth conditions, corresponding to the two kinds of intentionality. Phenomenal intentionality involves narrow truth conditions. These have "built into" them all the phenomenally constituted grounding presuppositions governing the given thought's externalistally reference-purporting thought-constituents. Insofar as such narrow truth conditions are formulable linguistically (whereof more below), the formulation will employ only these kinds of vocabulary: (i) logical expressions, (ii) predicative expressions

10

We say capable o f these two kinds o f intentionality because possession of externalistic intentionality requires cooperation from the e x p e r i e n c e d environment: grounding presuppositions must be satisfied.

Phenomenal Intentionality and the Brain in a Vat

313

designating properties and relations to which the experiencer can mentally refer nonexternalistically, and (iii) certain first-person indexical expressions. Take, for instance, a thought that you could express in language by saying "That picture is hanging crooked" where 'that picture' purports to refer to a picture directly in front of you. The narrow truth conditions for this thought would be expressible something like this: "There is a unique object x, located directly in front of me and visible by me, such that χ is a picture and χ is hanging crooked (relative to my visual/kinesthetic up/down axis)." Your own picture-thought, your Twin Earth phenomenal duplicates picture-thought, and your BIV duplicate's picture-thought all match with respect to these truth conditions; i.e., the truth conditions are just the same, apart from the different referents of the first-person indexical. Externalistic intentionality, on the other hand, involves wide truth conditions. The wide truth conditions of a given thought incorporate the specific satisfiers (if any) of the phenomenally constituted grounding presuppositions governing the externalistic reference-purporting constituents of that thought. In order for the thought's wide truth conditions to be met, those specific satisfiers must play a suitable truth-making role. Thus, your picture-thought and your Twin Earth phenomenal duplicate's corresponding picture-thought have non-matching wide truth conditions, because they respectively involve different pictures—in one case, the picture in front of you, and in the other case, the picture in front of your Twin Earth duplicate. As for your BIV phenomenal duplicate's picture-thought, it lacks wide truth conditions altogether (and hence lacks externalistic intentionality), because the pertinent grounding presuppositions are not satisfied. (Likewise, mutatis mutandis, for your own thought that you could express linguistically by saying "Water is good to drink," and the corresponding thoughts of your Twin Earth phenomenal duplicate and your BIV phenomenal duplicate.) For a thought that you could express in a sentence of your own language, normally the thought's wide truth conditions can be formulated homophonically, via that very sentence—because normally a word or phrase expressing an externalistically referential thought-constituent will be externalistically referential itself. 11 Is it in general possible to provide compact, cognitively surveyable, formulations of the narrow truth conditions for thoughts and for other intentional mental states? We doubt it, and we deny that this needs to be possible. Let us take up these points in turn: why we doubt it, and why the lack of such formulations would not be a problem. Here are two reasons to doubt that narrow truth conditions can always be tractably, compactly, formulated. First, ordinary perceptual experience is normally so phenomenally-intentionally rich that its intentional content defies complete description. The phenomenally constituted grounding presuppositions constituted by perceptual experience inherit this feature, and often one's ongoing intentional states and processes rely heavily upon enormous portions of this presuppositional background. (Think, for instance, of the presuppositional background at work in your mental life as you apparently hike along an apparent mountain trail—apparently constantly positioning your feet in

11

It bears emphasis, in connection with this observation, that wide truth conditions are the kind that belong to the that-clauses of many, perhaps most, propositional-attitude attributing statements.

314

Critiques o f C o n t e n t Externalism

optimal ways given the apparent local terrain both near your apparent feet and further up the apparent trail, apparently pausing periodically to take in the apparent view and examine various specific apparent objects, etc.) Second, many of the putatively referential thought-constituents deployed in thought, especially those thought-constituents involving apparently social aspects of the apparent world presented in experience, are heavily interconnected with other putatively referential constituents of other potential thoughts, in a holistically ramifying web of interlocking grounding presuppositions. So the presuppositional background of a given thought is often enormous, involving a significant portion of the web. (Think, for instance, of the vast presuppositional background lying behind the thought you could express linguistically by saying "Even some Republicans in the U.S. Congress are apt to balk at Bush's plan to eliminate federal taxes on investement dividends, because such a plan would be likely to harm the U.S. economy more than help it.") What exactly are narrow truth conditions, if they typically are not compactly formulable linguistically in a way that can be neatly plugged into the right side of statements of the form "Statement'S' is true iff,.."? Well, the truth conditions for a thought are the various possible ways the world might be each of which would accord with the thought's intentional content. As long as there is a reasonably determinate range of ways the world might be that count as potential truth-makers for the thought, then the possibilities within this range collectively constitute the thought's truth conditions. (We say reasonably determinate because intentional content quite often exhibits some degree of vagueness. 12 ) Truth conditions, thus understood, certainly need not be expressible in compact linguistic formulations; indeed, they need not be fully expressible even in infinitary linguistic formulations. 13 If, as we are suggesting, the background presuppositions figuring in the narrow content of intentional mental states typically cannot be spelled out in any tractable way, and if indeed these presuppositions typically are too complex and too ramified to be cognitively surveyable, do these facts prevent humans (and BIV duplicates of human brains) from possessing and competently deploying thought-constituents governed by such background presuppositions? Not at all. Rather, what's required for cognitive/ conceptual competence is that the cognitive agent make judgments, and possess judgment-forming dispositions, that accord reasonably accurately with the narrow truth conditions of the judgments—modulo available evidence. Humans meet this standard, and so would BIV duplicates of human brains. 14 The BIV duplicates' judgments would

12

W h a t counts as a grounding presupposition is often likely to be a somewhat vague matter itself. There need not be any sharp divide between the background assumptions that play a reference-fixing role visa-vis a given thought-constituent and those that do not.

13

Possible-world semantics, a familiar and widely invoked approach to truth conditions, certainly does not impose such expressibility requirements. Instead it typically construes the truth conditions for a given statement as constituted by a set o f possible worlds. T h e same goes for so-called two-dimensional modal semantics, an approach that has s o m e kinship to our own, including the positing o f two kinds of truth c o n d i t i o n s — o n e kind narrow and the other kind wide. See Davies and H u m b e r s t o n e (1980); Chalmers ( 1 9 9 6 ) , section 2.4, especially pp. 6 3 - 6 5 ; Jackson ( 1 9 9 8 ) , chapters 2 and 3, especially pp. 75-77; and Chalmers ( 2 0 0 2 ) .

14

S o m e o n e might maintain that the computational

conception o f human cognition requires that back-

Phenomenal Intentionality and the Brain in a Vat

315

very often be mistaken, of course, but nonetheless would be the right judgments to make given the available experiential evidence. Because thoughts with externalistic reference-purporting constituents are capable of two kinds of truth conditions—narrow and wide—they also are capable of two kinds of truth or falsity. First, such a thought is narrowly true just in case there exist entities that (i) satisfy the operative grounding presuppositions that govern the thought's externalistic reference-purporting constituents, and (ii) have the features that the thought attributes to them. It is narrowly false just case it is not narrowly true. Second, if (and only if) there do exist entities satisfying the operative grounding presuppositions, the thought has wide truth conditions too.15 So if the grounding presuppositions are satisfied, then the thought is widely true just in case the referents of its externalistically referring thought-constituents have the features that the thought attributes to them, and it is widely false just in case those referents do not have the features that the thought attributes to them. (If the thought's grounding presuppositions are not satisfied, then it does not have externalistic truth conditions; so it is neither widely true nor widely false. 16 ) When grounding presuppositions are satisfied, narrow truth and falsity normally run smoothly in tandem with wide truth and falsity: a thought with externalistic reference-purporting constituents is narrowly true just in case it is widely true, and is narrowly false just in case it is widely false. Consider, for instance, the thought that you could express linguistically by saying "Water is good to drink." This thought is narrowly true just in case (i) there exists a kind of stuff uniquely satisfying the grounding presuppositions governing your 'water' thought-constituent, and (ii) this stuff is good to drink. And the thought is widely true just in case that very s t u f f (Viz., H 2 0 ) is good to drink. 17 On the other hand, when the grounding presuppositions are not satisfied for a thought with externalistic reference-purporting constituents, then the thought is narrowly false and is neither widely true nor widely false. It is narrowly false because the

15

16 17

ground presuppositions be compactly and explicitly formulable; otherwise, it might be argued, there could be no programmable rules for manipulating mental representations in ways that suitably reflect their narrow truth conditions. This may well be right. But if so, we say, then so much the worse for the computational conception of human cognition. For arguments against the computational conception and in favor of a specific non-computational alternative, see Horgan and Tienson (1996). This is something of an oversimplification. It is more accurate to say that the thought has wide truth conditions just in case, for each of the thought's externalistic reference-purporting constituents, there exists a unique entity that comes close enough to satisfying the operative grounding presuppositions that the given thought-constituent actually refers to that entity. An alternative approach to wide falsity would treat a thought as widely false if its grounding presuppositions are not satisfied, rather than treating it as neither widely true nor widely false. Even when grounding presuppositions are satisfied, narrow truth and wide truth can still diverge in interesting ways with respect to certain kinds of modal and counterfactual reasoning. For instance, someone who doesn't know the chemical composition of water might contemplate each of the two epistemic possibilities that water is H20 and that water is XYZ, might form judgments about the comparative likelihood of these two possibilities, might have different counterfactual beliefs pertaining to each of these possibilities, and so forth. In effect, such reasoning holds narrow truth conditions constant across the possibilities under consideration, while varying wide truth conditions. This general theme is explored and developed in the literature on two-dimensional modal semantics; cf. note 13.

316

Critiques of Content Externalism

grounding presuppositions are built directly into the narrow truth conditions themselves, and fail to be met. And it is neither widely true nor widely false because it lacks wide truth conditions (and wide falsity conditions). This is precisely the situation, on a massive scale, with respect to your BIV phenomenal duplicates thoughts that have externalistic reference-purporting thought-constituents. Those thoughts all are narrowly false; and because they lack wide truth conditions, they all are neither widely true nor widely false.

4. Conclusion The position we have outlined meets the desiderata we set forth at the outset: it accommodates and vindicates Cartesian intuitions about the BIV, it accommodates the externalistic factors that play a constitutive role in the mental reference of many thought-constituents, and it is independently plausible both phenomenologically and theoretically. The key to this package is the recognition that thoughts with externalistic referencepurporting constituents normally have two kinds of intentionality, each with its own truth conditions: phenomenal intentionality and externalistic intentionality. Phenomenal intentionality is narrow, being wholly constituted by phenomenology alone. Your BIV duplicate has phenomenal intentionality that matches your own—with matching, phenomenally constituted, truth conditions. Externalistic intentionality on the other hand, is wide: its truth conditions incorporate the specific satisfiers, if any, of the phenomenally constituted grounding presuppositions that govern externalistic referencepurporting thought-constituents. Your BIV duplicate lacks externalistic intentionality, because the grounding presuppositions governing its thoughts are not satisfied. On our account, the BIV's thoughts with externalistic reference-purporting constituents are narrowly false, and are neither widely true nor widely false. This captures well the radical kind of nonveridicality that intuition attributes to the BIV's external-world beliefs. Their falsity is not the ordinary kind, in which a belief's grounding presuppositions are satisfied but its satisfiers do not have the features that the belief attributes to them. (In ordinary cases, a false belief is both narrowly and widely false.) Rather, it is a deeper kind of falsity, in which the grounding presuppositions themselves fail to be satisfied. Our account rests heavily and essentially upon the contention that mental reference to many properties and relations—including various spatiotemporal-location properties, shape-properties, size-properties, artifact-properties, and personhood-involving properties—is wholly constituted by phenomenology alone. Even systematically nonveridical phenomenology, as in the case of the BIV, provides reference-constituting acquaintance with such properties and relations. These properties and relations, in turn, are the ones figuring in the phenomenally constituted grounding presuppositions at work in narrow, phenomenal, intentionality. Some philosophers recently have put forth views that resemble ours in positing some form of phenomenal intentionality, but that differ from ours by not honoring the Cartesian intuition that the BIV has beliefs that are systematically nonveridical.

Phenomenal Intentionality and the Brain in a Vat

317

These views depart from our own position either (i) by treating all mental reference externalistically (e.g., Loar, forthcoming a, forthcoming b), or else (ii) by narrowing the extent of phenomenally constituted mental reference in such a way that the BIV's thoughts allegedly refer to what goes on in the computer that generates its inputs, and hence are actually veridical (e.g., Chalmers, forthcoming). Addressing these alternative approaches to phenomenal intentionality is a task for another occasion, but meanwhile the following point deserves emphasis: our own approach has the significant advantage, over these others, of thoroughly accommodating Cartesian intuitions. Versions of strong externalism about mental reference and mental intentionality have been very influential in recent philosophy, and strong externalism poses a challenge to Cartesian intuitions about the BIV. We have argued here that this challenge can be met, in a way that is independently plausible, vindicates Cartesian intuitions, and incorporates the reference-fixing roles often played by externalistic factors. If the position described here is correct, then strong externalism itself is seriously in error. We maintain that recent insights about external aspects of reference should be disentangled from strong externalism, and should be tethered instead to the quite different philosophical picture we have outlined. The most fundamental kind of mental intentionality is narrow, phenomenal, intentionality. 18

References Chalmers, D. (1996). The Conscious Mind, Oxford University Press. Chalmers, D. (2002). "The Components of Content," in D. Chalmers, ed., Philosophy of Mind: Classical and Contemporary Readings, Oxford University Press. Chalmers, D. (forthcoming). "The Brain in a Vat as a Metaphysical Hypothesis." Davies, M. and Humberstone, I. (1980). "Two Notions of Necessity," Philosophical Studies 38, 1-30. Flanagan, O. (1992). Consciousness Reconsidered, MIT Press. Horgan, T. and Tienson, J. (1996). Connectionism and the Philosophy of Psychology, M I T Press. Horgan, T. and Tienson, J. (2002). "The Intentionality of Phenomenology and the Phenomenology of In-

tentionality," in D. Chalmers, ed., Philosophy of Mind: Classical and Contemporary

Readings, Oxford

University Press. Jackson, F. (1998). From Metaphysics to Ethics: A Defense of Conceptual Analysis, Oxford University Press. Loar, B. (forthcoming a). "Phenomenal Intentionality as the Basis of Mental Content." Loar, B. (forthcoming b). "Transparent Experience and the Availability of Qualia." McGinn, C. (1989). Mental Content, Blackwell. McGinn, C. ( 1 9 9 1 ) . The Problem of Consciousness: Essays Towards a Resolution, Blackwell. Pitt, D. (2004). "The Phenomenology of Cognition, or: W h a t is It Like to Think that P?", Philosophy and

Phenomenological

Research 69, 1-36.

Putnam, H. (1981). Reason, Truth and History, Cambridge University Press. Searle, J. (1992). The Rediscovery of the Mind, M I T Press. Siewert, C. (1998). The Significance of Consciousness, Princeton University Press. Stich, S. and Warfield, T., eds. (1994). Mental Representation: A Reader, Blackwell. Strawson, G. (1994). Mental Reality, MIT Press.

On an Argument from Properties of Words to Broad Content FRANK JACKSON

I Beliefs play highly distinctive causal roles. Beliefs are, as we often say, responses to the environment that tend to cause behaviour satisfying subjects' desires in cases where subjects' beliefs are true. This is far too crude of course. For starters, it neglects the importance of degrees of belief and degrees of desire, and the consequent fact that we should, more precisely, think of the behaviour caused by belief and desire as increasing expected utility rather than necessarily realising desires in cases where beliefs are true. Those with the equivalents of "Field Marshall batons in their packs" behave in a way which increases their chances if their beliefs are true but only very rarely do they behave in a way which in fact realises their lofty ambitions. But the rough sketch is enough to tell us that beliefs are internal states of persons. It is an empirical fact about subjects that the states that play anything like these kinds of causal roles are located inside subjects' skins; indeed, are located in the brain. It may be that in the future certain kinds of cognitive deficits will be repaired by appropriately connecting one's brain to a computer strapped to one's back (or located at I T headquarters). In such a case, belief might (depending on the details) be a state partly on one's back (or at I T headquarters). But the way things are currently, all beliefs are located inside brains somewhat like ours and those of the higher animals, for only in these locations are the right kinds of causal roles being played. Although beliefs are inside us, they often concern how things are outside us. They are representational states, and how they represent things to be often concerns how things are outside us. For example, my (true) belief that the World Cup was recently played in Japan and Korea represents something about how things are a long way outside me. And what makes it true that it represents how things are a long way outside me is the fact that its truth derives from the way things were in Japan and Korea, and Japan and Korea are a long way away from where I am. In this sense, much belief is wide or broad or externalist. In this sense, belief is a state of subjects that can, and very often does, reach outside of subjects. The same goes for desire. Many of our desires are that things outside us be a certain way—that there be fewer famines or peace in the Middle East, as it might be. But we will conduct our discussion in terms of belief. How can someone be in belief states with the capacity to reach outside their skin? This is a large, highly controversial question but many, perhaps a majority nowadays, hold that we know this much about the answer to this question. The capacity is realised in part through the way beliefs qua internal states of subjects relate to subjects' surroundings, and this relation is such that how belief represents things as being, the content of belief, fails to supervene on subjects' internal natures. It is determined by

320

Critiques of Content Externalism

a combination of how subjects are and how their environments are. Two subjects can be identical from the skin in and yet have different beliefs in the sense that how their beliefs represent things as being are different. The subjects will respond in the very same way to each and every question of the form, Do you believe that there is a tiger outside the door? But, all the same, what they believe will in general differ if their surroundings differ appropriately. In this sense, belief content is broad or wide or external, not narrow. It is broad not only in the sense of typically concerning how things are outside believing subjects, but in the sense that subjects' surroundings play a key role in determining the contents of their beliefs. Putting the view in terms of the content of belief (and thought more generally, if it comes to that) being broad, though common enough, is a little misleading. The issue is not really the nature of belief content as such. A belief's content is how things are being represented to be by the belief, and, though there is great controversy about how to understand what sort of animal "how things are being represented to be" might be, the controversy is not over whether it is a broad or a narrow animal. The many who hold that the content of belief is broad or wide differ from those who hold that it is narrow over what makes it the case that a given belief has the content it does, not over what it is that may or may not be in part a function of subjects' environments. It might, as Robert Stalnaker suggests,1 be better to characterise the issue as over whether belief a broad (wide etc.) state, but ships take a long time to turn around. I will go with the majority and discuss the issue in terms of whether or not belief content is broad. No harm is done provided one is clear about what is at issue. This paper is about one very popular line of argument for the view that belief content is broad (and that the content of contentful states in general is broad, but we will focus on belief) that takes off from the semantic properties of certain words and sentences we use to report what we believe. Indeed, it is largely about the best-known such argument, the argument from Twin Earth. I will be suggesting that the line of argument is unpersuasive. But first I do some clearing of the decks. There are some relatively straightforward matters that should not be confused with the difficult issue of broad content.

II First, it is sometimes said with very little ado that the folk theory of content makes content broad or wide on the ground that folk theory treats content as truth conditional, typically with truth conditions that advert to how things are outside the subject. ...the folk view [is] that people have t h o u g h t s with rich representational properties and semantic properties; in particular, people have mental states with truth-conditional content. Cognitive psychology must explain the interaction of thoughts with each other and the world by laws that advert to these semantic properties: psychology must be 'wide'. 2

1 2

See, e. g., p. 196 of'Narrow Content', reprinted in his Context and Content, Oxford: Oxford University Press, 1999, pp. 194-209. Michael Devitt, Ά Narrow Representational Theory of Mind' in W. G. Lycan, ed., Mind and Cognition, Oxford: Basil Blackwell, 1990, pp. 371-398, p. 371.

On an Argument from Properties of Words to Broad Content

321

The possibility of narrow content is then said to turn on whether or not we can find a non-truth conditional notion of content important for psychology and cognitive science, perhaps tailored to play a distinctive causal explanatory role. It is regarded as close to axiomatic, or anyway very obvious, that truth conditional content is broad; the live issue is whether or not there is a narrow, non-truth conditional notion of content of importance in psychology.3 However, this is not the issue we will be concerned with. 'Content' for us is a term for how belief represents things as being, and is thereby ipso facto truth conditional—being true if and only if how things are represented as being is how things are in reality. We will, that is, be challenging one line of argument for a view regarded as close to axiomatic by many. It may be that the reason it is regarded as close to axiomatic that content conceived of in the representational-cum-folk-cum-truth conditional way is broad is the conviction that, only in that case, would it be possible for content, as we put it earlier, potentially to reach outside a subject's skin. How could a state inside the skin potentially represent how things are outside the skin independently of the skin's surroundings? But we have to allow that states may represent how things are in regions remote from themselves independently of how their surroundings are, on pain of a vicious infinite regress. Suppose that it is never true that S represents that P. All that is true is that S in surroundings S' represents that P. But 5 and S-—the aggregation—is itself a state. Repeating the reasoning, therefore, leads to the conclusion that the aggregation does not represent that P. Instead, S+S' in surroundings S" represents that P; and so on and so forth. Of course, this infinite regress does not show that belief content is narrow. It shows only that the question of whether it is broad or narrow is not closed by the fact that it may, and typically does, represent how things are outside believers. Second, it is sometimes observed, correctly, that content cannot be an intrinsic, or relatively intrinsic, property of a belief. How could the shape, say, of a belief, that is, of the relevant brain state, in itself possibly represent that the sun is a long way away? And the same goes for the belief's mass, temperature, size, location, electrical conductivity and so on, and indeed for all the properties sometimes called 'syntactic'. But, continues the argument, if content is a relational property, surely it must be a relation to the environment and thereby broad? Once upon a time, some took seriously the idea that a relation to Plato's forms or God's ideas or some kind of abstract entity might confer content on belief but we now know that the right relations must be to the naturalistic. A relation to an abstract entity might well be how we represent or 'measure' content but that is another question. Here is how Robert Stalnaker puts the key idea (his emphasis) Isn't it obvious that semantic and intentional properties generally, are relational properties... A n d isn't it obvious that relations depend, in all but degenerate cases, on more than the intrinsic properties of one of the things related. 4

3 4

See also Jerry Fodor, Psychosemantics, Cambridge Mass.: M.I.T. Press, 1987, ch. 2. Robert Stalnaker, 'On What's in the Head', reprinted in Context and Content: Collected tionality in Speech and Thought, Oxford: Clarendon Press, 1999, see p. 170.

Papers on

Inten-

322

Critiques of Content Externalism

However, there are properties which are not intrinsic, which relate the things that have them to their surroundings in some good sense, but which are shared between duplicates. One example is being among the tallest ten percent of people who exist timelessly. If X and Y have the very same height, then X and Y are alike in regard to their membership or otherwise of the tallest ten percent. All the same, being among the tallest ten percent is a relation to the people who exist at one time or another. Of course, the situation will be quite different if Xand Fare in different possible worlds, but our concern here is with world-bound versions of broadness. Our concern is with whether what a person believes supervenes on how they are from the skin in in the sense that, within their world, changing their surroundings, might or might not change what they believe.5 An example closer to home, in the sense of being a property many hold is roughly of a kind with the properties that confer content, is being soluble in water. If X and Y are alike from their surface in, they are alike in whether or not they are water soluble. The same is true generally of dispositional properties and functional properties: they supervene on relatively intrinsic nature. The reason dispositional properties are invariant across duplicates despite the fact that they concern relations—to immersion in water in the case of water solubility, to deformation in the case of elasticity, to being dropped in the case of fragility, and so on, is that, although their possession is a matter of facts about potential interaction patterns and what sustains these interaction patterns internally, their possession is independent of which elements, if any, of the interaction patterns are manifested. A sugar lump that is never put in water may be every bit as soluble as one that it is and dissolves; a rubber ball that is never deformed may be just as elastic as one that is and bounces back. Third, there is a sense in which much belief content is broad, a sense which should be, and typically is these days, acknowledged by even the most enthusiastic supporter of narrow content. Much belief concerns how things are with, or from the perspective of, subjects. For example, I have lots of beliefs about how far from me various things are, whereas you have lots of beliefs about how far from you various things are. Enthusiastic supporters of narrow content will insist that if you and I are duplicates from the skin in, then if I believe, say, that a chair is two feet from me, you must believe that a chair is two feet from you. But that is, in a sense, to believe differently, to have beliefs with different contents, for our beliefs will be true under different conditions. My belief will be true if and only if there is a chair two feet from me; yours will be true if and only if there is a chair two feet from you. However, if the narrow content enthusiast is right, we believe alike in the following sense: we agree in how we take ourselves to be. As David Lewis would put it, the property we each self-ascribe is the same.6 We each selfascribe being two feet from a chair. From now on, and when appropriate, I'll treat the issue of whether content supervenes on how subjects' are from the skin in as whether content in the sense of how subjects' take things to be with them or from their perspective, supervenes on how subjects' are from the skin in. That is the live issue. 5 6

Also, we are confining ourselves to creatures whose thinking does not reply on the knapsack computers we spoke of at the very beginning. David Lewis, 'Attitudes De Dicto and De Se, Philosophical Review, 88: 5 1 3 - 5 4 3 .

On an Argument from Properties of Words to Broad Content

323

I now turn to the discussion of the argument for broad content from certain semantic properties of the words we use to report what we believe.

Ill It is widely agreed that the referential properties, and meanings more generally, o f many o f the words and sentences subjects employ in saying how they take things to be d o not supervene on subjects' natures from the skin in. Their surroundings play a major role in determining reference and meaning in m a n y cases. I will not be querying this. M y concerns relate to the passage from this premise to conclusions about the dependence of what they believe, how they take the world to be, the content o f what they believe, on their surroundings. I start with a simple example. Suppose I hear noises from the study one floor down. I have no idea who is making them but am confident that what I am hearing is a staff member giving a tutorial. As a result, I come to believe that the person I can hear in the room below is giving a tutorial. Suppose that in fact, though unknown to me, it is M a r y I can hear giving the tutorial. T h e n , in s o m e g o o d sense, I have a belief about Mary. However, it is not true that how I take things to be one floor down is that M a r y is giving a tutorial. Ex hypothesi I have no idea that it is M a r y who is giving the tutorial. This is reflected in the truth conditions o f the sentence ' T h e person I can hear in the room below is giving a tutorial'. It will be true if M a r y is giving a tutorial in the room below, but equally it will be true if Harry is. ( H o w then is it right to say that my belief is about Mary? T h e reason lies in the conjunction of: (a) m y belief is that the person I can hear in the room below is giving a tutorial, (b) there is a person I can hear in the room below, (c) Mary is that person, and (d) hearing is potentially the right kind of relation to constitute de re belief. 7 ) We can easily modify the sentence ' T h e person I can hear in the room below is giving a tutorial' to make it one that requires that it be M a r y giving the tutorial for it to be true. ' T h e person I can actually hear in the room below is giving a tutorial' differs from ' T h e person I can hear in the room below is giving a tutorial' in that the former contains a rigid designator of Mary, and the latter does not. ' T h e person I can actually hear in the room below' designates in every possible world the same person, namely the person I can hear in the room below as things actually are; that is, Mary. Whereas, ' T h e person I can hear in the room below' designates M a r y in the actual world, but designates other people, e. g., Harry, in other worlds. In every world in which ' T h e person I can actually hear in the room below is giving a tutorial' is true, M a r y is giving the tutorial, but worlds in which ' T h e person I can hear in the room below is giving a tutorial' is true may or may not have Mary giving the tutorial. Under what epistemic circumstances am I in a position to utter ' T h e person I can actually hear in the room below is giving a tutorial'? Surely Gareth Evans is right when he says that this sentence is epistemically equivalent to ' T h e person I can hear in the

7

I here follow David Lewis, 'Attitudes De Dicto and De Se, Philosophical Review, 88: 513—543.

324

Critiques of Content Externalism

room below is giving a tutorial'. 8 T h e reason is that it is a priori that the person I can actually hear in the room below is giving a tutorial iff the person I can hear in the room below is giving a tutorial. It is true under any and every hypothesis about which world is actual. It follows that the epistemic circumstances in which one might justifiably say ' T h e person I can actually hear in the room below is giving a tutorial' are one and the same as those in which one might justifiably say ' T h e person I can hear in the room below is giving a tutorial'. T h e upshot is that the belief I report with the sentence ' T h e person I can hear in the room below is giving a tutorial' is the same belief as the belief I report with the sentence ' T h e person I can actually hear in the room below is giving a tutorial'. Despite the difference we noted above in the referential behaviour of the subject terms of the two sentences and the consequent difference in their truth conditions, the belief we report with one sentence is the same as the belief we report with the other sentence. This tells us that we must be careful when moving from the semantics of a sentence one might properly use to report what is believed to belief content. 9 T h e sentence ' T h e person I can actually hear in the room below is giving a tutorial' is true if and only if Mary is giving the tutorial. 1 0 Its truth requires that she be giving the tutorial, but how I take things to be when I use that sentence to report what I believe does not require Mary to be taking the tutorial. I have, as we said, no idea who is taking the tutorial. It would be bizarre to hold that by using the word 'actually', I somehow enrich my conception of how things are one floor down. I think the points we have just made with our simple example apply equally to the famous argument from Twin Earth. 1 1 Putnam offered the argument as one about the semantics of natural kind terms. T h e extension to the conclusion that belief and

8

More precisely, he says that 'Actually Q' is epistemically equivalent to 'Q'; see p. 210 of'Reference and Contingency', reprinted in his Collected Papers, Oxford: Clarendon Press, 1985, pp. 178-213. 9 As others have noted but for different reasons and more in the context of responding to Tyler Burge's arthritis case in 'Individualism and the Mental' in P. French et al., eds, Midwest Studies in Philosophy, vol. 4, Minneapolis: University of Minnesota Press, 1979. See Brian Loar, 'Social Content and Psychological Content', in R. Grimm and P. Merrill, eds, Contents of Thoughts, Tucson: University of Arizona Press, 1988, pp. 99-110, and Sarah Patterson, 'The Explanatory Role of Belief Ascriptions, Philosophical Studies, 59 (1990): 313-332. 10 This is true only for truth conditions in the sense corresponding to C-intensions (or secondary intensions as they are sometimes called). The /1-intension (or primary intension in the alternative terminology) of 'The person I can actually hear in the room below is giving a tutorial' takes the value true at any world where the person I can hear in the room below is giving a tutorial, whether or not it is Mary giving the tutorial. (For the distinction between C and /1-intensions, see Frank Jackson, From Metaphysics to Ethics, Oxford: Clarendon Press, 1998. The alternative terminology is used by David Chalmers, The Conscious Mind, New York: Oxford University Press, 1996 and elsewhere.) So why not draw the moral that when we seek to extract belief content from belief sentence, we should look to sentences' ^-intensions? I have no quarrel with this policy, quite the contrary in fact; but many view ^-intensions with deep suspicion. For example, Robert Stalnaker gives the flavour of 'On Considering a Possible World as Actual', Proceedings of the Aristotelian Society, Supp. Vol. 75 (2001), pp. 141-156, by displaying 'Hell is paved with primary intensions' on the first page. My aim here is to argue the issue without recourse to /4-cum-primary intensions. 11 Hilary Putnam, 'The Meaning of "Meaning"' in Language, Mind and Reality, Cambridge: Cambridge University Press, 1975, pp. 215-271.

On an Argument from Properties of Words to Broad Content

325

mental content in general is broad came later. It is the extension we will be querying. If we are right, it would have been better to leave matters where they stood after the publication of 'The Meaning of "Meaning"'.

IV Twin Earth is a remote planet very like Earth in many ways but isolated from it. Tjackson lives on Twin Earth. He is identical from the skin in with me. I speak English, he speaks Twin English. The big difference between Earth and Twin Earth is that the kind that plays the role played by H 2 0 on Earth is played by XYZ on Twin Earth. Call that role in both places the watery role. It includes, along with being a clear potable liquid at room temperature that falls from the sky etc., being the watery stuff we (in our case) and the Twin Earthians (in their case) are acquainted with and use the word 'water' for. The fine detail of how to spell out the watery role will not be important. 1 2 We all agree that, despite the two of us being identical from the skin in, 'water' in my mouth refers to H 2 0 and that 'water' in his mouth refers to XYZ. His water is not my water and, when necessary, we mark the point by using 'twater' for his water. And the reason for the difference lies in the difference in the environments we inhabit (as we are the same internally, there is nowhere else for the difference to come from). What has all this got to do with the contents of our beliefs? The contention is that the belief about how things are I express with, say, the sentence, 'Water is abundant' is different from the belief Tjackson expresses with the very same sentence, so we have a difference in belief content driven solely by a difference in surroundings. But wherein lies the difference? One obvious difference is that I have a belief about how things are around me and he has a belief about how things are around him; in consequence, the truth conditions of our beliefs differ. As we observed near the beginning, even the most enthusiastic supporter of narrow content must allow that doppelgänger may differ in the contents of belief in this sense. But this is compatible with their believing alike concerning how things are around them; they agree in the sense of ascribing the very same nature to the different parts of space-time that surround them. In any case, as we said above, this is not the issue on the table; and if it were, it was settled once and for all long before we were told about Twin Earth. Moreover, the difference in belief is not the result of the difference in surroundings or the subjects' environment. It is the result of a difference in the subjects; it is because I am not Tjackson, not because Earth and Twin Earth are different in regard to what plays the watery role. A way to bring this out is to note that had we considered a variant Twin Earth case where we contrast the actual situation with the counterfactual one where I am on Twin Earth-—I not Tjackson—the difference in question would disappear.

12 I have compressed the example in view of its extreme familiarity but not, I trust, in ways that bias the discussion. I borrow 'watery stuff' from David Chalmers, The Conscious Mind, New York: Oxford University Press, 1966, see p. 57.

326

Critiques o f C o n t e n t Externalism

Another difference between Jackson and Tjackson is in what our beliefs are about. My belief is about H 2 0 , and his is about XYZ. This indeed is the way the matter is very often put. But, as we saw in our discussion of the Mary case above, a difference in 'aboutness' is not enough to make for a difference in content. What matters is how things have to be for the beliefs to be true, not what they are about. At this point it will be said—perhaps with some impatience at my refusal to see what so many have agreed to—that, in order that my belief be true, H 2 0 needs to be abundant, whereas what needs to be true for Tjackson's belief to be true is for XYZ to be abundant. That's the difference. But what reason do we have to believe this contention? On the face of it, it is very implausible. As it happens, Jackson knows about H 2 0 and in particular that water is H 2 0 . But the Twin Earth argument is independent of this fact. We could have supposed Jackson to be living in 1700 and knowing nothing about H 2 0 , let alone know that it is water. The suggestion that how the Jackson of 1700 takes the world to be in any aspect, let alone that aspect he uses the sentence 'Water is abundant' to report, requires for its truth that H 2 0 be abundant implausibly makes out that the Jackson of 1700 anticipated Lavoisier—though, of course, he would get little credit for this, as it would be true of everyone on Earth who has 'water' beliefs. The same goes for Tjackson. He might not know about XYZ and in particular that twater is XYZ. He might be living before Tlavoisier. This makes it very implausible to hold that how things have to be for his belief to be correct is that there be XYZ. Many will say that the reason I am asking for to justify the contention that Jackson's 'water' beliefs require the existence of H 2 0 whereas Tjackson's require the existence of XYZ, lies in what we learn from the Twin Earth example about the semantics of'Water is abundant'. We learn that in Jackson's mouth it has different truth conditions from those it has in Tjackson's mouth. But our discussion of the example of Mary above tells us to look on inferences from the semantics of belief sentences to conclusions about the contents of beliefs reported by means of them with caution, and the Twin Earth case is, it seems to me, precisely one where we can go wrong. The key to seeing this is to note that it is contingent and a posteriori that 'water' has the semantics revealed by the Twin Earth case. The word 'water' in my mouth, and in the mouths of Earthians generally, might have been short for 'the watery stuff or some similar term. In this case, 'water' refers to H 2 0 when uttered by Earthians because H 2 0 is indeed the watery stuff around Earthians; and 'water' refers to XYZ when uttered by Twin Earthians because XYZ is the watery stuff around Twin Earthians. And in this case, the belief expressed by 'Water is abundant' in my mouth would have been the same as that expressed by the sentence in Tjackson's mouth, in the sense that we would have had the same belief about how things are around our respective selves. We would both have been expressing our belief that the watery stuff around us is abundant. The difference between this possible situation and the actual one, as revealed by the Twin Earth thought experiment, is that 'water' is a rigid designator of the watery stuff. It picks out the stuff around us that is actually watery. But what we learn from the Mary example above is that rigidification does not change belief expressed. We have, as Evans put it, a case of epistemic equivalence. What it takes to believe true 'The watery stuff around us is abundant' is true is exactly what it takes to believe true 'The actually watery stuff around us is abundant'. Little wonder

O n an Argument from Properties of Words to Broad Content

327

that the belief we express with the second sentence is precisely that expressed by the first.

V I think that there are many cases where we cannot automatically read off the content of beliefs from the semantics of belief sentences. We need to adjust in one way or another. Let me conclude with a simple case that does not involve rigidification and which demonstrates very clearly the way recent views about meaning and reference, whatever their merits, have the implication that care needs to be taken when using the semantics of sentences used by speakers to report what they believe to bolster conclusions about belief content. I believe that quarks exist and use the sentence 'Quarks exist' to express what I believe when speaking English. However, all I know about quarks is that they are fundamental particles called 'quarks' by physicists in my language community. (I exaggerate the extent of my ignorance—philosophers' license.) Speaking for myself, I see no problem with saying that 'quark' in my mouth means something like 'fundamental particle called "quark" by physicists in my language community'. If this is right, a grasp of what 'Quarks exist' means in my mouth tells you what I believe when I use the sentence to characterise my belief. However, many theorists of meaning deny that 'quark' in my mouth means 'fundamental particle called "quark" by physicists in my language community'. They hold that a proper appreciation of the lesson of the division of linguistic labour, along with the fact that 'quark' is a word in a public language, means that the word 'quark' in my mouth means what it means in the mouths of physicists. Suppose that they are right. Then the truth conditions of, say, 'There are quarks' as produced by me will be one and the same as the truth conditions of the sentence when it comes out of the mouth of a physicist in my language community. But it would clearly be a mistake to conclude that what I believe is the same as what physicists believe when we each say that quarks exist. I would very much like to be in the same epistemic state as physicists in regard to quarks, but that would take hard work on my part. Using a word they use is not enough, and nor is being in the same language community as physicists. I would have to do some serious reading and study. Perhaps it will be objected that my belief is about quarks and that that is why it cannot be true unless quarks exist. But I have beliefs about tables. Tables are aggregations of molecules—as a matter of fact according to some, as a matter of metaphysical necessity according to others. But we should not infer from this fact that my belief cannot be true unless there are molecules. Acquiring beliefs that represent that there are molecules took more work than acquiring beliefs that represent that there are tables. I have both as it happens but not in the one fell swoop.

Reference, Causal Powers, Externalist Intuitions and Unicorns GABRIEL M . A . SEGAL

0. Introduction In this chapter, I will compare and contrast singular concepts with what I call 'nature concepts', these being lay concepts of natural phenomena such as liquids (water) and types of animal (tiger). I will argue that the reference of singular concepts, which is obviously a variety of wide content, has a role to play in psychology. Singular concepts, I will claim, have both wide and narrow contents. In my view, no other concepts have wide contents. I shall give some reasons for supposing that psychology does not need to recognize any kind of reference-like content for nature concepts.

1. Wide Content and Causal Powers I'll begin by looking at an argument from Fodor (1987) that is meant to show that, in general, wide content should be kept out of psychology. Fodor argued more or less as follows. "Categorization in science is characteristically taxonomy by causal powers. Identity of causal powers is identity of causal consequences across nomologically possible contexts." (Fodor 1987 44). Psychology is a science. It taxonomizes by content. Causal powers supervene on local microstructure. So, Fodor concluded, content supervenes on local microstructure. If the argument were sound, it would follow that, in general, 'characteristic' scientific taxonomies would not distinguish micro-structural duplicates. But, it's obvious that plenty of scientific taxonomy does distinguish micro-structural duplicates and takes relational properties into account. For example 'island', 'crocodile' and 'planet' all apply to an object partly in virtue of relational factors (being surrounded by water, having the right ancestry, being gravitationally bound to a star). Fodor was of course aware of this. He attempted to account for it by recognizing those relational taxonomies whereby the relational properties 'affect causal powers'. Being a planet is thus taxonomic because 'whether you are a planet affects your trajectory, and your trajectory determines what you can bump into; so whether you are a planet affects your causal powers, which is all that individualism [i.e. internalism] asks for' 1 (1987 43).

1

I prefer 'internalism' to 'individualism' since the former better evokes ideas o f intrinsicness and narrow content. 'Individualism' in philosophy o f psychology might be taken merely to contrast with social theories o f content.

330

Critiques o f C o n t e n t Externalism

But that's not all internalism asks for. Internalism is about classifying micro-structural duplicates together. 'Planet' distinguishes micro-structural duplicates: a duplicate of, say, Jupiter that was not gravitationally bound to a star would not be a planet. An internalist about astronomy would demand an astronomical equivalent of narrow content. That would require there to be some astronomical property shared by all and only planets and twin planets. And I doubt that there is one, since just about any reasonably robust, non-self-illuminating object could be a planet, if it got into the right position vis-ä-vis a star. I'd guess that the property (if any) shared by all and only such objects is not taxonomic in any science, let alone in astronomy. Let us, then, not be internalist about astronomy. And, if we like internalism (as I do) let us not rely on the soundness of any arguments using premises that cite perfectly general features of science. So, something must be wrong with Fodor's argument. Either taxonomy in science is not characteristically done by causal powers or causal powers are not locally supervenient. I am not sure about the first disjunct. But the second one looks right. Fodor's official view is that 'identity of causal powers is identity of causal consequences across nomologically possible contexts.' On the other hand, he also says that being a planet affects your causal powers. That would suggest that, say, Jupiter and Twin Jupiter have different causal powers in spite of being micro-structural duplicates. Jupiter, being a planet, can bump into things in ways that Twin Jupiter cannot. If we use 'causal powers' in the spirit of that thought, then identity of causal powers is partly context-dependent and causal powers aren't locally supervenient. Suppose that Lara can bench-press 250 pounds on Earth. If she went to the moon, she could there bench-press 1,500 pounds. Would her causal powers thereby be changed? 'Yes and no' seems to be the best answer. We can distinguish two senses of 'causal power', one for a context-independent notion, the other for a context-dependent one. We might express the first by 'causal potentialities' and the second by active causal powers'. Lara has the same causal potentialities on Earth and on the Moon, but her active causal powers are greater on the Moon. Both notions are perfectly good. But one has to be clear which one is at work when discussing arguments about internalism. There is nothing wrong with taxonomies that respect relational properties. In some cases, although not necessarily all, they work because they reflect active causal powers. 'Planet' is one example. 'Moving' is another, in statistical mechanics: a molecule in motion has different active causal powers from one at rest. It can help heat things up. I would even argue with Fodor (1987) over what he took to be an uncontentious example of a relational property that fails to affect causal powers (for reasons recognized by Fodor himself in his (1991)). Fodor defined 'is an Η-particle' so that it is satisfied by a particle at t iff Fodor's dime is heads-up at t. Correspondingly a particle is a T-particle at t iff Fodor's dime is tails-up at t. Fodor thought it was obvious that he can't change the causal powers of all the particles in the universe just by flipping his coin. He said (1987 ρ 34) "whether something is an H- (T-)-particle is irrelevant to its causal powers. To put it a little more tensely, if an event e is caused by an Η-particle p, then that same event e is also caused by ρ in the nearest nomologically possible world in which ρ is Τ rather than H". That's wrong. It all depends what kind of event e is. One could build an Η-particle detector. The machine could work by detecting a particle, checking whether Fodor's dime was heads- or tails-up and classifying the particle

Reference, Causal Powers, Externalist Intuitions and Unicorns

331

accordingly. If the machine has classified ρ as an Η-particle and Fodor turns his coin over, then the machine reclassifies ρ as T. Suppose that e is an event of an Η-particle, p, causing the machine to classify it as H. Counterfactually, had ρ been T, then it would not have caused e. Evidently, just about any relational property can affect causal powers: there just needs to be a possible detector for it. 2 Of course, 'Η-particle' and 'T-particle' are not scientifically useful terms. But that's not because the properties they pick out could not 'affect causal powers'. Rather, it's just because, as a matter of fact, nature doesn't have lots of H - and T-particle detectors, so there aren't any interesting generalisations about H - and T-particles that we might want to describe and understand. So Fodors argument fails. Perhaps, then, wide content has a place in psychology because it 'affects causal powers'. In the next sub-section, I shall give some reasons to think that it does, and say something about how.

1.1 Reference in Psychology I tentatively believe that there is one and only one kind of wide content and that it has a place in psychology: the reference of singular concepts, those concepts typically expressed by proper names and demonstratives. I'll illustrate the sort of role I have in mind by a story involving a proper name. Lara is an archaeologist who comes across a papyrus apparently describing a ritual performed by an ancient Egyptian priestess called "Nefertoti". Lara becomes interested in Nefertoti. She pursues her research, and comes to believe that Nefertoti's remains might be found amongst certain ruins near Luxor, in Egypt. She makes her way there, and, after a thorough excavation, successfully locates a sarcophagus containing the remains. One would suppose that the semantic relation of reference holding between the concept Lara expresses by "Nefertoti" (her "'Nefertoti' concept") and Nefertoti, has a role to play in explaining why Lara ended up successfully locating the remains. It would be very surprising if it were an accident that the remains Lara discovered just happened to be those of the referent of a concept that features in some of the beliefs and desires that led her to Egypt. But what kind of role is reference playing here? Actually, the story is a bit complicated. I will now try to tell some of it. I take it that the supervenience base of the referential relation is partly intrinsic to Lara and partly relational. The intrinsic part presumably includes Lara's general competence with names, her (partly tacit) cognition of how the reference of a name is fixed, her knowledge of the particular term "Nefertoti" and maybe more. The relational aspect consists in a chain of linguistic events. At the start of the chain, a practice of using "Nefertoti" with the intention of referring to a particular individual, A, is set up among

2

As Fodor (1991) notes. However, Fodor (1991) continues to endorse the cross-contexts test for causal powers. This contradicts the line taken here and the results of a contingency test for difference of causal power that he proposes and makes much use of in that paper.

332

Critiques o f C o n t e n t Externalism

some people who can identify A independently of their knowledge of the name. Gareth Evans (Evans 1982) calls these 'producers'. At some point, a producer uses the name in the company of someone who doesn't know A. This person, a 'consumer', acquires the name and uses it with the intention of referring to the person referred to by the producer. 3 At some point, the name passes to the scribe who scripted Lara's papyrus, and thence to Lara herself. The intrinsic and relational aspects of this supervenience base provide different aspects of the explanatory role of reference. The relational aspect is purely historical and so does not directly make any relevant difference to Lara's or her concept's causal potentialities or her active causal powers. To see this, compare the situation originally described to one in which everything is exactly the same, except in respect of the historical relations linking Lara to Nefertoti. Imagine that Nefertoti was a priestess called "Nefertoti" and that various documents featuring "Nefertoti" exist in Lara's time. However, bizarrely, the manuscript that Lara originally reads features an occurrence of the name "Nefertoti" that is unconnected with the practice of using "Nefertoti" to refer to Nefertoti. The manuscript is a fake, created by someone who had never heard of Nefertoti, and who came up with "Nefertoti" by coincidence. In such a case, Lara's causal powers would be just as they are in the original case. She still ends up locating Nefertoti's remains near Luxor. Of course, in that bizarre situation, the initial desire she expresses by "I want to locate Nefertoti's remains" fails to get satisfied. But, even so, the desire has the same causal powers as the one that does get satisfied in the original case. Moreover, the concept's having its reference relates in obvious ways to factors that do enhance its active causal powers. The historical chain exists as part of the name-using practice. And that practice left marks that are still present in Lara's era: in particular, all the written records that contain the name "Nefertoti". The existence of those records enhances Lara's and her concept's active causal powers by providing her with an intellectual path leading to knowledge of the whereabouts of Nefertoti's remains. The internal aspect of the supervenience base figures in Lara's causal potentialities. The intrinsic properties of her cognitive endowment that allow her to understand "Nefertoti" are part of what explains her searching as she does. Putting these aspects together, we could say that Lara's concept is associated with her having a 'cognitive grip' on Nefertoti. . Very roughly speaking, a concept c associates with a cognitive grip on ο if and only if c's owner is in a position to form a large number of specific beliefs involving c, and these beliefs constitute knowledge. The beliefs must be 'specific' in the sense that they concern properties exhibited by ο in particular and not by the majority of other objects of o's kind. 4 The cognitive grip itself supervenes on two things: the intrinsic aspect of the supervenience base of reference

3

4

O r : with the intention o f referring to the person w h o is the d o m i n a n t source o f information associated with the name-using practice in which the producer was participating, to use Evans's ( 1 9 7 3 ) more complicated and more accurate proposal. M a r k Textor pointed out to me that it's important that we don't normally name boring objects, objects about which we couldn't or wouldn't f o r m m a n y specific beliefs. (I name this water molecule "Caesar"). If we did, then maybe reference would dissociate f r o m cognitive grip and lose its value for psychology.

Reference, Causal Powers, Externalist Intuitions and Unicorns

333

and—not the relational aspect of the supervenience base, but—current elements of the name-using practice that those relational factors are part of.5 Segal (forthcoming a) claims that singular concepts endow their subjects with a cognitive grip on their objects. But that is too strong. Suppose, for example, that just after Lara learns "Nefertoti", all traces of the name-using practice, except those in Lara's memory, were to disappear: the name is no longer written or spoken anywhere. Then Lara wouldn't have much of a cognitive grip on Nefertoti. But her concept would still have its reference. I would tentatively suggest that the reference of a singular concept is often enough associated with a cognitive grip to make reference worthwhile for psychology. Consider an example from Peacocke (1992) of a subject, S, pointing to a man, M, that he (S) can see in a garden. If Μ had been in another position in the garden, then S would have pointed there instead. That sort of situation occurs frequently. S has a visually based singular concept referring to Μ and this fact enters into the explanation of why S points as he does. The general structure of this second case seems parallel to that of the first one. Once again it is plausible that the referential relation supervenes on intrinsic and extrinsic factors. The extrinsic factors are historical (it's because it's Μ that caused the formation of the concept that the concept refers to M), hence do not directly affect S's current active causal powers in any relevant way. S's capacity to point to Μ is not directly affected by the fact that M, historically, had a causal role in the formation of the concept. (Suppose, for example, that another man, M2, had caused it (or one just like it) and was its referent, and there had been a quick switch whereby Μ took M2's place. S would still have the capacity to point to M.) But the historical factors are closely linked with present and future phenomena that do directly enhance S's causal powers. Quick switches don't happen very often and it's no accident that Μ is still standing there, reflecting light etc.. And those latter phenomena enhance S's active causal powers. There is a class of singular concepts for which the set-up is importantly different: those whose reference is fixed by description. Both name-type and demonstrative-type concepts can work that way: "Let us call the inventor of the magnifying glass 'Zippy'", "A man defied the King last night. That man will suffer." In these cases, the semantics is still likely to be associated with a cognitive grip. But the relevant semantic property is quite different: it is reference-conditions, not reference. Please consider (1) and (2): (1) Nefertoti refers to Nefertoti (2) (x)(Zippy refers to χ iff χ alone invented the magnifying glass). (1) describes a semantic relation that can only obtain if Nefertoti exists and if certain real relations obtain between her and possessors of the concept. (2) describes a semantic property of a "Zippy" concept that does not essentially involve any real relations to Zippy and does not require Zippy to exist. If you have a non-empty "Zippy" concept then you will have some cognitive grip on the referent, just because you grasp the reference-fixing description. In these cases,

5

I got the idea o f cognitive grip f r o m Evans ( 1 9 8 2 ) . I think that Evans would see the cognitive grip as grounding reference and hence necessary for it. I don't.

334

Critiques of Content Externalism

the descriptive content does all the work, and the referent itself is idle. If the subject is prone to form and revise beliefs about the referent in a way that is sensitive to the referent's actual properties, this is not due to their having causal influence. It's just because the subject is thinking about something as the possessor of those properties. The manner in which reference matters to psychology, if the above ideas about cognitive grip are correct, seems to involve two related factors. First, if you have a concept that refers to something, then you are more likely to have desires relating to that something, than if you don't. Second, since reference is associated with cognitive grip, you have a non-negligible chance of satisfying those desires, since you are in a position to gather truths about the thing. Thus, in general, if you have a concept that refers to something, there is non-negligible chance that you will end up trying to relate to it in certain ways and that your endeavours will be successful. The chances might be non-negligible enough for it to be worth a psychologists paying attention to the relation of reference. 6 If that is a correct account of reference in psychology, then it follows that reference can be a causally efficacious property of a concept. So if one says, for example, that it is in part because she has a concept that refers to Nefertoti that Lara located Nefertoti's remains, the "because" is a genuine causal "by cause of". Or so I will argue, following up on an idea from Segal and Sober (1991). Segal and Sober offer a sufficient condition for a property to be causally efficacious with respect to a particular effect. Essentially, the idea is that a macro-property F of an event, c, is causally efficacious in respect of some effect of event e's having property G, if (a) F events are (ceteris paribus) nomologically sufficient for G events and (b) c's possession of F mereologically supervenes7 on some set, M, of micro-properties of c such that c's possession of at least one of Μ is causally efficacious in respect of e's being G. Segal and Sober claim that according to their condition, representational content can be causally efficacious. Peacocke (1992) objects. He takes up Segal and Sobers example of the presence of air being efficacious in respect of the lighting of a match. The criterion legitimates the air's claim to be efficacious even though it is the presence of oxygen that is doing the causal work, not the presence of the nitrogen that is also in the air. Peacocke says that according to Segal and Sobers treatment, semantic properties of a mental representation are no more efficacious in respect of the effects of that representation than is the presence of the nitrogen in the air efficacious in respect of the match's lighting. Peacocke offers no argument for this claim and it appears to be false. If you apply Segal and Sobers condition to the atmosphere's property of containing nitrogen, then it does not license the efficacy of that property. Now suppose that we apply the condition to the reference of a singular concept. And let's suppose it's a case where the supervenience base of reference has intrinsic and extrinsic components, as discussed.

6 7

For more on the explanatory role of wide content, see Williamson (1998). Mereological supervenience is meant to be the converse of the 'makes it the case that' relation featuring in claims like: the way the diamonds crystals are bound together makes it the case that the diamond is hard. Mereological supervenience is, by definition, a synchronic relation.

Reference, Causal Powers, Externalist Intuitions and Unicorns

335

Then reference meets at least one relevant requirement: an aspect of the supervenience base, the internal one, is itself causally efficacious in respect of relevant effects. And that seems right. Reference is not analogous to the presence of nitrogen in the atmosphere, but to the presence of air. It is the historical component of the supervenience base of reference that is analogous to the presence of nitrogen. It is actually the other requirement, ceteris paribus nomological sufficiency, that appears more problematic. Consider the pointing case. It seems true that, ceteris paribus, anyone who has a visually based demonstrative concept, referring to an object o, who tries to point to o, will point in o's direction. But in Lara's case it's not so clear. Would anyone who has a name-concept referring to ο and who wants to find ο (or what's left of her) be likely to find o? That's not obvious. If what reference is doing here is enhancing people's chances of success, then the ceteris paribus nomological sufficiency requirement might be asking too much of it. It's not plausible that if we have a concept of something then, ceteris paribus, we'll be successful in our endeavours to relate to it. Personally, I find that my endeavours are often in vain. Rather, it has more to do with non-negligible chances, as discussed above. 8 Thus someone who has a concept referring to χ and tries to relate to χ in way w is more likely to relate in w to χ than someone who doesn't have a concept referring to x. But, anyway, it's a fine line between 'ceteris paribus' generalisations and increased probabilities. There's no huge difference between, 'Ceteris paribus, if ρ then q' and 'p increases the probability of q'. A version of Segal and Sobers condition that replaced the nomological sufficiency requirement with one framed in terms of raised probabilities looks as plausible as the original. And referential relations can meet this new condition for causal efficacy. Before moving on to the contrast between singular concepts and nature concepts, I pause to note that all of the above is consistent with reasonable versions of Fregeanism and internalism. Thus, consistently with the above suggestions about reference, we may suppose that singular concepts have, in addition to reference, some other kind of content resembling Fregean sense, in some or other important respects. 9 We may also suppose that, in addition to reference, they possess narrow content. And, in an Ockhamist spirit, we might go so far as to suppose that sense-like content is narrow and consists in, or determines, extension conditions. In the next section I will argue for a contrast between non-empty singular concepts and concepts of other sorts.

2. Nature Concepts, Externalist Intuitions and Unicorns The reference of certain singular concepts, then, may be a relation that matters to psychology. However, I would argue, reference is not the only kind of content required by psychology. Reference is a real relation, a relation that is only instantiated when both 8 9

See Segal forthcoming b for related discussion. See Segal forthcoming a and b for arguments in favour of sense-like content.

336

Critiques of Content Externalism

relata exist. O n the face of it, there are plenty of singular concepts that don't refer but that still have cognitive content. For example, suppose that an ancient Egyptian, Ahmenphut, prayed to Ra to end a heat wave, because he believed that Ra was the god who was responsible for the Sun, and hence the god with most control over the temperature. Ahmenphut's "Ra" concept has a cognitive content that it contributes to the beliefs that motivate Ahmenphut's action. Evidently this content is not its reference. So we must recognize cognitive content that isn't reference. 10 O n the face of it, the same reasoning appears to apply to a very wide range of nonsingular concepts: vampire, for example, has an empty extension. And while concepts like tax, taxon, tomfoolery and tomorrow, have non-empty extensions, it does not seem plausible that their having those extensions has any relevance to their cognitive content. That is to say, the concepts could have had the same cognitive content, had they had different or empty extensions. If that's right, then it looks as though non-empty singular concepts may be rather exceptional. Perhaps they are the only ones whose actual extensional relations are of interest to psychology. In fact, I think they are, although I will not argue the point here. The diametrically opposed view would, I suppose, be along the following lines. Reference of non-empty singular concepts is the paradigm of content, and we should seek an account of psychological content that is modelled on that paradigm. Empty singular concepts should be regarded as aberrations of some kind, and left out of consideration in theories of content. When it comes to general concepts, we need to find some analogue of reference to serve as content. That is, we need some real relation between concept and something else (say, a property or kind or substance) that can do some work in psychology analogous to that done by singular reference. 11 An intermediate view would be that there are some types of non-singular concepts that enter into semantic relations with real things in the world in a way that is psychologically significant and other types that don't. The obvious candidates for non-singular analogues of singular concepts are nature concepts. A popular view of the functioning of these concepts, arising from work of Putnam (1975) and Kripke (1972) sees nature concepts as very much like name concepts. I shall state a case against the intermediate view. If that case is successful, it should cast some doubt on the extreme view. If we can't even find a role for the reference of nature concepts in psychology, then what hope is there for the others? 12 I will proceed as follows. First, I will briefly sketch the Putnam- Kripke-inspired account. Then I will briefly state some of the chief arguments against the view that have

10 T h e ideas in this paragraph are elaborated and defended Segal (1989), see also Segal (2000) chapter 5 for an extended version of the argument applied to nature concepts. 11 See Fodor (1994) and Millikan (2000) for views in this ballpark. 12 This line of argument against the extreme view is suggestive rather than demonstrative, since the intermediate view relies on modelling the reference of nature concepts very closely on that of name concepts, and of thinking of natural kinds as scientific kinds. Note however that Fodor (1994) is devoted to trying to explain how and why reference matters for psychology and fails in that endeavour (see Segal 1997 for the argument). Millikan (2000) also deserves detailed discussion, which I reserve for another place.

Reference, Causal Powers, Externalist Intuitions and Unicorns

337

been developed elsewhere by others and myself. I will then focus on what I take to be one important source of support for the view: externalist intuitions about the semantics of nature terms. I'll argue that there are specific reasons why those particular intuitions should not be given much credence.

2.1 The Alleged Analogy between Name- and Nature Concepts and a Tempting Thought A nature concept is a lay concept of some natural phenomenon such as water, aluminium, polio, lemon or tiger. I begin by recalling the chief points of the Putnam- and Kripke- inspired account of nature concepts, staying closer to Putnam's version. Be so kind as to envisage a lay subject, Mathilda, in 1250, and her understanding of the term "nightingale". She is reasonably good identifying nightingales, knows what a typical nightingale looks like and knows something about typical nightingale habits. She uses the term "nightingale" with the intention of referring to a particular species of bird, members of which she and others on Earth have interacted with. On a twin Earth, there are birds that look just like nightingales, but they share no ancestry with Earth nightingales and have different genes. Twin nightingales are not nightingales, they just look like them. If Mathilda were to encounter a twin nightingale and say "Lo, a nightingale!", she would be saying something false in her language. That is because she takes her term to be true of members of the nightingale species, not of anything that happens to resemble such birds. W h e n she says "nightingale", she means nightingale, just as we modern sophisticates do. Mathilda does not know what makes a nightingale a nightingale. But she regards the species as having a hidden real essence. This essence is something that could be—and indeed later will be—discovered by science. So Mathilda uses "nightingale" as if it had been introduced to the language by a dubbing ceremony: "We shall call those birds 'nightingales', and the term is also to be true of all and only those birds that belong to the same species of bird as do they." The real essence, which determines the sameness relation, is not known to the dubbers. But there is a fact of the matter about it all the same, and it might be discovered by empirical investigation. It is tempting to think that "nightingale" works in much the same way as "Nefertoti". Speakers have a "nightingale" concept the extension of which is essentially related to a single species. This concept functions in such a way that it associates with a cognitive grip on the species. Thinkers deploy the concept in thought in such a manner that they form and revise beliefs involving that concept in way that is sensitive to the actual properties of the species. We can make the suggested parallel with singular concepts more explicit: if nightingales, qua nightingales have feature F, then there is a reasonable chance that someone with a concept that essentially extends over the nightingale species will, if the F-ness of nightingales matters to her, come to believe that nightingales are F and will therefore have a fair chance of interacting successfully with particular nightingales. If that suggestion is right, then we can see one reason why it makes sense to see nature concepts as having their extensions essentially limited to natural kinds rather than

338

Critiques of Content Externalism

motleys. For natural kinds are the kinds that do best in inductive inferences. Nightingales and twin nightingales might look alike. But that's just a superficial similarity. Since they are different species, the underlying difference is more important than the superficial similarity. In the longer term, the differences between nightingales and twin nightingales will lead to cognitive differences between Earth and twin Earth thinkers: the former will come to know about and adapt their behaviour to nightingales, the latter will come to know different things about the other sort of bird, and will modify their behaviour appropriately. Hence it makes sense to assign difference content to Mathildas and Twin Mathilda's "nightingale" concepts. The thought that nature concepts work like that might be tempting. But I think it is wrong. Outside science, we simply do not have much in the way of cognitive grips on most of the natural kinds whose samples we bump into. The result is that we would get little predictive mileage by assuming that nature concepts have their extensions essentially tied to natural kinds. The analogy with name concepts fails: (1) We are good at keeping track of individuals whose name we know: we are good at re-identifying them (finding out about their properties at different times) and we don't often confuse them with other individuals. By contrast, before the relevant science is done we are often bad at re-identifying natural kinds. When Archimedes figured out how to distinguish gold from fool's gold, he made history. In the normal run of things, for most kinds of natural kinds, we have little idea which individuals belong to the same natural kind and which don't. 13 (2) We are very interested in keeping of track of individuals. Our interests in kinds are many and various and typically not best served by carving nature at its joints. (3) There are actual dubbing ceremonies for individuals. Outside of science, there are no dubbing ceremonies for kinds. (4) We know what kind of thing a name is supposed to refer to: an individual. But even if there were dubbing ceremonies for kinds, we would have no good idea of what kind of kind we would be referring to—subspecies, species, genus, family? Further, as has often been pointed out, the Kripke-Putnam model makes a false prediction about actual usage. Suppose that, pre-scientifically, people apply a term to samples nearly all of which belong to the same natural kind, but occasionally fail to apply it to samples of the kind and/or apply it to some things that are not of the kind. Then the science is done and the facts discovered. The model predicts that, in such cases, we would revise our usage accordingly and correct what would seem to have been classificatory errors. In fact, though, this is not what typically happens. Rather we just carry on as before. 14 "Cat", in its most general sense, is a good example of this. It applies to members of the family Felidae, but also to a few non-members, such as civet cats. In the final sub-sections, I will suggest an alternative account of nature concepts, and go on to say something about the role that intuitions ought to play in adjudicating between the two accounts.

13 We are very competent with respect to plant and animal kinds. I will discuss those shortly. 14 For arguments and examples supporting the claims of the last two paragraphs, see Segal forthcoming b, Segal 2000 chapter 5 and references there cited.

Reference, Causal Powers, Externalist Intuitions and Unicorns

339

2.2 A Motley Alternative The Kripke-Putnam account, then, is highly problematic. Let me offer an alternative. I suggest that nature concepts, rather than extending over natural kinds, extend over motleys. I expect that "motley" is itself something of a motley term. But the idea is that what groups samples together under a motley concept is not a hidden real essence. Rather, members might be grouped either by features explicitly known to those whose concept it is, or by features that are determined in more (or less) complicated ways by features of the thinkers' cognition. 1 5 T h e concept of tree as expressed by the Oxford English Dictionary, definition b., is a good example: "a. a perennial plant with a woody self-supporting main stem or trunk when mature and usu. unbranched for some distance above the ground, b. any similar plant having a tall erect usu. single stem e.g. palm tree". The key point about motley concepts, when it comes to the issue of the role of reference in psychology, is that they act like descriptive concepts. It is their extension conditions, not their actual extensions, that matter to psychology. One can have a motley concept whether or not the motley exists, and the extension conditions of the concept are independent of the nature of any samples that there might be. Any cognitive grip on the motley that associates with possession of a motley concept supervenes on features of the thinker's cognition that are not essentially dependent on real relations between thinker and members of the motley. And a key point about motleys when it comes to the debate over internalism is that it's plausible that motley concepts are shared by twins. Mathilda's and Twin Mathilda's "tree" concepts have the same extension conditions. If their "nightingale" concepts are motley concepts, then it's reasonable to suppose that they would both extend over both nightingales and twin nightingales.

2.3 Unicorns, Externalist Intuitions and Chachalacas One key source of support for the Kripke-Putnam account comes from certain externalist intuitions, for example, intuitions generated by classical twin Earth thought experiments. So, for instance, we are supposed to have the intuition that if Mathilda were confronted with a twin nightingale and said "Lo, a nightingale!", she would be saying something false and expressing a false belief. I think that there are specific reasons why we should not trust those intuitions, which I will explain as we proceed. To begin with, contrast those intuitions with analogous ones about name concepts. Suppose that Mathilda's best friend is Mary, and that on Twin Earth, Mary has an identical twin. If Mathilda were to come across Twin Mary and say "That's Mary" then she would be saying something false and expressing a false belief. Now that is intuitively

15

Notice the "might". There might be no metaphysically interesting answer to the question "what groups members under motley concept FV other than "their being F".

340

Critiques of Content Externalism

obvious. 16 But the key thing is that it is obvious to just about everybody, including Mathilda herself. Our intuitions about our own use of names and the functioning of our own name concepts provide evidence about their meanings and reference. By contrast, twin Earth intuitions relating to nature concepts are notoriously unstable, varying with how they are elicited and whom you ask. Does your intuition really tell you that Mathilda's concept isn't true of twin nightingales, that it doesn't extend over the motley, rather than the species? Wouldn't we do better to ask Mathilda? I suspect that both Putnam and Kripke draw on their intuitions about the extension conditions of concepts and mistakenly think that their intuitions are ours', that they are representative of those of all sensible, reflective humans. I think that there is a dangerous element of a priorism in that approach to the individuation of contents. The remainder of the paper will elaborate on that thought. I begin with some claims made by Kripke in his unpublished (1973). In the course of an argument for the conclusion that unicorns could not have existed, Kripke made some remarks about the medieval use of the term "unicorn". Kripke introduced the possible term "shmunicorn", which he defined as "member of the species of animals, if there in fact is one, which is identified by the following surface characteristics, having one horn and looking like a horse". He went on to claim that he doubted that this was the way the term "unicorn" was used by the medievals, who took the term seriously. He said that they probably did believe that there was only one species that looked like this. But they would acknowledge the possibility of species that had the relevant surface characteristics but that would not be unicorns. Kripke asked his audience to suppose that the medievals believed that Sir Galahad had met a unicorn. Then, according to Kripke, a medieval would agree that, if in some part of the world there was a species of animals other than that of which a specimen was met by Sir Galahad, but which looked like it, these animals would not be unicorns. The thing to note is that Kripke was making substantial empirical claims about how the medievals used or would have used the term "unicorn". But Kripke provided no empirical evidence in favour of these claims. I would speculate that he thought empirical research wasn't necessary because he assumed that his own intuitions about species and our words for them would be a reliable guide to medieval intuitions about unicorns and "unicorns". But given the huge intellectual differences between 20 t h century academics and medieval lay persons, such an assumption would be unwarranted. I have done a little of the required research and it does not confirm Kripke's claims. First, it is widely accepted among medieval historians that the term "unicorn" was used by the medievals as a broad motley term meaning quadruped with one hornΡ So, for example, there are two species of unicorn described in Androvandus' 12 th century Bestiary (from the translation, White 1954). Here are extracts: "Unicornis the Unicorn, also called rhinocerous by the Greeks, is of the following nature. He is a very small animal like a kid, excessively swift, with one horn in the

16 17

If you don't find it obvious, then assume that the twin is not called "Mary" and focus just on the belief Mathilda expresses. My source is medieval historian Brigitte Resl.

Rcfcrcnce, Causal Powers, Externalist Intuitions and Unicorns

341

middle of his forehead and no hunter can catch him." (p. 20) That sort of unicorn can only be caught by a lone virgin in the woods. And the second extract tells us: "The monocerous 18 is a monster with a horrible howl, with a horse-like body, with feet like an elephant and with a tail like a stag's. A horn sticks out from the middle of its forehead with astonishing splendour" (p. 44). Unicorns of that sort cannot be caught at all, although they can be killed. 19 The term "unicorn" was also used more specifically to designate a specific type of unicorn. In that usage, would it have expressed a motley concept or a natural kind one? Kripke's claim was that a medieval would agree that if there were animals that looked like members of the species a specimen of which was encountered by Sir Galahad, but belonged to a different species, then they would not be unicorns. Perhaps in some sense Kripke is right about that. But if so, then it would not show that the medieval species concept of unicorn was a natural-kind concept rather than a motley concept. There are two related reasons for this, one to do with Kripke's implicit choice of question and one to do with the term "species". Kripke's implicit question is framed in terms of appearance: would these animals that looked like unicorns be unicorns? But motley concepts are not typically appearance concepts: being an F might involve a great deal more (and/or less) than looking like an F, even if F-hood is not a natural-kind property. And the problem with "species" is that we are used to using the term in its contemporary scientific sense. The medievals did not use it that way. It is therefore not particularly clear how a medieval would have interpreted a question framed in terms of "species", nor is it clear that we could have asked them about conspecificity without first teaching them biology. I suggests we consider a variation on Kripke's thought experiment. Suppose then, we were to ask Mathilda this: "In a land far away, there are some animals that resemble the unicorn encountered by Sir Galahad. This type of animal has one horn in the middle of its forehead, it ventures to attack elephants, so sharp is the nail in its foot that with one blow it can rip the belly of a beast. Hunters can catch the animal only by placing a young virgin in its haunts. No sooner does the animal see the virgin, than he runs towards her and lies down at her feet and so suffers himself to be captured by the hunters. 2 0 But these animals are of a different species from the one encountered by Sir Galahad. Would they be unicorns?" How might Mathilda have reacted? I think she might have been rather mystified. She might wonder how an animal with so many of the characteristic properties of unicorns could be of a different species. In virtue of what would they not be unicorns? Thus she might have said: "The animal you have described is certainly a unicorn. Your description is most precise. Indeed, the unicorn is the only animal so brave as to venture to

18 19 20

"Monocerous" is just the Greek translation of "unicorn". In fact, many medieval bestiaries follow the elder Pliny's Natural History and list three species. This description, which is pretty standard, is adapted from Le Bestaire Divin de Guillaume, Clerc de Normandie, 13 th century, quoted in Brewer's Dictionary of Phrase and Fable, Millennium Edition, ed, Adrian Room.

342

Critiques o f C o n t e n t Externalism

kill an elephant. 21 Please explain in what sense they belong to a different species from the one encountered by Sir Galahad". An answer in terms of shared ancestors or genes or interbreeding would not have meant much to Mathilda, since the medievals did not think of species in those terms. In sum, Kripke's claims about what a medieval would agree to are based on his own intuitions. And one would suspect that those intuitions are partly based on his reading key features of his own contemporary, scientifically-informed concept into the pre-scientific medieval concept. Kripke provides no empirical evidence in support of his claim, and certain evidence suggests that it is false. I move on now to some more evidence bearing on the nature of nature concepts. Anthropologist Scott Atran and his collaborators made an extensive ten-year (and counting) study of what they call "folkbiology", this being the way normal humans think about plants and animals. Atran studied, in particular, people of various tribes with little or no scientific education, living in natural habitats around the world. Atran and his collaborators found interesting universals. In general, such people are very good at classifying (middle-sized and large) flora and fauna. Folk classifications for these correspond closely to scientific ones in respect of their actual extensions. The folk classification scheme is always organised as a ranked taxonomy, that is, a genuine hierarchical taxonomy that allows for generalisations over the different ranks themselves. And, finally, the folk believe in individual essences. Each individual plant or animal has an essence that determines its identity and persistence conditions. (See e.g. Atran 1999). Pinker (1994 423) recounts: in rural Nigeria, some students disguised a pawpaw as a pineapple and asked natives whether the result was pawpaw or pineapple. A typical response was 'It's a pawpaw because a pawpaw has its own structure from heaven and a pineapple its own origin. One cannot turn into the other.' Such essences are speciesspecific and passed from parent to offspring. Interestingly, these apparent universals of folkbiology are not apparent in children who grow up in cities and who are explicitly taught biology (see Carey 1985). But city dwellers are exceptional in their biological ignorance. Atran's work indicates that there is an innate, species-specific folkbiology module of some kind, which module functions in humans who grow up in our natural environment but which atrophies in humans who grow up in cities and receive formal education Kripke and Putnam tried to find out about human nature concepts, including specifically those of plants and animals, by consulting their own intuitions. But they, too, grew up in cities and received formal education. This means that their intuitions may be far from representative. In particular (as Atran thinks) the view that natural kinds are individuated by hidden, real essences may be the fruit of an illegitimate union between the folk idea of heaven-sent essences and a respect for science as the discoverer of deep truths. But heaven-sent essences are not the kinds of things that science can tell us about. They are other-worldly.

21

W h i c h it is, according to Guillaume's Bestaire.

Reference, Causal Powers, Externalist Intuitions and Unicorns

343

Putnam's and Kripke's intuitions are suspect. The intuitions of the folk themselves, however, can provide useful evidence about the nature of the folks' own concepts. We can't ask Mathilda about her intuitions. But we can ask the folk. By consulting people from unscientific cultures living in nature, we have a chance to study nature concepts in something like their natural habitat. Such a study might provide evidence relevant to adjudicating between the natural-kind hypothesis and the motley hypothesis. Atran, his collaborators and I have made start. Atran and I constructed a questionnaire designed to tap relevant twin-Earth intuitions among tribespeople. The questions were put to Maya subjects in the Yucatan, Mexico. Unfortunately, at the time of writing some of the data remain inaccessible, due to floods and hurricanes in Mexico. As a result, we have not properly analysed such data as we have. I conclude with an anecdote arising from our pilot study. The key question that we asked was as follows: "In a far away land there are some birds. These birds look exactly like the chachalacas22 here in Yucatan and move and sound exactly like the chachalacas here. But these birds do not come from the same ancestors or lineage as the chachalacas here. Are the birds of the far away land chachalacas or not?" What answers might a proponent of the Putnam-Kripke account predict? Perhaps one sensible answer, under this view, would be "maybe". The question would be whether the twin chachalacas had the same hidden, real essence as Yucatan chachalacas. The Maya, not knowing the relevant science, might abstain on what the essence is or how you could tell when something had it. Another sensible answer would be: "No". If species are natural kinds at all, then it's likely that species boundaries are in fact determined by ancestry, and the Maya are here being given the relevant information. They might already think of ancestry as crucial. Or they might expect that whatever the essence consists in, ancestry would be a good guide on the subject. O n e would not predict, however, that they would answer "yes". Given that there is an important difference between chachalacas and twin chachalacas, the possibility that the twin birds are not chachalacas, but merely resemble them, ought to be salient. The Maya subjects, then, would not be expected to answer "yes". The results were: "yes" 20, "no" 17 and "maybe" 4. These data certainly support the motley hypothesis and count against the natural-kind hypothesis. O f course, they are not conclusive. The issue turns on the best overall account of Maya cognition. Other things the Maya think would also need to be taken into consideration. But surely these data, which tap Maya intuitions about animals, should be given considerably more weight than Putnam's intuitions about Oscar's "water" concept and Kripke's intuitions about medieval "unicorn" concepts. 23

22 23

Chachalacas are local birds. I presented an early version of this work to an audience at the Philosophy, Psychology and Neuroscience programme at Washington University, St. Louis. I am grateful to participants there for helpful discussion. I am much indebted also to Scott Brocket, Shamik Dasgupta, Nicole Harris, Keith Hossack, David Papineau, Mark Sainsbury, Susanna Siegel, Nick Shea, Maja Spener and Mark Textor for helpful comments on early drafts, to Brigitte Resl for much help with medieval concepts of unicorns and to Scott Atran for arranging the interviews with Yucatan Maya and for discussion.

344

Critiques of Content Externalism Bibliography

Atran, S., 1999 "Itzaj Maya Folkbiological Taxonomy: Cognitive Universals and Cultural Particulars", in Medin, D., and S. Atran (eds) Folkbiology, Cambridge MA: MIT Press. Carey, S., 1985 Conceptual Change in Childhood, Cambridge MA: MIT Press. Evans G., 1973 "The Causal Theory of Names", Proceedings of the Aristotelian Society, Supp Vol 47 187208. Evans, G., 1982 The Varieties of Reference, Oxford: Oxford University Press. Fodor, J. A, 1987 Psychosemantics, Cambridge MA: MIT Press. Fodor, J. A. 1991 "A Modal Argument for Narrow Content", The Journal of Philosophy 88: 5-25. Fodor, J. A. 1994 The Elm and the Expert, Cambridge MA: MIT Press. Kripke, S., 1972 Naming and Necessity, in D. Davidson and G. Harman (eds.) Semantics for Natural Languages, Dordrecht: Reidel 253-355. Kripke, S., 1973 unpublished, John Locke Lectures, Oxford. Millikan, R., 2000 On Clear and Confused Ideas. Cambridge: Cambridge University Press. Peacocke, C. A. B., 1992 "Externalist Explanations", Proceedings of the Aristotelian Society. Pinker, S., 1994 The Language Instinct, Penguin. Putnam, H., 1975 "The Meaning of'Meaning'", in K. Gunderson (ed.) Language, Mind and Knowledge, Minnesota Studies in the Philosophy of Science 7. Segal, G„ 1989 "The Return of the Individual", A W 9 8 , 39-57. Segal, G., 2000 A Slim Book about Narrow Content, Cambridge MA: MIT Press. Segal, G., 1997 "Content and Computation: A Critical Notice of Jerry Fodors The Elm and the Expert", Mind and Language: 491 -501. Segal forthcoming a "Intentionality", in Jackson., F., and M. Smith (eds.) The Blackwell Companion to Contemporary Analytic Philosophy, Oxford: Blackwell. Segal forthcoming b "Keep Making Sense", in Crane, T., (ed.), Contemporary Philosophy in Focus: Jerry Fodor, Cambridge: Cambridge University Press Segal, G., and Sober., E., 1991 "The Causal Efficacy of Content," Philosophical Studies 63, 1-30. Williamson, T., 1998 "The Broadness of the Mental: Some Logical Considerations", in Tomberlin, J., (ed.), Language, Mind and Ontology, Philosophical Perspectives 12, 389-410, Oxford and Boston: Blackwell. White, Τ. H., 1954 The Book of Beasts: a Translation from a Latin Bestiary of the Twelfth Century. London: Jonathan Cape.

ν An Exemplary Debate about Content

Millikan's (Un?)Compromised Externalism 1 GEORGES REY

1. Introduction In the past twenty-five years Externalism seems to have become something of an orthodoxy in the theory of conceptual content. Although I am sympathetic to many of the impulses in the work of Putnam (1965/75, 1975), Kripke (1972/80) and Bürge (1979, 1986) that originally motivated it, I am worried by the extremes to which it seems currently to have been taken. Elsewhere (Rey 1992, 1995, 1998) I have expressed these worries with regard to the externalisms espoused by Fodor (1990), Lycan (1987), Dretske (1995) and Tye (1995). In this paper I want to consider what I take to be an even more extreme form than any of theirs, namely the form defended by Ruth Millikan (2000), her recent book, On Clear and Confused Ideas ("CCI"). Millikan defends there what she characterizes as an uncompromisingly externalist view. W h a t makes a thought be about a certain substance is not merely in the mind, nor any mere disposition of the mind, not even a wide disposition, but the thoughts origin—an external causal historical relation between the concept and the substance.—(CCI, p. 9 5 )

My general worry about this view is that it seems to me to presume a much greater attunement to the real world than we have any reason to suppose, and Millikan's arguments for it are inadequate to override strong prima facie evidence to the contrary (§2). It also seems to me to fly in the face of serious explanatory practice not only in psychology, but in a selectionist biology itself (§4). Moreover, like externalist views in general, it has a problem with vacuous concepts, like [devil] or [angel], that fail to refer to anything—certainly nothing in the selectional history of any organism. Millikan's proposal would actually exclude much of the psychology of the very White Queen (who makes a habit of believing contradictions) with whom Millikan (1993, p. 280) expressed much sympathy (§5). All these problems seem to me to point to the need for something more internal. Millikan herself notices a plausible theory needs to identify contents at least in part by some internal workings (§6). However, her "uncompromising" externalist commitments prevent her from dealing with these internal workings with sufficient care, and this, I think, leads her to make needlessly rash claims about conceptual competence, the language of thought, and the non-existence of identity judgments (§7). A more compromising externalism might invite a more sensitive internalism that a nonetheless naturalistic explanation might need.

1

An expression enclosed in square brackets refers to the concept expressed by the expression.

348

An Exemplary Debate about Content

A quasi-verbal issue that is of some interest in the discussion. It might be thought that dispositional (e.g. discrimination, indicator) accounts of content, like those of Dretske (1980) or Fodor (1991), are also externalist, since their concept ascriptions advert to phenomena outside an agent's skull. However, even though that amount of reliance on the external world can be problematic, I think Millikan rightly senses that appeals to merely possible external phenomena are not really all that seriously externalist. After all, dispositions generally might well be regarded as internal conditions: although solubility is identified by what a substance would do in water, the facts about the actual world on which ascriptions of solubility depend are in fact entirely internal to the substance. Salt would be soluble in water even in a world in which water never existed. Similarly, on a dispositional (e.g. indicator) semantics, someone might have the concept unicorn even in a world in which there were no unicorns. The view Millikan seems to want to defend is much more radically externalist in that conceptual content depends upon historical facts in the actual world.

2. Millikans

Externalism

A presumption and, I think, a problem for currently fashionable "externalist" theories of concepts is the idea that people are largely on the right track about the world. This presumption appears first in quasi-externalist theories, like Quine's and Davidson's, which incorporated Neil Wilson's "Principle of Charity." As Davidson (1973/84) so touchingly put it: In our need to make [a person] make sense, we will try for a theory that finds him consistent, a believer of truths, and a lover of the good.—(p. 253)

I must say, I've always found this idea implausible (see Rey (2002) for discussion). Unlike Davidson, I am impressed by how really bizarre most people's views are, from religion and ethics, to psychology, medicine, politics, history, physics, statistics, mathematics. They seem often more like the early-morning thoughts of the White Queen than like the considered views of Einstein (consider the ordinary concept, [person], and its apparent commitments to a virtually incoherent freedom, identity and possible disembodiment). Now, Millikan may not agree with Davidson that it's always truths that we track; but her externalism does seem to lead her to presume that the contents of our thoughts involve real external properties, and it's this presumption that I find as problematic as Davidson's. Millikan writes (I italicize the real world commitments): It is not the purposes of individuals, but the biological functions—the unconscious purposes—of their inborn concept-tuning mechanisms that connect substance concepts with certain extensions. ... My claim then will...be that the extension of a substance concept is

whatever substance in the world it is the job ofthat particular concept, given its particular phylogenetic and ontogenetic history of development, to be re-identifying or conceptually "tracking."—(CCI, pp. 49-50)

Millikan's (Un?)Compromised Externalism

349

[T]he cognitive systems are designed by evolution and tuned by experience to find real world substances, not randomly logically possible ones. Close attention needs to be paid to the details

of real world ontology, to the principles that hold real substances together, and the relevance of experiments using artificial objects and kinds should be carefully j u s t i f i e d . — ( C C I , p. 8 3 )

I very much doubt that evolution "designed" (or selected) cognitive systems for any such thing—Millikan certainly doesn't present any evidence that it did. But it's a little hard to see why people interested in concepts should pay "close attention" to the external world independently of what the subjects of psychological research may happen to think, understand, or discriminate, advice that Millikan presses with surprising explicitness: [ U n d e r s t a n d i n g of the ontological principles...that ground a substance can certainly help in tracking it... but [psychologists] have not been clear that understanding o f this sort is not necessary to having a concept o f a substance, and that having or lacking this understanding

need make no difference to the extensions of one's substance concepts.—(CCI, p. 82, emphasis mine)

Millikan seems to be suggesting that, instead of muddling around with the odd ideas rattling around in their subjects' heads, psychologists should roll up their sleeves and assign concepts and extensions to the subjects' thoughts by doing, say, some serious chemistry—or perhaps string theory?—learning "the principles that hold the real world substances together." This seems more than a little presumptuous about the work of psychologists (who might understandably balk at the advice), but also about the minds of these subjects: should psychologists really override even the opinion of, say, an expert ornithologist about what he's talking about, should the selectional history of one of his concepts, say, [hawk], turn out to have involved something else? Similarly, it seems to me pretty wild to credit small children wanting hamburgers with concepts of certain proteins, even should those proteins have played a crucial role in their evolutionary history. Again, I don't see why we should suppose, ä la Davidson, that subjects are by and large thinking about the real properties of the world. 2 Of course, as Kripke (1972/80) and Putnam (1965/75, 1975) reminded us, it's true that people often times defer to the world in determining the defining conditions of the things they are referring to. But even that fact is an interesting internal feature of some deployments of some terms and concepts in some circumstances; which deployments, when and by whom, being a matter of the internal investigation of people's minds. 3 Kripke, Putnam—and really even Millikan herself—are not really basing their conclusions only on, e.g., chemistry. As Kripke (1972/80) rightly points out, what's crucial in

2

3

It is interesting to note just how widespread this externalist presumption is these days. Not only does it seem to be presumed by philosophers as otherwise different as Davidson and Millikan, but even among philosophers one would have thought were opposed to the views of these figures and to those of each another: for example, Fodor (1990, 1998) and Bonjour (1998) both seem to assume that there's a real property for at least every primitive concept (see Rey (2001) for discussion). Thus, as many have pointed out, asking for water in the fridge isn't asking for H 2 0 . What the KripkePutnam observations show is that we have available a use of many terms to describe the world independently of our interests and perspectives. But we manifestly don't always exploit that use. It seems to me an entirely open question whether our evolutionarily crucial ancestors did.

350

An Exemplary Debate about Content

his discussion relies on his linguistic "intuitions"—the chemistry is simply what those intuitions point him to (in at least the case of substance terms). And, whether or not those intuitions reflect cognitive procedures that were evolutionarily selected, the point of considering artificial objects and kinds is to determine precisely the contours of those intuitions and procedures. I respect the possibility, however, that Millikan might have an argument that could override initial intuitive judgments. What argument does she provide?

3. Millikan's Argument I take Millikan's argument for her externalism to be as follows: (1) conceptual possession is ability possession (2) ability possession is due to historical learning/selection. therefore (3) conceptual possession is due to historical/learning selection. Although Fodor (1998, chapter 1) has challenged premise (1), I suspect I join many in thinking the jury's still out on it. It's premise (2) that is far more problematic. It is presented as a piece of philosophical analysis of the notion of an ability, and her argument for it is that there is no way to distinguish a genuine ability from an accident: In the case of innate abilities, no matter what dispositions a mechanism happens to have, what determines its abilities is what it was selected for doing... To know how to do something as a result of learning, one must have a disposition to succeed in doing it under the conditions one learned under that afforded previous successes, and this learning or tuning must have been of the kind the learning mechanisms were selected for doing. Otherwise one does not know how, but does what one does by accident.—(CCI, p. 63)

But I can't see that there mightn't be a lot more alternatives to consider. Reflection on the "abilities" of, say, plants to withstand novel environmental pollutants, porpoises to be trained to do circus tricks, and people to follow mathematical arguments, does rather suggest that many abilities far outstrip whatever was selected. Moreover, people who learn to do something under particularly arduous conditions—say, how to clean a rifle in the midst of a battle, or how to swim by having been thrown into a swelling sea—don't seem to lose these abilities if they could no longer perform them under those arduous conditions. Or consider an evolutionary example: the ability to run may well have been selected for the advantage to outrun certain animals, at least to the nearest tree. I very much doubt I would succeed at doing that. But I have the ability to run nonetheless; I really do. And I presume it's no accident that I do. The explanatory story is likely just far more indirect than Millikan insists it must be, involving, for example, spandrals and/or the way the trait occasioned by some mutation may be complexly related to what happened to be required at a crucial point in evolutionary history. In any case, the important point is that abilities can be identified and studied in considerable detail in complete ignorance of their selectional history. For spectacular

351

Millikan's (Un?)Compromised Externalism

example, Chomsky and others have produced impressive evidence that grammar is largely unlearnt, and, moreover, given its many peculiarities, has speculated that it may very well not have been selected. Conclusive evidence in this latter regard, of course, isn't available either way (and maybe never will be; see Lewontin 1990); but it certainly wouldn't serve Millikan's "naturalistic perspective" (CCI, p. xii) to claim, a priori, that it was selected. But if it wasn't, would the ability to understand and produce language turn out to not be an ability after all? Surely it would hamstring science to hold our abilities hostage to such selectionist fortune.

4. Selectional Explanations that Presuppose

Content

Selectionist fortune can work in devious ways, and it's important to see how it could easily undermine Millikan's proposal. Modulo Lewontin's scepticism about knowledge in this area, it could turn out that some of our zanier concepts, like that of [god], [devils], [angels] and the like, if they were selected for anything at all, were selected to be keep people happy and obedient (they certainly aren't there to track anything real). But, although happiness and obedience may have been the "social meaning" of religious concepts, these roles don't begin to provide them with their semantic content, and for a quite straightforward reason: these social meanings would standardly be explained by their semantic content and not vice versa. It's because [god] has the semantic content it does that people are kept happy and obedient; it does not have the semantic meaning [happiness and obedience] because it got selected for having those effects. Although there might well be some examples that don't work this way, the point seems to me to be pretty general, and raises a fundamental, what I will call (speaking roughly) a "circularity" problem for any teleosemantics. It was admirably illustrated by Paul Pietroski's (1990, 1992) example of the kimus and the snorfs that Millikan discusses both in her (1990) and in the present volume (CCI, p. 235). Pietroski's example and his discussion of it is quite rich, but he doesn't focus specifically on my moral, and, to do so, I'll fill in some further inessential details: The kimus live on a plain near a rocky hill. Their only predators are blood-sucking s n o r f s — extremely tiny insects that inhabit the plain but not the higher hill. But, alas, the snorfs are too tiny for the kimu's poor sight, and so they die after being bitten without knowing what hit them. But one kimu, Jack, manages to be born color-sighted, i.e. he is disposed to enter internal state, B, in the presence of red light, and he is specifically caused to enter this state as a consequence of the dawn's early light. He develops a fondness for various red things, and this causes him to run up the hill and fortuitously avoid the daily predations of the snorfs. In time, all the other kimus are killed o f f by the snorfs; only Jack and his descendants survive and proliferate, by virtue, it w o u l d appear, o f their sensitivity to and preference for red light.

However, according to Millikan, this isn't quite right. Standing fast to the consequences of her theory, she insists that what Jack had a preference for was not "red," but for "fewer snorfs this way" (Millikan 1990), or "the snorf -free direction" (CCI, p. 236); i.e. the state Β has contents involving [snorf], not [red], since it was the lack of snorfs that in the circumstances was differentially responsible for Jack's survival and proliferation.

352

An Exemplary Debate about Content

But saying that Jack survived the snorfs because he became sensitive to the snorffree direction is patently shallow explanation, one, moreover, for which there is no independent support (hence the appearance here of a "circularity"). Although Jack is demonstrably color-sighted (which, pace Millikan's concern at CCI, p.236, needn't involve qualia), he, like other kimus, is completely insensitive to the presence of snorfs. They are too small for him to see, and he doesn't have concepts of [things too tiny to see] to help him out. Especially in the case as I've filled it out, we have absolutely no reason independent of Millikan's theory to suppose that he has any representation whatever with the content [snorf]. To the contrary, we have independent reason to think he doesn't. Experiments trying to condition him to the presence of snorfs would persistently fail. Aside from Millikan's reading of the case, there would be no evidence that he is capable of actions, judgments, preferences, reasonings involving snorfs (he wouldn't, for example, flee in the presence of snorfs as, let us suppose, he does in the perceived presence of other non-kimu animals). Jack patently hasn't the faintest idea about snorfs, any more than he does about amino acids, despite the role of snorfs— and amino acids—in his evolution. Moreover—and this is the point I want to stress here—our explanation of why the plain-dwelling kimus perish relies essentially on this perceptual and conceptual insensitivity to snorfs: were they able to detect the (let us suppose) languorous snorfs, they would readily swat them dead—and this counterfactual is true also of Jack and his descendants, despite their being color-sighted. On the other hand, our explanation of Jack's success does depend upon crediting him with the content [red], both in his perceptions and in his preferences. Unlike the [snorf] content, the [red] content would enter into a multitude explanatory patterns: Jack could be conditioned by the presence of red things; his preference for red things could lead him to be attracted also by apples, roses and pomegranates, and to avoid red birds and worms; and his navigations might well be aided by keeping track of red landmarks (to simplify the discussion, let us suppose that these patterns were never essential to his survival and so didn't develop "functions" of their own). And there is a straightforward explanation of how he acquired this new concept: his retinas became red sensitive—a fact that in any but this accidental context has nothing whatsoever to do with snorfs. 4 In any case, to repeat the point I made about [God]: the fact that state Β means [red] explains the role it plays; the role it plays doesn't explain its content. Indeed, it

4

In reply to this point, Karen Neander (personal communication) raised the possibility that the explanation could go a different way: color sightedness might initially provide Jack not with a state that meant [red], but merely with a state that sufficiently co-varied with (the property) red to cause Jack to run up the hill; only after this covariation was exploited for some survival sensitive purpose, such as avoiding snorfs, would it come to mean [red] (since, for a teleosemanticist, only then would there be a possibility of error). I agree that this would, strictly speaking, seem to be a way to avoid my "circularity" objection. But it would entail that the initial point at which the state merely indicates red isn't a content state for Jack, and so consequently it wasn't his preference for red things that led him up the hill. But then how did the red covariation get him moving? I suspect that this explanation is really just a verbal variant of Pietroski's, showing either that a weaker, mere "co-variant" notion of 'means' is adequate for attitude content—or that the insistence on selection is otiose; for note that, on this account, what the state indicates and subsequently means is still [red], not [snorf-free].

Millikan's (Un?)Compromised Externalism

353

would appear that there is an obvious trade-off between teleosemantics and successful selectionist explanation: the more a state is constituted by the role it plays, the less likely it would be that one can seriously explain its presence by its having been selected to play that role: if there were a state of being adapted to environment e, it would hardly be explanatory of something's being in that state that it permitted it to adapt to environment e. One needs an independent take on the state. Just so: one needs a take on a semantic state independent of the role it plays if the state is to play a role in a selectionist explanation. In this connection it is interesting to note that Neander's (1995) version of teleosemantics avoids this problem, constructing it so as to go hand in hand with a Cummins-style functional analysis: it is by representing black dots that a frog detects flies, and by detecting flies that it detects food; similarly, it's by representing [red] that Jack survives and proliferates; and it's by failing to represent snorfi that his conspecifics perished. I'm not convinced by Neander's account either, but it seems to me at least to recognize how, at a fairly early stage of selectional explanation, ascription of content may well need to precede ascription of function, and so not be identified by it. In reply to Pietroski's example, Millikan (CCI, p. 236) offers an analogy with tortoises, whose attraction to green things takes them to edible vegetables, with which greenness happens to be accidently correlated. Millikan asks: Can the green mean "chow over there" to the tortoise? Obviously not in so many words. But your percept of an apple doesn't mean "there's an apple over there" in so many words either. If the green didn't mean chow over there to the tortoise, then what on earth could mean chow over there to anyone?—(CCI, p. 236)

She doesn't pause for an answer to this last rhetorical question, but I think she should have. For starters (and putting aside natural (or indication) meaning as irrelevant here), animals like us pretty clearly have a concept of [chow over there] as distinct from [green] since, e.g., we can readily tell them apart, and think green is a color while chow is what you eat. None of these further abilities, we may suppose (by analogy with the kimus), is available to the tortoise that merely lumbers towards all and only green things and eats them. I'm not saying these further abilities are constitutive of having the concept chow. I'm very prudently not in the business here of trying to provide any theory of content. I'm only answering Millikan's rhetorical question. That is, my complaint with Millikan is not that I believe that content is constituted by discrimination abilities, information, ideal co-variation, conceptual role or messages from God. I, myself, think it's premature to be trying to advance a serious theory of intentionality before we have an understanding of the diverse theoretical work intentionality will need to do, in particular, of the specific kinds of psychological mechanisms in which it will figure (imagine a theory of force prior to an understanding of physics). I'm only calling attention to what some seems to me some pretty strong prima facie evidence about some work it would need to perform in, inter alia, the very sort of selectionist explanations that Millikan favors. The example of the kimus and the tortoises again raises the issue of whether animals are to be regarded as tracking real properties and substances. There are those who are

354

An Exemplary Debate about Content

reasonably skeptical that the concepts [red] and [green] actually correspond to any real properties. And Millikan herself seems similarly skeptical about [chow]: 5 Is it really plausible that there could be a genuine informationL channel open to us any of us...that would communicate the informationL that there was chow on the table? Does human chow, as such, figure in any causal laws?—(CCI, p. 236)

Now perhaps neither chow nor green figure in any serious causal laws—other than specifying what's picked out by creatures like us with concepts of [red], [green] and [chow]. Leave aside the (to my mind unclear) issue of whether these could serve as contents even for Millikan (who, at CCI, pp. 196-7, seems to claim that content properties had better appear in laws). W h a t concerns me is merely the possibility that both the kimus and the tortoises are representing unreal properties. As I said earlier, this seems to me a commonplace about human cognition, particularly any that is supposed to accommodate the likes of the White Queen. So how is Millikan to deal with such concepts?

5. Vacuous Concepts Vacuous concepts are the bane of any externalist view, and so it is not surprising that Millikan gives them short shrift: there can also be, of course, semi-opaque descriptions of various "would-be" means [of identifying substances] but that fail to support real substance concepts. That is, unbeknownst to their possessors, they are not the means for any real recognition abilities. These substance concepts are "empty" or, more accurately, they are not substance concepts at all. An ability that is not an ability to do anything is not an ability at all. Empty substance concepts result from failures of the mechanisms designed develop substance concepts. They are "concepts" only in that their biological purpose was to have been concepts.—(CCI, pp. 175-6)

But, whether or not they are to be counted as bonafide "concepts" for Millikan, one would have thought they have semantic content, since they manifestly enter into all manner of intentional interactions in a persons mind—e.g., into their thoughts, desires, (delusional) perceptions, emotional reactions. Millikan seems uneasy about granting this. In the next sentence, she admonishes us: Nor should they be confused with concepts we merely pretend to have, such as the concept of Santa Claus after we are grown.—(CCI, p. 176)

Speaking for myself, I don't think / am "pretending" to have a concept of Santa Claus when I talk about him, or deny his existence 6 . But fine; maybe self-conscious fictions

5 6

"InformationL" for Millikan is, roughly, information one state of the world, B, carries about another, A, as a result of a natural law, "A => B", but "purified of all mere frequencies" (CCI, p. 223). I'm not sure exactly what Millikan has in mind by this remark, but I presume she is alluding to Evans' (1982, chapter 10) curious view that in negative existentials we are only pretending (for half the sentence) to have the concepts whose extension we are denying. Evans didn't live to work this view out in plausible detail, but in any case I thought he was concerned, both there and elsewhere in that book largely with singular terms/concepts, not concepts in general.

Millikan's (Un?)Compromised Externalism

355

are a special case. But [God] and [angel] et al are not intended as fictions by their standard users, and surely they are still contentful, even if extensionless. Millikan says no more in the present book about such cases, but in her (1984), she denied they have sense: Consider for example a vacuous proper name—'Pegasus'. It undoubtedly has an intension. But a vacuous name cannot have a sense. For it cannot have a history of having performed its proper functions due to having mapped onto something in the context of sentences; e.g., "bewitch', 'bewitched' and ' dephlogistonated' lack sense.—(1984, p. 112)

But, again, they still have some sort of content·, so apparently this must be their "intension." In the present book she replaces this (indeed, confusing) term, with 'conception (CCI, p. 12, fn. 6), and seems to have in mind "ways one has of identifying" something (1984, pp. 111-2,148; CCI, p. 12). 7 So the semantic content of a vacuous concept is apparently the way—or wayr?—one has, or would have, of identifying something as satisfying it. I'm not sure I have a good grip on these identificatory ways in such cases (how would ont identify angels??), but, whatever they are they certainly don't seem external. And if they're not, then it is hard to resist asking: if such internal conditions suffice for the semantic contents of vacuous concepts, why don't they suffice for the non-vacuous ones as well? In particular, if we can frame content explanations using conceptions in the vacuous cases, it's hard to see why we shouldn't be able to do so in the non-vacuous ones, especially for processes (e.g. reasoning) which don't depend upon actual extensions. Reflection on the content of vacuous terms would seem to land us squarely in the court of internalism.

6. Some Internalist

Compromises

For all her "uncompromising externalism", I was surprised to find Millikan, herself, recognizing the need of some substantial internal conditions. Note for starters that, in her announcement of her "uncompromising externalism," she does write: 7

In correspondence since the original presentation of this paper, Millikan claims that I am misinterpreting her here. Contrary to this 1984 view just quoted, she says she wants to claim that empty terms are not "real terms," that they are like "an ability that is not an ability to do anything," that it "makes no sense to talk of a mode of presentation that doesn't present anything," and that, indeed, when reference "fails utterly there is no thought at all." This does seem an extreme generalization of the Evans view about empty singular terms, noted in my previous footnote, and one much more open to objection than the view I consider in what follows. In the first place, it makes perfect sense to talk of a mode of presentation that doesn't present anything: Evans' unusual view to one side, mistaken thoughts about imaginary things—and true, negative existential ones denying them—are a commonplace. But, more importantly, if apparent "thoughts" with empty terms are not really thoughts at all, then how are we to explain the often rational, content sensitive behavior of people who seem to have them? Ponce de Leon arranged an elaborate journey to Florida; people pray, make sacrifices, and engage in often elaborate reasonings about gods, devils and angels; and astronomers once intelligently sought evidence for the (non-existent) ether, and the planet Vulcan: in all such cases, empty thoughts interact in myriad inferential ways with nonempty ones (e.g. about churches, Florida, light or Mercury), ways that surely require the empty ones to possess some kind of intentional content. How else to make sense of, e.g. "evidence"?

An Exemplary Debate about C o n t e n t

356

...an uncompromisingly externalist view. What makes a thought be about a certain substance is not merely in the mind...—(CCI, p. 95, emphasis mine) suggesting that some internal conditions might also be important. More explicitly, she writes: It should not be thought that concepts of properties are always involved in conceptual tracking of substances. Having concepts of properties, I am assuming, would be to represent properties, as such, in thought. The thought of a property is not just a reaction caused by a property; it must play an appropriate conceptual role.—(CCI, p. 78) Conceptual roles are about as internalist as one gets these days, and the questions that immediately arise for them are "Which ones? Why?" Millikan directs us to §7.4, where she adds (I italicize the requisite roles): 8 Acquiring concepts of these various substances fsuch as gold, frog\ must involve some understanding of which predicate contrary spaces are correlative to them, that is, of the "substance templates" under which they fall... such as person, animal, animal species, plant, plant species, mineral, and so forth.—(CCI, p. 108) Indeed: Having a concept ofa substance requires a grasp of what kinds of things can be learned about that substance. It requires understanding from which kinds of experienced practical success to generalize to new encounters with the substance, or, if the concept is usedfor gathering information, it requires understanding what sorts of predicates will remain stable over encounters with the substance, that is, what some of the meaningful questions are that can be asked about the substance. You can ask how tall Mama is, but not how tall gold is... Having a concept of a substance does not involve knowing an essence. Rather, it involves understanding something of what recognition of the substance might be good for, in the context of either developing practical skills or theoretical knowledge. To have the concept of any individual person, you must know what kinds of questions can be asked and answered about individual people; to have the concept of any individual species, you must know some of the questions that can be asked about species... The primary interest of groupings like persons, species, and chemical elements is not that they themselves correspond to substances, but that they bring with them "substance templates."...—(CCI, pp. 9-10, italics mine) WTaich sorts of questions are relevant? T h e only examples of these templates that Millikan provides are of some pretty technical chemistry: all chemical elements have, for example, some atomic number or another, some particular chemical combining properties or others, some electrical conductivity or other.—(CCI, p. 10) If these are the templates that come with concepts of chemical elements, the above passages seem to entail that someone has a concept of a chemical element only if they know to ask about its atomic number, combining properties, and electrical conductivity! O f f hand, I would have thought children and others (cooks, jewelers) unacquainted with modern

8

Millikan uses * substance' broadly to include "individuals..., stuffs...., and natural kinds" (CCI, p. 1). "Contrary spaces" seem to be the parameters over which a given substance can vary, e.g., size and shape in the case of animals (CCI, pp. 107-8).

Millikan's (Un?)Compromised Externalism

357

chemistry have perfectly good concepts of salt, sugar, gold, silver, and so forth, without the slightest inkling of these sorts of questions. Knowing the right sort of questions to ask about substances is part and parcel of learning the right theory about them, and, again, I see no reason to assume with Davidson that people are on the whole on to the right theories of very many things at all. Note that its not merely some social practice of ascription that is at stake here. I would have thought the explanation of many of these peoples behavior would involve ascriptions of chemical elements in the absence of their knowing the right questions to ask (consider, for example, the position of the inquiring student or chemical researcher, or the positions of the ancient Greeks who thought fire was a substance and stars, holes in the heavenly canopy). In any case, one can't just impose such strong conditions without a suggestion of an argument. But I'm afraid I can find none. It's almost as though Millikan thinks that, once one does go inside, it doesn't much matter what one says. Of course, the usual way to argue for or against such requirements has been by thought experiments: thus, many say that truth is a condition on knowledge, since the idea of false knowledge seems contradictory. But would it be a contradiction for someone (for example, some modern proponent of alchemy, who thought chemical theory was hooey) to think that nothing had atomic numbers? Millikan, however, is not friendly to this traditional method of reasoning: I will show no interest...in what a person might be "credited with" referring to, or thinking of, or having a concept of, and so forth in possible worlds cases, or even in queer actual cases. Such questions rest, I believe, on false assumptions about the kind of phenomena that reference and conception are and tend to be philosophically destructive.—(CCI, p. xiii)

Unfortunately, this seems to be due to her thinking that analysis is somehow incompatible with vagueness: if we are working with biological phenomena, then we are working in an area where the natural divisions are divisions only de facto and are often irremediably vague. These divisions do not apply across possible worlds; they are not determined by necessary and sufficient conditions.—(CCI, p. xiii)

I want to lodge as strong an objection against this fallacious reasoning as polite company will tolerate. It apparently can't be emphasized enough—Dan Dennett is always making this mistake as well (see, e.g, his 1991, pp. 107,119,126,275)—that the viability of analyses into necessary and sufficient conditions, and even "essences," is wholly compatible with vagueness. All that's required is that an analysis match an analysandum "umbra for umbra, penumbra for penumbra", as Quine (I960, p. 41) so eloquently put it. Putting vagueness to one side, it's a little hard to see exactly what's wrong with the traditional method of searching for defining conditions, and testing them against counterexamples, sometimes far flung. As I and others have emphasized elsewhere, the activity is rather like the activity of linguists when they consult native speakers' intuitions about what one can and can't say in a language—or when they test a syntactic rule by consideration of equally far-flung syntactic constructions (e.g. multiple centered embeddings). The procedure is continuous with the kinds of more controlled interviews psychologists like Carey (1985), Keil (1987) and Atran (1990) administer to children and other people to learn about conceptual diversity and development.

358

An Exemplary Debate about Content

7. Responsibilities

of

Internalism

Of course, insofar as Millikan does opt for some internalist conditions, she faces a number of difficulties. She will owe us a more plausible account than she has provided of just which specific "conceptual roles", or "conceptions", or ways of telling, are genuinely essential and part of a concept's semantic content, which ones any two people must share (or one person maintain over time) in order for them to be using the same concept. And, in doing this, she will need to meet Quine's challenge to provide some principled basis for deciding such issues. Indeed, Millikan would seem to be very much in the position in which she finds Neander (1995) who, we observed earlier, bases her teleosemantics on a "by" relation in addition to direct selectional mechanisms. Millikan rightly worries (CCI p. 231) that such a position threatens Neander with a verificationist reduction of distal representations to the proximal ones "by" which the distal are applied. However much Neander's program has this implication, certainly Millikan's reliance on "ways of identifying" does, both for vacuous concepts, but then, by parity of reasoning, for the rest of people's substance concepts as well. It's not hard to sympathize with the impulse to avoid these problems by, so to say, retreating to the external world. However, as I said, I worry that Millikan's retreat to externalism leads her to not take these lingering internalist problems seriously, and to not pay the kind of "close attention" to inner mechanisms that she thinks psychologists ought to pay to outer ones (CCI, p. 83). At any rate, it seems to me that her claims that the language of thought couldn't handle identities (CCI, pp. 166-7)—and that there really are no judgments of identity at all (CCI, p. 172)!—suffer from a similar failure to consider internal workings in detail. 9 As in the case of her proposed internalist conceptual requirements, Millikan's moves here seem to me far too quick and unargued, and I commend to her much of the traditional work that has been done on the logic of identity in a formal system. But here my recommendations outrun the available space to make them.

8.

Conclusion

I have argued that Millikan's selection based "externalism" faces serious explanatory problems: it threatens to deprive experts of tracing the real properties they understand, and to credit too many creatures with contents of which they haven't a clue. It also gets the order of explanation wrong: for many representational states, their selectional role is determined by their content, not their content by their role. And, last but not least, it deprives us of a means of dealing with the psychology of creatures, like the White Queen and the devoutly religious, who make a habit of thinking vacuously. Dealing

9

For the life of me, I can't make out the argument at CCI, pp. 166-68 that a language of thought would have to treat all terms in identity statements as tokens of the same type (what about the role of such terms in oblique contexts?), or why identity judgments couldn't involve both abilities and explicit representations that play a role in mediating inference.

Millikan's (Un?)Compromised Externalism

359

with these difficulties would appear to require making at least some substantial compromises with internalism. As we've seen, Millikan does make some passing suggestions of such compromises; but she doesn't argue for the specifically strong conceptual conditions she suddenly imposes. They seem rash, as, I think, do her other remarks about internalist resources. I submit that psychology would on the whole be better served by thinking more carefully about what's presently going on inside a mind instead of what long ago went on outside of it, particularly given that the latter is lost in the probably forever unknowable evolutionary history of our concepts. Acknowledgments: A shorter version of this paper was presented in an "Author Meets Critic" session at the March 2002 Pacific Meeting, in Seattle (WA), of the American Philosophical Association. Other "critics" were Louise Antony and Kristin Lawler. I'm grateful to them, as well as to Dan Blair, Paul Pietroski and especially Karen Neander for comments and useful discussion.

References Atran, S. (1990). Cognitive Foundations of Natural History: Towards an Anthropology of Science, Cambridge University Press. Bonjour, L. (1998). In Defense of Pure Reason, University of Cambridge Press. Bürge, Τ. (1979). "Individualism and the Mental", Midwest Studies in Philosophy, IV, pp. 73-121. Bürge, Τ. (1986). "Individualism and Psychology", Philosophical Review, XCV 51, pp. 3-46. Carey, S. (1985). Conceptual Change in Childhood, Cambridge (MA): M I T Press. Dennett, D. (1991). Consciousness Explained, Boston: Little Brown & Co. Dretske, F. (1995). Naturalizing the Mind, Cambridge (MA): M I T Press. Fodor, J. (1998). Concepts: Where Cognitive Science Went Wrong, Cambridge: MIT. Keil, F. (1987). "Conceptual Development and Category Structure," in Neisser, U. (ed.) Concepts and Conceptual Development: Ecological and Intellectual Factors in Categorization, Cambridge University Press, pp. 175-200. Kripke, S. (1972/80). Naming and Necessity, Cambridge: Harvard University Press. Lewontin, R. (1990). "The Evolution of Cognition", in Osherson, D. and Smith, E. (eds.) An Invitation to Cognitive Science, vol. 3, Cambridge: M I T Press. Lycan, W. (1987). Consciousness, Cambridge (MA): MIT. Millikan, R. (1984). Language, Thought and Other Biological Categories, Cambridge: M I T (Bradford). Millikan, R. (1990). "Clarifications on Language, Thought and Other Biological Categories", Annals of Scholarship, 7, pp. 147-9 Millikan, R. (1993). White Queen Psychology and Other Essays for Alice, Cambridge: MIT. Millikan, R. (2000). On Clear and Confused Ideas, University of Cambridge Press. Neander, K. (1995). "Misrepresenting and Malfunctioning", Philosophical Studies, 79, pp. 109-41. Pietroski, P. (1990). "Meaning and Evolution", Annals of Scholarship, 6, pp. 318-30. Pietroski, P. (1992). "Intentionality and Teleological Error", Pacific Philosophical Quarterly, 73, pp. 767-82. Putnam, H., (1965/75). "Dreaming and Depth Grammar", in his Phibsophical Papers, vol. 2, Cambridge University Press. Putnam, H. (1975). "The Meaning of 'Meaning"', in his Philosophical Papers, vol. 2, Cambridge University Press. Quine, W. (1956/76). "Carnap and Logical Truth", in his Ways of Paradox and Other Essays, 2 n d ed., Cambridge, MA: Harvard University Press. Rey, G. (1992). "Semantic Externalism and Conceptual Competence", Proceedings of the Aristotelian Society, pp. 315-333.

360

An Exemplary Debate about Content

Rey, G. (1995). "Keeping Meaning More in Mind", Intellectica, 2-21, pp. 65-80. Rey, G. (1998). "A Narrow Representational Account of Qualitative Experience", i n j Tomberlin (ed.) Philosophical Perspectives 12. Language Mind and Ontology, Atascadero, Ridgeview Press, pp. 435-57. Rey, G. (2002). "Physicalism and Psychology: a Plea for Substantive Philosophy of Mind", in Carl Gillet and Barry Loewer (eds.) Physicalism and Its Discontents, Cambridge University Press, pp. 99-128. Rey, G. (2001). "Digging Deeper for the A Priori". Commentary on Laurence Bonjour, in Defense of Pure Reason, Philosophical and Phenomenological Research Tye, M. (1995). Ten Problems of Consciousness: a Representational Theory of the Phenomenal Mind, Cambridge (MA): M I T Press.

Comments on "Millikan's (Un?)Compromised Externalism" R U T H GARRETT MILLIKAN

Georges Rey's commentary on On Clear and Confused Ideas; An Essay about Substance Concepts contains what seem to me to be many misreadings and misunderstandings of my position in O C C I . I will try to straighten out some of these, hoping that others will straighten themselves out for readers who approach the book independently and without prejudice. Concerning the title themes in O C C I —what "substance concepts" are (the reference is to Aristotle; both his primary and his secondary substances are included) and what "confused ideas" are—Rey is silent. Interested readers will have to turn to the book itself for enlightenment. A virtue, I believe, of the definition of function ("proper function") that I have employed in many of my papers and books is that this kind of function depends in no way on what happens on the average. Davidson's stand that people must mostly believe what is true is not merely more extreme than my position. It is directly antithetical to my position (Rey, §2). I have reiterated many times over the years that there are lots of items that have proper functions that they very seldom perform. So my claim that the proper function of the mechanisms of concept formation in acquiring substance concepts is to acquire concepts of real substances has no bearing on how often they succeed in this. Indeed, empty substance concepts, as I explain them, are exactly those substance concepts that don't correspond to anything real. In his §2 Rey seems to think I don't allow for enough of these concepts; in §5 he objects to my allowing any. (To avoid confusion, perhaps I should mention that I nowhere discuss "properties" in O C C I . Properties are not substances. O C C I actually has nothing whatever to say about whether or not any of our apparent concepts of properties are of real ones. But we can let that pass.) The claim that "understanding the ontological principles... that ground a substance" is important for psychologists studying substance concepts but not necessary for the possessor of a concept to have in order that their concept have a definite extension (Rey, §2) is reflected in the experimental work, for example, of the psychologist's Medin, Gelman, Keil, Gopnik and Meltzoff (cited in O C C I p. 81-82). They have shown that both adults and children appreciate that there must be some general underlying ontological principles forming each of the various kinds of substances into a unity, but without understanding exactly what these principles are. And these psychologist's have themselves speculated on the form these underlying principles take. I claimed that psychologists do indeed need to think about the ontology of various kinds of substances, as I have done in O C C I Chapter Two. To confuse doing basic ontology with doing chemistry and physics is not my mistake but Rey's (Rey, §2). I claim, for example, "... dogs are quite good at recognizing their masters, and babies at recognizing their mothers, even though it is quite certain that neither conceives of a criterion of identity for

362

An Exemplary Debate about Content

persons over time" (OCCI p. 75). I do not think that claims of that sort encroach on the turf of any particular science. On the other hand, it is true that my claims about the ontology of biological species are taken directly from modern biological theory. That is because biologist's themselves have been doing some ontology. That we need to turn to post Darwinian biology to discover what holds biological species together is another nice example of the fact that a layman's substance concepts need not depend on a correct grasp of the basic ontological structures supporting those concepts. (Perhaps I should mention here (c.f., Rey last para of §2) that I have not used in OCCI any of the kinds of, yes, "internalist" arguments that were employed by Putnam and Kripke. For a discussion of the central importance of that matter, see my "Cutting Philosophy of Language Down to Size".) Preparatory to remarking on Rey's §3 on "abilities," I should explain that the following quotation (Rey §5) was not introduced as an argument against the program of conceptual analysis: If we are working with biological phenomena, then we are working in an area where the natural divisions are divisions only de facto and are often irremediably vague. These divisions do not apply across possible worlds; they are not determined by necessary and sufficient conditions.

The passage was presented as a characterization of the way in which I proposed to discuss certain phenomena such as substance concepts, abilities, and reference, all of which I assimilated to biological phenomena. The emphasis was meant to be on the de facto nature of the natural divisions involved, not on their vagueness. The idea was that I would suggest theoretical definitions of these phenomena rather than describing usage. In line with this, Chapter Four on abilities is not a "philosophical analysis of the notion of an ability" (Rey §3) but a theory about the natural biological structures that underlie abilities. What's wrong with "the traditional method of searching for defining conditions, and testing them against counter examples..." (Rey §5)? What's wrong is that it assumes precisely what needs to be proved. The method is circular if the question is how meaning and reference are determined. For a careful discussion of this, see "Cutting Philosophy of Language Down to Size." A comment on Rey's reading of my thesis on abilities may be helpful. I point out in later chapters (not, however, in the basic chapter on abilities) that "ability" is not a count noun. You cannot count a person's abilities. Almost any ability of a person you can name rests on countless more general abilities, combined and recombined for various purposes. The history of any particular ability that a person has is very diffuse, and the selectionist history of the aspects that combine to make that ability stretches from the evolutionary history of each of the various kinds of learning mechanisms that underlie it through the most recent episodes of individual practice that have helped to perfect it. I have argued elsewhere, with care, that it a mistake to suppose that our abilities to do mathematics and such are not accounted for by natural selection, given that one describes those abilities in the right way (Millikan 2002). Rey ignores my characterization of "conditions one learned under," namely, "They are conditions necessary to completing the mechanisms by which past successes were reached by the systems or programs responsible for the abilities" (OCCI p. 61). That is, they are the

Comments on „Millikan's (Un?)Comromised Externalism"

363

"Normal conditions for proper functioning" of these systems or programs, as defined in Chapters O n e and Two of Language, Thought and Other Biological Categories (Millikan 1984). Rey does not discuss the arguments I give for the thesis on abilities. The reader may find that acquaintance with these arguments is a help in understanding the conclusions reached. The example of the Kimus (Rey §4) concerns a side issue that arose in Appendix Β of O C C I . Appendix Β is a reprinted paper originally written in another context for another purpose, included as an afterthought in O C C I because it contains a description of natural information that is useful in filling out the thesis on substance concepts. The Kimu example is an interesting, though very complicated, one and I would be glad to discuss it in another context on another occasion, but I think it not helpful or appropriate to do so here. None of the necessary background has been filled in by Rey, nor is it filled in in O C C I . Rey's claim that O C C I does not deal with sufficient care with the "internal workings" of concepts (§1) puzzles me very much. I don't know what"internal workings" of a substance concept would be other than what I call "conceptions." A very large proportion, indeed, I should think most, of O C C I is about what I call "conceptions," about how they are acquired, about exactly what relation they have to the extensions of substance concepts, about exactly what relations they have to words for substances, about exactly what relations they have to the "meanings" of words for substances, about how these conceptions are developed in individual ontogeny, about exactly what relations they have to Fregean senses, and so forth. But Rey has especially in mind the question of the "inner workings" of empty concepts, to which he turns in §5. Of empty concepts, Rey says "one would have thought they have semantic content, since they manifestly enter into all kinds of intentional interactions in a persons mind." Well, here Rey assumes exactly what needs to be proved, namely, that having a dispositional or causal role is sufficient to give a would-be substance concept a semantic content. A crucially central claim, argued for at length, running throughout O C C I , is precisely that dispositional or causal role is not a determinant of content. Surely Rey needs to do something more than just assert the contrary. Yes, O C C I is completely explicit many places that both equivocal substance terms and empty substance terms have conceptions attached, and that conceptions, in the case of valid substance terms, are ways of (re)identifying substances. But where has Rey shown that possessing inner mechanisms that were developed by learning devices/methods whose proper functions are to produce ways of reidentifying things, but that have failed in this task in the particular case, is possessing a term with content? That, of course, is exactly what is at issue. As to how one would identify angels, I am explicit in many places that one of the many ways things are identified is through descriptions. But of course no matter how many lovely descriptions one has in one's mind of angels, if there aren't any angels these descriptions won't correspond to any real ability to identify angels. That is precisely the point. Perhaps this what Rey is missing. Of course if you think that causal role determines content, causal role without content will seem a contradiction to you. One way crudely to put the basic thesis argued in OCCI, however, adding to it also the position on properties in Millikan 1984, might be to say this. There is no content that is not either (1)

364

An Exemplary Debate about Content

reference, or (2) whatever is reference's correlate in the case o f thoughts o f properties and relations (I don't mind speaking o f properties as "referents"), or (3) built up out o f prior more basic thoughts that have referents (as in the case, for example, o f concepts o f classes). Real kinds (these are one kind o f "substance"), O C C I argues, are not classes but are in important ways similar, ontologically, to individuals. They too can be said to have referents. T h e claim then is that terms for substances (and it does help to be told what substances are—see all o f O C C I Chapter Two) are as basic as any thoughts get. T h e y are not thoughts o f classes. They don't have definitions. They do not correspond to property clusters or to paradigms. They are not defined by their roles in theories. Thus empty thoughts o f substances don't have semantic content, though they do have causal roles. ( I f Rey understands this position but finds it objectionable, he does not tell us why he finds it objectionable.) T h e term "externalism" has had various uses. However, although I may have overlooked someone, but I can't think o f any philosopher who has termed himself an "externalist" who doesn't think that thoughts have some kind o f "cognitive" or "conceptual" or "causal" roles. What is "internalist" is not "conceptual roles", but supposing that conceptual roles determine content. Let me pull out o f Rey's long quotation in §5 the point I was trying to make: Having a concept of a substance requires a grasp of...some of the meaningful questions that can be asked about the substance...it involves understanding something of what recognition of the substance might be good for...To have the concept of any individual species, you must know some of the questions that can be asked about species; to have the concept of any chemical element, you must know some of the questions that can be asked and answered about chemical elements.

Please note the various "some's and the "something". T h e remainder o f the paragraph tells about some o f the questions that can be asked and answered about chemical elements. It does not say that you have to understand the relevance o f any particular ones o f these questions to have a concept of something that is a chemical substance. (Example: " W h a t color is it, what consistency is it, and what does it taste like?" are questions that can generally be asked and answered —though not necessarily safely—about things that are chemical substances in solid form.) That I give no other examples o f questions that can be asked and answered about other kinds o f substances is a serious misunderstanding. Another example, namely, questions that can be answered about Mama, is given within the very passage that Rey quotes. More important, a large section of chapter Two is about the various kinds o f stable properties substances in different ontological categories have, hence about relevant questions one can ask and answer about them. Other examples are given throughout the book as well. (I find Rey's reading o f my thesis on substance templates just as bizarre as he does!) It was my claim that, typically, the causal roles that correspond to thoughts o f the same substance vary greatly from person to person. T h e substance templates that different people use for the same types o f substances also vary greatly. Nor is anything about the causal role o f a substance concept for a given person analytic o f that concept or definitional o f the substance even for that person. Indeed, I follow Quine in denying the relevance o f the analytic/synthetic distinction in this context, so o f course I feel no obligation to explain which specific conceptual roles or conceptions are "genuinely es-

Comments on „Millikan's (Un?)Comromised Externalism"

365

sential and part of a concept's semantic content, which...any two persons must share... in order for them to be using the same concept" (Rey § 6). Indeed, my claim is explicit and loud that the only way in which people share "the same (substance) concept" is by each having their own conception of what is, in fact, the same substance (e.g., OCCI, p. 11-13). It is the substance that is the same across thinkers or speakers, not the conception of the substance. But perhaps what is troubling Rey, or might trouble someone, is not seeing how there could be words in a public language that express empty substance concepts. Empty substance concepts cannot be handed down from person to person by each learning separately how to identify the same nonexistent substance. The answer is that there is often a great deal of overlap among different person's conceptions of the same substance, and that large portions of conceptions of substances very often are handed down. In the case of empty public terms what is handed down is, of course, only conceptions. No semantic values are handed down. (There are interesting cases in which some people grasp this and others don't, as with "Santa Claus.") I don't know why Rey thinks I am not familiar with "the logic of identity," or what light he thinks "the logic of identity" could possibly shed on what happens in people's heads when they hear identity sentences. Neither, apparently, did Strawson, who originally proposed my model for what sentences asserting identity do to people's heads.

References

Cited

Millikan 1 9 8 4 Language, Thought and other Biological Categories (Cambridge MA: The M I T Press) Millikan 2 0 0 0 "Biofunctions: Two Paradigms" in R. Cummins, A. Ariew and M. Perlman, eds., Functions:

New Readings in the Philosophy of Psychology and Biology, Oxford University Press (2002) Millikan "Cutting Philosophy of Language Down to Size," Philosophy (July 2 0 0 1 ) . Reprinted in at the New Millennium, A. O'Hear ed., Cambridge University Press 2 0 0 1 , pp. 1 2 5 - 1 4 0 .

Philosophy

VI Self-Knowledge

Believing that You Know and Knowing that You Believe SVEN BERNECKER

External-world skepticism claims that we cannot rule out, on the basis of how things seem, a range of possibilities concerning how they really are, all of which are incompatible with what we take ourselves to know about reality. Arguably the most exciting line of argument against the skeptic is based on semantic externalism, that is, the view that mental contents are (partly) determined by factors external to the subject. Antiskeptical arguments from semantic externalism question the skeptical assumption that how things seem and how they really are can logically come apart. The idea is that if reality were entirely different—as the skeptic suggests—we wouldn't have beliefs or at least not the kind we do have. Anti-skeptical arguments from semantic externalism presuppose that we can know, just by reflection, that we possess beliefs. Given that having beliefs is a necessary condition for having a mind, and given that zombies are creatures devoid of beliefs and other propositional attitudes but who are functionally and phenomenologically indistinguishable from content-possessing beings, anti-skeptical arguments from semantic externalism assume that we can know non-empirically that we are not zombies. 1 This paper argues that anti-skeptical arguments from semantic externalism are inconsistent or, at best, ineffective, since non-empirically knowledge of whether we have beliefs or are zombies conflicts with semantic externalism. Sections 1 and 2 examine Hilary Putnam's brain-in-a-vat argument and an analogous argument by Fred Dretske and show that anti-skeptical arguments from semantic externalism presuppose that we can know non-empirically that we possess beliefs and thus aren't zombies. In section 3 I argue that, given semantic externalism, we cannot non-empirically know whether we have beliefs or are zombies. Section 4 spells out the consequences of this position for Putnam's and Dretske's anti-skeptical arguments.

I. According to semantic externalism, beliefs are in the head, but what gives them their relational content, what makes them about something, are the relations of these internal states to external affairs, namely to certain aspects of the believer's environment. The contents of an individual's thoughts depend on relations of the individual to certain

1

I am focusing exclusively on propositional zombies, leaving aside the issue of phenomenal zombiehood.

370

Self-Knowiedge

aspects of his physical and social environment. T h e dependency relation of internal states (e.g., brain states) on environmental states is construed as a nomic relation that supports counterfactuals. Putnam (1981, ch. 1) uses semantic externalism to argue that there is a certain sense in which we cannot be disembodied brains in a vat being fed electrical signals by a computer that is programed to stimulate our nerve endings as to induce in us experiences indistinguishable from the ones we actually have. Putnam's strategy is to show that there is a logical connection between certain external facts—that fact that we are not brains in vats—and certain internal facts—that fact that our utterance "We are not brains in vats" means that we are not brains in vats. Though brains in vats can say anything we can, their utterances cannot express the thoughts that our utterances express, because they lack the appropriate sort of causal contact with the things that we refer to. A brain in a vat is incapable of expressing that it is a brain in a vat because its words 'brain and vat' don't refer to what our words refer to. 'Brain' uttered by a brain in a vat refers to an aspect of a computer program and 'vat' means a vat-internaJ ersatz vat. So when a brain in a vat says "I am a brain in a vat", it is saying something false, for that is not what it means by 'a brain' in what it means by 'a vat'. Given that we are not brains in vats, the statement "We are brains in vats" is obviously false. And if we were brains in vats, then, since brains in vats are by hypothesis brains in vats, and these truth-conditions are incompatible with them being aspects of a computer program in a vat-internal ersatz vat, our statement "We are brains in vats" would also be false. Putnam therefore concludes that the statement "I am a brain in a vat" is necessarily false. Many formulations and alleged improvements on Putnam's anti-skeptical argument have appeared in the literature. A rough reconstruction of the argument runs as follows: (1) If I was a brain in a vat, I couldn't believe that I am a brain in a vat. (2) I believe that I am a brain in a vat. (C) Therefore, I am not a brain in a vat. Premise (1) gives expression to the doctrine of semantic externalism. Premise (2) rests on the doctrine of privileged self-knowledge, that is, the idea that a person's sincere first-person present tense claims about his own thought contents have an epistemic authority no second or third person claim can have. (I use 'privileged' and 'non-empirical' interchangeably.) If the premises of an anti-skeptical argument could only be known empirically·, one would assume empirical knowledge to show that we are capable of having empirical knowledge. Thus, for an argument against external-world skepticism not to beg the question, its premises need to be knowable non-empirically. So let's examine whether premises (1) and (2) are indeed knowable just by reflection. Semantic externalism is not based on empirical beliefs concerning either our actual practice of belief attribution or the actual connections between mental states and environmental conditions. Since semantic externalism is a piece of arm-chair philosophy, it is usually stated in hypothetical terms: if a normal person and a brain in a vat would both

Believing that You K n o w and K n o w i n g that You Believe

371

say "I am a brain in a vat", then they would express different thoughts. Thus, assuming that semantic externalism is true, premise (1) is indeed knowable just by reflection. Let's turn to premise (2). To non-empirically know premise (2) one has to have privileged access to the contents of one's beliefs. Recently a number of philosophers have worried that semantic externalism conflicts with privileged self-knowledge. Elsewhere I have argued that semantic externalism and privileged self-knowledge are compatible, but that the notion of privileged self-knowledge reconcilable with externalism is weaker and less interesting than the Cartesian notion of self-knowledge (cf. 1996, 1998, 2000). Here I will not defend the compatibility of externalism and self-knowledge, but rather assume its truth. Yet even when the truth of compatibilism is presupposed, the nonempirical knowability of premise (2) is not unproblematic. It is a truism that one knows relative to a set of relevant alternatives and that the set of alternatives is determined relative to a context. In the case of my claim to know that I believe that p there are two relevant alternatives: either I believe that q or r rather than ρ or I am a zombie and thus incapable of having any belief whatsoever. If I can non-empirically know that I believe that p and not q or r, then it looks as if I can infer the falsity of external-world skepticism. For if thought contents supervene on external conditions and if I am capable of knowing my specific belief contents, then I seem to be in a position to tell which world I inhabit—the ordinary world or the skeptic's demon-world. 2 Even if we grant Putnam that I can non-empirically know that I believe that p in the sense that I know that I don't believe that q or r it remains an open question whether I can non-empirically know that I believe that p in the sense that I can rule out that I am a zombie. Semantic externalism designates a position that not only concerns the conditions for individuating belief contents, but also the conditions for having beliefs. Semantic externalism accounts not only for the differences among beliefs, but also for the difference between beliefs and states lacking mental content. Given semantic externalism, for a brain state to have any determinate content there have to be things that systematically cause it to be tokened. If ontological solipsism were true and the only things that exist were my brain states, there would be nothing in the outside world for these brain states to supervene on and hence they would lack content. 3 Thus, to non-empirically know that I believe that p in the sense of knowing that I am not a zombie amounts to a refutation of ontological solipsism. To see that Putnam's brain-in-a-vat argument doesn't refute ontological solipsism, consider a zombie saying "I am a brain in a vat". Since the zombie's use of 'brain' and 'vat' are completely uncaused, Putnam's externalism can get no foothold; it doesn't enable us to evaluate as true any of the zombie's sentences. These sentences will have to be regarded as lacking determinant meaning. Putnam's argument only allows us to

2

In ( 2 0 0 0 ) I argue that McKinsey-style ( 1 9 9 1 ) refutations of external-world skepticism on the basis o f semantic externalism plus privileged self-knowledge are defective because they erroneously assume that we can non-empirically know that we are not zombies incapable o f haveing beliefs.

3

Ontological solipsism has to be distinguished f r o m epistemological solipsism which holds that the self is the origin o f knowledge o f existence.

372

Self-Knowledge

differentate between scenarios in which my words have determinate meaning, but it doesn't allow us to rule out the hypothesis that our terms are empty and that we are zombies. Only if it is assumed that we are not zombies does Putnam's reasoning enable us to rule out that we are not brains in vats. In sum, Putnam's brain-in-a-vat argument has a point against external-world skepticism but it is ineffective when it comes to stronger forms of epistemic skepticism such as ontological solipsism. Unless Putnam can complement his brain-in-a-vat argument with an argument to the effect that we know non-empirically that we are not zombies he makes no real progress with the skeptical challenge. For as long as such an additional argument is not provided, an external-world skeptic can dismiss the conclusion of the brain-in-a-vat reasoning by asking how we know that we are not zombies living in the solipsist s world. But what are the prospects of developing such an additional argument against the zombie-hypothesis? In section 3 I will argue that semantic externalism is incompatible with non-empirical knowledge to the effect that I am not a zombie. If this argument can be believed it follows that skepticism triumphes over Putnam's anti-skepticism.

II. I will now focus on an anti-skeptical argument from semantic externalism by Fred Dretske (1981, ch. 9; 2000). Unlike Putnam's brain-in-a-vat argument, Dretske's antiskeptical argument explicitely assumes non-empirical knowledge of the fact that one has beliefs and thus isn't a zombie. The key notion of Dretske's version of semantic externalism is information. Information is an objective commodity, existing independently of any cognitive agent. We only have information if what we learn from it is true. Concepts and beliefs—special kinds of representations—are internal structures designed to carry information. Since a person can only learn the concept f h y being exposed to the information that χ is f and since a signal can only carry the information that χ i s f i i in fact χ isf, his concept y~must refer to something—it cannot have an empty extension. True beliefs involve the tokening of a concept with respect to a relevant perceptually present object. False beliefs are caused by the mistokening of a concept. According to Dretske's epistemic externalism, for a person to know that χ is f , he needs to believe that it is f , and the belief must be caused by the information that χ is f . The person doesn't need to know that he is receiving information for his information-produced belief to qualify as knowledge. From Dretske's semantic externalism it follows that unless a person can receive the information that x's are f , and in fact has received (during learning) the information that is used to form the concept^ he cannot represent χ as being f . Being able to form a concept presupposes being causally connected to the requisite kind of information at least once, viz. during the learning period. The way Dretske ties a person's contents and concepts down to his actual environment blocks the crucial starting assumption of the skeptic—that everything about his beliefs could be exactly the way it is now, even if the world were completely different from the way it seems to him. In Dretske's

Believing that You Know and Knowing that You Believe

373

picture, beliefs are about an actually existing source o f information, and they are tied to that source constitutively, so that the belief could not exist without the very source having existed at least once. Granted that during the learning situation things must seem how they are, and granted that knowledge is nothing but information-produced belief, when a person learns the concept ^ a n d first comes to believe that χ is f , he also knows that χ is f . Thus, just as we can only know what we have the resources o f believing, we can only believe what we have (or have had) the resources o f knowing. Dretske's reasoning can be parsed into the following steps: (3) To form the concept f , I must have processed the information that χ is f . (4) To form the belief that χ is f , I must possess the concept f . (C) Therefore, if I believe that χ is f , then I am, in principle, capable o f knowing that χ is f. By turning the conclusion o f this argument into premise (5), we get the following transcendental argument against external-world skepticism: (5) I f I have beliefs, I have the possibility o f knowledge. (6) I have beliefs. (C) Therefore, I have the possibility o f knowledge. Assuming that the external-world skeptic doesn't want to challenge premise (6), Dretske concludes that "the view that nothing can be known is demonstrably false" ( 1 9 8 1 , p. 229). Dretske's anti-skeptical argument is limited to simple, ostensively acquired concepts. A skeptic can, o f course, ask just how widespread beliefs involving simple concepts are and how complex concepts are manufactured out o f simple ones. But instead o f examining the scope o f Dretske's anti-skeptical argument I will focus (once again) on the question o f whether premises (5) and (6) can be known non-empirically. For premise (5) to be knowable non-empirically, the justificatory basis o f semantic externalism must be independent o f empirical beliefs concerning the actual connections between mental states and utterances on the one hand, and the physical and social world on the other. In the previous section I have already stated my reason for holding that semantic externalism is knowable non-empirically. Next let's now look at premise (6). Unlike Putnam's premise (2) which concerns a specific belief content, Dretske's premise (6) states the general fact that one has beliefs and thus isn't a zombie. As was explained in section 1, I assume that semantic externalism is compatible with non-empirical knowledge o f one's specific belief contents. T h e crucial question, however, is whether semantic externalism also allows for nonempirical knowledge o f the general fact that I am a belief-possessing being rather than a zombie.

III. Given that semantic externalism is true and that mental content consists in the nomic dependence o f brain states on environmental conditions, a type o f configuration in my

374

Self-Knowledge

brain is a belief about, say, water only if it is normally tokened in the presence of H 2 0 . But how can I know—from a first-person perspective—that there is any systematic relationship between my internal states and those of an external kind? This I cannot know from the apparent regularities of my inner states, for, according to semantic externalism, the intrinsic character of my states isn't relevant for content-determination. Thus, just by reflection, I cannot tell whether my so-called beließ have content or are contentless states. For all I can tell by reflection, the internal state I refer to as the belief that, say, water is wet could be nothing but an episode of heartburn or some other state lacking mental content. Since reflection doesn't provide me with knowledge that my so-called beliefi are in fact beliefs, i.e., depend on nomic relations—whether engendered by the ordinary objects I normally suspect, or the machine which maintains my brain in a vat—I am unable to non-empirically know that I am having beliefs. Semantic externalism implies that we are unable to non-empirically know that we are minded beings rather than zombies. A critic might raise the following objection: if I were a zombie, then, by definition, I couldn't have any beliefs, let alone the belief that I am not a zombie. Thus I couldn't believe that I am not a zombie, unless it was true. The belief that I am not a zombie is self-verifying. Now the reliability theory of knowledge claims that for a state of believing to qualify as knowing the belief must be non-accidentally true. This idea is explicated in terms of a counterfactual condition: if p were not true, the subject would not believe that p. The belief must 'track', 'hook up with', or 'indicate' the facts that make it true. The belief that I am not a zombie is an example of a truth-tracking belief. I would not—indeed could not—believe ρ unless ρ was true. Hence given the reliability theory of knowledge, semantic externalism allows me to non-empirically know that I am a minded being rather than a zombie. The problem with this objection is that not every self-verifying belief is an instance of knowledge—not even on a reliabilist conception of knowledge. To see this, consider self-fulfilling beliefs. If a belief could count as knowledge simply by covarying with the facts it is about any self-fulfilling belief would have to count as knowledge. To disqualify self-fulfilling beliefs from the ranks of knowledge one only has to combine reliabilism about knowledge with reliabilism about justification. Reliabilism about justification claims that what makes a belief justified is the fact that either it is produced by a reliable method or that it is based on a reliable reason. Following the latter proposal, my non-empirical belief that I am a minded being fails to qualify as knowledge because we are unable to distinguish from the 'inside' our possession of a mind from relevant compatible states of affairs. None of the internal states that I may use to support the belief that I am not a zombie indicate (carry the information) that I am not a zombie. And hence not even on a reliabilist conception of knowledge can we be attributed nonempirical knowledge of the fact that we are not zombies.

IV. In the previous section it was argued that, given semantic externalism, we are unable to non-empirically know that we are minded beings rather than zombies. The question

Believing that You Know and Knowing that You Believe

375

I want to explore in this section is how this result affects Putnam's and Dretske's antiskeptical arguments from semantic externalism. Let's begin with Dretske. If the reasoning in the previous section hit its mark, Dretske is not entitled to assume non-empirical knowledge of premise (6), for, given semantic externalism, reflection doesn't teach us whether we have beliefs or are zombies. An external-world skeptic can dismiss Dretske's argument by pointing out that the doctrine of semantic externalism underlying premise (5) conflicts with the non-empirical knowability of premise (6). I should note that Dretske has changed his position since he developed the antiskeptical argument under discussion. He now maintains that semantic externalism conflicts with non-empirical knowledge of the fact that we are minded beings. See Dretske (2003) and his contribution to this volume for his most recent views on this topic. Unlike Dretske, Putnam does not presuppose that we can rule out, just by reflection, that we are zombies devoid of beliefs. Premise (2) of Putnam's argument concerns a specific belief content rather than the general fact that one has beliefs. This poses the question of whether it is possible to non-empirically know premise (2) whilst premise (6) cannot be known non-empirically. Prima facie, the idea that semantic externalism is incompatible with non-empirical knowledge of whether one has beliefs conflicts with the idea that we can know about the specific contents of our beliefs. For if we cannot non-empirically know that we are having beliefs with some determinate content, how can we know the specific contents of our beliefs? If we cannot know that we have a mind, how can we know what is in our mind? The thesis that semantic externalism doesn't allow us to know, just by reflection, whether we are zombies appears to undermine the possibility of privileged self-knowledge understood as knowledge of the specific contents of one's mind. Given this reasoning, semantic externalism conflicts with the non-empirical knowability of premise (2) of Putnam's anti-skeptical argument. Elsewhere I have argued that the apparent conflict between the idea that the zombiehypothesis cannot be ruled out non-empirically, on the one hand, and the doctrine of privileged self-knowledge of specific belief contents, on the other, rests on an illegitimate

utilization of the principle ofebsure

under known implications (2000, pp. 11-16). The

closure principle says that if S knows p, and S knowns that ρ entails q, then S knows q. Those who argue that the commitment to the doctrine of privileged self-knowledge of specific belief contents (premise (2)) conflicts with the claim that premise (6) isn't knowable non-empirically have to rely on the closure principle to make their case. The proposition that I am having a particular belief, say, that I am a brain in a vat (p) implies the falsity of the proposition that I lack beliefs (q). Given the closure principle, if I am aware of the implication between ρ and q, I need to non-empirically know q to be able to non-empirically know p. And if my argument to the effect that q cannot be known non-empirically is correct, then ρ cannot be known non-empirically. It follows that the doctrine of privileged self-knowledge of one's belief contents is a myth and so is compatibilism. Regarding knowledge of the outside world, most epistemologists agree that the closure principle is too strict to be convincing. If knowing that p would require the elimination of every known alternative to p, as suggested by the closure principle, we could never know anything about the world around us. A more plausible view is that

376

Self-Knowledge

knowledge requires the elimination of only relevant alternatives. So if, for some reason, the hypothesis that we are mindless zombies should become relevant, we would have to eliminate it to know what our beliefs are about. And assuming that we would fail to eliminate this hypothesis, we would lack privileged self-knowledge of our specific belief contents. Ordinarily, however, the hypothesis that we are zombies isn't relevant and need not be eliminated—not even if we are aware of the fact that "I believe that ρ (rather than q or r)" implies the falsity of "I lack beliefs because I am a zombie". So usually we can know what we are believing—the specific contents of our beliefs—without having to know that we are believing—that our so-called beliefs are states with content. Provided we have minds, semantic externalism doesn't rob us of our privileged self-knowledge of specific belief contents. Self-knowledge (a belief) is consistent with our lacking the ability to rule out, via reflection, the possibility that we don't have any beliefs. Therefore premise (2) is knowable non-empirically although premise (6) is not. 4 Even though Putnam's anti-skeptical argument from semantic externalism is coherent—unlike Dretske's argument—it is nevertheless ineffective. For as was explained in section 1, unless Putnam can complement his brain-in-a-vat argument with an argument to the effect that we know non-empirically that we are not zombies, the skeptic has an easy job of refuting the brain-in-a-vat-argument; all he has to do is to entertain the possibility of ontological solipsism. But such an argument to the effect that we non-empirically know that we are not zombies is not to be had, at least not if semantic externalism is taken for granted.5

References Bernecker, S. 1996: "Externalism and the Attitudinal Component of Self-Knowledge", Nous 30, pp. 262-275. Bernecker, S. 1998: "Self-Knowledge and Closure", in P. Ludlow & N. Martin (eds.), Externalism and Self-Knowledge, Stanford, CSLI Publications, pp. 333-349. Bernecker, S. 2000: "Knowing the World by Knowing One's Mind", Synthese 123, pp. 1-34. Dretske, F. 1981: Knowledge and the Flow of Information, Cambridge/MA, MIT Press. Dretske, F. 2000: "The Epistemology of Belief", in his Perception, Knowledge, and Belief, Cambridge, Cambridge University Press, pp. 64-79. (Reprinted from Synthese 55, 1983, pp. 3-19). Dretske, F. 2003: "Externalism and Self-Knowledge", in S. Nuccetelli (ed.), New Essays on Semantic Externalism, Skepticism, and Self-Knowledge, Cambridge IMA, MIT Press, pp. 131-142. Dretske, F. 2004: "Knowing what You Think vs. Knowing that You Think it", in this volume. McKinsey, M. 1991: "Anti-Individualism and Privileged Access", Analysis 51, pp. 9-16. Putnam, H. 1981: Reason, Truth and History, Cambridge, Cambridge University Press. 4

5

For the distinction between knowing what and knowing that, see my (2000, pp. 11-6). To avoid misunderstanding, I should note that the relevant alternative analysis of knowledge is not committed to denying closure tout court. Just because one cannot know that one has a mind by way of knowing what is in one's mind (plus semantic externalism) doesn't mean that one cannot, for example, know that one is petting a mammal by means of knowing that one is petting a dachshund. If one couldn't know any of the known consequences of the things one knows, reasoning in general would be in jeopardy. One of the difficulties facing the relevant alternative account of knowledge is to come up with a general criterion by which to determine the number and kind of counter-possibilities that are ruled out by a particular knowledge claim. This paper was written during a visiting appointment at the Philosophy Department at the University of Minnesota, Twin Cities in the fall of 2002. I wish to thank the members of the department for their hospitality. For helpful comments I am grateful to Peter Baumann, Fred Dretske, Thomas Grundmann, Kaveh Kamooneh, and Wade Savage.

McKinsey Redux? ANTHONY BRUECKNER

Michael McKinsey's "Anti-Individualism and Privileged Access" has generated an interesting literature.1 In this paper, I will discuss McKinsey's recent contributions. In his seminal paper, McKinsey argued that the following is an inconsistent triad: (1) Oscar knows a priori that he is thinking that water is wet. (2) The proposition that Oscar is thinking that water is wet conceptually implies E. (3) The proposition Ε cannot be known a priori, but only by empirical investigation. Ε is some "external proposition" describing "the relations that Oscar bears to other speakers or objects in his external environment". 2 By "a priori knowledge", McKinsey means "knowledge obtained independently of empirical investigation" of the external world. (1) expresses the view that Oscar has privileged access to his thought that water is wet. (2), according to McKinsey, expresses the core idea of anti-individualism (a.k.a. content externalism) as it applies to Oscars thought about water. The triad can be seen to be inconsistent in the following way: Suppose (1) that Oscar knows a priori that he is thinking that water is wet. T h e n by...[(2)], Oscar can simply deduce E, using only premisses that are knowable a priori, including the premises that he is thinking that water is wet. Since Oscar can deduce Ε from premises that are knowable a priori, Oscar can know Ε itself a priori. But this contradicts (3), the assumption that Ε cannot be known a priori. Hence (1), ...[(2)], and (3) are inconsistent. And so in general, it seems, anti-individualism is inconsistent with privileged access. 3

This reasoning depends upon McKinsey's understanding of the notion of conceptual implication (which in recent work he also calls logical implication). He says: ...a proposition ρ conceptually implies a proposition q if and only if there is a correct deduction of q from p, a deduction whose only premises other than ρ are necessary or conceptual truths that are knowable a priori, and each of whose steps follows from previous lines by a self-evident rule of some adequate system of natural deduction. 4

1

See McKinsey 1991. Some of the literature spawned by this article: Boghossian 1997, Brown 1995, 1999, 2001, Brueckner 1992, 1995, 1999, 2000, 2 0 0 2 , Davies 1998, 2 0 0 0 , Falvey 2 0 0 0 , McKinsey 1994, McLauglin and Tye 1998, Miller 1997, Nuccetelli 1999, R a f f m a n 1998, Sawyer 1998, Wright

2000. 2 3 4

McKinsey 1991, p. 1. McKinsey 1991, p. 15. McKinsey 1991, p. 14.

378

Self-Knowledge

1. A Reply In the literature's first reply to McKinsey's seminal paper, I focused upon the work of Tyler Bürge, as did McKinsey. 5 I maintained that for the Burgean anti-individualist, there is no "external proposition" Ε which makes both (2) and (3) true. So there is no threat to the privileged access thesis (1). For example, (2) is false, according to the anti-individualist, if Ε is interpreted as ( E l ) Oscar inhabits an environment containing H 2 0 and not XYZ. For the Burgean anti-individualist, there is no conceptual/logical connection between (a) Oscar's thinking that water is wet, and (b) the existence of H 2 0 . Consider the conditional (4)

If Oscar is thinking that water is wet, then E l .

(4), if true, expresses a metaphysical necessity that is knowable only a posteriori, since its truth would depend upon the a posteriori necessity that water is H 2 0 . Thus, if Ε in (2) is interpreted as El, then (2) is false. The anti-individualist can hold on to the privileged access thesis asserted in (1) while affirming (3) and rejecting (2). Further, I pointed out that according to the Burgean anti-individualist, it is possible for Oscar to think that water is wet in a world lacking H 2 0 . So (4) does not even express an a posteriori, metaphysical necessity in the first place, much less an a priori truth. There is thus all the more reason to reject (2) if Ε is interpreted as E l . Bürge does seem to commit himself, however, to the view that the following is a metaphysical necessity: (N) If Oscar is thinking that water is wet, then either (i) water exists, or (ii) Oscar theorizes that H 2 0 exists, or (iii) Oscar is part of a community of speakers some of whom theorize that H 2 0 exists. (2) is once again false if Ε interpreted as the disjunctive consequent of (N) (call this E2). Even if (N) is a metaphysical necessity, it is not knowable a priori, since its truth would depend upon the a posteriori metaphysical necessity that water is H 2 0 . The anti-individualist can again affirm both (1) and (3) while rejecting (2), on this interpretation of E. Though the matter is controversial, (2) may be true if Ε is interpreted as (E3) The external world of physical objects exists.

5

See Brueckner 1992, and Bürge 1 9 8 2 .

McKinsey Redux?

379

In this case, though, (3) would, arguably, be false. There might well be a transcendental argument, whose premises are knowable a priori, that connects thinking that water is wet with the existence of a physical world. 6 O n this interpretation of E, then, the anti-individualist can affirm (1) and (2) while rejecting (3). Thus, McKinsey has not demonstrated the incompatibility of anti-individualism and privileged access.

2. Reconsideration of McKinsey s Position As we have seen, in "What an Anti-Individualist Knows A Priori", I assumed that if ρ conceptually/logically implies q, then the conditional If p, then q. is knowable a priori. Since the conditionals (4) and If Oscar is thinking that water is wet, then E2. 7 are not knowable a priori, it would follow that the antecedent of the conditionals does not conceptually/logically imply the consequents. (2), then, would come out false on the pertinent interpretations of E, allowing the anti-individualist to hold on to privileged access while affirming (3). In recent work on these issues, McKinsey maintains that his original argument has been misinterpreted by various writers in the literature. Critics, and supporters as well, have made a crucial, unwarranted assumption about McKinsey s view of the connection between a prioricity and conceptual/logical implication. Before describing the misinterpretation in question, I will discuss McKinsey's overall strategy in his new paper, "Forms of Externalism and Privileged Access".8 Whereas his original paper has often been seen in the literature as an attempted reductio of anti-individualism, McKinsey points out in the new paper that in the original paper, he in fact argued only for the inconsistency of (l)-(3), without taking a stand on which member of the triad is to be rejected. In the new paper, he maintains that semantic externalism is true, and that it entails (2): (SE) Many de dicto-structured predicates of the form 'is thinking that p' express properties that are wide, in the sense that possession of such a property by an agent logically or conceptually implies the existence of contingent objects external to that agent. (200)

6 7 8

See Brueckner 1992 and Brueckner 1999 for discussion of this issue. The second paper makes a case against the existence of such a transcendental argument. This is just (N) rewritten. See McKinsey 2002b. See also McKinsey 2002a.

380

Self-Knowledge

Given the inconsistency of our triad and the plausibility of (3) and (SE), McKinsey says, (1) is seen to be the false member of the triad. More generally, the following principle ("Privileged Access to Content") is shown to be false: (PAC) It is necessarily true that if a person χ is thinking that p, then χ can in principle know a priori that he himself, or she herself, is thinking that p. (199) McKinsey thinks that "the semantic facts about cognitive predicates containing smallscope names, indexicals, and natural kind terms show that (SE) is true and therefore that (PAC) is false". He holds that since ordinary proper names have no descriptive meaning, the sentence (6) Dave is thinking that Larry is a janitor. is relational with respect to the referent of'Larry'. On the first approach to this relationality, (6) ascribes to Dave and Larry "the relation that an object χ bears to an object y just in case χ bears the relation of thinking to the singular proposition that y is a janitor". On the second, related approach (which McKinsey favors) "the relation ascribed [by (6)] is that which an object χ bears to an object y just in case χ has a thought about y to the effect that y is a janitor". (204) On this approach, "the structurally de dicto sentence...[(6)] turns out to be semantically de re". (203) The upshot is that the cognitive property expressed by 'is thinking that Larry is a janitor' is wide, because its possession by an agent " l o g i c a l l y , not just metaphysically, implies the existence of the contingent property with respect to which the property is relational". (204) McKinsey says, "Thus, the proposition that Dave thinks that Larry is a janitor logically implies that Larry exists". (204) These points establish (SE), according to McKinsey: there are many de dicto-structured predicates, such as 'is thinking that Larry is a janitor', which turn out to express wide properties. Let us return to the question of how exactly the inconsistency of the triad (l)-(3) is supposed to be established. Both supporters of McKinsey's position and critics as well have interpreted him in the following way. Assume that (1) is true: Oscar knows that he is thinking that water is wet. Assume that (2) is true: the proposition that Oscar is thinking that water is wet conceptually/logically implies E. Then Oscar can know a priori a corresponding conditional: (2*) If Oscar is thinking that if water is wet, then E. The following principle ("Closure under A Priori Knowable Implication") is very plausible: (CAK) Necessarily, for any person χ and any propositions Ρ and Q, if χ can know a priori that Ρ and χ can know a priori that if Ρ then Q, then χ can know a priori that Q. (207) From (1), (CAK), and the a priori knowability of (2*), it follows that

McKinsey Redux?

381

not-(3) The proposition Ε can be known a priori (by Oscar). Many of McKinsey's critics object to the claim that the triad (l)-(3) is inconsistent by denying that (2*) is knowable a priori. This was my strategy in considering the conditionals (4) and (5) above. In response, McKinsey, in the new paper, correctly points out that he never committed himself to the a priori knowability of (2*) in his original paper. In the new paper, he explicitly denies that (2*) is knowable a priori, even given the truth of (2). He affirms the truth of (CAK), while maintaining that it does not apply in the present case, since the conditional (2*) is not knowable a priori. This means that the foregoing way of arguing for the inconsistency of the triad is not McKinsey's way. So how is the inconsistency supposed to be established? According to McKinsey, the pertinent closure principle is not (CAK) but rather the principle of "Closure of A Priority under Logical Implication": (CA) Necessarily, for any person x, and any propositions Ρ and Q, if χ can know a priori that P, and Ρ logically implies Q, then χ can know a priori that Q. (207) Given (1) and (2), (CA) yields not-(3) without appeal to the a priori knowability of (2*). Since (2) and (3) are both true, says McKinsey, (1) is false. If Oscar could know a priori that he is thinking that water is wet, he would, given (CA) and (2), be able to know Ε a priori. But this he cannot do. Thus, one cannot object to McKinsey's treatment of the triad (l)-(3) by denying the a priori knowability of (2*).

3. Problems for

McKinsey

One would have thought, however, that if ρ conceptually/logically implies q, the corresponding conditional If p, then q. is going to be knowable a priori. If ρ logically/conceptually implies q, then given McKinsey's understanding of such implication, it would seem that we can construct a correct conditional derivation of q from p, each of whose steps consists in an a priori knowable deduction from previous lines (which may contain additional a priori knowable premises other than ρ itself). The existence of such a derivation appears to be guaranteed by the assumed conceptual/logical implication (given McKinsey's understanding of such implication), and knowledge of the derivation would yield a priori knowledge of the conditional. According to McKinsey, however, "externalist claims like (2) constitute an important class of exceptions to the assumption that logical relations are knowable a priori". (208) In order to illustrate this point, McKinsey returns to his claim:

382

Self-Knowledge

(7) The proposition that Dave is thinking that Larry is a janitor logically implies that Larry exists. I quote McKinsey at length: N o w . . . [ ( 7 ) ] is not knowable a priori. For even t h o u g h . . . [ ( 7 ) ] is a meta-proposition about the logical implication o f o n e proposition by another, it is also a proposition that is singular w i t h respect to Larry, and it ascribes a c o m p l e x logical property to Larry. T h i s is the property that a n y object χ has if and o n l y if the proposition that D a v e is thinking that χ is a janitor logically implies the proposition that χ exists. S o the meta-proposition expressed b y . . . [ ( 7 ) ] is itself a singular proposition that does n o t exist unless Larry does, and hence its truth cannot be k n o w n a priori. Similarly, the conditional proposition that if Dave is thinking that Larry is a janitor then Larry exists, is also singular with respect to Larry, and so it too cannot be k n o w n a priori. ( 2 0 8 )

McKinsey goes on to say that for the same reason, the conceptual/logical implication asserted in (2) does not imply that the corresponding conditional (2*) is knowable a priori. There are a number of problems with McKinseys reasoning in the foregoing passage. Before considering them, I will try to reconstruct this reasoning. First, McKinsey appears to have assumed the principle (8)

If (i) the proposition that ρ is singular, and (ii) S knows that p, then S knows that the proposition that ρ exists.

If (8) is true, then it is plausible to hold (9)

If (i) the proposition that ρ is singular, and (ii) S knows a priori that p, then S knows a priori that the proposition that ρ exists.

Since one cannot know a priori whether Larry exists, one cannot know a priori whether there exists a singular proposition that is expressed by (10) If Dave is thinking that Larry is a janitor, then Larry exists. Thus, even though (10) is in fact true, and even though its antecedent conceptually/logically implies its consequent, one cannot know (10) in an a priori manner, given (8) and (9). Now the following variant of (8) is quite plausible: (8') If S knows that p, then the proposition that ρ exists. However, the higher-order knowledge required by (8), regarding the existence of a proposition, is not possessed by ordinary knowers whose conceptual repertoire does not contain the concept of a proposition. But such knowers clearly can know singular propositions despite lacking the higher-order knowledge required by (8). Further, (8) leads to a problematic regress unless (8) is somehow qualified. The proposition that the proposition that ρ exists is itself a singular proposition, according to the views expressed in the previous long quotation from McKinsey. Thus, in order

McKinsey Redux?

383

to know this proposition (call it p*) concerning the existence of the proposition that p, S will have to know that p* exists, according to (8). The same reasoning applies to the proposition that p* exists (call this proposition p**). Since p** is itself a singular proposition, in order to know that p**, S will have to know that p** exists. And so on without end. Let us now return to McKinseys claims about the holding of the alleged conceptual/logical implications expressed in (2) and (7). According to McKinseys (7), there is a correct deduction of (11) Larry exists, from (12) Dave is thinking that Larry is a janitor. such that (a) each step in the deduction follows from a previous line by a self-evident rule of inference, and (b) any premises other than (12) are knowable a priori. Recall that according to McKinsey, (12) ascribes to Dave and Larry that relation "which an object χ bears to an object y just in case χ has a thought about y to the effect that y is a janitor". (204) If we symbolize this relation by the two-place predicate 'R(xy)', and we let 'D' denote Dave and 'L' denote Larry, we can represent (12)'s logical form by (12*)

R(DL).

From (12*) and (13) L=L we can deduce (14) R(DL) & L=L. From (14), we can deduce (15) 3x[R(Dx) &x=L], That is, there exists something that is Larry and is R-related to Dave. Larry exists! So we have made a case for the conceptual/logical implication of (11) by (12), based on the relationality of 'is thinking that Larry is a janitor'. We cannot, however, in an analogous fashion correctly deduce (16) Pegasus exists, from (17) Dave is thinking that Pegasus flies.

Self-Knowledge

384

This is because (17) does not have a relational logical form analogous to that given in (12*). Indeed, given McKinsey's remarks about why the conditional (10) If Dave is thinking that Larry is a janitor, then Larry exists. is not knowable a priori, his view is presumably that (17) fails to express a proposition. That is, according to McKinsey, one cannot know (10) a priori, because one cannot know a priori whether or not there exists a proposition that is expressed by (10). One cannot know a priori whether or not Larry exists, and if he does not, then neither the antecedent nor the consequent of (10) expresses a proposition. Similarly, (17) presumably fails to express a proposition, in virtue of the nonexistence of Pegasus. It is worth noting in this connection that it follows from the foregoing considerations that there are some sentences of the form If p, then p. which are such that one cannot know a priori whether the sentence expresses a logical truth. For example, it follows from McKinsey's views that one cannot know a priori whether the following sentence expresses a logical truth: (18) If Pegasus flies, then Pegasus flies. This is because one cannot know a priori whether (19) Pegasus flies. expresses a proposition. It is an a posteriori matter whether Pegasus exists, and thus it is an a posteriori matter whether there exists a proposition that is expressed by (19). Let us now return to the question whether McKinsey is right in holding that (2) is true. If (2) is true, then there must be a correct deduction of Ε (whatever "external proposition" this turns out to be) from (20) Oscar is thinking that water is wet. As we saw earlier, McKinsey holds that both 'is thinking that Larry is a janitor' and 'is thinking that water is wet' express wide psychological properties. Presumably, then, McKinsey believes that the deduction of Ε from (20) parallels the deduction of (11) from (12). Thus, where 'R*(xy)' symbolizes a relation between a thinker and the external objects whose existence is asserted by Ε, Ό ' denotes Oscar, and ' W ' denotes the external objects in question, the logical form of (20) is, for McKinsey, presumably given by (20*)

R*(OW).

As before, from (20*) and

McKinsey Redux?

385

(21) W = W we can deduce (22) R*(OW) and W=W. From (22) we can deduce (23) 3x[R*(Ox) & x=W], Ε follows from (23), since ' W ' denotes whichever objects are asserted to exist by E. Even granting this view of (20)'s logical form and the concomitant deduction of Ε from (20), McKinsey's defense of the inconsistency of (l)-(3) in the end runs into an insuperable difficulty. Recall that that defense hinged on the principle (CA). I will now argue that (CA) is false given McKinsey's apparent view of the deduction of Ε from (20), whose existence is required by the conceptual/logical implication asserted in (2). We treated the deduction involving the allegedly wide psychological predicate 'is thinking that water is wet' as being on a par with the deduction involving 'is thinking that Larry is a janitor'. Accordingly, we ought to hold that just as 'is thinking that Pegasus flies' fails to express a wide psychological property in virtue of the nonexistence of Pegasus, similarly 'is thinking that phlogiston is plentiful' fails to express a wide psychological property in virtue of the nonexistence of phlogiston. So just as the existence of Pegasus is not deducible from the premise that Dave is thinking that Pegasus flies, similarly there is no proposition E* concerning external objects such as phlogiston that is deducible from the premise that Oscar is thinking that phlogiston is plentiful. Let us grant that there is a correct deduction of Ε from (20) that flows from the logical form of (20). O n e can know a priori that Ε is deducible from (20) only if one can know a priori that (20) has the logical form given in (20*). Now, one cannot know a priori that (12) has the logical form given in (12*). For all one knows a priori, (12) might be on a par with (17), which fails to express a proposition at all. Similarly, for all one knows a priori, (20) is on a par with (24) Oscar is thinking that phlogiston is plentiful. and thus (a) lacks the logical form given in (20*), and (b) fails to express a proposition. Even if (20) in fact has the logical form given in (20*), which would allow for the deduction of E, this is something that one cannot know a priori. To put the point in a different way: one cannot know a priori whether or not 'water' is a successfiil natural kind term, rather than a term which merely purports to denote an existing natural kind. For all one knows a priori, 'water' is like 'phlogiston' and fails to pick out an existing natural kind. These points show that given McKinsey's apparent conception of how (20) conceptually/logically implies E, he is barred from embracing

386

Self-Knowledge

(CA) Necessarily, for any person x, and any propositions Ρ and Q, if χ can know a priori that P, and Ρ logically implies Q, then χ can know a priori that Q. In some cases in which a proposition ρ conceptually/logically implies q, one cannot know a priori that this implication holds. In such cases, then, given that one knows a priori that p, it does not follow that one can know a priori that q. For this to follow, one would have to be able to recognize the conceptual/logical implication of q by ρ in an a priori manner. Thus, McKinsey cannot appeal to (CA), as he does, in arguing for the inconsistency of the triad (l)-(3). Oddly enough, McKinsey himself makes the foregoing point upon which my objection to (CA) relies, viz. that some conceptual/logical implications are not knowable a priori. As we saw, he makes this point in connection with (7) The proposition that Dave is thinking that Larry is a janitor logically implies that Larry exists. His justification for the point seemed to involve the dubious principle (8) and the related (9). The point, though, can be seen to hold by examining, as we have done, McKinsey's rationale for the deducibility of (11) from (12) and his apparent rationale for the deducibility of Ε from (20). Given the existence of counterexamples to (CA), McKinsey is left in an untenable position with respect to the triad (l)-(3). In order to show that (1) and (2) imply not-(3) The proposition Ε can be known a priori (by Oscar). McKinsey in the end needs to maintain that (2) implies the a priori knowability of (2*)

If Oscar is thinking that water is wet, then E.

Then (CAK), which McKinsey accepts, would yield not-(3), given (1): (CAK) Necessarily, for any person χ and any propositions Ρ and Q, if χ can know a priori that Ρ and χ can know a priori that if Ρ then q, then χ can know a priori that Q. Since (CA) is false, McKinsey is forced to depend upon (CAK) in order to demonstrate the inconsistency of (l)-(3). However, I have argued that there is no interpretation of Ε on which (2*) is knowable a priori and on which (3) is true. McKinsey has nowhere countered these claims about the a posterioricity of (2*). Indeed, his main thrust in his new work has been aimed at establishing the inconsistency of the triad (l)-(3) without challenging the claims in the literature regarding the a posterioricity of (2*). I conclude that the considerations that McKinsey has brought forward in his recent work do not vindicate his position on the triad (l)-(3). The anti-individualist can affirm privileged access while rejecting the interpretations of (2) on which (3) is true. For all that McKinsey has shown, anti-individualism is compatible with privileged access.

McKinsey Redux?

387

References Boghossian, P. 1997. What the Externalist Can Know A Priori. In C. MacDonald, B. Smith, and C. Wright (eds.) Knowing Our Own Minds, pp. 271-84. Oxford: Oxford University Press. Brown, J. 1995. The Incompatibility of Anti-Individualism and Privileged Access. Analysis, 55, pp. 14956. 1999. Boghossian on Externalism and Privileged Access. Analysis, 59, pp. 52-9. 2001. Anti-Individualism and Agnosticism. Analysis, 61, pp. 213-24. Brueckner, A. 1992. What an Anti-Individualist Knows A Priori. Analysis, 52, pp. 111-18. 1995. The Characteristic Thesis of Anti-Individualism. Analysis, 55, pp. 146-48. 1999. Transcendental Arguments from Content Externalism. In R. Stern (ed.) Transcendental Arguments: Problems and Prospects, pp. 229-50. Oxford: Oxford University Press. 2000. Externalism and the A Prioricity of Self-Knowledge. Analysis, 60, pp. 132-7. 2002. Anti-Individualism and Analyticity. Analysis, 62, pp. 87-91. Bürge, Τ. 1982. Other Bodies. In A. Woodfield (ed.) Thought and Object: Essays on Intentionality, pp. 97120. Oxford: The Clarendon Press. Davies, M. 1998. Externalism, Architecturalism, and Epistemic Warrant. In C. MacDonald, B. Smith, And C. Wright (eds.) Knowing Our Own Minds, pp. 321-61. Oxford: Oxford University Press. 2000. Externalism and Armchair Knowledge. In P. Boghossian and C. Peacocke (eds.) New Essays on the A Priori, pp. 384-414. Oxford: The Clarendon Press. Falvey, K. 2000. The Compatibility of Anti-Individualism and Privileged Access. Analysis, 60, pp. 137-42. McKinsey, M. 1991. Anti-Individualism and Privileged Access. Analysis, 51, pp. 9-16. 1994. Accepting the Consequences of Anti-Individualism. Analysis, 54, pp. 124-28. 2002a. Review of Knowing Our Own Minds. Philosophical Quarterly, 52, pp. 107-16. /2002b. Forms of Externalism and Privileged Access. Philosophical Perspectives, 16, pp. 199-224. McLaughlin, B. and Tye, M. 1998. Externalism, Twin Earth, and Self-Knowledge. In C. Mac Donald, B. Smith, and C. Wright (eds.) Knowing Our Own Minds, pp. 285-320. Oxford: Oxford University Press. Miller, R. 1997. Externalist Self-Knowledge and the Scope of the A Priori. Analysis, 57, pp. 67-75. Nuccetelli, S.1999. What Anti-Individualists Cannot Know A Priori. Analysis, 59, pp. 48-51. Raffman, D. 1998. First-Person Authority and the Internal Reality of Beliefs. In C. MacDonald, B. Smith, and C. Wright (eds.) Knowing Our Own Minds, pp. 363-70. Oxford: Oxford University Press. Sawyer, S. 1998. Privileged Access to the World. Australasian Journal of Philosophy, 76, pp. 523-33. Wright, C. 2000. Cogency and Question-Begging: Some Reflections on McKinsey's Paradox and Putnam's Proof. Philosophical Issues, 10, pp. 140-63.

Knowing What You Think vs. Knowing that You Think It FRED DRETSKE

I am an externalist about meaning. I am an externalist about all forms of representation. That includes mental representation. What gives something intentional content, what makes it represent, mean, or say something about other affairs are not its intrinsic properties, but, rather, something about its purpose or function in an informational scheme. That is why alcohol in a glass tube—an ordinary household thermometer—is able to mean or say, truly or falsely as the case may be, that the temperature is 70° F. It has the function (a function we give it) of telling us, providing us with information, about temperature. Remove this informational purpose, this indicator function, and the glass encased liquid becomes representationally lifeless. Alcohol bottled for medicinal purposes on a medicine cabinet shelf does not represent temperature. It still expands and contracts as the temperature varies, of course, thereby providing information (if anyone cares to use it) about temperature, but it no longer says anything about temperature that (like a statement or a belief) could be false.1 The same is true of the splotches of ink in books and newspapers and the sounds we produce when talking to one another. It is not their shape, color, size, volume, or wave length (intrinsic properties), but rather something, broadly speaking, about their role in a system of communication that gives them their meaning. Except for the source of the functions (natural vs. conventional), the same is true, I submit, of events in our brains. They become representations, they acquire intentional content, by developing via some appropriate history an informational function. I have just described my own brand of metaphysical externalism about the mind. Like other forms of externalism, it denies that thought supervenes on the neurobiology of thinkers. The facts that determine what you think, are (some of them anyway) facts about the relations you, or your thoughts, bear (or bore) to external, affairs. The thoughts are in your head (just as words are in books), but what gives them their content (just as what gives words their meaning) isn't there. This, we are told, creates a problem. It is supposed to be incompatible with the kind of knowledge we have of what is going on in our own mind. How can we know, in the special, authoritative, way that we do, that we are thinking about peanuts if

*

1

Thanks to the members of El Instituto de Investigaciones Filosoficas, Universidad Nacional Autonoma de Mexico, Mexico City, for the tough questions they asked when I read an early version of this paper. I'm especially grateful to Maiate Ezcurdia for her questions about the scope of knowledge attributions. Thanks, also, to Andreas Kemmerling, Heidelberg, for his constructive skepticism about this material. The liquid continues to have what Grice calls "natural" meaning (it continues to indicate temperature), but, lacking an indicator function, it fails to have what he calls "non-natural" meaning. In talking about the content of thought, we are always talking about non-natural meaning.

390

Self-Knowledge

thinking about peanuts consists, in part, o f relations that exist between the thinker (or thought) and other parts o f the world? If, as some externalists hold, you cannot have peanut thoughts—cannot, that is, think that, or wonder whether, something is a peanut—without standing (or having stood) in causal relations to peanuts, then it seems to follow that to know, in that special authoritative way we call introspection, that you are thinking about peanuts is to know, in that same authoritative way, that there are (or were) peanuts. But one cannot know, not at least in that way (by, so to speak, gazing inwardly) that there actually are (or were) peanuts. That would be like trying to figure out whether you were married or wealthy by looking at yourself in a mirror. So thinking about peanuts cannot consist o f relations (causal, functional, informational, or whatever) to peanuts. It cannot depend on there being, or having been, peanuts. It cannot consist o f any relations o f the sort externalists propose since this would imply that we could know, in the same way we know that we are thinking about peanuts, that we stand (or stood) in these relations to the independently existing conditions that (according to externalism) make such thoughts possible. 2 I f one has a certain view o f knowledge, this objection to externalism doesn't have much bite. Suppose, for instance, that one is a reliabilist who thinks that any reliable connection to the facts (whether or not one knows or can determine that it is reliable) enables one to know these facts. Combine this with the entirely plausible assumption that some relations between an object and its environment are reliably reflected in the intrinsic properties o f the object. We can, for example, learn a great deal about the history o f a tree (its age, the rainfall and drought conditions during its growth, perhaps beetle infestations it survived) by examining its intrinsic properties—the pattern o f rings in its trunk. This is all one needs to given an account o f how a person might come to know, by looking inward (at his own intrinsic properties), what (some of) his relations to the environment are (or were). W h y can't introspection be like that for an externalist? 3 As a reliabilist, I'm attracted to this account. But I set it aside here for two reasons. ( 1 ) 1 don't think it really works (see Dretske 2 0 0 3 a for my reasons: basically, we are not aware o f the intrinsic properties that might reveal the relevant relationships). (2) I'm interested in convincing everyone, whether or not they are a reliabilist, that externalism about the mind creates no epistemological problem o f first-person authority. I f there is a problem about self-knowledge (I think there is, but more o f this later) it's a problem everyone has—whatever their view o f the minds constitution.

2

One's thoughts about peanuts may, o f course, depend, causally, on one's past transactions with peanuts, but externalist theory requires more than this. It requires the thoughts to consist o f these relations to peanuts. O n e cannot (logically) have peanut thoughts (want peanuts, look for peanuts, etc.) in a peanutfree world. This is a bit o f an exaggeration since externalists can manufacture peanut thoughts out o f external relations to things other than peanuts. O n e can, after all, think about (look for, be afraid of) unicorns. T h e basic point remains, though. Peanut thoughts require one to stand in relations to external things, if not peanuts, then whatever external elements are required for possession of this concept.

3

My thanks to David Sanford, Güven Güzeldere, and Murat Aydede for calling my attention to this possibility.

Knowing What You Think vs. Knowing that You Think It

391

This epistemological argument against externalism has by now spawned a substantial literature. I do not propose to review this literature (many o f the seminal articles are collected in Ludlow & Martin 1998) because, as I see it, the debate rests on a false assumption, an assumption that (with few exceptions) both parties to the debate, both externalists and their critics, make. They assume, mistakenly, that what we know by introspection is not only (in the case o f thought 4 ) what we think, the content o f our current prepositional attitude, but also that we think it, the fact that we occupy a mental state having a proposition as its content. I f this assumption is false, if it is possible to know by introspection that it is peanuts one is thinking (wondering, worrying, deciding) about without knowing, at least not in the same way, that one is thinking (wondering, worrying, or deciding) something with this content, then there is no epistemological threat to externalism. W h a t the externalist says is constituted by external relations to peanuts—thinking about peanuts—is not what introspection reveals about thought—that it is peanuts one is thinking about. To my dismay, however, I have discovered that many people (even a few o f the non-philosophers I've tried it on) find this thesis unintelligible. They tell me it just doesn't make sense. We cannot know what we are thinking about without knowing, in the same way, that we are thinking about it. I f we know, by introspection, that it is peanuts we are thinking about, then we must know, by introspection, that we are thinking about peanuts. I can live with implausibility but unintelligibility is something else. So I take this criticism seriously. I mean to address it in this essay. What I hope to show is that it is possible to know, by introspection (or by whatever method it is one knows it), that it is ρ one thinks without knowing, by that same method, that ρ is something one thinks. Although it won't quite get me to where I want to go, let me begin by talking about young children—people who think but do not know, do not even understand, what it means to think. Psychologists tell us that a typical three-year-old does not have a developed concept o f belief. These children have beliefs, o f course, and it is easy enough to find out, by asking them the right questions, what it is they believe, but that they believe or think these things is beyond their comprehension. They lack the concept o f thought, but they are, nonetheless, authorities about what they think. They enjoy a special kind o f access to the content o f their thought. I f you want to know whether three-year old Suzy thinks Daddy is home or the dog is loose, just ask her. Is Daddy home? Is the dog loose? Her answers will tell you, quite unerringly, what it is she believes . You cannot, to be sure, ask Suzy directly, and in just these words, what it is she believes because (we are assuming) Suzy doesn't yet understand, at least she haven't fully mastered, what it means to believe something. But there are indirect ways o f finding out. I f Suzy understands what it means for the dog to be loose, you can find out whether she thinks the dog is loose by asking her whether the dog is loose. She is an

4

Throughout I use (like Descartes) "thought" as a very general category. It (for me) includes all prepositional attitudes. In wondering whether P, in hoping, fearing, or regretting that P, and in wanting or desiring that P, one is (in this broad sense) thinking that P. To know that you think (in this broad sense) that P, then, is to know that you occupy a mental state with intentional content. It is not merely to know that you (for instance) think rather than fear that P.

392

Self-Knowledge

authority, mind you, not on whether Daddy is home or the dog is loose (about these topics you may know better than she), but on whether she thinks Daddy is home or the dog is loose. Her authority on these topics is in no way diminished by her ignorance of what it means to think such things. Children who think (and say) that the dog is loose must, of course, understand what it means for the dog to be loose but they needn't understand what it means to think it. They would not—indeed, they could not—say (or even think) that they know what they think since saying (or thinking) this requires them to refer or pick out what they think (that the dog is loose) as something they think and, lacking an understanding of what it means to think, they are unable to do this. Nonetheless, we can certainly describe what they know in this way. The child knows what it thinks—viz., that the dog is loose—despite not knowing (not even thinking) that it thinks this. I will be accused of playing fast and loose with the scope of knowledge attributions. It may be intelligible—even true—to say that Suzy knows something about the proposition—that the dog is loose—she believes, but that is not the same as saying that Suzy knows what she believes. That would be like saying that Suzy knows the answer to a difficult mathematical problem—that it is, say, 24—simply because she knows what number is written on the board (viz., 24) when 24 happens to be the answer to the problem. Suzy may know that the number 24 is on the board, but unless she knows this number under the description, "the answer to the problem" she doesn't know what the answer to the problem is. Likewise, Suzy may think that the dog is loose, and she may know this proposition—that the dog is loose—under some description or other (perhaps as "what I told Mommy"), but unless she knows it under the description "what I think" she doesn't know what she thinks. All Suzy really knows is what she told Mommy. What she told Mommy is what she thinks, of course, but she doesn't know this. 5 This is a fair objection. At least it is an objection that any philosopher who has come this far is likely to make. So I concede the point. It is why I said at the beginning that consideration of children, people who lack the concept of thought, would not get me quite where I wanted to go. It only gets me to the point of having established that there is some sense in which children can know what it is they think without knowing that they think it. They can know what they think without knowing they think it in the same sense you can know what my brother is doing without knowing it is my brother doing it. But this, as my critic has been quick to point out and as I am now willing to concede, only shows that Suzy knows of what she thinks that it is that the dog is loose (the phrase "of what she thinks" kept carefully outside the that-clause that fixes what it is Suzy really knows. It does not show that Suzy knows that what she thinks is that the dog is loose (the phrase "what she thinks" here occurring inside the scope of the knowledge attribution). So it does not, not in any relevant sense, show

5

This point could also be expressed by talking about the difference (in the context of knowledge or belief attributions) between attributive vs. referential uses of descriptions. Following Boer and Lycan's (1986, p. 18) description of the difference between a referential sense of knowing who the murderer is (in which it is not necessary to know the murderer murdered anyone) and the attributive sense (where this is necessary), we could say that children know what they think in the referential sense (where it is not necessary to know they think it), not the attributive sense (where it is necessary to know this).

Knowing What You Think vs. Knowing that You Think It

393

that Suzy knows what she thinks—much less that she knows what she thinks without knowing she thinks it. So, to make the next step in my argument, let me shift to a person who, unlike a child, possesses the relevant concepts and the relevant beliefs. I will describe an analogous situation—knowing what someone said—and suggest that it provides a plausible model for knowing what one thinks. Clyde gets a telephone call from his good friend Harold. Harold tells him that he is going on vacation for two weeks. Clyde hears him say this and, let us suppose, hears him say it under ideal telephonic conditions (no static, clear articulation, etc.), the kind of conditions that would ordinarily prompt us to say that Clyde knows what Harold said. So far, I hope, there is nothing suspicious. Now the twist. There are several people, all practical jokers, who, quite unknown to Clyde, enjoy telephoning Clyde and imitating Harold. They are very good at it. As far as Clyde can tell, the call he received from Harold could have been from any one of these other people. It sometimes is one of these other people, Unaware of the past deceptions, and, therefore, the very real present possibilities they create, Clyde not only believes (correctly as it turns out), without doubt or hesitation, that it is Harold he is talking to, but (incorrectly as it turns out) that he knows it is Harold. 6 I have Gettierized7 Clyde's belief that it was Harold on the phone while leaving intact his evidence for what it was that Harold said to him. The question I'm interested in is this: does the fact that Clyde does not know that it was Harold who said he was going on vacation mean that he doesn't know what Harold said to him? If asked ("What did Harold say?"), Clyde will tell you, confidently and truthfully, exactly what Harold said. And if asked whether he knows—and, if so, how he knows—that this is what Harold said, Clyde will tell you, once again confidently and (I submit) truthfully, that he knows this because he heard Harold say it. If anyone ever knows what another person says on the phone, Clyde, given the circumstances, surely, knows what Harold said. Yet, Clyde doesn't know it was Harold who said it. Clyde thinks he knows. This, indeed, is why he so confidently reports what he knows by referring to the caller as Harold. But the truth of the matter is that Clyde is ignorant about who called him. Unlike the earlier case of the child, we now have an example in which the agent, Clyde, does understand the phrase ("what Harold said") being used to pick out the proposition that he heard expressed on the phone. He not only understands it, he confidently (and truly!) believes it refers to what he heard. That is why he describes what he heard the caller say as "what Harold said." Unlike the case of the child who does not believe, does not even understand, that "what I think" (when said or thought by her) is a correct description of the mental state whose content (viz., that the dog is loose) she has special access to, Clyde does understand—indeed, he truly and confidently believes,

6

7

I here assume that if Clyde can't tell the difference between Harolds voice on the phone and the voices of several other people, any one of which might be calling, then, whether or not he realizes it, he doesn't know it is Harold. He certainly can't hear that it is Harold. That is, I have described conditions in which Clyde has a justified true belief (that it is Harold he is talking to) that does not constitute knowledge.

394

Self-Knowledge

that "what I heard Harold say" is a correct description of the content he has special access to, the proposition he heard expressed on the telephone. Why isn't this enough to know not (once again) that it was Harold who said he was going on vacation, but that what Harold said was that he was going on vacation? If it is enough, then, it seems, we have an attractive externalist model of introspection. Just as Clyde can know (by hearing) what it was Harold said without knowing, at least not by hearing, that it was Harold who said it, why can't a person know what it is he thinks (by, say, introspection) without knowing, not by introspection, that he thinks it?8 I will be told (I have been told) by impatient skeptics that I am still ignoring subtle, but nonetheless quite valid, distinctions of scope. Since Clyde doesn't know it was Harold who said he was going on vacation, it would be wrong to put the phrase "what Harold said" inside the that- clause that expresses what Clyde knows. Clyde doesn't know that what Harold said is that he is going on vacation. All he really knows is something of or about what Harold said—that it is that he is going on vacation. We can say that Clyde knows what the caller said, and the caller was Harold, but since Clyde doesn't know the caller was Harold, he doesn't know what Harold said. Clyde believes the caller was Harold, and he is (we may suppose) fully justified in this belief, but (given the special circumstances) he doesn't know it. So it would be wrong to describe him as knowing that what Harold said is that he was going on vacation. Clyde doesn't—not really, not strictly—know what Harold said. I don't believe in this real, this strict, form of knowledge9, but I'm willing, once again, to concede the point to those who believe that, strictly speaking, nothing belongs inside the scope of a knowledge attribution that isn't known to belong there by the agent to whom the knowledge is ascribed. I am, after all, interested in convincing even a stubborn internalist that there is a workable—indeed, a plausible—externalist model of self-knowledge. So I'll work with what I'm given. My third (and final) example—a slight variation on the second—is meant to comply with these more demanding strictures on scope.

8

9

Strictly speaking, the analogy with knowing what you think vs. knowing that you think it should contrast knowing what Harold said with knowing that he said it—not, as I have done, with knowing that it was Harold who said it. With a few minor alterations this could be done. All we need to imagine are things—programmed sound synthesizers, for example, or (thanks to Doug MacLean for this suggestion) parrots—that can produce the same sounds as Harold when he says that he is going on vacation without actually saying or asserting anything. I assume that parrots and machines who make the sounds, "I am going on vacation" are not actually saying they are going on vacation. They utter the words (and, therefore, perhaps, in direct discourse say "I am going on vacation"), but they do not, by producing these sounds, say, as Harold does by intentionally producing these sounds, that (indirect discourse) they are going on vacation. When Clyde hears Harold on the phone saying that he is going on vacation, therefore, he can know (by hearing) what Harold said without knowing (at least not by hearing) that anything was said. Substituting birds for impersonators of Harold as relevant possibilities, it could (for all Clyde can tell) be a parrot he heard on the phone. I have chosen to run the analogy as I have in the text because it is simpler and more intuitive and it makes the point equally well. The important point, once again, is that the way you know the χ is y may be, and often is, quite different from the way you know that it is χ that is y. Readers familiar with my views on contrastive statements (Dretske 1972), closure (Dretske 1970, 1971, and forthcoming) and the incremental character of perceptual claims (Dretske 1969, Chapter III) will understand why I don't accept it.

Knowing What You Think vs. Knowing that You Think It

395

We have to remember that we needn't suppose that Clyde can know what Harold said without knowing it was Harold who said it. All I need to show is that the way Clyde knows it was Harold who said it can be different from the way he knows what Harold said. If this is possible, then the way is clear to concede (to the yet unconverted) that although you cannot know what you think without knowing that you think it, the way you know what you think may be entirely different from the way you know that you think it. You can know what you think by introspection, but introspection may not the way you know you think it. To know that you are thinking about peanuts may require—or so externalists are free to maintain—a more indirect method, a method (empirical investigation?) compatible with an externalist theory of what it takes to think about peanuts. All that introspection tells you is that it is peanuts you are thinking about.. To illustrate this possibility we need imagine only a small modification of our last example. Clyde (overly suspicious from too much philosophy) finds out who called him by tracing the call. He discovers that the call originated from a phone to which only Harold had access. So he knows it was Harold who called him. He knows both what Harold said (viz., that he was going on vacation) and that it was Harold who said it, but his way of knowing the one is different from the way he knows the other. He heard what Harold said, but he did not—indeed (given the sound-alike impersonators) could not—have heard (by listening to his voice on the phone) that it was Harold who was saying it. Given the conditions, there is nothing distinctive about Harold's voice to enable one to know, by hearing him talk on the phone, that it is Harold. Clyde knows it was Harold (who said that he was going on vacation) as a result of an empirical investigation. But no investigation was required to find out what Harold said. Clyde heard him say he was going on vacation. Using this slightly modified example as a model for introspection, then, the proposal is that the way we find out what it is we think (desire, wonder, fear, expect, etc.) is different from the way we find out that we think (desire, wonder, fear, and expect). The first method we call introspection. Whatever, exactly, introspection comes down to, it does not involve empirical investigation of external circumstances. That is what makes it zrairospection. But this is quite consistent with an empirical investigation being required to find out that the content revealed by introspection is the content of a mental state, a state whose possession of content is constituted, in part, by a network of external relations. Given our model, this should be as sensible as saying that Clyde heard what Harold said but needed an investigation (tracing the telephone call) to find out it was Harold who said it. Or, to give an example that might appeal to baseball fans, it is like needing a program to find out that it was Lou at bat, but not needing a program to know that Lou hit a home run. You saw him hit a home run. You know what he did by direct perception. But you know who did it—that it was Lou—indirectly, by looking at your program. We have, then, the following picture of self-knowledge: when thinking about peanuts, we can know, with a special kind of first person authority, what it is we are thinking about—peanuts. Nothing illicit is smuggled into the scope of the knowledge attribution since we know both that it is peanuts we are thinking about and that we are thinking about them. So we can, in both word and thought, and with full knowl-

396

Self-Knowledge

edge, pick out and refer to what we are thinking about — peanuts—as something we are thinking about. Nonetheless, our way of knowing that it is peanuts we are thinking about is, or it may be, quite different from the way we know that we are thinking about them. Although we enjoy first person authority about the first fact, we may enjoy no privileged access to the second fact. It may be, as externalists have it, that to know you are thinking about peanuts requires information not obtained by looking exclusively inward. Introspection doesn't tell you that you think, only what you think. If this is right, we have a viable theory of self knowledge. Special authority about, and privileged access to, one's own thoughts and experiences is compatible with an externalist theory of thought and experience. The only remaining question is whether this theory gives, besides a coherent account of self-knowledge, a plausible account of self-knowledge. Is this really all that, introspection yields? Do we, in fact, use a different method to find out that we think from the method (if it is a method) that tells us what we think? My purpose was only to argue for the intelligibility of a hypothesis, not to argue for its truth. Since I've already done this, I should quit now. Nonetheless, I cannot resist adding a few remarks—three will do—about the plausibility of this picture of the mind's knowledge of itself. As the above examples show, we often know that χ is y by some direct method (hearing, seeing, introspection) without being able to know, by that same direct method, that it is χ that is y. If we know that it is χ that is y, our way of knowing this may be, and often is, quite different from our way of knowing that χ is y. I don't have to see, even be able to see, that it is water that is boiling to see that the water is boiling. There is, after all, nothing about water to distinguish it from gin, vodka, and a variety of other liquids. I needn't be able to see that it is water in order to knowingly refer to what I see to be boiling as water. So if it is water, and if I reasonably and truly believe it is water, there is nothing to prevent me from saying I can see that the water is boiling. That, I submit, is how I know that the water is boiling. If I know it at all, though, that isn't how I know it is water that is boiling. If I actually know it is water, I probably know that in some way other than the way I know it is boiling. The fact that I came to know (or believe) that it is water by chemical analysis doesn't mean I can't see that the water is boiling. Why shouldn't the same be true of /«inspection. The fact that I found out I think by having someone (parents? teachers? friends? Descartes?) tell me doesn't mean I can't now discover what I think by simple introspection. The analogy with ordinary perception can, I think, be pushed a little farther with, I hope, illuminating results. Perception of ordinary dry goods tells us what is in the physical world, not that there is a physical world. I see that there are cookies in the jar, people in the room, and (by the newspapers) continued violence in the Middle East. That is how I know there are cookies, people, and violence in these places. Cookies, people, and violence are physical things, things that exist independently of my perception of them. Do I, therefore, know, by perception, by seeing, that there are things that exist independently of my perception of them? Can I see that there is a material world and that, therefore, solipsism is false? I don't think so. It seems more reasonable to say that

Knowing W h a t You T h i n k vs. Knowing that You T h i n k It

397

assuming there is a physical world, or assuming we know (in some other way) that there is such a world, perception tells us what sorts o f things are in it—cookies, people, and violence. Visual perception has the job o f telling me what (physical objects) I see, not that I see physical objects. I f my perceptual faculties had the latter job, the job o f telling me that I was (in effect) not hallucinating, not aware o f some figment o f my own imagination, then they would be incapable o f discharging their responsibilities. For, as we all know, hallucinatory cookie jars can, and sometimes do, look much the same as real cookie jars. You can't see the difference. I f it's a real object you see, perception can tell you whether it's an orange or a banana (a difference that is plainly visible), but perception cannot tell you whether it's a real orange or just a figment o f your imagination. That difference isn't visible. Memory has a similar structure. Memory tells us what happened in the past—the specifics, as it were, o f personal history. It does not tell us there is a past. I can remember (hence, know) what I had for breakfast this morning. No trick at all. I distinctly remember that it was granola. Nonetheless, despite this implying that the past is real (if it isn't real, I didn't have breakfast this morning; hence, do not remember having granola for breakfast this morning) this doesn't mean I can remember that the past is real. I f I know the past is real, I don't know this by remembering that it is real. That isn't a way to answer Russell's skeptical question about the past. 10 I f I know the past is real, I know it in some way other than by memory. Memory is a faculty that tells me what occurred in the past given that there was a past just as perception tells me what is in the material world given that there is a material world. Maybe I have to know the past is real in order to remember what I had for breakfast this morning (I doubt it, but let it pass), and maybe I have to know there is a physical world to see whether there are cookies in the jar (let that pass too), but the point is that I do not have to know these things by memory and vision in order for memory and vision to tell me (give me knowledge of) what I had for breakfast this morning and what is in the cookie jar. Introspection is like that. Introspection tells me what is in my mind, what it is I am thinking, wanting, hoping and expecting. It doesn't tell me I really have a mind, mental states with such content. I f I know that at all, I know it in some way other than by introspection, the faculty that, given that I have thoughts and feelings, tells me what I'm thinking and feeling. This account o f introspection also squares neatly with a representational theory o f mind. This won't persuade anyone who isn't antecedently sympathetic with a representational scheme, but it may tip the balance for those who are. Think about a simple representational device, a measuring instrument, say, whose job it is to provide information about some magnitude—temperature, say. A pointer registers 70° on a scale calibrated in "degrees Fahrenheit" and, thereby, represents the

10

Russell's question: H o w do you know the world and all its contents were not created a few m o m e n t s ago complete with m e m o r y traces, fossils, history books, etc.—complete, that is, with all the indicators you rely on to tell you about the past?

398

Self-Knowledge

temperature to be 70° F. In very crude terms, and setting aside for the moment the difference between a conventional (in which the indicator functions derive from us) and a natural (where they derive from evolution or learning) forms of representation, this state of the thermometer (the pointer pointing at "70") is the thermometer's "belief" that the temperature is 70° F. The instrument occupies a state that represents, says, or means (in a way that might be false) that the temperature is 70° F. Even if we refuse (as we should) to credit instruments with knowledge (since knowledge of the sort now in question implies belief, and instruments do not have beliefs), we can ask what kind of information they have about their own representational states. Does the thermometer have information about how it is representing the temperature? Does it have the information that it is representing temperature? If thermometers knew things, could it know it was representing the temperature to be 70° F? With respect to temperature, this thermometer is like the child, Suzy (described above), with respect to the dog's whereabouts. The instrument "thinks" the temperature is 70°, but it has no concept of representation. It, therefore, has no way to represent itself as thinking about (i.e., representing) temperature. It has been given the power of thinking about temperature (this is what it reports on) but it hasn't been given the power to think about (i.e., represent) itself as thinking about temperature. It reports on temperature, not on itself representing temperature. Nonetheless, though it does not represent itself as representing temperature, it does carry (in a somewhat degenerate sense) information about the way it represents temperature. The way we would find out how the instrument is representing the temperature, after all, is by looking at the instrument. Is the pointer pointing at the numeral "70" on a scale suitably calibrated in "degrees Fahrenheit?" Yes? Then it is representing the temperature to be 70° F. The actual temperature may not be 70° F, of course, but if the pointer is pointing at "70," then the instrument is saying, perhaps falsely, that the temperature is 70° F. If that is where we look to see how the instrument is representing the temperature, to see what the instrument "says" or "thinks" about temperature, then that is where the instrument itself should look if it wants to find out what it "thinks" about the temperature. In that sense, if the instrument wants to know what it thinks, it should look inward, at itself. It should introspect. That's where the information is about what it thinks, about the way it is representing the world. But though the instrument, given that it is representing temperature, necessarily has information (in its own intrinsic state—i.e., pointer position) about how it is representing the temperature, about what it is representing the temperature to be, it is not so clear that it has—in fact I think it clearly does not have—the information that it is representing temperature or, indeed, representing anything at all. If somebody wanted to be sure that this device was a thermometer (something that actually represents temperature) and not a badly designed wind gauge, a door stop, or an object d'art (things that do not represent temperature) she clearly shouldn't look at it. Looking at it wouldn't provide the information because, typically, there is nothing about a device that tells one what its purpose or function is. To find out what the purpose or function of an object is, or whether it even has a purpose or function, one has to know (in the case of artifacts) the intentions and purposes of its designers and builders or, perhaps, the way the object is normally used. These facts though, are extrinsic, relational, facts, that are

Knowing W h a t You T h i n k vs. Knowing that You T h i n k It

399

not, or need not, be evident in the object itself. There is nothing about a gauge itself, its intrinsic properties, that tells you whether it is a pressure gauge, an altimeter, a badly designed (absolutely dysfunctional) speedometer, or a fancy paperweight. 11 If this is correct (I argue it at greater length in Dretske 2003b), then every representational device necessarily has, in its own intrinsic state, information about how it represents the world, but little—-possibly no—information that it represents the world. It can look inward to see what it represents the temperature as—about this subject, the content of its representational states, it has privileged and authoritative access—but it has to look outward to find out whether, in fact, it represents anything at all. That, I submit, is an accurate picture of self knowledge, and it is, I remind you, a picture that externalism recommends.

References Boer, S. E. and Lycan, W. 1986. Knowing Who. Cambridge, MA; M I T Press, A Bradford Book. Dretske, F. 2003a. " H o w D o You Know You Are N o t A Zombie?" in Privileged Access and First Person Authority, Brie Gertler, ed. Ashgate Press. Dretske, F. 2003b. "Externalism and Self -Knowledge," New Essays on Semantic Externalism and Self-Knowledge, S. Nuccetelli, ed., Cambridge, MA, M I T Press. Dretske, F. forthcoming (c). " T h e Case Against Closure," Contemporary Debates in Epistemology, Matthias Steup and Ernest Sosa, editors. Blackwell Publishers. Dretske, F. 1969. Seeing and Knowing. Chicago; University of Chicago Press. Dretske, F. 1970. "Epistemic Operators Journal of Philosophy, LXVIII, 24; 10097-1023. Dretske, F. 1971. "Conclusive Reasons," Australasian Journal of Philosophy, 49.1: 1-22. Dretske, F. 1972. "Contrastive Statements," Philosophical Review, LXXXI, 4: 4 1 1 - 4 3 7 Ludlow, P. and N . Martin, eds. 1998. Externalism and Self-Knowledge. Stanford, CA; CSLI Publications.

11

T h e instrument may, of course, have a label that describes its function—ALTIMETER—and a scale calibrated in dimensions ("feet above sea level" rather than "pounds/square inch") that serve to identify what the instruments function is. In this case an intrinsic property (the label) indicates an extrinsic property (the instrument's function). Brains, though, don't come with labels, and even if they did, we would have to be able to see the labels (are we aware of our brain when we introspect?) and figure o u t what these labels said (another extrinsic fact) in order to come to know in this way,

Do We Know how We Know Our Own Minds yet?

£

PIERRE J A C O B

Introduction: the incompatibilist argument One aspect of the contemporary philosophical situation is puzzling. O n the one hand, few if any of the features of the special epistemic authority granted by both the traditional empiricist and the traditional rationalist pictures of introspective self-knowledge have survived recent philosophical scrutiny. O n the other hand, several philosophers—the incompatibilists—assume that the alleged special epistemic authority granted to introspective self-knowledge by traditional epistemology can bear the burden of an argument against content externalism. In response, several externalists have tried to argue for the compatibility between content externalism and the alleged special epistemic authority of introspective self-knowledge. 1 Content externalism is the view that the content of an individual's thought, prepositional attitude and perceptual experience does not (always) supervene only upon the internal cognitive resources of the individual. 2 Nor does it (always) supervene only upon the internal physical, chemical and biological properties of the individual's brain. Content externalism comes in two broad varieties: social and non-social externalism. According to the latter, the content of an individual's mental representation may depend upon the individual's non-social environment. According to the former, it may also depend upon what other members of her community think. Notice that social externalism seems tailor-made for the conceptual contents of an individual's thoughts and propositional attitudes, not for the nonconceptual contents of the individual's perceptual experiences. 3 In a nutshell, the argument put forth for the incompatibility between content externalism and the special epistemic authority of introspective self-knowledge has the general following structure (see e.g., McKinsey, 1991 and Boghossian, 1997). If introspective self-knowledge has special epistemic authority, then content externalism cannot

*

1 2

3

I am very grateful to Fred Dretske for extensive email exchanges about his views on self-knowledge. I also wish to thank Gabriele Usberti and Kirk Ludwig for extensive and penetrating comments o n this paper, Max Kistler for a useful conversation and audiences at both C U N Y and the university of Gainesville for illuminating comments. See e.g., Bürge (1988) and Davidson (1987). I simply assume without argument both a representational view o f the mind and the distinction between the conceptual content of thoughts and propositional attitudes and the nonconceptual content of perceptual experiences. What an individual thinks and believes may depend on what members of his community think and believe. But I assume that what an individual experiences does not depend on what members of his community think, believe or experience.

402

Self-Knowledge

be true. Introspective self-knowledge has special epistemic authority. Therefore: content externalism cannot be true. In a little more detail, the incompatibilist argument assumes that one can know with special epistemic authority that one believes that, for example, water is a liquid. Let us say that one knows a priori that one believes that water is a liquid. But one could not believe that water is a liquid unless one had the concept WATER. 4 It follows that one knows a priori that one has the concept WATER. According to content externalism, however, one could not have the concept WATER unless one stood in some appropriate relation to water. It follows that one can know a priori that one stands in relation to water and thus that there is water in one's environment. But this seems incredible: one cannot know a priori that (or have special epistemic authority over whether) ones environment contains water, which is, according to content externalism, necessary for having the concept WATER. Whether one's environment contains water (not something else) is not something one can know a priori. In summary, two assumptions seem needed for the incompatibilist conclusion that content externalism cannot apply either to the concept WATER or to the belief that water is a liquid. First of all, one must know a priori with special epistemic authority that one believes that water is a liquid. Second of all, the concept WATER must make the same contribution to the simpler content (or truth-conditions) of one's first-order belief that water is a liquid and to the more complex content (or truth-conditions) of one's introspective higher-order belief that one believes that water is a liquid. 5 The reason I find the contemporary situation perplexing is that I take externalism about the contents of an individual's first-order thoughts about the world to be more plausible-—not less plausible—than anything we may think about introspective selfknowledge. On the one hand, content externalism—at least non-social externalism—is a doctrine about first-order human mental representations of the external world. It is a view about the processes—some of which may be common to humans and to nonhuman animals, e.g., perception and memory—which allow humans to achieve some knowledge of the external world. On the other hand, introspective self-knowledge consists in higher-order representations about first-order mental representations of the world. It is at least conceivable that a creature might have the cognitive resources required for forming reliable beliefs about the external world, even though it lacks the cognitive resources for forming introspective beliefs about its own beliefs about the world. 6 Imposing top down constraints on the contents of first-order mental representations about the world from assumptions about the alleged epistemic status of introspective self-knowledge sounds to me like putting the cart before the horse. 7

4 5 6

7

I use words in capital letters to refer to concepts. I shall come back to the second assumption in the conclusion. In order not to beg any question in favor of epistemological externalism and against epistemological internaiism, I purposefully put the last point in terms of reliable beliefs, not knowledge, of the external world. An epistemological externalist might want to claim, and an epistemological internalist might want to deny, that this is sufficient for knowledge of the external world. It sounds preposterous to impose internalist epistemological constraints, not on knowledge of the world, but on the contents of beliefs about the world.

Do We Know how We Know Our Own Minds yet?

403

The idea that introspective beliefs about facts involving one's own psychological properties enjoy a unique epistemic authority or privilege has played a different role in traditional rationalist epistemology and in traditional empiricist epistemology. In rationalist epistemology, the primary target of introspective self-knowledge are thoughts. In empiricist epistemology, the primary target of introspective self-knowledge are sense-data or perceptual experiences. Whether one and the same mechanism—introspection—could satisfy both rationalist and empiricist desiderata is far from clear. On the one hand, in rationalist epistemology, psychological self-knowledge was taken to be the paradigm of both a priori and infallible human knowledge. Rationalist epistemology has three ingredients. First, it is of the essence of the Cartesian mind both that it entertains or forms thoughts (as opposed to having e.g., perceptual experiences). Secondly, thoughts have concepts or ideas as constituents. Thirdly, the mind is transparent to itself: one cannot have a thought of which one is not aware. In rationalist epistemology, what secures the a priority and infallibility of one's introspective awareness of one's psychological properties is that all the psychological properties a mind can exemplify are properties of thoughts or judgments, not experiences. Given that the mind is transparent to itself or that thoughts are reflexive in the sense that one cannot entertain a thought (or make a judgment) unless one is aware that one is, it follows that introspective knowledge of one's own mind is a priori and infallible. In empiricist epistemology, on the other hand, the most primitive and elementary constituents of minds are perceptual experiences or sense-data, not concepts. Concepts (of either psychological or non-psychological properties) are logical constructions out of sense-data. According to much traditional empiricist epistemology from Locke to Russell, knowledge of the external world—knowledge of mind-independent facts—is twice dependent on psychological self-knowledge. First of all, knowledge of mind-independent facts depends on the epistemologically antecedent knowledge of mental or psychological facts about oneself (such as that one is having a particular perceptual experience, sense-datum or idea). Secondly, one's knowledge of mental or psychological facts about oneself derives in turn from one's direct quasi-perceptual acquaintance with some mental entity present to or in one's mind, i.e., the sense-datum or perceptual experience. If one's awareness of one's sense-data consists in being acquainted with them, then one is made aware of one's sense-data by some kind of quasi-perceptual process or peering inside at one's own perceptual experiences. Both the rationalist and the empiricist pictures of introspective knowledge have come under serious criticism in contemporary philosophy. On the one hand, the empiricist model of a quasi-perceptual process whereby one becomes self-aware of one's own perceptual experiences raises at least three issues. First of all, if one's knowledge of psychological facts about oneself derives from some quasi-perceptual acquaintance with one's own perceptual experiences, then it is questionable whether self-knowledge can still meet the epistemic requirements of a priority and infallibility. Secondly, as Shoemaker (1994) and other philosophers have noticed, whereas vision, audition, olfaction, touch and proprioception can be used to get information about mind-independent facts (involving one's own body and the bodies of others), no inner sense organ provides information about one's own perceptual experiences, let alone about one's thoughts. Finally, as Harman (1990), Tye (1992) and many others have noticed, perceptual ex-

404

Self-Knowledge

periences are introspectively transparent. In other words, the phenomenology of the introspection of e.g., one's own visual experience of e.g., a bush of blue lavender just is the phenomenology of one's visual experience of a bush of blue lavender. What it is like to introspect and to be aware of one's own visual experience of a bush of blue lavender is nothing but what it is like to have the visual experience of a bush of blue lavender. Presumably, if introspection of one's visual experience of anything involved some quasi-perceptual process, then introspective awareness of one's visual experience would have a phenomenology of its own—in addition to and above that of the visual experience itself. O n the other hand, the Cartesian picture of introspection raises at least two issues. First, the asymmetry between first-person and third-person mindreading that results from a Cartesian picture of introspective self-knowledge raises a genuine puzzle. As Davidson (1984, 1987) recognizes, the asymmetry between first-person and third-person mindreading takes it for granted both that claims to know one's own mind are made independently of any empirical evidence and that they enjoy an epistemic authority of which third-person claims to know the minds of others are deprived. The puzzle is: why should claims without evidential support have more epistemic authority than claims based on evidence? The second question is: given that the Cartesian assumption that the mind is transparent to itself has come under heavy attack, what is left of the Cartesian picture of the special epistemic authority of introspective beliefs? Since Freud, it is commonly accepted that one may be blind to some of one's own beliefs and desires. Furthermore, a human mind does not merely entertain thoughts and prepositional attitudes; it also has perceptual experiences. On the Cartesian picture, the immunity to error of the mind's introspective beliefs about itself was secured by the joint assumptions that it is of the essence of the mind to entertain thoughts and that one cannot entertain a thought unless one knows that one is doing so. But how could such assumptions entail that one's introspective beliefs are both exhaustive and infallible? How could such assumptions secure infallible introspective beliefs about one's perceptual experiences at all? How could they secure infallible introspective beliefs about each of one's propositional attitudes—both one's occurrent prepositional attitudes and one's dispositional propositional attitudes? 8 In the first section of the paper, I sketch Fred Dretske's (1995) view of a restricted subset of the set of one's introspective beliefs, i.e., one's introspective beliefs about one's own perceptual experiences, based on the model of displaced perceptual knowledge. In the second section of the paper, I examine the question whether the naturalistic view of the contents of first-order mental representations on which it is based has the resources to accommodate the contents of introspective metarepresentations of one's first-order mental representations. Finally, in the third and last section of the paper, I argue that it is a mistake for an externalist to accept wholesale the premisses of the incompatibilist argument and to try to accommodate externalism to these premisses.

8

These points are made by Boghossian (1989).

Do We Know how We Know Our Own Minds yet?

1. The Displaced Perception Model of Introspective

405

Self-Knowledge

In this section, I will sketch what is to my mind a very plausible externalist account of the introspective process whereby one comes to form introspective beliefs about one's own perceptual experiences, i.e., Dretske's (1995) displaced perception model of introspection. What I call a little misleadingly 'the displaced perception model of introspective beliefs' has really two ingredients: the theory of displaced perceptual knowledge proper and the general informationally based teleosemantic (henceforth, IBT) account of the contents of first-order mental representations of the world (from Dretske, 1988, 1995), which I will presently sketch very briefly. According to IBT, no system can represent anything unless it has a function (a design or a purpose), i.e., the function to indicate or carry information about the presence of some property, e.g., property F. A system could not indicate the presence of property F unless it were correlated with property F. Carrying information about property F, however, is necessary but not sufficient for representing property F. Unless a system has the function to carry information about F, it cannot misrepresent, and hence represent F: if a system has the function to carry information about F, then it can represent someting as F even though it fails to carry information about F because what it represents as F may fail to be F. Importantly, the IBT account of mental content entails a principle, which I shall dubb "the principle of the reflexivity of content", and which can be formulated thus (see Dretske, 1995: 52): Reflexivity of content: A system cannot represent things to be F without carrying information about it's being F that it is representing. System 5 cannot represent something to be F unless it has the function to carry information about F. If 5 represents correctly something to be F, then it does carry information about F. Suppose now that S represents incorrectly χ to be F. Since it is 5s function to covary with F, in misrepresenting χ as F, S is correlated with the property that would be instantiated were S performing its function according to its design. This property is no other than F. Thus, even if S incorrectly represents χ as F, still S carries information about property F, i.e., the property that would be instantiated if S were doing its job properly. It follows that by representing something (whether correctly or incorrectly) to be F, a device has available information about it's being F that it is representing. The principle of the reflexivity of content is, as we shall see in section 3, an important step in Dretske's (2003) reply to the incompatibilist argument. Here, I just note that the principle of the reflexivity of content entails that a representational device has available information about which property it is representing. Now for a device to have such available information is not the same thing as knowing that it is representing something as F. A representational device has available information about the content of its representation, but it does not thereby know what it is doing. To know the latter, it must be able to represent the fact that what it is doing is representing. Unless it has the concept REPRESENTATION, a device cannot represent the fact that what it is doing is representing.

406

Self-Knowledge

I now turn to displaced perceptual knowledge. 'Displaced perception' is Dretske's (1995) word for what Dretske (1969) called 'secondary epistemic perception'. I shall give a few examples. You hear the dog bark. You thereby come to believe that the dog barks. The dog would not bark unless there was someone at the door. You believe that the dog would not bark unless there was someone at the door. You thereby come to believe that someone is at the door. By hearing the dog bark, you thereby hear that someone is at the door. You see hoof prints in the snow at t. You thereby come to believe that there are hoof prints with a particular shape in the snow at t. There would not be hoof prints with such a particular shape in the snow at t unless a horse had walked on the snow at t — 1. You believe that there would not be hoof prints with such a particular shape in the snow at t unless a horse had walked on the snow at t — 1. You thereby come to believe that a horse walked on the snow at t - 1. Although you did not see the horse at t - 1, by seeing the hoof prints at t, you see that a horse walked on the snow at t — 1. One can represent the general structure of displaced perception in the following sequence of steps. S has displaced perceptual knowledge of the fact that object o' is G iff (1) (2) (3) (4) (5) (6)

ο 5 S ο S S

is F (intermediate fact). has a perceptual experience of o's being F. believes that ο is F (intermediate belief). would not be F unless o' were G (correlation between facts). believes (4) (connecting belief). believes that ο is G (from (3) and (5)).

In order to extend the model of displaced perceptual knowledge to introspective knowledge, I shall introduce some of Dretske's own terminology. Step (2) results from step (1) as a matter of perceptual psychology. With Dretske (1969, 1978), we may call nonepistemic or simple perception step (2) and primary epistemic perception step (3). Only a creature with some concept of property F could move from step (2) to step (3). In Dretske's (1995) terminology, the fact that ο is F is the intermediate fact. 5"s belief that ο is F is S s intermediate belief. (4) is a correlation between two distinct facts and (5) is Ss connecting belief. The fact that S comes to believe via displaced perception—the fact that o' is G—is the target fact. Consider now the application of the model of displaced perceptual knowledge to introspective knowledge. Suppose that one has the visual experience of a triangular object: one has the visual experience of ο as triangular. S has introspective knowledge that S has the visual experience of a triangular object iff (1*) (2*) (3*) (4*) (5*) (6*)

ο ο 5 ο S 5

is F looks F to 5. believes (2*). would not look F to S unless S had a visual experience of ο as F. believes (4*). believes that S is having a visual experience of ο as F (from (3) and (5)).

D o Wc Know how We Know Our O w n Minds yet?

407

As Dretske (1995: 60-61) notes, there is one disanalogy between displaced perceptual knowledge and introspective knowledge. One cannot come to believe that o' is G by seeing that ο is F unless ο is F. In other words, displaced perceptual knowledge that o's is G requires that (1) obtains—that ο indeed be F-— and that 5s intermediate belief (3) that ο is F be true. By contrast, 5s introspective belief that 5 is having a visual experience of ο as triangular might be true even though 5s visual experience of ο as triangular is non-veridical. 5s introspective belief that 5 is visually experiencing .Fdoes not require either that (1 *) obtains or that 5 correctly believes that ο is F. Whether (1 *) obtains is optional, for (2*) may obtain although (1*) does not. 5 may falsely believe that object ο is triangular even though ο might not be a triangle at all. Object ο may look triangular to 5 (as in (2*)) even though either ο is not really triangular or there is no object at all: either 5 may misperceive object ο as triangular or 5 may have a visual hallucination of a triangular object. Two features of the displaced perception model of introspective knowledge of one's perceptual experiences are worth emphasizing. First, it is an externalist account since one comes to learn facts about one's own perceptual experiences by having the experiences. Granted, the experiences need not be veridical. But on the assumption that one would not have any visual experiences at all unless natural selection had provided the human visual system with the function to carry information about the visual attributes of objects instantiated in the environment of ancestors of humans, it follows that one's non-veridical visual experiences are parasitic on one's veridical visual experiences. O n e could not either misperceive a triangle or have a visual hallucination of a triangle unless one could visually perceive triangles. If so, then one learns facts about one's own perceptual experiences by perceiving mind-independent objects, properties and facts in the external world, not by experiencing—or by peering at—one's own perceptual experiences (as the traditional empiricist model of self-knowledge would have it). Secondly, according to the displaced perception model of introspective knowledge of one's own perceptual experiences, one comes to know that one has perceptual experiences by forming beliefs about oneself. Although it is necessary to perceive mindindependent objects, it is not sufficient for introspective self-knowledge. Presumably, one can have visual experiences whether or not one can form the connecting belief (5*). But unless one can form the connecting belief (5*), one could not form the belief that one is having a visual experience. 9 Arguably, one could not form the connecting belief (5*) unless one had the higher-order concept VISUAL EXPERIENCE. Nor could one come to believe, as in (6*), that one is having a visual experience of a triangular object unless one had this higher-order concept. Similarly, one could not come to believe that one has the belief that ο is a triangle unless one had the concept BELIEF.

9

On some interpretation, Rosenthal's (1986, 1993) higher-order thought ( H O T ) theory of conscious mental states would deny that one could have a conscious perceptual experience if one could not form, if not the connecting belief (5*), at least some close cousin of (5*)—a higher-order thought to the effect that one is having a perceptual experience.

408

Self-Knowledge

2. Displaced Perceptual Knowledge, IBT and Metarepresentations The displaced perception model of introspection raises an interesting question about the compatibility of IBT and the appeal to metarepresentations. IBT offers a (presumably naturalistic) account of the contents of first-order mental representations of the external world based on the notions of information and function. No doubt, introspective beliefs about one's own perceptual experiences are higher-order mental representations of one's first-order perceptual representations of the world. Unlike the latter, the former are metarepresentations. Does a full naturalistic account require an extension of IBT to the contents of introspective metarepresentations? If so, can it be so extended? This question has been raised as a challenge to a naturalistic approach to the puzzles of mental content by Kemmerling (1999: 323-24), who writes: "How could such a thing—a metarepresentational belief—show up in Dretske's framework? [....] what is lacking, is an account of how a natural system may come to need information which is specifically about the content of its own ground-floor representations [...] an account of natural metarepresentational systems [ ] a job which is clearly separate from any job of any ground-floor representation". Can the IBT account be extended from the contents of "groundfloor" representations to the contents of introspective higher-order representations? Or should it? My response to Kemmerling's (1999) challenge will come in two steps. First, I will argue that introspective beliefs notwithstanding, displaced perceptual knowledge itself is already infected by metarepresentations. So the challenge could be directed to displaced perceptual knowledge as well: is displaced perceptual knowledge compatible with a naturalistic semantics? Conversely, if displaced perceptual knowledge is immune to the challenge, so should the displaced perception model of introspective knowledge. Secondly, to parody Jerry Fodor (1994), I will argue that the challenge involves a confusion between semantics and epistemology. First, in the previous section, I sketched a simple example of displaced perceptual knowledge in which one sees that a horse walked on the snow at t - 1 by seeing hoof prints in the snow at t. It does not seem to me out of the question at all that many non-human preys and predators are capable of such displaced perceptual knowledge. Now, consider in more detail some of Dretske's (1995) own examples of displaced perceptual knowledge. One comes to learn that the gas tank of one's car is empty by seeing the pointer of the gas gauge. One comes to learn something about the temperature of a liquid by seeing the level of mercury of a thermometer immersed in the liquid. One comes to learn "what is happening on the other side of the world" by reading a newspaper or watching television (Dretske, 1995: 41). I doubt very much that nonhuman preys and predators can come to achieve displaced perceptual knowledge of any of these last three kinds. In the sequel, I will distinguish between metarepresentational and first-order displaced perceptual knowledge. Unless one knows the grammar of the natural language to which the perceived tokens of printed or spoken sentences belong, one will not be able to form the intermediate belief about the propositional content of the linguistic inscriptions or utterances one reads in the newspaper or one hears on television. In other words, unless one knows the grammar of some natural language, one will not grasp what sentences mean or

Do We Know how W e Know Our Own Minds yet?

409

what they are used to say. Here, I will leave aside the question whether knowledge of the grammar of some natural language or other is necessary for one either to learn that the gas tank in ones car is empty by seeing the gas gauge or to learn what the temperature of a liquid is by seeing the level of mercury on a thermometer. Rather, I want to emphasize the complexity of the cognitive resources imposed by metarepresentational displaced perceptual knowledge of any kind. In coming to learn that a horse walked on the snow at t - 1 from seeing hoof prints in the snow at t, one merey needs to form first-order beließ about mind-independent facts, i.e., the fact that there are hoof prints in the snow and the fact that a horse walked in the snow. In other words, the target belief, the connecting belief and the intermediate belief are first-order mental representations of mind-independent facts. This is why it is plausible that non-human preys and predators can achieve such first-order displaced perceptual knowledge. Not so in the three examples of metarepresentational displaced perceptual knowledge. In all three examples, the target or displaced belief is a first-order mental representation of some fact.10 But the intermediate belief is not. The intermediate fact about which one must form an intermediate belief in each of the three cases is itself a non-mental representation of some state of affairs, which need not be mental either.11 In order to form a target belief about the level of gas in the tank of one's car, one must form an intermediary belief about the representation of the level of gas yielded by the gas gauge. In order to form a target belief about the temperature of the liquid, one must from an intermediary belief about the representation of the temperature yielded by a thermometer. In order to form a target belief about some event happening on the other side of the world, one must form an intermediary belief about the representational content of some linguistic expression. All three intermediary beliefs are about a representation or the state of some representational device. Of course, if some intermediate belief about a representation is metarepresentational, then so is the antecedent of the relevant connecting belief. Notice that in each three cases, the relevant intermediate belief is not about some intrinsic property of the non-mental representation. One must perceive the shape, orientation and color of either the pointer of the gas gauge or the tube containing the mercury in order to achieve belief either about the level of gas in the tank or the temperature of the liquid. Similarly, one must perceive either the shape, orientation and color of the symbols printed in the newspaper or the acoustic properties of the sound structure of the utterances in order to achieve belief about some event on the other side of the world. But the relevant intermediate belief must be about the content, not the intrinsic local properties, of the representation. It must be about what the representation is about or what it stands for. The perceptual experience of the intrinsic properties

10

11

I do not say 'mind-independent' fact in order not to get entangled in the issue (that is irrelevant to the present discussion) of whether the fact that the gas tank is empty is a mind-independent fact. O f course, the gas tank is part of the car, which is itself an artifact that would not exist if it had not been manufactured by humans with minds. I say purposefully 'state of affairs', not 'fact', for the representation need not be veridical and the state of affairs need not obtain.

410

Self-Knowledge

of the representation gives rise to what Dretske (1969) calls nonepistemic perception. T h e intermediary belief about the content of the representation is epistemic perception that the representation means so and so. Dretske (1995: 42) notes that "perceptual displacement enlarges the number of facts one perceives without a corresponding enlargment of the number of objects one perceives. [ ] O n e see more facts, not by seeing more objects, but by expanding one's knowledge of what the objects one can already see signify about the objects one cannot see. This is what connecting beliefs (e.g., well-confirmed theories) provide". O n the one hand, what Dretske did not emphasize is the extent to which, if I am right, some of his own examples of non-introspective displaced perceptual knowledge do presuppose the power to represent representations as such, i.e., metarepresentational resources. O n the other hand, the metarepresentational resources involved in the above examples of displaced perceptual knowledge are really the tip of the iceberg of the full battery of h u m a n metarepresentational resources. Arguably, unlike other animals, humans derive more, not less, of their first-order representations of the world from the testimony of others (i.e., from verbal communication with their conspecifics) than from perception and memory. For example, an addressee located in Paris would not come to believe that it is raining in San Francisco from his understanding of the speaker's utterance of the sentence 'It is raining' in San Francisco unless the hearer had come to believe that the speaker believes that it is raining in San Francisco and intends to induce in her Parisian hearer the belief that it is raining in San Francisco by means of her utterance. 1 2 Thus, human verbal communication requires third-person metarepresentational capacities, i.e., the ability to do third-person mindreading. From an evolutionary standpoint, the adaptive benefits of third-person mindreading among humans seem indeed quite obvious. N o t only are the beliefs of one's conspecifics a useful source of information about aspects of the world that one cannot directly perceive (displaced perceptual knowledge), but social cooperation is an important source of mutual (social) benefit. Now, if one individual is contemplating the choice between cooperation and competition with another human agent, then it is useful for the former to be able to detect accurately the latter's goals, intentions, desires and vice-versa. Arguably, if one did not know the content of one's own mind, one could not even contemplate the choice between cooperation and competition. If so, then perhaps the metarepresentational resources necessary for introspective self-knowledge ride piggyback on the metarepresentational resources for knowing the minds of others. Alternatively, given that the adaptive advantages of both third-person and first-person mindreading may well stand or fall together, they may have co-evolved in tandem. 1 3 I now turn briefly to my Fodorian claim that semantics is not part of epistemology. T h e task of IBT, or so I claim, is to provide a semantic account of the contents of

12 13

This paper is not trying to clarify the complexity of communicative intentions. Notice that here I am considering the relation between the metarepresentational resources required respectively for introspective self-knowledge and for knowing the minds of others. I am not considering the sort of evidence relevant to each kind of knowledge, let alone the sort of epistemic authority that should be associated with each kind of knowledge claims.

D o We Know how We Know O u r O w n M i n d s yet?

411

first-order (or groundfloor) mental representations of the world. It is neither the task of IBT to provide an epistemic account of their justification, nor to provide an account of how one comes to acquire such higher-order concepts as REPRESENTATION, EXPERIENCE, BELIEF, and so on. From a semantic standpoint, the content of a metarepresentation includes and depends systematically (or compositionally) upon the content of the representation metarepresented. 14 The content of the latter is a proper part of the former. If and when IBT has provided a naturalistic account of the content of the embedded first-order representation, its job is over. Presumably, one can see a horse and have a visual experience of a horse even though one does not know what horses are—even though one does not possess the concept HORSE. O n e cannot, however, believe that there is a horse nearby unless one knows what horses are or one has the concept HORSE. The task of IBT is to offer an account of both the non-conceptual content of a percept of a horse and the content of the concept HORSE. Arguably, one can have either the visual experience of a horse or the belief that there is a horse nearby without believing (or knowing) that one does. One cannot, however, believe that one has either the visual experience of a horse or the belief that there is a horse nearby unless one has the concepts VISUAL EXPERIENCE and BELIEF. Furthermore, one cannot believe that either one has a visual experience of a horse or the belief that there is a horse nearby unless one has the concept HORSE. If one cannot have the belief that one believes that there is a horse nearby unless one believes that there is a horse nearby and if one cannot have the latter first-order belief unless one has the concept HORSE, it follows that one cannot have the former metarepresentational belief unless one has the concept HORSE. Although one can have the visual experience of a horse without having the concept HORSE, one cannot, however, have the metarepresentational belief that one has the visual experience of a horse unless one has the concept HORSE. Given that one may have the visual experience of a horse without having the concept HORSE, how is it that one cannot have the metarepresentational belief that one has the visual experience of a horse unless one has the concept HORSE? Unless one knows what horses are—unless one has the concept H O R S E —, one cannot entertain the higher-order concept VISUAL EXPERIENCE O F A HORSE. Similarly, one cannot believe that a piece of paper is a photograph of a horse unless one knows both what horses are and what cameras are—i.e., unless one has the concept H O R S E and the concept CAMERA. The tough epistemological questions are: how does one acquire such higher-order concepts as REPRESENTATION, BELIEF or EXPERIENCE? How does one know that such concepts apply to oneself? To have the concept BELIEF, for example, is to know that beliefs, unlike intentions and desires, have, in Anscombe's (1957) and Searle's (1983) terminology, a mind-to-world direction of fit, not a world-to-mind direction of fit, or that they have truth-conditions. Correlatively, one cannot have the concept

14

I limit myself to second-order metarepresentations of first-order representations of the world. But in verbal communication, the h u m a n metarepresentational faculty can ascend to higher levels.

412

Self-Knowledge

BELIEF unless one knows that, unlike states of knowledge, beliefs can be false. Does a human child learn these features of BELIEF by some ontogenetic process? Has the human brain been endowed by the phylogenetic evolution of the species with mastery of a set of higher-order concepts including the concept BELIEF with such characteristics? In this paper, I certainly do not pretend to offer any response to this question, which is presently the object of much empirical investigation in evolutionary psychology and in developmental psychology.15 Rather, I merely wish to divide the content of a metarepresentational belief into two pieces, one of which is the content of the first-order representation of the world that is metarepresented, and the other of which is the conditions for applying the relevant higher-order metarepresentational concept to the content of the first-order representation. I submit that the task of IBT, which deals with semantic mind/world relations, is to offer an account of the first piece. I submit that it is the task of epistemology, which deals with the conditions for believing and knowing something, to offer an account of the second piece.

3. Compatibilism Revisited At the beginning of this paper, I discussed the argument for the incompatibility between content externalism and the special authority of introspective self-knowledge. One can know a priori (with special first-person authority) that one believes that water is a liquid. It follows from externalism that one could not believe that water is a liquid unless one stood in relation to water. If so, then it follows from externalism that one can know a priori (with special first-person authority) that one stands in relation to water. But it is false that one can know a priori (with special first-person authority) that one stands in relation to water. It follows that externalism should be rejected. I noted that the incompatibilist argument can go through only on two assumptions. First, it must be assumed that, for any introspective belief one may have, this belief is a priori and/or it has the very special first person epistemic authoritative features that were attributed to it by either the rationalist or the empiricist epistemological traditions or both. Secondly, it must be assumed that one's WATER concept makes one and the same semantic contribution to the content of both the metarepresentation and the first-order representation metarepresented. In the last section of this paper, I want to examine one compatibilist strategy that is based on the distinction between two sorts of self-knowledge: knowledge of what one thinks, believes or experiences—i.e., knowledge of the content of one's mental representations—and knowledge—of the fact—that one thinks, believes or experiences whatever it is that one thinks, believes or experiences. As Dretske (forthcoming) has recently put it, "knowing what you think is easy". Knowing that you think is not.

15 See e.g., Baron-Cohen (1995).

D o We Know how We Know Our Own Minds yet?

413

Armed with this distinction, the compatibilist strategy argues that what one knows introspectively with special first-person epistemic authority is not the fact that one thinks, believes or experiences what one thinks, believes or experiences. Rather, what one knows introspectively with special first-person epistemic authority is what one thinks, believes or experiences. W h a t one may know by introspection is that it is water that one has beliefs about, not that one has beliefs about water. I f ones introspective a priori knowledge is not that one believes that water is a liquid, then even if externalism is true o f W A T E R , it still does not follow that one can know a priori—in the way one has introspective knowledge about what one believes—that one stands in relation to water (not to something else). Although I think that this strategy, which has recently been championed by Dretske (1995, 2 0 0 3 ) , is a perfectly coherent strategy, I do not approve it. Since I think it is a perfectly coherent strategy, I will not provide any knock-down argument against it. Rather, to show why I do not approve it, I will argue that its cost upsets its utility. As section 1 o f the present paper made clear, Dretske's (1995) model o f introspection has two basic ingredients: one is the model o f displaced perceptual knowledge, the other is the principle o f the reflexivity o f content, which in turn derives from the I B T account o f the contents o f first-order mental representations o f the external world. According to the principle o f the reflexivity o f content, one cannot represent the presence o f property F unless one has the information that the represented property is F (see section 1). Thus, it is a consequence o f the principle o f the reflexivity o f content that by virtue o f representing the presence o f F, one has the information that F is the represented property. T h e combination o f the displaced perception model o f introspection and the principle o f the reflexivity o f content yields the following result: one cannot have the visual experience o f a triangle unless one has the information that triangularity is what one is experiencing. By visually perceiving a triangular object, one is provided with information about oneself. O n e gets information about oneself, by turning ones visual attention, not to oneself, but to some triangular object. This externalist view o f introspective knowledge o f one's own perceptual experiences seems open to the following objection. First o f all, having the information that F is the represented property is not the same thing as knowing it. According to the principle o f the reflexivity o f content and the I B T account o f first-order representations, from the fact that it is representing the presence o f F, any representational device will have the information that F is the represented property. A thermometer will have the information that temperature is what it represents. From the fact that it is perceiving a cat, a dog will have the information that cathood is the perceived property. Still, neither a thermometer nor a dog can resaonably be credited with introspective metarepresentational knowledge o f the contents o f their first-order representations. Secondly, one could not know what it is one is experiencing—e.g., triangularity—unless one could apply to oneself the concept E X P E R I E N C E . O n e could not know what it is one has beliefs about—e.g., water—unless one could apply to oneself the concept BELIEF. Undoubtedly, thermometers and even dogs lack the mastery o f such higher-order concepts as E X P E R I E N C E and BELIEF. T h e reason why neither thermometers nor dogs can know what it is that they are representing would be that they cannot know that they are representing—something they are deprived o f by the lack o f the higher-order

414

Self-Knowledge

concept REPRESENTATION. Does not this show that one could not know what one experiences unless one knew that one has experiences? Does not this show that one could not know what one believes unless one knew that one has beliefs? This twofold objection relies on a subtle confusion between semantics and epistemology. It is a semantic truth that one cannot know either what one experiences or what one believes unless one applies to oneself either the concept EXPERIENCE or the concept BELIEF (and hence unless one possesses this concept). This is consistent with the epistemological point that having information about a represented property is different from knowing which property is being represented. One could not move from having the information to knowing that triangularity is the experienced property unless one applied the concept EXPERIENCE to oneself. O n e could not move from having the information to knowing that water is the property one has a belief about unless one could apply the concept BELIEF to oneself. But now the combination of the displaced perception model of self-knowledge and the principle of the reflexivity of content raises a further epistemological issue. Given that one cannot know either what one experiences (and/or believes) or that one has experiences (and/or beliefs) unless one can apply to oneself the concepts EXPERIENCE (and/or BELIEF), can one's self-knowledge be decomposed into two separable epistemological components: the knowledge of what one experiences (or believes) and the knowledge that one experiences (or believes) it? The above semantic truth provides no answer to this question. The present epistemological question arises in the context of introspective selfknowledge. But as Dretske (2003) argues convincingly, the very same epistemological question can be raised about perceptual knowledge of the world. Dretske (1969: 93-99) argued that one can know that the water is boiling by seeing it boil. One's grounds for believing that the water is boiling are that one sees it. Before seeing the water boil, one did not believe—let alone know—that it was boiling. O n a reliabilist view of what it takes to know the fact that the water is boiling, given that one's visual system is reliable, then by seeing the water boil, one can thereby come to know that it does. However, from the fact that one can see the water boil—and hence know that the water is boiling —, it does not follow that one can see that what is boiling is water. That what is boiling is water may be something one learnt not by seeing the water boil but otherwise. Presumably, one cannot tell by visual perception alone whether something is water, gin or gas. One may believe that what is boiling is water because either one was told that it was or because one tasted it. If so, then presumably one's grounds for knowing that what the water is doing is boiling should not extend to one's grounds for believing that what is boiling is water. Conversely, the question arises whether a (skeptical) challenge directed towards one's grounds for knowing that what is boiling is water could defeat one's claim to know that what the water is doing is boiling. 16

16

Dretske (2003) Has another convincing example to the same effect: Clyde comes to know that Harold told him on the p h o n e that he was going on vacation from hearing Harold's voice, understanding English and retrieving Harold's communicative intention. Clyde may not be able to know by using the same resources that Harold was the person who told him that he was going on vacation. H e may not be able to recognize Harold's voice. He may come to know (or form a justified belief) that it was Harold who told him that he was going on vacation by tracing the p h o n e call back to Harold.

Do We Know how We Know Our Own Minds yet?

415

Thus, the question raised by such examples is this: from the fact that some complex piece o f knowledge can be decomposed into two separable components because each can be traced to a particular epistemic pedigree, does it follow that one can know one piece without knowing the other? From the fact that some knowledge claim can be decomposed into two components, one of which is learned in one way and the other of which is learned in another way, does it follow that one can know either without knowing both? This question is perfectly general and it applies to perceptual knowledge as well as to psychological self-knowledge. Arguably, if the answer to the question is positive in the case o f perceptual knowledge, then so should the answer be in the case of psychological self-knowledge. Clearly, Dretske (1969, 2 0 0 3 ) thinks that a positive answer can be given to the question in the case o f perceptual knowledge and Dretske (2003) concludes that a positive answer can be given to the question in the case of psychological self-knowledge. Dretske (1969) did endorse the view that one can see (and hence know) that the water is boiling even though one does not know—not in the same way—that what is boiling is water. Dretske (2003) endorses the view that Clyde can hear (and hence know) that Harold told him that he was going on vacation even though he does not know—not in the same way—that the person who told him so was Harold. Understandably, Dretske (2003) argues that one can know by introspection what one is representing, even though one may not know—not in the same way—that what one is doing is representing. On this view, although one cannot know either what one thinks or that one thinks it unless one applies to oneself the concept T H O U G H T , still one can know what one thinks and fail to know that one thinks it. For example, one knows that the concept WATER applies to the content o f one's belief (that water is a liquid). Although one has the concept BELIEF, one merely believes (perhaps justifiably so) that the concept BELIEF applies to oneself (or to what one is doing while one believes that water is a liquid). But conceivably, the conditions for knowing that B E L I E F applies to oneself might not be met. O n this view, direct introspective knowledge is only o f what one thinks, not that one thinks it. As I said above, I think both that this is a coherent picture o f introspective selfknowledge and that it undermines the argument for the incompatibility between the special epistemic authority o f introspective self-knowledge and externalism. Nonetheless, I want to point out the cost o f this picture and why it may be superfluous. The cost of the picture is the denial of closure, i.e., the principle that knowledge is closed under known implication. Consider how the argument for the denial of closure works. First, it is noticed that in a complex conjunctive belief K, one can sort out two components B1 and B 2 — Κ = B1 & B 2 — o n the grounds that the epistemic justification o f one component is different from the epistemic justification o f the other. For example, one learns B l , that the water is boiling by visual perception, and one learns B2, that what is boiling is water, by some other method, e.g., by tasting it. Furthermore, one's grounds for believing one component, B l , meet the requirements for knowledge and one's grounds for believing the other component, B2, do not. It is uncontroversial that one's epistemic justification for believing Β1, the component one is in a position to know, should not carry over to one's epistemic justification for believing B2, the component one is not in a position to

416

Self-Knowledge

know. In other words, given the hypothesis, the fact that one knows B l , that the water is boiling, does not entail that one thereby knows B2, that what is boiling is water. It is also uncontroversial, I think, that the epistemic justification of the complex belief Κ — consisting of the two component beliefs, Bl and B2—should not exceed the epistemic justification of the lowest of the two component beliefs, i.e., B2. W h a t is controversial, I think, is the contraposition: namely, that a skeptical challenge directed towards one's justification for the weakest belief, B2, cannot ipso facto defeat one's justification for the stronger belief, B l . Certainly, a skeptical challenge so directed should defeat one's justification for the complex belief K. W h a t is controversial is precisely the denial of closure. H o w loose can ones epistemic standards be for believing that what is boiling is water, B2, consistent with one's knowledge that what the water is doing is boiling, Bl? Is skepticism about the fact that one knows that one has beliefs and experiences compatible with one's knowing what one believes and experiences? Now, given the costs of the two-tiered analysis of psychological self-knowledge, I want to reexamine the strategy an externalist may want to choose in order to avoid the incompatibilist conclusion. In the first section of the paper, I have expressed doubts about one premiss of the incompatibilist argument: namely that we know enough about the process of psychological self-knowledge to accept the various strands of the picture inherited from traditional rationalist epistemology and from traditional empiricist epistemology. Now, I want to question the conditional premiss: if the traditional picture of self-knowledge is right, then externalism cannot be right.

4. Conclusion: How to Question the Conditional Premiss of the Incompatibilist Argument Although the conditional premiss does not say so, it does imply, and seems motivated by, the claim that, unlike an externalist view of mental content, an internalist view would be consistent with the traditional claims made on behalf of the special epistemic authority of self-knowledge. This implication of the conditional premiss is puzzling. 17 According to a physicalist version of internalism, the content of one's mental states is constituted not by the history of one's brain nor by extrinsic relations between one's brain and properties instantiated in one's environment, but by the current internal physical (chemical and biological) structure of one's brain. If so, then how could one know a priori with special first-personal epistemic authority the physical structure of one's brain any better than one can know a priori with special first-personal epistemic authority the history of one's brain or the nature of the extrinsic relations between one's brain and properties instantiated in one's environment? Indeed, assuming the truth of physicalism, what difference does it make to the traditional picture of self-knowledge

17 As Heil (1992: 174) insightfully put it, "If the contents of one's thoughts depended entirely on the state of one's brain, for instance, why should that fact alone render our access to those contents any less direct or problematical?".

D o W e K n o w how W e K n o w O u r O w n M i n d s yet?

417

whether one is an externalist or an internalist? Perhaps ontological dualism was a better match for the traditional special epistemic authority of self-knowledge than physicalism of either an externalist or an internalist variety. 18 On the one hand, the account of introspective self-knowledge of one's perceptual experiences based on the principle of the reflexivity of content and the displaced perceptual knowledge model is an elegant externalist account. First, it clearly shows that only if one has metarepresentational resources can one become aware of the fact that one is having a perceptual experience. Secondly, it shows that one becomes aware that one is having a perceptual experience, not by peering inside at ones perceptual experience, but by having the perceptual experience itself, i.e., by experiencing the world. Thirdly, if the displaced perception model of self-knowledge is right, then so is the Cartesian thesis that self-knowledge involves higher-order (or metarepresentational) thought, not some quasi-perceptual mechanism. But its vindication of the Cartesian model of introspective self-knowledge is not unmitigated. Whereas the Cartesian model of introspective self-knowledge applies to one's knowledge that one has thoughts, the displaced perception model explains how one knows that one has perceptual experiences, not that one has beliefs or desires. 19 On the other hand, Dretske (2003) takes both premisses of the incompatibilist argument seriously enough to choose to deny closure in order to block the incompatibilist conclusion. Given the costs incurred by the denial of closure, the externalist might consider questioning both premisses of the incompatibility argument. One option is to question the conditional premiss on the grounds that physicalist internalism does not seem easier to accommodate with special epistemic authority of self-knowledge than externalism. Another option is to reject the assumption that the concept WATER makes the same contribution to the content (or truth-conditions) of one's first-order belief that water is a liquid and to the content (or truth-conditions) of one's metarepresentational belief that one believes that water is a liquid. On the second option, Oscar's concept WATER on Earth and Twoscar's concept TWATER on Twin-Earth would have, in Kaplan's (1989) terminology, different contents, but they would have one and the same character. Oscars concept WATER would contribute its content to the truth-conditions of his belief that water is a liquid. Twoscar's concept TWATER would contribute its content to the truth-conditions of his belief that twater is a liquid. Thus, Oscar's and Twoscar's first-order beliefs would have different truth-conditions. Suppose now that Oscar's concept WATER contributes its character, not its content, to the truth-conditions of Oscar's introspective belief that he believes that water is a liquid. Suppose that Twoscar's concept TWATER contributes its character, not its content, to the truth-conditions of Twoscar's introspective belief that he believes that twater is a liquid. Suppose that the character of WATER is the same as the character of TWATER. Then Oscar's introspective belief and Twoscar's introspective belief would have the same-truth-conditions. If so, then externalism would

18

Something Heil ( 1 9 9 2 : 1 7 4 ) expresses doubts about.

19

This seems to me vindicated by higher-order introspection: one's belief that one has such or such a perceptual experience is generally more reliable than one's belief that e.g., one has a particular desire.

418

Self-Knowledge

be true of one's first-order beliefs about the world, not of one's introspective beliefs about one's own beliefs. Whether the resulting picture is still externalist is a topic for another paper.

References Anscombe, G.E. (1957) Intention, Oxford: Blackwell. Baron-Cohen, S. (1995) Mindblindness, Cambridge, Mass.: M I T Press. Boghossian, P.A. (1989) "Concent and self-knowledge", in Bernecker, S. & Dretske, F. (eds.) (2000) Knowledge, Readings in Contemporary Epistemology, Oxford: Oxford University Press. Boghossian, P. A. (1997) "What the externalist can know a priori", in Wright, C., Smith, B. C. & Macdonald, C. (eds.) Knowing Our Own Minds, Oxford: Oxford University Press. Bürge, Τ. (1988) "Individualism and self-knowledge", in Bernecker, S. & Dretske, F. (2000) Knowledge, Readings in Contemporary Epistemology, Oxford: Oxford University Press. Davidson, D. (1984) "First-person authority", in Davidson, D. (2001) Subjective, Intersubjective, Objective, Oxford: Oxford University Press. Davidson, D. (1987) "Knowing one's own mind", in Davidson, D. (2001) Subjective, Intersubjective, Objective, Oxford: Oxford University Press. Dretske, F. (1969) Seeing and Knowing, Chicago: Chicago University Press. Dretske, F. (1988) Explaining Behavior, Cambridge, Mass.: M I T Press. Dretske, F. (1995) Naturalizing the Mind, Cambridge, Mass.: M I T Press. Dretske, F. (1999) "The mind's awareness of itself", in Dretske, F. (2000) Perception, Knowledge and Belief, Cambridge: Cambridge University Press. Dretske, F. (2003) "Knowing what you think vs knowing that you think it", in Schantz, R. (ed.) The Externalist Challenge: New Studies on Cognition and Intentionality, Berlin: Walter de Gruyter. Dretske, F. (forthcoming) "How do you know you are not a zombie" in Gertler, B. (ed.) Privileged Access and First-Person Authority, Ashgate Publishing Co. Fodor, J.A. (1994) The Elm and the Expert, Cambridge, Mass.: M I T Press. Harman, G. (1990) "The intrinsic quality of experience", in Block, N., Flanagan, O. & Güzeldere, G. (eds.) The Nature of Consciousness, Cambridge, Mass.: MIT Press. Heil, J. (1992) The Nature of True Minds, Cambridge: Cambridge University Press. Kaplan, D. (1989) "Demonstratives", in Almog, J., Perry, J. & Wettstein, Η. (eds.)(1989) Themes from Kaplan, New York: Blackwell. Kemmerling, A. (1999) "How self-knowledge can't be naturalized (some remarks on a proposal by Dretske" Philosophical Studies, 95, 311-328. McKinsey, M. (1991) "Anti-individualism and privileged access" in Chalmers, D.J. (ed.)(2002) Philosophy of Mind, Classical and Contemporary Readings, Oxford: Oxford University Press. Rosenthal, D.M. (1986) "Two concepts of consciousness", Philosophical Studies, 99, 3, 329-59. Rosenthal, D.M. (1993) "Thinking that one thinks", in Davies, M. and Humphreys, G.W. (eds.) Consciousness: Psychological and Philosophical Essays, Oxford: Blackwell. Searle, J. (1983) Intentionality. An essay in the philosophy of mind, Cambridge University press, Cambridge. Shoemaker, S. (1996) The First-Person Perspective and Other Essays, Cambridge: Cambridge University Press. Tye, M. (1992) "Visual qualia and visual content", in Crane, T. (ed.)(1992) The Contents of Experience, Cambridge: Cambridge University Press.

What Was I Thinking? Social Externalism, Self-Knowledge, and Shifting Memory Targets* PETER L U D L O W

In Ludlow (1995a, 1995b) I considered an argument by Boghossian (1989) that was designed to show the incompatibility of externalism and self-knowledge. The Boghossian argument turned on so-called slow-switching thought experiments in which an agent unknowingly moves between Earth and Twin Earth and is later informed that switches had been taking place. Would such an agent have authoritative knowledge of his mental states? According to Boghossian, externalism appears to allow an agent S to know his thoughts at time t, forget nothing, yet at some time later than t (having been informed of the possibility or prevalence of slow-switching) be unable to say what the contents of his thoughts were at t. As Boghossian argued, The only explanation, I venture to suggest, for why S will not know tomorrow what he is said to know today, is not that he has forgotten but that he never knew. Burge's self-verifying judgments do not constitute genuine knowledge. W h a t other reason is there for why our slowly transported thinker will not know tomorrow what he is said to know directly and authoritatively today?

In Ludlow (1995b), this argument is reconstructed as follows, (1) If S forgets nothing, then what S knows at tj, S knows at t 2 (2) S forgot nothing (3) S does not know that Ρ at t 2 (4) therefore, S did not know that Ρ at tj One natural response to this argument, offered by Warfield (1992) for example, would contend that switching cases are rather fantastic episodes (rare, if not nonexistent) and hence that they don't constitute the sort of relevant alternatives that might undermine our knowledge. Hence, on such a view, premise (3) must fail. In Ludlow (1995a) I offered that while switches to Twin Earth may be very rare indeed, more pedestrian cases of "slow switching" are altogether commonplace—at least for cases of social content. So, for example, consider the case of Richie, who uses the term 'cool' (in the social sense, not the thermodynamic sense) having only partial knowledge of its individuating conditions, and deferring to others in his social environment. Unknown to Richie, as he slides between different circles of friends he is now deferring to one individual

*

Thanks are due to David Chalmers, Peter Graham, and Norah Martin for helpful discussion. In addition, 1 am grateful to Anthony Brueckner, Stacie Friend, and Peter Smith for comments on earlier drafts of this paper.

420

Self-Knowledge

(say Fonzie) and a bit later to another (say Ralph), without any clear sense that the individuals he is deferring to employ different individuating conditions for 'cool'. Richie has become an unwitting victim of slow-switching without ever traveling to Twin Earth (or even leaving town, for that matter). Bürge (1998) has suggested that there are different reasons for rejecting premise 3. O n Burge's view, even if one cannot discriminate water thoughts from twater thoughts, one can still know what one was thinking at the earlier time t r This is possible, because memory preserves the contents of our original thoughts and we can access those contents—indeed, our second order thoughts will just have those contents—even if we cannot distinguish, say, water contents from twater contents. Although I have catalogued some reservations about this move in Ludlow (1999a), 1 Burge's way out remains a candidate solution to the puzzle. In this paper I do not want to dismiss the Bürge way out, but rather to explore an alternative path—one that I initially investigated in Ludlow (1995b). In that paper I held that the weak premise in Boghossian's argument was (1). The reasoning was that if the contents of our mental states are determined by our social environment it is natural to suppose that the contents of our memories will depend upon our social environment. As we move from one environmental condition to another (perhaps without even noticing the environmental change), the contents of our memories will shift accordingly. This line of reasoning was applied to Boghossian's argument in the following manner. ...let's say that at time tj I know that I am thinking that chicory is bitter. Suppose that at time t 2 later than tj, I recall that initial thought about chicory, but, due to undetected changes in my linguistic community, the content of my thoughts ... have shifted. Boghossian is arguably correct in asserting that I do not know at t 2 what I knew at tj, but he is incorrect in supposing that "the only explanation" for this is that I "never knew" my thoughts in the first place. It is entirely consistent with the social externalist view of memory that I forgot nothing, but that the contents of my memories have nonetheless shifted.

There are, however, some interesting consequences to this move that were initially pointed out by Hofmann (1995). Among those consequences, it appears that memory is not reliable regarding the contents of my earlier thoughts. Obviously in the scenario just envisioned, my initial thought at tj was about chicory, while my subsequent t 2 recollection of that thought involved, let's say, twicory. It seems that my memory has got things very wrong. In Ludlow (1998) I responded that while memory is not faithful in certain respects, in most important respects it will be. So, for example, the veracity of most of my beliefs about water will survive the switching cases. If I believe at tj that water is wet, then at

1

For example, one might want to argue that in order for us to know the content of a term it must be possible (in principle) for us to recover the individuating conditions of the term, and that this might require us to be in a position to inquire of the experts that we defer to. Arguably, if a shift has cut us off from certain experts that we might otherwise defer to, then the corresponding contents are now lost to us and not preserved at all. If this is right, then contents can only be recovered from the current environment and current deference relations.

W h a t Was I Thinking?: Social Externalism, Self-Knowledge, and Shifting Memory Targets

421

t 2 I will recall2 thinking that twater is wet. If my earlier beliefs were on the whole reliable, then my later beliefs should be as well. As for the fact that my recollection of the initial thought is that it was a twater thought, we simply needed to rethink the point of having memories in the first place. As argued in that paper: But is there any reason to suppose that it is the job of memory to "record the contents" of past mental episodes? On the face of it, this assumption begs the central question at issue. According to the externalist conception of memory that I have proposed, it is not the job of memory to record contents, but rather to provide information about past episodes relative to current environmental conditions. Even if there were a mechanism which could, as it were, freeze the contents of an initial mental episode and carry it in memory indefinitely, I'm not sure that it would have any utility. Indeed, it would be a way of preserving the content of a thought which we could no longer have.

Brueckner (1997) argued that the situation is actually worse than Hofmann made out, for if memories are to be individuated by their contents then it's not clear that my t 2 mental state even counts as a memory at all. It is rather a "false memory" or a pseudo-memory. In Ludlow (1999a) I argued that we can avoid this conclusion if we show some flexibility in how memories are to be individuated. If memories are to be individuated solely by their external contents, then Brueckner's conclusion seems inevitable. But most formulations of externalism hold only that some of our mental states are individuated at least in part by environmental relations. For example, following Putnam (1975) and many subsequent writers, one might opt for a "dual aspect" theory in which mental states can be individuated both by external and by internal contents. If so, then one can still identify the tj and t 2 mental events by their having the same non-relational contents. Intuitively the idea would be that in this case the tj and t 2 memory tokens could be identified as being the same memory type because they are instantiated by the same data structure, or perhaps because they play the same functional role.3 In hindsight, I think my responses to Hofmann and Brueckner got things all wrong. Or rather, I think the responses didn't go far enough. In the remainder of this paper I want to float an alternative (and speculative) response: the contents of memories shift, yet they always reflect the contents of the initial recollected thoughts. How could that be possible? The key idea is that the social externalist needn't suppose that the content of a thought is frozen at the moment it is tokened. Rather, the externalist can argue that such thoughts can have multiple time-indexed contents and that the time-indexed contents will track the environments inhabited by the agent as the agent moves through time, shifting as the agent moves through different environments! This is the provocative way to put the point. A more careful elucidation of the idea is necessary if we are to avoid absurd consequences here. Suppose that at tQ I have an initial thought about water. Now suppose that at t, I shift to Twin Earth and at some still later point, say t 2 , my original thought comes to have a twater content. Prima facie this sounds absurd, for we are used to supposing that thoughts are individuated by their contents and that there is a one-to-one cor2 3

Scare quotes around 'recall' are optional here, depending on how one wants to individuate memories. More on this point below. See Nagasawa (forthcoming) for further discussion of the Ludlow/Brueckner dispute on this point.

Self-Knowledge

422

respondence between thoughts and contents—i.e., thoughts are individuated by their contents (or at least in part by their contents). We are also used to supposing that the content that a thought has is fixed at the time the thought occurs. T h e externalist needn't be wedded to these suppositions, however. T h e alternative conception that I am floating here would replace the first supposition (that there is a one-to-one correspondence between thoughts and contents) with the idea that thoughts are individuated by a series of contents, not a single content, and that the contents will vary across time. T h u s the contents (note the plural) o f what I was thinking at tQ would depend upon where I happen to be (i.e. what social and environmental situations I find myself in) at later points of time. To illustrate, a given a single episode of thought Ε occurring at tQ, might be assigned different contents at different times, as follow. time

TO

E's content

H2O

2

H20

3

4

R

L

£

XYZ

XYZ

XYZ

H20

Returning to the original Boghossian argument, we find that the reconstruction of the argument given above suppresses some important details. T h e full argument should have been the following. (0) (1) (2)

If a first order thought Ε has content Ρ at tj, then it has content Ρ at t 2 If S forgets nothing, then what S knows at tj, S knows at t 2 S forgot nothing

(3) (4)

At t 2 , S does not know the (t 2 ) content of Ε to be Ρ therefore, at t p S did not know the (tj) content of Ε to be Ρ

Obviously the previously hidden premise (0) is suspect on this picture, since Ε may well have the content Ρ at time t ] ; but something else altogether, say Q , at t 2 . Premise (1) is suspect as well, of course, since what I know at tj is E's tj content, and what I know at t 2 is E's t 2 content. N o longer knowing the tj content of Ε doesn't count as forgetting any more than the fact that at t 2 I can no longer know that Ε is happening now. But the crucial thing to see is that while my 2 n d order knowledge of my thoughts shifts over time, that shift is crucial, since it allows me to track the shifting contents of my first order thoughts. Memories that are sensitive to environmental conditions are crucial for us to keep track of what we were thinking. To illustrate, consider how the content of the memory will shift with the environment. time

ti

t2

t3

t4

t5

E's content

H2O

H2O

XYZ

XYZ

XYZ

H20

Recollected Content of Ε

H2O

H20

XYZ

XYZ

XYZ

H20

W h a t Was I Thinking?: Social Externalism, Self-Knowledge, and Shifting Memory Targets

423

Now of course it follows that I will often be in error about the facts of what happened. If the tQ event that gave rise to my thinking that water is wet was my falling into some water, then my subsequent recollection of what actually happened will in fact be errant to some degree (at t 2 I will recall having fallen in twater), but memory errors about what happened are routine in any case. What is saved here is the idea that barring the loss of some sort of memory data structure, I will remember what I was thinking. This maneuver, which is also entertained in Ludlow (1999), seems to dispatch Boghossian's argument while keeping Hofmann (1995) and Brueckner (1997) at bay, but it also has a number of interesting consequences which remain to be explored. One such consequence involves the issue of what is to happen when we begin to consider the reliability of second and third person belief reports. Imagine a case where telephone calls of some sort are possible between Earth and Twin Earth, and that every time Oscar uses the telephone on Twin Earth, he unknowingly dials through to his old friends on Earth. If Oscar says over the telephone Ί fell in some water today', what should his friends (unaware of his switch) take him to be saying? On the whole, they have found Oscar a reliable informant, and they have found such testimony in general to be reliable. Should they not take him to be reporting a water experience, even though, from Oscar's current point of view, it is a twater experience? It will not do to simply say that Oscar's testimony is generally reliable so we can overlook switching cases, for as argued in Ludlow (1995a), switching cases are entirely commonplace. Moreover, such "telephone conversations" should also be commonplace, since in the kinds of undetected shifts envisioned (for example, when walking from Fonzie's house to Ralph's house) communication remains possible (indeed widespread). How is it that slow switching is commonplace, communication with our pre-switch community is widespread, and yet we can be justified in relying upon the testimony of others? This question is drawn into even sharper focus when we consider the reliability of memory and the kinds of shifting contents proposed in this paper. Imagine that at time tQ you hear me say Ί think water is wet' and thereby justifiably (if not authoritatively) come to believe that I am thinking water is wet. I subsequently shift to Twin Earth and after time passes recollect my tQ mental episode and recollect it as a twater thought. You, meanwhile, have not undergone a shift, and continue to consider my remark as reporting a water thought. Who is correct? Was my original thought a water thought (as you now think) or a twater thought (as I now think)? Couldn't we both be right? It is not just a question about memory, of course. On the view advanced here, the content of my original thought shifts behind me as I slide between communities over time (even if I never bother to recollect the original thought). Does the content of my thought also shift (or remain constant) as you and other witnesses to my mental episode slide between communities over time? And if so, how can everyone be right about the content of my thought? In Larson and Ludlow (1993) it was argued that we lapse into a certain confusion when we suppose that the aim of a belief report is to use a that-clause to characterize some single object in the agent's head. Rather, it was argued that belief reports are constructed for the benefit of the hearer H, to help Η construct a theory of the agent's mental life relative to the interests and concerns of H. On that view, the extension of

424

Self-Knowledge

'believes' was "fat" in a certain sense, allowing that any number of contents might correctly characterize an agent's belief relative to the interests of possible hearers. 4 Applying that basic idea to the current puzzle, a tQ mental episode Ε is in fact best thought of as a matrix of different contents for different times and witnesses (where a witness is someone privy to a reliable first or second person belief report). So for each time t and witness w, Ε specifies a content. Schematically, the picture is as follows—a kind of wO is the first person witness: E's content

times r

l

3

r

4

£

5

h2o

h20

XYZ

XYZ

XYZ

h20

1

h2o

h20

h20

h20

h20

h20

w2

h2o

h2o

h2o

XYZ

XYZ

XYZ

XYZ

XYZ

XYZ

XYZ

XYZ

w

witnesses

2

o

W

W

3

—

This may seem a remarkable conclusion. Episodes of belief would no longer be stable objects defined relative to a single individual, but would be highly dynamic and sensitive to the witnesses of those beliefs and the communities that those witnesses inhabit over time. But on the other hand, granted social externalist assumptions, how could such a possibility be dismissed out of hand? If the Cartesian conception of belief is abandoned for one in which the contents are sensitive to the social environment, and if we allow a plurality of social environments, then the real surprise would be finding a path that avoids just these conclusions. Notice finally, that while authoritative self-knowledge is thus saved from Boghossian's memory argument, it survives in a much more modest form. The first person witness wO has authoritative and perhaps a priori knowledge of the content of his/her mental states, but only of one aspect of those mental states. wO may have little or no knowledge of the content of that same mental state as viewed at other times, nor indeed, by other witnesses. Put another way, if we think of thoughts as being characterizable by "matrix propositions" like the above, we will indeed have authoritative access to one of the many contents in the matrix, but only to one—its content for a given time and witness. The above proposal is obviously programmatic (critics might add that it is also problematic). What I hope to have accomplished here is simply point out a position in the logical space that remains to be explored. It is certainly a counterintuitive position at first blush, but for all that it is one that can probably be sustained if supported by

4

This idea bears some similarity to more recent "swarm theories" of linguistic content in which it is held that an utterance does not merely express a single proposition, b u t many. T h e semantics puts constraints on what propositions might be expressed by the utterance, but does not isolate a single proposition. T h e idea in this paper, as will be clear below, departs in the following way: an utterance expresses a single proposition that is composed of many contents (contents that are indexed to times and hearers).

What Was I Thinking?: Social Externalism, Self-Knowledge, and Shifting Memory Targets

425

careful maneuvering. In this respect the position resembles m a n y other positions that have been developed within the externalist framework, and indeed the f r a m e w o r k itself. W h a t is alluring about externalism is its ability to escape successive charges o f i n c o m patibility and contradiction and emerge with surprising n e w avenues o f research. W h i l e debates about the relative merits o f externalism a n d individualism m a y have played out, it seems that questions about the available avenues for developing externalism and the kinds o f strategies available w i t h i n the f r a m e w o r k are very m u c h alive.

References Boghossian, Paul, 1989. "Content and Self-Knowledge", Philosophical Topics, 17, pp. 5-26. Brueckner, Anthony, 1997. "Externalism and Memory", Pacific Philosophical Quarterly, 78, pp. 1-12. Bürge, Tyler, 1979. "Individualism and the Mental", Midwest Studies in Philosophy, 4, pp. 73-122. Bürge, Tyler, 1988. "Individualism and Self-Knowledge", Journal of Philosophy, 85, pp. 649-63. Bürge, Tyler, 1998. "Self-Knowledge and Memory", in P. Ludlow and N. Martin (eds.) Readings in Externalism and Self-Knowldge. Stanford: CSLI Publications. Davidson, Donald, 1987. "Knowing Ones Own Mind", Proceedings of the American Philosophical Association, 60, pp. 441-458. Hofmann, Frank, 1995. "Externalism and Memory", manuscript, Dept. of Philosophy, University ofTuebingen, Germany. Larson, Richard, and Ludlow, Peter, 1993. "Interpreted Logical Forms", Synthese, 95, pp. 305-355. Ludlow, Peter, 1995a. "Externalism, Self-Knowledge, and the Prevalence of Slow-Switching", Analysis, 55, pp. 45-49. Ludlow, Peter, 1995b. "Social Externalism, Self-Knowledge, and Memory", Analysis, 55, pp. 157-159. Ludlow, Peter, 1998. "Social Externalism and Memory: a Problem?" Acta Analytica, 14, reprinted in P. Ludlow and N. Martin (eds.) Readings in Externalism and Self-Knowldge. Stanford: CSLI Publications. Ludlow, Peter, 1999a. "First Person Authority and Memory." in M. de Caro (ed.) Interpretations and Causes: New Perspectives on Donald Davidsons Philosophy. Kluwer. Ludlow, Peter, 1999b. Semantics, Tense, and Time: an Essay in the Metaphysics of Natural Language. Cambridge: MIT Press. Nagasawa, Yujin, 2002. "Externalism and the Memory Argument", Dialectica, 56, 335-346. Putnam, Hilary, 1975. "The Meaning of'Meaning'", in Gunderson (ed.) Language, Mind and Knowledge. Vol. 7, Minnesota Studies in the Philosophy of Science. Minneapolis: University of Minnesota Press. Warfield, T. 1992. Privileged self-knowledge and externalism are compatible. Analysis, 52, pp. 232-37.

Anti-Individualism and Minimal Self-Knowledge: A Dissolution of Ebbs's Puzzle B R I A N P. M C L A U G H L I N

There is a large and still growing literature concerned with the epistemic consequences of conjoining various anti-individualist theses with various theses about self-knowledge.1 Gary Ebbs's "A Puzzle About Doubt" (2003) is a recent addition to this literature. He calls our attention to the following two theses: Anti-Individualism·. "What a person believes and thinks is not settled by his linguistic dispositions, internal physical states, or phenomenal experiences, described independently of his social and physical environment" (p. 143). 2 Minimal Self-Knowledge·, "we each know without empirical investigation what thoughts our own utterances express" (p. 143). The conjunction of these theses, he argues, is incompatible with the standard analysis of epistemic possibility. It is his recommendation that we embrace both theses and reject that analysis. What is the standard analysis of epistemic possibility? Given that it is Ebbs's target of criticism, it is unfortunate that he offers no definitive statement of it. What he does instead is to cite certain claims that he says are consequences of it. He makes it clear, however, that the notion of epistemic possibility that concerns him is the one introduced by Saul Kripke, who famously claimed that it can be epistemically possible that not-p, even when it is metaphysically necessary that ρ (Kripke 1972/1980). (Thus, for instance, it is epistemically possible that water is not H 2 0, even if it is metaphysically necessary that water is H 2 0.) There is, I believe, nothing that answers to the description 'the standard analysis' of this notion, for there is no consensus as to how, exactly, the notion should be analyzed.3 Perhaps Ebbs would agree. By 'the standard analysis of epistemic possibility' he seems to mean 'the standard conception of epistemic possibility', where the standard conception is understood to be the Kripkean conception (however, exactly, that conception is to be analyzed). There are some points of agreement as concerns that conception of epistemic possibility. One is that it is epistemically possible that ρ only if it is not a priori knowable that not-p; equivalently, if it is a priori knowable that not-p, then it is not epistemically possible that p. Another point of agreement is that if one knows by introspection that

1 2

See, e.g., the essays in Wright, Smith, and Macdonald 1 9 9 8 ; the essays in Ludlow and Martin 1 9 9 8 ; and those in Nuccetelli 2 0 0 3 . Unless otherwise specified, all pages references are to Ebbs 2 0 0 3 .

3

See, e.g., some o f the essays in Gendler and Hawthorne 2 0 0 2 .

428

Self-Knowledge

ρ, then not-p is not epistemically possible for one. 4 As will become apparent, given these points of agreement, we will be able to evaluate Ebbs's criticisms of the Kripkean notion of epistemic possibility, despite the fact that there is no standard analysis of it. When introducing "the standard analysis of epistemic possibility" 5 flagged as his target of criticism, Ebbs tells us: A c c o r d i n g to the standard analysis o f e p i s t e m i c possibility, n o o n e can k n o w [just] b y reasoning or i n t r o s p e c t i n g — w i t h o u t empirical i n v e s t i g a t i o n — w h i c h o f her subjectively equivalent worlds she is actually in. (p. 145)

This invites the question: What are subjectively equivalent worlds? While the term 'subjectively equivalent world' is Ebbs's own term of art, he nowhere in the article offers a definition of it. One thing is clear, however: he takes subjectively equivalent worlds to be possible worlds, and the actual world to be one of them. Notice that if the actual world were not one of one's subjectively possible worlds, then it would be trivially true that no one can know just by reasoning or introspecting—without empirical investigation—which of one's subjectively equivalent worlds one is actually in. The reason would be that one is not actually in any of them. The alter reader will have already realized that given simply that subjectively equivalent worlds are possible worlds, it is trivially true that no one can know just by reasoning or introspecting—without empirical investigation—which of one's subjectively equivalent worlds one is actually in. To know what possible world one is in, one would have to know every contingent fact. It is trivial that one cannot know every contingent fact, let alone know every one without empirical investigation. As we shall see shortly, however, despite flagging this consequence as a target, Ebbs's does not in fact challenge the claim that one cannot know just by reasoning or introspecting—without empirical investigation—which of one's subjectively equivalent worlds one is actually in. He instead challenges other claims that he alleges are entailed by the standard analysis. Of them, more shortly. While Ebbs fails to defines 'subjectively equivalent worlds', he seems to have the following in mind: A possible world w is a subjectively equivalent world for a subject S if and only if in w S has exactly the same history of intrinsic physical states and exactly the same history of phenomenal experiences that S has in the actual world. 6 It is because S has exactly the same history of phenomenal experiences in all such worlds that they are called subjectively equivalent worlds for S. There are, of course, subjectively equivalent worlds for one that differ in respect of the types of social and/or physical environment that one is in. For example, there are subjectively possible worlds in which one was born, raised, and now live on a Putnamian twin earth (Putnam 1975), in which the rivers, lakes, oceans, etc. are filled with xyz rather than H 2 0. 7

4 5 6 7

Some hold that introspective knowledge is a priori. For discussion, see McLaughlin 2000 and McLaughlin 2003. Hereafter, I shall drop the scare quotes. I base this on his discussion on pages 148-150. In email correspondence, Ebbs informed me that he is now reluctant to characterize subjectively equivalent worlds by appeal to phenomenal experiences since he now wants to leave it open whether anti-in-

W h a t Was I Thinking?: Social Externalism, Self-Knowledge, a n d Shifting M e m o r y Targets

429

Ebbs case against the standard analysis o f epistemic possibility divides into two independent arguments. Each purports to show that the standard analysis is false because it counts some types o f situations as epistemically possible that are not. In what follows, I shall examine both arguments try to say why they fail. Ebbs's first argument relies on a would-be reductio ad absurdum o f an alleged consequence of the standard analysis. H e presents the following line o f reasoning that we can each run through from our own first-person perspective, and be lead, thereby, to contradiction: Premise One. "I may actually be in any o f m y subjectively equivalent worlds." (p. 150) Premise Two. " M y utterances o f 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature." (p. 151) Premise Three. "In all o f my subjectively equivalent worlds, my utterances o f 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature." (p. 151) Premise Four. "In some o f my subjectively equivalent worlds, m y utterances o f ' W a ter is a liquid at room temperature' do not express the thought that water is a liquid at room temperature." (p. 151) Conclusion. In all o f my subjectively equivalent worlds, my utterances o f 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature and it is not the case that in all o f my subjectively equivalent worlds, my utterances of 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature. Ebbs claims that the above reasoning presents us with "a puzzle" (p. 153). says:

He

T h e puzzle...is this. It seems that without empirical investigation, we are each epistemically entitled to accept [Premise One], because it follows from the standard analysis o f epistemic possibility; [Premise Two], because we have minimal self-knowledge; and [Premise Four], because it follows from the Twin Earth thought experiments. But [Premise O n e and Prem-

dividualism is true o f them. I d o not see why this leads him to want to avoid appeal to phenomenal experiences in the characterization o f subjectively equivalent worlds. First o f all, the above definition leaves open whether anti-individualism is true of phenomenal experiences. Secondly, even if anti-individualism is true o f phenomenal experiences, it will still be the case that there are subjectively equivalent worlds for one that differ in respect o f the types o f social and/or physical environment that one is in. T h e types o f social and physical environments that will vary would still include the types with which E b b s is concerned—e.g., Twin Earth environments with X Y Z rather than H 2 0 . In any case, in the email correspondence, Ebbs suggested the following definition o f 'subjectively equivalent worlds': a possible world w is a subjectively equivalent world for S if and only if in w S has the s a m e history o f intrinsic physical states that S has in the actual world. It is puzzling why these worlds are called 'subjectively equivalent', unless we are to assume that phenomenal experiences supervene on intrinsic physical states, a n d so that anti-individualism is false of phenomenal experiences. But, in any event, suffice it to note that everything I say below about subjectively equivalent worlds can be recast for this notion o f subjectively equivalent worlds. T h u s , nothing turns on whether we understand subjectively equivalent worlds for S to be those that meet the definition given above in the body o f the text, or whether we instead understand them to be all and only those worlds in which S has the s a m e history o f intrinsic physical states that S has in the actual world.

430

Self-Knowledge

ise Two] together entail [Premise Three], which is the negation of [Premise Four]...We are therefore apparently committed to each of [Premise One] though [Premise Four], including the contradictory pair [Premise Three] and premise [Premise Four], (p. 153)

Since the reasoning leads to contradiction we must reject at least one of the premises. The question is which premise(s) should be rejected. Ebbs contends that "the weakest premise...is the one that almost everyone accepts without reflection—[Premise One]" (p. 153). He holds that Premise One should be rejected by reductio ad absurdum. And he maintains that since Premise One is false, the standard analysis of epistemic possibility is mistaken. Ebbs's first argument against the standard analysis of epistemic possibility is, thus, a modus tollens argument: (pi) if the standard analysis of epistemic possibility is correct, then I may be in any of my subjectively equivalent worlds. (p2) It is false that I may be in any of my subjectively equivalent worlds. (This premise is to be justified by the above reduction (c) Therefore, the standard analysis of epistemic possibility is incorrect. I shall argue that Ebbs's would-be reductio argument for (p2) fails. I shall grant for the sake of argument that Premise One follows from the standard analysis of epistemic possibility, given that the phrase Ί may be in of Premise One means 'it is epistemically possible that I am in'.8 Thus, I shall grant that if Premise One is false, then the standard analysis is indeed mistaken. I shall also grant that Premise Four is supportable by certain Twin Earth thought experiments for anti-individualism; and I shall grant, in addition, that anti-individualists are committed to Premise Four. Premise Three and Premise Four are indeed incompatible. Thus, no one can accept both on pain of contradiction. Since anti-individualists must accept Premise Four, it follows that they must reject Premise Three. Ebbs claims that since anti-individualists must reject Premise Three, they must reject either Premise One or Premise Two. He claims this because he holds that Premise One and Premise Two entail Premise Three. Rejecting Premise Two would, he maintains, commit them to rejecting the minimal self-knowledge thesis. And he holds that this is not a rational option for them since the Twin Earth thought experiments invoked to support anti-individualism presuppose the minimal self-knowledge thesis. He thus holds that anti-individualists must reject Premise One. Ebbs is mistaken in claiming that Premise One. I may actually be in one of my subjectively equivalent worlds, and Premise Two. My utterances of 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature, jointly entail Premise Three. In all of my subjectively equivalent worlds, my utterances of 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature.

8

For reasons I shall give later, I in fact think that proponents of the standard notion of epistemic possibility are not ipso facto committed to Premise O n e .

What Was I Thinking?: Social Externalism, Self-Knowledge, and Shifting Memory Targets

431

They do not. Thus, an anti-individualist can reject Premise Three without rejecting either Premise One or Premise Two. To see that Premise One and Premise Two fail to entail Premise Three, consider first of all the following line of reasoning: (pa) I may actually be in any one of my subjectively equivalent worlds, (pb) My books are on my bookshelf. (c) In all of my subjectively equivalent worlds, my books are on my bookshelf. As should be apparent, this argument is patently invalid. The conclusion is false: in some of my subjectively equivalent worlds, it is not the case that my books are on my bookshelf. But that is perfectly compatible with both (pa) and (pb) being true. The argument from Premise One and Premise Two to Premise Three, I claim, fares no better. Now Ebbs holds that "it seems that without empirical investigation, we are each epistemically entitled to accept...[Premise Two], because we have minimal self-knowledge" (p. 153). The minimal self-knowledge thesis is, you will recall: Minimal Self-Knowledge·, "we each know without empirical investigation what thoughts our own utterances express" (p. 143). This thesis does not entail Premise Two. One reason is that the thesis is (allegedly) necessary, while Premise Two is clearly contingent. Nevertheless, the reductio argument is, presumably, supposed to be available to anyone for whom Premise Two is true. And Ebbs seems to assume that, given the minimal self-knowledge thesis, if Premise Two is true of someone, then that person can know independently of empirical investigation that his or her utterances of 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature. This invites the speculation that Ebbs thought that Premise One and Premise Two entail Premise Three because he made the following two assumptions. First, he assumed that given the minimal self-knowledge thesis (and the fact that I have the requisite concepts), Premise Two entails: Premise Two.5. I can know independently of empirical investigation that my utterances of 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature. Second, he assumed that Premise Two.5 entails Premise Three. Given that the minimal self-knowledge thesis is necessarily true, were these two assumptions correct, Premise One and Premise Two would entail Premise Three. But notice first of all that even were these two assumptions correct, they would be of no help to Ebbs in his attack on Premise One. Were they correct, Premise Two alone would entail Premise Three (given the necessity of the minimal self-knowledge thesis and the fact that I have the requisie concepts). As we noted, given that anti-individualists are committed to Premise Four, they must reject Premise Three. Thus, were the two assumptions correct, anti-individualists would have to reject Premise Two. They could not avoid doing that by rejecting Premise One instead. They would have to reject Premise Two whether or not they rejected Premise One. Secondly, the two assumptions are not both correct: it is either not the case that given minimal self-knowledge, Premise Two entails Premise Two.5 or it is not the case that Premise Two.5 entails Premise Three. For it is surely not the case that in accepting

432

Self-Knowledge

Premise Two. My utterances of 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature, I am committed to Premise Three, Premise Three. In all of my subjectively equivalent worlds, my utterances of 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature, and so committed to rejecting anti-individualism. Now given that 'by know independently of empirical investigation' is meant 'know just by reasoning or introspecting', then Premise Two.5 arguably entails Premise Three. For it seems that what one can know in fact just by reasoning or introspecting cannot conflict with one's history of phenomenal experience or one's history of intrinsic physical states. Gien, however, we are to understand 'know independently of empirical investigation' in this way, Premise Two.5 is patently false. It, like Premise Two, concerns all of my utterances, whether past, present, or future of'Water is a liquid at room temperature'. I cannot know just on the basis of reasoning and/or introspections that all of our utterances—whether past, present, or future—of 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature. Indeed, the minimal self-knowledge thesis offers no support to Premise Two.5 Presumably, the thesis of minimal self-knowledge is not supposed to be understood as the thesis we each know independently of empirical investigation what thoughts our own utterances express, whether those utterances are past, present, or fiiture. The thesis, presumably, is that that we each know independently of empirical investigation what thoughts our own utterances express when we make them. Given that the minimal self-knowledge thesis is to be so understood, it is not the case that given the thesis, if Premise Two is true, then Premise Two.5 is. Perhaps Ebbs had something like the following in mind when he stated Premise Two: Premise Two-R(evised). My utterances of'Water is a liquid at room temperature' in the language I am now actually speaking as I am now actually using it express the thought that water is a liquid at room temperature. 9 Arguably, if Premise Two-R is true of me (and I have the requisite concepts), then: Premise Two-R.5. I can know independently of empirical investigation that my utterances of 'Water is a liquid at room temperature' in the language I am now speaking as I am now using it express the thought that water is a liquid at room temperature. Premise Two-R, however, will not help Ebbs in his attempt to provide a reductio of Premise One. Given that 'know independently of empirical investigation means 'know just by reasoning or introspection', Premise Two-R.5 alone entails: Premise Three-R. In all of my subjectively equivalent worlds, my utterances of'Water is a liquid at room temperature' in the language I am actually now speaking as I am actually now using it express the thought that water is a liquid at room temperature.

9

In email correspondence, Ebbs acknowledged that he had something like Premise Two-R in mind, though he wondered why the phrase as I am now using it' is necessary. T h e reason it is needed is to accommodate ambiguity and contextual shifts in meaning in the language in question.

What Was I Thinking?: Social Externalism, Self-Knowledge, and Shifting Memory Targets

433

But even in conjunction with Premise One, it does not entail: Premise Three. In all of my subjectively equivalent worlds, my utterances of 'Water is a liquid at room temperature' express the thought that water is a liquid at room temperature. For it does not entail the patently false claim that in all my subjectively equivalent worlds, all of my utterances of 'Water is a liquid at room temperature' are in the language I am actually now speaking as I am actually now using it. Moreover, unlike Premise Three, Premise Three-R is compatible with Premise Four. In some of my subjectively equivalent worlds, my utterances of 'Water is a liquid at room temperature' do not express the thought that water is a liquid at room temperature. An anti-individualist can accept both Premise Three-R and Premise One. If antiindividualism is true, then there are subjectively equivalent worlds for me in which I use 'water' to refer to a chemical substance XYZ. And if Premise One is true, then I may be in a subjectively equivalent world in which I use 'water' to refer to a chemical substance XYZ. But Premise Three-R is compatible with my not being able to know independently of empirical investigation that I am not in a world in which I use 'water' to refer to a chemical substance XYZ. The reason is that I cannot know independently of empirical investigation whether water is XYZ. Since XYZ is by stipulation not H 2 0 , if water = H 2 0, then it is metaphysically impossible that water is XYZ. But according to the standard conception of epistemic possibility, it is nonetheless epistemically possible that water is XYZ. Consider, once again, Ebbs's first argument against the standard analysis of epistemic possibility: (pi) If the standard analysis of epistemic possibility is correct, then I may be in any of my subjectively equivalent worlds. (p2) It is false that I may be in any of my subjectively equivalent worlds, (c) Therefore, the standard analysis of epistemic possibility is false. We see now that Ebbs fails to justify (p2). His would-be reductio of the claim that I may be in any of my subjectively equivalent worlds fails. Let us turn, then, to Ebbs's second line of argument against the standard analysis of epistemic possibility. He claims that the standard analysis is mistaken since it counts certain types of situations as epistemically possible for us that are not. He claims, in particular, that according to the standard analysis, it is epistemically possible for us that we were born, raised, and now live on Twin Earth. And he argues that this is not epistemically possible for us. His second line of argument against the standard analysis is, then, in board outline: (pi) If the standard analysis of epistemic possibility is correct, then it is epistemically possible that one was born, raised, and now live on Twin Earth, (pii) It is not epistemically possible that one was born, raised, and now live on Twin Earth, (c) Therefore, the standard analysis of epistemic possibility is incorrect. I shall argue that Ebbs, at very best, succeeds in justifying (pii) only in a sense of epistemic possibility in which (pi) is false. In the sense of epistemic possibility in which (pi) may well be true, Ebbs fails to justify (pii). Ebbs thus commits the fallacy of relevance. To see this, let us first ask why Ebbs thinks that (pb) is true. He tells us that we can each come to know it by working through the following argument from our own first-person perspective:

434

Self-Knowledge

First Premise. "I am now using this sentence to express the thought that water is a liquid at room temperature." (p. 156) Second Premise. "If I were actually in the [kind of] world in which I was born, raised, and now live on Twin Earth, I could not use the sentence in question to express the thought that water is a liquid at room temperature." (p. 156) 1 0 Conclusion. "Therefore, I am not actually in the [kind of] world in which I was born, raised, and now live on Twin Earth." (p. 156) The argument is valid. He holds that, given the minimal self-knowledge thesis, we can each know the First Premise independently of empirical investigation. And he holds that we can each know the Second Premise independently of empirical investigation on the basis of Twin Earth thought experiments that support anti-individualism. 11 N o w the first point to note is that the standard analysis of epistemic possibility entails that it is epistemically possible for one that one is in a world in which one was born, raised, and now live on Twin Earth only on the assumption that one cannot know a priori that one is not in a world in which one was born, raised, and now live on Twin Earth. For, as noted earlier, according to the standard analysis, it is epistemically possible that ρ only if it is not a priori that not-p. Thus, if Ebbs has succeeded in showing that one can know a priori that one is not in the kind of world in which one was born, raised, and now live on Twin Earth, then according to the standard analysis of epistemic possibility, it is not epistemically possible that one is actually in a world of that kind. And so premise (pi) would be false. Ebbs does not, however, claim that by working through the argument from the First and Second Premises to the Conclusion, one can know the conclusion a priori. He acknowledges that one cannot know a priori that one is not actually in the kind of world in which one was born, raised, and now live on Twin Earth (p.···). He dis-

10

I have inserted '[kind o f ] ' here since there is more than one possible world in which I was born, raised, and now live on Twin Earth. 11 As Ebbs acknowledges, Michael McKinsey (1991) was the first to call our attention to this sort of line of argument. H e employed such an argument in a would-be reductio of compatibilism, the view that anti-individualism and self-knowledge are compatible. H e took it to be absurd to think that statements such as the Conclusion can be known independently of empirical investigation. This, however, is not Ebbs's view. Unlike McKinsey, Ebbs is a compatibilist. There are three sorts of compatibilists responses to McKinsey-style reductios of compatibilism. O n e is to maintain that while there are anti-individualist theses that are supportable by Twin Earth thought experiments, none of them entail that the Second Premise is knowable independently of empirical investigation. Ebbs calls this 'the orthodox reply'. (See, e.g., McLaughlin and Tye 1998a, and McLaughlin and Tye 1998b). A second sort of compatibilist response, one that goes unmentioned by Ebbs, is to deny that one can come to know the conclusion independently of empirical investigation by working through the above line of argument. According to this second sort of response warrant fails to transmit from premises to conclusion in such arguments. (See, e.g., Davies 1998; Wright 2003; and Davies 2003. McLaughlin 2 0 0 0 and 2003 endorse this line, where weak, as opposed to strong, a priority is concerned.) T h e third sort of response is to maintain that we can indeed come to know the Conclusion of the argument independently of empirical investigation by working through the argument. (See Warfield 1998; and Sawyer 1998.) It is this third response that Ebbs takes to be the most plausible. In making it, compatibilists must, he holds, reject the standard analysis of epistemic possibility. For, he holds, it mistakenly entails that it is epistemically possible that one is in a world in which one was born, raised, and now live on Twin Earth.

W h a t Was I Thinking?: Social Externalism, Self-Knowledge, and Shifting M e m o r y Targets

435

tinguishes "the claim that we k n o w a given statement a priori and the claim that we k n o w the statement w i t h o u t empirical investigation" (p. 159). It is his position that by working through the a r g u m e n t f r o m the First a n d Second Premise to the C o n c l u s i o n , we can k n o w that we are not actually in the kind of world in question i n d e p e n d e n t l y of empirical investigation, even t h o u g h we are unable to k n o w it a priori. W h a t , then, does Ebbs mean by' knowledge that is independently of empirical investigation in his discussion of this argument? H e says: Many of the statements we know without empirical investigation are statements that we cannot know a priori, in any standard sense of that term. What we know without empirical investigation includes what we are entitled to believe without going through any special empirical investigation, (p. 159) H e offers the following examples: Consider my belief that physicists in my linguistic community have detected quarks. I have read this in authoritative books, but have never undertaken any empirical investigation into whether it is true. By ordinary standards, I am epistemically entitled to believe that physicists in my linguistic community have detected quarks; if this is true, then I know it independently of empirical investigation. Similarly, an unusually sheltered person who is told that members of her linguistic community have seen or touched water might be epsitemically entitled to believe this without undertaking any empirical investigation into whether the person who told her this is trustworthy or whether it is true; if it is true, then by ordinary standards she knows it independently of empirical investigation, (p. 148) I am uncertain why Ebbs thinks that reading authoritative books fails to c o u n t as e m pirical investigation in the ordinary sense. But, in any case, I a m sympathetic with the view we can be epistemically entitled to belief things we have never empirically investigated, a n d that such entitlement (combined with truth) can be e n o u g h for knowledge. 1 2 I think that such entitlement is highly context dependent; b u t I have n o reason to think that Ebbs would resist that. T h e second point to note for present purposes, then, is that in this weak sense of 'knowable independently of empirical investigation', the standard analysis of epistemic possibility does not entail that if ρ is epistemically possible, then ρ is n o t knowable i n d e p e n d e n t l y of empirical investigation. A p r o p o n e n t of the standard analysis of epistemic possibility thus need not deny that one can k n o w independently of empirical investigation in Ebb's weak sense that one is n o t actually in the kind of world in which one was born, raised, and n o w live on Twin Earth, even if the p r o p o n e n t holds that it is epistemically possible that one is actually in the kind of world in w h i c h o n e was born, raised, a n d n o w live on Twin Earth. So, even if Ebbs were right t h a t via the a r g u m e n t f r o m the First and Second Premise to the Conclusion one can k n o w in this weak sense independently of empirical investigation that one is not in the kind of world in which one was born, raised, a n d n o w live on Twin Earth, it w o u l d n o t follow that the standard analysis of epistemic possibility is mistaken. 1 3

12

See the discussions of "weak a priori knowledge" in McLaughlin 2 0 0 0 and McLaughlin 2 0 0 3 .

13

I should note that I argue elsewhere that this sort of line o f argument will fail to establish the sort o f conclusion in question in either a weak or a strong sense o f a priority. See McLaughlin 2 0 0 0 and McLaughlin 2 0 0 3 .

436

Self-Knowledge

Consider, once again, then, Ebbs's second argument against the standard analysis: (pi) If the standard analysis of epistemic possibility is correct, then it is epistemically possible that one was born, raised, and now live on Twin Earth, (pii) It is not epistemically possible that one was born, raised, and now live on Twin Earth, (c) Therefore, the standard analysis of epistemic possibility is incorrect. Ebbs has, at very best, then, succeeded in justifying premise (pii) only if from the fact that not-p is knowable independent of empirical investigation in his weak sense, it follows that it is not epistemically possible that p. But as we noted, the standard analysis does not entail that if it is epistemically possible that p, then, in the weak sense of knowing independently of empirical investigation, it is not knowable that not-p independently of empirical investigation in Ebbs's weak sense. In the sense of'epistemically possible' in which (pi) is true, Ebbs has not justified (pii). In the sense of 'epistemically possible' in which Ebbs has (at very best) justified (pii), (pi) is false. The argument commits the fallacy of relevance. 14 Rather than raising an objection to the standard analysis of epistemic possibility via his appeal to the argument from the First and Second Premise to the Conclusion, Ebbs seems instead just to be invoking a different notion of epistemic possibility from the standard notion. He seems to be proposing that if not-p is knowable independently of empirical investigation in his weak sense, then ρ is not epistemically possible. Ebbs is, of course, free to introduce his own technical notion of epistemic possibility. Suffice it to note, however, that his notion won't serve the purposes for which the Kripkean notion of epistemic possibility is intended. To see this, note that in Ebbs's sense of independence from empirical investigation, the vast majority of people who know that water = H 2 0, know it independently of empirical investigation. But the notion of epistemic possibility is intended to capture the intuition that there is a sense in which it is possible that water is not H 2 0, even though it is metaphysically necessary that water is H 2 0. The idea is that it is epistemically possible that water is not H 2 0, even though it is metaphysically impossible that water is not H 2 0. In conclusion, Ebbs has given us no reason whatsoever to think that anti-individualists who accept the minimal self-knowledge thesis must reject the Kripkean notion of epistemic possibility (however exactly that notion should be analyzed). 15

References Davies, M. 1998. "Externalism, Architecturalism, and Epistemic Warrant." In Wright, Smith, and Macdonald 1998, pp.321-361. Davies, M. 2003. "The Problem of Armchair Knowledge." In Nuccetelli 2003, pp.23-56. Dretske, F. 1995. Naturalizing the Mind. Cambridge: M I T Press. Ebbs, G. 2003. "A Puzzle About Doubt." In Nuccetelli 2003, pp. 143-169.

14

15

I should note that I argued in McLaughlin 2 0 0 0 and McLaughlin 2003, that twin earth thought experiments won't provide "strong a priori warrant"—empirically indefeasible warrant—for claims like the Second Premise. I wish to thank Gary Ebbs for comments.

What Was I Thinking?: Social Exiernalism, Self-Knowledge, and Shifting Memory Targets

437

Field, H. 1996. "The A Prioricity of Logic." Proceedings of the Aristotelian Society 96:359-376. Gendler, T., and Hawthorne, J. (eds.) 2002. Conceivability and Possibility. Oxford University Press. Kripke, S. 1972. "Naming and Necessity." In Semantics of Natural Languages, eds. D. Davidson and G. Harman, Dordrecht, Netherlands, Reidel, pp.253-355. Reprinted as Naming and Necessity. Cambridge: Harvard University Press, 1980. Ludlow, P., and N. Martin. 1998. Externalism and Self-Knowledge. Stanford: CSLI. McKinsey, M. 1991. "Anti-individualism and Privileged Access." Analysis 51: 9-16. McLaughlin, B. P., and M. Tye. 1998a. "Is Privileged Access Incompatible with Content-Externalism?" Philosophical Review 107: 349-380. McLaughlin, B.P., and M. Tye. 1998b. "Externalism, Twin Earth, and Self-Knowledge." In Wright, Smith, and Macdonald 1998, pp.285-320. McLaughlin, B.P. 2000. "Self-Knowledge, Externalism, and Skepticism." Proceedings of the Aristotelian Society, supp. vol. 74: 93-117. McLaughlin, B.P. 2003. "McKinsey's Challenge, Warrant Transmission, and Skepticism." In Nuccetelli 2003, pp.79-96. Nuccetelli, S. 2003. New Essays on Semantic Externalism and Self Knowledge. MIT/Bradford. Putnam, H. 1975. "The Meaning of'Meaning'." In his Philosophical Papers, vol.2: Mind, Language, and Reality, pp.215-271. Cambridge: Cambridge University Press. Sawyer, S. 1998. "Privileged Access to the World." Australasian Journal of Philosophy 76: 523-533. Tye, M. 2000. Consciousness, Color, and Content. Cambridge: M I T Press. Warfield, T. 1998. "A Priori Knowledge of the World: Knowing the World by Knowing Our Minds." Philosophical Studies 92: 127-147. Wright, C., B.C. Smith, and C. Macdonald. 1998. Knowing Our Own Minds. Oxford: Oxford University Press. Wright, C. 2003. "Some Reflections on the Acquisition of Warrant by Inference." In Nuccetelli 2003, pp.57-78.

VII The Epistemic Significance of Perception

Explaining Perceptual Entitlement CHRISTOPHER PEACOCKE*

How do perceptual experiences entitle us to form beliefs about the objective world? It is this question, perennial and fundamental, that I will be addressing. I will be offering an answer to the question that brings to bear externalism in the theory of intentional content upon the theory of the normative relation of entitlement. The account draws on some more general considerations about explanation and rationality, considerations applicable in many domains far beyond the mental. If the principles underlying the account are correct, they will apply also to entitlement in some other areas, including beliefs about our own actions. The very first step in addressing our question must, however, involve some clarification of the several levels at which issues about the entitlement relation can be addressed.

1. Entitlement: The Three Levels The three levels at which we may characterize the entitlement relation are levels which we can distinguish for any property or relation. For any property or relation, there is an increase in generality and explanatory power as one proceeds through these three levels of characterization. Level (1). The first of the three levels I distinguish is simply the level of instances or examples of the entitlement relation. So characterizations at this level comprises true statements of the form 'a thinker in such-and-such circumstances with so-and-so background information is entitled to judge that />'. These examples involve a specification of types of circumstances in which an entitlement exists. The types may concern the thinker's environment, his other conscious states characterized in terms of content and his general capacities. If a thinker is entitled to make the judgement, of a seen object, that it is curved, when he visually experiences it as curved, and when there is no reason for doubting his senses, then that would be a statement included at this first level of instances. So would statements about the entitlement to rely on apparent personal memory in making judgements about one's own past. It is an important matter what we take to be the entitling state. We cannot take it to be something of the kind perceiving that p, since that state implies that its subject already believes (indeed arguably knows) that p. A transition with something of this

*

Christopher Peacocke retains the copyright in this paper.

442

The Epistemic Significance of Perception

kind as the specified entitling state would be vacuous as a means of rationally reaching new beliefs. It could be employed only if it were redundant. But there are kinds of perceptual states with the content that ρ that are factive (imply the truth of p) without implying that the subject already believes that p. One of these is the state perceives χ to be F, or perceives χ andy to stand in relation R. One can perceive the room to be square without believing it is square (one may falsely believe that one is subject to a perceptual illusion). Such a factive perceptual state can be the entitling state for a perceptual judgement without vacuity ensuing. An alternative approach would be to take as the entitling states perceptual experiences whose contents can be false, such as having an experience as of χ being F, or even having an experience as of there being something in front of one that is F. Perceptual entitlement is often said to be defeasible. One's entitlement to judge, on the basis of a perceptual state, that ρ can be defeated in the presence of further information, such as that one is in a room with strange lighting conditions, or that one is, unbeknownst, a participant in an experiment in the psychology of perception. It is important, here as in other cases, to distinguish two kinds of defeasibility. 1 Something one thought to be a mathematical proof may turn out not to be so; or reasonable doubt may be cast on it by the most distinguished experts. I call this 'defeasibility of identification'. What is in question in these cases is the correctness of the identification of something as a certain kind of ground. One may have good reason for wondering whether something really is a proof. But if it really is a proof, no additional information can establish that its conclusion is not true. If something really is not a proof, it is not a proof all along, whatever we think about it. By contrast, in what I call 'defeasibility of grounds', defeating conditions show that the grounds one has are not strong enough. One may make an inductive inference, on fairly extensive grounds, to a generalization that all F s are G. Ones entitlement to this conclusion is defeated if someone can show that all one's extensive inferential basis of singular cases of F's that are G are all special cases in some way. In such a case, one always knew ones grounds were not conclusive. Which kind of defeasibility one regards perceptual entitlement as possessing varies with whether one takes the entitling state to be factive or not. When the entitling state is factive, the defeasibility of perceptual entitlement can consist only in defeasibility of identification, not defeasibility of grounds. If someone really does perceive χ to be F, then it will be true that χ is F. What can be defeated is only one's identification of one's own state as a perceiving χ to be F. When the entitling state is not factive, we have genuine defeasibility of grounds: an experience as of its being the case that p may be shown, in the light of further information, to coexist with its not being the case that p, or to be in the context entirely sufficient to make it rational to accept that p. Later in this paper I will be arguing that there are good reasons for taking the entitlement as formulated using factive perceptual states to be more fundamental than those formulated using non-factive states of perceptual experience. For the moment,

1

This distinction is further discussed in my paper 'The A Priori', forthcoming in The Oxford Handbook of Analytical Philosophy, ed. F. Jackson and M. Smith (Oxford: Oxford University Press, 2005).

Explaining Perceptual Entitlement

443

we can remain neutral on the issue. Both the factive and the non-factive states seem, in the case of a range of representational contents that p, to give prima facie reason for judging that p, in the absence of corresponding defeating conditions. Level (2). The second level is the level of generalizations about the entitlement relation. This level consists of true generalizations which, in the presence of additional information determined by the generalizations, have statements at level (1) as instances. It may be helpful to think of the relation of this level to others as the same as that exhibited by grammatical generalizations to other levels in the theory of grammars. In his 1965 theory, Chomsky wrote that a grammar "is descriptively adequate to the extent that it correctly describes the intrinsic competence of the idealized native speaker".2 In the same spirit as Chomsky's use of the term, we could call this second level 'the level of descriptive generalization. A descriptively adequate grammar for a language will have theorems specifying instances of the property of grammaticality for the language. We equally operate at this second level of characterization when, in developing a logic for some particular expression in natural language, we move from particular valid transitions containing the expression but not containing schematic letters to the stage of formulating general schemata that are valid. That is a move to a level of descriptive adequacy. As in the other cases, it can be an important step towards theoretical understandings. It is, for example, illuminating, and a step towards an explanatory theory, to note that though transitivity does not hold for the counterfactual conditional, the schema 'If A were the case, then Β would be the case; if A & Β were the case, then C would be the case; so if A were the case, C would be the case' is generally valid.3 In the logic case, as in the grammar case, people are capable of making all sorts of mistakes about which are the correct generalizations at this second level. Van McGee has argued that modus ponens, often taken by many philosophers as an example of an obviously valid rule, fails when the consequent of the conditional is itself a conditional.4 It is one thing to make judgements for reasons which are entitling states, in a variety of conditions. It is another to be able to articulate correctly generalizations about the conditions under which one is entitled to make judgements of a given kind. This is simply the application to the notion of an entitled transition of a point long recognized about the possible inability of those who correctly use the concept chair to offer a correct definition of the concept (even when there is one). Having a classificatory ability is one thing; it is another to be able to collect correctly in thought in a non-trivial fashion just the conditions under which the classification applies. What we need to be responsive to good reasons is an ability to respond to entitling conditions with appropriate judgements. Making correct generalizations about entitling conditions is a further ability, not implied by the former capacity.

2 3 4

Aspects of the Theory of Syntax (Cambridge, Mass: MIT Press, 1965), p.24. R. Stalnaker, Ά Theory of Conditionals', Studies in Logical Theory ed. N . Rescher (Oxford: Blackwell, 1968); D. Lewis, Counterfactuals (Oxford: Blackwell, 1973), esp. pp. 32-5. McGee's example: for someone speaking in 1980 of the then upcoming election in the United States, utterances of the premises 'If a Republican wins, then if Reagan doesn't win, Anderson will win' and Ά Republican will win' are true, but an utterance o f ' I f Reagan doesn't win, Anderson will win' is false. See Vann McGee, Ά Counterexample to Modus Ponens', Journal of Philosophy 82 (1985) 462-71.

444

T h e Epistemic Significance of Perception

The generalizations about the relation of entitlement at this second level may be more or less extensive, and correspondingly more or less illuminating. The generalizations may use theoretical notions in classifying circumstances, contents, and capacities. It is information about the extension of these classifying notions that one will need if one is to use the generalizations to derive truths at level (1) about instances of the entitlement relation. Level (3). The third level is the level of explanation. This third level consists of explanations of the true generalizations at level (2). If certain theoretical notions seem to be involved in capturing the correct generalizations at that second level, then one of the tasks at this third level is also to explain why they do so. In linguistic theory, Chomsky distinguished a level of explanatory adequacy, a level of theory at which one aims to explain why the child selects a particular descriptively adequate grammar. 5 Explanations at that level would be empirical explanations of acquisition in the linguistic case. In the case of the development of a logic, formulations of explanations at this third level would involve the development of a semantical theory that explains the patterns of validity and invalidity captured in schemata at the second level. As these examples show, explanations at this third level may be either empirical or a priori, according to the nature of the explananda in the given case. In the case of the explanatory level for the entitlement relation, we are concerned with philosophical explanations. In contrast at least with the linguistic case, the task of characterizing this third level for the relation of entitlement is that of explaining timeless generalizations rather than historical (extended) events of acquisition of a grammar. In the particular case of explaining the entitlements provided by perception and memory, the task is to explain the true generalizations about defeasible entitlement, and to say why those generalizations (and nothing weaker or stronger) captures the extension of the entitlement relation. The development of a third-level characterization of the entitlement relation further divides into two parts. First, one must specify (with arguments) some very general property which is required for an arbitrary transition to be one to which the thinker is entitled. A tendency to lead to true judgements must be one element of an account of what makes a transition one to which a thinker is entitled. 6 This first part can be called goal-specifying. The second part of a characterization at this third level is then showing that, and showing why, generalizations about the entitlement relation at the second level have the general property specified in the first part. This second part can be described as proof of fulfillment of the goal. One would similarly distinguish these two parts in a level-three characterization of a logic. One would first specify a semantic property that a valid inference must possess; one would then show, in a semantic theory, why the general principles identified as intuitively valid at the second level have this property, and why the invalid ones lack it.

5 6

Aspects, pp.25-7. For further discussion, see my paper 'Three Principles of Rationalism', European Journal of Philosophy 10 (2002) 375-397.

Explaining Perceptual Entitlement

445

2. The Second Level for Perceptual Entitlement How would we describe these three levels in more detail for the case o f perceptual entitlement? At the level o f examples, I wrote, like James Pryor, o f an entitlement to take the perceptual content o f experience at face value, in the absence o f reasons for doubting it. 7 For a wide range o f perceptual contents, there is such an entitlement. It exists for many spatial contents, temporal contents, contents relating to surface texture, colour and illumination, and a range o f material properties. There are, however, also cases where it is much less plausible that perceptual experience alone can supply the entitlement. Consider furniture that looks Swedish; appliances that look like Mac computers; or the properties o f looking sad or looking delighted. Are we entitled, in the absence o f reasons for doubt, to judge on the basis o f such experiences that some furniture we see is Swedish, that some presented object is a Mac computer, or that someone is sad or delighted? In some o f these cases, one can follow a strategy o f divide-and-rule. O n e can explain the apparently perceptual phenomenon thus. There is some kind such that the thing or person appears to be o f that kind, and the person judges that things o f that kind are (say) Swedish people, or Mac computers. T h e perceptual entitlement holds only for the kind which is given in the content o f perception, as opposed to the content o f the judgement. But such a division is not plausibly available in all cases. It does not, for example, fit the case o f perception o f the expression o f an emotion. To describe, when seeing the face o f a person, the experience in which they look sad in non-emotional terms is not to capture its distinctive representational content. There is no kind, described without reference to the emotions, o f which one can say that the facial expression appears to be o f that kind, and it is merely an additional judgment on the part o f the person that people looking that way are sad. It is tempting to say that the purely perceptual entitlement holds only for observational concepts. That may well be true; but it is hardly an illuminating generalization at level (2), unless we have some independent account o f observationality. We are in danger o f moving in a circle, for it is only too plausible to say that observational concepts are those that can be applied with entitlement simply on the basis o f perceptual experience, and without further information. Unless we have some characterization o f observational concepts distinct from that, then to say that the perceptual entitlement holds only for observational concepts will become the vacuous claim that it holds for those contents for which it holds. I will suggest later on that a by-product o f a proper characterization o f the second level will be a starting point for characterizing the relevant notion o f observationality without circularity. Suppose we can formulate a sufficiently wide-reaching true generalized conditional about the conditions under which perceptual entitlement holds. It will have the form ' I f such-and-such conditions hold for the content ρ and for the

7

J. Pryor, ' T h e Skeptic and the Dogmatist', Nous 3 4 ( 2 0 0 0 ) 5 1 7 - 4 9 , at pp. 536ff.. This position is in the same spirit for entitlement in perception as Burges on the entitlement to accept the utterances o f interlocutors: see T. Bürge, 'Content Preservation', Philosophical Review 1 0 2 ( 1 9 9 3 ) , 4 5 7 - 8 8 .

446

T h e Epistemic Significance o f Perception

thinker's circumstances, then the thinker is entitled to take the content ρ of his perceptual experience at face value'. One way to characterize the non-observational will be as contents not meeting the antecedent of that conditional.8 The task of formulating such a generalization about perceptual entitlement lies at the second of the three levels. So I aim to characterize, without using the notion of an observational concept, a relation which holds between a perceptual experience and a particular content p which it represents as being the case, a relation with the following property: the holding of that relation is sufficient for a subject who enjoys the experience to have a perceptual entitlement to judge that p, in the absence of reasons for doubt. Various other perceptual entitlements, I will later argue, have their status as such in virtue of the relations in which they stand to this sufficient condition. In some cases, and to a first approximation, what is constitutive of an experience's having a certain representational content is that when the thinker's perceptual apparatus is functioning properly, in a normal environment, experiences with that content are caused by the holding of the condition which is in fact the correctness condition for that content. This is plausible for the spatial representational contents of perception: the representational contents concerning such matters as distance, direction, shape and size. In Being Known, I argued that the same is true for the temporal contents of perception.9 When all is working properly, in a normal environment, and in the most fundamental cases, a subject's perception of temporal order and magnitude is caused by instances of those order-relations and temporal magnitudes. (The perception of temporal magnitudes, like the perception of spatial magnitudes, is commonly unit-free.) Suppose we agree that it is constitutive of a particular kind of experience's having a spatial or temporal content that such experiences have certain causes in specified conditions. It does not follow that it is constitutive of that content that it feature in experiences of that kind in any perceiver capable of having states with that content. Experience of different kinds, in more than one sense modality, may have the same spatial representational content. A given subject may be capable of having experiences in only one of those modalities. Furthermore, the given content may also feature in proprioception. The feeling of moving one's arm in a straight line may involve the same content straight line as also features in visual or tactile experience. The spatial content straight line can also feature in the non-proprioceptive awareness of acting that can be present even when one's limbs are anesthetized. Indeed, the very fact that perceptual experiences with these contents are individuated in part by facts about their causes in certain circumstances opens up the possibility of the occurrence of such contents in other conscious states, both in perception and in action. The cause that is involved in the individuation may cause other experiences too. Equally, the state of affairs that is the cause may be mentioned in constitutive accounts in which it features as an effect, as in the non-proprioceptive awareness of action. In short, we must be careful not to

8

Pryor uses the notion o f propositons which our experience 'basically represent': these are propositions we seem to perceive to be so, but not in virtue o f seeming to perceive other propositions to be so (p.539). T h i s will give a wider class o f entitling states than results from application o f the criterion o f observationality outlined below.

9

Being Known

(Oxford: O x f o r d University Press, 1999), Chapter 3.

Explaining Perceptual Entitlement

447

overstate the constitutive principle which links the individuation of some perceptual contents with the holding of those contents in the perceiver's environment. What is true, however, is that the spatial and temporal contents of experience are in a certain sense constitutively basic with respect to these experiences. That is, these experiences do not have these elements of their content in virtue of the experiences' having certain other relations to other states with the same contents. One can contrast this with contents of perceptual experience which seem to use such concepts as soldier or judge, as when one says that it looks as if there is a soldier guarding the building, or a judge speaking from the bench in a courtroom. Such experiences, if that is their literal content, have those contents by virtue of their having contents which also feature in the ability to come to believe that someone is a soldier or a judge; and these capacities in turn have to do with some knowledge, perhaps rudimentary, of what it is to be a soldier or a judge. Unlike the case of contents concerning the properties of being a soldier, or of being a judge, perceptual experience can provide a thinker's fundamental fix on spatial and temporal properties and relations. The experiences of which I am writing are sometimes called externally or anti-individualistically individuated. For present purposes, this not an ideal label. For the essential characteristics of the phenomenon are present in, for instance, proprioception of limb position, and the disposition of one's own body in space. One has an awareness that represents one's limbs and body as being a certain way spatially. It is highly plausible that what gives this awareness the content it has is that, when all is functioning properly, an awareness as of one's arm being straight is caused by one's arm being straight. This is not a relation between conditions external to the perceiver's body and the perceiver's own mental states. So I prefer to speak of perceptions which are instance-individuated with respect to certain of their contents. What makes these perceptions have the content they do is the fact that when the subject is properly related to the world, the holding of these contents causally explains the subject's experience as of their holding. Even when the subject is properly connected to the world, and the environment is normal, still some spatial, temporal and bodily contents of these sorts do misrepresent. There are some reliable illusions—such as the Miiller-Lyer—which occur even in ordinary circumstances when the embedding of subject is as proper as it is ever going to be. In these cases, the experiences have the contents they do because of their relations to those experiences that are directly instance-individuated with respect to perception. These cases of illusion can be described as derivatively instance-individuated with respect to perception. We can then formulate this generalization at the second level about the relation of entitlement: A perceptual experience which represents a content as correct, and which is instanceindividuated with respect to that content, is also one which entitles a thinker to judge that content, in the absence of reasons for doubting that he is perceiving properly. This needs some adaptation if perception has nonconceptual content. Suppose it does. Then for conceptual contents for which there is a perceptual entitlement to judgment, there will be a range of nonconceptual contents of experience which generate an en-

448

T h e Epistemic Significance of Perception

titlement to judge such a conceptual content. Call these 'the range of nonconceptual contents which canonically correspond to the conceptual content'. The generalization at the second level would then be formulated by saying that: A perceptual experience which represents a nonconceptual content as correct, and which is instance-individuated with respect to that nonconceptual content, is one which entitles a thinker to judge a conceptual content as correct, in the absence of reasons for doubt, when the nonconceptual content is in the range which canonically corresponds to the conceptual content. Under either variant, the generalization at the second level is equivalent to something simple and intuitive. The generalization is in effect saying that when making perceptual judgments, one is entitled to take it, in default of evidence to the contrary, that one is in the circumstances with respect to which one's perceptions are instance-individuated with respect to the contents in question. So I will call this thesis about the second level 'the Individuation Thesis about Perceptual Entitlement'. This Individuation Thesis suggests an approach to the issue of what makes something an observational concept. The intuitive idea is that a non-observational concept will not be instance-individuated, because it has commitments which go beyond what is involved in instance-individuation. An experience of something as a Mac computer, or as a PET-scanner, cannot be purely instance-individuated, because that would not capture the commitments of these concepts, the commitment that objects falling under them are capable of carrying out certain functions. Such a development of a criterion for observationality has to be carried out with some care, because instance-individuation is not to be taken as meaning that nothing more than causal interaction is involved in an experience's having a certain content. There is causal interaction in ordinary visual experience with patterns of light reaching the eye; with retinal stimulation patterns; and with the state of the optic nerve. None of these matters enter the representational content of ordinary visual experience. The representational contents of visual experience also serve as input to the subject's construction of a conception of the layout of the objective spatial world around him. This is a feature of perceptual experiences with instance-individuated contents. So the criterion for a concept to be non-observational might be better formulated thus. Non-observational concepts have commitments going beyond the minimal conditions for objective content which are met by the contents of experiences which are instance-individuated. I simply note the possibility of this approach to observationality, as one by-product of the Individuation Thesis about Perceptual Entitlement. It would take us too far off our main path to pursue here the further elaboration which would be necessary to develop the criterion in detail, and with a rationale. Our actual entitlement to perceptual judgements rests on far more than the Individuation Thesis alone, applied atomistically experience-by-experience. Consider a stream of experiences, each of which entitles a thinker to believe some corresponding content. If these objective contents cohere, each being a spatial content of a perception reasonably expected to follow its predecessors, then the resulting entitlement to judge each content is massively greater than if each experience had occurred in isolation. Holism

Explaining Perceptual Entitlement

449

of confirmation is as pervasive in the sphere of perceptual judgements as it is in other areas. Even the proposition that objects have rears is something that requires perception from more than one angle, and cannot be confirmed by a single view. All the same, it seems to me that this holism of confirmation serves to increase a prior level of entitlement that can already exist in the individual case before additional perceptions or evidence are brought in. T h e additional perceptions or evidence are important because they can give further reasons for thinking things are as an initial perception represents them as being. (They can also serve to show that certain kind of defeating conditions do not hold.) In a sequence of coherent experiences, the later experiences themselves give defeasible reasons for making certain judgements, independently of the occurrence of the earlier experiences. It is only because this is so that the later experiences can then give further confirmation of the judgements supported by the earlier experiences. There is an abstract, structural argument that if rational, entitled thought is to be possible at all, some concepts must be such that one is default-entitled to presume that one is in the circumstances in which they are individuated. Maybe there could be a concept whose possession-condition makes reference to applications in circumstances one is not default-entitled to presume are one's own. Perhaps there could be a concept which, as a matter of its nature, is to be applied to objects which look a certain way, but only under a certain kind of abnormal illumination. Any entitled application of this concept on the basis of experience will require inference, or some other entitled transition, to the conclusion that the illumination is of the special abnormal kind. Now could it always be that inference, or some other entitled transition, has to be made before we are entitled to apply a concept? It seems that this could not be so, if entitled application is ever to get started. This abstract, structural argument seems to me to be sound. But abstract arguments by themselves have only abstract conclusions. It is one thing to know that default entitlements must exist. It is another to explain how they are possible in the first place, and to explain why they have the particular character and contents they do. T h e abstract argument does not give us an understanding, of any particular generalization about the entitlement relation, of why it, rather than some other principle, holds. To move towards such understanding in the perceptual case is the purpose of my next question, which is at the third of the three levels I distinguished. The question is: if the Individuation Thesis about Perceptual Entitlement is a true generalization about entitlement, what explains its truth? There would be no further task of answering this explanatory question if the Individuation Thesis about Perceptual Entitlement were derivable simply from the truths about the individuation of perceptual content together with principles about the nature of entitlement in general. But I cannot see what such a derivation from those premises alone would be like. The generalization which is the Individuation Thesis about Perceptual Entitlement does not itself explain why one is, in the default case, entitled to accept what would hold in the circumstances with respect to which perceptual content is individuated. Those circumstances may be special from the standpoint of the theory of the individuation of content; but what is so special about them for epistemology and the theory of entitlement? Why is one entitled, in the default case, to form perceptual

450

The Epistemic Significance of Perception

beliefs as if one were in the circumstances with respect to which the content of the perception is individuated? To achieve philosophical understanding of these issues, we have to undertake the further substantive philosophical task of explaining the epistemic significance of facts about the individuation of perceptual content. It is part of the task of connecting the epistemology of the theory of content with its metaphysics.

3. The Argument Outlined, and the First Step I now turn to the level-three task of explaining the second-level generalization about perceptual entitlement that we have identified in the previous section. T h e explanation will have the following three steps. Step 1 consists in the formulation and defence of a general principle about the explanation of complexity. Step 2 is the claim that the general principle in Step 1 applies to the explanation of the occurrence of experiences with instance-individuated contents, and does so in such a way as to support the presumption that the contents of such experiences are correct. Step 3 is the claim that this application to perceptual states provides the philosophical explanation, at Level 3, of why one is entitled to take experiences with instance-individuated contents at face value. T h e argument has two extensions. T h e first extension applies the argument to all as-if' states—to all states which are states in which it as if some condition obtains. A second extension, even more general in character, applies the argument to all nonconclusive entitlements. Let us take the first step. Here are some illustrations of what I mean by complexity, with some discussion of each case. (a) Snowflakes I start with the humble example of the snowflake. Although no two snowflakes have the same shape, almost all exhibit six-fold symmetry. Each undamaged snowflake repeats a pattern around its center, with a repeat each 60-degree segment. This is an example of complexity or order that needs explanation. (The problem of what the explanation is attracted the attention of Kepler, who wrote a pamphlet about it. 10 ) T h e explanation is as beautiful as the patterns it explains. 11 It is well-known that the most efficient way

10 J. Kepler, The Six-Cornered Snowflake, tr. C. Hardie (Oxford: Oxford University Press: Oxford, 1976). 11 For an overview, see I. Stewart, Life's Other Secret: The New Mathematics of the Living World (London: Allen Lane, 1998), 30-35; and for further details, R. Davey and D. Stanley, 'All about ice', New Scientist December 18, 1993, 33-37. 12 G. Nicolis and I. Prigogine, Exploring Complexity: An Introduction (New York: Freeman, 1989), pp-8-

Explaining Perceptual Entitlement

451

of packing spheres on a plane results in a hexagonal arrangement. Oxygen molecules in frozen water are roughly spherical, they are arranged on a plane, and the frozen crystals grow in a way that minimizes energy. They do not grow in solid hexagons, however, because long sides tend to break. Growing arms also develop other arms, by the same principles. T h e differences between individual snowflakes are explained by the varying conditions of temperature, moisture flow and so forth in which they are formed. What matters for present purposes about this explanation is that the complexity or order at the level of the whole snowflake is explained in a way that does not simply presuppose a similar complexity or order in the states or conditions that do the explaining. There is no appeal here in the explanation to objects that themselves exhibit a similar six-fold symmetry that is simply taken for granted; and it would hardly provide a satisfying explanation if there were. Any such alleged explanation, for instance one saying that snowflakes are built on skeletons that themselves exhibit six-fold symmetry, would simply leave unanswered a major part of the question of how this species of complexity came about. 'And why do those things have six-fold symmetry?', one would reasonably ask. The correct explanation of the shape of snowflakes does not leave us with the same complexity again at another level. It reduces—in this case it eliminates—that kind of complexity. (b) Benard cells in convection patterns A second example concerns convection currents in heated liquids. Take a closed volume of a liquid that receives heat from a source below it. Within a certain range of quantities of heat applied from below, this liquid will reliably divide into a set of what are known as Benard cells, within each of which the liquid circulates in a rotating pattern, when one takes a vertical cross-section of the liquid. Consider a small quantity within the fluid that is moving upwards. This small quantity enters a cooler region of greater density than it has itself, and so is buoyed further upwards. The opposite applies to a falling quantity. Particles at the top of the liquid are constrained to move horizontally by top of the container. 12 This explanation of how the convection cells arise does not appeal to any prior division of the liquid into cells or volumes. T h e complexity or order is explained in a way that does not beg any such questions. (c) Amoebic behaviour: order over time Sometimes the complexity or order to be explained has to be characterized in terms of the environmental relations of the system or object in question. A group of amoebae whose supply of nutrients is reduced tend to aggregate into a single multicellular body. This environmentally described regularity needs explanation. The explanation involves the emission by starved amoebae of a chemical (cAMP) which attracts other amoebae to it, and causes them to emit the same substance. 1 3

15. 13 Nicolis and Prigogine, 31-36. 14 See his Complexity and the Function of Mind in Nature (Cambridge: Cambridge University Press, 1996),

452

The Epistemic Significance of Perception

This explanation does not appeal to any principle that starved organisms tend to coalesce. No such principle is true. Even if it were, it would still need explanation of how organisms to which it applies succeed in coalescing. The correct explanation shows why the complex coalescence occurs without leaving a residue of unexplained complexity. It of course needs explanation why organisms displaying such complexity survive, and why this behaviour is adaptive; that is a good question, but a different one, and it can equally be answered. One needs both an explanation of how complexity occurs, and another—and different—explanation of why objects or organisms which display such complexity continue to exist and reproduce. These three initial examples give a first fix on the notion of complexity or order, and what is needed for a satisfactory explanation of particular instances of complexity. The general phenomenon of order needing a certain kind of explanation has illustrations in widely diverse areas. Illustrations range from the biochemical phenomenon of replication of a molecule, through biological phenomena of reproduction, to such economic phenomena as economic cycles of growth and recession. All of these involve a kind of order. All the satisfying explanations of these cases of order do not simply reproduce the complexity to be explained. But can we say something general about the nature of this complexity? What is it, and why does it have to have a certain kind of explanation? The kind of complexity which needs explanation cannot be a matter of the improbability of the particular arrangement of particles whose distribution constitutes the complex state on a given occasion. The instantaneous distribution of particles at a given moment in a case of convection currents in Benard cells is no less improbable than an equally detailed specification of the distribution of particles in a case of random Brownian motion. The same applies to sequences of such distributions over time. Correspondingly, the task of explaining this kind of complexity should not be described merely as the task of explaining the occurrence of the very improbable. Nor does complexity consist in the objective improbability of the occurrence of the particular kind oi arrangement of particles that makes us classify a case as one involving complexity. On the contrary: a good explanation of the six-fold symmetry of snowflakes shows that that kind of pattern is highly likely, almost inevitably, occurs. A good empirical explanation of the characteristic feature of a particular instance of complexity can hardly be one that shows the case is not one of complexity after all. Both the preceding attempts to explain complexity in terms of improbability misarticulate what seems to me the correct account of the phenomenon exemplified in the initial examples. The improbability in question is apparent, rather than real. Complex phenomena are ones which instantiate kinds that are apparently improbable; but in fact there is an explanation of why those kinds are instantiated. The challenge is to say what the explanation is. An explanation must consist in showing how what seems unlikely can in fact come about with less difficulty than seemed to be the case. It may be almost inevitable, as with the symmetry of snowflakes. The explanation shows why the empirically possible—as opposed to geometrically possible—shapes for a snowflake all exhibit six-fold symmetry. All the examples of complexity their explanation that we have considered so far conform to this description: there is a wide variety of apparently possible states for some object, event or process, but there is an explanation of why the

Explaining Perceptual Entitlement

453

actual instances all fall within some restricted, specifiable narrow range o f this wide variety o f apparently possible states. T h e fact that all o f the many undamaged snowflakes exhibit six-fold symmetry is evidence that there is some underlying uniform explanation. But frequency o f occurrence should not be written into the characterization o f what complexity is, or into the conditions under which we need an explanation o f it. As Roger White remarked to me, even if there were only one snowflake ever in the universe, there could and would in fact still be an explanation o f its six-fold symmetry. W h e n complexity has an explanation, the explanation can apply equally whether the instances are frequent or whether they are rare. T h e complexity which has an explanation (if it is not merely coincidental) is not a merely mind-dependent property. It is true that the six-fold symmetry o f snowflakes is perceptually salient, and that Benard cells are correspond to perceptual groupings when they are illustrated. But complexity or order can exist, and we can have evidence that it exists and has an explanation, even when it is not perceptually salient. T h e Fibonacci series is widely exemplified in nature, so widely that it is plausible that there will be an explanation, or explanations, o f why it is so. But one does not need to be able to perceive the locations or angles at which (for instance) new shoots grow as instances o f the Fibonacci series for the phenomenon to need explanation. T h e instantiation o f a complex property can have, and need, an explanation whether or not that property is perceptually salient. When there is an explanation o f a complex property o f some object or event, there is an explanation o f why the object or event has a property which falls within a narrow range o f the space o f possible properties o f that object or event. Shapes with hexagonal symmetry form a small subset o f the geometrically possible shapes for a quantity o f a frozen liquid. W h a t needs to be explained is why the shape o f actual snowflakes fall within that narrow subset. In this description o f the task, neither what is to be explained, nor what it is to be an explanation o f it, seem to me to be mind-dependent. T h e notion o f complexity I am using is not the only extant notion, nor the only significant one. Peter Godfrey-Smith, for example, explains a notion o f complexity which is intuitively a notion o f heterogeneity, and puts it to good theoretical use. 14 This is just a different notion from complexity as a certain kind o f order in the range o f actual phenomena, which is the notion I am pursuing. T h e range o f geometrically possible shapes which might be taken by a solid built from water molecules (independently o f miminization o f use o f energy) goes vastly beyond those with six-fold symmetry. T h a t is, it is more heterogeneous than the actual range o f shapes o f snowflakes, and so more complex in Godfrey-Smith's sense. But the actual range, with its orderliness, is more complex in the sense with which I am concerned, and which has been intended in the tradition in which Nicolis and Prigogine, for instance, are writing. Any explanation o f complexity as it is conceived here must explain more complex states by less complex states. An explanation which did not do so would not have shown how the apparently unlikely could easily come about, or could come about in less com-

15

at pp.24-28. In his Brainstorms (Montgomery, Vt.: Bradford Books, 1978).

454

T h e Epistemic Significance of Perception

plex ways than one might have expected. If a proposed explanation simply reproduces the complexity to be explained, one will have explained the apparently unlikely in terms of the equally apparently unlikely. One will not have shown how the complexity could easily have come about. Similarly, in cases in which a complex kind is frequently instantiated, one will not have shown why it should be so frequently instantiated. I summarize this point in the 'Complexity Reduction Principle': Genuine explanations of complexity explain the more complex in terms of the less complex; they reduce complexity. In accepting the Complexity Reduction Principle as a constraint on good explanations, one is committing oneself to the intuitive principle that it rational, other things equal, to hold that things come about in easy, rather than more unlikely, ways. That it is rational to hold that things come about in the ways they are more likely to seems to me an a priori principle. It is not something extracted from any particular science, but is presupposed in all empirical sciences. What various empirical sciences tell us is not that things tend to come about in the easier ways. Rather, they tell which ways are the easier ways. Which ways are the easier ways is a wholly empirical matter. Particular sciences may have surprises about which ways are the easier ways. It is an objective matter how easy it is for an event or state of affairs of a given kind to occur in given circumstances. It is not an epistemic matter. In making judgements about easiness, we should be aiming to get this objective matter right. One way is an easier way than a second for a certain state of affairs to come about if the first way is exemplified in a wider range of initial conditions that could bring about the state of affairs—where this 'could' is empirically (rather than being the 'could' of pure metaphysical possibility). In assessing how easy it is for an event of a given kind to occur in given circumstances, we have to draw on all sorts of information about those circumstances. The easiest way for a piece of inanimate matter in outer space to move may be for it to be pushed by some other piece of matter, and not for it to be controlled by states with informational content. But for a piece of matter such as human arm, that we know is under the control of states with intentional content, the easiest way for it to move in given circumstances is for its owner to move it. In giving examples of the explanation of complexity, I considered the illustrations provided by complex spatial patterns, by complex relational behaviour, and by complexity over time. In some cases, complexity is encoded or implicit in some relatively simply described state. Many psychological states with intentional content are like that. The state is simply specified by giving the kind of state in question, together with its intentional content. For a state to have a given intentional content, however, it must stand in an extremely complex network of relations to other states and to the subject's environment. Any explanation of how the subject comes to be in that state, an explanation that accounts empirically for the presence of this complexity, must not simply presuppose similar intentional complexity. It is for this reason that the spuriousness is so widely acknowledged of a purported explanation of someone's ability to recognize his grandmother by postulating a 'grandmother' neuron. In fact the whole methodology described by Dennett in 'Artficial In-

Explaining Perceptual Entitlement

455

telligence as Psychology and as Philosophy' of explaining intelligent capacities in terms of less intelligent ones can be regarded as the application, to the special case of certain psychological capacities, of the Complexity Reduction Principle. 15 This methodology, both in artificial intelligence and in psychology, involves the explanation of rich intellectual capacities in terms of the activities of subsystems involving progressively less intelligence, until we eventually reach a level at which the subsystems involved have no more capacities than can be explained in physical terms such as the firing of assemblies of neurons in response to certain patterns of stimulation of those assemblies. A purported explanation that contained an ineliminable violation of this reduction in intellectual capacities would also be a violation of the Complexity Reduction principle. It would be offering an explanation of some intellectual, or more generally content-involving, capacity in terms that presupposed that capacities of a similar degree of intellectual richness were already present. This would involve unreduced, unexplained complexity given the relatively uncontroversial premise that possession of some intellectual capacity is a complex state needing explanation. We can no more accept intellectual capacity as an unexplained, primitive feature of an organism than we could accept primitive, unexplained six-fold symmetry of certain arrangements of matter in a purported explanation of the shape of snowflakes. In the case of evolutionary biology, the appeal to step-by-step evolution to explain such matters as the existence of as complex and subtle an organ as the eye, and more generally the ability of organisms to survive and reproduce equally involve an application of the Complexity Reduction Principle. Richard Dawkins is explicit about some aspects of this methodology (I have demurred from the point in the following passage about probability, which he himself later qualifies): " A complicated thing is one whose existence we do not feel inclined to take for granted, because it is too 'improbable'. It could not have come into existence in a single act o f chance. W e shall explain its coming into existence as a consequence o f a gradual, cumulative, stepby-step transformations from simpler t h i n g s . . . " . 1 6

Dawkins' 'could not' here is not that of empirical implausibility, rather than that of metaphysical impossibility.

4. The Second Step: The Application to Perceptual Experience Consider a particular occurrence of a perceptual experience with an instance-individuated content that p. This is an event of considerable complexity. Its complexity is in part relational, in two respects. First, to have this instance-individuated content that p, the experience must be of a kind which, when the subject is properly connected to the world, has its instances caused by the fact that p (or else it is derivatively instance-individuated), in the sense discussed earlier. The spatial and temporal experience of organisms, even relatively 16

R. Dawkins, The Blind Watchmaker

(Harlow, Essex: Longman, 1986), p. 14.

456

The Epistemic Significance of Perception

primitive ones, with well-developed perceptual systems, will in fact have many contents concerning spatial and temporal properties and relations that are instance-individuated in this sense. There is much about the particular experience under this first head alone that requires empirical explanation. There is a second respect in which this experience displays a relational complexity. An experience with spatial representational content, for instance, must be one whose content is capable of contributing to its subject's conception of the spatial layout of the world around him. Without this, the experience would not amount to having a spatial content at all. T h e content of the experience must be capable of integrating with other spatial representations in confirming or disconfirming the subject's conception of the layout of the world. The same applies pari passu to the temporal contents of perceptual experience, and to their role in building up simultaneously the subject's conception of his history, and the history of the world around him. So in these spatial and temporal examples, the requirement of causal sensitivity to instances of the properties and relations represented as instantiated are only necessary conditions for a perception to have a content of one of these kinds. They are not sufficient conditions. These most recent holistic elements in the possession of spatial and temporal perceptual content all contribute further to the complexity of the property of having an experience with a given spatial representational content that p. Taking both these kinds of complexity together, we have another case of complexity that needs empirical explanation; and the Complexity Reduction Principle will apply here too. I suggest that the explanation of the occurrence of a perceptual experience with the instance-individuated content that p which most reduces complexity is that the experience is produced by a device which has evolved by natural selection to represent the world to the subject. This involves, other things being equal, the perceptual experiences produced in such a subject being predominantly correct. The occurrence of an experience with the representational content that ρ would hardly be a coincidence if its representational content held of the world, and the subject had a properly functioning perceptual system whose holistic complexities were adapted to its spatial embedding in the world. For a subject with such a perceptual system, its being the case that ρ would in the predominance of cases then explain the occurrence of an experience with whatever complex relational property is involved in representing it as being the case that p. It is a relatively a priori truth that since subjects rely substantially on their perceptual systems in the formation of belief, there will be selection for roughly accurate perceptual systems. As always in evolution by natural selection, there are trade-offs. Some, perhaps considerable, inaccuracy may be traded for speed or range of representations. But perceptual states generated by a system produced and sustained by natural selection can be expected to be in large part correct in their representational contents. T h e explanation by natural selection of the existence of roughly accurate perceptual systems reduces complexity. The explanation succeeds by citing states of affairs of lesser complexity than that which is to be explained. The explanation does not postulate the occurrence of other intentional states in the production of the perceptual experience. Nor does it postulate other unexplained states of the same relational complexity as those to be explained.

Explaining Perceptual Entitlement

457

Why does the explanation of entitlement apply only to the contents of instanceindividuated perceptual experiences? It would not be a satisfying answer to this question to say: "The entitlement does not, when we consider the cases intuitively, apply beyond the instance-individuated". If the proposed complexity-reduction explanation really were to apply also beyond the instance-individuated cases, that would then constitute a serious objection to this theory. For by the lights of the complexity-reduction account, there ought also to be a corresponding purely perceptual entitlement in the cases of contents of experience that are not instance-individuated—when actually there is no such entitlement. In fact it seems to be intrinsic to the complexity-reduction account that it applies only to the instance-individuated cases. The complexity displayed by experiences with instance-individuated contents is one which involves a causal sensitivity, when the subject is properly connected to the world, to instances of the very properties and relations represented in the experience. This is not true of those elements of the representational content of the experience which are not instance-individuated, such as soldier, clock or computer. It is neither necessary nor sufficient for an experience to represent someone as being a soldier, or something as being a clock, or a computer, that experiences of that kind be caused, when the subject is properly connected to the world, by soldiers, or clocks, or computers. It is entirely sufficient that the perceived features which also cause the subject to perceive something as a soldier, clock or computer be ones which the subject has evidence, good or bad, are sufficient for something to be a soldier, clock or computer, and whose sufficiency for this has been, by some general mechanism, absorbed into the content of the perception itself, so that things are perceived as being soldiers, clocks or computers. This is certainly an easy way for an experience to come to have the complex property of having a representational content involving these concepts soldier, clock or computer. This is an explanation which does not, in itself, imply that experiences with these representational contents are likely to be veridical. It is neutral on that issue (though it does not preclude that further arguments to that conclusion could be developed). The same explanation could not, in the nature of the case, be given for the occurrence of experiences with instance-individuated contents. In these cases, it cannot be true that the perceiver has evidence (good or bad) that certain perceived features are sufficient for something's being straight, or curved, or, for instance, to be moving slowly. There is no further, more fundamental level of representational content of which it is true that things represented as being certain ways at that level of content are also, as an empirical matter, straight, or curved, or are moving slowly. We are, at the instanceindividuated level, already at the most basic level of representational content. There is nothing more primitive to fall back upon, and which might be merely empirically associated with these instance-individuated contents. This is why the complexity-reducing explanation that appeals to selection of roughly accurate perceptual mechanisms really is limited to the instance-individuated cases. For experiences with contents that are not instance-individuated, that need not be the explanation that most successfully reduces complexity. 17 17

O n the perception of the expression of emotions, I've equally said that there is no fallback, more primi-

458

T h e Epistemic Significance o f Perception

I will be arguing that the various sceptical hypotheses fail to reduce complexity. My thesis will be that it is not a good explanation, one reducing complexity, to suppose, in the absence of specific information to the contrary, that an experience with the complex property of having instance-individuated representational contents occur without their being predominantly explained by the correctness of those representational contents. Sceptical hypotheses about the explanation of perceptual experience can be divided initially into two broad classes. There are those that hypothesize that some agent is intentionally causing non-veridical experiences. Descates' evil demon and the scientist who controls your brain in a vat fall into this first class. The other class of hypotheses do not involve any intentional production of illusions, but suggest either random or coincidental physical events that result in combinations of matter that produce illusory experiences; or else they suppose, for instance, that the universe has always consisted or one or more envatted brains. Sceptical hypotheses that postulate intentional agents as the source of the experiences cannot reduce complexity. If the agents producing the illusions themselves have experiences, complexity is manifestly not reduced. The explaining states are as complex as those whose complexity is in need of empirical explanation. If the agents have other complex attitudes, or attitude-like states, without having experiences themselves, these attitudes still display a form of relational complexity that needs empirical explanation. T h e emergence of propositional attitudes other than experience can be given in evolutionary theory and its byproducts, in a way that does not leave a residue of empirically unexplained complexity. T h e sceptical hypotheses as normally formulated do not offer such empirical, complexity-reducing explanations of the emergence of the attitudes of the deceivers. It would of course be possible to modify them to provide such an empirical explanation of the complexity of the hallucination-producers own states. But if that is then counted as an improved explanation because it reduces complexity, it seems that a greater improvement would be made by not introducing the sceptical hypothesis in the first place. What of the sceptical hypotheses in which the hallucinations are not intentionally produced? The aim of complexity-reduction in explanation is to show how the apparently unlikely can easily come about after all. This aim can hardly be met by the hypothesis of chance or coincidental motions of matter that produce hallucinations. This is to explain the apparently unlikely in terms of the genuinely unlikely. What of the hypothesis that the universe has always consisted of one or more envatted brains? Although this does not involve an unlikely kind of event or a coincidence at a particular moment, this still involves initial conditions in the universe that that are highly complex. Why should the initial conditions (vats producing hallucinations) involve conscious events with the complex relational properties we described? This remains a case of empirically unexplained complexity.

tive level. T h i s could be the starting point for a demonstration that the easiest way for these experiences to c o m e about involves not just causation by instances, but involves the subject's capacity to experience the s a m e emotion, and to express it.

Explaining Perceptual Entitlement

459

T h e hypothesis of a world in which there are and always have been permanently envatted brains does raise another challenge for the present account. Wouldn't it be true of such a world that the easiest way for a perceptual experience to occur is for it to be caused by stimulation of one of the brains in the vats? Such an experience would be a hallucination. Doesn't this point mean that the whole approach to explaining perceptual entitlement by means of the notion of the easiest way in which something can come about, and by Complexity Reduction, is undermined? I agree that there is a reading under which it is true that in the world of permanently and eternally envatted brains, the easiest way for it to come about that a perceptual experience occurs will make that experience a hallucination. It would also completely undermine the present approach, or else lead to scepticism, if one were to reply to this point by drawing on empirical knowledge that the world is not in fact that way. That response has two defects. First, it cannot fully explain the empirical entitlement for the knowledge that the world is not in fact that way—its status as knowledge is just taken for granted in this response, rather than being explained. In the face of the objection, appeal to the easiest way in which something can come about would be begging the question. Second, any such response would not leave the status of perceptual entitlement as relatively a priori. It would rather be empirical; and then it would not be clear how one could become entitled to believe anything about the observable world at all. A better response is to distinguish between what can easily be the case, given that certain conditions hold, from what can easily be the case absolutely, without such relativization to certain conditions. The easiest way for it to come about that an experience occurs, given that the universe contains permanently envatted brains, is for one of these brains to experience a hallucination. It does not follow that the easiest way for a perceptual experience to occur, given no information about the conditions in the world, is for there to be envatted brains and for an experience to occur in one of them. That is precisely what I have been arguing against. There are many other cases in which we would draw the distinction between the unrelativized and relativized statements of the easiest way for something to come about. Given that D N A molecules already exist, both in natural organisms, and in laboratories, the easiest way for a molecule of a specified type of D N A to come into existence is for one of the existing molecules to be copied, with a certain modification. Without relativization to the condition that D N A molecules already exist, the easiest way for the specified kind of molecule to come into existence is by some kind of chemical evolution. There is similarly relativization to various conditions in statements of apparently non-epistemic, objective probabilities, for instance statements about radioactive decay. The objective probability that an alpha-particle will be emitted in a given time-interval from a given quantity of matter is greatly increased relative to the condition that the matter is plutonium, rather than being some naturally-occurring substance. This relativization to specified conditions should not be taken to show that the notion of the easiest way for something to come about is epistemic. T h e relativization merely cuts down the range of worlds one considers in assessing how easy it is for as given condition to be met. Within that restricted range, it is still a non-epistemic matter how easy it is for the condition to be fulfilled.

460

The Epistemic Significance of Perception

It is a claim of the unrelativized kind I intend when I say that the easiest way for a perceptual experience to occur is one in which it is unlikely to be a hallucination. T h e case for this claim, whether right or wrong, is made on philosophical grounds, and does not rely for its justification on empirical information attained by perception. There is no relativization in this claim to conditions which are known to hold only on empirical grounds. It has been noted in several discussions of sceptical hypotheses, and emphasized particularly by Jonathan Vogel, that there are many arbitrary elements as these sceptical hypotheses are filled out in more detail. 18 There are so many 'unexplained explainers': why should someone want to deceive undetectably? What is their history? Why should there be these envatted brains from eternity? O n the present account, this excess of unexplained explainers, this general lack of constraint, is a consequence of failing to respect the requirement of Complexity Reduction in good explanations. Once that requirement is abandoned, the proposed explanation of the occurrence of perceptual experiences may be arbitrarily baroque. By contrast, there is a parameter that is not adjustable on natural-selection explanations of the emergence of perceptual systems. T h e explanation, now, of the occurrence of a range of experiences with instance-individuated contents is that the world is, predominantly, as they represent it to be. This talk of an excessive number of adjustable parameters in the sceptical hypotheses may remind one of the Akaike theorem about the distance from the truth of a family of curves in the curve-fitting problem. 1 9 Akaike's measure of the distance from the truth of a family of curves contains a term Ikrj2 where k is the number of adjustable parameters in the equation for the family of curves in question, and σ 2 is the variance of the distribution of errors around the true curve. It may be tempting to elucidate all my talk of reduction in complexity in terms of reduction in the number of adjustable parameters in a range of theories. T h e Akaike result is of great interest, and there may be further connections to be elucidated between what I am talking about and his parameter k—but I do not think they can be quite the same thing. There is a sensitivity of the measure of the distance from the truth in Akaike's measure only in the case in which O 2 is not zero—that is, in the case in which there are some errors in the data points. But as far as I can see, the intuitive notion of complexity reduction I have been emphasizing gets a grip even when we entirely prescind from errors in our data points. Even if we have a set of true statements about experiences, with no errors about which experiences are occurring, still there is a truth-related reason to prefer theories which reduce complexity—in this case, theories which explain the relational complexity of experiences. Sceptical hypotheses do have excessive numbers of parameters compared with their non-sceptical competitors, and in the case in which there are errors in our data about which experiences occur, a family of sceptical hypotheses will, other things

18 19

'Cartesian Skepticism and Inference to the Best Explanation', Journal of Philosophy 87 (1990) 658-66. For an exposition for philosophers, see M. Förster and E. Sober, 'How to Tell when Simpler, More Unified, or Less Ad Hoc Theories will Provide More Accurate Predictions', British Journal for the Philosophy of Science 45 (1991) 1-35. See also Section 4 of E. Sobers article 'Simplicity (in scientific theories)' in the Routledge Encyclopedia of Philosophy (London: Routledge, 1998).

Explaining Perceptual Entitlement

461

equal, thereby be further from the truth than s o m e families o f non-skeptical hypotheses under the Akaike measure. B u t I d o not think this gives us an account o f w h a t it is for a theory to fail to reduce complexity, as o p p o s e d to a consequence o f such failure. T h e s e claims I have m a d e a b o u t the easiest way in which perceptual experiences with instance-individuated contents m a y c o m e a b o u t seem to put m e in disagreement with at least part o f what is intended in two widely held theses a b o u t simplicity a n d confirmation. T h e y are widely held a m o n g s t thinkers w h o disagree on m u c h else. T h e first o f these theses states that simplicity is irrelevant to c o n f i r m i n g s o m e t h i n g as true. Bas van Fraassen, for instance, regards simplicity o f a theory as a p r a g m a t i c virtue, o n e speaking specifically to h u m a n concerns. O f the p r a g m a t i c virtues in general, he writes that "In so far as they go beyond consistency, empirical adequacy, and empirical strength, they do not concern the relation between the theory a n d the world, b u t rather the use and usefulness o f the theory; they provide reasons to prefer the theory independently o f questions o f t r u t h " . 2 0 Lycan writes "Simplicity? A b s e n c e o f mess? W h y not prettiness and niceness? W h y should these virtually aesthetic properties, which s m a c k in any case o f laziness a n d corner-cutting, be thought to c o u n t in a n y way towards truth ( J o h n Keats notwithstanding)?". 2 1 O n e could q u o t e m a n y other writers to similar effect. As an essential preliminary to considering this view, I suggest that w e need to distinguish simplicity o f an explanation f r o m simplicity o f a state o f affairs. In particular, a simple explanation need not cite an explaining c o n d i t i o n whose truth consists in a simple state o f affairs. U n d e r o n e o f the no d o u b t m a n y kinds o f simplicity, a s i m p l e explanation is o n e that reduces complexity, in the way we have been discussing. In a g o o d explanation the cited explaining conditions are, other things equal, less c o m p l e x than those being explained in the sense o f complexity I tried to elucidate earlier. T h e explaining condition in an explanation that reduces complexity m a y not be a s i m p l e state o f affairs on all intuitive notions o f simplicity. In fact, in s o m e cases it w o u l d be problematic were it to be so. All o f the following are simple states o f affairs in an intuitive sense: the state o f affairs o f the northern half o f a globe being entirely covered with ocean, a n d the southern half being entirely dry land; the state o f affairs o f a plane surface containing just o n e black square, the rest being entirely white; the emission o f a particle o f a given kind at regular one-second intervals by s o m e quantity o f matter. T h e s e are all orderly states o f affairs. T h e y w o u l d b e c o u n t e d as simple on the a c c o u n t o f simplicity given by Elliot S o b e r . 2 2 In the sense o f complexity we were relying o n earlier, all o f these s i m p l e states o f affairs are complex! In the range o f all possible states o f the entities involved, they instantiate a distinctive property f o u n d in only s o m e o f those possible states. S u c h simple states meet the earlier characterization o f complexity, a n d their occurrence requires explanation. For those w h o regard complexity as a k i n d o f orderliness, given that simplicity is itself a k i n d o f orderliness, it is clear that c o m plexity a n d simplicity m u s t in many, if indeed not all, cases overlap.

20 The Scientific Image (Oxford: Oxford University Press, 1980), p. 88 21 Judgement and Justification (Cambridge: Cambridge University Press, 1988), p. 134 22 E. Sober, Simplicity (Oxford: Oxford University Press, 1975).

462

The Episcemic Significance of Perception

I have suggested that good explanations of the occurrence of complex states involve a reduction in complexity from what is to be explained to the explaining conditions they postulate. I suggest too that the reduction in complexity makes it more likely that such an explanation is true. There was a reduction in complexity, and corresponding simplicity of explanation, in the explanation of the six-fold symmetry of snowflakes, of the existence of convection patterns in heated liquids, and in evolutionary explanations of the existence of complex capacities and organs. It is not credible that these explanations are preferred merely on pragmatic grounds, or as a matter of aesthetic preference. Nor is it credible that a proper preference for these explanations has nothing to do with truth. Insofar as simple explanations are ones that reduce complexity, the simplicity of an explanation is not something mind-dependent, or something speaking merely to human concerns. I am therefore in disagreement with those writers on simplicity quoted earlier if their views do entail that the simplicity of the explanations in the snowflake or Benard cell examples is something mind-dependent. The second extensively-held thesis about simplicity and confirmation with which I am committed to disagreeing is that the evil-demon hypothesis about the explanation of perceptual experience is equally well-confirmed as the real-world hypothesis. When what is to be explained is not just the occurrence of a series of experiences, but the occurrence of something with the complexity of the relations in which they must stand to have the representational contents they do, the evil-demon and other sceptical hypotheses, I argued, fail to reduce complexity. That is their defect, rather than failing to have the occurrence of the experiences in question as a consequence of their hypotheses. Those who hold this second thesis sometimes see it as a result of their acceptance of the first hypothesis. Sometimes too scepticism is thought of as a consequence of the view that only truth-relevant considerations can contribute to the confirmation relation. William Lycan calls this view of confirmation 'the spartan view'. He says "It is also the spartan v i e w that drives evil-demon scepticism about the external world: by h y pothesis, the evil-demon theory makes exactly the same observational predictions as does the realist external-world theory, so both are equally probable or well-confirmed on our evidence, so w e have not reason to believe the external-world theory to the exclusion o f the o t h e r " . 2 3

On the position I am defending, there can actually be acceptance of what Lycan calls the spartan view. What I do dispute is that the spartan view (that holds that only truth-relevant considerations are pertinent to the confirmation relation) must always omit anything to do with simplicity. If what I have argued is correct, the fact that an explanation reduces complexity counts in favour of its confirmation, because it is an explanation that does not make it hard or excessively improbable for the postulated explaining condition to be true.

23

'Theoretical (Epistemic) Virtues', in Routledge Encyclopedia of Philosophy, vol. 9, p.34l; note also the emphasis in this passage on the explanation of the occurrence of the experiences, rather than the complexity of the relations in which they must stand. Crispin Wright, in a seminar at NYU, has also argued that the evil-demon hypothesis is equally well-confirmed as the real-world hypothesis.

Explaining Perceptual Entitlement

463

5. The Third Step: The Philosophical Explanation of Perceptual Entitlement The third step of the argument is that the preceding considerations in Steps 1 and 2 provide the philosophical explanation of a thinker's entitlement to take at face value the instance-individuated content of a perceptual experience. Explanations that reduce complexity are more likely, other things equal, than those that do not. What explains the entitlement in question is the fact that explanations of the occurrence of experiences with instance-individuated contents which succeed in reducing complexity will also result in the representational contents of those instance-individuated experiences being predominantly correct. Such representational contents are predominantly correct in the case that is most likely, that of the complexity-reducing explanation which appeals to the evolution of a perceptual system through natural selection. The argument is open-ended in that I have not shown that explanations by natural selection of the existence of perceptual systems provide the only satisfactory explanation of complexity that succeeds in reducing complexity. I have not proved that there are no others: I have merely not been able to construct any. It is, however, striking that the wide range of sceptical hypotheses that have been presented over many centuries now do not succeed in reducing complexity. Complexity-reducing hypotheses other than those that lead to predominantly genuine perception of the world have not been easy to come by. The complexity-reduction explanation of the entitlement to take certain perceptual experiences at face value has two properties that we should require of any such explanation. First, it is a priori. We have not appealed to empirical features of our external environment in the actual world in explaining why this entitlement exists. If the explanation had done so, it would have failed to explain why the (defeasible) transition from instance-individuated representational content to judgement of the correctness of that content—or to a suitably related conceptual content—is itself an a priori transition, not resting on other empirical information. Second, as required, the explanation explains the defeasibility of the entitlement. Complex states may in fact have complex explanations. What is less likely is not thereby impossible. Further information that makes it rational no longer to take a perceptual experience at face value is information which also shows that the complexity of the perceptual experience does not have a maximally complexity-reducing explanation. Perceptual states have specifically representational (not merely intentional) content. In this they differ from states such as imagining or wishing, which have intentional content but not representational content. In being in a state with representational content, the subject of the state is thereby under the impression that the world is a certain way. Intuitively, it seems essential to the entitling character of perceptual states that their content is representational. But has this specifically representational character featured in the explanation offered at level 3 of perceptual entitlement? This feature is playing an essential role in the explanation. The complexity-reducing explanation of the occurrence of perceptual states invokes the adaptive advantages of having roughly correct perceptual states. But this is an advantage only if the representational content of these states is indeed taken at face value. The states would have no adaptive advantage otherwise if their contents were not taken at face value. But it

464

T h e Epistemic Significance of Perception

is only because they have representational content at all that perceptual states even have a 'face value'. O u r task was to explain empirically the complexity of perceptual experiences with representational content. The kind of explanation possible for these states will not be available for states without representational content. For states without that sort of content, some additional mechanism, involving inference or some other operation, would have to be invoked before any adaptive advantage can be established. I should note that in making this feature of representational content part of the level-3 explanation of perceptual entitlement, I am assuming that it is a priori that perceptual experiences, in default of reasons to the contrary, tend to produce corresponding acceptance of their representational contents. Only to the extent that this is a priori is the present explanation of perceptual entitlement also a priori. It is one thing to hold that complexity-reduction and its consequences provide a level-3 explanation of the entitlement to make perceptual judgements in the case of instance-individuated experiences. It would be another, and in my view false, to claim that ordinary thinkers who make judgements in an entitled way must know that this is the explanation of the entitlement (or even to know these points about complexityreduction without knowing that they explain entitlement). This is partially parallel to the case of valid transitions involving certain operators in logic. Ordinary thinkers are entitled to make certain transitions involving logical constants, or the counterfactual conditional, and so forth. They will recognize instances of certain transitions as valid, and others as invalid. It does not follow that they can state the explanation of why transitions of these various forms are valid or are invalid. For that, as we noted, a semantic theory is required. O n e difference stands out from the logical case. A case can be made that ordinary thinkers have some form of tacit knowledge of the axioms of a semantic theory, have (as I would say) implicit conceptions whose content is that of the axioms of a recursive semantic theory. Ordinary thinkers have the following actual abilities: to evaluate certain sentences as true or as false with respect to certain circumstances; to assess certain argument forms as valid and others as invalid; to appreciate the validity of new primitive forms of transition. These abilities are best explained by the postulation of tacit knowledge of the semantic contribution made by an expression to the truthconditions of the complex sentences in which it occurs.2^ But corresponding points do not apply to the explanation I have offered of the entitlement to take the contents of certain perceptual experiences at face value. What are the capacities or judgements of a thinker whose best explanation would be tacit knowledge of those points about complexity-reducing explanations? Of course thinkers will offer some kind of reaction when presented with sceptical hypotheses. But what they say in such circumstances is likely to have as little, or even less, to do with the explanation of why they are entitled to make perceptual judgements than the ordinary thinker's stabs at the formulation of grammatical rules have to do with the correct explanation of his grammatical and semantical competence. We should note also that the level-3 explanations of entitle-

24

See my paper 'Implicit Conceptions, Rationality and Understanding', Philosophical Issues 9 (1998).

Explaining Perceptual Entitlement

465

ment have to do with the philosophical explanation of normative truths, rather than the explanation of empirical capacities. W h a t is the explanation of this difference between the logical case and the observational case? A straightforward answer is that the possession-condition for the logical concept requires the thinker to have an implicit conception with a certain semantic content, while there is no such requirement for the existence of an implicit conception on the part of the thinker in the possession-condition for the observational concept. This seems to me to be a true answer—but the worry motivating the question is likely to need more if it is to be allayed. T h e worry is likely to be that we need to have a closer relation between the rationality of a transition and the fact that a thinker makes it than I have provided for in my account of the third level for perceptual entitlement. Here it seems to me important to avoid an illusion about the nature of the connection between the rationality of a judgement or transition and the fact that someone makes it. Consider a paradigm case of rationality of judgement, say a normal thinkers acceptance of the thought that 31 + 1 =32. Does he accept this because it's rational to do so? A full explanation of the rationality of this transition is a complex thing. In my view, a full explanation would mention the fact that the same principles involving '+' that we find primitively compelling are the ones which individuate the very function of addition on the natural numbers (by the recursion equations). 2 5 This philosophical explanation of the rationality is not anything the ordinary thinker knows. Nor is the philosophical explanation something that enters the content of what we tacitly know. There are no philosophical impressions or judgements made by the ordinary person for the explanation of which we need to postulate such tacit knowledge about the elucidation of the rationality of a judgement or a transition. It is certainly legitimate to say 'We make the judgement because it's rational to do so'; but I suggest that this means roughly 'We are rational; and rationality here involves finding this content primitively compelling'. This does not imply that the fact that something is a content it is rational to accept enters the causal explanation of why we accept it. If these points are correct, the connection between the rationality of a transition and the fact that someone makes it is as close in the perceptual case as it is in such a clear case of rationality as a simple arithmetic judgement. In the perceptual case, as in the arithmetical, rationality involves finding certain contents compelling (under the additional condition that perception is taken at face value in the observational case). In both cases, this is written into the possession-condition for the concept. In neither case does the philosophical explanation of the rationality of the transition or the judgement causally explain why one makes it. We can now return to issues about the particular account of perceptual entitlement I offered. The argument I have developed applies only to experiences with instance-individuated representational contents. But is there not equally an entitlement to take at

25

See 'Explaining the A Priori: T h e Programme o f Moderate Rationalism', in New Essays on the A Priori (Oxford: Oxford University Press, 2 0 0 0 ) , ed. P. Boghossian and C. Peacocke.

466

The Epistemic Significance of Perception

face value the content of experiences that represent something as a soldier, as a clock or as a computer? There certainly is such an entitlement as things actually are; but on my view that entitlement does not have the same source as the entitlement to take instance-individuated contents at face value. There can be more than one source of an entitlement to take a given component of the representational content of an experience at face value. In more detail, I would distinguish at least the following varieties of perceptual entitlement. (1) There is basic perceptual entitlement, which exists for cases in which the thinker takes an instance-individuated content of an experience at face value. (2) There is sequentially corroborative perceptual entitlement. As we noted earlier, a sequence of experiences, even a sequence with purely instance-individuated contents, can give a thinker additional entitlement to take its later members at face value, simply because the later members are as one would expect them to be if indeed the contents of the earlier members of the sequence are veridical. (3) There is informational entitlement. This is a wide category, covering many different sub-types. In the case of contents that are not instance-individuated, a thinker may have memories, knowledge from which he can make inferences, testimony and any variety of background information and informational states that make it reasonable to believe that what is in front of him is, say, a real computer, or a real clock, and not, for instance, a stage-prop. It is information of this breadth James Cornman was drawing on when he wrote "...a common explanation of why I have a visual experience of mail in my box, and, indeed, see this mail, is that a postman puts mail in my box in the morning whenever he has mail for my address, and this morning a postman had mail for my address...[...].. .because it has remained there till now when I am looking in the box, I have now a visual experience of mail in the box...". 2 6 Although informational entitlement for the veridicality of an experience can come from virtually any background knowledge, we ought to distinguish in thought between one's entitlement to think that one's environment is a certain way, and one's entitlement to think that it has come to be that way as a result of a certain history. O n e can be entitled to think, and can know, that one's environment is a certain way without having any idea of how it came to be that way. T h e principles one uses in establishing how one's environment came to be the way it actually is will involve general-purpose principles about testimony, memory and other sources of information that are not specific to one kind of conceptual content. Basic perceptual entitlement, by contrast, involves principles of entitlement that are specific to the instance-individuated content in question. I suggest that as things actually are, we are indeed entitled to take a wide range of representational contents of our experiences at face value even when those contents are not instance-individuated; but this is so only because there is informational entitlement, in the sense displayed above, to take such contents at face value. This seems to me to correspond to an intuitive distinction. It is a highly intuitive position that while you can tell just by looking that some line is curved, you need more background information

26

Skepticism, Justification and Explanation (Dordrecht: Reidel, 1980), p.255-6.

Explaining Perceptual Entitlement

467

to be entitled to judge that the thing you see in front of you is designed to carry out the functions of a computer. You need yet more information to be entitled to accept a particular explanation of how there came to be a computer at that location. This is one of the many points at which my position differs from those who offer entirely undiscriminating arguments to the effect that the best explanation of the occurrence of our experiences is that they are veridical. Bertrand Russell famously, and admittedly briefly, developed such a position in The Problems of Philosophy.27 As against his view, I do not think that there is any sound unrestricted argument, applicable to an arbitrary component of the representational content of experience, however theoretical, to the conclusion that the simplest explanation of why such experiences occur is that their contents are correct. Such a conclusion always needs some additional information when we move beyond the instance-individuated contents. This point applies equally to several other varieties of explanationist epistemologies. A rather different kind of objection claiming that the account of perceptual entitlement I have given is too restricted appeals to a different kind of case. These are cases in which it is not plausible that the way in which the experience represents something involves further commitments like the soldier or computer case. T h e cases in question are also ones in which additionally, because the property represented as instantiated is regarded purely dispositionally, it is also not possible to hold that experiences in which the property is represented as instantiated are caused, in any cases, by its instantiation. Some have held such purely dispositional views of colour-properties, but a more plausible range of such cases is given by such a property as sparkling, as when one sees the sunlight sparkling on the lake. It is very plausible that there is a defeasible entitlement to take such experiences at face value. It is quite implausible that this entitlement relies on informational entitlement of the general kind mentioned in (3) above. For the light to be sparkling does not require the holding of further commitments about function, structure or the like that require additional information before there exists an entitlement to judge, on the basis of perceptual experience, that the light is sparkling on the water. Yet it is plausible to treat the property of sparkling as purely dispositional: it is no more than the property of producing such experiences as of sparkling in properly perceiving subjects. T h e same arguably applies to various taste properties, and certain other sensory properties. If that is so, then the presence of the pure disposition cannot be a causal explanation of manifestations of the disposition. It follows that experiences of something as sparkling are not instance-individuated in the meaning given, since that requires precisely such causal explanation when the perceiver is properly connected to the world. So the question arises of whether the entitlement to take such contents at face value has been accounted for adequately. I think this entitlement does exist, and is purely perceptual; and that it can be captured consequentially under the present approach. It is a consequence of what I have been arguing that the thinker is entitled to accept, in the absence of reasons for doubt, that he is perceiving properly. But if he is perceiving, and has an experience in which some surface is represented as sparkling, then under the dispositional account of this

27

O U P : Oxford, 1 9 7 3 reprinting, at p p . 1 0 - 1 1 .

468

T h e Epistemic Significance of Perception

property it will be true that the perceived surface is sparkling. So the thinker will in these circumstances be entitled to the judgement that the presented object is sparkling. This explanation draws, in the rationalist spirit, on the truth-condition for 'That surface is sparkling'. This reasoning is available both to those who think that these dispositional properties involve some species of non-representational properties of experience, like the so-called 'sensational' properties of my Sense and Content, and to those who think that the relevant properties of experience are purely representational. 28 All that matters for the point is the dispositional nature of the property, not the nature of the properties of experience that are manifestations of that disposition. As promised I note two extensions of the argument of this section so far. T h e First Extension is to a wider range of 'as-if' states. Suppose a state or kind S of event is individuated by the relations in which its instances stand to other events and objects. Suppose also that we can draw a distinction between genuine instances of the state S, instances that stand in the required relations, and 'merely as i f states or events which are in a quite specific sense qualitatively similar to and parasitic upon to those genuine instances. T h e 'merely as if' states do not stand in those relations as things actually are. They are also parasitic in the sense that they are given as states which, although they do not in fact stand in the required relations, it is as if they do. For illustration, we can use a helpful example of Crispin Wright's. 29 Suppose, remarkably, a group of people ran around kicking a ball for 90 minutes, without any idea or intention of playing soccer, but engaging in the same bodily movements that would be involved in a game of soccer. We can say that their movements are as if they are playing soccer, even though those movements do not have the right relations to their own and to others' mental states for it to be a game of soccer. Similarly, a visual hallucination does not stand in the right relations to things in the environment to be a genuine perception; but it is, subjectively, as if it were so related. O n the way I will use the terms here, we will say that genuine soccer games and genuine perceptions are as-if states, though of course they are not merely as-if states. T h e First Extension of the argument I propose then states this: in a significant range of cases, given just the information that an as-if state qualitatively similar to an instance of S occurs, the easiest way for this to be the case is for it to be a genuine instance of S, and not a mere as-if state. For it to be a genuine instance of S is the explanation that most successfully reduces complexity. Let us take Wright's soccer example again. (We will in fact end up with a very different position on the issues from Wright's, and I will eventually be drawing different conclusions from his own examples.) For the movements of a set of 22 people to replicate those of a soccer match without their having any idea of soccer is not metaphysically impossible. It would, however, involve a massive, extraordinary series of coincidences. An explanation that proposes that there is accidental replication of movements of the same kind as would occur in a real soc-

28 29

Sense and Content (Oxford: Oxford University Press, 1983). C. Wright, '(Anti-)Sceptics Simple and Subtle: G.E.Moore and John McDowell', forthcoming in Philosophy and Phenomenobgical Research.

Explaining Perceptual Entitlement

469

cer match is a much less simple explanation than that they are playing soccer, and that their bodily movements are controlled by the intentions that are made understandable by their meaning to play the soccer. It would not be a coincidence that the agents' movements matched those of a game in the other case Wright mentions, that in which they are under the control of a movie director who wants his movie to represent a game taking place. But the hypothesis, given only that people are moving as if playing soccer, that they are under the control of a movie director, seems to me more complex, to demand more of the world, than that they are simply playing soccer. It demands not just that the agents have the notion of soccer, but that they all be influenced by some further individual. A Second Extension of the argument, which I leave as a thesis for development in other work, is that the argument of this paper is applicable to all non-conclusive a priori entitlement. The idea is that in every case in which we have a non-conclusive a priori entitlement to make a certain judgement on the basis of certain states or evidence, this is so because the easiest way for those states to come about or for that evidence to obtain is for the content judged to be true. This would, for instance, assimilate the philosophical explanation of sound inductive argument to the same model that we have applied in the non-conclusive, defeasible perceptual case.

6. Features of the Treatment (1) The approach does not say that it is a priori that hallucinations are rare, nor that an evil-demon world is impossible. Hallucinations may be frequent, and there are genuinely possible worlds in which there is a deceiving evil demon. The present position is only that the explanations of the experiences in those worlds do not reduce complexity in the way that complexity is reduced in the explanation that the experiences occur because they predominantly represent the world correctly, and occur in a perceptual system that has evolved by natural selection. (2) The argument I have offered goes far beyond the very modest position which states that if we're going to commit ourselves to anything about the relations of experience to the nonmental world, the perceptual hypothesis is best, but that it is preferable outright just to remain neutral on whether the experience stands in any such relations at all. This very modest position is offering no explanation at all of the complexity involved in the occurrence of an experience with a representational content. The conclusion of our argument is not just that //we say anything at all about the environmental relations of the experience, then the explanation that most reduces complexity is one that entails that perceptual experiences are likely to be predominantly veridical. What needs explanation is that a perceptual experience with a certain representational content occurs at all, with the complex of relations to the nonmental world this requires in the case in which its subject is properly connected to the world (and the complex of relations to other mental states whether or not the subject is so connected). Remaining neutral on the experience's relations to the environment is no explanation of these relations at all. Correct application of the Complexity-Reduction Principle takes

470

The Epistemic Significance of Perception

us all the way from the mental world to the nonmental, unconditionally—though, as always, defeasibly. If this approach is correct, it suggests that a widely-accepted dichotomy is false. The dichotomy is well-expressed by John Pollock, though he is by no means unique in accepting it. He writes: According to internalism, the justifiedness of a belief is a function exclusively of internal considerations, so internalism implies the denial of both belief and norm externalism. That is, the internalist maintains that epistemic norms must be formulated in terms of relations between beliefs or between beliefs and nondoxastic internal states (e.g., perceptual states), and he denies that these norms are subject to evaluation in terms of external considerations. 30

The argument I have offered suggests that what Pollock counts as an internal state, an experience as of something's being so, can in some cases give defeasible reason for thinking that something holds of the world outside the mind of the thinker. If this is so, norms formulated in terms of perceptual experience may help towards the fulfillment of goals which are formulated externally, in particular in terms of truth. If the dichotomy is false, the possibility opens up of a general accommodation of norms which mention internal states in Pollock's sense and an overarching aim of judging only what is true. (3) If the argument of this paper is correct, the following passage from Daniel Dennett seems prescient. It concerns what Dennett calls 'skyhooks'—procedures, capacities or information that are not conceived as resulting from earlier selection processes and testing of the sort envisaged in natural selection. Dennett writes "The renunciation of skyhooks is, I think, the deepest and most important legacy of Darwin in philosophy, and it has a huge domain of influence, extending far beyond the skirmishes of evolutionary epistemology and evolutionary ethics." 31 If the theses of the present paper are right, the Darwinian legacy is of significance even in the relatively a priori domain of theories about the normative notion of entitlement. This significance does not result from a confusing of the normative and the descriptive. Rather, the claim is that a proper philosophical explanation of certain truths about the normative—the entitlement relation—must be accounted for by the special explanatory status of Darwinian mechanisms. (4) How can the Complexity-Reduction Principle be the basis of perceptual entitlement when, for instance, it seems to be so different from the notion of validity which underlies logical transitions to which one is entitled? I offer three remarks in reply. First, when we consider other transitions to which a thinker is entitled but which are not conclusive, moving to the least complex hypothesis seems to play a role. It would be hard to deny that in the entitlement to inductive inference by enumeration, supposing that not all F's are G when all the encountered F s are G is more complex that supposing that all F's are G. If we think such cases are to be explained as tacit inference to the best explanation, avoiding complexity still plays a part in the choice of hypothesis, and of what one takes as needing explanation. Second, I will be arguing below that a

30 Contemporary Theories of Knowledge (Totowa, NJ: Rowman and Littlefield, 1986), p. 126. 31 'In Darwin's Wake, Where am I?', Proceeedings and Addresses of the American Philosophical Association 75 (2001), at p.23.

Explaining Perceptual Entitlement

471

range of nonperceptual entitlements that involve relying on psychological states are also ones in which the Complexity-Reduction Principle is implicated. If this is right, the perceptual case is not unique. Third, if we see a spectrum of cases ranging from conclusive entitlement through strong but nonconclusive, to weaker nonconclusive cases, we can see conclusive entitlement as the special case in which the entitling grounds give a sufficient condition of truth without needing to appeal to complexity-reduction in our philosophical explanations. With the nonconclusive, complexity-reduction needs to be brought into the philosophical account, but it is still serving a purpose which is uniform across the conclusive and the nonconclusive cases—contributing to the determination of which grounds are really reasons for thinking something to be true. (5) Some widely-discussed approaches in epistemology characterize normative notions in ways that seem to leave the challenge of explaining generalizations about perceptual entitlement unaddressed. A distinguished, well-known treatment is that of of Alvin Goldman, in his book Epistemology and Cognition?2. Goldman requires of a legitimate method only that it reliably produce truth in 'normal' worlds. He writes: "Imagine, the objection goes, that our actual world turns out to be an evil demon world. (Or imagine that we are actually brains in a vat being deceived by scheming scientists.) Intuitively, our beliefs would still be justified; yet the belief-forming processes being deployed are not reliable. Again the case is easy to handle. Its apparent strength rests on the assumption that the justificational status of the beliefs is determined by the reliability of their causal processes in the actual world. But this does not accord with our theory. Reliability is measured in normal worlds; and in this case, the actual world is an abnormal world!" (p. 113).

So Goldman is classifying worlds in which the thinker is a brain in a vat as abnormal. Now I would not want to say that they are normal. But there seems to be a legitimate question: 'Why should I rely on a method which yields true belief only in worlds which I haven't, on Goldman's theory, been given reason to think are actual?' A more satisfying treatment must give the thinker some reason, if only a defeasible one, for thinking that he is not the brain in a brain-in-a-vat world. Otherwise, we will be back with sceptical conclusions. The Complexity-Reduction Principle aims to meet this need.

7. Links and Applications (a) The Complexity-Reduction Principle is pertinent to wider issues about the relations between rationality and truth. It can be deployed in arguments over the issue of whether such notions as the default entitlement or default reasonableness of a method or rule can be elucidated in terms of its tendency to yield true beliefs. Hartry Field regards such an elucidation as "thoroughly implausible, on numerous grounds". 33 One of his grounds is

32 33

Cambridge, Mass.: Harvard University Press, 1986. 'Apriority as an Evaluative Notion', in New Essays on the A Priori,

at p. 124.

The Epistemic Significance of Perception

472

"The standard 'internalist' criticism: it is implausible to hold that our methods (assuming them reliable in the actual world) would be straightforwardly unreasonable in a 'demon world' (a world designed to make those methods unreliable, but undetectably so)." (Op. cit., p.125).

The treatment I have been proposing still permits a truth-based elucidation of default reasonableness for the practice of taking certain experiential contents at face value. Although this method is certainly not productive of truths in a demon-world, we argued that that world provides a more complex explanation of why there are experiences than does a world in which there is no such demon. The default-reasonableness of taking certain perceptual experiences at face value can be elucidated in terms of the tendency of doing so to produce true beliefs in the worlds which have the least complex explanation of why such an experience occurs, the ones in which it comes about in an easy way that there is such a perceptual experience. This is only the first step in accounting for default reasonableness in terms of a more elaborate relation to the production of true beliefs: but it is an essential first step. (b) O n the present approach, a thinker is entitled, in the absence of reasons for doubt, to take certain perceptual experiences that represent it as being the case that p at face value, and to judge that p. In the right circumstances, this judgment can be knowledge. Under this account, the transition is from perceptual experience to knowledge. There is no reliance on a premise to the effect that this experience, or experiences of some kind under which it falls, is perceptual. If there were such reliance, it is not clear that perceptual knowledge would ever be possible at all. If the transition to perceptual knowledge were even partly inferential, it could yield knowledge only if the premises of the inference were also known. But how is the premise that this experience (or experiences of such-and-such a kind) are perceptions to be known? It is not known a priori. If it is known a posteriori, it must rest on other cases of perceptual knowledge. But how are these other cases of perceptual knowledge to be attained, if they themselves rely on some premise to the effect that the experiences they involve are perceptual? This way lies infinite regress. Entitlement will never be attained unless some perceptual entitlement is non-inferential. Gilbert Harman has argued to the contrary, that what he treats as Gettier examples in the area of perception give us reason to say that ordinary perceptual beliefs are based on inference. He writes that "If we were to suppose that direct perceptual knowledge does not involve inference, these Gettier examples would require special treatment of an obscure sort". 34 The kind of case Harman cites is that in which someone looks, has an experience as of a candle ahead of him, and comes to believe that there is a candle ahead of him. There is a candle ahead of him, but it cannot be seen because it is behind a mirror, which is reflecting a similar candle off to the right. This thinker is justified in believing that there is a candle in front of him, and the belief is true, but it is not knowledge. Harman's diagnosis of why this is not knowledge is that the belief is reached inferentially, by a false premise about explanation. He writes of this thinker: "He infers that it looks to him as if there were a candle before him because there is a

34

Thought

(Princeton, NJ: Princeton University Press, 1973), p.174.

Explaining Perceptual Entitlement

473

candle there and because of the normal connection between the way things look and the way things are. Since that explanation is essential to his conclusion but is false, he does not come to know that there is a candle before him even though his belief is justified and true". 3 5 There is an alternative explanation of why the belief that there is a candle in front of him is not knowledge. This existentially-quantified content, that there is a candle in front of him, rests on a perceptual-demonstrative belief with the content 'That candle is in front of me', where the perceptual-demonstrative 'that candle' refers to the candle he sees—which is the one off to the right. But that candle, the one presented in his perception and demonstratively thought about, is not in front of him. So his existentially-quantified belief that there is a candle in front of him does rest on a false belief, about that candle. T h e false belief upon which it rests is not a hypothesis about the explanation of his perceptual state. Far from this philosophical account of the example being of some obscure sort, this explanation of why the belief is not knowledge appeals to the same compelling principle that Harman himself identified and uses, the 'no false lemmas' requirement to the effect that a belief which is inferred from a false premise is not knowledge. 3 6 If the mirror interposed between the subject and the candle reflected a second mirror, which in turn reflects that candle, then the subject's belief 'that candle is before me' would refer to the candle that is before him. T h e belief would be true, and the 'no false lemmas' requirement would not apply. But the defeasible entitlement to take perceptual experience at face value holds only in normal circumstances, and such an arrangement of mirrors makes the circumstances abnormal. This is not to imply that normality of the circumstances is a premise on which observational beliefs rest. O n the contrary, if it were a premise, it would need empirical support—and regress would again be threatened. Q u i t e generally, the conditions under which an entitlement exists should not be identified with premises used by the thinker in judging in accordance with that entitlement. If perceptual knowledge is not inferential, that fact bears on the correct diagnosis of what is wrong with the most famous argument of G. E. Moore's 'Proof of an External World'. 3 7 Moore wrote, "I can prove now, for instance, that two human hands exist. How? By holding up my two hands, and saying, as I make a certain gesture with the right hand, 'Here is one hand', and adding, as I make a certain gesture with the left, 'and here is another'. And if, by doing this, I have proved ipso facto the existence of external things, you will all see that I can also do it now in numbers of other ways: there is no need to multiply examples." 3 8 This 'Proof is sometimes criticized on the

35

36 37

38

Thought p. 174. Examples o f this k i n d — t h o u g h used for a different p u r p o s e — t r a c e back at least to H . P. Grice, ' T h e Causal T h e o r y o f Perception', repr. in The Philosophy of Perception ed. G . Warnock ( O x f o r d : O x f o r d University Press, 1967): see especially his example o f the pillar ahead o f the perceiver, obscured by a mirror which reflects a numerically different though similar pillar, at p. 104. Thought Chapter 3, section 6. In Moore's Philosophical Papers ( L o n d o n : George Allen and Unwin, 1959), and reprinted in G. E. Moore: Selected Writings, ed. T. Baldwin ( L o n d o n : Routledge, 1993). Page references are to Baldwin's collection. 'Proof of an External World', pp. 165-6.

474

The Epistcmic Significance of Perception

ground that Moore's own perceptual experience entitles him to judge 'This is a hand' only in the presence of the additional premise that his experience is produced (in the right way) by the external world. 3 9 Under this diagnosis, since this is what the sceptic is questioning, Moore's 'Proof fails because its conclusion is already taken for granted in one of the argument's (suppressed) premises. If what I have said is right, this cannot be the correct diagnosis. Perceptual entitlement does not rely on such additional premises. Does this mean that the present approach is committed to accepting Moore's 'Proof' as successful? Here we must distinguish between the existence of an entitlement, and having a dialectically effective reply to the sceptic. If there are not in fact any good reasons for Moore to doubt his perceptual experience, he is entitled to judge that he has two hands, and to move from this to the conclusion that material objects exist. Entitlement is preserved throughout Moore's line of thought. If, however, the sceptic is challenging whether there really is an entitlement to rely on perceptual experience, then to offer Moore's reasoning and nothing more is to beg the question. One needs at the very least to say more about why there is an entitlement to rely on perceptual experience. If the sceptic has more specific grounds for doubt, those too must be addressed. The important point is that it is entirely consistent to acknowledge that Moore's argument should not by itself rationally convince the skeptic, whilst also holding that an entitlement to perceptual judgment is not a matter of inference. For any given application we make of the earlier claims about entitlement, we can ask: is that application dependent only upon the first level of characterization, or does it additionally dependent upon the second level, or upon the third as well? If an application depends only upon a given level, that application is neutral on theses about the deeper levels. Theories that disagree on correct characterizations at the later levels may still agree on the application in question. In these initial remarks about the diagnosis of Moore's 'Proof', the application depends only on the idea that there is defeasible perceptual entitlement of a non-inferential nature. This particular application is not dependent upon any one theory of the second and third levels. Theorists who disagree about those levels may nevertheless agree on a diagnosis of Moore's 'Proof which does not construe perceptual knowledge as inferential (and does not attribute that construction to Moore either). (c) T h e generalization I formulated about the conditions under which a thinker is entitled to take for granted the representational content of his perceptual experience bears on the relationship between entitlement and factive states. I suggest that the generalization I formulated at level (2) supports the view that we can in some cases formulate the conditions under which a thinker is entitled to make a judgment in terms of his sensitivity to factive states, such as genuinely perceiving something to be thus-and-so, rather than formulating them in terms of a sensitivity to experiences whose content may or may not be correct.

39

See C . Wright '(Anti-)Sceptics', op.cit.

Explaining Perceptual Entitlement

475

Let us take an observational content Fa, and consider the conditions under which a thinker has a perceptual entitlement to accept it. We have so far been considering a defeasible rule, concerning the nonfactive state of perceptual experience (D): (D) Nonfactive-defeasible: A thinker is entitled to judge Fa if perceptual experience represents it as being so, where Fa is observational, in the absence of good reasons for doubting he is perceiving properly. How should we formulate a rule that mentions a factive state? There may be an initial temptation to say this: a thinker is entitled tout court to judge Fa, where Fa is observational, if he perceives a (so given) to be F. That is, however, an implausible conditional when we consider the case in which the thinker is perceiving a to be F, but does not realize he is, and also has good (though misleading) grounds for thinking that he may well be hallucinating. In those circumstances, he is, contrary to this formulation, not entitled to judge that Fa, even though he perceives a to be F. A better formulation of a rule involving a factive state is this Rule (O): (O) Factive-outright: When you perceive a (so given) to be F, then, if the question arises, judge that a is F. What is the relation between the factive-outright rule (O) and the defeasible rule (D)? In the domain of rules in general—not just epistemic rules—we can distinguish between an objective rule and its subjective counterpart. The objective rule for making a chicken casserole has the form: obtain chicken and vegetables, then prepare them and cook them in a certain way. The subjective counterpart of this rule is: obtain what you believe to be chicken and vegetables, then do what seems to you to be preparing them and cooking them in the specified way. Someone who aims to follow the objective rule will also be conforming to its subjective counterpart, since doing so is the best he can do by way of trying to obey the objective rule. But the only rationale he would have for following a subjectively formulated counterpart rule is precisely that it would be a way he could hope to follow the objective rule. The subjective rule has no relevant rationale independently of that fact. A rule formulated in subjective terms that was not the reflection of some objective rule would, in particular, have no connection with the goal of making only true judgements. For these reasons, the objective rule is explanatorily more fundamental than its subjective counterpart. I suggest the same is true of the relation between the objective factive-outright rule (O) and the nonfactive-defeasible rule (D). The latter is the subjective counterpart of the former, and has a rationale only insofar as judging in accordance with it will respect the objective rule (O). The complexity of (D) is precisely what one would expect if the thinker were aiming to make his judgments sensitive to his genuinely perceptual states. It is not as if one would find intelligible a statement of perceptual entitlement that had defeating clauses relating not only to reasons for thinking that one is not perceiving properly, but rather to some other arbitrary condition not having to do with one's perceptual mechanisms. The practice of taking one's perceptual states at face value is

476

The Epistemic Significance of Perception

the practice of taking it that they are delivering factual information about the world. Anything which makes it rational not to take them at face value must be something which undermines the proposition that one's senses are delivering factual information about the world. This receives partial confirmation from the fact that the qualification in (D), that the thinker is entitled only if he has no reason for doubting he is perceiving properly, could not be replaced by something more general, to the effect that there is no reason for doubting that Fa. If that were a correct formulation of a principle about entitlement, one could never be entitled to set to rest one's doubts about whether it is the case that Fa by coming to perceive a to be F. One is frequently entitled to do just that. Doing so seems to be a paradigm of rationality. The fact that the qualification in the correct formulation of (D) concerns reasons for doubting that one's experiences are genuinely perceptual further highlights the fact that this defeasible condition is simply aiming to make judgments to which it counts one as entitled sensitive to whether one's experiences are genuine perceptions of the way the world is. There is an argument that, under the level 2 generalization I formulated, the nonfactive-defeasible rule (D) and the factive-outright rule (O) are in a certain sense equivalent. That sense is that: a thinker is entitled to judge the observational content Fa in exactly the same circumstances whether he is following the defeasible rule (D) or the factive/outright rule (O). We consider two cases, according as (1) the thinker does not, or (2) does, have reasons for doubting his perceptual states. In case (1), where the thinker has no reasons for such doubts, consider the case in which he judges that Fa by rule (D). Then, by our earlier arguments, he is entitled to assume he is in the circumstances with respect to which the observational content of his experience is individuated, and these, I argued earlier, are circumstances in which his experience is a genuine perception. So in the absence of reasons for doubt, our thinker is entitled to treat his experience as perceptual. Hence he will equally be entitled to judge that Fa if he is following rule (O). Conversely, and trivially, if our thinker is entitled to judge Fa when following rule (O), he must be entitled to take it that he is perceiving, which he would not be if there were good reasons for doubt. So he will equally be entitled to judge Fa if following rule (D). If there were any reasons for doubting that he is perceiving, then he would not be entitled to judge Fa under rule (O) either. In case (2), where the thinker has reasons for doubting that he is perceiving properly, he will not be entitled to judge Fa under either rule. The argument could be refined, without essential alteration, to treat the case in which the thinker has reasons for doubting just certain of his perceptual states. We would just consider cases according as states with the perceptual content Fa are in the doubtful category. The claim of equivalence for (D) and (O) will have analogues for other informational states in cases in which there are contents which stand to those states as observational content stands to perceptual states. It is a plausible principle that what a thinker is entitled to judge, and what is justified and what is rational, depends only what seems to the thinker to be the case, and not on which factive states he stands in. The truth of this principle (if it is true) does not

Explaining Perceptual Entitlement

477

imply that principles of entitlement, justification and rationality cannot mention factive states. 40 It does not follow, because the thinker who responds to seeming-, non-factive states, may be doing so because he thinks or takes for granted that they are perceptual. The rules he is trying to follow may still mention factive states. If the claim of equivalence of (D) and (O) is correct, then at least some entitlement rules mentioning only non-factive states are equivalent to rules mentioning factive states. If the equivalence holds, it must be false that there are entitlements which are captured by rules mentioning non-factive states, but not by rules mentioning factive states. (d) The above explanation of perceptual entitlement bears upon the theses of my book Being Known. There I argued that certain concepts can be individuated by the conditions under which certain contents containing them are not merely rationally judged, but are known. This was part of the 'Linking Thesis' of Chapter 2 of that work. The idea was, for instance, that the concept a Babylonian expressed by 'Hesperus' is distinct from the concept he expressed by 'Phosphorus' because there are certain circumstances in which he can come to know that Hesperus is F without thereby being in a position to know that Phosphorus is F. Now if a sceptic questions whether a thinker is entitled to take perceptual experience at face value, his scepticism will extend to this Linking Thesis too. If perceptual knowledge is not possible, it follows that it is not possible to know by perception that Hesperus has some property without knowing that Phosphorus has it. Much of the discussion of Being Known would then collapse. One would be left only with sceptical responses to the challenge of integrating metaphysics and epistemology, rather than the ones attempted in that book which aimed to show that we really do know much of what we think we know, and without weakening our conception of truth for the propositions in question. Being Known thus presupposed that some answer to scepticism exists, without actually supplying that answer. I take the theses about perceptual entitlement at the third level of explanation to be the start of such an answer. They can be seen as a contribution to the task of explaining why, in the perceptual cases, the Linking Thesis is true. In a discussion essay written after Being Known, I spoke of a 'Second Linking Thesis', linking instance-individuation with entitlement and, thereby, with knowledge. 41 The theses at the second and third levels in the present work certainly say more than the Linking Thesis of Being Known. They are, however, contributions to the tasks of explaining the conditions under which the Linking Thesis is true, and of explaining why it is true, rather than being autonomous, additional theses. The claims of the present paper also bear on the somewhat creaky discussion of the 'rationally nondiscretionary' in Being Known. The rationality of judging an observational content on the basis of perceptual experience requires the rationality of two things: the rationality of moving from the content of the perceptual experience to the content of the observational judgment (in theories under which these are distinct contents); and the rationality of taking perceptual experience at face value in the first place. What I

40 Contrast R. Wedgwood, 'Internalism Explained', Philosophy and Phenomenobgical Research 65 (2002) 349-369. 41 'The Past, Necessity, Externalism and Entitlement', Contribution to a Symposium on Being Known, Philosophical Books Μ (2001) 106-17.

478

The Epistemic Significance of Perception

have been offering is an explanation of why this second thing is rational; the approach

of Being Known is incomplete without it. (e) The existence of defeasible but non-inferential entitlement structures goes far beyond such cases as perception and the various forms of memory, and possibly testimony. 42 The existence of defeasible, non-inferential entitlement relations can also provide more room for maneuver in the philosophical account of some areas of moral thought. One example is provided by the discussion in an important recent paper by Allan Gibbard, 'Normative and Recognitional Concepts'. 43 In the part of his paper concerned with 'Thick Recognition', Gibbard observes that one's understanding of a situation may be 'heavy with demands for action (p. 163). One may, for instance, perceive an unjust act as demanding rectification. Such cases pose a problem for views that sharply separate how things are and what to do. Gibbard observes, acutely, that it would be completely unacceptable to think that the thing to do is to act on every impression that the situation demands a certain action. We are all subject to prejudices, and there can be illusions of demands. He says that "the principle we'd need to accept" in order to take the apparent demands as face value is "appalling" if it means we should act on any impression of any demand. His conclusion is that we should just take it as part of our situation that we have this sense, and this is "a psychological aspect, not plan-laden in itself" (p. 164). These cannot be the only two possibilities, if defeasible but non-inferential entitlement structures exist. Taking the seeming-demands of a situation at face value in deciding what to do may be something to which one is prima face entitled, an entitlement which can be defeated if the seeming-demands are promoting a course of action one has reason to think is morally wrong. The "appalling" principle Gibbard cites is the analogue of the epistemically wholly unacceptable principle that one should always take perceptual experience at face value—even an experience one knows to be of a perfect trompe l'oeil, or knows to have an inconsistent content. Putting the fact that there are certain seeming-demands into the specification of one's situation is the analogue, for the practical case, of the treatment of perception that says one has a premise to the effect that one is having an experience of a certain kind. That approach has made it impossible to see how perceptual knowledge could be attained. The believer in the importance of thick concepts can and should insist that the apprehension of demands for action should not be assimilated to a model that has proved unworkable in the perceptual case. The defender of thick concepts and their significance should invoke the structure of a defeasible, non-inferential entitlement relation. I am not necessarily endorsing this position: my point is just that defeasible, non-inferential entitlement relations makes available this position in logical space.

42 43

On testimony, see Bürge, 'Content Preservation'. Philosophy and Phenomenological Research 64 (2002) 151-67.

Explaining Perceptual Entitlement

479

8. Further Extensions Some of the features o f this account o f perceptual entitlement generalize beyond perception to cases in which the direction o f the relation o f causation between world and mind is the opposite o f that in perception. It is not crucial to the general form o f the account of entitlement I outlined in the perceptual case that the entitling mental state be caused by the conditions to which one is entitled. T h e general structure o f the account can still get a grip provided that the mental state is individuated by certain o f its relations to the conditions mentioned in a statement o f the entitlement. In my view, this applies in the case o f action—as one might well expect from the many symmetries, now widely recognized, between perception and action. For basic bodily action-types φ, the mental event-kind o f trying to φ is individuated by the fact that events o f that kind tend to produce φ-ings, when the subject's central control system is properly connected to his body. Now thinkers normally know what they are doing. In fact, they have a distinctive phenomenology of action. It can seem to the subject that he is φ-ing, and this apparent awareness can be present even in the subject who is acting with an anaesthetized or damaged limb from which there is no proprioceptive feedback (nor any illusion o f such perceptual states). In such cases, the distinctive apparent awareness a subject has o f his own actions seems to be a result o f his tryings. T h e fact that he tries to φ causally explains his impression that he is φ-ing. Some striking experiments by Tony Marcel suggest that apparent awareness of one's φ-ing can be produced by one's trying to φ , when one is not in fact φ-ing, even if the limb employed is neither anaesthetized nor damaged. 44 T h e experimenter induces in the subject an illusion about the location o f his hand. T h e subject is then asked to move his hand to a new location. This new location is chosen in such a way that for his hand to move to it from its actual present location, it has to move in one direction (clockwise, say); while it is in an opposite direction (anticlockwise) that it would have to move if his hand were at its apparent initial location. Subjects succeed in moving their hand to the new location, but they have the impression that they have moved it in the opposite (anticlockwise) direction, a direction in which o f course they have not so moved it. T h e content o f their trying (or some event causally related to it) seems to cause the content of their impression of action, even though the actual motor instruction issued requires, and produces, movement in the opposite direction. The question then arises: how can this distinctive awareness yield knowledge, on the part of the subject, that he is acting a certain way? A reliabilist would say that in circumstances in which the agent does know, trying to φ is reliably correlated with φ-ing. But there are strong objections in other cases to pure reliabilism. Is there some explanation o f how we have knowledge o f what we are doing which is not dependent upon perception o f ourselves, or upon proprioceptive feedback, but which does not

44

See the description of his vibro-tactile experiments in his paper 'The Sense of Agency—Ownership and Awareness of Action' in Agency and Self-Awareness, ed. J. Roessler and N. Eilan (Oxford: Oxford University Press, 2003). For further philosophical discussion of the issues, see also my paper 'Action: Awareness, Ownership and Knowledge' in the same volume.

480

The Epistemic Significance of Perception

involve a reversion to reliabilism? I suggest that there is, and that it relies on a generalized version of the principle on which we have relied in the account of perceptual entitlement. The event-type of trying to φ is individuated by its relation to cp-ings in the case in which the agent's control center is properly connected with his body. The thinker is entitled to take it that he is in the circumstances with respect to which these event-types are individuated. (Again, an explanation in terms of complexity-reduction could be given of why he is so entitled; though the present argument for knowledge of action requires only that such an entitlement exists.) But in these circumstances, tryings to φ do produce φ-ings. The awareness that is produced by the trying to φ can then, with entitlement, be taken at face value. In suitable circumstances, this can yield knowledge on the subject's part that he is φ-ing. If we reject this approach, it would be a real task to explain philosophically how we have knowledge of actions we are performing, without reverting to reliabilism. One cannot simply apply the perceptual model straight, since as we saw, it is not the bodily action itself that causes the apparent awareness of action. It is tempting to apply the same generalization to other cases too, outside the realm of perception and action. Consider, for instance, the entitlement to self-ascribe beliefs on the basis of one's own judgments. Judgment is individuated as an event of a kind which, when all is working properly, leads to belief. So one can explain how one can know what beliefs one has by making self-ascriptions that are sensitive to one's one judgments—even though this is certainly a fallible method. This explanation does not involve a reversion to pure reliabilism. T h e method is a rational one. Again too, one could develop an appeal to complexity-reduction to explain why the entitlement exists, given the complex relations an event must stand in if it is to be a judgment with a given intentional content. At this point, there are many tasks for further work. Here are some of the questions which arise. Can every case of entitlement by a state with intentional representational content be assimilated to the present model, or to some natural extension thereof? If so, how? If not, why not? And if not, what is the correct explanation of entitlement for cases which no extension of the present model can capture? There is evidently massive further work to be done. But I do conjecture that entitlements that are rather different from that of the perceptual case, or any extension thereof, are possible only because perceptual entitlement is also possible. The cases in which the Complexity-Reduction Principle applies are fundamental. 4 5

45

I thank Tyler Bürge, Stephen Schiffer, Scott Sturgeon, Roger White, and the members of David Chalmers' and David Hoy's 2 0 0 2 N E H Institute in Santa Cruz, and of my N Y U Seminar on Reasons, 2002, for valuable comments and advice.

Externalism and Disjunctivism JAMES V A N CLEVE

In what follows I explore the increasingly prominent conception of perceptual experience known as disjunctivism relating it to the divides in epistemology and philosophy of mind between internalism and externalism. I shall focus on McDowell's version of disjunctivism in particular,1 and I shall argue that although his case seems initially to involve an implausibly strong version of epistemological internalism, the thrust of McDowell's internalism is considerably modified by its being combined with a form of content externalism. I shall also highlight one additional and noteworthy way in which perceptual experience involves an external element for McDowell.

I

McDowell advances his disjunctivism as a response to the argument from illusion, one version of which runs as follows:2 1. For any veridical experience, there is a possible illusory or hallucinatory experience phenomenologically indistinguishable from it. 2. To account for illusions and hallucinations, we must introduce special mental items (sense data or whatnot) to be the objects of these experiences. (The drunkard is seeing something, so if there are no real rats before his eyes, there must at least be rat-like sense data displayed on his mental screen.) 3. When experiences are phenomenologically indistinguishable, they must have the same ontological assay. 4. Therefore, even veridical experience must have sense data as its proper objects. Our experience never reaches the external world. When the argument is put that way, McDowell would block its conclusion by rejecting its third premise. He opposes the idea that veridical experience and hallucinatory experience must have any shared ingredient or "highest common factor" as he calls it. What is common to veridical and nonveridical experience can only be captured in a disjunction: one is either having an experience in which the fact that ρ is made mani-

1

2

John McDowell, "Criteria, Defeasibility, and Knowledge," in Perceptual Knowledge, edited by Jonathan Dancy (Oxford: Oxford University Press, 1 9 8 8 ) , pp. 2 0 9 - 1 9 . McDowell cites as his forerunners J . M . Hinton, Experiences (Oxford, 1 9 7 3 ) , and Paul Snowdon, "Perception, Vision, and Causation," Proceedings of the Aristotelian Society, 81 ( 1 9 8 0 - 8 1 ) , 1 7 5 - 9 2 (reprinted on pp. 1 9 2 - 2 0 8 of Dancy). I shall not discuss here any modifications to McDowell's disjunctivism that may be needed to adapt it to illusions as opposed to hallucinations.

482

The Epistemic Significance of Perception

fest to one (let us symbolize this as Mp) or an experience in which it merely appears to one that ρ (let us symbolize this as Ap). There is nothing common to the two but the disjunction Mp ν Ap—which is a way of saying there is nothing in common at all. At any rate, there is no Xsuch that we could analyze Mp as Xp & p and Ap as Xp & -p. What is wrong with the common factor model? In some places, McDowell suggests that the common factor model is bad because it leads to veil-of-ideas skepticism·, it posits "mere appearances," things that interpose themselves between the subject and the fact, or the like. 3 To this I would reply: not at all! The common factor could be a sensory state rather than a sensory object (as in adverbial as opposed to sense-datum theories), and such states are arguably vehicles of perceptual knowledge (or perhaps simply accompaniments of it) rather than veils hanging before it. Indeed, those who advocate adverbial theories of sensory experience often do so in order to block the argument from illusion at its second premise—one step earlier than McDowell. They say that the victim of delirium is not sensing any objects, but is just sensing in a certain way, and it is this way that is common to veridical and nonveridical perception. McDowell seems quite oblivious to the distinction between thing and state versions of what the highest common factor might be. This indifference on his part turns out to be inconsequential, however, for his deepest objection to common factor views is an objection that applies to thing and state versions alike. To this deeper objection I now turn. 4 In the common factor model, knowledge of ρ is supposed to be yielded by a nonconclusive (or as McDowell more often says, "defeasible") basis for believing p: the basis for believing p is an experience that falls short of entailing p, 5 The basis on which one knows there is a tree before one will be an experiential state that might equally well occur in the absence of any tree. But a genuine basis for knowledge, McDowell believes, must be conclusive. If the relation between grounds or criteria and what they are criteria for is not conclusive, w e have n o real e x p l a n a t i o n o f h o w the relation c o u l d be knowledge-sustaining: f o r even in the m o s t favorable cases, it remains possible, f o r all o n e k n o w s , [that the g r o u n d s constituting one's basis] are b o t h true a n d k n o w n to be true, w h i l e the c o n c l u s i o n s w h i c h rest o n t h e m are false a n d hence n o t k n o w n at all. 6

Notice how far out of the mainstream McDowell is on this point. The vast literature devoted to solving the Gettier problem all takes for granted that knowledge that ρ can be based on grounds that do not entail ρ—otherwise, the problem would never arise. So

3

4 5

6

The language of mere appearances, things that interpose themselves, things that intervene, and the veil of ideas is all to be found on p. 2 1 1 of "Criteria, Defeasibility, and Knowledge." (My page references are to the Dancy volume.) See especially the first and the tenth paragraphs of "Criteria" as reprinted in Dancy. I believe we should distinguish, as McDowell does not, the statuses of being defeasible and being nonconclusive. If Pollock's category of the "incorrigibly justified" is nonempty, there are nonentailing reasons for believing things that, despite not being logically conclusive, are nonetheless immune from defeat. (See John L. Pollock, Knowledge and Justification (1974), ch. 2.) What McDowell is really against is the idea of knowledge based on nonconclusive grounds, whether these be defeasible or not. "Criteria," p. 209.

Externalism and Disjunctivism

483

before going a n y further, it m a y be w o r t h w h i l e to explore a possible a r g u m e n t in f a v o r o f M c D o w e l l ' s u n c o m p r o m i s i n g p o s i t i o n o n t h e r e l a t i o n o f k n o w l e d g e t o its g r o u n d s .

II It is t e m p t i n g to t h i n k t h a t if k n o w l e d g e can b e based o n i n c o n c l u s i v e g r o u n d s , t h e r e can be cases o f k n o w l e d g e a n d cases o f n o n k n o w l e d g e t h a t d i f f e r o n l y in t h e t r u t h v a l u e o f w h a t is believed. W i t n e s s N o r m a n M a l c o l m : As philosophers we may be surprised to observe that it can be that the knowledge that ρ is true should differ from the belief [based on qualitatively identical grounds] that ρ is true only in the respect that in one case ρ is true and in the other false. But that is the fact. 7 J o h n T i e n s o n agrees w i t h M a l c o l m t h a t the possibility o f k n o w l e d g e based o n i n c o n clusive g r o u n d s i m p l i e s t h e possibility o f cases d i f f e r i n g o n l y in t h e w a y M a l c o l m notes. U n l i k e M a l c o l m , h o w e v e r , T i e n s o n t h i n k s this i m p l i c a t i o n c o n s t i t u t e s a n objection

to

c o n c e p t i o n s o f k n o w l e d g e ( s o m e t i m e s called "weak c o n c e p t i o n s o f k n o w l e d g e " ) t h a t a l l o w f o r k n o w l e d g e based o n i n c o n c l u s i v e g r o u n d s : 8 The difference between knowing and not knowing that ρ cannot be just the truth or falsity of/>. . . .[If we allow that the truth requirement of knowledge is independent of the evidence requirement], we are forced to allow, counter-intuitively, that a person can have the same evidence for two propositions, but know only one of them. 9 W h y is it o b j e c t i o n a b l e t o say t h a t a p e r s o n k n o w s in o n e case b u t n o t in a n o t h e r , d e spite h a v i n g e x a c t l y t h e s a m e e v i d e n c e f o r his belief? T i e n s o n seems t o t h i n k t h a t this result is c o u n t e r - i n t u i t i v e in its o w n right. B u t h e also n o t e s w h a t h e takes to be a f u r t h e r u n d e s i r e a b l e c o n s e q u e n c e o f it: "It will n o t be p o s s i b l e f o r a p e r s o n t o tell w h a t p a r t o f his a p p a r e n t k n o w l e d g e is k n o w l e d g e . . . . [ H e o r s h e w i l l n o t be] in a p o s i t i o n to d e t e r m i n e w h a t h e k n o w s . " 1 0 I f p is true, the p e r s o n k n o w s ; i f n o t , h e does n o t ; a n d h o w is the p e r s o n t o t e l l ? 1 1

7 Norman Malcolm, "Knowledge and Belief," in Knowledge and Certainty (Prentice-Hall, 1963), pp. 58-72; reprinted in Knowing, edited by Michael D. Roth and Leon Galis (New York: Random House, 1970), pp. 17-32, at p. 19. In Malcolm's example, someone says "I know there will be water in Cascadilla Gorge, because I saw a lot of water flowing there this morning." If we go to the gorge and find water, we say that the person knew there would be water; if we find it dry, we say he did not; and according to Malcolm, the truth or falsity of what the person believed is the only relevant difference between the two cases. 8 For discussion of "strong" and "weak" conceptions of knowledge, see Laurence Bonjour, Epistemobgy: Classical Problems and Contemporary Responses (Lanham, MD: Rowman & Littlefield, 2002), pp. 41-43. The terms derive originally from the Malcolm article cited in the previous note. 9 John Tienson, "On Analysing Knowledge," Philosophical Studies, 25 (1974), 289-93. InTiensons example, a detective has qualitatively the same evidence for thinking White innocent that he has for thinking Black innocent, and in fact White is innocent and Black is not. Tienson thinks it intolerable to say that the detective knows White is innocent, but does not know Black is innocent. 10 Tienson, pp. 290 and 291. 11 I find at least the hint of a similar argument in McDowell: "Consider a situation in which someone supposedly has such knowledge [knowledge based on a nonentailing ground or criterion], as compared with a situation in which someone [has the same experiential ground] but, since the state of affairs for

484

T h e Epistemic Significance of Perception

Are Malcolm a n d T i e n s o n right in maintaining that the weak conception of knowledge permits knowledge to differ from nonknowledge only in the truth value of what is believed? I think not. It is a lesson of the Gettier problem that any proponent of a weak conception of knowledge must add a fourth condition of knowledge alongside justification, truth, and belief. Symbolizing the fourth condition, whatever it may be, as 4p, we have Kp iff Jp & Tp & Bp & 4p, or Κ = J T B 4 . In Plantingas work on knowledge, 'warrant' is used as a name for whatever has to be added to true belief to get knowledge. 1 2 If Jp does not entail Tp, what you have to add is Jp & 4p\ in other words, warrant equals J plus 4. In a recent article, Trenton Merricks has convincingly argued that warrant in this sense of the term must entail truth. If it does not, a strategy similar to the one Gettier used to refute the J T B analysis of knowledge can be used all over again to refute J T B 4 analyses. 1 3 T h e upshot for present purposes is the following: it is not true, pace Malcolm and Tienson, that a weak conception of knowledge must allow that a case of knowledge and a case of nonknowledge can differ only in the truth of what is believed. Suppose subject S knows ρ in one case but not in another, owing to the falsity of p in the second case. It follows from what Merricks has shown that the subjects belief, though warranted in the first case, is not warranted in the second: S may have equal justification in the two cases, but cannot have equal warrant. So the two cases do not differ merely in the truth value of what S believes; if alike in justification, they must differ in regard to the fourth condition. T h i s removes one o f the consequences Tienson found counter-intuitive in weak conceptions of knowledge, but what about the other consequence? It was this: a person would not be able to determine what he knows. What requirement is Tienson invoking when he says this is a bad consequence? Is it the K K thesis—that anyone who knows ρ should also know that he knows p, or at least be in a position to know that he knows? That requirement is satisfiable even by weak conceptions o f knowledge, just so long as we let the requirements for second-order knowledge be of the same weak sort. O n e could know that he knows ρ on the strength of inconclusive reasons for believing that he meets the conditions of first-order knowledge. So perhaps Tienson is implicitly invoking the stronger requirement laid down in Prichard's famous dictum: that a person should always be able to know whether he knows. 1 4 This goes further than the K K thewhich it is a criterion does not obtain . . . he lacks the knowledge in question: can the blankly external obtaining o f the fact, in the first case, make it intelligible that the subject knows, though he is supposedly indistinguishable, in the reach of his experience, f r o m the subject who does not know?" ("Criteria," p. 2 1 3 ) . In the end, however, I a m not sure that M c D o w e l l s concerns are the s a m e as Tiensons. 12

Alvin Plantinga, Warrant: The Current Debate (Oxford: O x f o r d University Press, 1993), p. 3 and throughout.

13

Trenton Merricks, "Warrant Entails T r u t h , " Philosophy and Phenomenological Research, 55 ( 1 9 9 5 ) , 84155. Besides giving a general argument for his thesis, Merricks notes that the leading candidates on offer to serve as the fourth condition o f knowledge (for example, that one's justification be undefeated) do in fact, when conjoined with justification, entail truth. I a m indebted to J u a n C o m e s a n a for pointing out to me the relevance o f Merricks's work to McDowell; he has discussed the matter further in his P h . D . dissertation, The Ways of Reason (Brown University, 2 0 0 3 ) .

14

"We must recognize that when we know something we either do, or by reflecting, can know that our condition is one o f knowing that thing, while when we believe something, we either do or can know that our condition is one o f believing and not knowing: so that we cannot mistake belief for knowledge or vice versa." H . A . Prichard, Knowledge and Perception (Oxford: Clarendon Press, 1950), p. 88.

Externalism and Disjunctivism

485

sis, implying not only that a subject who knows can know that he knows, but also that a subject who does not know can know that he does not know. Prichard's requirement is arguably not satisfiable by any weak conception of knowledge, for there is no reason to think that someone who believes something falsely but on good evidence will be in a position to know that he does not know. But it is also a requirement whose intuitive credentials are by no means as firm as those of the simple KK thesis.

Ill The arguments we have considered for the thesis that knowledge must always be based on conclusive grounds are themselves inconclusive. Nonetheless, let us grant McDowell that thesis so we may continue our exploration of his disjunctivism. McDowell believes that the grounds on which one knows any truth must always entail that truth, and he thinks that the disjunctive account of perceptual experience enables subjects to live up to this requirement in cases of perceptual knowledge. In veridical perceptual experience, the object of one's experience is not any highest common factor that falls short of />, but the very fact that p. In the experience we confront the fact or have it made manifest to us. And now the experience that is our basis for knowing ρ does not fall short of p—it entails p. Readers of McDowell may suspect that he has squashed down a bump in the carpet at one place only to have it reappear at another. To avoid having knowledge hinge on objectionably "external" facts, he insists that knowledge of ρ be based on a state Mp that entails p. But now whether Mp (rather than Ap) obtains turns on whether p obtains, which may seem to be equally external. So what have we gained? Perhaps in anticipation of misgivings along such lines, McDowell articulates an objection to his view. The state Mp that entails ρ and gives us knowledge of it is, he admits, indistinguishable from Ap. So how could we know which of the states we are in? And if we could not know this, how could the difference between them make a difference to what else we could know? As McDowell puts it, 'Ex hypothesi a mere appearance can be indistinguishable f r o m what you describe as a fact made manifest. So in a given case one cannot tell for certain whether what confronts one is one or the other o f those. How, then, can there be a difference in what is given to experience, in any sense that could matter to epistemology?' 1 5

McDowell prefaces his reply to this objection by saying, "There is something gripping about the 'internalism' that is expressed here." 16 Since people are sometimes in the grip of false views, we do not know from that remark alone whether McDowell accepts whatever form of internalism is in question. But I believe the ensuing discussion makes clear that McDowell does accept the "gripping" internalism. 17 He responds to

15 16 17

"Criteria," p. 2 1 4 . "Criteria," p. 2 1 5 . At least he does not reject it. H e says he is "detaching" it from its s u p p o s e d consequences (p. 2 1 5 ) and "disarming" it (p. 2 1 6 , n.l 1).

486

T h e Epistemic Significance o f Perception

the objection not by rejecting the internalism, but by arguing that his view does not run afoul of it. McDowell presents (and as I believe, endorses) the following argument for the "gripping" variety of internalism: The root idea [of the gripping internalism] is that one's epistemic standing on some question cannot intelligibly be constituted, even in part, by matters blankly external to how it is with one subjectively. For how could such matters be other than beyond one's ken? And how could matters beyond one's ken make any difference to one's epistemic standing? 18

Rewriting the rhetorical questions as assertions, we obtain the following argument, which I shall refer to as the internalist syllogism: 1. Blankly external matters are matters beyond one's ken. 2. Matters beyond one's ken cannot make any difference to one's epistemic standing. Therefore, 3. Blankly external matters cannot make any difference to one's epistemic standing. To evaluate this syllogism, we need to explicate its three terms: 'blankly external', 'beyond one's ken, and 'makes a difference to one's epistemic standing'. I propose the following: ρ is beyond S's ken. The obvious implication here is that one does not know ρ or, more strongly, is not in a position to know p, which I shall symbolize as - P K p , suppressing the reference to S. If ρ is a matter genuinely beyond one's ken, perhaps we should add that one is not in a position to know - ρ either, but I shall ignore that complication.

q makes a difference to S's epistemic standing with respect to p. One possibility is that S's knowing p depends on q's being the case, which we may symbolize as Kp —» q. Another is that S's knowing p hinges on whether q, which we may symbolize as Kp q. The weaker symbolization will be adequate for our purposes below. In either case, we are taking "epistemic standing" with respect to p to be a matter of whether you know ρ or not. We could get further possibilities by taking epistemic standing to be a matter of whether you are justified to a certain degree or not, but I shall ignore those complications. q is blankly external to S. McDowell's full phrase is "blankly external to how it is with one subjectively." Here again there are two possibilities. We could say (i) that whether q obtains makes no difference to what S's total subjective state is like—a change in the truth value of q would entail no change in S's subjective state. In other words, S's total subjective state entails neither q nor -q. The other alternative is to say (ii) that q is not knowable on the basis of S's subjective states alone. If we operate with McDowell's assumption that knowledge of q requires a basis for believing q that entails it, (i) entails (ii). It is arguable that (ii) does not entail (i), but McDowell may believe otherwise, as we shall see below. In the meantime, I shall operate with the prima facie stronger (i) as my explication of blank externality. 19 18

"Criteria," p. 2 1 5 .

19

Since 'blankly external' occurs unnegated in the antecedent o f a premise and negated in the consequent o f the conclusion, choosing the stronger reading of it yields weaker—thus more plausible—versions o f premise and conclusion.

Externalism and Disjunctivism

487

Using the symbolism I have suggested, the overall form of the internalist syllogism may be represented thus: 1. If q is blankly external to S, then -PKq. 2. If Kp —> q and Kp, then PKq. Therefore, 3. If Kp —» q and Kp, then q is not blankly external to S. To arrive at this form of the argument, I have contraposed lines 2 and 3 in the English version above before rendering them into symbols. I have also added the second occurrence of lKp in the antecedents of 2 and 3 as something that clearly seems required if 2 and 3 are to have any chance of being true. It is only provided I do know ρ that I must be in a position to know those things on which my knowing p depends. 20 If we now explicate blank externality in the first of the ways I suggested, we obtain the following as our internalist syllogism: 1. If S's total subjective state entails neither q nor -q, then -PKq. 2. If Kp —^ q and Kp, then PKq. Therefore, 3. If Kp —» q and Kp, then S's total subjective state entails q. What immediately follows from 1 and 2 is a weaker version of 3 in which the consequent is 'S's total subjective state either entails q or entails -q . But given the antecedent of 3, S's total subjective state cannot entail -q, so the argument as I have given it is valid. The argument at which we have now arrived appears at first sight to impose unabashedly Cartesian restrictions on knowledge. One premise implies that any knowledge-making or knowledge-enabling factors must themselves be known (or at least knowable). 21 The other premise says that nothing can be known save what is entailed by one's own subjective states. 22 Together, these requirements may seem to add up to a strict version of epistemological internaJism indeed. We shall see, however, that these initial appearances are misleading.

IV Let us now return to the objection McDowell formulates against his own view. Recall that his view is that knowledge of facts in one's environment cannot be based on any factor common to veridical and hallucinatory experience, since such a factor would fall short of entailing what it is supposed to yield knowledge of. It must instead be based on experiences in which the very fact to be known is made manifest to one—experiences of the sort we are symbolizing as Mp. The objection McDowell anticipates is that

20

21

22

T h e addition o f ' K p ' does not automatically fall out o f the symbolizations I propose above, but it w o u l d be authorized by any o f several more complicated symbolizations o f ' p is beyond one's ken', e.g., '~[(p - > Kp) & ( - p K-p)]'. T h e second premise implies a qualified version of the K K thesis. If we substitute 'Kp' for both occurrences of 'q' in 2 and let a tautology drop out of the antecedent, we obtain the thesis that if Kp, then P K K p — y o u never know w i t h o u t being in a position to k n o w that you know Perhaps McDowell w o u l d only impose this requirement for basic knowledge, allowing that inferential knowledge may be based on nondeductive modes o f inference.

488

T h e Epistemic Significance o f Perception

such facts, since they are indistinguishable from the "mere appearances" we are symbolizing as Ap, would be blankly external to us and thus incapable of contributing to our knowledge of anything else. We may put the objection thus: 1. If on some occasion I know ρ in virtue of p's being made manifest to me (thus: Kp and Kp —» Mp), then Mp is not blankly external to me. 2. But Mp is blankly external to me. 3. Therefore, I never know ρ in virtue of its being made manifest to me. As the parenthetical symbols make clear, the first premise in this objection is an instance of the conclusion of the internalist syllogism—an argument that McDowell does not challenge. So McDowell needs to deny the second premise in the objection. And that is exactly what he does: When someone has a fact made manifest to him, the obtaining o f the fact contributes to his epistemic standing on the question. But the obtaining o f the fact is precisely not blankly external to his subjectivity, as it would be if the truth about that [i.e., his subjectivity] were exhausted by the highest c o m m o n factor. 2 3

What he claims to be "precisely not blankly external" in that passage is actually not Mp, but p. But I believe that p and Mp must have the same status as far as blank externality or its opposite goes, so the difference does not matter. Is McDowell right to claim that ρ and Mp are not blankly external? Given the explication of 'blankly external' we settled on above, the question comes to this: are p and Mp (where ρ is an environmental fact and Mp its being made manifest to the subject) entailed by one's total subjective state? Since p is entailed by Mp, the answer will be yes if Mp itself counts as a subjective state. So our next question must be: what is a subjective state? There are two natural answers to this question that will not serve McDowell's purposes. O n e answer is that subjective states are states a subject can know he is in simply through introspection. Mp is not plausibly regarded as a subjective state in this sense, since it entails an environmental fact, which puts it outside the traditional ambit of introspection. 2 4 In any case, I believe it is foreign to McDowell's intent that subjective states be characterized in terms of introspective accessibility. Another natural answer is that subjective states supervene on phenomenology: if two persons (or the same person on different occasions) are in states that are experientially indistinguishable, then they are in the same subjective state. But this, too, is clearly foreign to McDowell's intent. Mp entails p and Ap does not, so they cannot be the same subjective state, despite being experientially indistinguishable; nor can they involve a c o m m o n subjective state, or we would have a c o m m o n factor. McDowell cannot have a phenomenological criterion for subjective states.

23 24

"Criteria," pp. 2 1 5 - 1 6 . For an argument to the contrary, however, see Ted A. Warfield, "A Priori Knowledge of the World: Knowing the World by Knowing O u r M i n d s , " in Philosophical Studies, 1998; reprinted in Skepticism: A Contemporary Reader, edited by Keith D e R o s e and Ted A. Warfield ( O x f o r d : O x f o r d University Press, 1999), pp. 7 6 - 9 0 .

489

Externalism and Disjunctivism

What emerges from these considerations, I believe, is that McDowell espouses a thoroughgoing externalism about mental content. Subjective states are not individuated by what it is like to be in them, but at least in part by what environmental states of affairs they lay hold of. That is why he can announce that p and Mp are "precisely not blankly external to [one's] subjectivity." In his reckoning, Mp counts as a subjective state even though one could not be in that state unless the environmental fact ρ obtained. The resulting combination of views—content externalism plus epistemological internalism—may seem surprising. It is often assumed that content externalism makes trouble for epistemological internalism.25 Could McDowell be confounding expectations by using content externalism in the service of epistemological internalism? 2 6 I believe that McDowell is indeed combining content externalism with an epistemological internalism of sorts, but it turns out not to be one of the usual sorts. Let us run through some of the doctrines typically held by those who call themselves internalists about knowledge and see where McDowell fits in. First, there is the claim that knowledge is subject to higher-order requirements: the factors that make a belief justified or convert it to knowledge must themselves be known or at least knowable. A special case of this is the KK thesis—knowledge requires knowing that you know, or at least being in a position to know that you know. McDowell does accept a higher-order requirement on knowing: factors that contribute to one's epistemic standing must be within one's ken, and a knower must be in a position to know that he knows. That was one of the premises in the internalist syllogism. There need not be anything distinctively internalist about this requirement, however. Paradigmatic externalists can accept it, too, provided that the higher-order knowledge is reachable by an externalist route. For example, one who says that knowledge is true belief produced by a reliable mechanism may say that a subject knows that he knows ρ (or that his belief in ρ was reliably formed) just so long as he has a true higher-order belief that was itself produced by a reliable mechanism. Second, there is the doctrine that knowledge-making factors must not only be knowable, but knowable in a special way: they must be accessible to introspective reflection. This doctrine is sometimes called accessibilism.27

25 26

27

See Laurence Bonjour, "Externalism/Internalism," in A Companion to Epistemology, edited by J . D a n c y and E. Sosa (Oxford: Blackwell, 1992), pp. 132-36. Juan C o m e s a n a has pointed out to me that in "Knowledge and the Internal" (Phibsophy and Phenomenological Research, 55 ( 1 9 9 5 ) , 8 7 7 - 9 3 ) , M c D o w e l l seems to repudiate the interpretation I a m advancing here. In response to a suggestion by S i m o n Blackburn that McDowell avoids a shrinkage in what is known by letting the mind "expand to 'embrace' all sorts o f worldly states o f affairs," M c D o w e l l says he never meant to propose any such thing. T h i s disavowal leaves m e at a loss. If I look just at "Criteria, Defeasibility, and Knowledge," I cannot see how to understand what McDowell is saying in any other way. I should note as well that T i m o t h y Williamson seems to have a similar understanding o f M c D o w ell, at least putting him in the c a m p o f those who let an externalist conception o f the m i n d make it possible to construe knowledge as a mental state; see the Introduction and Chapter 1 o f Knowledge and its Limits (Oxford: O x f o r d University Press, 2 0 0 0 ) . Richard Feldman and Earl Conee, "Internalism D e f e n d e d , " American 1-18.

Philosophical

Quarterly,

38 (2001),

490

T h e Epistemic Significance o f Perception

The accessibilist does offer a distinctively internalist higher-order requirement, but I believe it is one that McDowell would repudiate. What gives one knowledge of ρ is Mp, and McDowell apparently denies that a subject can know on the basis of introspection alone that his state is Mp rather than Ap.28 Third, there is the doctrine that knowledge-making factors must be internal in the sense of being internal to the subject's mental life. This doctrine has been called mentalism.2