Rationality and the Subject’s Point of View

by

Declan Smithies

A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Philosophy New York University September, 2006

Paul Boghossian

© Declan Smithies All Rights Reserved, 2006

ACKNOWLEDGMENTS

As Wittgenstein once remarked, a philosopher who avoids discussion is like a boxer who never enters the ring. I am fortunate to have had the chance to shape up my ideas with a truly exceptional group of trainers and sparring partners. I thank them for throwing me so many jabs, body shots and occasional heavy rights. Above all, I would like to thank my principal consiglieri, Paul Boghossian, and the other members of my committee, Ned Block, Christopher Peacocke and Crispin Wright. I would also like to single out James Pryor, who has helped me since arriving at NYU last year, and Bill Brewer, who supervised me during my year as a visiting student at the University of Warwick. All of them have had a huge influence on this dissertation through their feedback and their own writing. During my time in New York, I have been helped by numerous discussions with Yuval Avnur, Ray Buchanan, Greg Epstein, Dana Evan, Geoff Lee, John Morrison, Anna-Sara Malmgren, Karl Schafer, Josh Schechter, Nico Silins, Sebastian Watzl and Massahiro Yamada. Back in the UK, I benefited from talking with Stephen Butterfill, Naomi Eilan, Susan Hurley, Hemdat Lerman, Rory Madden, Richard Price, Johannes Roessler, Susanna Schellenberg and Nick Shea. Finally, I would like thank my parents, Joan and Christopher, for their kindness and encouragement. This dissertation is dedicated to my father, who first inspired my interest in philosophy. Since we first discussed relativism in matters of taste and

iii

morals more than twenty years ago, I have enjoyed many philosophical conversations with him and I am immensely grateful.

iv

TABLE OF CONTENTS

Acknowledgments

iii

Introduction

1

Chapter 1. Rationality and the Subject’s Point of View

12

1.1. The Core Insight

12

1.2. Clairvoyance and Brains in Vats

13

1.3. Variations on Blindsight

17

1.4. Phenomenology, Attitude and Content

24

1.5. The Phenomenology of Belief

28

1.6. The Problem of Forgotten Evidence

31

1.7. The Problem of Stored Beliefs

33

1.8. Blind-Reasoning

37

Chapter 2. The Rational Autonomy of Personal Level Psychology

42

2.1. Personal and Subpersonal Levels

42

2.2. The Modular and the Conceptual

51

2.3. Functional Zombies

57

2.4. Analytical Functionalism and the Rationality Constraints

61

2.5. Normative Functionalism

71

Chapter 3. Rationality, Reflection and Epistemic Responsibility

74

3.1. Why Care about Rationality?

74

3.2. Rationality and Epistemic Responsibility

76

3.3. An Argument from Epistemic Responsibility

80

3.4. The Reflective Access Requirement

83

3.5. Explaining the Reflective Access Requirement

89

3.6. The Internalism/Externalism Debate

95

Chapter 4. Rationality and the Problem of Truth-Conduciveness 4.1. The Problem of Truth-Conduciveness

v

103 103

4.2. The Reliabilist Strategies

105

4.3. The Rational Presupposition Strategy

110

4.4. An Analysis of Knowledge

115

Chapter 5. Rationality, Knowledge and Luck

122

5.1. Anti-Luck Epistemology

122

5.2. Reliabilist Theories of Knowledge

124

5.3. The Super-Gettier Cases

129

5.4. Dispositions and Counterfactuals

132

5.5. A Dispositional Theory of Knowledge?

136

5.6. McDowell on the Composite Conception

139

5.7. The Value of Knowledge

144

Bibliography

147

vi

INTRODUCTION

The Significance of Consciousness For much of the last century, consciousness was a neglected topic, not only in philosophy, but also in the empirical sciences. Fortunately, that period of neglect has now come to an end. Recent decades have seen plenty of interesting work on consciousness, including empirical work on the neural correlates of consciousness and philosophical work on the metaphysics of consciousness.1 Still, the question of the significance of consciousness has received less attention than it deserves.2 Indeed, it is quite widely assumed that consciousness carries rather little in the way of theoretical significance, aside from questions about its seat in the brain and its place in the physical world.3 In my view, this is a massive oversight. The significance of consciousness extends well beyond metaphysics into the core of epistemology and the theory of cognition. The principal thesis of this dissertation is that there is a fundamental constitutive connection between consciousness and rationality. This, I argue, has important consequences for the nature of personal level psychology, epistemic responsibility and knowledge. My overall aim, then, is to show that consciousness has a significance that thoroughly permeates our cognitive lives.

1

See Block, Flanagan and Guzeldere (1997) for a representative selection. Siewert (1998) and Campbell (2002) are exceptions, but their claims differ from mine. 3 This assumption is made explicit in Fodor (1975) and Chomsky (1976). See Chapter Two. 2

1

Consciousness and Rationality Consciousness and rationality are the key concepts for this dissertation. So I will begin by putting an intuitive gloss on each of them. Consciousness is the property that mental states have when there is something it is like for the subject to be in them (Nagel, 1974). This characterization is not intended to provide anything like a reductive analysis of consciousness in terms of more basic ingredients. It is merely intended to capture the pre-theoretic notion that conscious states are those that have a phenomenology or subjective character. Like many other concepts, the concept of consciousness is best learned by means of examples – the paradigms in this case being perceptual experiences, bodily sensations, emotions, memories and the like. A distinction is often drawn between this phenomenal kind of consciousness and other purely functional kinds of consciousness.4 In my view, the phenomenal kind of consciousness is the only kind there is.5 But this view does not play a very crucial role in what follows, since my proposals are explicitly framed in terms of phenomenal consciousness. Moreover, I do not make any commitments regarding the metaphysical nature of phenomenal consciousness. For instance, I am neutral, at least at the outset, as to whether it is ultimately functional in nature.6 As far as I can see, there is a whole range of positions concerning the metaphysics of phenomenal consciousness that are consistent with the claim that there is a fundamental constitutive connection between phenomenal consciousness and rationality. 4

See Block (1995) for the distinction between phenomenal consciousness and access consciousness. See Chapter One in support of this claim. 6 See Chapter Two for some reasons to reject a functionalist theory of phenomenal consciousness. 5

2

For the purposes of this dissertation, I restrict my attention to the case of epistemic rationality – the rationality of belief. In my view, the main arguments generalize fairly straightforwardly to the case of practical rationality – the rationality of action. However, an explicit discussion of practical rationality will have to wait for another occasion. Rationality is a certain kind of normative property, which can be elucidated in terms of the notion of a reason. To say that a belief is rational is to say that it is reasonable – that is, it is held on the basis of good reasons. As for the metaphysical nature of reasons themselves, we can remain neutral, since our interest in reasons will be restricted to their role in determining what it is rational for the subject to believe. A good reason for a belief is just whatever it is that makes it rational for the subject to hold the belief when he has such a reason.7 It may be rational for a subject to hold a certain belief even if he does not in fact hold the belief rationally.8 A belief is rational only if it is held on the basis of the reasons that the subject has for holding it. But a subject may have reasons to hold a belief that he does not in fact hold. Alternatively, he may hold a belief that he has reasons to hold but not on the basis of the reasons he has for doing so. The basing relation can be understood as a certain kind of causal relation that holds between the belief and other mental states that provide the subject with his reasons for holding it.9 7

It is standard practice to draw a distinction between “prima facie” and “all things considered” notions of reasons and rationality. 8 In the jargon, this is a case of “propositional” without “doxastic” rationality. 9 The basing relation must be a non-deviant causal relation, so as to rule out deviant causal chains of the kind described by Davidson (1980, p.79).

3

We need not decide whether the mental states that provide subjects with their reasons for belief thereby constitute the subject’s reasons.10 For our purposes, the important question is how mental states provide subjects with reasons for holding certain beliefs and thereby make it the case that it is rational for the subject to hold those beliefs. What are the properties of a subject’s mental states that determine what it is rational for him to believe? An Overview In Chapter One, I argue for the thesis of Phenomenal Dependence, which is the thesis that what it is rational for the subject to believe depends solely on the phenomenal properties of his mental states. In other words, mental states provide subjects with reasons for belief in virtue of their phenomenal properties. The phenomenal properties of a mental state are the properties in virtue of which that state is phenomenally conscious in the way that is, when it is. Thus, Phenomenal Dependence purports to articulate a deep constitutive connection between rationality and phenomenal consciousness. I argue for Phenomenal Dependence on the basis of intuitive judgements about cases, such as the imaginary cases of Clairvoyance and the Brain in a Vat, as well as various hypothetical variations on the empirical phenomenon of blindsight. The intuitions prompted by these various cases manifest what I call the “core insight” about rationality – namely, that rationality is a normative property that is sensitive in a distinctive way to the limitations of the subject’s point of view on the world. In 10

See Dancy (2000) for an extended discussion of the issue of whether reasons are mental states.

4

particular, these intuitions support the claim that reliability is neither necessary nor sufficient for rationality, since reliability is not sensitive in the relevant way to the limitations of the subject’s point of view. I argue that Phenomenal Dependence provides the best explanation of these intuitions. Moreover, given that we can elucidate the notion of a subjective point of view by appealing to phenomenal consciousness, Phenomenal Dependence provides a way of articulating the core insight that rationality is sensitive to the limitations of the subject’s point of view. In Chapter Two, I argue that Phenomenal Dependence supports a conception of personal level psychology as rationally autonomous from subpersonal level psychology. The key idea is that personal level states have phenomenal properties, in virtue of which they are available to phenomenal consciousness, whereas subpersonal states do not. Given Phenomenal Dependence, it follows that personal level states are suited to figure in psychological processes that are evaluable for rationality, whereas subpersonal states are not. This explains why inferential reasoning at the personal level is evaluable for rationality, whereas computational information processing at the subpersonal level is not. I go on to argue that Phenomenal Dependence rules out the possibility of a rational functional zombie. Moreover, there is at least a prima facie case to made against ruling out the possibility of a functional zombie altogether. These claims provide the basis for an argument that analytical functionalism is inconsistent with the existence of rationality constraints on beliefs and other propositional attitudes. I sketch an alternative conception of functionalism – normative functionalism – that is

5

consistent with the rationality constraints. But I argue that it can be explained as a consequence of Phenomenal Dependence, since it is in virtue of their phenomenal properties that propositional attitudes play the rational functional roles laid down by the rationality constraints. In Chapter Three, I argue that Phenomenal Dependence explains the truth of the Reflective Access Requirement, which states that what it is rational for the subject to believe is accessible to the subject by reflection alone. Moreover, the Reflective Access Requirement plays an indispensable role in making sense of our conception of epistemic responsibility in reflective critical reasoning, which is central to our conception of ourselves as persons in contrast with other rational but unreflective animals. Phenomenal Dependence itself is perfectly compatible with the existence of unreflective rationality, but it also plays an indispensable role in explaining distinctive features of reflective critical reasoning, which are deeply important for our own self-conception. In Chapter Four, I address the Problem of Truth-Conduciveness, which is the problem of explaining how it is that following rational belief-forming methods is conducive to the aim of believing the truth. This problem is particularly urgent for any theory on which reliability is not a necessary condition for rationality. I propose an alternative solution, which depends on the claim that reliability is a “rational presupposition” for any rational method. The rational presuppositions of a rational method may or may not be satisfied, but if they are, then the method in question is reliable in the circumstances. I go on to argue that knowledge is also a rational

6

presupposition for any rational method and I conclude with an analysis of knowledge as belief that is rationally held when all of its rational presuppositions are satisfied. In Chapter Five, I argue that this analysis illuminates the value of knowledge by explaining the “anti-luck” intuition that knowledge is incompatible with luck. We value knowledge because it is the kind of cognitive achievement that results when a subject’s beliefs achieve the aim of truth in a way that is immune from epistemic luck. I argue that the anti-luck intuition cannot be explained purely in terms of reliability, but only in a way that makes essential appeal to the notion of rationality. A belief is immune from epistemic luck if and only if it is rationally held when all of its rational presuppositions are satisfied. Thus, rationality is essential for knowledge and not merely a desirable, optional extra. In summary, Phenomenal Dependence articulates a conception of rationality that is not only intuitively motivated, but is also highly theoretically significant in the sense that it plays a central and pivotal role in our conceptual scheme. The fundamental

constitutive

connection

between

rationality

and

phenomenal

consciousness is not a fact of merely incidental significance, since it plays an important theoretical role in explaining the nature of personal level psychology, epistemic responsibility and knowledge. Methodology Let me say a few words about methodology. I take it that any philosophical theory of the nature of a property is subject to a pair of constraints. First, it should

7

explain the presumed correctness of our intuitive judgements about whether the property is instantiated in particular actual or hypothetical cases. And second, it should explain our interest in the property by illuminating its wider theoretical significance. The theory of rationality that I propose is designed to satisfy both of these constraints. How do our intuitive judgements about cases provide data for a philosophical theory? Here, it is important to distinguish a philosophical theory of the nature of a property from a psychological theory of our concept of the property or of our dispositions to make judgements about the property.11 A philosophical theory of rationality, for example, is a theory of the nature of rationality itself. However, our intuitive judgements about rationality can provide us with data for such a theory only in the context of a defeasible presumption that they constitute a reliable guide to the nature of reality. To explain the data provided by an intuitive judgement, then, is to explain the presumed correctness of the judgement in terms of a theory of the nature of the property in question. Our intuitive judgements are not sacrosanct. In order to make sense of the theoretical significance of a property, a philosophical theory may be forced to explain away some of our intuitive judgements as intelligible errors. In the worst case scenario, such a theory may hold that there is in fact no theoretically significant

11

Pace Goldman: “Once it is granted that conceptual analysis has the task of laying bare the semantical features of items that are ‘hidden’ within a cognitive system – items that are neither directly observable nor directly introspectible – it should become credible that scientific investigation is in principle relevant to the task.” (1999, p.22)

8

property that explains even a substantial proportion of the intuitive judgements we are inclined to make. But an error theory of this kind is a last resort, to be adopted only when the alternatives have been found wanting. The default presumption is that our intuitive judgements track a theoretically significant property, whose nature can be more deeply understood by reflecting on those judgements. I will sometimes refer to our intuitive judgements about cases as “intuitions”, but I do not thereby intend to beg any questions about their provenance, such as whether they have their source in some a priori faculty of rational intuition. Intuitions are simply classification judgements and we can remain neutral on whether they have an a priori or an a posteriori status.12 It does matter, though, that they are “intuitive” in the sense that they are relatively pre-theoretical. After all, they are to serve as data for a theory, so they cannot simply be read-off from the theory in question. But they need not be entirely pre-theoretical, since they can be informed by theoretical considerations without being read-off from any particular theory.13 Typically, intuitions are not snap judgements, but considered judgements. This is crucial for understanding the nature of “intuition-pumping” as a resource for philosophical theory-building. Theoretical considerations come into play not only in determining which considered judgements to accept and which to reject, as in Rawls’ (1971) method of reflective equilibrium, but also in the process of reaching considered judgements themselves. For this reason, I have not taken the opportunity

12 13

See Williamson (2005) for some discussion of this issue. For example, consider the discussion of the problem of forgotten evidence in Chapter One.

9

to conduct surveys in order to gauge the intuitions of ordinary folk in order to determine whether they are invariant across social class, ethnicity and sexual orientation.14 The intuitions of ordinary folk (and philosophers, too, for that matter) are relevant only to the extent that they manifest good philosophical thinking about cases. Grasping the relevant concepts may be enough for correctly classifying the easy cases, but some cases are harder than others. In order to classify the hard cases correctly, reflective judgement may be required, which involves an ability to consider variations on cases, to spot salient similarities and differences between them, to draw out the implications of judgments about cases for others like it, to consider the alternatives, assess their merits, and so on. On this conception, intuition-pumping is an essentially philosophical enterprise. When I make a claim about our intuitions with respect to some case, I do not thereby commit myself to any empirical speculations concerning what ordinary folk would say when presented with the case. Rather, I am committing myself to a normative claim about what they should say, given the presumption that they share the same set of basic concepts. We cannot translate this normative claim into a descriptive claim about what ordinary folk would say without helping ourselves to a very considerable degree of idealization.15 Therefore, it is far from clear how surveying the responses of ordinary folk would bear on the question of what their responses should be. In

14 15

For this methodology, see Weinberg, Nichols and Stich (2001). Compare Kauppinen (forthcoming), who suggests that correct classification requires drawing on a semantic competence with the relevant concepts in the absence of distraction by pragmatic factors and given full attention to the relevant details of the case. But see also the footnote below.

10

order to make progress on the normative question of what we should say, rather than the descriptive question of what we would say, there is no substitute for engaging in the modes of reflective enquiry that characterize the philosophical tradition.16

16

Burge (1986) suggests that philosophical reflection, and other modes of enquiry, can deepen our understanding of concepts that we only incompletely understand. The suggestion is that incomplete understanding derives most fundamentally not from deference to experts, as is often supposed, but rather from deference to reality. I hope to develop this point in the future.

11

CHAPTER ONE

RATIONALITY AND THE SUBJECT’S POINT OF VIEW

Chapter 1. Rationality and the Subject’s Point of View

1.1. The Core Insight Rationality is a normative property that is sensitive to the subject’s point of view on the world. Whether a belief is true depends on the state of the objective world. By contrast, whether a belief is rational depends on mental states of the subject that constitute his subjective perspective or point of view on the world. This is why a belief can be rational even if it is not true. The norm of rationality characterizes what the subject ought to believe and do given not the state of the objective world, but rather given the limitations of his subjective point of view on the world. My aim in this chapter is to show that we can elucidate this core insight about rationality by appealing to the notion of phenomenal consciousness. For it is plausible that we can elucidate the notion of the subject’s point of view by appealing to the notion of phenomenal consciousness. And, as much of this chapter is devoted to arguing, there is also a deep constitutive connection between rationality and phenomenal consciousness. In particular, what it is rational for the subject to believe depends on the phenomenal properties of his mental states. My contention is that this provides the most fundamental articulation of the core insight that rationality is sensitive to the subject’s point of view.

12

1.2. Clairvoyance and Brains in Vats We can illustrate the core insight that rationality is sensitive to the subject’s point of view by appealing to some well-known cases that provide intuitive counterexamples to the necessity and sufficiency of reliability as a condition for rationality. The notion of reliability can be glossed in modal terms: a way or method of forming beliefs is reliable in a possible circumstance C just in case it would have yielded true beliefs in possible circumstances sufficiently similar to C. Pure reliabilism, then, is the extreme view that reliability is both necessary and sufficient for rationality: •

Pure Reliabilism: necessarily, a belief is rational if and only if it is formed on the basis of a belief-forming method that is reliable in the circumstances.

A counterexample to necessity is provided by the case of the Brain in a Vat, while the case of Clairvoyance provides a counterexample to sufficiency:1 •

The Brain in a Vat Case: While you are sleeping, a malicious scientist removes your brain from your body and places it in a vat where he stimulates it with electrodes. You do not notice the change, and your conscious mental life continues more or less as it would have done if the scientist had not intervened. But the method of forming beliefs on the basis of perceptual experience is now unreliable, though it would have been reliable if the scientist had not intervened.

1

These cases are adapted from Cohen and Lehrer’s (1983) so-called “New Evil Demon Problem” and Bonjour’s (1985, Ch.3) case of the Clairvoyant, Norman.

13

•

The Clairvoyance Case: While you are sleeping, a benevolent scientist implants a mechanism in your brain which causes you to form beliefs about distant states of affairs on the basis of reliable hunches. You do not notice the change, and your conscious mental life continues more or less as it would have done if the scientist had not intervened. But the method of forming beliefs on the basis of hunches is now reliable, though it would have been unreliable if he had not intervened.

These two cases share a common structure. In each case, the interventions of the scientist affect the reliability of the subject’s belief-forming methods without impacting on the conscious life of the subject. Intuitively, however, they do not thereby affect the rationality of those methods. In the case of the Brain in a Vat, for example, the method of forming beliefs on the basis of perceptual experience is intuitively rational, despite the fact that it is no longer reliable. In the case of Clairvoyance, by contrast, the method of forming beliefs on the basis of mere hunches (that is, mere conscious inclinations towards judgement) is intuitively irrational, despite the fact that it is now reliable. If we take these intuitions seriously, then we should conclude that reliability is neither necessary nor sufficient for the rationality of a belief-forming method. And this is because reliability is not sensitive in the right kind of way to the limitations of the subject’s point of view on the world. These counterexamples to pure reliabilism share a common structure and they can also be given a common explanation in terms of the following supervenience principle:

14

•

Phenomenal Supervenience: what it is rational for a subject to believe at a time supervenes on the phenomenal properties of his mental states at that time.

Let me explain. The phenomenal properties of a mental state are the properties in virtue of which that state is phenomenally conscious in the way that it is, when it is. And a state is phenomenally conscious just in case there is something it is like for the subject to be in that state. In other words, when there is something it is like for the subject to be in a state, what it is like depends on the phenomenal properties of that state. Now, what Phenomenal Supervenience states is that, for any pair of triples of subjects, worlds and times, there can be no difference in what it is rational for a subject to believe at any given time without some difference in the phenomenal properties of his mental states at that time. That is to say, phenomenal duplicates, who are alike in respect of the phenomenal properties of their mental states, are also rational duplicates, who are alike in respect of what it is rational for them to believe. The intuitive counterexamples to pure reliabilism can now be explained by appeal to Phenomenal Supervenience. First, let us compare the Brain in a Vat with his normally embodied and embedded phenomenal duplicate. Intuitively, it is rational for a normal subject to form beliefs on the basis of perceptual experience. But then it must also be rational for the Brain in a Vat to form beliefs on the basis of perceptual experience, even though this method will be unreliable given his abnormal circumstances. Similarly, let us compare the Clairvoyant with his phenomenal duplicate in normal circumstances. Intuitively, it is irrational for a

15

normal subject to form beliefs about distant states of affairs just on the basis of hunches. So, it must also be irrational for the Clairvoyant to form beliefs on the basis of hunches, although this method will be reliable given his abnormal circumstances. Phenomenal Supervenience imposes substantial constraints on the kinds of difference that make for a difference in what it is rational for the subject to believe. In particular, it rules out anything that fails to make for a difference in the phenomenal properties of the subject’s mental states. Since differences in reliability do not suffice for differences in the phenomenal properties of a subject’s mental states, they do not suffice for differences in what it is rational for the subject to believe.

This

is

how

Phenomenal

Supervenience

explains

the

intuitive

counterexamples to pure reliabilism. However, as always with supervenience principles, this one stands in need of explanation too. Why is it that there can be no difference in what it is rational for the subject to believe without some difference in the phenomenal properties of his mental states? My proposal is that the supervenience principle is explained by a stronger principle of dependence: •

Phenomenal Dependence: what it is rational for a subject to believe at a time depends solely on the phenomenal properties of his mental states at that time.

The dependence relation is stronger than the supervenience relation, since it implies an explanatory connection. What it is rational for a subject to believe depends on the phenomenal properties of his mental states in the sense that any difference in what it is rational for the subject to believe requires a corresponding difference in the phenomenal properties of his mental states which explains that difference. The

16

suggestion here is not merely that mental states must possess phenomenal properties if they are to provide reasons for belief. It is the stronger claim that mental states provide reasons for belief in virtue of their phenomenal properties.2 Phenomenal Dependence is a thesis that is pitched at fairly high level of generality. It does not provide answers to more specific questions concerning which phenomenal properties make it rational for the subject to form which beliefs, and why. However, Phenomenal Dependence provides the most fundamental articulation of the core insight that rationality is sensitive to the subject’s point of view. After all, we can elucidate the notion of the subject’s point of view by appealing to the phenomenal properties of the subject’s mental states. It is only in virtue of having phenomenal properties that states count as figuring within the subject’s point of view. And it is only in virtue of having states with phenomenal properties that an individual counts as a subject at all — that is, as an individual with a subjective point of view. So, the constitutive connection between rationality and phenomenal consciousness provides us with a way of articulating the core insight that rationality is sensitive to the subject’s point of view. 1.3. Variations on Blindsight So far, I have argued that we can explain the core insight that rationality is sensitive to the subject’s point of view by appealing to phenomenal consciousness.

2

See Alston’s (1988) “internalist externalism” for a view on which reason-giving states have phenomenal properties, but they do not play their reason-giving roles in virtue of their phenomenal properties.

17

However, it remains to be seen whether the appeal to phenomenal consciousness can be avoided in explaining the core insight. In what follows, I will argue that the appeal to phenomenal consciousness is indispensable because it is required for explaining the intuitions prompted by various hypothetical variations on the empirical phenomenon of blindsight. Patients with blindsight lose conscious visual experience in “blind” regions of the visual field, owing to lesions in the visual cortex.3 As a result, they are unable to initiate reasoning, action or verbal reports with respect to stimuli presented in the blind field. However, under conditions of prompting and given a range of alternatives, they are able to draw on phenomenally unconscious visual information in responding reliably to stimuli presented in the blind field. At first, patients tend to dismiss their reports as mere guesswork and express surprise when they learn of their reliability. But we can imagine a blindsighter who reports with complete confidence on stimuli presented in the blind field, despite lacking any conscious visual experience of them and having no inductive basis for believing that he is reliable. Let us suppose that he is not merely guessing, but actually believes his reports. Intuitively, his beliefs are no more rational than blind guesswork, even though they are formed on the basis of reliable visual information.4 Phenomenal Supervenience explains our intuitions about blindsight in the same way that it explains our intuitions about Clairvoyance. Consider a phenomenal

3 4

See Weiskrantz (1986) for an overview. See Eilan (1998) and Dretske (2006) for similar verdicts on the blindsight case.

18

duplicate of the blindsighter who forms beliefs about stimuli even though he does not consciously experience them or receive any unconscious visual information about them. Since his beliefs are certainly no more rational than blind guesswork, it follows from Phenomenal Supervenience that the beliefs of the blindsighter are not rational either. But why is it that visual experiences can provide reasons for belief, whereas phenomenally unconscious visual states cannot? According to Phenomenal Dependence, mental states provide reasons for belief in virtue of their phenomenal properties. But the phenomenally unconscious visual states of the blindsighter do not have any phenomenal properties, and this is why they fail to provide him with reasons for belief. Is there any way to avoid the appeal to phenomenal consciousness here? The most promising strategy is to offer an explanation in purely functional terms. We can state the overall shape of this proposal by replacing Phenomenal Dependence with the following functionalist alternative: •

Functional Dependence: what it is rational for the subject to believe at a time depends on functional properties of his mental states at that time, i.e. the second-order properties of having first-order properties that play functional role R.

Different versions of the functionalist proposal will offer differing characterizations of functional role R. But if any such proposal is to provide a genuine alternative to the appeal to phenomenal consciousness, then it will have to conform to certain conditions. Role R cannot be specified in terms that explicitly presuppose the

19

presence of rationality, since this would be to beg the question. Nor can it be specified in terms that explicitly presuppose the presence of phenomenal consciousness, since this would fail to provide a genuine alternative. Indeed, functional role R cannot even be regarded as providing a functionalist analysis of phenomenal consciousness. The proponent of the functionalist strategy must concede that it is possible for a state to occupy role R without thereby becoming phenomenally conscious or instantiating any phenomenal properties. This point will be crucial in what follows. Ned Block (1995) has argued that our ordinary concept of consciousness is ambiguous, or even confused, since it fails to distinguish between phenomenal consciousness and various purely functional notions of consciousness, such as access consciousness and reflective consciousness. Therefore, on behalf of the proponent of the functionalist strategy, we might consider whether the core insight about rationality can be explained by appeal to one of these purely functional notions of consciousness, rather than phenomenal consciousness. For these purposes, though, one would have to assume, as Block does, that phenomenal consciousness itself cannot be analyzed in terms of any such purely functional notion of consciousness.5 One would have to allow that they could be satisfied by a zombie – a creature that lacks any phenomenally conscious states at all. In my view, this undermines Block’s claim that any of these purely functional notions amounts to a genuine notion of consciousness, since it is not plausible to suppose that a zombie is conscious in any 5

This is a commitment of Block’s (2002) phenomenal realism.

20

sense at all.6 For the purposes of the functionalist strategy, however, it does not matter whether or not the favoured functional notion amounts to a genuine notion of consciousness that is distinct from phenomenal consciousness. Let us begin with Block’s own definition of access consciousness: A is access-consciousness. A state is access conscious if it is poised for direct control of thought and action. To add more detail, a representation is Aconscious if it is poised for free use in reasoning and for direct “rational” control of action and speech. (The “rational” is meant to rule out the kind of control that obtains in blindsight.) (1995, p.382) If this is to provide a genuine and non-question-begging alternative to an appeal to phenomenal consciousness, then the appearance of “rational” in the definition should be understood as Block intends — that is, merely as a placeholder for a purely functionalist condition that rules out the kind of control that obtains in blindsight. As we have seen, blindsight patients are unable to initiate responses to stimuli in the blind field except under conditions of forced choice. So, this kind of control can be ruled out by means of a condition that requires responses to be initiated by the subject automatically – in other words, not merely under conditions of forced choice. The proposal we want to consider, then, is that the blindsighter’s beliefs are irrational because they are based on visual states that fail to satisfy the purely functional conditions for access consciousness. In order to test this proposal, let us consider Block’s super-blindsighter, who is just like the regular blindsighter except that he responds to stimuli presented in the 6

Compare Burge (1997, p.428): “Any being that is not phenomenally conscious is not conscious in any sense. There are no zombies that lack phenomenal consciousness but are conscious in some further way.”

21

blind field automatically, and not merely under forced choice conditions. Like the regular blindsighter, he has no conscious visual experience of stimuli presented in the blind field and, we may assume, he lacks any inductive basis for believing that he has reliable means for responding to them.7 Unlike the regular blindsighter, though, his responses are based on visual information which is access conscious, although it is not phenomenally conscious. However, this difference makes no difference for the rational status of his beliefs. Intuitively, his beliefs are no more rational than blind guesswork, despite the fact that they are formed on the basis of reliable visual information that also satisfies the functional conditions for access consciousness. Therefore, the proponent of the functionalist strategy might appeal to a more demanding functional notion, such as reflective consciousness, which can be defined as follows. A state is reflectively conscious just in case it is not only access conscious, but is also the target of a reliably formed higher-order thought, which is formed without mediation by inference or observation.8 The proposal now is that we can explain the irrationality of the beliefs of both the blindsighter and the superblindsighter in terms of the claim that they fail to satisfy these purely functional conditions for reflective consciousness. One obvious problem with this proposal is that it seems to rule out the possibility of unreflective rationality.9 But we can leave this issue aside, since the proposal fails in any case to explain the relevant intuitions. 7

Block’s super-blindsighter is “trained to prompt himself at will” (1995, p.385), but I omit this detail, since it is crucial for my purposes that the super-blindsighter lacks the kind of background beliefs about the reliability of the belief-forming method that would result from such training. 8 Compare Rosenthal’s (1997) higher-order thought theory of consciousness. 9 See Chapter Three for further discussion of the issue of unreflective rationality.

22

Consider the hyper-blindsighter, who is just like the super-blindsighter except that he also has a reliable mechanism that yields higher-order thoughts about his phenomenally unconscious visual states, but without mediation by inference or observation. Thus, the hyper-blindsighter is reliable not only about the stimuli presented in his blind field, but also about the phenomenally unconscious visual states that carry information about those stimuli. Intuitively, though, his higher-order thoughts themselves are no more rational than blind guesswork. After all, they are not based on any phenomenally conscious states, but merely pop into his head out of the blue. But surely we cannot turn irrational beliefs into rational ones by adding more irrational beliefs? If not, then his beliefs about stimuli presented in the blind field are no more rational than blind guesswork, despite being formed on the basis of reliable visual information that also satisfies the functional conditions for reflective consciousness. In fact, it doesn’t matter how much we complicate functional role R, so long as it remains insufficient for phenomenality. For we can always imagine an uberblindsighter, who forms beliefs on the basis of visual information that occupies functional role R, but which is not available to phenomenal consciousness. Intuitively, such beliefs can be no more rational than blind guesswork and the intuitions are best explained in terms of the appeal to phenomenal consciousness. For the uber-blindsighter will have a phenomenal duplicate whose beliefs are based on states that are not available to phenomenal consciousness, but which do not occupy functional role R either. Since his beliefs are certainly no more rational than blind

23

guesswork, then it follows from Phenomenal Supervenience that neither are those of the uber-blindsighter. A deeper explanation is provided by Phenomenal Dependence, since the phenomenally unconscious visual states of the uber-blindsighter do not have any phenomenal properties and this is why they do not provide him with reasons for belief. 1.4. Phenomenology, Attitude and Content Phenomenal Dependence states that what it is rational for a subject to believe depends on the phenomenal properties of his mental states. Surely, though, what it is rational for a subject to believe depends on which kinds of attitudes and contents are instantiated by his mental states. For example, it is prima facie rational for me to believe that there is a red cube before me if I am having a perceptual experience as of a red cube before me, but not if I am merely imagining a red cube before me or if I am having a perceptual experience as of a blue sphere before me. This raises an issue about the relationship between the phenomenal properties of a conscious mental state and the kind of attitude and content that it instantiates. What I want to suggest is that differences in attitude or content suffice for differences in what it is rational for the subject to believe if and only if they suffice for differences in the phenomenal properties of the subject’s mental states. Let us begin with attitude kinds. Intuitively, the phenomenal difference between experiencing and imagining explains why it is rational for me to believe that there is a red cube before me on the basis of experience, though not on the basis of

24

imagination.10 When I have an experience as of a red cube before me, it really does seem to me just in virtue of having that experience that there is red cube before me. But this is not so when I merely imagine it. What it is like to have the experience is quite different from what it is like merely to imagine it, which explains why states of these different kinds play different rational roles. More generally, what it is like to judge that p is different from what it is like to hope that p, to suppose that p, to fear that p, and so on, and it is the phenomenal difference between these kinds of attitudes that explains the difference in their rational roles. It may be objected that this proposal is subject to counterexamples involving pairs of token states, such experiencing and dreaming, that have the same phenomenal properties, but which play different reason-giving roles because they are states of different kinds. However, even if dreams are sometimes very difficult to distinguish from genuine experiences, it does not follow that they have the same phenomenal properties. It may be plausible that the phenomenal properties of dreams are sufficiently similar to those of genuine experiences that they are difficult to distinguish, especially under the conditions of reduced cognitive functioning that usually come with sleep.11 But assuming that they do not have the very same phenomenal properties, this explains why they do not have the same reason-giving roles. The objector may insist that it is possible for episodes of experiencing and dreaming to have the very same phenomenal properties. Perhaps this is true, perhaps 10 11

Compare Pryor (2000, fn.37) and (2004, p.357). In his unpublished John Locke lectures, Sosa argues that there is no entailment from dreaming that one is having a perceptual experience to having a perceptual experience while one is dreaming.

25

not. But supposing it is, the right response is to deny that phenomenally identical episodes of experiencing and dreaming play different rational roles. A subject who takes such realistic dreams at face value is no less rational than the Brain in a Vat. So, differences in attitude suffice for a difference in rational roles if and only if they also suffice for a difference in phenomenal properties. Likewise, I claim, differences in content suffice for a difference in rational roles if and only if they suffice for a difference in phenomenal properties. For example, it is the phenomenal difference between an experience as of a red cube and an experience as of a blue sphere that explains why experiences with these different contents play different rational roles. However, differences in content do not always suffice for differences in phenomenal properties. Let us define content externalism as the thesis that some of the contents of a subject’s conscious mental states fail to supervene on the phenomenal properties of his mental states. This allows for the possibility that there can be pairs of phenomenal duplicates whose mental states have different contents in virtue of their embeddings in different kinds of environments. The principal source of motivation for content externalism derives from “Twin Earth” cases of the following kind.12 Oscar lives on Earth where the stuff in the rivers and lakes is composed of H2O molecules, while his phenomenal duplicate, Toscar, lives on Twin Earth where the stuff in the rivers and lakes is composed of XYZ molecules. Since water is essentially composed of H2O molecules, the stuff in the rivers and lakes on Twin 12

Adapted from Putnam (1975).

26

Earth is not water, but rather a different kind of substance – twin-water. Now, Oscar and Toscar know nothing of the chemical composition of the stuff in the rivers and lakes. All the same, it is intuitively plausible that they express different beliefs by the sentence, “water is wet”. After all, Oscar’s belief is true just in case water is wet, whereas Toscar’s belief is true just in case twin-water is wet. Their beliefs have different truth-conditions, and hence they have different contents, since contents have their truth-conditions essentially. Oscar believes the content that water is wet, whereas Toscar believes the content that twin-water is wet. Even though Oscar and Toscar are phenomenal duplicates, their beliefs have different contents. But since they are phenomenal duplicates, it seems right to say that Oscar thinks about water in the same type of way that Toscar thinks about twinwater. Their beliefs may have different truth-conditions in virtue of their embeddings in different kinds of environment, but these are contents of the same general type. The suggestion here is that contents may be typed by the phenomenal properties of the mental states that have them. Thus, whenever phenomenal duplicates have mental states that differ in their truth-conditional contents, there is a more general phenomenal type under which they fall.13 By appealing to phenomenal ways of typing contents, it is possible to reconcile content externalism with the theory of rationality that I am proposing. Let us assume that neither Oscar nor Toscar is guilty of any irrationality, so that all of their beliefs 13

Wedgwood (2006) makes the point that even if there is no type of truth-conditional content that is narrow, there may still be narrow ways of typing contents. If phenomenology is narrow, as it is natural to assume, then one narrow way of typing contents is by their phenomenology.

27

are rational beliefs. Therefore, although they are phenomenal duplicates, they are not duplicates with respect to their rational beliefs, since their beliefs are different. Nevertheless, there is no conflict with Phenomenological Supervenience, since they are duplicates with respect to what it is rational for them believe. The crucial point here is that the requirements of rationality are in a certain sense context-independent. Environmental context plays a role in determining which contents a subject believes, but not in determining which contents it is rational for him to believe. And we can flesh this out as follows. The phenomenal properties of a subject’s mental states determine which general types of contents it is rational for him to believe. And environmental context plays a role in determining which instances of those general types the subject believes. But the requirements of rationality themselves are context-independent – they are neutral with respect to which instances of those general types it is rational for the subject to believe. 1.5. The Phenomenology of Belief It is often claimed that there is a sharp distinction to be drawn between sensory states on the one hand and propositional attitudes on the other. In its starkest form, the claim is that sensory states have phenomenal properties, whereas propositional attitudes do not. Thus, for example, Richard Rorty writes: Beliefs don’t feel like anything – they don’t have phenomenal properties…. The attempt to hitch beliefs and pains together seems ad hoc. (1979, p.22) In my view, this view results from an overly narrow conception of phenomenology. I concur with Christopher Peacocke, who writes:

28

Perceptual experiences and sensations, on the one hand, and so-called occurrent propositional attitudes, on the other, differ in many respects. But there is one property they share. They both contribute to what, subjectively, it is like for the person who enjoys them. (1998, p.64) Here, we need to distinguish between standing beliefs, which are certainly not conscious in any sense, and their conscious manifestations as occurrent thoughts and judgements, which surely do affect the overall phenomenology of the subject. Moreover, we cannot characterize the subjective phenomenology of these conscious episodes in purely functional terms, such as access or reflective-access, unless we are able to give a functional reduction of phenomenal consciousness across the board. For if we cannot give any such functional reduction, then we must allow that the relevant functional notions are satisfied by a functional zombie. But a functional zombie has no subjective phenomenal life of any kind. So, the appeal to phenomenal consciousness is indispensable for characterizing the phenomenology of belief. Let it be granted that beliefs have phenomenal properties, at least in their conscious manifestations as occurrent thoughts and judgements. Still, it might be doubted that their phenomenal properties are essential to them, rather than being merely contingently associated with them. In this vein, David Chalmers writes: The most substantial requirements for having a specific belief will lie elsewhere than in the phenomenal. One could even subtract any phenomenal component out, leaving a concept of pseudobelief that resembles belief in most important respects except that it does not involve the concept of consciousness. Indeed, it is plausible that pseudobelief could do most of the explanatory work that is done by the concept of belief. (1996, p.20) Again, this view is encouraged by a conception of phenomenology as essentially sensory in nature. For the natural suggestion is that the phenomenology of occurrent

29

belief is exhausted by contingently associated sensory states, such as imagery or subvocalized speech. On this view, the phenomenal properties of an occurrent belief are not essential to its being the kind of belief it is. Someone could have the very same type of occurrent belief while having very different phenomenal properties or even none at all. This view is not adequate to the phenomenology of occurrent belief. First, it is far from obvious that any associated imagery or subvocalized speech has to be involved when it occurs to me that I have forgotten to consider a certain objection to my work, or when I judge that there may also be further objections that I have failed to consider. But even if we leave that point aside, the occurrence of such imagery is surely not sufficient to exhaust the phenomenology of occurrent thought. It is a familiar point that hearing the very same string of sounds involved in the utterance of a sentence can produce phenomenally different experiences depending on whether, and if so how, you understand the uttered sentence. So, for example, an utterance of the German sentence, “Empedocles liebt,” will produce phenomenally different experiences in, respectively, German speakers, English speakers and speakers of neither language. The same applies when the sentence is merely imagined, rather than heard. In such a case, it is not at all obvious how the phenomenal differences can be explained in terms of associated sensory imagery. A more plausible explanation is that an essential contribution to phenomenology is made by the intentional content of the occurrent belief — in this case, the content that

30

Empedocles leaped for English speakers and the content that Empedocles loves for German speakers. 1.6. The Problem of Forgotten Evidence Against Phenomenal Dependence, it may be objected that what it is rational for the subject to believe at a time depends on not only on the properties of her mental states at that time, but also on the properties of her mental states at earlier times. For instance, Alvin Goldman (2001) appeals to the so-called “problem of forgotten evidence” in support of the claim that the historical properties of a subject’s mental states can be relevant in determining what it is rational for the subject to believe. He argues that this is required for explaining how it can be rational for thinkers to retain beliefs even after they have irretrievably forgotten the rational grounds on which those beliefs were originally formed, as in the following case: Last year, Sally read a story about the health benefits of broccoli in the “Science” section of the New York Times. She then justifiably formed a belief in broccoli’s beneficial effects. She still retains this belief but no longer recalls her original evidential source (and has never encountered either corroborating or undermining sources). Nonetheless, her broccoli belief is still justified, and, if true, qualifies as a case of knowledge. Presumably, this is because her past acquisition of the belief was epistemically proper. (1999, p.215) In response, I want to grant that it is still rational for Sally to believe in the health benefits of broccoli even after she forgets the rational grounds on which she

31

originally formed the belief. But I deny that the rationality of her belief at the later time is explained in terms of the way in which it was formed it at the earlier time.14 Suppose that, last year, Sally also read a story about the health benefits of beer in the National Enquirer, and despite her rational doubts about the source, wishful thinking led her irrationally to form a belief in the health benefits of beer. A year later, she still holds the belief, but she has irretrievably forgotten the irrational grounds on which she originally formed it. At this later time, is her belief in the health benefits of beer any less rational than her belief in the health benefits of broccoli? According to Goldman, it is less rational because it was formed in an irrational way. But surely it would be not be rational for Sally to abandon her belief in the health benefits of beer, while retaining her belief in the health benefits of broccoli. So, if either belief is rational, then both are. Otherwise, we have the implausible consequence that Sally has an irrational belief that it would be irrational for her to abandon. What this means is that Sally’s irrationally formed belief in the health benefits of beer can be rationally preserved if she loses all memory of the irrational basis on which she formed it. That may sound surprising, but its plausibility emerges from reflection on the analogies between memory and testimony. Just as it can be rational to take testimony at face value in the absence of defeaters, so it can be rational to take memory at face value in the absence of defeaters. But when Sally forms her

14

My response follows that of Conee and Feldman (2001), although it was Rory Madden who suggested that I explore the helpful analogy between memory and testimony.

32

belief on the basis of testimony, she has a defeater that she loses when she later preserves her belief on the basis of memory — namely, her belief that the information about the health benefits of beer was derived from a dubious source. Certainly, there is something epistemically defective about Sally’s belief, but this can be explained in terms of a failure of knowledge, rather than rationality. Suppose that beer does in fact carry significant health benefits, so that Sally’s belief is not only rational, but also true. Nevertheless it falls short of knowledge, because intuitively it is merely a matter of luck that it is true. Again, the analogies between memory and testimony are illuminating. If I form two beliefs on the basis of testimony, derived from two distinct testimonial sources, my beliefs might be equally rational, even though only the first amounts to knowledge, because the second is derived from an irrational testimonial source. Similarly, if I preserve two beliefs in memory, they might be equally rational, even though only the first amounts to knowledge, because the second was formed in an irrational way. 1.7. The Problem of Stored Beliefs Another objection raised by Goldman (2001) is the so-called “problem of stored beliefs”. What it is rational for a subject to believe at a time often depends not only on which states are occurrent in phenomenal consciousness at that time, but also on which beliefs are stored in memory. For example, it may be rational for me to believe that the Tate Modern is in the capital of England because I believe that the Tate Modern is in London and that London is the capital of England. Moreover, it

33

will be rational for me to believe that the Tate Modern is in the capital of England whether or not these beliefs are occurrent in phenomenal consciousness. So, what it is rational for me to believe does not depend solely on what is occurrent in phenomenal consciousness. One might respond by restricting the application of Phenomenal Dependence to occurrent belief, rather than standing belief.15 But this strategy has two major defects. First, the significance of Phenomenal Dependence is dramatically reduced if its application is restricted in this way. And second, it is not clear that the restriction succeeds in solving the problem anyway. For it is not plausible that what it is rational for the subject to occurrently believe depends solely on what is occurrent in phenomenal consciousness. Peacocke (1998) gives examples of “NICS” selfascriptions, which can be rational even if they are formed on the basis of first-order beliefs that do not become occurrent in consciousness before the self-ascription is made. More generally, it is plausible that much of our conscious reasoning can be rational even if it depends on background beliefs that are not occurrent in consciousness. Conscious reasoning need not be fully articulated in order to be rational. My response is different. Phenomenal Dependence states that what it is rational for the subject to believe depends on the phenomenal properties of his mental states, but this does not imply that it depends solely on what it is occurrent in phenomenal consciousness. For a state can have phenomenal properties even if it is not occurrent 15

This suggestion is put forward by Conee and Feldman (2001).

34

in phenomenal consciousness, so long as it is available to become so. As I define them, the phenomenal properties of a state are the properties in virtue of which it is phenomenally conscious in the way that it is, when it is. Phenomenal properties are individuated by their potential contributions to phenomenal consciousness. But a state can have phenomenal properties even if there is nothing it is like for the subject to be in that state, so long as there are conditions under which there would be something it is like to be in it. So, I can allow that what it is rational for the subject to believe may depend on the phenomenal properties of beliefs that are stored in memory. It is often assumed that states have phenomenal properties only if they are occurrent in phenomenal consciousness.16 This is perhaps because it is so often assumed that only sensory states have phenomenal properties, since sensory states are plausibly taken to be phenomenally conscious insofar as they exist at all.17 But, as I have argued, beliefs also have phenomenal properties and they have them essentially. Moreover, it seems to be part of our commonsense concept of belief that one and the same token belief can be stored unconsciously in memory at one time and then brought to consciousness as an occurrent thought or judgement at another. Certainly, one can draw a distinction between the standing state and its occurrent manifestations in consciousness. But these are best understood as the components of

16 17

Notable exceptions include Burge (1997) and Rosenthal (1997). Burge and Rosenthal both suggest that a pain can persist without remaining phenomenally conscious because of distractions. However, we might equally say that pains are phenomenally conscious, but unattended, during periods of distraction.

35

a complex, evolving state.18 The standing state is a categorical state that has categorical properties that determine what it is like for the subject when it is occurrently manifested in phenomenal consciousness.19 These are the phenomenal properties of the belief. It may be objected that the proposed account of phenomenal properties is so permissive that it applies even to the subpersonal visual states of the blindsighter, thereby undermining the proposed explanation of his irrationality. After all, we can surely imagine counterfactual conditions in which the lesions in his visual cortex are healed and the very same subpersonal visual states play a role in determining what it is like for him to enjoy his visual experience. So, why don’t the subpersonal visual states of the blindsighter count as having phenomenal properties? The answer is that these states are not individuated by their potential contributions to phenomenal consciousness. Rather, they are individuated by their role in explaining the reliability of his verbal reports. Compare Davies’ (1989) example of a scenario in which states of tacit knowledge of syntax surface in phenomenal consciousness as distinctive itches or tickles. Such a case would not be one in which the tacit states are individuated by their potential contributions to phenomenal consciousness. Rather, it would be a case in which a systematic causal connection holds as a matter of contingent empirical fact. By contrast, a standing belief is individuated by its 18 19

Compare Goldie (2000, Ch.2) on the emotions. Stored beliefs are categorical states, to be contrasted with purely dispositional states, such as your belief (before you read this) that deer do not wear overcoats in the wild. Searle (1990) is wrong to claim that there is nothing for the ontology for an unconscious state to consist in except a disposition or its neural ground.

36

potential contributions to phenomenal consciousness. This explains why it is so plausible to regard the belief as a complex state comprising the standing state along with its occurrent manifestations in consciousness. 1.8. Blind-Reasoning Since blindsight is a specifically perceptual phenomenon, one might question whether it illustrates a connection between rationality and phenomenal consciousness that holds more generally. Does the connection hold only for the rationality of beliefs based on perceptual experience or does it apply also, say, to beliefs based on inferential reasoning? I will argue that the more general conclusion is supported by the conceivability of blind-reasoning, which is the analogue of blindsight in the case of reasoning.20 But first I want to consider a strategy for resisting the generalization move, which seeks to explain the connection between rationality and phenomenal consciousness in terms of a connection between reference and phenomenal consciousness. John Campbell (2002) appeals to the blindsight case in motivating a connection between reference and phenomenal consciousness. Campbell’s thesis is that grasping a perceptual-demonstrative concept of an object requires knowledge of reference in the form of acquaintance – that is to say, it requires phenomenally conscious attention to the object that is the reference of the concept. Intuitively, a patient with blindsight or super-blindsight is unable to grasp perceptual-demonstrative concepts 20

Boghossian (2003) uses the terminology of “blind reasoning” somewhat differently, to signify reasoning in the absence of critical reflection.

37

of an object in his blind field because does not know, in the relevant kind of way, which object is in question.21 He lacks the kind of “experiential highlighting” of the object that is required for knowledge by acquaintance. Therefore, he is unable to think perceptual-demonstrative thoughts about the object and, a fortiori, he is unable to think them rationally. It might seem, then, that we can explain the connection between rationality and phenomenal consciousness in terms of the connection between reference and phenomenal consciousness. But this way of explaining it certainly does not generalize across the board. After all, the connection between reference and phenomenal consciousness derives from distinctive features of perceptualdemonstrative concepts, which are not shared by concepts of other kinds. It is not true in general that there is a phenomenon of conscious attention to objects, properties and relations that plays a constitutive role in explaining our capacity for conceptual thought about them. There is, for example, no phenomenon of conscious attention to the semantic values for logical concepts that plays a constitutive role in explaining our capacity for logical reasoning. In any case, the connection between rationality and phenomenal consciousness cannot be explained in this way. Although the blindsighter is unable to think perceptual-demonstrative thoughts about objects in his blind field, he may be 21

Evans (1982) imposes a “knowing which” requirement, which he glosses in terms of discriminating knowledge. But this is a more demanding requirement, since conscious attention to an object is possible in the absence of an ability to discriminate it from others, as in Evans’ example of the indiscriminable spinning spheres. In my view, Kelly’s (2004) critique of Campbell trades on a confusion of these two “knowing which” requirements.

38

perfectly capable of thinking existential thoughts about them. All the same, it is not rational for a blindsighter to form the existential belief that there is an X or an O presented in the blind field solely on the basis of phenomenally unconscious visual information. Similarly, it is not rational for a Clairvoyant to form the belief that it is raining in Nepal without reliance on background information. To explain this, we need to appeal to a more general connection between rationality and phenomenal consciousness that is independent of Campbell’s thesis about the requirements for grasping perceptual-demonstrative concepts. Moreover, there are conceivable analogues of blindsight and clairvoyance in the case of inferential reasoning. For example, consider the following case: •

The Blind-Reasoning Case: While you are sleeping, a benevolent scientist implants a mechanism in your brain which computes the square roots of very large numbers. You do not notice the change, and your conscious mental life continues just as it would have done if the scientist had not intervened. But the method of forming beliefs on the basis of hunches about square roots is now reliable, though it would have been unreliable if the scientist had not intervened.

As in the case of clairvoyance and blindsight, the beliefs of the blind-reasoner are intuitively no more rational than guesses, despite their reliability. It is not rational for the blind-reasoner to form beliefs about square roots on the basis of the phenomenally unconscious computations of the implanted mechanism because the

39

relevant computational states do not have any phenomenal properties in virtue of which they might provide reasons for belief. Although blind-reasoning is conceivable, and presumably possible too, there are no good reasons to believe that it is actual. For instance, blind-reasoning is to be contrasted with familiar phenomenon of being suddenly arrested by the solution to a problem. This phenomenon comes in various different forms, but all of them can be accounted for in a way that brings out the contrast with blind-reasoning. In one kind of case, the subject thinks about the problem while he is sleeping and when he wakes up, he remembers the solution, but not how it was reached. In such a case, the reasoning that led to the solution may have been conscious even though the subject is unable to recover the reasoning upon later reflection. In another kind of case, the subject thinks about the problem while his attention is distracted by something else, for example grocery shopping. In this kind of case, the thinking may be conscious even if it does not occupy enough of the subject’s attention to figure in the foreground of conscious awareness. Hitting upon the solution may then engage the subject’s attention in such a way as to bring it back into the foreground of awareness. In other cases, there may even be a wholly unconscious process of inference, which results in the solution to a problem surfacing in phenomenal consciousness. But there remains a contrast to be drawn between this kind of unconscious inference and the imaginary case of blind-reasoning. After all, the propositional attitudes involved in unconscious inference have phenomenal properties in virtue of which they are available to be brought to consciousness under the right conditions. By

40

contrast, the computational states involved in blind-reasoning do not have any phenomenal properties and so they are not available to consciousness even in principle. For this reason, blind-reasoning is not really a case of genuine inference at all, but merely a case of computational information-processing. The contrast between inference and computation is explored in more detail in the next chapter.

41

CHAPTER TWO

THE RATIONAL AUTONOMY OF PERSONAL LEVEL PSYCHOLOGY

Chapter 2. The Rational Autonomy of Personal Level Psychology

2.1. Personal and Subpersonal Levels A core commitment of commonsense psychology is that thinkers have the capacity for performing deductive inferences. This is what underwrites the platitude that if a thinker believes that P and believes that if P, then Q, then other things being equal, he will also believe that Q. Any thinker who has the capacity to perform modus ponens and modus tollens inferences is causally sensitive to the fact that Q is entailed by the conjunction of P and if P, then Q. So, if he believes that P and believes that if P, then Q, then other things being equal, he will either infer the belief that Q or he will abandon one of his other beliefs. It is also a core commitment of current mainstream scientific psychology that thinking subjects have information-processing sub-systems that perform inferencelike computations.1 Consider Noam Chomsky’s (1965) proposal that our capacity to recognize and produce syntactically well-formed sentences of our native language is best explained in terms of the claim we have tacit knowledge of a syntactic theory for the language. According to the standard account of tacit knowledge, it consists in the existence of causal structure in our psychological states that mirrors the

1

Of course, computation is more like inference on classical than on connectionist models.

42

derivational structure of the theory and is sensitive to the storage of information stated by the theory.2 The idea is that when we recognize a sentence as syntactically well formed, this is best explained as the result of an inference-like computation that mirrors the steps in the derivation of its syntactic well-formedness in the abstract theory. For another influential example, consider David Marr’s (1982) proposal that visual experience of the environment is best explained in terms of a series of inference-like computations defined over representational states, such as the primal sketch and the 2.5-D sketch, which take as their input the representation of light intensity in the two-dimensional retinal image and yields as their output a threedimensional representation of shapes and their spatial organization in an objectcentred frame of reference. Inference and computation are in many ways alike. Both inference and computation are causal transitions between psychological states that have intentional contents. And both inference and computation are causal transitions between psychological states that are made in virtue of the intentional contents of those states. These points are sometimes disputed. Some philosophers, including John Searle (1990) and John McDowell (1994a), have claimed that the ascription of intentional content to subpersonal computational states in cognitive science is not to be taken seriously, but merely as a useful fiction. Others, including Fodor (1975), have claimed that while subpersonal computational states have intentional contents, the computational transitions between those states are performed not in virtue of their 2

See Evans (1981) and Davies (1987).

43

intentional contents, but rather in virtue of their syntactic properties. However, we can recognize a level of purely syntactic computation while also insisting on the indispensability of a notion of genuinely semantic computation. The appeal to semantic content plays an indispensable role in computational explanation when it is invoked to explain the semantic contents of intentional states, such as visual experience. Intentional data need intentional explanation.3 Even if these points are granted, however, there remains a crucial point of difference between inference and computation. Inferences are not always rational, but they are always evaluable for rationality. Mere computations, by contrast, may be evaluable for reliability, proper functioning, and so on, but they are not evaluable for rationality. It makes no sense to ask whether it is rational for a subject, or his visual system, to perform a computation that takes the retinal image as input and yields a three-dimensional representation of the environment as output. Nor does it make sense to ask whether it is rational for a subject, or his visuo-motor system, to perform a computation that fine-tunes the spatial parameters of a visually guided action. Moreover, the point generalizes. Inference is not the only kind of causal process that is evaluable for rationality. And there may be many different kinds of computational processes that are not evaluable for rationality. So, there is a more general distinction to be drawn here, which we can formulate in the terminology of personal and subpersonal levels. The paradigm examples of personal level processes 3

Peacocke (1994) argues the case for the semantic conception of computational explanation.

44

are evaluable for rationality. These include not only inference, but also forming beliefs on the basis of perceptual experience, deliberating about how to act, and so on. Not all personal level processes are evaluable for rationality – consider daydreaming and memory association. But these personal level processes involve states that are suited to figure in other rationally evaluable processes. By contrast, there are no subpersonal level processes that are evaluable for rationality. Since we are dealing with technical terms, we can simply stipulate that this is how the terminology of personal and subpersonal levels is to be understood. But this stipulation should not disguise the fact we have here a substantial datum that stands in need of explanation. Why is it that some psychological processes are evaluable for rationality, whereas others are not? The natural suggestion is that there is some relevant difference between the kinds of psychological states that figure in psychological processes of these respective kinds. The distinction in respect of rational rationality between personal and subpersonal level processes is to be explained in terms of the distinction between personal and subpersonal level states. On this way of understanding the distinction between personal and subpersonal levels, it is a distinction between different kinds of states and processes, not merely a difference between different kinds of descriptions of states and processes of one and the same kind. This is certainly how Daniel Dennett understood the distinction when he first introduced it: When we have said that a person has a sensation of pain, locates it and is prompted to react in a certain way, we have said all there is to say within the scope of this vocabulary. We can demand further explanation of how a

45

person happens to withdraw his hand from the hot stove, but…if we do this we must abandon the explanatory level of people and their sensations and activities and turn to the sub-personal level of brains and events in the nervous system. But when we abandon the personal level in a very real sense we abandon the subject matter of pains as well. (1969, p.93-4) As he puts it, when we abandon the personal level of description we thereby abandon its subject matter, too. Psychological states at personal and subpersonal levels are states of different kinds. But what exactly does the difference consist in? Here, Dennett’s choice of terminology is somewhat misleading. Given that pains are paradigm examples of personal level states, it is not plausible that personhood is the crux of the issue, at least in the Lockean sense in which a person is “a thinking intelligent being that has reason and reflection and can think of itself as itself, a thinking intelligent thing, in different times and places”. Moreover, the kind of reflective self-consciousness that is required for fully-fledged personhood is not a requirement for having psychological processes that are evaluable for rationality.4 Nor is it plausible that the distinction can be cashed out simply in terms of states that are properly ascribed to the whole organism as opposed to its parts or subsystems. After all, why can’t an organism be in a complex subpersonal state involving sufficiently many of its subsystems as to make it impossible to ascribe the state to any particular subsystem as opposed to the whole organism? In my view, there is an important distinction in the vicinity, but it does not correspond to the distinction between persons and their subsystems or even between organisms and their subsystems. Rather, the distinction is between psychological 4

See Chapter Three for more on this point.

46

states that figure within the subject’s point of view in virtue of their possession of phenomenal properties and psychological states that do not. So, we might even choose to re-label it as a distinction between “subjective” and “sub-subjective” states, although I will persist with the standard terminology. The proposal, in short, is this. Personal level states have phenomenal properties in virtue of which they are either phenomenally conscious or available to become so. Subpersonal states, on the other hand, do not have any phenomenal properties and so they are neither phenomenally conscious nor available to become so. How does this proposal explain the difference between personal and subpersonal level processes in respect of rational evaluability? The answer is that it does so by invoking Phenomenal Dependence. We can assume that a transition is evaluable for rationality only if it is based on psychological states of a kind that are relevant for determining what it is rational for the subject to believe. But, according to Phenomenal Dependence, what it is rational for the subject to believe depends solely on the phenomenal properties of his mental states. And, according to the proposed distinction, personal level states have phenomenal properties, whereas subpersonal states do not. Therefore, psychological processes are evaluable for rationality only if they are located at the personal, rather than the subpersonal, level. In Chapter One, we saw that Phenomenal Dependence provides the most fundamental articulation of the core insight that rationality is sensitive to the subject’s point of view. What we have seen here is that when we descend from the personal level to the subpersonal level, we leave behind the subject’s point of view and so we thereby leave behind the

47

space of reasons and rationality. In this way, Phenomenal Dependence supports a conception of personal level psychology as rationally autonomous from the subpersonal level. A number of influential theorists in philosophy and cognitive science have expressed skepticism concerning the theoretical significance of a distinction between psychological states that is defined in terms of consciousness. So, for example, in an early discussion of this issue, Jerry Fodor writes: That will depend upon whether there are generalizations which hold (just) for conscious mental states, and that depends in turn on whether the conscious states of an organism have more in common with one another than with the unconscious states of the nervous system of the organism. It is, in this sense, an open question whether conscious psychological states provide a natural domain for a theory… (1975, p.52, fn.19) Chomsky’s writings contain remarks in a similar vein, for example: It may be expected that conscious beliefs will form a scattered and probably uninteresting subpart of the full cognitive structure. (1976, p.163) The picture that emerges from these writers is one on which the personal level states constitute a mere fragment of a larger psychological structure, much of which remains permanently below the surface of phenomenal consciousness. On this view, there is not much of theoretical interest to the distinction between the parts of the cognitive structure that are available to surface in phenomenal consciousness and those that are not, aside from the fact that we happen to have introspective access to the parts that are available at the surface. Fodor sometimes puts the point in terms of the claim that personal level states fail to “exhaust a natural kind” (1987, p.25-6). According to Fodor’s (1975) account of the structure of science, psychology as a

48

whole is an autonomous special science, in the sense that the predicates figuring in psychological laws and explanations pick out natural kinds that are not also picked out by the predicates figuring in the laws of lower-level sciences. But he would deny that personal level psychology is autonomous from subpersonal level psychology, since he holds that personal level states do not exhaust a natural kind, but merely form a fragment of a larger natural kind that also includes subpersonal states. In his view, personal level psychology is not an autonomous theory with its own distinctive subject matter, but is merely a proto-theory to be refined and augmented by subpersonal level psychology. By contrast, I have been arguing, in effect, that personal level psychology is autonomous from subpersonal level psychology in the sense that it has its own distinctive subject matter. What is at issue in this debate, as Fodor observes, is whether personal level states have any more in common with one another than they do with subpersonal states. What we have observed is that personal level states do share a property that is lacked by subpersonal states – they are suited to figure in psychological processes that are evaluable for rationality. Moreover, it is in virtue of their phenomenal properties that they are suited to play this role. Other philosophers have emphasized the rational autonomy of personal level psychology – most notably, Donald Davidson (1980) and John McDowell (1985, 1994b). But they have not correctly identified its source. For example, consider the following passage in which McDowell elaborates Davidson’s claim that the

49

normative ideal of rationality plays a constitutive role in propositional attitude explanation: To recognize the ideal status of the constitutive concept [of rationality] is to appreciate that the concepts of the propositional attitudes have their home in explanations of a special sort: explanations in which things are made intelligible by being revealed to be, or to approximate to being, as they rationally ought to be. This is to be contrasted with a style of explanation in which one makes things intelligible by representing their coming into being as a particular instance of how things generally tend to happen. (1985, p.328) This passage contains a positive claim and a negative claim. The negative claim is that propositional attitude explanations are not law-like: they do not explain events by subsuming them under laws. But it is hard to see how this can be true, assuming that propositional attitude explanations are causal explanations, since causal explanations support counterfactuals and the relevant counterfactuals are explained by laws (in this case, ceteris paribus laws) that state how things generally tend to happen. However, McDowell’s view, and presumably Davidson’s too,5 is that while propositional attitude explanations cite causes, they are not causal explanations. As McDowell puts the point, “it would be a mere prejudice to suppose that citing a cause can be explanatory only by exploiting that possibility” (1985, p.335). This is where the positive claim comes in. McDowell’s contention is that propositional attitude explanations explain events by showing them to be rational or at least approximately so. On the face of it, however, this massively underestimates the scope of propositional attitude explanation. For example, we can explain 5

Davidson (1980) claims: (1) propositional attitude explanations cite reasons as causes; (2) every true singular causal statement is backed by a strict law (the Nomological Character of Causality); (3) there are no strict psychophysical or psychological laws (the Anomalism of the Mental).

50

someone’s belief in terms of wishful thinking, even though the belief is not formed in a way that is so irrational that it is not even approximately rational. And we can explain someone’s thought in terms of a chain of associations that are not even evaluable for rationality. In contrast, I am quite happy to allow that personal level explanation is a species of causal explanation that explains events by subsuming them under ceteris paribus laws. Certainly, explanations at the personal level often explain events in a way that shows them to be rational, or approximately rational, but this is not always the case. In my view, the fundamental issue is not the kind of intelligibility that is delivered by explanations at the personal level, but rather the metaphysical nature of psychological processes at the personal level. In particular, the kinds of states that figure in these processes have phenomenal properties, in virtue of which they are suited to play a role in psychological processes that are evaluable for rationality. The rational autonomy of personal level psychology has its source in the phenomenal properties of personal level states. 2.2. The Modular and the Conceptual We began with a distinction between inferences, which are evaluable for rationality, and mere computations, which are not. I have proposed that the distinction in respect of rational evaluability is to be explained by the fact that beliefs are personal level states, which have phenomenal properties in virtue of which they are available to phenomenal consciousness, whereas computational states are

51

subpersonal level states, which do not. However, I do not claim that this is the only theoretically significant distinction between beliefs and computational states. Beliefs are conceptual states and, as such, they essentially conform to constraints on functional integration and representational generality. By contrast, our paradigm computational states, such as Chomsky’s tacit knowledge of syntactic rules and Marr’s primal and 2.5-D sketch, are modular states, which do not essentially conform to these constraints. So, we have here a theoretically significant distinction which is drawn in broadly functional terms. But this is not in competition with my proposed distinction between personal and subpersonal levels. In fact, as I will argue, it is orthogonal, since it is not the case that all personal level states are conceptual states nor that all subpersonal states are modular states. The point is often made that it is an essential feature of conceptual states, such as beliefs, that they conform to constraints on representational generality and functional integration.6 Rather than attempting to articulate these constraints in full generality, it will suffice to provide the following substitution-instances: •

The Generality Constraint (GC): if the whole system tokens a state with the content that a is F and a state with the content that b is G, then it must have also the representational resources for tokening states with the contents that b is F and that a is G.

6

See Stich (1978) for the Integration Constraint and Evans (1982) for the Generality Constraint. See also Davies (1989) for an excellent discussion. Evans (1981) also claims that beliefs are functionally integrated with desires in the sense that they are “at the service of many distinct projects”.

52

•

The Integration Constraint (IC): if the whole system tokens a state with the content that a is F and a state with the content that a = b, then it must be disposed to token a state with the content that b is F.

But what is the explanatory connection between these constraints? Presumably, it is not just a metaphysical coincidence that beliefs essentially conform to both of them. In fact, we can explain GC in terms of IC, since conformity to GC is required for non-accidental conformity to IC. For example, suppose some system tokens states with the contents that a is F and that a = b; it will then be disposed to token a state with the content that b is F only if it has the representational resources to token a state with that content. So, it is plausible that if it is essential to a system that it should conform to IC, then it is also essential that it should conform to GC. But why is it an essential feature of conceptual states that they should conform to IC? This can be explained in terms of the requirements for possessing concepts in the first place. For example, it is a requirement for possessing the concept of identity that, other things equal, the thinker should be disposed to infer the conclusion that b is F from the premises that a is F and that a = b. As elsewhere, the disposition may be masked in such a way that it does not exhibit its characteristic response when its triggering conditions obtain. But such cases are to be explained as failures in performance, rather than conceptual competence. Beliefs and other conceptual states may be contrasted with computational states that are modular in the sense that they operate within computational modules. These modules are functionally encapsulated computational systems that are “domain

53

specific” in the sense that each is dedicated to solving its own distinctive computational problem, such as syntactic parsing, early visual processing, and so on.7 Therefore, it is not an essential feature of the computational states that operate within these modules that they should conform to constraints on representational generality or functional integration. If module M1 tokens a state with the content that a is F in solving computational problem P1 and module M2 tokens a state with the content that b is G in solving a quite different computational problem P2, there is no reason to suppose that the system as a whole will have the representational resources for tokening states with the contents that b is F or that a is G. After all, there may not be any computational problem whose solution requires the capacity for tokening such states. Similarly, if module M1 tokens a state with the content that a is F in solving computational problem P1 and module M2 tokens a state with the content that a = b in solving a quite different computational problem P2, there is no reason to suppose that the system as a whole will have any disposition to token a state with the content that b is F, even if it happens to have the relevant representational resources. Again, there may not be any computational problem whose solution requires having and exercising any such disposition. There is, then, a theoretically significant distinction between conceptual states, such as beliefs, and modular states, such as Chomsky’s tacit knowledge of syntactic rules and Marr’s primal and 2.5-D sketch. However, it would be a mistake to think

7

Fodor (1983) proposes various additional marks of modular systems, but functional encapsulation and domain specificity are the most important.

54

that we can invoke this distinction in explaining the difference between inference and computation in respect of rational evaluability. This is because there can be rationally evaluable processes at the personal level that are not conceptual, and there can be computational processes at the subpersonal level that are not modular. The personal-subpersonal distinction and the conceptual-modular distinction are simply orthogonal. First, then, there can be rationally evaluable processes at the personal level that are not conceptual. As Susan Hurley (2006) has argued, non-linguistic animals often exhibit “islands” of practical rationality, which are context-bound in the sense that they involve representational resources and functional dispositions that do not generalize in the way that would be required for conformity to the constraints on conceptual content. Hurley gives several compelling examples, including this one: For example, a primate could have reasons in social contexts that she cannot generalize to nonsocial but logically similar contexts. Suppose a monkey observes that conspecific A is dominant over B and that B is dominant over C and, never having observed A and C together, registers that A is dominant over C, and is able to use this information in instrumentally appropriate ways in relation to various goals. Nevertheless, she might be unable to generalize the ability to make transitive inferences to foraging contexts, such as: tree A has more fruit than tree B, which has more than tree C, so tree A has more fruit than tree C. In response, it may be objected that such cases involve failures of performance, rather than conceptual competence. But this objection needs to be supported by independent evidence that favours the attribution of the relevant conceptual capacities in the first place. As Hurley observes, linguistic capacities provide one kind of evidence for conceptual competence, but this kind of evidence is obviously

55

not available in the case of non-linguistic animals. In the absence of any such evidence, the objection simply rules out by stipulation what seems to be a live empirical possibility – namely, that animals can have rational capacities in the absence of full conceptual capacities. Second, there can be computational processes at the subpersonal level that are not modular. Fodor (1983) has proposed that there is a central system of highly integrated and recombinable subpersonal states, which is involved in solving computational problems that take as inputs the outputs from many distinct computational modules. For example, he argues that practical decision-making is underpinned by computations in the language of thought, which calculate the expected utility of a decision the basis of assignments of probabilities and preferences to its various possible outcomes. These calculations do not occur within domain specific and functionally isolated computational modules, since they draw on a wide range of preferences, perceptual information and background beliefs. However, in the normal case, they do not take place at the personal level in the form of explicit inferential reasoning. Rather, what we have here is a case of subpersonal computation that is not modular, but functionally integrated and representationally general. At the very least, this is a live empirical possibility which should not be ruled out by stipulation.8

8

Indeed, it can be argued (see 2.4 below) that inferential causal processes at the personal level are empirically explained by functionally isomorphic causal processes at the subpersonal level.

56

In conclusion, then, the personal level is not essentially conceptual and the subpersonal level is not essentially modular. Therefore, the theoretical significance of the distinction between modular and conceptual states does not eclipse the theoretical significance of the distinction I have drawn between personal and subpersonal level states in respect of availability to phenomenal consciousness. 2.3. Functional Zombies Could there be a functional zombie? And, if so, could there be a rational functional zombie? On the first question, Phenomenal Dependence is neutral. It is compatible with a wide range of views about the metaphysical nature of phenomenal consciousness, including functionalist views that rule out the metaphysical possibility of functional zombies. If there could not be a functional zombie, then of course there could not be rational functional zombie either. But as I will explain, there is an independent case to be made against ruling out the possibility of a functional zombie. However, even if we grant the possibility of a functional zombie, Phenomenal Dependence rules out the possibility of a rational functional zombie, since it is committed to the existence of a fundamental constitutive connection between rationality and phenomenal consciousness. A zombie is a creature whose mental states do not have any phenomenal properties, so his mental states are neither phenomenally conscious nor available to become so. A functional zombie is a zombie whose mental states play a causal role that is functionally isomorphic to the causal role played by the mental states of

57

creatures, like us, whose mental states have phenomenal properties in virtue of which they are either phenomenally conscious or available to become so. The existence of a functional isomorphism means that there is a one-to-one mapping of our phenomenal states onto non-phenomenal states of the zombie such that for every causal relation among our phenomenal states there is a causal relation among corresponding nonphenomenal states of the zombie. A functional zombie need not be functionally isomorphic with us “all the way down”, but only at the relatively superficial level of our phenomenal states. In other words, a functional zombie is only superficially functionally isomorphic with us. Ned Block (2002) gives the example of Commander Data, who is superficially functionally isomorphic with us, but physically very different – instead of neural tissue, his head is filled with silicon chips. On the face of it, we do not have sufficient reason to deny that the case is metaphysically possible. Moreover, when we consider the case, we do not have sufficient reason to rule out the possibility that Commander Data is a functional zombie. Block argues, plausibly in my view, that it remains an open question, epistemically speaking, whether or not he is conscious. On the one hand, the fact of his superficial functional isomorphism with us provides us with defeasible grounds for attributing consciousness to him. But on the other hand, these grounds are defeated by the consideration that he lacks any relevant kind of physical overlap with us. The lack of physical overlap does not give us sufficient grounds for attributing the absence of consciousness to Commander Data, since it would be chauvinistic to insist on these grounds alone that he is a functional zombie.

58

Nevertheless, it remains a live possibility that he is a functional zombie and it is not clear that anything could provide us with grounds to rule it out. Therefore, there is at least a prima facie case to be made against ruling out the metaphysical possibility of a functional zombie. When we consider the case of Commander Data, it remains an open question whether or not he is conscious. Likewise, I would suggest, it remains an open question whether or not he is rational. Certainly, if we were ever to encounter a creature like Commander Data, we would have good pragmatic reasons for treating him as if he were rational in explaining and predicting his behaviour. But if we cannot foreclose on the question of whether or not he is conscious, then by the same token, we cannot foreclose on the question of whether or not he is rational. If he is conscious, then certainly he is rational. But if he is not conscious, then intuitively he is not a rational subject at all, but rather a sophisticated kind of robot or computing machine. The intuition here is that only subjects can be rational, where being a subject is a matter of having a subjective perspective or point of view on the world. Robots and computers, however sophisticated, do not qualify as rational thinkers, since they do not have a subjective point of view on the world. And the very same point applies to the functional zombie. This intuition is best explained as follows. First, we can explain what it is to have a subjective point of view on the world by appealing to the phenomenal properties of the subject’s mental states. For it is only in virtue of having phenomenal properties that states count as figuring within the subject’s point of

59

view. And it is only in virtue of having states with phenomenal properties that an individual counts as a subject at all – that is, as an individual with a subjective point of view. Next, we can appeal to Phenomenal Dependence in articulating the core insight that rationality is sensitive to the subject’s point of view on the world. According to Phenomenal Dependence, what it is rational for a subject to believe depends on the phenomenal properties of his mental states. So, for a functional zombie whose mental states do not have any phenomenal properties, there is nothing that it is rational for him to believe. He is not a subject and so the requirements of rationality simply do not apply. In short, the functional zombie is neither rational, nor irrational, but arational. Notice that the explanation that I am offering here does not appeal to any purported connection between consciousness and intentionality. According to John Searle’s (1990) Connection Principle, psychological states have genuine, rather than merely “as if”, intentional content only if they are available to phenomenal consciousness. Since it is plausible that what it is rational for the subject to believe depends solely on mental states that have genuine intentional content, this provides a rival explanation of the connection between consciousness and rationality that rules out the rationality of a functional zombie. However, the Connection Principle itself is highly questionable because of its implications for the role of computational explanation within contemporary cognitive science. Computational explanations work by explaining certain cognitive capacities, such as object recognition and face recognition, in terms of computations defined over a series of unconscious

60

intentional states. Searle’s view, in effect, is that explanations that appeal to unconscious intentional states are no more than placeholders for explanations given in purely neurobiological terms. According to him, the appeal to intentional content plays no essential role in explanation at the subpersonal level and is at best a useful fiction, a heuristic that aids the search for genuine neurobiological explanations. However, instrumentalism about intentional explanation at the subpersonal level is no more plausible than it is at the personal level. For instance, it is very plausible that the appeal to intentional content is indispensable for explaining the relational properties of behaviour.9 The Connection Principle implies that intentional contents cannot be ascribed to the states of a functional zombie, since they are not available to phenomenal consciousness. But the ascription of intentional states is indispensable for explaining the relational properties of the functional zombie’s behaviour. If this is right, the ascription of intentional contents to deeply unconscious states is not a purely instrumentalist matter, but is answerable to perfectly objective constraints on explanation. Searle may be right that consciousness matters for rationality, but he is wrong about why it matters. 2.4. Analytical Functionalism and the Rationality Constraints Intuition and theory are in stable equilibrium on the question of whether a functional zombie would be rational – a functional zombie would be neither rational, nor irrational, but arational. But if a functional zombie is not rational, then it follows 9

For this argument, see Peacocke (1994). In fact, Peacocke goes further in arguing for the indispensability of an appeal to externally individuated content.

61

that he does not have any beliefs or other propositional attitudes, since there are rationality constraints on the propositional attitudes.10 I have suggested that the ascription of intentional states may be indispensable for explaining the relational properties of the functional zombie’s behaviour, but these states are not propositional attitudes. As I am using the terminology, propositional attitudes are intentional states that have conceptual contents, and it is partly in virtue of their conceptual contents that they are subject to rationality constraints. The rationality constraints on the propositional attitudes can be explained in terms of a corresponding rationality constraint on concept possession: •

The Rationality Constraint on Concept Possession: possessing a concept requires some disposition for rational use of the concept.

A thinker can token a propositional attitude with a certain conceptual content only if he possesses the concepts that compose that content. But, by the rationality constraint on concept possession, he possesses those concepts only if he has some disposition for using them rationally. So, a thinker can token propositional attitudes only if he has some basic rational dispositions. This is a very weak constraint. First, it allows for the possibility of plenty of local irrationality in the use of concepts, though only against the background of a certain amount of minimal rationality.11 Moreover, it does not imply any commitment to the stronger claim that necessary 10

I remain neutral on the question of whether it is part of the essential nature of propositional attitudes states to conform to the rationality constraints, or whether it is merely an a priori, but contingent matter, as many functionalists will claim. See, for example, Loar (1981, p.20-25). 11 For example, it allows for the possibility of delusions. Campbell (1999) argues that such cases present problems for the claim that there are rationality constraints on belief.

62

and sufficient conditions for possessing a concept can be given in terms of a noncircular specification of its rational conceptual role; nor to the claim that the content of a concept is to be explained in terms of its rational conceptual role, rather than vice versa. This constraint would be rejected only by proponents of the extreme view that there are no rational constraints on concept possession at all.12 In the eighties, there was a debate over the issue of whether or not analytical functionalism provides a strategy for naturalistic reduction of the propositional attitudes that is compatible with the existence of rationality constraints on the propositional attitudes. Brian Loar argued that, given a functionalist interpretation, the propositional attitudes can be reduced to the physical states whose causal role mirrors the rational role laid down by the rationality constraints: Can a functionalist theory…recognize the constitutive force of rationality? Of course it can, and must…. In particular, a system of physical state-types satisfies the constraints on rationality provided they are all related counterfactually as the theory says the beliefs to which they correspond ought rationally to be related. (1981, p.22) Nothing like the constraints of rationality shows up in physical theory; but counterfactual constraints that are isomorphic to rationality constraints may show up in the physical facts. (1981, p.23) In response, John McDowell conceded the point that if the propositional attitudes can be given a functionalist interpretation, then they can indeed be given a naturalistic reduction. But he argued instead that: …a functionalist interpretation of propositional attitudes is part of what the constitutive role of rationality excludes. (1985, p.325) 12

See Fodor (2004). One question here is whether there is any good motivation for rejecting the weak constraint once it is clearly distinguished from the stronger claims in the vicinity.

63

In what follows, I will argue that McDowell’s conclusion is correct, although the argument that I will be offering is quite different in character from his own. Analytical functionalism provides a strategy for naturalistic reduction of the propositional attitudes. The basic idea is that concepts of the propositional attitudes are in some sense implicitly defined by their role in a commonsense theory of propositional attitude psychology. There is scope here for disagreement about whether the theory is explicitly or tacitly known, whether it is known a priori or a posteriori, and so on. For our purposes, these issues can be left to one side. The key point is that the theory specifies causal roles for the various types of propositional attitudes to which it is committed, and whichever states in fact occupy these causal roles are fixed as the references of the concepts in terms of which the theory is formulated. Suppose that we come to discover that the causal roles of the propositional attitudes that are specified by our commonsense psychology are in fact occupied, more or less exactly, by certain types of physical states. Then, it will be claimed, these physical states are to be identified with our propositional attitudes. So, the template for naturalistic reduction is as follows: (1)

Propositional attitude P = the occupant of causal role R

(2)

The occupant of causal role R = physical state Q; therefore

(3)

Propositional attitude P = physical state Q.13

This leaves us with one crucial issue that remains to be settled. How is the causal role R of a propositional attitude P to be specified? 13

This template is adapted from Lewis (1972).

64

Analytical functionalism is analytically reductive in its ambitions. It aims to provide a reductive analysis of our commonsense psychological concepts in topicneutral terms. Therefore, the causal roles of the propositional attitudes are to be specified not in terms of the concepts of the commonsense psychological theory, but rather in topic-neutral terms, by means of the Ramsey-sentence for commonsense psychology. The Ramsey-sentence is derived from a statement of the theory by systematically replacing the distinctive vocabulary of commonsense psychology with variables bound by existential quantifiers. The result is a complex definite description that purports to specify the causal role of the propositional attitudes in topic-neutral terms. Now, the objection I want to develop is that a topic-neutral specification of the causal role of the propositional attitudes is not sufficiently discriminating to yield any unique specification of reference for the propositional attitudes. The Ramsey-sentence for commonsense psychology is satisfied not only by the propositional attitudes, but also by any system of states that plays a causal role that is isomorphic with the rational causal role played by the propositional attitudes. Consider the functional zombie. He does not have propositional attitudes because his states do not satisfy the rationality constraints. He has states that play a causal role that is functionally isomorphic with the rational causal role that our propositional attitudes play in us. In other words, there is a one-to-one mapping of our propositional attitudes on to states of the zombie such that for every rational causal relation involving our propositional attitudes there is a causal relation between corresponding states of the zombie. But this just goes to show that not every causal

65

role that is functionally isomorphic with a rational causal role is itself a rational causal role. In short, the functional zombie does not have propositional attitudes because his states do not play a rational causal role. Nevertheless, he satisfies the Ramsey-sentence for commonsense psychology because his states play a role that is functionally isomorphic with a rational causal role. Hence, the Ramsey-sentence for commonsense psychology fails to capture the rationality constraints on the propositional attitudes. In response, the analytical functionalist might invoke the point that a unique specification of reference for the Ramsey-sentence is achieved only relative to context.14 Like other definite descriptions, the Ramsey-sentence is a non-rigid designator whose reference may vary from context to context. Since our context is different from that of the functional zombie, the Ramsey-sentence may yield different specifications of reference for the concepts of personal-level states in the respective contexts. Therefore, our propositional attitudes may be very different in nature from those of the functional zombie. So, even if it is not an essential feature of the propositional attitudes of a functional zombie that they should conform to the rationality constraints, it may still be an essential feature of our propositional attitudes that they should do so. The suggestion, then, is that once we relativize to our own context, the Ramsey-sentence for commonsense psychology is sufficiently discriminating to yield a unique specification of reference for the propositional attitudes. Moreover, it does so in a way that is compatible with the claim that it is 14

See Lewis (1980). I am grateful to Ned Block for suggesting this kind of response.

66

essential to the kind of propositional attitudes we have they should conform to the rationality constraints. However, this relativization strategy also fails. For it is plausible that some rational causal processes at the personal level are empirically explained in terms of the way in which they are underpinned by functionally isomorphic causal processes at the subpersonal level.15 We can flesh this out in more detail by considering what it is to perform a deductive inference in virtue of its logical form. At a minimum, the inference must be causally explained as the exercise of a more general disposition to perform inferences which have the same logical form. And here, exercises of the same disposition must have some common causal explanation in terms of a common causal factor that grounds the disposition. However, much of our deductive inference is unreflective, in the sense that it is not mediated by any conscious appreciation of logical form.16 Therefore, the common causal factor that grounds the inferential disposition must be a state or processor that is located at the subpersonal level. And, presumably, the way it works is by mediating causal processes involving other subpersonal states that are systematically related to the personal level propositional attitudes involved in the inference. In other words, there is a one-to-one mapping from the personal level beliefs involved in an inference on to underlying subpersonal mental representations that preserve the causal relations between them.17 But it

15

See Fodor (1975), (1987) and especially Davies (1991). This premise is challenged by Brewer (1995). I discuss this in detail elsewhere. 17 Note that purely dispositional beliefs are not involved in inference before they are triggered, and so they do not need to be underpinned by mental representations. 16

67

would be a kind of category mistake, or a confusion of levels, to suppose that the personal level beliefs can be identified with the underlying subpersonal mental representations.18 The suggestion, in effect, is that we have “zombie systems” within us. We have subpersonal states that play a causal role which is functionally isomorphic with the rational causal role that is played by our personal level states. However, they do not themselves play a rational causal role because, like the states of the zombie, they do not have any phenomenal properties in virtue of which they are available to phenomenal consciousness. Therefore, the problem of uniquely specifying a reference for the Ramsey-sentence arises within our own context and so the appeal to context-relativity does not help. Relative to our own context, the Ramsey-sentence fails to specify a unique reference, since it is satisfied both by our personal level states, which are available to phenomenal consciousness, and by a subset of our subpersonal states, which are not. The analytically-reductive functionalist is therefore forced to claim either that it fails of reference altogether or that it has ambiguous reference.19 But neither option is plausible. In conclusion, the analytical functionalist strategy for naturalistic reduction of the propositional attitudes fails. The first premise of the argument for naturalistic reduction is false, at least when it is interpreted in the analytically-reductive manner

18

When Fodor (1987, Ch.1) states the Representational Theory of Mind, he does not formulate it as an identity thesis, but he says in a footnote, “you could do if you were so inclined” (fn.7, p.156). 19 Lewis (1972) claims that if the Ramsey-sentence fails to determine a unique reference, then it fails of reference altogether. But in his (1994), he concedes that it may have ambiguous reference.

68

proposed by analytical functionalism. The concepts of the propositional attitudes cannot be defined, or have their references fixed, by means of a topic-neutral specification of their causal role in the form of a Ramsey-sentence for commonsense psychology. Therefore, McDowell is correct in his assertion that “a functionalist interpretation of propositional attitudes is part of what the constitutive role of rationality excludes.” However, the argument that I have given is quite different in character from McDowell’s own. McDowell’s key claim is that it is impossible to capture the full normative force of the requirements of rationality, and deductive rationality in particular, in purely non-intentional terms: [The structure of deductive rationality] cannot be abstracted away from relations between contents, or forms of contents, in such a way that we might hope to find the abstracted structure exemplified in the interrelations among a system of items described in non-intentional terms. And in this case the claim is actually susceptible of something like proof. Someone who denied the claim would find it hard to explain how his position was consistent with the fact that there is no mechanical test for logical validity in general. (1985, p.327) Since analytical functionalism specifies the causal role of propositional attitudes in purely non-intentional terms, it follows that it fails to capture the full normative force of the requirements of rationality, as opposed to some “thinned-down surrogate” comprising only very minimal rational constraints. But why should that be considered a problem? Ordinary thinkers are only imperfectly rational after all. However, McDowell claims that analytical functionalism must capture the full normative force of the requirements of rationality if it is to account for the way in

69

which rationality functions as a constitutive ideal in propositional attitude explanation. I confess that I find this claim difficult to evaluate, but I suspect that it depends on any implausibly demanding conception of the rationality constraints on propositional attitude psychology. In any case, the overall strategic differences should now be clear. McDowell’s claim is that analytical functionalism fails to capture the full normative force of the requirements of rationality, whereas my claim is that it fails to capture even the most minimal rationality constraints. One final comment is in order. The argument that I have given rules out the strategy for naturalistic reduction of the propositional attitudes that is proposed by the analytical functionalist. However, there is nothing in this argument that rules out the possibility of naturalistic reduction in principle. Phenomenal Dependence is compatible with a range of views about the metaphysics of phenomenally conscious states and their phenomenal properties, including naturalistic views that identify them with neural states and properties. The significance of ruling out the analytical functionalist strategy is that it blocks an argument for naturalistic reduction of the personal to the subpersonal that can seem inevitable given the analytical functionalist strategy. Thus, it makes logical space for the plausible view that rational causal processes at the personal level are empirically explained in terms of the way in which they are underpinned by distinct, but functionally isomorphic, causal processes at the subpersonal level.

70

2.5. Normative Functionalism The core insight behind functionalism in the philosophy of mind is that it is an essential feature of the propositional attitudes that they are disposed to play a certain kind of causal role — and in particular, a more or less rational causal role. However, this functionalist insight can be divorced from the analytically reductive ambitions that are characteristic of analytical functionalism.20 I have argued that the concepts of the propositional attitudes cannot be defined, or have their references fixed, by means of a topic-neutral specification of their causal role in the form of a Ramseysentence for commonsense psychology. But that is not to say that the concepts of the propositional attitudes cannot be given an informative elucidation by specifying their rational causal role in the vocabulary of commonsense psychology. This makes room for a kind of normative functionalism, according to which it is an essential feature of the propositional attitudes, and other personal level states, that they are disposed to play a certain kind of rational causal role, which may be specified in terms of the concepts of commonsense psychology.21 However, these normative functional roles are not essential in the constitutive sense, since they are not fundamental, but are rather consequential on something else.22 The truth of normative functionalism can be explained a consequence of Phenomenal Dependence, which states that what it is rational for the subject to believe depends 20

Block’s (1978) problem of the specification of inputs and outputs raises an independent worry about the analytically reductive ambitions of analytical functionalism. 21 Peacocke’s (1992) account of the possession conditions for concepts is a normative functionalist account, since it specifies rational causal roles for attitudes with particular conceptual contents, but it does so in terms of the concepts of propositional attitude psychology. 22 See Fine (1995) for the distinction between constitutive and consequential senses of essence.

71

on the phenomenal properties of his mental states. After all, a subject’s belief is rational only if it is based on whatever makes it rational to have a belief of that kind. Therefore, a subject’s belief is rational only if it is based on the phenomenal properties of his mental states that make it rational for him to have a belief of that kind. So, it is in virtue of their phenomenal properties that a subject’s mental states provide his beliefs with a rational causal basis. The phenomenal properties of mental states are what ground and explain their rational causal roles. In Chapter One, we considered the view that mental states can be divided into phenomenal states on the one hand and propositional attitudes on the other. This bifurcationist view is sometimes developed in terms of the claim that functionalism is false as a theory of phenomenal states, but true as a theory of the propositional attitudes.23 But functionalism here can be understood either as analytical functionalism or as normative functionalism. Either way it is false, though for different reasons. Analytical functionalism is not only false as a theory of phenomenal states, but also false as a theory of propositional attitudes because the propositional attitudes are not uniquely identified by means of a specification of their causal roles in topic-neutral terms. Normative functionalism, on the other hand, is not only true as a theory of the propositional attitudes, but also true as a theory of phenomenal states. This is because phenomenal states can be uniquely identified by means of a specification of their rational causal roles in terms of the concepts of personal level psychology. 23

See Block (1978) for this suggestion.

72

Jaegwon Kim poses the following challenge concerning the unity of our conception of the mental: What do our pains and beliefs have in common in virtue of which they fall under the single category of “mental phenomena”? To the extent that we lack a satisfying answer to this question, we fail to have a unitary conception of what mentality consists in. (1996, p.23) We are now in a position to give the following response. Both pains and beliefs have phenomenal properties and it is in virtue of their phenomenal properties that they play their rational functional role. So, we do have a unitary conception of what mentality consists in, at least at the personal level. As for the nature of the relationship between mental states at personal and subpersonal levels, this will have to wait for another time.

73

CHAPTER THREE

RATIONALITY, REFLECTION AND EPISTEMIC RESPONSIBILITY

Chapter 3. Rationality, Reflection and Epistemic Responsibility

3.1. Why Care about Rationality? In Chapter One, we began with the core internalist insight that rationality is sensitive to the limitations of the subject’s point of view on the world. I argued that this core insight can be articulated in terms of a fundamental constitutive connection between rationality and phenomenal consciousness. In particular, I argued for Phenomenal Dependence, which states that what it is rational for a subject to believe depends on the phenomenal properties of his mental states. It is sometimes suggested that the core insight can be articulated in terms of a Reflective Access Requirement, which states that what it is rational for the subject to believe is accessible to the subject by reflection alone. So, for example, Laurence BonJour writes: The person for whom a belief is justified must indeed himself have access to the justification, for otherwise he has no reason for thinking that the belief is likely to be true, even though others may, and hence his acceptance of it is not justified. (1985, p.23) In this chapter, I will argue that the Reflective Access Requirement is true, though not fundamental. Rather, it is to be explained by appeal to Phenomenal Dependence, which provides a more fundamental characterization of the core insight. Indeed, I will argue that the Reflective Access Requirement plays no role in a constitutive

74

account of what makes it the case that a belief is rational, although it contributes to an account of why we care about the concept of rationality. I have claimed that the connection between rationality and phenomenal consciousness is constitutive and fundamental. Therefore, it cannot be given a reductive explanation in terms of any more fundamental constitutive truths, since there are none to be found. However, this is not to say that we can achieve no reflective understanding of the connection. Let it be granted, at least for the sake of argument, that there exists a fundamental constitutive connection between rationality and phenomenal consciousness. We can still go on to ask, from a reflective standpoint, why it is that we care about the notion of rationality. This question is best conceived as posing a kind of challenge.1 After all, we can surely define a notion – schmrationality – which is just like rationality except that it does not bear the proposed connection with phenomenal consciousness. Why not simply abandon our internalist notion of rationality in favour of this externalist replacement notion? What, if anything, would we lose by doing so? The challenge, in effect, is to provide a reflective understanding of the internalist notion of rationality by articulating the significance of its role within our cognitive lives. My strategy for responding to this challenge appeals to a distinction between rationality and epistemic responsibility. Rationality is a kind of normative status that even unreflective thinkers can achieve, whereas epistemic responsibility can be achieved only by reflective thinkers, who have the capacity to engage in critical 1

I am grateful to Jerry Fodor for persuading me of the importance of this challenge.

75

reasoning. Epistemic responsibility is plausibly central to our conception of what is distinctive of ourselves as persons in contrast with other rational but unreflective animals. As such, it is clearly something that we care very much about. My strategy, then, will be to argue that we cannot make sense of our conception of epistemic responsibility except in terms of the Reflective Access Requirement, which is explained in turn by Phenomenal Dependence. The claim is that rationality must be reflectively accessible if we are to make any sense of our conception of epistemic responsibility and that Phenomenal Dependence explains why it is. This is offered as an additional line of argument for Phenomenal Dependence, which both supplements the appeal to intuitions about cases and explains their wider theoretical significance. 3.2. Rationality and Epistemic Responsibility Let me begin by explaining the distinction between rationality and epistemic responsibility. The crux of the distinction is that epistemic responsibility requires the capacity for reflecting on the rational credentials of one’s beliefs, whereas rationality does not. The contrast between rational and irrational belief-forming methods can be applied even to purely unreflective creatures, such as animals and small children. But epistemic responsibility is a more demanding normative status. It is not merely a matter of unreflectively forming one’s beliefs in rational ways, but also requires one to deliberate about what it is rational for one to believe, to reflect on the rational standing of one’s beliefs and to revise or sustain them accordingly. That is to say, it requires engaging in what Tyler Burge calls “critical reasoning”:

76

Critical reasoning is reasoning that involves an ability to recognize and effectively employ reasonable criticism or support for reasons and reasoning. It is reasoning guided by an appreciation, use, and assessment of reasons and reasoning as such. As a critical reasoner, one not only reasons. One recognizes reasons as reasons. (1996, p.246) As Burge explains, the point of critical reasoning is that it enables reflective thinkers to acquire a kind of rational self-control over their own habits of reasoning. And epistemic responsibility is the kind of normative status that consists in having this kind of rational self-control. To achieve it, thinkers need not be constantly engaged in the activity of critical reasoning, so long as they are disposed to engage in critical reasoning when it is appropriate. I will not attempt to say anything more specific about when it is appropriate for reflective thinkers to engage in critical reasoning. The key point is simply that a capacity for reflection is a prerequisite for engaging in critical reasoning in the first place and hence for acquiring the kind of self-control that is characteristic of epistemically responsible thinkers. It is central to our conception of ourselves as persons that we hold ourselves to be epistemically responsible for our beliefs and practically responsible for our choices.2 We distinguish persons from other rational animals on the grounds that they can be held to be responsible for their beliefs and choices. This is evidenced by the fact that we do not normally regard it as appropriate to adopt reactive attitudes, such as resentment, towards animals that are not persons.3 There is no suggestion here that responsibility can be explained in terms of the appropriateness of adopting 2

Here, and in what follows, there are various points of contact with Burge’s (1998) discussion of the relationship between rationality and the first-person concept. 3 Strawson (1962) is the seminal discussion of the reactive attitudes.

77

reactive attitudes. Indeed, I take it that the correct direction of explanation is the other way around. But for current purposes, it suffices that it is appropriate to adopt reactive attitudes towards subjects only if they can be held responsible for their beliefs and choices. Persons can be held responsible for their beliefs and choices, whereas other merely rational animals cannot. And this is why it is appropriate to adopt reactive attitudes towards the former, but not the latter. But what explains why rational animals cannot be held responsible for their beliefs and choices? The most natural explanation is that a capacity for acknowledging one’s rational commitments is required before one can be held responsible for one’s beliefs and choices. In other words, what is required is a capacity not just for reasoning, but for critical reasoning — for recognizing one’s reasons as such.4 And it is plausible that persons are distinguished from other rational animals, at least in part, by virtue of their capacity for the kind of reflection that is involved in critical reasoning.5 Mere animals may be subject to the requirements of rationality and so they may be held to be irrational insofar as they fall short of satisfying those requirements. However, they cannot legitimately be held

4

The point is often made in terms of the claim that responsibility requires freedom, where this is construed in terms of a certain kind of responsiveness to reasons. For example, Susan Wolf (1990, p.94) suggests that “…the freedom necessary for responsibility consists in the ability (or freedom) to do the right thing for the right reasons.” But I insist that it requires more — namely, the ability to do the right thing for the right reasons recognized as such. The same point applies to McDowell’s (1994b, p.5) claim, in the context of his discussion of the Kantian notion of spontaneity, that “the space of reasons is the realm of freedom”. 5 The idea that personhood requires not only rationality, but also reflective self-consciousness, can be found in Locke’s (1689, II xxvii 9) famous definition of a person as “a thinking intelligent being that has reason and reflection and can think of itself as itself, a thinking intelligent thing, in different times and places.”

78

responsible for doing so, since they lack the reflective capacities for explicitly acknowledging what it is that rationality requires of them. Critical reasoning comes in degrees, some more reflective than others. In reasoning critically, thinkers must conceptualize at least some of the factors that are relevant in determining what it is rational for them to believe, but they may be sensitive to others without conceptualizing them. Fully reflective critical reasoning is self-conscious, in the sense that it involves an ability to self-ascribe one’s own psychological attitudes. But there can also be a more primitive kind of critical reasoning which involves reasoning about logical and evidential relations between contents in the absence of any capacity for self-ascription of attitudes.6 This more primitive form of critical reasoning will require some kind of systematic causal sensitivity to the differences between types of psychological attitudes, since they play different rational roles, but it does not require any capacity for conceptualizing those differences. Fully reflective critical reasoning, by contrast, involves reasoning about relations between contents which are simultaneously conceptualized as the contents of psychological attitudes that the subject self-ascribes. I have claimed that persons are distinguished from merely rational animals by the fact that they can be held responsible for their beliefs and choices in virtue of possessing the reflective capacities required for critical reasoning. But how reflective does a person’s capacity for critical reasoning need to be? A natural view is that, just as the notion of reflective critical reasoning comes in degrees, so do the notions of 6

This point is developed in Peacocke’s (1996) reply to Burge (1996).

79

responsibility and personhood. If a creature lacks conceptual self-consciousness, but is nevertheless capable of engaging in more primitive forms of critical reasoning, then to the extent that he appreciates his rational commitments, he may be capable of achieving a degree of responsibility for his beliefs and choices and hence a degree of personhood. However, a self-conscious creature with the capacity for fully reflective critical reasoning is able to achieve a more fully developed kind of responsibility for his beliefs and choices, since he has the conceptual resources for explicitly acknowledging his rational commitments as his own. For this reason, a selfconscious critical thinker has the capacity to become a more fully-fledged person. This is enough to motivate an interest in the fully reflective, self-conscious version of the capacity for critical reasoning, which will be the exclusive focus of the discussion to follow. 3.3. An Argument from Epistemic Responsibility Now we are in a position to see how our conception of epistemic responsibility supports the Reflective Access Requirement. An essential feature of epistemic responsibility in reflective critical reasoning is that it is possible for any sufficiently sophisticated thinker to achieve it, just on the basis of reflection alone, whatever his contingent circumstances and state of information. There may also be a more comprehensive notion of epistemic responsibility that requires thinkers to engage in various kinds of empirical enquiry with a view to improving their state of information. But relative to every possible state of information, there is a purely

80

reflective kind of epistemic responsibility that any sufficiently sophisticated thinker can achieve just on the basis of reflection alone. Necessarily, then, it is possible for any sufficiently sophisticated thinker to become epistemically responsible solely by engaging in reflective critical reasoning. But reflective critical reasoning can contribute this kind of positive normative status only if it is itself rational. So, necessarily, it is possible for any sufficiently sophisticated thinker to engage in reflective critical reasoning in a way that is rational – that is, to form rational reflective beliefs about what it is rational for him to believe. This is already a significant result.7 But I want to suggest that we can squeeze even more out of our commonsense conception of epistemic responsibility. In what follows, I argue that it is possible for any sufficiently sophisticated thinker to form reflective beliefs about what it is rational for him to believe that are not merely rational, but also true. Thinkers who engage in reflective critical reasoning are epistemically responsible not just for the rationality of the reflective process itself, but also for the rationality of the beliefs they are reflecting upon. Moreover, they are responsible for those beliefs in an especially intimate way. In the following passage, Burge makes the point suggestively: The [simple observational] model implies that we are in reviewing our reasons only derivatively responsible for objects of review, as one might be 7

Burge (1996, p.249) presents an argument of this kind as a transcendental deduction of the existence of a rational entitlement to form introspective beliefs about our own psychological states. But I am suggesting that it serves equally as a transcendental deduction of the existence of a rational entitlement to form beliefs about the rational status of our own psychological states.

81

responsible for the actions of one’s child or a dog…. But one is not epistemically responsible for the thoughts one reflects upon in critical reasoning the way one is responsible for something one owns or parents. One’s responsibility in reflecting on one’s thoughts is immediately for the whole point of view. (1996, p.258) The suggestion here is that the way in which one is responsible for one’s beliefs is very different from the way that one is responsible for something one owns or parents, such as a dog or a child. This is just a metaphor, of course, but it can be cashed out as follows. Your responsibility for a pet dog is constrained by the nature of your epistemic access to it, as well as by the nature of your control over it, and both are highly imperfect. In particular, your beliefs about the dog are subject to what we might call “brute errors” – they can be false, through no fault of your own, even when you reflect in a fully rational and attentive manner.8 For example, we can allow for the possibility of a case in which you are fully rational, but brutely mistaken, in believing that your dog will not chew the rug. In such a case, there will be a sense in which you cannot be held responsible if your dog does in fact chew the rug. You might be expected to buy a new rug, but you cannot be held rationally responsible, since you were fully rational in believing on reflection that the dog would not chew it up. Matters are quite different when we consider the nature of a thinker’s responsibility for his own beliefs and reasoning. We do not seem to allow for the possibility of a case in which a thinker forms his beliefs in an irrational way, but

8

Burge (1996) also uses the term “brute error”, but in a slightly different way. Here, brute errors are errors that cannot be explained in terms of deficiencies in rationality or attention.

82

cannot be held responsible for doing so because he is fully rational, though brutely mistaken, in his reflections on what it is rational for him to believe. What explains this is that brute errors are impossible with respect to reflection on what it is rational for one to believe. Any such error is to be explained in terms of deficiencies in either rationality or attention. Suppose we were to allow for the possibility of brute error, so that I might be fully rational and attentive, but brutely mistaken, in forming the reflective judgement that it is rational for me to believe that P. Then I would be fully rational in believing that it is rational for me to believe that P, even though it would not be rational for me to believe that P. If I were to form the belief that P, my belief would be irrational. And yet I could not be held responsible for this, since the irrationality of the belief would not be accessible from my own perspective in reflective critical reasoning. But this seems absurd. So, by reductio, brute errors are impossible with respect to reflection on what it is rational for one to believe. 3.4. The Reflective Access Requirement What we have established is that necessarily, it is possible in principle for any sufficiently sophisticated thinker to form reflective beliefs about what it is rational for him to believe that are not only rational, but also true. This is tantamount to the Reflective Access Requirement, which may be formulated as follows: •

The Reflective Access Requirement: necessarily, what it is rational for the subject to believe is accessible to the subject by reflection alone.

83

It may be helpful to unpack this a little. To access a fact is to form a rational belief that the fact obtains. But it is important that access is a factive state: it can be held only in relation to facts.9 To say that a fact is accessible is to say that it is possible to access it at least in principle. There is more to be said about the notion of “in principle” possibility, but we will return to this issue in due course. Finally, to access a fact by reflection alone is to access it on solely on the basis of introspection and a priori reasoning. The Reflective Access Requirement is to be distinguished from a stronger claim, which I call the Guidance-by-Access Requirement: •

The Guidance-by-Access Requirement: necessarily, a belief is rational only if it is formed on the basis of access to what it is rational for the subject to believe.10

The Reflective Access Requirement is a claim about which propositions it is rational for the subject to believe, whereas this is a claim about the rationality of beliefs that the subject has actually formed. The claim here is that if a belief is to be rational, then what it is rational for the subject to believe must be not merely accessible, but actually accessed, and the belief in question must be formed on the basis of such

9

I do not say that accessing a fact requires knowing it. Williamson (2000, Ch.4) has argued that no non-trivial conditions are “luminous” in the sense that the subject in a position to know that they obtain. His argument exploits a tension between luminosity and a reliability requirement for knowledge. However, rational access is not subject to a reliability requirement. 10 BonJour (1985, Ch.2) endorses a reflective version of this requirement, which plays a role in his argument against foundationalism. Brewer (1995, 1999) endorses a non-rflective version, which he deploys in his (1999, Ch.5) argument for conceptualism about the content of perceptual experience. I discuss Brewer’s view in more detail elsewhere.

84

access. So, this claim implies that the notion of access enters into a constitutive account of what makes it the case that a belief is rational. The Reflective Access Requirement, by contrast, depends on a fairly sharp distinction between the constitution and the accessibility of the facts concerning what it is rational for a subject to believe. After all, it reduces to the following conditional schema: •

Necessarily, if it is rational (/irrational) for the subject to believe that P, then the fact that it is rational (/irrational) for the subject to believe that P is accessible to the subject by reflection alone.

The facts about rationality that are mentioned on the left-hand-side of the conditional reappear explicitly on the right-hand-side, so the conditional itself does not provide any kind of analysis of the relevant facts. The Reflective Access Requirement should be understood not as part of a constitutive account of what it is rational for a subject to believe, but rather as a constraint of adequacy on such an account. A constitutive account of what it is rational for the subject to believe should be logically prior to the Reflective Access Requirement and should explain its truth as a consequence. In order to explain why the Reflective Access Requirement is true, we need to explain what makes it possible for the subject to access what it is rational for him to believe on the basis of reflection alone. In other words, what we need is an account of the grounds of the possibility of access. Here, though, we need to distinguish between two readings of the modal strength of the notion of accessibility in terms of which the Reflective Access Requirement is formulated. On the weak reading, what

85

is required is accessibility in principle, whereas on the strong reading, it is accessibility in practice. According to the strong reading of the Reflective Access Requirement, the possibility of access is grounded in the reflective capacities of rational subjects. Rational subjects are required to have the reflective capacities that make it possible in practice for them to access what it is rational for them to believe. In other words, the strong reading rules out the possibility of unreflective rationality – that is, rationality in the absence of any capacity for rational reflection. But this conflicts with the empirical evidence that three-year olds can know things without being able to know how they know them, and can form rational beliefs without being able to form rational second-order rational beliefs about what makes their first-order beliefs rational.11 One possible response here is to insist on a more demanding conception of rationality that requires the capacity for reflection. However, this would be to overlook the distinction I have drawn between rationality and epistemic responsibility. Rationality is a kind of normative status that even unreflective thinkers can achieve, whereas epistemic responsibility can be achieved only by reflective thinkers, who have the capacity to engage in critical reasoning. With this distinction in play, the more demanding conception of rationality loses any motivation it might otherwise have had. In any case, there is an independent argument against the demanding conception of rationality, which I will briefly sketch here. As we saw in Chapter Two, it is 11

For example, see Gopnik and Graf (1988).

86

plausible that there are rationality constraints on concept possession, such that possessing a concept requires some minimal rational capacity for using the concept in thought and judgement. So, the demanding conception of rationality implies that unreflective creatures cannot possess any concepts at all. For if rationality requires the capacity for reflection, then unreflective creatures do not satisfy even the most minimal rationality constraints on concept possession. This consequence is implausible because it rules out a model of our reflective capacities as an “overlay” on more basic conceptual capacities that we share with unreflective creatures. Why should we accept the overlay model?12 The answer is that it is supported by explanatory considerations, since it provides the resources for a plausible explanation of the development of reflective capacities. If unreflective creatures can share the concepts possessed by reflective creatures, then they can also draw on these conceptual resources in coming to acquire reflective capacities. For instance, they can redeploy the concepts employed in basic perceptual beliefs in self-ascribing the perceptual experiences which make it rational to form those beliefs. On this kind of account, the capacity for reflecting on the rationality of one’s perceptual beliefs depends asymmetrically on more basic rational but unreflective capacities for forming those very perceptual beliefs. By contrast, the demanding conception rules out the existence of any common conceptual basis which might be drawn on in making the developmental transition from not having the relevant reflective 12

The overlay model is challenged in Bermudez (2003) and Millar (2004). Related issues are at play in McDowell’s (1994b, p.64) denial of the claim that our perceptual experiences have contents that are shared by the perceptual experiences of creatures that do not possess any concepts.

87

capacities to having them. In effect, it forces us to explain the transition not in terms of conceptual enrichment, but rather in terms of radical conceptual change. I want to suggest that the correct understanding of the Reflective Access Requirement is weak, rather than strong. What it is rational for the subject to believe must be accessible by reflection alone, if not in practice, then at least in principle. We can gloss this in counterfactual terms – what it is rational for the subject to believe would be accessible to the subject given sufficient idealization of the subject’s reflective capacities. This is perfectly compatible with the existence of unreflective rationality. Given appropriate idealization, even creatures that lack reflective capacities would be capable of accessing what it is rational for them to believe. Concerns about trivialization may arise here, but they can be forestalled as follows. The first point is that when we evaluate the counterfactual, we must hold fixed the phenomenal properties of the subject’s mental states that determine what it is rational for him to believe.13 And the second point is that the relevant kind of idealization applies only to the subject’s reflective capacities. It does not trivialize the requirement because it is not the case that facts of every kind are made accessible by idealizing on the subject’s reflective capacities alone. In order to explain why the Reflective Access Requirement is true, we need to explain what makes it possible for the subject to access what it is rational for him to

13

Can idealization affect what it is rational for the subject to believe? A plausible view is that idealization does not affect what it is rational for the subject to believe, but only whether he is able to conform to its requirements. After all, rationality is an ideal that imperfect thinkers realize to a greater or lesser degree.

88

believe on the basis of reflection alone. On the strong reading, the possibility of access is grounded in the reflective capacities of the rational subject. But on the weak reading, we need a different account. In what follows, I will argue that Phenomenal Dependence plays an indispensable role in explaining why the weak reading of the Reflective Access Requirement is true. The possibility of access is grounded by the phenomenal properties of the subject’s mental states, which determine what it is rational for him to believe. A sufficiently reflective subject has access to what it is rational for him to believe because he has access to the phenomenal properties of his mental states. 3.5. Explaining the Reflective Access Requirement The explanation that I am proposing depends on getting the right account of the relation between phenomenal judgements and the phenomenal properties of the mental states they concern. Two points are relevant here. First, subjects can have phenomenally conscious states without forming any phenomenal judgements about them and without even having the capacity for doing so.14 Second, the phenomenal properties of phenomenally conscious states typically play a role in the causal explanation of phenomenal judgements about those states.15 Thus, phenomenal properties are not constitutively dependent on phenomenal judgements. However, there are limits on the possibility of error with respect to a subject’s phenomenal judgements about the phenomenal properties of his phenomenally 14 15

Pace Rosenthal’s (1997) higher-order thought theory of phenomenal consciousness. For this point, see Peacocke (1998) and Campbell (2002, Ch.7).

89

conscious states. In particular, brute errors are impossible. Whenever a subject forms a false judgement about the phenomenal properties of his phenomenally conscious states at that time, he is either less than fully attentive or less than fully rational. This allows for cases in which a subject’s attention is primed in such a way as to lead him misidentify his sensations, as when the subject is taken to the torture room where an ice cube is placed on the back of his neck and he mistakes the icy cold sensation for an electric shock. It also allows for cases in which subjects are lured into misidentifying their sensations by various kinds of fallacious reasoning. But if a subject, who is fully attentive and fully rational, forms a belief that he is in a phenomenally conscious state with certain phenomenal properties, then he is; and if he forms a belief that he is not in a phenomenally conscious state with certain phenomenal properties, then he is not. There are no corresponding limits on the possibility of brute error concerning perceptual judgements. Consider a fully attentive and rational subject who is forming perceptual judgements about the external world on the basis of his perceptual experience, without reliance on any background beliefs. If he judges that there is a dagger before him, it cannot be inferred that there is a dagger before him, since he may be hallucinating. Therefore, perceptual judgements are not immune from brute error – they can be false even when the subject is fully attentive and fully rational. It is a distinctive feature of phenomenal judgements that they are immune from brute error. What explains this? According to a judgement-dependent conception, a judgement to the effect that one is in a state with certain phenomenal properties can

90

make it true that one is in a state with those properties.16 But we have already seen grounds for denying that phenomenal properties are constitutively dependent on phenomenal judgements. So, there is a challenge to explain why phenomenal judgements are immune from brute error, but without appealing to some kind of mysterious super-mechanism that is just like the mechanisms involved in perception except that it never malfunctions.17 The special feature of phenomenal judgement is this. When it is made in a way that is fully attentive and fully rational, it is based on the phenomenal properties of the phenomenally conscious state that it is about. The reflective judgement does not make it true that the subject is in that state; rather, it is made true by the occurrence of the phenomenally conscious state on which it is based. By contrast, a perceptual judgement is not made true by the occurrence of the perceptual experience on which it is based. Rather, it is made true by the fact that the world is the way it is represented as being by the perceptual experience on which it is based. Hence, there is scope for brute error in the perceptual case because it is compatible with the subject’s rationality and attentiveness that his perceptual experience should fail to represent the world veridically. But there is no analogous scope for brute error in the case of phenomenal judgement. There is a further, related difference between perceptual and phenomenal judgements. The possibility of brute error in the perceptual case means that

16 17

See Wright (1989) for a version of the judgement-dependent conception. Martin (2006) presses this challenge.

91

perceptual experiences only provide defeasible reasons for perceptual judgements. These reasons may be defeated by background beliefs to the effect that the perceptual experiences in question fail to represent the world veridically. So, a perceptual experience does not suffice to make it rational for a fully rational and attentive subject to form a perceptual judgement, since he may have defeating reasons. By contrast, brute error is impossible in the case of phenomenal judgement, and so phenomenal states provide conclusive reasons for phenomenal judgements. There are no background beliefs that can defeat a reason of this kind, at least in the case of a fully rational and attentive subject. Therefore, having a phenomenally conscious state with certain phenomenal properties is sufficient to make it rational for a fully attentive and reflective subject to form the phenomenal judgement that he is in that state. What we have shown so far is that it is possible, at least in principle, for any sufficiently reflective subject to form a phenomenal judgement about the phenomenal properties of his mental states that is not only rational, but also true. In other words, the phenomenal properties of a subject’s phenomenally conscious states are accessible in principle by reflection alone. Moreover, any state that has phenomenal properties is available to phenomenal consciousness and so it is possible, at least in principle, for the subject to bring it into phenomenal consciousness. So, we have the following: (1)

The phenomenal properties of a subject’s mental states are accessible in principle by reflection alone

92

Phenomenal Dependence states that: (2)

What it is rational for the subject to believe depends on the phenomenal properties of his mental states

If we also assume the plausible claim that: (3)

The rational significance of the phenomenal properties of a subject’s mental states is accessible by reflection alone

Then we can infer: (4)

What it is rational for the subject to believe is accessible in principle by reflection alone

And this, of course, is exactly what is stated by the Reflective Access Requirement. Let us take stock. Phenomenal Dependence posits a fundamental constitutive connection between rationality and phenomenal consciousness. Therefore, it is subject to the challenge of explaining why we should care about rationality, rather than schmrationality, which is just like rationality except that it does not bear the proposed connection with phenomenal consciousness. I have been arguing that it makes sense of our conception of epistemic responsibility in reflective critical reasoning by playing a role in explaining the truth of the Reflective Access Requirement. It is therefore crucial that Phenomenal Dependence plays an indispensable role in the explanation. The argument will be undercut if it turns out that the Reflective Access Requirement can be explained instead by appealing to the kind of functionalist alternative to Phenomenal Dependence that was discussed in Chapter One:

93

•

The Functionalist Constraint: what it is rational for the subject to believe depends on functional properties of his mental states, i.e. the second-order properties of having first-order properties that play functional role R.

It might be thought that we can explain the Reflective Access Requirement as a shallow, definitional truth by building accessibility by reflection alone into the specification of functional role R. But this proposal is faced with a dilemma. If reflective access is understood as a kind of rational, higher-order belief, then the Functionalist Constraint is merely a redescription of Phenomenal Dependence, since only phenomenal properties are suited to occupy rational functional roles. But if it is not, then the Functionalist Constraint fails to explain the Reflective Access Requirement, since the notion of reflective access that figures in the latter is rational, higher-order belief. The Functionalist Constraint would be satisfied by a functional zombie. But as we saw in Chapter Two, a functional zombie is neither rational, nor irrational, but arational, and so he fails to satisfy even the most minimal rationality constraints for having beliefs and other propositional attitudes. He does not engage in reflective critical reasoning and so he does not achieve the kind of epistemic responsibility that is required for personhood. In response, it may be said that the functional zombie is not only schmrational, but also a schmepistemically schmresponsible schmerson. Personally, I have no wish to withhold any such compliment from the functional zombie. My aim here has been to understand why we care about the notion of rationality by drawing out its connections with other important aspects of our conceptual scheme, such as the kind

94

of epistemic responsibility in reflective critical reasoning that is required for personhood. The aim has been to show that the way in which we think about rationality is not some isolated peculiarity, but rather plays a crucial and pivotal role in our conceptual scheme. If we were all functional zombies, then perhaps there would be no significant use for these concepts. Fortunately, however, we are not functional zombies. It may be that our concepts are significant for us because of contingent features of our psychological make-up, but they are none the less significant for that. 3.6. The Internalism/Externalism Debate In this chapter, I have proposed two very different kinds of argument for the Reflective Access Requirement. First, there was a broadly transcendental argument to the effect that the Reflective Access Requirement must be true if we are to make any sense of our conception of epistemic responsibility in reflective critical reasoning. And second, there was an explanatory argument that derived the Reflective Access Requirement from Phenomenal Dependence along with further plausible assumptions. We need to distinguish carefully between these different kinds of argument because it is one thing to argue that something must be true and quite another to explain what makes it true.18

18

This distinction is relevant for assessing Burge’s (1996) claim that the source of our entitlement to introspective self-ascription is its role in the reflective critical reasoning that it makes possible. In my view, Burge’s transcendental argument plausibly establishes the existence of an entitlement to introspective self-knowledge, but it does not explain the source of that entitlement. As Peacocke observes, “an entitlement to self-knowledge is one of the sources of critical reasoning, rather than vice versa” (1996, p.128).

95

The function of the transcendental argument is principally to explain the significance of the concept of rationality by articulating its pivotal role in our conceptual scheme. The function of the explanatory argument, by contrast, is to articulate the way in which the rationality of a belief must be constituted so as to play this role. Notice that the notion of epistemic responsibility has not figured in the explanatory argument for the Reflective Access Requirement, but only in the transcendental argument. I do not appeal to the notion of epistemic responsibility in explaining what makes it the case that a belief is rational. In my view, this would be to reverse the correct direction of explanation. Epistemic responsibility is to be explained in terms of rationality, but not vice versa. My overall argumentative strategy can be broken down into three stages. It begins at stage one with the development of a theory of rationality that is motivated by reflection on intuitive judgements about cases. At stage two, this theory of rationality is invoked in explaining the Reflective Access Requirement. And at stage three, the Reflective Access Requirement is invoked in explaining our conception of epistemic responsibility in reflective critical reasoning. This may be usefully contrasted with the strategy that is pursued by BonJour in his classic (1985) development of an internalist theory of rationality. At stage one, he proposes a broadly deontological conception of rationality, according to which epistemic responsibility is a requirement for rationality: The idea…of being epistemically responsible in one’s believings, is the core of the notion of epistemic justification. (1985, p.8)

96

At stage two, he attempts to derive the Reflective Access Requirement from this broadly deontological conception of rationality: If a given putative knower is himself to be epistemically responsible in accepting beliefs in virtue of their meeting the standards of a given epistemic account, then it seems to follow that an appropriate metajustification must, in principle at least, be available to him. (1985, p.10) And finally, at stage three, he attempts to explain our intuitive judgements about cases, such as the case of the clairvoyant, Norman, in a way that makes essential appeal to the Reflective Access Requirement: Norman’s acceptance of the belief…is epistemically irrational and irresponsible, and thereby unjustified, whether or not he believes himself to have clairvoyant power, so long as he has no justification for such a belief. Part of one’s epistemic duty is to reflect critically upon one’s beliefs, and such critical reflection precludes believing things to which one has, to one’s knowledge, no reliable means of epistemic access. (1985, p.42) In short, BonJour reverses the direction of explanation that I have been recommending. As a result, he is faced with very serious objections at every stage. To begin at stage one, rationality does not require epistemic responsibility. As I have emphasized several times now, rationality is a kind of normative status that even unreflective thinkers can achieve, whereas epistemic responsibility can be achieved only by reflective thinkers, who have the capacity to engage in critical reasoning. Here, it is important to resist the following seductive, but ultimately fallacious, line of reasoning. Let us define a normative concept to be one whose application entails certain ought-claims. Rationality, then, is a normative concept, since it is a truism that subjects ought to do what rationality requires of them. These ought-claims express rational obligations or duties and it is the thinker’s epistemic

97

responsibility to discharge them. However, a thinker cannot be held responsible for discharging his obligations and duties unless he is capable of acknowledging the fact that they bind him. So, the requirements of rationality apply only to thinkers who have the reflective capacities for acknowledging that they apply, while satisfying them is a matter of discharging one’s epistemic duties in an epistemically responsible manner. This argument trades on an equivocation between two senses of the word “ought”. While there is indeed a “thin” sense which applies in all areas of normative discourse, it is to be distinguished from a “thick” sense which applies only in connection with a family of deontological concepts, such as duty, responsibility, praise and blame. The argument threatens to reduce all normative concepts to deontological concepts, but there are certain normative concepts which are clearly not deontological. For example, it is plausible that the visual system is constrained by the norm of veridically representing the environment of the perceiver. If the visual system malfunctions in such a way as to misrepresent the environment, then there is a clear sense in which it is not functioning as it ought to. The visual system is failing to satisfy norms by which it is constrained, but it would be absurd to gloss this in terms of an irresponsible failure to discharge its duties. Therefore, it would be a mistake to assume that rationality is a deontological concept simply on the grounds that it is a normative concept. Even if we grant BonJour’s deontological conception of rationality, there remains a serious problem at stage two. We may assume that a thinker cannot be

98

held responsible for discharging his obligations and duties unless he is capable of acknowledging the fact that he is bound by them. So, if we conceive rationality in terms of epistemic responsibility, then a thinker cannot be held responsible for discharging his epistemic duties unless he is capable of acknowledging the fact that he is bound by them. But this falls short of the Reflective Access Requirement, as Goldman (2001) has observed, since it fails to motivate the epistemic restriction in terms of which the requirement is formulated — namely, that what it is rational for the subject to believe must be accessible by reflection alone. The problem is that BonJour has reversed the correct direction of explanation. The Reflective Access Requirement is not derived from a prior and independent conception of epistemic responsibility. On the contrary, our conception of epistemic responsibility in reflective critical reasoning is itself informed by the Reflective Access Requirement. Finally, at stage three, it is a mistake to suppose that our intuitions about cases, such as the clairvoyance case, are to be explained in a way that makes essential appeal to the Reflective Access Requirement. BonJour’s suggestion is that Norman’s clairvoyant beliefs are irrational because it is epistemically irresponsible of him to form them in ways that, on reflection, it is not rational for him to believe to be rational. However, the irrationality of clairvoyance applies not only to reflective subjects, but also to unreflective subjects who are incapable of engaging in the kind of reflective critical reasoning that must be in place before epistemic responsibility can become an issue.

99

It is widely held that BonJour’s deontological conception of rationality is implausibly demanding, since it rules out the possibility of unreflective rationality. One common line of response endorses a two-level account according to which rationality requires either epistemic responsibility or a mere absence of epistemic irresponsibility.19 The idea here is that the second disjunct makes room for the possibility that unreflective creatures may form rational beliefs in a way that is neither epistemically responsible nor irresponsible, since they lack the reflective capacities required for these kinds of epistemic evaluation to get a grip. However, a mere absence of epistemic irresponsibility is not sufficient for rationality, since a creature may be programmed by nature or nurture to employ irrational beliefforming methods, such as gambler’s fallacy, without being epistemically irresponsible in so doing.20 Moreover, I doubt that any purely externalist account can provide a satisfactory account of the irrationality of clairvoyance at the unreflective level.21 I want to recommend a different approach. We need to begin with a distinctively internalist theory of unreflective rationality that is motivated by reflection on intuitions about cases, such as the clairvoyance case. Then we can invoke this internalist theory of unreflective rationality in explaining our conception of epistemic

19

See Boghossian (2001) and Wright (2001) for this suggestion. See Pryor (2001), and references therein, for examples of this kind. 21 Ernest Sosa (1991) proposes a distinction between apt belief, which is formed on the basis of intellectual virtue (roughly, a belief-forming disposition that is reliable in the circumstances) and reflectively justified belief, which is part of a coherent higher-order perspective on one’s beliefs. Sosa’s view is that clairvoyant beliefs are apt, but not reflectively justified. However, this fails to capture the irrationality of unreflectively formed clairvoyant beliefs. 20

100

responsibility in reflective critical reasoning. Paul Boghossian sums up a widely held view when he writes, “The core distinction between externalism and internalism in the theory of justification is properly characterized in terms of the notion of epistemic responsibility” (2001, p.41). However, I am claiming that the core distinction applies at the more fundamental level of unreflective reasoning, where considerations of epistemic responsibility and irresponsibility fail to get a grip. Moreover, our conception of epistemic responsibility in reflective critical reasoning is informed by an internalist conception of rationality which applies at this more fundamental, unreflective level. The debate between internalism and externalism about rationality tends to oscillate between two extremes. On the one hand, internalist theories tend to impose implausibly demanding requirements on the reflective capacities of rational subjects, thereby ruling out the possibility of unreflective rationality. On the other hand, externalist theories tend to make room for the possibility of unreflective rationality only by obscuring the distinction between rationality and reliability. This dialectical situation is exacerbated by the fact that we lose sight of the significance of this distinction insofar as we restrict our focus to cases of unreflective rationality alone. The point and interest of the concept of rationality fully emerges only in the context of its role in an epistemically responsible practice of reflective critical reasoning. However, it does not follow that the rationality of a belief is constituted in a way that makes essential reference to the subject’s reflective capacities. In order to reach a satisfactory resolution of the debate between internalism and externalism, we need a

101

theory of rationality that can make sense of the possibility of unreflective rationality without losing the connection with the kinds of reflective critical reasoning which illuminate the significance of the concept of rationality. The theory I am proposing is intended to occupy this elusive middle ground.

102

CHAPTER FOUR

RATIONALITY AND THE PROBLEM OF TRUTH-CONDUCIVENESS

Chapter 4. Rationality and the Problem of Truth-Conduciveness

4.1. The Problem of Truth-Conduciveness It is often claimed that truth is a constitutive aim of belief. This claim can be understood in various different ways, not all of which are plausible. However, there is a way of understanding it on which it is extremely plausible – that is, as a distinctive kind of normative claim.1 To say that belief has an aim is to say that it is subject to normative assessment. To say that truth is an aim of belief is to say that truth sets a normative standard of correctness for belief, which it satisfies if and only if it is true. And to say that truth is a constitutive aim of belief is to say that it is a fundamental part of the essential nature of belief that it is subject to the normative standard of truth. So understood, the claim that truth is a constitutive aim of belief is both true and important. Other attitudes, such as imagining or supposing for the sake of argument, do not have truth as a constitutive aim, since there is nothing normatively defective about them in the case where they are not true. It is a distinctive feature of the attitude of believing that there is a normative standard of correctness which it

1

See Wedgwood (2002a) for this interpretation. Velleman (2000) offers a different approach.

103

satisfies if and only if it is true. Moreover, this is essential to the nature of belief — it is part of what makes belief the kind of state it is. Truth is not the only normative standard for belief.2 For example, rationality is a further normative standard, which a belief can fail to satisfy even if it is true. Moreover, it is plausibly an essential feature of belief that it is subject to the normative standard of rationality. However, rationality is not a constitutive aim of belief because it is not sufficiently fundamental, since its status depends on its relationship to the more fundamental aim of truth.3 This is not to say that the norm of rationality can be derived from the norm of truth, plus other non-normative truths. Rather, the suggestion is that a full explanation of the norm of rationality requires appealing to its relations to the norm of truth, whereas the converse claim does not hold, and in this sense the norm of truth is more fundamental in the order of philosophical explanation. Why think that the norm of rationality must be explained by its relation to a more fundamental norm of truth? This can be seen by reflecting on the question: Why suppose that rationality is a norm or aim for belief at all? What makes it the case that the rationality of our beliefs is something worth caring about? Surely, this is not just a brute fact. Rather, we need to explain the normative status of rationality by appealing to its relation to the more fundamental norm of truth. In particular, we

2

Perhaps truth is not the only constitutive aim of belief. For all I have said, belief may have additional constitutive aims, such as power. See Goldman (1986) for more on the notion of power. 3 Compare Fine’s (1995) distinction between constitutive and consequential notions of essence. The basic idea is that consequential essence is explanatorily dependent on constitutive essence.

104

need to explain how it is that forming beliefs in rational ways is, in some sense, conducive to the aim of believing the truth.4 If, on reflection, we are unable to make any sense of the claim that rationality is relevantly truth-conducive, then we will have failed to make any sense of why rationality is worth caring about. This is a point that BonJour has insisted on: 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. It is only if we have some reason for thinking that epistemic justification provides a path to truth that we as cognitive beings have any motive for preferring epistemically justified beliefs to epistemically unjustified ones. (1985, p.8) In sum, it is a constraint of adequacy on any theory of rationality that it should explain how it is that forming beliefs in rational ways is conducive to the aim of believing the truth. This is the Problem of Truth-Conduciveness. 4.2. The Reliabilist Strategies One of the attractions of reliabilist theories of rationality is that they provide a straightforward solution to the Problem of Truth-Conduciveness. This applies not only to pure reliabilism, on which reliability is both necessary and sufficient for rationality, but also to moderate reliabilism, on which reliability is necessary but not sufficient for rationality. After all, if reliability is necessary for rationality, whether

4

An explanation that takes this form will only be partial, since there may be non-rational ways of forming beliefs that are, in some relevantly similar sense, conducive to the aim of belief. To complete the explanation, we need to motivate an interest in the specifically rationally truthconducive ways. I hope to have made some progress on this task in Chapter Three.

105

or not it is sufficient, then by forming our beliefs rationally we thereby form them reliably, and reliably formed beliefs tend to be true. In Chapter One, we saw that the Brain in a Vat (BIV) case provides an intuitive counterexample to the claim that reliability is necessary for rationality. However, the proponent of a reliabilist theory of rationality might respond as follows: “The BIV is not epistemically rational, but merely blameless. For given his circumstances, his belief-forming methods are not truth-conducive. Of course, he cannot legitimately be blamed or held responsible for being in these circumstances. But this kind of blamelessness is not sufficient for rationality, since genuinely rational methods also have to be truth-conducive in some way.” It is not clear that this response succeeds in accommodating the relevant intuitions. After all, there is an intuitive contrast to be drawn between (i) the envatted counterparts of normal rational subjects and (ii) the envatted counterparts of subjects who are irrational, but still blameless, because their irrationality has been pre-programmed by nature or nurture in such a way that they cannot legitimately be held accountable.5 In other words, our intuitions about the BIV are sensitive to the distinction between genuine epistemic rationality and mere blamelessness. Nevertheless, the reliabilist response raises a genuine challenge. Without some alternative solution to the Problem of Truth-Conduciveness, we will have failed to reach any theoretical understanding of the intuitive distinction between rationality and mere blamelessness. And if no such account is forthcoming, then the reliabilist is 5

For this point, see Pryor (2001, p.117).

106

in a position to make a plausible case for abandoning the intuitions that motivate this distinction. A rather different strategy for responding to the Problem of TruthConduciveness is suggested by Ernest Sosa. Sosa (2003, Ch.9) distinguishes between two notions of reliability – aptness and adroitness. A belief-forming method is apt if and only if it is reliable in the circumstances in which it is employed, whereas a belief-forming method is adroit if and only if it is reliable in our actual circumstances. In other words, aptness is de facto reliability, whereas adroitness is rigidified reliability. The significance of this contrast emerges when we consider non-actual, but possible, counterfactual scenarios – such as the BIV case and the Clairvoyance case. The belief-forming methods of the BIV are reliable in our actual circumstances, even though they are unreliable in his counterfactual circumstances, so they are adroit, but not apt. Meanwhile, the belief-forming methods of the Clairvoyant are unreliable in our actual circumstances, even though they are reliable in his counterfactual circumstances, so they are apt, but not adroit. So, there is a case to be made that our intuitions about rationality track a notion of reliability that is rigidified to our actual circumstances — not aptness, but adroitness. This rigidification strategy depends crucially on the assumption that our actual world is not a BIV world. For if we assume that our actual world is a BIV world, then it follows that the belief-forming methods of any actual or counterfactual BIV would be neither apt nor adroit, and so not rational either. Intuitively, though, the rationality of belief-forming methods is not in this way dependent on what kind of

107

world the actual world is. For our perceptual beliefs are still rational even if our actual world is a BIV world. In response, Sosa claims that he can accommodate this intuition as follows. He argues that the conditional, if our actual world is a BIV world, then our perceptual beliefs are still rational, is ambiguous between indicative and subjective readings, but that it comes out true either way. If it is read in the indicative mood, then it is trivially true because its antecedent is false – our actual world is not a BIV world. But if it is read in the subjunctive mood, then it is also true because our actual world is not a BIV world, and for any counterfactual BIV world, our perceptual beliefs would have been formed on the basis of belief-forming methods that are reliable in our actual world. In responding this way, Sosa refuses to consider a natural reading of the conditional. The subjunctive conditional is ambiguous between two readings, depending on whether the BIV world is considered as actual or as counterfactual, and Sosa considers only the latter. Following Peacocke (2004), we can use a technique of double-indexing in order to distinguish between these two readings, where P (w1,w2) abbreviates the claim that a proposition P, when evaluated from the standpoint of the external state of w1, holds with respect to w2: (A) For any world w, P (w,w) (B) For any world w, P (@,w) On the B-reading, the first parameter is rigidly determined by the state of the actual world, whereas the A-reading allows for the first parameter to be flexibly determined by a range of non-actual possible worlds. Hence, the A-reading allows us to consider

108

as actual a range of non-actual possible worlds, including the BIV world, whereas the B-reading allows us to consider such worlds only as counterfactual. Now we can use this apparatus in reformulating the objection against Sosa’s proposal. Even if it can explain why our perceptual beliefs would be rational in a BIV world considered as counterfactual, it cannot explain why they would be rational in a BIV world considered as actual. Intuitively, though, our perceptual beliefs would be rational even in a BIV world considered as actual. Another way of making the same point is that basic epistemic principles, including those governing perceptual rationality, are, if true, not only necessarily true, but also fixedly actually true – that is to say, they are true in the actual world, whichever world is considered as actual.6 However, Sosa’s proposal cannot explain the status of basic epistemic principles of perceptual rationality as fixedly actually true, because it implies that our perceptual beliefs would not be rational in a BIV world considered as actual.7 Alvin Goldman (1986, Ch.5) anticipates Sosa’s rigidification strategy, but rejects it, opting instead for the proposal that rational methods are reliable in normal worlds, which he defines as “worlds consistent with our general beliefs about the actual world” (p.107). His idea is that if our actual world were a BIV world, then our general beliefs about the actual world would be false, but our perceptual beliefs would nevertheless be rational, since they would be formed on the basis of methods

6 7

For this terminology, see Davies and Humberstone (1980). Compare Peacocke’s (2004) argument that mind-dependent theories of morality cannot explain the status of basic moral principles as fixedly actually true, because they imply that such principles are not true in a world considered as actual in which we have different morality-generating attitudes.

109

that are reliable in counterfactual worlds in which our general beliefs about the world are true. He sums up the proposal as follows: …beliefs are deemed justified when (roughly) they are caused by processes that are reliable in the world as it is presumed to be. Justification-conferring processes are ones that would be reliable in worlds like the presumptively actual world, that is, in normal worlds. (1986, p.108) Goldman considers himself a reliabilist, but this proposal makes a significant departure from the spirit of a reliabilist theory of rationality. After all, it is plausible that our beliefs about the nature of the actual world play a role in determining what it is rational to believe only if it is rational for us to make them. But if it is rational for us to make them even in worlds considered as actual in which they are dramatically false, such the BIV world, then the rationality of these presumptions cannot be explained in terms of either de facto or rigidified reliability. Trivially, of course, they are reliable in normal worlds, since normal worlds are simply defined as the kinds of worlds in which they are true. But if this is reliabilism, what isn’t? This is not intended as an objection to the proposal considered on its own terms. Indeed, in what follows, I will develop a proposal that is, in many respects, similar to Goldman’s. This proposal provides a strategy for solving the Problem of TruthConduciveness which is entirely compatible with the internalist theory of rationality presented in earlier chapters. Indeed, in crucial respects, it depends upon it. 4.3. The Rational Presupposition Strategy The key idea for my proposal is that the rationality of a method essentially depends upon certain rational presuppositions concerning the circumstances in which

110

it is employed. A method is simply a way of forming beliefs and rational methods are simply rational ways of forming beliefs.8 We can say that R is a rational presupposition for a rational method M just in case it is essential for the rationality of employing M that it is rational for the subject to believe that R. A couple of points are worth emphasizing here. The first point is that a subject can be rational in employing a method M without having formed a rational belief that its rational presuppositions are satisfied and without even having the conceptual capacities required for doing so. Therefore, the claim that rational methods have rational presuppositions is perfectly compatible with the existence of unreflective rationality. The second point is that I am not committed to the claim that a subject’s rational basis for believing R should be independent of M. So, for all that has been said, it may be possible for a subject to form a rational belief that R in a rule-circular way that depends essentially on the employment of M.9 We can illustrate the notion of a rational presupposition by considering the Reflective Access Requirement, which we argued for in Chapter Three: •

The Reflective Access Requirement: necessarily, what it is rational for the subject to believe is accessible to the subject by reflection alone.

To restate this in terms of methods, necessarily, if it is rational (/irrational) for the subject to believe that P on the basis of method M, then the fact that it is rational (/irrational) for the subject to believe that P on the basis of method M is accessible

8 9

Goldman (1986) distinguishes between processes and methods, but I treat them as equivalent. See Pryor (2000), (2004) on independence and Boghossian (2001) on rule-circular justification.

111

by reflection alone. The Reflective Access Requirement therefore implies an iterativity principle, which says that if it is rational for a subject to form beliefs on the basis of M, then it is rational for him to form the higher-order belief that it is rational for him to do so. So, for any rational method M, it is a rational presupposition that M is rational. We can now invoke this result in arguing for the more interesting claim that, for any rational method M, it is a rational presupposition that M is reliable. We begin with the principle of the iterativity of rationality, which we can formulate as follows:10 (1)

M is rational → RB (M is rational)

Now, the following seems very plausible. If it is not rational for the subject to believe that a method M is reliable in the circumstances, then it is not rational for him to believe that it would be rational for him to employ M. By contraposition, it is rational for the subject to believe that M is rational only if it is rational for him to believe that M is reliable in the circumstances: (2)

RB (M is rational) → RB (M is reliable)

And given the iterativity principle, we can exploit the transitivity of the conditional in a way that effectively collapses this second-order claim into a corresponding firstorder claim. Thus, M is rational only if it is rational for the subject to believe that M is reliable in the circumstances:

10

As I hope is clear, “RB” means “it is rational for the subject to believe that…”, while the arrow represents necessitation, rather than mere material implication.

112

(3)

M is rational → RB (M is reliable)

That is to say, for any rational method M, it is a rational presupposition that M is reliable in the circumstances in which it is employed. My proposal, then, is that the truth-conduciveness of rational methods consists in the fact that they are reliable in the circumstances in which their rational presuppositions are satisfied. Of course, it does not follow that rational methods are reliable in the circumstances in which they are employed, since their rational presuppositions may or may not be satisfied – and in the BIV case, for example, they are not. Claim (3) is therefore to be contrasted with the very different claim that is endorsed by the reliabilist: (4)

M is rational → M is reliable

The rationality of the BIV is compatible with (3), but not with (4). For even though his rational methods are not in fact reliable, it is rational for him to believe that they are. This claim is supported by Phenomenal Supervenience, which we argued for in Chapter One. After all, presumably it is rational for us to believe that our rational belief-forming methods are reliable, and it follows that so is the BIV, since phenomenal duplicates are duplicates in respect of what it is rational for them to believe. My proposed strategy for solving the Problem of Truth-Conduciveness is intentionally modest. What I have claimed to be essential for rationality is not that rational methods are in fact reliable, but merely that it is rational to believe that they are. Is this really enough to explain how following rational belief-forming methods is

113

conducive to the aim of believing the truth? In order to see why it is, recall the way in which we motivated the Problem of Truth-Conduciveness as a problem that arises from the reflective standpoint. To repeat, the claim was that if, on reflection, we are unable to make any sense of the claim that rationality is relevantly truth-conducive, then we will have failed to make any sense of why rationality is worth caring about. In order to solve this distinctively reflective problem, it should be enough to show that any method that it is rational for us to employ is one that it is also rational for us to believe, on reflection, to be reliable. We do not need to go so far as to argue that, from the reflective standpoint, we can rule out the very possibility of rationality in the absence of reliability. There is another respect in which my proposed solution is modest, since I have not attempted to address the problem of scepticism. For many of our intuitively rational methods, there are sceptical hypotheses such that it is rational for us to believe that if they were to obtain, then our methods would not be reliable. Therefore, it is rational for us to believe that our methods are reliable only if it is also rational for us to believe that these sceptical hypotheses do not obtain.11 However, I have done nothing to defend the claim that it is in fact rational for us to believe that these sceptical hypotheses do not obtain. For all I have said, the correct view is the sceptical view that none of our belief-forming methods are rational ones. But if we assume that scepticism is false, then there remains a problem of explaining how we

11

This move depends on a plausible closure principle, which says that if it is rational for a subject to believe that P and that P entails Q, then it is also rational for him to believe that Q.

114

can be rational in believing that sceptical hypotheses do not obtain. Various antisceptical proposals have been advanced in the literature.12 It would take us beyond the scope of the current project to evaluate them in any detail, but I take it that there is a whole range of anti-sceptical strategies that are consistent with the proposal developed here. 4.4. An Analysis of Knowledge I have been arguing that it is a rational presupposition for any rational method that it is reliable in the circumstances in which it is employed. But the very same form of argument can be used to establish a stronger conclusion – namely, that it is a rational presupposition for any rational method that it yields knowledge in the circumstances in which it is employed. As before, we begin by invoking the iterativity principle: (1)

M is rational → RB (M is rational)

Next, we reflect on the plausibility of the following claim. If it is not rational for the subject to believe that a method M yields knowledge in the circumstances, then it is not rational for him to believe that it would be rational for him to employ M. By contraposition, it is rational for the subject to believe that M is rational only if it is rational for him to believe that M is knowledge-yielding in the circumstances: (2)

12

RB (M is rational) → RB (M is knowledge-yielding)

Various anti-sceptical strategies have recently been proposed, including those of Pryor (2000) and (2004), Peacocke (2004) and Wright (2004).

115

And given the iterativity principle, we can exploit the transitivity of the conditional in a way that effectively collapses this second-order claim into a corresponding firstorder claim. Thus, M is rational only if it is rational for the subject to believe that M is reliable in the circumstances: (3)

M is rational → RB (M is knowledge-yielding)

The upshot is that, for any rational method M, it is a rational presupposition that M is knowledge-yielding in the circumstances and hence that M yields knowledge when all its rational presuppositions are satisfied. It may be objected that there are some rational methods such that it is obvious, on reflection, that they cannot yield knowledge. For example, many philosophers claim that in a lottery, it is rational for me to believe that I won’t win, although I don’t know that I won’t win. However, if these are the facts of the case, then presumably they are accessible on reflection. And so, it is rational for me to believe that the methods I use in such a case are rational, though not knowledge-yielding in the circumstances. There may be various possible ways of responding, but here is one that I find plausible.13 First, there may be some, less demanding contexts in which it is true to say not only that it is rational for me to believe that I won’t win, but also that I know that I won’t win. But in more demanding contexts in which it is not true to say that I know that I won’t win, it is also not true to say that it is rational for me to believe outright that I won’t win. Rather, in such a case, it is rational for

13

See Wedgwood (2002) for this response to the lottery case.

116

me to have a high partial degree of belief that I won’t win and to believe outright that I probably won’t win.14 This forces an important qualification. I have claimed that it is a rational presupposition for any rational method that it yields knowledge in the circumstances in which it is employed. But this claim is meant to apply only to rational methods that yield outright belief. Of course, there are also rational methods that yield partial degrees of belief and these may have rational presuppositions, such as the rational presupposition that they are reliable in the circumstances in which they are employed. But it is not a rational presupposition of such methods that they yield knowledge in the circumstances in which they are employed. What I am claiming, in effect, is that outright belief aims at knowledge, whereas partial degrees of belief do not. The discussion that follows is restricted to methods that yield outright belief. A rational presupposition for any such method is that it is knowledge-yielding in the circumstances. Therefore, it is a sufficient condition for a belief to be knowledge that it is formed on the basis of rational methods all of whose rational presuppositions are satisfied. But is it also a necessary condition? To deny this is to hold that a rational method M can have a rational presupposition R such that M can yield knowledge even if R is false. Here, one might appeal to “hyper-reliable” propositions, which are propositions that are true whenever they are thought because they are made true by the very fact that they are thought.15 An example is the

14 15

See Williamson (2000, p.99) on the distinction between outright belief and high degrees of belief. See Pryor (2006) for further discussion.

117

proposition, I am thinking a thought that water is wet. One might argue that, necessarily, it is rational for any subject to believe any hyper-reliable proposition and hence that every hyper-reliable proposition is a trivial case of a rational presupposition for every rational method. However, the truth of every such proposition is certainly not a necessary condition for every rational method to yield knowledge. For example, it is rational for me to take perceptual experience at face value whether or not it is true that I am thinking a thought that water is wet. I have two objections against this proposed counterexample to necessity. The first is that it is not plausible that, necessarily, it is rational for any subject to believe any hyper-reliable proposition. James Pryor has given convincing counterexamples, such as the proposition, I am thinking some thought that includes quantification and more than two unsaturated components. It is not rational for a subject to believe this proposition unless he has reflected on it and appreciated that by thinking it, he makes it true. Moreover, this treatment plausibly generalizes to other hyper-reliable propositions. So, it is not clear that there are in fact any propositions such that, necessarily, it is rational for any subject to believe them. Even if we suppose that there are some such propositions, a second problem remains. If R is a rational presupposition for a rational method M, then it is not merely a necessary condition, but an essential condition, for the rationality of M that it is rational for the subject to believe R. What this means is that the rationality of believing R must either explain or be explained by the rationality of M. This condition is not met in the case of hyper-reliable propositions. Even if we grant the

118

claim that, necessarily, it is rational for the subject to believe any hyper-reliable proposition P, we do not have essential conditions for the rationality of a method without the relevant explanatory relations. Presumably, though, the rationality of believing P would be explained in terms of its hyper-reliability, rather than in terms of its relations to rationality of M. By contrast, every rational presupposition R that I have proposed so far can be explained by appealing to the rationality of M, along the following lines: (1)

M is rational → RB (M is rational)

(2)

RB (M is rational) → RB (M is knowledge-yielding)

(3)

RB (M is knowledge-yielding) → RB (R)

Therefore, (4)

M is rational → RB (R)

I want to suggest that all rational presuppositions for a rational method M can be explained in the same way. This proposal is intended not as mere stipulation, but as a substantive claim that is open to counterexample. So far as I can see, every rational presupposition for every rational method can be explained along these lines. If this is right, then the most fundamental rational presupposition for any rational method is that it is knowledge-yielding in the circumstances. All other rational presuppositions are to be explained derivatively by the fact that their satisfaction is essential for the method in question to yield knowledge. So, for example, reliability is a rational presupposition for any rational method because the reliability of a method is

119

essential for it to yield knowledge in the circumstances. And likewise, I suggest, for all the others. I conclude that it is both necessary and sufficient for a belief to be knowledge that it is rationally held on the basis of rational methods when all of their rational presuppositions are satisfied. This yields an analysis of knowledge:16 •

An Analysis of Knowledge: a belief is knowledge just in case it is rationally held on the basis of rational methods when all of their rational presuppositions are satisfied.

This analysis is rather different in character from many of the analyses proposed in the post-Gettier tradition, in that it does not attempt to specify which conditions are the rational presuppositions of rational methods. Of course, we know that the analysis is extensionally adequate just in case all the rational presuppositions for any given rational method are entailed by the proposition that the method in question is knowledge-yielding in the circumstances. But this does not provide us with any reductive specification of which conditions are the rational presuppositions of rational methods. Nevertheless, the analysis is not trivial, since it provides us with a non-trivial elucidation of the nature of the conceptual connection between rationality and knowledge. Moreover, in the next chapter, I argue that it illuminates the value of

16

Wedgwood (2002) proposes a similar analysis: “knowledge is a rational belief that results from the thinker’s following…a rule or set of rules that worked as it was supposed to,” in the sense that, “everything that it must have been rational for her to believe, in order for it to be rational for her to make that belief revision through following those rules, really was the case,” or, in other words, “all of the ‘rational presuppositions’ of the thinker’s making that belief revision through following those rules were true”.

120

knowledge by providing the correct articulation of the intuition that knowledge is incompatible with luck.

121

CHAPTER FIVE

RATIONALITY, KNOWLEDGE AND LUCK

Chapter 5. Rationality, Knowledge and Luck

5.1. Anti-Luck Epistemology Anyone who knows that something is so thereby has a true belief that it is so. But not just any true belief amounts to knowledge. For instance, a belief based on mere guesswork fails to amount to knowledge, even if it happens to be true. Intuitively, this is because it is merely a matter of luck that it is true. So, we can ask, what does it take for a true belief to amount to knowledge? An initially plausible answer is that a true belief is knowledge if and only if it is not merely a matter of luck that it is true.1 Ultimately, I think this answer is correct, but it stands in need of considerable refinement. Not every kind of luck is incompatible with knowledge. For example, you can know that you’ve won the lottery, even though it is lucky that the proposition you believe is true. And you can have knowledge even if you were lucky to form your belief in the first place — for example, you can know that a grand piano narrowly missed falling on your head.2 You can even have knowledge if you’re lucky to be in

1

Compare Unger: “For any sentential value of p (at a time t), a man knows that p if and only if at t it is not at all accidental that the man is right about its being the case that p.” (1968, p.158) 2 Unger qualifies his proposal in the light of similar examples: “…a complete absence of the accidental is claimed, not regarding the occurrence or existence of the fact known nor regarding the existence or abilities of the man who knows, but only as regards a certain relation between the man and the fact.” (1968, p.159)

122

circumstances in which you form a true belief, rather than a false belief — to adapt Nozick’s (1981) example, you might knowledgeably recognize the masked bandit as Jesse James only because his mask happens to slip; otherwise, you would have falsely supposed him to be Billy the Kid. So, clearly, we need a more discriminating account of which kinds of luck are incompatible with knowledge. In what follows, I will use the term “epistemic luck” to denote the kinds of luck that are incompatible with knowledge. In addition, I will assume that we can give an account of knowledge by giving an account of the conditions for epistemic luck, which we can call the project of “anti-luck epistemology”. My main aim in this chapter is to argue that the conditions for epistemic luck are captured by the analysis of knowledge proposed in the previous chapter: •

An Analysis of Knowledge: a belief is knowledge just in case it is rationally held on the basis of rational methods when all of their rational presuppositions are satisfied.

My proposal is that a belief is immune from epistemic luck if and only if it is rationally held when all of its rational presuppositions are satisfied. Moreover, this helps to illuminate the value of knowledge. We value knowledge because it is a certain kind of cognitive achievement. In particular, it is the kind of cognitive achievement that results when a subject’s beliefs achieve the aim of truth in a way that is immune from epistemic luck. Many philosophers believe that we can give an account of the conditions for epistemic luck solely in terms of reliability, without making any appeal to the notion

123

of rationality. On this kind of view, rationality may be a desirable property for a belief to have, but it is not essential for knowledge.3 However, I will argue that such views are unable to capture the anti-luck intuition. Rationality is not merely an optional extra for knowledge. On the contrary, rationality is essential for a belief to be immune from epistemic luck in the way that is required if it is to amount to knowledge. 5.2. Reliabilist Theories of Knowledge Reflection on guesswork provides some initial support for the claim that rationality is required for a belief to be immune from epistemic luck. If a belief that is based on guesswork happens to be true, then it is merely a matter of luck that it is true. And presumably this is because a belief that is based on guesswork is not rationally justified. However, the requirement of rational justification is not sufficient to rule out all the conditions for epistemic luck. After all, there are well known Gettier-style counterexamples to the sufficiency of an analysis of knowledge as justified, true belief, which turn on the fact that it fails to capture the conditions for epistemic luck. Here are a few representative examples: •

False Lemmas: you have applied for a job, but you are justified in believing that Jones will get the job. You are also justified in believing that Jones has ten coins in his pocket. So, you infer that whoever will get the job has ten coins in his pocket. In fact, however, you are the one who will get the job

3

For this view, see Pritchard (2005).

124

and, unbeknownst to you, there are ten coins in your pocket, so your inferred belief is both justified and true. But you could easily have had fewer coins in your pocket, in which case your belief would have been justified, but false.4 •

Stopped Clocks: you look at the clock, which reads twelve o’clock and you form a justified, true belief that it is twelve o’clock. Unbeknowst to you, the clock stopped exactly twelve hours ago. But you could easily have looked at the clock an hour earlier, in which case your belief would have been justified, but false.5

•

Fake Barns: you see a barn in front of you and form a justified, true belief that there is a barn in front of you. Unbeknownst to you, you are surrounded by fake barns. So, you could easily have seen a fake barn, in which case your belief would have been justified, but false.6

In each of these cases, it is plausible to say that the belief in question fails to amount to knowledge because it is merely a matter of luck that it is true.7 Moreover, in each case, the subject could easily have formed a false belief in the same way. The same point applies in the case of guesswork, since a belief that is formed on the basis of guesswork might happen to be true, but the subject could easily have formed a false belief in the same way. This suggests an alternative account of epistemic luck in

4

Adapted from Gettier (1963). Adapted from Russell (1948). 6 Adapted from Goldman (1976), who attributes the example to Carl Ginet. 7 Hetherington (1998) claims that these are cases of lucky knowledge in which a subject actually knows, but could have easily failed to know. Intuitively, though, not all kinds of luck are compatible with knowledge, even though some kinds are. The challenge is to articulate this distinction in a way that makes good sense of our concept of knowledge. 5

125

counterfactual terms, according to which a true belief is infected by epistemic luck just in case the subject could easily have formed a false belief in the same or a relevantly similar way. As Nozick (1981) observes, the relativization to “ways” or “methods” of forming beliefs helps to explain why epistemic luck is not involved in the Jesse James case. For you have a recognitional capacity which is brought to bear in the case where his mask slips, but not in the nearby case where it stays on, and so your beliefs are formed in relevantly different ways in each case. Therefore, your belief can be knowledge when the mask slips, even though your belief would be false in the nearby case in which it stays on. Of course, this raises what Conee and Feldman (1998) have called the “generality problem”, which is the problem of determining which are the relevant ways or methods.8 But it is not clear that we should be expected to provide a general algorithm for determining which ways are relevant, so long as we can decide the issue on a case-by-case basis. Presumably, the issue involves a certain degree of vagueness and context-sensitivity. Certainly, we will need the resources for providing a principled response to ad hoc or gerrymandered accounts of which ways are the relevant ways. But this problem, if it is a problem, is endemic in any attempt to characterize the epistemic properties of beliefs in terms of the way they are formed or sustained.

8

There is also a related problem of determining, for any given case, which possible worlds are sufficiently similar or “nearby” to qualify as worlds that could easily have obtained.

126

The counterfactual account of epistemic luck provides a rationale for an alternative to the traditional analysis of knowledge as justified, true belief. On this alternative kind of analysis, the extra condition for a true belief to be knowledge is not rational justification, but rather a counterfactual condition that is designed to rule out epistemic luck. According to a reliabilist analysis of knowledge, for example, a true belief is knowledge if and only if it is reliable, in the sense that the subject could not easily have held a false belief in the same or any relevantly similar way.9 There is no commitment here to any kind of reliabilist analysis of the notion of justification that figures in the traditional analysis of knowledge. For the proponent of a reliabilist analysis of knowledge can accept an internalist conception of justification, while denying that it constitutes a necessary condition for knowledge. On this view, rational justification is not essential for knowledge, but is merely a desirable, but optional, extra. However, a reliabilist analysis of knowledge is vulnerable to some of the very same counterexamples that have been raised for a reliabilist analysis of rational justification. What is more, the intuitive force of these counterexamples seems to turn on the fact that the reliability condition fails to capture the conditions for epistemic luck. For example, consider clairvoyance. Intuitively, clairvoyant beliefs do not amount to knowledge because it is merely a matter of luck, considered from 9

Counterfactual theories of knowledge have been proposed most notably by Goldman (1976), Nozick (1981) and Sosa (1999), but they differ in their formulation of the counterfactual requirement. As I have formulated it, reliability is closer to Sosa’s notion of safety than Nozick’s more demanding notion of sensitivity. As Sosa articulates the distinction: “A belief is sensitive iff had it been false, S would not have held it, whereas a belief is safe iff S would not have held it without it being true.” (1999, p.146)

127

the subject’s point of view, that his beliefs are true. If we abstract away from the subject’s point of view, then of course we lose sight of this element of luck, since the beliefs in question are formed on the basis of a reliable method. But as BonJour insists, we cannot abstract away from the subject’s perspective in evaluating the rationality or justifiability of his beliefs: One reason why externalism may seem initially plausible is that if the external relation in question genuinely obtains, then Norman will in fact not go wrong in accepting the belief, and it is, in a sense, not an accident that this is so: it would not be an accident from the standpoint of our hypothetical external observer who knows all the relevant facts and laws. But how is this supposed to justify Norman’s belief? From his subjective perspective, it is an accident that the belief is true. And the suggestion here is that the rationality or justifiability of Norman’s belief should be judged from Norman’s own perspective rather than one which is unavailable to him. (1985, p.43-4) By the same token, it seems that we cannot abstract away from the subject’s perspective in evaluating whether or not his beliefs amount to knowledge. For if a belief is not rationally justified, then it can be no more than a matter of luck, considered from the subject’s perspective, if it happens to be true. Intuitively, however, this kind of luck is incompatible with knowledge and so rational justification is necessary for knowledge. The proponent of a reliabilist theory of knowledge may not be convinced by this kind of example. He may grant that Clairvoyance involves a “subjective” kind of luck, which is not ruled out by the counterfactual condition on reliability. However, he may also deny that this kind of luck is incompatible with knowledge, even if it is incompatible with rational justification. To suppose otherwise would be to assume what he denies – namely, that rational justification is a necessary condition for

128

knowledge. Thus, we have something of a stand-off here. However, in what follows, I want to undermine this response by showing that we can modify the Gettier-style cases in such a way as to reveal that they too involve a subjective kind of luck, which is not ruled out by the counterfactual condition on reliability, but which is clearly incompatible with knowledge. 5.3. The Super-Gettier Cases Let us say that a Super-Gettier case is one in which the subject has a true belief that is both rational and reliable, but which does not amount to knowledge because it is merely a matter of luck that it is true.10 If we can construct such a case, then we will have shown that reliability alone is not sufficient to rule out epistemic luck. But there is in fact a recipe for constructing such cases. First, take a standard Gettier case in which a subject has a rational, true belief that falls short of knowledge and then gerrymander the background conditions in such a way that the subject also forms justified, true beliefs of the same kind in nearby possible worlds. By following this procedure, we can turn any standard Gettier case into a Super-Gettier case. For example, in the case of “False Lemmas”, we may suppose that, unbeknownst to you, there is a mischievous company representative who always slips ten coins into the pockets of job candidates. So, your belief that whoever gets the job has ten coins in his pocket could not easily have been false. Nevertheless, it 10

On the face of it, lotteries are cases in which subjects have beliefs that are both rational and reliable, but intuitively fall short of knowledge. However, this is debatable. For example, Lewis (1996) suggests that reliability stands or falls with knowledge in a way that is sensitive to context: in demanding contexts, the relevant beliefs are not sufficiently reliable for knowledge, whereas in relaxed contexts, they are.

129

does not amount to knowledge because it is merely a matter of luck, considered from your own subjective perspective, that your belief is true. In response, it may be objected that the reliability requirement for knowledge has not been satisfied in this case. Even if the belief in question is such that it could not easily have been false, it is nevertheless formed in a way that could easily have yielded other false beliefs.11 For instance, you could easily have been justified in believing that Jones had nine coins in his pocket, in which case you would in a relevantly similar way have formed the justified, but false, belief that whoever gets the job has nine coins in his pocket. However, this objection can be dealt with very easily. For we can always just elaborate the case in such a way as to rule out the possibility that you could easily have formed a justified, false belief in a relevantly similar way. We may stipulate that, for some reason or other, you could not easily have been justified in believing that Jones had nine coins in his pocket; or alternatively, that if you could have been, then there would have been a mischievous company representative who always surreptitiously slips nine coins into the pockets of job candidates. The general point is that we can always stipulate the details of the case in such a way that a subject’s belief is not only rational, but also reliable, though intuitively it fails to amount to knowledge. Perhaps the clearest case is that of the “Super-BIV”. This is a rational Brain in a Vat (BIV) whose experiences match the world in such a way that many of his

11

This corresponds to Goldman’s (1986) distinction between local and global reliability.

130

perceptual beliefs are true.12 We may suppose that his experiences are being produced by a benevolent neuroscientist who aims to simulate the experiences the BIV would have if he were normally embedded in his immediate environment. Moreover, we may suppose that the neuroscientist is sufficiently dedicated and skilled in this task that he could not easily have failed to produce a successful simulation. Thus, the perceptual beliefs of the Super-BIV are not only rational and true, but also reliable, since they are formed in a way that could not easily have yielded false beliefs. However, considered from the subject’s perspective, it is merely a matter of luck that his perceptual beliefs are true and so they do not amount to knowledge. In this way, we can construct Super-Gettier cases by gerrymandering the background conditions of a standard Gettier-style case in such a way as to render the subject’s beliefs reliable, but without thereby eliminating the element of epistemic luck. The reliabilist analysis of epistemic luck, then, is clearly inadequate. For that matter, so is the account of epistemic luck in terms of rational justification. Indeed, these cases show that even the conjunction of rationality and reliability is insufficient to rule out epistemic luck. It is not enough merely to conjoin the reliability condition by means of a rationality condition — as an afterthought, as it were. Rather, what is required is some way of integrating them. So, we are faced with an Integration 12

Even if his perceptual-demonstrative beliefs are not true because they lack reference, his descriptive perceptual beliefs will be true. Can’t we explain why these beliefs are not knowledge on the ground that they are formed on the basis of inference from lemmas that are, if not false, then at least not true? Only on the questionable assumption that descriptive perceptual beliefs must be formed on the basis of perceptual-demonstrative beliefs in order to be rational.

131

Problem — how must rationality and reliability be integrated if a true belief is to be knowledge? The perceptual beliefs of the Super-BIV are both rational and reliable, but they do not amount to knowledge because there is not the right kind of relationship between the rationality and the reliability of his beliefs. In particular, his beliefs are reliable not in virtue of their rationality, but rather in virtue of the interventions of the benevolent neuroscientist. In what follows, I consider the suggestion that we can think of knowledge as a true belief that is not only rational and reliable, but also reliable in virtue of being rational. This raises the question of how we to understand the “in virtue of” relation. Rather than attempting to provide any general analysis of this relation, I propose to draw attention to our intuitive understanding of the relation by briefly considering what it is for an object to exhibit a response in virtue of possessing a certain disposition.13 5.4. Dispositions and Counterfactuals The ascription of a disposition to an object implies that certain counterfactuals are true of it. But it is not easy to formulate the relevant counterfactuals in a way that is immune to counterexamples. According to a simple counterfactual analysis of dispositions, an object O has a certain disposition D just in case there are non-trivial

13

This approach occurred to me as a result of Crispin Wright’s suggestion that the counterexamples to Nozick’s counterfactual analysis of knowledge may be viewed as just more counterexamples to the crude counterfactual analysis of dispositions.

132

triggering conditions C such that if they were to obtain, then the object would exhibit some characteristic response R: •

The Simple Counterfactual Analysis: O has D just in case: if C were to obtain, then O would exhibit R.

As is well known, this Simple Counterfactual Analysis is subject to various kinds of counterexamples. For example, there are “finkish” cases in which the object in question would either gain or lose the disposition if the relevant triggering conditions were to obtain. But there are also non-finkish cases in which the object’s characteristic response would be “mimicked” or “masked” if its triggering conditions were to obtain, without either losing or gaining the disposition in question.14 Masking cases are counterexamples to necessity, in which if C were to obtain, then O would have D but would not exhibit R, while mimicking cases are counterexamples to sufficiency, in which if C were to obtain, then O would lack D but would nevertheless exhibit R. The recipe for constructing these kinds of counterexamples is to ensure that triggering conditions C occur against the background of interfering conditions C* which reliably affect whether or not O would exhibit R in C. Consider the analysis of fragility in terms of the claim that a fragile object would break if it were dropped on a hard surface. The idea is that the interference of some extraneous natural or supernatural phenomenon could either 14

See Martin (1994) for finkish cases and Johnston (1992) for cases of mimicking and masking. These kinds of cases can be distinguished in terms of whether or not the object in question gains or loses the intrinsic properties that serve as the causal basis of the disposition. As Bird (1998) argues, Lewis’ (1997) proposed amendment to the counterfactual analysis – roughly, that O has D just in case if C and O retains causal basis B, then O exhibits R – deals with the finkish cases, but not cases of mimicking and masking.

133

mimic the disposition, by reliably causing a non-fragile object to break when dropped on a hard surface, or it could mask it, by reliably preventing a fragile object from breaking when dropped on a hard surface. Given the Simple Counterfactual Analysis, it is natural to assume that if O has D and exhibits R in C, then it does so in virtue of having D. However, we can also construct “super-mimicking” cases in which O has D and reliably exhibits R in C, but does not do so in virtue of having D. The recipe for producing such cases is to take a regular masking case in which O’s disposition to exhibit R in C would be masked by background interfering conditions C*, were it not for the fact that their masking effect is over-ridden by additional background interfering conditions C** which reliably ensure that O nevertheless exhibits R in C. For example, if I drop a fragile object onto a hard surface, C* may reinforce its structural properties in such a way that it would not break, were it not for the fact that C** cancels out C*, not by neutralizing its reinforcing effect, but by overriding it – for instance, by emitting very powerful rays that shatter the object. In such a case, the object does not shatter in virtue of its fragility, but rather in virtue of the extraneous background conditions C**. The truth of the counterfactual that the object O would exhibit R in C is explained by background conditions C**, rather than by the structural properties of the object that make it fragile. How should we amend the Simple Counterfactual Analysis in the light of such cases? An obvious move is to introduce a “ceteris paribus” clause:

134

•

The Amended Counterfactual Analysis: O has D just in case: ceteris paribus, if C were to obtain, then O would exhibit R.

For it was by stipulating background interfering conditions C* that we were able to generate masking cases in which O has D and yet fails to exhibit R in C and mimicking cases in which O does not have D and yet does exhibit R in C. But cases in which interfering conditions obtain are not cases in which the ceteris paribus clause is satisfied. Similarly, it was by stipulating a masking case against the background of interfering conditions C** which override the masking effects of C* that we were able to generate super-mimicking cases in which O has D and exhibits R in C but does not do so in virtue of possessing D. But, again, cases in which interfering conditions obtain are not cases in which the ceteris paribus clause is satisfied. Suppose, on the other hand, that the ceteris paribus clause is satisfied, so that interfering conditions do not obtain. In such a case, if O has D and C obtains, then O exhibits R and, moreover, it does so in virtue of having D. It may be objected that the ceteris paribus clause trivializes the analysis. After all, there may be an open-ended list of possible interfering conditions that is not amenable even in principle to any kind of finite, substantial specification. The worry is that in the absence of any substantial specification of what it is for the ceteris paribus clause to be satisfied, the amended analysis collapses into the vacuous claim that O has D just in case if C were to obtain, then O would exhibit R unless it doesn’t. But this objection rests on an overly demanding conception of what is required to avoid vacuity. It is not required that we should be able to give a finite,

135

substantial specification of all the possible interfering conditions whose obtaining would conflict with the satisfaction of the ceteris paribus clause. All that is required is that, for any given case in which the ceteris paribus clause is not satisfied, we should be able to explain why not by specifying interfering conditions which obtain in that particular case and which also pull explanatory weight in other cases.15 A substantial specification of interfering conditions on a case-by-case basis does not require a substantial specification of all possible interfering conditions, which may be open-ended and so not finitely specifiable even in principle. So, the introduction of a ceteris paribus clause provides us with a counterfactual analysis of dispositions, which is neither vacuous nor vulnerable to counterexamples. 5.5. A Dispositional Theory of Knowledge? After this brief diversion, let us return to the main thread. Our question is this: How must rationality and reliability be integrated if a true belief is to be knowledge? The suggestion I want to consider is that knowledge is true belief that is not only rational and reliable, but also reliable in virtue of being rational. Consider the SuperBIV. His perceptual beliefs are rational and reliable, but they are not reliable in virtue of being rational. This is because the truth of the relevant counterfactual (that he could not easily have formed false beliefs in the relevant ways) is explained by extraneous background interfering conditions (by the interventions of the benevolent 15

Pietrowski and Rey (1995) develop this line in defending the non-vacuity of ceteris paribus laws, but as they also point out, a proposed ceteris paribus law might be non-vacuous without thereby being true. On the relationship between dispositions and ceteris paribus laws, see Bird (2005) for the claim that ceteris paribus laws are grounded in the essentially dispositional nature of properties.

136

neuroscientist) rather than by the exercise of the subject’s rational belief-forming dispositions. There is a striking parallel with the case of super-mimicking a disposition. In a case of super-mimicking a disposition, the truth of the relevant counterfactual (that the object would exhibit response R if triggering conditions C were to obtain) is explained by extraneous background conditions C**, rather than by the exercise of any of the object’s dispositions. Indeed, we might even say that the Super-BIV case is a case of super-mimicking the dispositional-reliability of a rational belief-forming method. Let us say that a belief-forming method is dispositionally-reliable if and only if it is reliable when normal conditions obtain – that is, when the ceteris paribus clause is satisfied. Like any other dispositional property, the dispositional-reliability of a belief-forming method can be masked or super-mimicked in cases in which the ceteris paribus clause is not satisfied. With some plausibility, then, we might suggest that rational belief-forming methods have the property of dispositional-reliability, which can be masked or super-mimicked when, as in envatment cases, the ceteris paribus clause is not satisfied. In the case of the regular BIV, the dispositionalreliability of his rational methods is masked, with the result that his beliefs are rational, but not reliable. And in the case of the Super-BIV, the dispositionalreliability of his rational methods is super-mimicked, with the result that his beliefs are rational and reliable, but not reliable in virtue of being rational. The suggestion is that cases of masking and super-mimicking occur only when normal conditions do not obtain. But if normal conditions obtain, such that the

137

ceteris paribus clause is satisfied, then the dispositional-reliability of rational methods is neither masked nor super-mimicked. In such a case, the subject’s beliefs will be not only rational and reliable, but also reliable in virtue of being rational. And the natural suggestion is that this is both necessary and sufficient for knowledge. What we have here is a kind of dispositional analysis of knowledge, according to which a true belief is knowledge if and only if it is reliable in virtue of the dispositional-reliability of the rational methods on the basis of which it is held. Or, more succinctly, a true belief is knowledge if and only if it is reliable in virtue of being rational. According to the dispositional analysis of knowledge, the counterfactually specified reliability of the subject’s beliefs is not sufficient for knowledge unless the truth of the relevant counterfactuals holds in virtue of the exercise of the subject’s rational belief-forming dispositions.16 In my view, there is much insight to be gained from this dispositional analysis of knowledge. But the crucial question, of course, is what it takes for the ceteris paribus clause to be satisfied. What is it for conditions to be “normal” with respect to the employment of a rational belief-forming method? I do not see how to give any principled answer to this question except by appealing to the notion of rational presuppositions. Conditions are normal with respect to the employment of a beliefforming method if and only if its rational presuppositions are satisfied. For this 16

There are points of contact here with Sosa’s (1991) virtue epistemology, which aims to analyze knowledge in terms of reliable belief-forming dispositions or “intellectual virtues”. In the same tradition, John Greco proposes a view that he calls “agent reliabilism”, according to which “it is those [reliable] processes that have their bases in the stable and successful dispositions of the believer that are relevant for knowledge.” (1999, p.287)

138

reason, the dispositional analysis of knowledge is not in competition with the analysis of knowledge in terms of rational presuppositions, but rather gives us an alternative way of formulating it. According to the analysis of knowledge in terms of rational presuppositions, a belief is immune from epistemic luck if and only if it is rationally held when all of its rational presuppositions are satisfied. Every case of epistemic luck is a case in which either the subject’s belief is irrational or it is rational, but held in circumstances in which its rational presuppositions are not satisfied. The first condition covers the cases of guesswork and Clairvoyance, while the second covers the various Gettier and Super-Gettier cases. Plausibly, though, every instance of epistemic luck results from either a failure in rationality or a failure in the satisfaction of rational presuppositions. If this is correct, then the appeal to rationality is absolutely indispensable for an account of epistemic luck and so, given the project of anti-luck epistemology, it is also indispensable for an account of knowledge. Purely reliabilist theories of knowledge are radically mistaken in their view that rationality is not an essential component of knowledge, but rather a desirable, but optional, extra. 5.6. McDowell on the Composite Conception Before closing, I want to consider how my proposed analysis of knowledge fares in response to John McDowell’s (1995) critique of what he calls the “hybrid” or “composite” conception of knowledge. McDowell’s paper begins with the Sellarsian idea that knowledge is “a certain sort of standing in the space of reasons” and then

139

goes on to contrast this with the composite conception of knowledge, which is characterized as follows: A satisfactory standing in the space of reasons is a necessary condition for knowledge. But since the positions one can reach by blameless moves in the space of reasons are not factive, as epistemically satisfactory positions are, a satisfactory position in the space of reasons cannot be what knowledge is. (1995, p.400) According to the composite conception, rationality – that is, a satisfactory standing in the space of reasons – is a condition that is necessary but not sufficient for knowledge. On this view, knowledge is a composite status requiring not only rationality, but also further requirements (e.g. truth, reliability, etc.), whose satisfaction is not guaranteed by the satisfaction of the rationality requirement alone. The main problem for the composite conception is presented here: In the hybrid conception, a satisfactory standing in the space of reasons is only part of what knowledge is; truth is an extra requirement. So two subjects can be alike in respect of the satisfactoriness of their standing in the space of reasons, although only one of them is a knower, because only in her case is what she takes to be so actually so.… The extra that we need for knowledge…is, relative to the knower’s moves in the space of reasons, a stroke of good fortune, a favour the world does her. So if we try to picture epistemic status as constituted in the way the hybrid conception has it, we are vulnerable to a version of the familiar point that distinguishes knowledge from mere true belief. (1995, p.403-4) The problem is that the composite conception seems to impose incompatible requirements on the concept of knowledge. If the satisfaction of the rationality requirement fails to satisfy the further requirements for knowledge, then there can be pairs of subjects whose beliefs are alike in respect of their rational status, but which differ in whether or not they satisfy the further requirements for knowledge. But if

140

such pairs of subjects are possible, then it is merely a matter of luck from the subject’s point of view whether or not his rational beliefs amount to knowledge. However, if it is merely a matter of luck from the subject’s point of view whether or not his rational beliefs amount to knowledge, then they do not amount to knowledge, since knowledge is incompatible with luck. So, the composite conception renders knowledge unobtainable. How are we to rescue the possibility of knowledge? McDowell’s suggestion, as I understand it, is that we should reconceive the rationality requirement in such a way that rationality is not merely necessary, but also sufficient for knowledge.17 Thus, he writes: Once she has achieved such a standing [in the space of reasons], she needs no extra help from the world to count as knowing. (p.406) On this view, there cannot be pairs of subjects whose beliefs are alike in respect of their rational status, but which differ in whether or not they amount to knowledge. But this is a very extreme view. The point is not merely that it rules out the plausible view that the beliefs of a BIV can have the same rational status as those of his normally embedded phenomenal duplicate. For this is also ruled out by Williamson’s (2000, Ch.9) view that knowledge is what provides a subject’s beliefs with rational justification. But even Williamson’s view allows for the possibility of pairs of subjects who perform the very same inductive inferences from the very same 17

I am not confident that I am interpreting McDowell correctly here, since he allows for justification without knowledge in lottery cases. But it is not clear to me that there is any weaker view in the vicinity that survives the critique of the composite conception. See the discussion of Williamson’s view below.

141

knowledge, but only one of whom is situated in a kind of world such that the inference yields knowledge. On his view, the beliefs of these subjects may be alike in respect of their rational status, even though they differ in whether or not they amount to knowledge. McDowell’s argument against the composite conception view proves too much. It rules out everything short of the implausibly demanding view that rationality is not only necessary, but also sufficient for knowledge. And this is because it depends on an impossibly demanding conception of epistemic luck, according to which any condition whose obtaining is “not ensured by one’s standing in the space of reasons” is “relative to the knower’s moves in the space of reasons, a stroke of good fortune, a favour the world does her”. But it is not enough simply to abandon this conception of epistemic luck unless we have a plausible alternative to put in its place. It seems to me that McDowell’s critique of the composite conception can be reformulated as a version of the Integration Problem. If internalist rationality is a necessary, but insufficient, condition for knowledge, then we need an account of how it is integrated with the additional requirements for knowledge. If these requirements cannot integrated in any appropriate way — if their co-satisfaction is nothing more than a metaphysical coincidence — then in the context of the intuitions that motivate the rationality requirement, the satisfaction of any additional requirements can be no more than a matter of luck from the subject’s own perspective. But in that case, we will have failed to make any sense of the significance of the intuition that knowledge is incompatible with luck.

142

The analysis of knowledge in terms of rational presuppositions provides a way of solving this Integration Problem in a way that is compatible with the composite conception. For while rationality is not sufficient for knowledge, it is nevertheless a rational presupposition for any rational method that it yields knowledge in the circumstances and hence that any further conditions for knowledge are satisfied. On this view, rational duplicates can differ in respect of whether or not their beliefs amount to knowledge only if they also differ in whether or not the rational presuppositions of their rational methods are satisfied. But it is not merely a matter of luck from the subject’s point of view if the rational presuppositions of his rational methods are satisfied. On the contrary, it is a matter of bad luck if they are not satisfied. McDowell’s critique of the composite conception derives much of its appeal from the insight that a belief cannot amount to knowledge if that depends on the obtaining of conditions that are blankly external to the subject’s own perspective. But it would be a mistake to assume, as McDowell does, that any condition whose satisfaction is not entailed by the rationality of the subject’s beliefs is, in the relevant sense, blankly external to the subject’s perspective. For if the obtaining of some condition is a rational presupposition of a rational method, then it is not blankly external to the subject’s perspective, but it is not entailed by the rationality of the subject’s beliefs either. A satisfactory account of epistemic luck must supplement the appeal to rationality with an appeal to rational presuppositions. This is how we should reform the traditional analysis of knowledge as rationally justified, true belief.

143

5.7. The Value of Knowledge In the years following the publication of Gettier’s counterexamples to the traditional analysis of knowledge, a series of increasingly complicated and baroque analyses were proposed in an attempt to patch up the analysis in the face of a rapidly growing stock of counterexamples.18 More recently, the project of analyzing knowledge has met with a considerable degree of pessimism, not only concerning its prospects, but also concerning its significance. For example, Williamson (2000) has suggested that there may be no non-trivial way of elaborating a set of individually necessary conditions for knowledge into a set that is also jointly sufficient. But even if there is in fact a set of non-trivial necessary and sufficient conditions for knowledge, why should this be anything more than a metaphysical coincidence? Knowledge is something that we value and an analysis of knowledge should illuminate its value. But an analysis of knowledge that takes the form of a disunified and potentially open-ended list of divergent conditions does little to explain why we should care about their co-satisfaction. An analysis of knowledge should explain how the seemingly divergent conditions for knowledge hang together in such a way as to make sense of our interest in their co-satisfaction. If there is nothing that unifies them, then it becomes obscure why we should attach any special significance to knowledge over and above its various sub-conditions. What I want to suggest is that the analysis of knowledge in terms of rational presuppositions solves the Integration Problem by providing us with the organizing principle that unifies the various 18

See Shope (1983) for an overview of the literature of the first ten years or so.

144

conditions that are necessary for a true belief to be knowledge. What they have in common is that they are all rational presuppositions of rational methods. How does this illuminate the value or significance that we attach to knowledge? My proposal is that it provides the materials for a unified account of epistemic luck, according to which a belief is immune from epistemic luck if and only if it is rationally held when all of its rational presuppositions are satisfied. The problem of the value of knowledge was first posed by Plato in his dialogue, The Meno. Plato’s problem was to explain why we value knowledge over true belief, given that true belief is not sufficient for knowledge. The traditional answer, of course, was that knowledge is rationally justified, but in the light of the Gettier-style cases, the problem can be reformulated. The post-Gettier problem is to explain why we value knowledge over justified, true belief, given that the latter is not sufficient for knowledge.19 In response, a number of philosophers have emphasized the reliability of knowledge in one way or another.20 However, in the light of the SuperGettier cases, even this kind of response cannot be fully satisfying, since we need to explain why we value knowledge over true belief that is both rational and reliable, given that the latter is also insufficient for knowledge. This is the Super-Gettier version of the Value Problem.

19 20

See Kaplan (1985) for this challenge. For instance, Craig (1990) argues that the reliability of knowledge makes it a more valuable source of information in social contexts, whereas Williamson (2000) argues that the reliability of knowledge gives it a more robust role in explaining action.

145

What the Super-Gettier cases show is that even a true belief that is both rational and reliable can be infected with epistemic luck if it is not the case that its rational presuppositions are satisfied. According to my proposal, a belief is immune from epistemic luck, and so amounts to knowledge, if and only if it is rationally held when all of its rational presuppositions are satisfied. We value knowledge because it is a certain kind of cognitive achievement – it is the kind of cognitive achievement that results when a subject’s beliefs achieve the aim of truth in a way that is immune from epistemic luck.

146

BIBLIOGRAPHY

Alston, William (1988) “An Internalist Externalism,” Synthese 74: 265-283. Bermudez, Jose Luis (2003) Thinking Without Words, Oxford University Press. Bird, Alexander (1998) “Dispositions and Antidotes,” The Philosophical Quarterly 47: 143-158. Bird, Alexander (2005) “The Dispositionalist Conception of Laws,” Foundations of Science 10: 353-370. Block, Ned (1978) “Troubles with Functionalism,” reprinted in ed. Block (1980) Readings in the Philosophy of Psychology, volume I, Harvard University Press. Block, Ned (1995) “On a Confusion About a Function of Consciousness,” reprinted in eds. Block, Flanagan and Guzeldere (1997). Block, Ned (2002) “The Harder Problem of Consciousness,” The Journal of Philosophy 99: 391-425. Boghossian, Paul (1994) “The Transparency of Mental Content” in Philosophical Perspectives 8: 33-50. Boghossian, Paul (2001) “How Are Objective Epistemic Reasons Possible?” Philosophical Studies 106: 1-40. Boghossian, Paul (2003) “Blind Reasoning,” Proceedings of the Aristotelian Society, Supplementary Volume 77: 225-248.

147

Bonjour, Laurence (1985) The Structure of Empirical Knowledge, Harvard University Press. Bonjour, Laurence and Sosa, Ernest (2003) Epistemic Justification, Oxford: Blackwell. Brewer, Bill (1995) “Mental Causation: Compulsion by Reason,” Proceedings of the Aristotelian Society, Supplementary Volume, 64: 237-253. Brewer, Bill (1999) Perception and Reason, Oxford University Press. Burge, Tyler (1986) “Intellectual Norms and the Foundations of Mind,” Journal of Philosophy, 83: 697-720. Burge, Tyler (1996) “Our Entitlement to Self-Knowledge,” reprinted in eds. Ludlow and Martin (1998) Externalism and Self-Knowledge, CSLI, Stanford. Burge, Tyler (1997) “Two Kinds of Consciousness,” in eds. Block, Flanagan and Guzeldere, The Nature of Consciousness, MIT Press. Burge, Tyler (1998) “Reason and the First-Person,” in eds. McDonald, Smith and Wright, Knowing Our Own Minds, Oxford University Press. Campbell, John (1999) “Schizophrenia, The Space of Reasons, and Thinking as a Motor Process,” The Monist 82: 609-625. Campbell, John (2002) Reference and Consciousness, Oxford University Press. Chalmers, David (1996) The Conscious Mind, Oxford University Press. Chomsky, Noam (1965) Aspects of the Theory of Syntax, MIT Press. Chomsky, Noam (1975) Reflections on Language, Pantheon: New York.

148

Cohen, Stuart and Lehrer, Keith (1983) “Justification, Truth and Coherence,” Synthese 55: 191-127. Conee, Earl and Feldman, Richard (1998) “The Generality Problem for Reliabilism,” Philosophical Studies 89: 1-29. Conee, Earl and Feldman, Richard (2001) “Internalism Defended,” reprinted in ed. Kornblith (2001) Epistemology: Internalism and Externalism, Oxford: Blackwell. Craig, Edward (1990) Knowledge and the State of Nature, Oxford University Press. Dancy, Jonathan (2000) Practical Reality, Oxford University Press. Davidson, Donald (1980) Essays on Actions and Events, Oxford: Clarendon Press. Davies, Martin (1987) “Tacit Knowledge and Semantic Theory: Does a Five Percent Difference Matter?” Mind 96: 441-462. Davies, Martin (1989) “Tacit Knowledge and Subdoxastic States,” in ed. George, Reflections on Chomsky, Oxford: Blackwell. Davies, Martin (1991) “Aunty’s Own Argument for the Language of Thought,” in eds. Ezquerro and Larrazabal, Cognition, Semantics and Philosophy, Dordrecht: Kluwer. Davies, Martin and Humberstone, Lloyd (1980) “Two Notions of Necessity,” Philosophical Studies 38: 1-30. Dennett, Daniel (1969) Content and Consciousness, London: Routledge. Dretske, Fred (2006) “Perception without Awareness,” in eds. Gendler and Hawthorne, Perceptual Experience, Oxford University Press.

149

Eilan, Naomi (1998) “Perceptual Intentionality, Attention and Consciousness,” in ed. O’Hear, Contemporary Issues in the Philosophy of Mind, Cambridge University Press. Evans, Gareth (1981) “Semantic Structure and Tacit Knowledge,” in eds. Holtzman and Leich, Wittgenstein: To Follow a Rule, London: Routledge. Evans, Gareth (1982) The Varieties of Reference, Oxford University Press. Fine, Kit (1994) “Essence and Modality,” Philosophical Perspectives 8: 1-16. Fine, Kit (1995) “Senses of Essence,” in ed. Sinnott-Armstrong, Modality, Morality and Belief, Cambridge University Press. Fodor, Jerry (1975) The Language of Thought, Harvard University Press. Fodor, Jerry (1983) The Modularity of Mind, MIT Press. Fodor, Jerry (1987) Psychosemantics, MIT Press. Fodor, Jerry (2004) “Having Concepts: A Brief Refutation of the Twentieth Century,” Mind and Language 19: 29-47. Gettier, Edmund (1963) “Is Justified True Belief Knowledge?” Analysis 23:121-123. Goldie, Peter (2000) The Emotions, Oxford University Press. Goldman, Alvin (1976) “Discrimination and Perceptual Knowledge,” Journal of Philosophy 73: 771-791. Goldman, Alvin (1986) Epistemology and Cognition, Harvard University Press. Goldman, Alvin (1999) “A Priori Warrant and Naturalistic Epistemology,” Philosophical Perspectives 13: 1-28.

150

Goldman, Alvin (2001) “Internalism Exposed,” reprinted in ed. Kornblith, Epistemology: Internalism and Externalism, Oxford: Blackwell. Gopnik, Alison and Graf, Peter (1988) “Knowing How You Know: Children’s Understanding of the Sources of their Knowledge,” Child Development 59: 1366-1371. Greco, John (1999) “Agent Reliabilism,” Philosophical Perspectives 13: 273-296. Hetherington, Stephen (1998) “Actually Knowing,” The Philosophical Quarterly, 48: 453-469. Horgan, Terence and Tienson, John (2002) “The Intentionality of Phenomenology and the Phenomenology of Intentionality,” in ed. Chalmers, Philosophy of Mind: Classical and Contemporary Readings, Oxford University Press. Hurley, Susan (2006) “Making Sense of Animals,” in eds. Hurley and Nudds, Animal Rationality, Oxford University Press. Johnston, Mark (1992) “How to Speak of the Colours,” Philosophical Studies 68: 221-263. Kaplan, Mark (1985) “It’s Not What You Know That Counts,” Journal of Philosophy, 82: 350-363. Kauppinen, Antti (forthcoming) “The Rise and Fall of Experimental Philosophy”. Kelly, Sean (2004) “Reference and Attention: A Difficult Connection,” Philosophical Studies 120: 277-286. Kim, Jaegwon (1996) Philosophy of Mind, Westview Press.

151

Lewis, David (1972) “Psychophysical and Theoretical Identifications,” reprinted in his (1999). Lewis, David (1980) “Mad Pain and Martian Pain,” in ed. Block, Readings in the Philosophy of Psychology, volume I, Harvard University Press. Lewis, David (1994) “Reduction of Mind,” reprinted in his (1999). Lewis, David (1996) “Elusive Knowledge,” reprinted in his (1999). Lewis, David (1997) “Finkish Dispositions,” reprinted in his (1999). Lewis, David (1999) Papers in Metaphysics and Epistemology, Cambridge University Press. Loar, Brian (1981) Mind and Meaning, Cambridge University Press. Loar, Brian (2003) “Phenomenal Intentionality as the Basis of Mental Content,” in eds. Hahn and Ramberg, Reflections and Replies: Essays on the Philosophy of Tyler Burge, MIT Press. Marr, David (1982) Vision, New York: Freeman. Martin, C. B. (1994) “Dispositions and Conditionals,” Philosophical Quarterly 44:18. Martin, Michael (2006) “On Being Alienated,” in eds. Gendler and Hawthorne, Perceptual Experience, Oxford University Press. McDowell, John (1985) “Functionalism and Anomalous Monism,” reprinted in his (1998). McDowell, John (1994a) “The Content of Perceptual Experience,” reprinted in his (1998).

152

McDowell, John (1994b) Mind and World, Harvard University Press. McDowell, John (1995) “Knowledge and the Internal,” reprinted in his (1998). McDowell, John (1998) Meaning, Knowledge and Reality, Harvard University Press. McGinn, Colin (1997) “Consciousness and Content,” in eds. Block, Flanagan and Guzeldere, The Nature of Consciousness, MIT Press. Millar, Alan (2004) Understanding People: Normativity and Rationalizing Explanation, Oxford University Press. Nagel, Thomas (1974) “What is it Like to Be a Bat?” The Philosophical Review 83: 435-450. Nozick, Robert (1981) Philosophical Explanations, Oxford: Clarendon Press. Peacocke, Christopher (1992) A Study of Concepts, MIT Press. Peacocke, Christopher (1994) “Content, Computation and Externalism,” Mind and Language 9: 303-335. Peacocke, Christopher (1996) “Entitlement, Self-Knowledge and Conceptual Redeployment,” reprinted in eds. Ludlow and Martin (1998) Externalism and Self-Knowledge, CSLI, Stanford. Peacocke, Christopher (1998) “Conscious Attitudes, Attention and Self-Knowledge,” in eds. Wright, Smith and Macdonald, Knowing Our Own Minds, Oxford University Press. Peacocke, Christopher (2004) The Realm of Reason, Oxford University Press.

153

Pietrowski, Paul and Rey, Georges (1995) “When Other Things Aren’t Equal: Saving Ceteris Paribus Laws From Vacuity,” British Journal for the Philosophy of Science 46: 81-110. Pritchard, Duncan (2005) Epistemic Luck, Oxford University Press. Pryor, James (2000) “The Sceptic and the Dogmatist,” Nous 34: 517-549. Pryor, James (2001) “Highlights of Recent Epistemology,” British Journal for the Philosophy of Science 52: 95-124. Pryor, James (2004) “What’s Wrong With Moore’s Argument?” Philosophical Issues 14: 349-378. Pryor, James (2006) “Hyper-Reliability and A Priority,” Proceedings of the Aristotelian Society 106. Putnam, Hilary (1975) “The Meaning of ‘Meaning’,” reprinted in his Mind, Language and Reality: Philosophical Papers, vol.3, Cambridge University Press. Putnam, Hilary (1981) Reason, Truth and History, Cambridge University Press. Rawls, John (1971) A Theory of Justice, Harvard University Press. Rorty, Richard (1979) Philosophy and the Mirror of Nature, Princeton University Press. Rosenthal, David (1997) “A Theory of Consciousness,” in eds. Block, Flanagan and Guzeldere, The Nature of Consciousness, MIT Press. Russell, Bertrand (1948) The Analysis of Human Knowledge, New York: Allen and Unwin.

154

Searle, John (1990) “Consciousness, Explanatory Inversion and Cognitive Science,” Behavioural and Brain Sciences 13: 585-96. Shope, Robert (1983) The Analysis of Knowing: A Decade of Research, Princeton University Press. Siewert, Charles (1998) The Significance of Consciousness, Princeton University Press. Sosa, Ernest (1991) Knowledge in Perspective, Cambridge University Press. Sosa, Ernest (1999) “How to Defeat Opposition to Moore,” Philosophical Perspectives 13: 141-154. Sosa, Ernest (2003) “Beyond Internal Foundations to External Virtues,” in Bonjour and Sosa (2003) Epistemic Justification, Oxford: Blackwell. Stich, Stephen (1978) “Beliefs and Subdoxastic States,” Philosophy of Science 45: 499-518. Strawson, Peter (1962) “Freedom and Resentment,” Proceedings of the British Academy 48: 187-211. Unger, Peter (1968) “An Analysis of Factual Knowledge,” The Journal of Philosophy 65: 157-170. Velleman, David (2000) “On the Aim of Belief,” in Velleman, The Possibility of Practical Reason, Oxford University Press. Wedgwood, Ralph (2002a) “The Aim of Belief,” Philosophical Perspectives 16: 267-97.

155

Wedgwood, Ralph (2002b) “Internalism Explained,” Philosophy and Phenomenological Research 65: 349-369. Wedgwood, Ralph (2006) “The Internal and External Components of Cognition,” in ed. Stainton, Contemporary Debates in Cognitive Science, Oxford: Blackwell. Weinberg, Nichols and Stich (2001) “Normativity and Epistemic Intuitions,” Philosophical Topics 29: 429-460. Weiskrantz, Larry (1986) Blindsight, Oxford: Clarendon Press. Williamson, Timothy (2000) Knowledge and Its Limits, Oxford University Press. Williamson, Timothy (2005) “Armchair Philosophy, Metaphysical Modality and Counterfactual Thinking,” Proceedings of the Aristotelian Society 105: 1-23. Williamson, Timothy (forthcoming) “On Being Justified in One’s Head”. Wolf, Susan (1990) Freedom Within Reason, Oxford University Press. Wright, Crispin (1989) “Wittgenstein’s Rule-Following Considerations and the Central Project of Theoretical Linguistics,” in ed. George, Reflections on Chomsky, Oxford: Blackwell. Wright, Crispin (2001) “On Basic Logical Knowledge,” Philosophical Studies 106:41-85. Wright, Crispin (2004) “Warrant for Nothing (And Foundations For Free)?” Supplement to the Proceedings of the Aristotelian Society 78: 167-212.

156