EPISTEMOLOGICAL CONTEXTUALISM
Grazer Philosophische Studien INTERNATIONALE ZEITSCHRIFT FÜR ANALYTISCHE PHILOSOPHIE
GEGRÜNDET VON Rudolf Haller HERAUSGEGEBEN VON Johannes L. Brandl Marian David Leopold Stubenberg
VOL 69 - 2005
Amsterdam - New York, NY 2005
EPISTEMOLOGICAL CONTEXTUALISM
Edited by
MARTIJN BLAAUW
Die Herausgabe der GPS erfolgt mit Unterstützung des Instituts für Philosophie der Universität Graz, der Forschungsstelle für Österreichische Philosophie, Graz, und wird von folgenden Institutionen gefördert: Bundesministerium für Bildung, Wissenschaft und Kultur, Wien Abteilung für Wissenschaft und Forschung des Amtes der Steiermärkischen Landesregierung, Graz Kulturreferat der Stadt Graz
The paper on which this book is printed meets the requirements of “ISO 9706:1994, Information and documentation - Paper for documents Requirements for permanence”. Lay out: Thomas Binder, Graz ISBN: 90-420-1627-2 ISSN: 0165-9227 © Editions Rodopi B.V., Amsterdam - New York, NY 2005 Printed in The Netherlands
INHALTSVERZEICHNIS
TABLE OF CONTENTS
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
I
Duncan PRITCHARD: Neo-Mooreanism versus Contextualism
1
Krista LAWLOR: Living without Closure . . . . . . . . . . . . . . . . .
25
Patrick RYSIEW: Contesting Contextualism . . . . . . . . . . . . . . . .
51
Jessica BROWN: Comparing Contextualism and Invariantism on the Correctness of Contextualist Intuitions . . . . . . . . . . . . . . .
71
Adam LEITE: Some Worries for Would-be WAMmers . . . . . . . .
101
Martijn BLAAUW: Challenging Contextualism . . . . . . . . . . . . .
127
René VAN WOUDENBERG: Contextualism and the Many Senses of Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . . . .
147
Tim BLACK & Peter MURPHY: Avoiding the Dogmatic Commitments of Contextualism . . . . . . . . . . . . . . . . . . . . . . . . . .
165
Ram NETA: A Contextualist Solution to the Problem of Easy Knowledge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
183
Igor DOUVEN: A Contextualist Solution to the Gettier Problem
207
Peter BAUMANN: Varieties of Contextualism: Standards and Descriptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
229
Jesper KALLESTRUP: Contextualism between Scepticism and Common-Sense . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
247
This page intentionally left blank
Grazer Philosophische Studien 69 (2005), I–XVI.
INTRODUCTION: EPISTEMOLOGICAL CONTEXTUALISM Martijn BLAAUW University of Aarhus Can we know propositions about the external world? Can we, for instance, know that we have hands? The sceptic (Unger 1975) answers this question with a ‘no’: in order to know that we have hands, we have to eliminate the possibility that we are brains-in-vats, which cannot be done. The Moorean (Sosa 2000) answers this question with a ‘yes’: in order to know that we have hands, we have to eliminate the possibility that we are brains-in-vats, which can be done. The relevant alternatives theorist (Dretske 1970) answers this question with a ‘yes’ as well: in order to know that we have hands, we do not have to eliminate the — irrelevant — possibility that we are brains-in-vats at all. The contextualist (DeRose 1995; Cohen 2000; Lewis 1996), finally, answers this question with a ‘that depends’. We can know that we have hands in contexts in which ordinary standards for knowledge are in play; in such contexts, moreover, we can know that we are not brains-in-vats as well. But we cannot know that we have hands in contexts in which extraordinary sceptical standards for knowledge are in play; what is more, in such contexts we also fail to know that we are not brains-in-vats. Contextualist theories of knowledge are without a doubt among the most widely discussed theories in contemporary epistemology, especially since David Lewis published his seminal paper ‘Elusive Knowledge’ in 1996. It is an interesting question why contextualism enjoys this status. It might be due to the fact that contextualism promises to reconcile the sceptic and the Moorean in a way that respects the closure principle for knowledge. It might be due to the fact that most people share the intuition that contextual factors are important when it comes to evaluating whether someone has knowledge. It might also be due to the fact that epistemologists are getting tired of the ongoing debate about the correct analysis of knowledge and welcome any theory that departs from this. Interestingly, however, despite the enormous interest in contextualism,
most epistemologists seem to agree that extant contextualist theories do not give a correct account of ‘knows’.1 Up to now, I have left the term ‘contextualism’ unspecified. But specification is needed since ‘contextualism’ is a blanket term that covers several closely related, but distinct, theories. This is not the place to give an exhaustive overview of all the contextualist theories that have been offered in the epistemological literature. In what follows, then, I aim to give a very brief overview of the main contextualist positions that are of specific relevance to the papers in this collection.2 1. Contextualism in epistemology The versions of contextualism that have received most of the attention in the contemporary epistemological literature are the versions developed by DeRose (1992; 1995; 1999; 2002), Cohen (1999; 2000), Lewis (1996), and Williams (1992). These theories are related in at least the following two ways. First, they express the general position that what the truth-conditions of knowledge attributing and knowledge denying sentences are somehow is a contextual matter due to some characteristic of ‘knows’.3 Second, they are advertised as being able to solve the sceptical paradox (as well as related paradoxes, such as the lottery paradox) in a way that is superior to the other major contenders in the field.4 The version of the sceptical paradox contextualists usually respond to is the following one:
1. For a small sample of relevant critical literature, see Feldman (1999), Hawthorne (2004), Johnsen (2001) Klein (2000), Kornblith (2000), Pritchard (2001), Schaffer (2004), Schiffer (1996), Sosa (2000), Stanley (2004), and Williamson (2001). 2. For a useful general overview of key contextualist positions, see Norman (1999). 3. It is obvious that whether a subject knows a proposition depends on contextual features — this much will be accepted by the contextualist and non-contextualist alike. It is controversial, however, to argue that the context dependency of the truth-conditions of knowledge sentences can be traced back to ‘knows’ itself. 4. Such as safety accounts (Sosa 2000), sensitivity accounts (Nozick 1981), as well as other versions of contextualism.
II
(SP1) I cannot know that I am not a brain-in-a-vat. (SP2) If I cannot know that I am not a brain-in-a-vat, then I cannot know that I have hands. (SP3) I know that I have hands.5 These three intuitive claims are clearly incompatible. The contextualist, however, argues that the incompatibility is an illusion due to the fact that the contextual parameter is missing from these three claims. Once we add this parameter, the paradox vanishes. It is true that we cannot know that we are not brains-in-vats in sceptical contexts, and, given closure, it is true that we cannot know that we have hands in sceptical contexts either. But, the contextualist argues, this does not mean that we cannot know that we have hands in ordinary contexts where low standards for knowledge are in play. Nor does this mean that we cannot know that we are not brains-in-vats in ordinary contexts. The different contextualist theories are distinct, however, in that they take different positions as to (i) whose context is relevant, and (ii) what it is that changes in accordance with features of the context. As to (i), both DeRose, Cohen, and Lewis defend that the relevant context is the context of the attributer of knowledge. Thus, the context of the subject to whom knowledge is attributed is irrelevant when it comes to determining whether that subject knows or not. Appropriately, this version of contextualism is usually called ‘attributer contextualism’. The other position that is taken in response to (i) is ‘subject contextualism’. According to this view, it is the context of the subject to whom knowledge is attributed that is the relevant context. Thus, the context of the attributer of knowledge is irrelevant when it comes to determining whether the subject knows or not. This position has been defended by Williams (1992) and, more recently, by Hawthorne (2004).6 As to (ii), three main positions have been defended. The first position is ‘epistemic strength contextualism’. On this view, defended by DeRose, what changes in accordance with features of the conversational context is the strength of epistemic position that is required in order for the subject to count as knowing a proposition. In everyday contexts, the 5. Compare, for instance, DeRose (1995) for a formulation of this paradox. Recently, however, some commentators have proposed a stronger version of the sceptical argument in terms of an underdetermination principle. For discussion of this, see Brueckner (1994), Cohen (1998), and Pritchard (2005). 6. Though Hawthorne calls his position ‘subject sensitive invariantism’.
III
subject does not need to be in a very strong epistemic position to count as knowing a proposition. Accordingly, most knowledge attributions will be true in such contexts. In sceptical contexts, however, the subject does need to be in a very strong epistemic position to count as knowing a proposition. In such contexts, most knowledge attributions will be false. On DeRose’s view, the reason why the subject is unable to be in a very strong epistemic position with respect to a particular proposition p in sceptical contexts is that in such contexts it is required that the subject’s belief that p be sensitive. Roughly, a belief that p is sensitive if one wouldn’t continue to believe that p in case p were false. Crucially, however, the belief that I am not a brain-in-vat is not sensitive: I would continue to believe that I am not a brain-in-a-vat even if I were a brainin-a-vat. Combined with closure, this leads to the incapability to know ordinary propositions as well. The second position is what we may call ‘relevant alternatives contextualism’. On this view, defended by Lewis, what changes in accordance with features of the conversational context is the set of relevant alternatives the subject must be able to eliminate in order to count as knowing a proposition. A crucial characteristic of this view is that it is a form of infallibilism, though a restricted form of infallibilism. It is infallibilist insofar as the subject must eliminate all alternatives to a proposition. It is restricted insofar as what counts as ‘all’ alternatives to that proposition is a context-dependent matter. This immediately leads to an obvious question: what mechanism determines whether an alternative is relevant or irrelevant? Here, Lewis provides what may be the most detailed account of the mechanism of relevance to date, and proposes no less than seven rules of relevance: the rule of actuality, the rule of belief, the rule of resemblance, the rule of method, the rule of reliability, the rule of conservatism, and the rule of attention.7 The third position, finally, is what we may call ‘thresholds contextualism’. On this view, defended by Cohen, what changes in accordance with features of the conversational context is the threshold for justification one’s belief must reach in order for the belief to count as a ‘justified’ belief and, thus, as an instance of knowledge (if true). To illustrate this, consider that the notion of justification essentially is a degree notion: one can be justified to different degrees. Accordingly, the cut-off point between justified and unjustified beliefs fluctuates in accordance with the context. 7. For an interesting counterexample to Lewis’s theory, see Schaffer (2001).
IV
Interestingly, both Cohen and DeRose embed their version of contextualism in a linguistic thesis about ‘knows’: both argue that ‘knows’ is an indexical in that ‘knows’ has a fixed character but a shifty content. Cohen and DeRose differ, however, in what they take to be the indexical model for ‘knows’. According to DeRose, indexicals like ‘I’ and ‘here’ are the model for ‘knows’; according to Cohen, gradable terms like ‘tall’ or ‘flat’ are the model for ‘knows’. This linguistic claim has attracted a lot of attention.8 After having roughly sketched the contours of some key contextualist positions, I now turn to providing a detailed guide through the papers in this volume. 2. The papers This volume can be divided into four main sections. The papers that make up the first section (the papers by Duncan Pritchard and Krista Lawlor) respond to the contextualist (closure-preserving) dissolution of the sceptical paradox. The papers that make up the second section (written by Patrick Rysiew, Jessica Brown, Adam Leite, Martijn Blaauw, René van Woudenberg, and Tim Black & Peter Murphy) all deal with contextualism’s alleged capability of explaining our intuitions about puzzling cases such as the Bank Cases and the Airport Case. The general thrust of the papers in this section is to question to what extent these cases really support contextualism. In the papers in the third section (written by Ram Neta, Igor Douven, and Peter Baumann), novel versions of contextualism are proposed and are applied to important epistemological problems. Finally, the paper that makes up the fourth section (written by Jesper Kallestrup) ends this volume on a more positive note. Kallestrup argues that two recent arguments that have been advanced against contextualism cannot be accepted without problems. Section I: Contextualism, scepticism and closure The chief aim of Duncan Pritchard’s ‘Neo-Mooreanism versus Contextualism’ is to develop a neo-Moorean account of knowledge and to 8. See, for instance, Stanley (2004), and the papers by Van Woudenberg and Blaauw in this volume.
V
argue that this account has a dialectical advantage when compared to the versions of contextualism developed by DeRose, Cohen, and Lewis. The basic Moorean response to scepticism is to deny claim (SP1) of the sceptical paradox. But such a response faces at least two problems: (i) it does not offer any reasons not to accept scepticism, thus supporting a second-order scepticism; (ii) it seems to beg the question against the sceptic and thus is a dialectically inappropriate response. Pritchard argues that these problems can be evaded by modifying the basic Moorean response. To deal with the first problem, Pritchard proposes to add a safety-based theory of knowledge to the Moorean theory — on such an externalist theory, we can know the denials of sceptical hypotheses. To deal with the second problem, he argues that the inappropriateness of the Moorean response can be explained away by referring to the fact that the conversational implicatures concerning one’s evidential state can fluctuate in accordance with the context of attribution. It is inappropriate to assert ‘I know that I am not a brain-in-a-vat’ because this implies that one has special evidence in favour of the proposition that one is not a brain-in-a-vat, which one does not have. What results is a neo-Moorean view. However, Pritchard argues that this neo-Moorean account should be complicated in order to be able to capture the basic worry underlying the sceptical argument: that our beliefs about the external world are evidentially underdetermined. Here, Pritchard argues that we can make use of Neta’s recent contextualist account of ‘evidence’ without following him in accepting contextualism.9 Pritchard then compares this basic neo-Moorean view to contextualism and argues that contextualism is in a dialectically disadvantageous position: contextualism (but not neoMooreanism) makes a concession to the sceptic; contextualism makes two revisionistic claims (it rejects ‘invariantism’ and it denies the first claim of the sceptical paradox), whereas neo-Mooreanism just makes one (it only denies the first claim of the sceptical paradox); there is no conclusive linguistic evidence for contextualism. There are, Pritchard concludes, distinct advantages to being a neo-Moorean rather than a contextualist. In her paper ‘Living without Closure’, Krista Lawlor explores whether contextualists (and non-contextualists) should be committed to accepting the closure principle for knowledge. This is an important question, because if it appears that closure is false, then the sceptical 9. Also see the paper by Ram Neta in this volume.
VI
paradox fails to exists, and with it the contextualist response to it. In her paper ‘Living without Closure’, Lawlor argues that closure isn’t true of us in an important sense. Lawlor first explains what role closure plays in the contextualist response to scepticism, and concludes that accepting closure is crucial for the contextualist: without closure, there is no sceptical paradox to begin with. Lawlor then provides a version of ‘intuitive closure’ (IC) that formally states epistemic principles that can be expressed more informally, and, after a series of revisions, ends up with (IC*), concluding that life without IC* is unthinkable. We might now think that the contextualist should indeed be committed to accepting closure — as should we all. However, Lawlor continues to argue that it isn’t IC* that has played the key role in discussions about scepticism, but IC — and life without IC is thinkable. Indeed, Lawlor calls closure principles that can generate the sceptical paradox ‘strong closure’ principles and closure principles that cannot generate the sceptical paradox ‘weak closure’ principles, and argues that IC is a strong closure principle whereas IC* is a weak one. The argument of the paper now unfolds as follows: Lawlor argues that the contextualist — in responding to the sceptical paradox — is committed to accepting strong closure. However, strong closure principles are importantly not true of us, though weak closure principles are importantly true of us. But weak closure — the principle we should accept — isn’t able to generate the sceptical paradox. So in accepting the correct closure principle, the sceptical paradox vanishes, and with it, contextualism’s main motivation. Section II: Alternative explanations of the motivation for contextualism Alternative invariantist explanations of the intuitions about puzzle cases such as the Bank Case, the Airport Case, and the sceptical paradox, have quickly emerged. One of these explanations has been put forward by Patrick Rysiew (2001). The core of Rysiew’s proposal is to distinguish between the literal meaning of a sentence and the speaker-meaning of that same sentence. He argues that a sentence of the form ‘S knows that p’ literally expresses the proposition that S knows that p, and pragmatically conveys a proposition of the form ‘S’s epistemic position with respect to p is good enough to …’, where the ellipsis is completed in line with the purposes and interests of the speaker (e.g., ‘S’s epistemic position with respect to p is good enough for you to take my word on the
VII
matter’). Finally, Rysiew adds a relevant alternatives theory of knowledge to his ‘sophisticated invariantist’ framework and shows how his theory can explain the data that motivate contextualism. Essentially, Rysiew proposes that the assertion ‘I know that I am not a brain-in-vat’ is literally true. However, an utterance of this sentence will pragmatically convey the information that the subject can rule out the alternative that she is a brain-in-a-vat, which is false. DeRose (1999; 2002) has both expressed scepticism about the prospects of sophisticated invariantism in general and about Rysiew’s approach in particular. In his paper in the present volume — ‘Contesting Contextualism’— Rysiew responds to DeRose’s criticisms. In her ‘Comparing Contextualism and Invariantism on the Correctness of Contextualist Intuitions’, Jessica Brown aims to evaluate to what extent four invariantist theories can accommodate the following four intuitions about the contextualist puzzle cases: – the attribution intuition:
the intuition that it seems correct to attribute knowledge in low-standards contexts and that it seems correct to fail to attribute knowledge in high-standards contexts. – the denial intuition: the intuition that it seems correct to deny knowledge of a subject in highstandards contexts, and that it seems correct not to deny knowledge of a subject in low-standards contexts. – the assertion intuition: the intuition that it seems correct to assert a proposition in low-standards contexts and that it seems incorrect to assert a proposition in high-standards contexts. – the practical reasoning intuition: the intuition that it seems correct for a subject to practically reason from a proposition in low-standards contexts and that it seems incorrect for a subject to practically reason from a proposition in high-standards contexts.
VIII
Brown considers four invariantist ways to accommodate these intuitions: pure error theory, belief removal theories, warranted assertability maneuvers, and Williamson’s luminosity account. Brown concludes that it is only the pure error theory that provides an uncharitable treatment of all four contextualist intuitions. Furthermore, a warranted assertability maneuver can charitably treat the attribution and assertion intuition, while both the belief-removal account and Williamson’s luminosity account can charitably treat all intuitions but the denial intuition. Brown concludes that it isn’t obvious at all that, with respect to accommodating these intuitions, contextualism has the upper hand over invariantism. Contrary to Rysiew and Brown, Adam Leite is more pessimistic about the prospects of using a warranted assertability maneuver (or WAM) in order to explain our intuitions about the puzzle cases. Central in a WAM is that assertions such as ‘I don’t know that the bank will be open tomorrow’ that are made in high-standards contexts, seem true though they actually are false but warranted. Leite argues that in uttering such a sentence the speaker has intentionally asserted a sentence that she takes to literally express a falsehood and which she nonetheless intends to communicate something true and relevant. But now Leite argues that it is extremely difficult for the hearer of this sentence to get from the thought that the speaker has meant to communicate something else to what it is that the speaker intended to communicate. Indeed, Leite points out that there seems to be something unintelligible in the speaker’s behaviour: she is not communicating in a fully rational way. Leite then examines whether using Grice’s conversational maxims and Bach’s theory of conversational impliciture could solve this problem, and finds both options unsatisfying. He concludes that the prospects of a WAM against contextualism seem dim. In his ‘Challenging Contextualism’, Martijn Blaauw argues against two contextualist theses. First, he argues against the thesis that ‘knows’ is an indexical and concludes that this position isn’t plausible from a linguistic point of view. A variety of indexicality tests show that putting ‘knows’ in the same category as indexicals like ‘tall’, ‘I’, or ‘to like’ is unsupported by the relevant linguistic data. Second, he argues against the thesis that the truth-conditions of knowledge sentences fluctuate in accordance with features of the context. This thesis is supposed to be the best explanation of such puzzle cases as the Bank Cases. Blaauw shows that this explanation is rivalled: epistemological contrastivism — which is an invariantist position — can also explain these puzzle cases without
IX
being linguistically implausible. On contrastivism, knowledge sentences of the form ‘S knows that p’ are elliptical for contrastive knowledge sentences of the form ‘S knows that p rather than Q’, where ‘Q’ stands for a set of contrastive propositions. If we plug the set of contrastive propositions into the cases that are supposed to motivate contextualism (the Bank Cases, for instance), it will appear, or so Blaauw argues at any rate, that there is no incompatibility between the low-standards knowledge attribution and the high-standards knowledge denial. In René van Woudenberg’s contribution to this collection, entitled ‘Contextualism and the Many Senses of Knowledge’, the contextualist idea that knowledge is an indexical is also challenged. Instead, Van Woudenberg argues for the position that knowledge is a polysemous word: not only does knowledge have a shifting content, it also has a shifting character. In effect, this position is a form of epistemic pluralism akin to Alston’s epistemic desiderata approach: there isn’t one knowledge relation, but there are many knowledge relations. Crucially, moreover, Van Woudenberg also argues that this position is an invariantist position that can help to explain our intuitions about the puzzle cases that are the motivation for contextualism. Van Woudenberg gives two arguments in favour of his epistemic pluralism. First, it is supported by how we use ‘knows’ in ordinary cases: in some cases, for instance, knowledge seems equal to ‘justified true belief’, in other cases to ‘certain true belief’, and in yet other cases to just ‘true belief’. Second, Van Woudenberg argues that the never-ending discussion about the proper analysis of knowledge can also be seen as evidence for epistemic pluralism: there just is a plethora of epistemic qualities a belief can have. No wonder we were unable to isolate one set of conditions for ‘knowledge’. In ‘Avoiding the Dogmatic Commitments of Contextualism’, Tim Black and Peter Murphy also argue that the puzzle cases motivate invariantism not contextualism. They start out by showing that there are three positions we could take with respect to the puzzle cases: the highstandards position, according to which one high standard governs both cases; the low-standards position, according to which one low standard governs both cases; and the contextualist position, according to which different standards (one high and one low) govern the two cases. Stewart Cohen has argued against the low-standards view, and this is where Black and Murphy begin their argument. Illustrating the problem with Cohen’s Airport Case, the defender of the low-standards view must hold that utterances such as ‘We do not know that the flight stops in Chicago’,
X
uttered in a high-standards context, are actually false. And according to Cohen, this means that if Mary and John (the two subjects in the Airport Case) want to take into account their relevant epistemic situation, they must describe their situation as follows: ‘We know that the flight stops in Chicago, but still, we need to check further.’ Cohen thinks that such sentences are conceptually deficient and he furthermore thinks that the deficiency is to be located within the second conjunct of the above claim. In effect, this means that Cohen argues that if one knows, one does not need to check further. Black and Murphy argue that this commits Cohen to apathetic dogmatism, which is the view that knowers are permitted to be indifferent towards the active pursuit of new evidence. Furthermore, Black and Murphy argue that apathetic dogmatism is indefensible and propose an invariantist pragmatic way to avoid contextualism’s dogmatism. In essence, they argue that ‘S knows that p and S needs to check further’ sounds odd because the second conjunct pragmatically implicates that S needs to check further for the sake of knowing that p. And they propose that one can cancel this implication by asserting that S needs to check further not in order to know that p but in order to achieve another goal, for instance, in order to receive a research grant. This opens the way to accepting a non-sceptical version of invariantism. Section III: New versions of contextualism Recent years have seen the development of contextualist theories that are different from the ones proposed by DeRose, Cohen, and Lewis. One of these new theories is Neta’s evidential contextualism. In his ‘A Contextualist Solution to the Problem of Easy Knowledge’, Neta applies his type of contextualism to the problem of easy knowledge. This problem essentially is an objection to the ‘doctrine of basic knowledge’, which says that we can know propositions on the basis of a particular source without knowing that the source is reliable. According to the problem of easy knowledge, this doctrine cannot explain why it seems impossible to obtain knowledge on the basis of certain kinds of inferences, for instance, inferences that use one-premise closure and bootstrapping reasoning. After rejecting Cohen’s solution to the problem of easy knowledge, Neta shows that his version of evidential contextualism can solve it. Crucial in Neta’s contextualism is the idea that ascriptions of evidence are semantically context-sensitive. What may count as evidence in some contexts, may not count as evidence in other contexts. Applying eviden-
XI
tial contextualism to one-premise closure inferences, Neta shows (i) that in some cases we can achieve ‘easy knowledge’ by means of deduction, and (ii) that the reason we think that there is something wrong with certain cases is that we have moved into a context in which there are counterpossibilities that are uneliminated by one’s evidence. Applying evidential contextualism to bootstrapping inferences, Neta shows (i) that in some cases we can achieve ‘easy knowledge’ by means of bootstrapping inferences, and (ii) that the reason we think that there is something wrong with certain bootstrapping inferences is that there are hypotheses in play that our evidence cannot eliminate. Igor Douven presents another new version of contextualism in his paper ‘A Contextualist Solution to the Gettier Problem’. Douven argues that epistemologists have been too quick in abandoning the tripartite analysis of knowledge in terms of justified true belief. In his contribution to this volume, he develops a contextualized notion of deontic justification and argues that on this conception of justification many supposed Gettier cases will appear not to be cases in which the subject has a justified true belief at all. Furthermore, this conception of justification also shows that those supposed Gettier that are cases of justified true belief are actually cases of knowledge. Key to Douven’s suggestion is that it is a contextual matter whether someone has done her epistemic duty with respect to a proposition. To give theoretical support to this idea, he introduces the notion of ‘epistemic import’ (which measures how important it is for a particular subject to know whether or not a proposition is true), combines this notion with the expenditure of effort, time and money it takes to acquire better evidence for a proposition, and proposes that whether a body of evidence justifies (in the deontological sense) a proposition depends both on epistemic import and expenditure. Douven then applies this theory to Gettier cases. It appears, or so Douven argues, that some alleged Gettier cases are not cases of justified true belief after all. The famous stopped clock example, for instance, is not a case of justified true belief — given that it is important for the subject to know what time it is, and given that it is relatively easy to obtain further evidence, we must conclude that the subject did not fulfill her epistemic duty: she didn’t check for further evidence. Second, Douven concludes that some alleged Gettier cases in which the subject has a justified true belief and supposedly does not know, are actually cases where the subject has a justified true belief and does know. The fake barn case is used as an example here: Henry has a justified true belief that he sees a barn,
XII
and he knows this as well: given that this belief has little import, Henry does not need to obtain further evidence. Douven ends by considering the place of epistemic luck in his theory of justification and proposes that the fact that his theory must allow for some form of epistemic luck isn’t a shortcoming, provided that the luck in question is deserved: you fulfilled your epistemic obligations though the evidence in favour of your belief is adequate yet misleading. Peter Baumann, in his paper ‘Varieties of Contextualism: Standards and Descriptions’, begins by considering ‘standards’ contextualism, e.g., the type of contextualism that says that the standards for knowledge are context-dependent so that in different contexts the subject needs to be able to eliminate different sets of alternative possibilities. Baumann identifies various problems with this view, the most serious of them being that it is difficult to rank alternative possible worlds. Ranking possible worlds usually proceeds by determining to what extent possible worlds are similar to each other. But, Baumann worries, there is no context-independent way to determine this. What ranking we choose depends on contextual factors. And this, in turn, collapses the contextualist position, especially since, as Baumann argues, various replies to this worry are unconvincing. Baumann then goes on to focus on a new type of contextualism that is designed to be a solution to the ‘reference class’ problem for reliabilist theories of knowledge. This problem has to do with the reliability of a given method as measured against the background of a particular reference class. For instance, reliability is a time-sensitive phenomenon: one can be reliable during one time-interval, but unreliable during another time-interval, and the problem then becomes how to choose which time-interval (and thus, which reference class) is relevant. Baumann proposes that what the relevant time-interval is depends on the context of the attributer of knowledge. And thus, whether a knowledge sentence is true will depend on the context of the attributer of knowledge as well. Section IV: Defending contextualism Jesper Kallestrup, in the final contribution to this volume entitled ‘Contextualism Between Scepticism and Commonsense’, tries to defend DeRose-type contextualism against objections that have been put forward by Timothy Williamson and Crispin Wright. Williamson argues that contextualists are not able to assert their own view: in the
XIII
contextualist context (e.g., the context in which contextualists propose their theory), contextualists are committed to holding ‘p, but it is not known that p’, which is an instance of a Moorean paradox. Kallestrup responds to this complaint by saying that the contextualist should reject that the contextualist context is a sceptical context. Furthermore, he argues that the contextualist can reject this assumption because it conflicts with crucial contextualist commitments, in particular with the commitment that the contextualist wants to side with commonsense against scepticism. The second argument to which Kallestrup responds is put forward by Wright. Wright argues that the contextualist, while being in the contextualist context, knows that he knows in an ordinary context that p. Since knowledge is factive, he also knows that p is true in that ordinary context. And since contextualism does not imply relativism about truth, he knows that p is true in the contextualist context as well. And this, Wright claims, counters the contextualist hopes of providing a midpoint between scepticism and commonsense — we must side with the commonsense philosopher against scepticism. Kallestrup responds to this argument by saying that it misinterprets contextualism: contextualism is not the view that theories of knowledge are context-dependent; it is the view that ‘knowledge’ itself is context-dependent. Thus, Wright’s argument can be evaded. 3. Acknowledgements Some of the papers in this volume (the ones written by Baumann, Brown, Van Woudenberg, and Blaauw) were presented in an earlier form at the conference on epistemological contextualism that was held at the Vrije Universiteit Amsterdam on October 19th and 20th 2004. I would like to thank the board of the Department of Philosophy of the Vrije Universiteit for supporting this conference financially and Drs. Kiki Berk for providing invaluable practical assistance during the conference. I would also like to thank the people at Grazer Philosophische Studien for making the editing of this volume such a pleasurable experience. Finally, I would like to thank all the authors involved for their cooperation and enthusiasm.
XIV
REFERENCES Brueckner, A. (1994). ‘The Structure of the Skeptical Argument’, Philosophy and Phenomenological Research 54, 827–835. Cohen, S. (1998). ‘Two Kinds of Sceptical Argument’, Philosophy and Phenomenological Research 58, 143–159. — (1999). ‘Contextualism, Skepticism, and the Structure of Reasons’, Philosophical Perspectives 13, 58–89. — (2000). ‘Contextualism and Skepticism’, Philosophical Issues 10, 94–107. DeRose, K. (1992) ‘Contextualism and Knowledge Attributions’, Philosophy and Phenomenological Research 52, 913–929. — (1995). ‘Solving the Sceptical Problem’, Philosopical Review 104, 1–52. — (1999). ‘Contextualism: An Explanation and Defense’, The Blackwell Guide to Epistemology, (ed.) E. Sosa and J. Greco, 185, 203, Blackwell Publishing, Oxford. — (2002). ‘Knowledge, Assertion, and Context’, Philosophical Review 111, 167–203. Dretske, F. (1970). ‘Epistemic Operators’, The Journal of Philosophy 67, 1007–1023. Feldman, R. (1999). ‘Contextualism and Scepticism’, Philosophical Perspectives 13, 91–114. Hawthorne, J. (2004). Knowledge and Lotteries, Clarendon Press, Oxford. Johnsen, B. (2001). ‘Contextualist Swords, Sceptical Plowshares’, Philosophy and Phenemenological Research 62, 385–406. Klein, P. (2000). ‘Contextualism and the Real Nature of Academic Skepticism’, Philosophical Issues 10, 108–116. Kornblith, H. (2000). ‘The Contextualist Evasion of Epistemology’, Philosophical Issues 10, 24–32. Lewis, D. (1996). ‘Elusive Knowledge’, Australasian Journal of Philosophy 74, 549–567. Norman, A.P. (1999). ‘Contextualism: Its Past, Present, and Prospects’, Philosophia 27, 1–28. Nozick, R. (1981). Philosophical Explanations, Oxford University Press, Oxford. Pritchard, D. (2001). ‘Contextualism, Scepticism, and the Problem of Epistemic Descent’, Dialectica 55, 327–349. — (2005). ‘The Structure of Sceptical Arguments’, The Philosophical Quarterly 55, 37–53. Rysiew, P. (2001). ‘The Context-Sensitivity of Knowledge Attributions’, Noûs 35, 477–514.
XV
Schaffer, J. (2001). ‘Knowledge, Relevant Alternatives, and Missed Clues’, Analysis 61, 202–208. — (2005). ‘Skepticism, Contextualism, and Discrimination’, Philosophy and Phenomenological Research 69, 138–155. Schiffer, S. (1996). ‘Contextualist Solutions to Scepticism’, Proceedings of the Aristotelian Society 96, 317–333. Sosa, E. (2000). ‘Skepticism and Contextualism’, Philosophical Issues 10, 1–19. Stanley, J. (2004). ‘On the Linguistic Basis for Contextualism’, Philosophical Studies 119, 119–146. Unger, P. (1975). Ignorance: A Case for Scepticism, Clarendon Press, Oxford. Williams, M. (1992). Unnatural Doubts, Blackwell Publishers, Oxford. Williamson, T. (2001). ‘Comments on Michael Williams’ ‘Contextualism, Externalism and Epistemic Standards’, Philosophical Studies 103, 25–33.
XVI
Grazer Philosophische Studien 69 (2005), 1–24.
NEO-MOOREANISM VERSUS CONTEXTUALISM Duncan PRITCHARD University of Stirling Summary Attributer contextualism has undoubtedly been the dominant anti-sceptical theory in the recent literature. Nevertheless, this view does face some fairly serious problems, and it is argued that when the contextualist position is compared to a refined version of the much derided ‘Moorean’ response to scepticism, then it becomes clear that there are distinct advantages to being a neo-Moorean rather than a contextualist.
1. Introductory remarks —scepticism and Mooreanism When we teach epistemology to philosophy students for the first time, we tend to begin by introducing them to some form of the sceptical problem in order to grab their attention. The reason why we do this is obvious, since sceptical arguments, when well constructed at any rate, are paradoxes— that is, they proceed from apparently uncontentious premises and seemingly flawless reasoning to derive an utterly unacceptable and counterintuitive conclusion. Consider, for example, the following formulation of the sceptical problem, which has become fairly standard in the contemporary literature: S is unable to know that she’s not, say, a brain-in-a-vat (BIV). If S is unable to know that she’s not a BIV, then she is unable to know that she’s currently seated (and much else besides). Hence: SC S is unable to know that she’s currently seated (and much else besides).1 S1 S2
1. For a survey of the contemporary literature on scepticism which discusses in detail how this formulation of the sceptical problem is central to the recent debate on this topic, see Pritchard (2002c).
This argument is clearly valid. Moreover, both of its premises are highly intuitive. How could I know that I am not a BIV given that this errorpossibility is standardly described in such a way that one is unable to tell the difference between one’s experience of, say, being seated, and of merely being ‘fed’ the experience of being seated while in fact being envatted? Furthermore, given the highly plausible closure principle for knowledge— roughly, that if one knows one proposition and knows that it entails a second proposition, then one also knows the second proposition— the second premise ought to be equally unproblematic. Since one cannot, it seems, accept the conclusion, however, the puzzle is complete. What is interesting for my purposes here is that the initial reaction of most students to this problem is to employ a type of anti-sceptical tactic that is usually associated with G. E. Moore. In essence, what they argue is that of course they know that they are presently seated, and from this intuitive ‘fact’ they then directly infer that the first premise is false and that they do know that they are not BIVs after all. Although this line of argument doesn’t quite parallel Moore’s (1925; 1939) famous remarks on scepticism, it does share several key features. To begin with, the claim is that one can know the denial of the relevant sceptical hypothesis, despite sceptical claims to the contrary. (Moore famously argued that we can know that there is an external world, thereby claiming to know the denial of the sceptical/idealist hypothesis that there is no such world). Moreover, this is also meant to be a ‘commonsense’ response to scepticism— and in this sense a pre-theoretical response as well— in that the claim is that the sceptic hasn’t given us sufficient reasons to worry about whether we have knowledge in order to force us to engage with the problem theoretically. Instead, for all the sceptic has said, we can stick to our commonsense pre-theoretical intuitions and ignore the sceptical challenge with impunity; a very Moorean style of response to scepticism. Another salient feature of this commonsense line of argument that also mirrors the Moorean line is that it consists of one straightforwardly claiming to have the contested knowledge of both ‘everyday’ propositions— such as that one is seated (or that one has hands)— and also antisceptical propositions— such as that one is not a BIV (or that there is an external world)— where the epistemic status of the latter knowledge is dependent upon the former. That the knowledge is claimed in this way indicates that one believes that one has a certain special authority
2
about the subject matter in question. Moreover, note that the inferential move from everyday to anti-sceptical knowledge indicates that it is not part of this sort of response to scepticism to worry about the second premise of the sceptical argument— and thus to question the closure principle that is typically thought to underpin that premise— since this argument straightforwardly contraposes on the conditional contained within that premise. Henceforth, we will refer to this commonsense anti-sceptical approach as the ‘Moorean’ line, without concerning ourselves with the further question of just how much this approach has in common with Moore’s stance on scepticism. For our purposes we can let the tag ‘Moorean’ bear a relatively loose affiliation to the relevant work by Moore.2 Of course, most students backtrack from this commonsense line fairly quickly— possibly in response to prompting on our part as lecturers— and move instead towards a more theoretical response to the problem. There are a number of reasons for this, though surveying them all would take us too far afield.3 Instead, I want to focus on two interconnected problems which relate to the dialectical impropriety of this ‘Moorean’ style of argument as it stands. The first is that the Moorean line merely seems to offer us a draw with the sceptic, when what we wanted was a victory. Notice that the Moorean response to the sceptical argument is simply a modus tollens inference in response to the corresponding modus ponens inference employed by the sceptic, where both inferences have some intuitive force. The trouble is that since the Moorean response does not offer us any reason to prefer the modus tollens inference over the corresponding sceptical modus ponens inference— the Moorean does not try to explain away the intuitiveness of the sceptical argument, for example, by, say, offering a further theoretical story to back-up the general anti-sceptical line— it remains that we have no good reason to not to be sceptics, and this second-order sceptical conclusion seems almost as bad as its first-order cousin.4 The second difficulty is that the Moorean argument seems to be 2. For a useful overview of Moore’s contribution to epistemology, which hints at some differences between his approach to scepticism and the ‘Moorean’ line just described, see Baldwin (1992). 3. I survey the problems facing the Moorean response in detail in the ‘sister’ paper to this article, Pritchard (forthcoming b). 4. Indeed, some have argued that it is just as bad. See, for example, Wright (1991).
3
question-begging, in that it takes as a premise a claim— that one knows a paradigm ‘everyday’ proposition, such as that one is presently seated— which has already been called into question by the sceptical argument, the intuitiveness of which is not itself disputed by the Moorean. Given that the sceptic has offered an intuitive argument which calls this claim into question, however, it seems entirely dialectically inappropriate to cite this claim as a premise in one’s anti-sceptical reasoning. These two key problems, coupled with the other difficulties that face the Moorean response (the lack of a supporting epistemology to explain how we could know the denials of sceptical hypotheses for example), have ensured that the Moorean response has been quickly regarded by those engaging with scepticism for the first time as simply naïve. It is thus little wonder that the focus of attention soon moves to theoretical responses to the problem which make substantive epistemological claims, such as the kind of response offered by Fred Dretske (1970) and Robert Nozick (1981), who reject the closure principle (and thus the second premise of the sceptical argument). My interest here, however, is not with the arguments for the rejection of closure,5 or any of the other non-contextualist theoretical responses to the sceptical problem that have been presented in the recent literature, but rather with the response to scepticism offered by attributer contextualism (henceforth simply ‘contextualism’), a proposal that has been the focus of much of the recent work on scepticism. What I want to argue is that the problems facing this response indicate that we should look again at the Moorean treatment of scepticism. As we will see, the Moorean thesis, properly refined, has the resources within it to avoid some of the core problems facing contextualism, and is thus at a dialectical advantage relative to this view. Moreover, the Moorean can learn from the contextualist and exploit certain features of the contextualist position for her own ends. 2. Contextualism and scepticism Let us suppose, then, that closure is beyond reproach and that we have already considered the Moorean response to scepticism and found it 5. I have already discussed the arguments for non-closure at length elsewhere. See Pritchard (2002a; 2002b; 2005b, chapters 2, 3 & 6).
4
wanting. In such circumstances, it is easy to see the appeal of contextualism. In essence, this view holds that ‘knows’ is a context-sensitive term such that whether or not an assertion of an ‘ascription’ sentence which ascribes knowledge to an agent, such as ‘S knows that P’, is true depends on the epistemic standards in play in the context of utterance. Contextualists are therefore able to claim that while assertions of ascription sentences in everyday conversational contexts (henceforth simply ‘contexts’) where the epistemic standards are undemanding are generally true, in sceptical contexts in which we are taking the sceptical problem— and thus sceptical hypotheses— seriously, the epistemic standards get raised so that assertions of these same sentences will now express falsehoods. In this way, contextualists can hold that while it can be true to say of us from the perspective of an everyday context that we know that we are, say, currently seated, it can also be true to say of us from the perspective of a sceptical context (and where all that has changed is the context) that we do not know that we are currently seated. Furthermore, since contextualists retain the closure principle, it follows that in those contexts in which it is true to say of us that we know that we are not currently seated it will also be the case that we know that we are not, for example, BIVs (provided, of course, that we know the relevant entailment). Thus, since there are some epistemic standards relative to which we do know that we are not BIVs, it follows that contextualists, like Mooreans, reject the first premise that we are unable to know anti-sceptical propositions of this sort. This view— advanced by such figures as Keith DeRose (1995), David Lewis (1996), and Stewart Cohen (2000)— is attractive precisely because it accommodates the fact, so harmful to the Moorean response, that we find both the sceptic’s modus ponens argument and the corresponding modus tollens argument put forward by the Moorean compelling. These apparently irreconcilable facts about our epistemic intuitions in this regard can be made consistent with one another within the contextualist framework, since now there is a context in which the premises and the conclusions of the sceptical argument are all true, and also a context in which the premises and the conclusion of the Moorean argument are all true. We thus do not have to choose between two equally opposed sets of intuitions and thereby face the sceptical threat at second-order. Instead, we can simply explain away the problem of scepticism. There are problems with the view, however, three of which are particularly salient for our purposes. To begin with, note that the contex-
5
tualist treatment of scepticism is an essentially concessive one, in that it grants that there is a context in which what the sceptic claims is true and that this context is one in which more demanding epistemic standards are in play. The problem with this concession is that it raises the worry that perhaps it is the sceptic who is using the exacting epistemic standards, with the epistemic standards in play in everyday contexts reflecting merely an epistemic ‘looseness’ on our part. That is, the contextualist line prompts the thought that perhaps, strictly speaking, the sceptic is right after all, it is just that relative to the comparatively undemanding epistemic standards in play in everyday conversational contexts where we don’t concern ourselves with the austere epistemic requirements made by the sceptic we have a practice— useful, if not strictly correct— of treating ourselves as knowers. If there were a way of dealing with the problem without making the sort of concession that leads to this worry then this would naturally be preferable. Mooreanism is, of course, just such a view that doesn’t make a concession to the sceptic of this sort.6 The second problem is that contextualists make two revisionistic claims, even though one of these claims alone would seem to suffice to deal with the sceptical problem. Clearly, the main revisionistic aspect of contextualism is the contextualist thesis itself, which revises our natural non-contextualist (‘invariantist’) understanding of ‘knows’. The second revisionary aspect of the contextualist thesis is a by-product of the contextualist retention of closure, and this is the rejection of the first premise of the sceptical argument by allowing that we can know the denials of sceptical hypotheses after all (albeit only in undemanding contexts). The problem is, however, that if we can make sense of the denial of this premise then it is far from clear that we need to be contextualists, since the denial of this premise all by itself would suffice to block the sceptical argument. It is thus only if the denial of this premise is essentially connected to the wider contextualist thesis that this feature of the contextualist line does not undermine the general 6. In effect, the problem here is that contextualism seems to tempt one towards an infallibilist conception of knowledge, as defended, for example, by Unger (1971; 1975). Interestingly, in more recent work Unger (1984) has been more sympathetic to contextualism, though he maintains that there is no way of deciding between a context-sensitive understanding of ‘knows’ which limits the concession to scepticism, and an infallibilist treatment, which does not. As noted above, such second-order scepticism is at least almost as bad as its first-order cousin.
6
approach, and this is far from clear.7 Notice, however, that Mooreanism is precisely the sort of view that only makes the one revisionistic claim in this respect of denying the sceptic’s first premise. The third problem for contextualism which is salient for our purposes is that the linguistic evidence for the view seems at best inconclusive. We have already noted that there is an intuitive resistance to treating ‘knows’ as a context-sensitive term. More serious for the contextualist, however, is the fact that the apparent context-sensitivity that contextualists appeal to in order to motivate their view seems to be explicable in terms of a very different view which continues to treat ‘knows’ as an invariant term while offering a more nuanced account of the assertibility of sentences involving this term. In this way, one can account for why it might seem appropriate to assert a sentence involving this term in one context and yet also inappropriate to assert that same sentence in another context (even though all that has changed has been the context), without having to appeal to the context-sensitivity of ‘knows’. Indeed, one could simply treat both assertions as true (or both false for that matter). Moreover, since there is very little that is counterintuitive about the suggestion that it is sometimes inappropriate to assert a sentence which is literally true, this appeal to the context-sensitivity of the propriety of assertions seems to be a more natural way of accommodating apparent context-sensitivity than any appeal to the context-sensitivity of the terms involved. Clearly, such a view would be welcomed by the Moorean since it would enable the proponent of this position to explain away the apparent context-sensitivity of knowledge— and thereby also some of our sceptical intuitions— without conceding that ‘knows’ is a context-sensitive term (or conceding anything substantive to the sceptic for that matter). Although there are other problems with the contextualist thesis,8 what these three difficulties point to, as we will see, is a principled ground on which one could resurrect the Moorean anti-sceptical thesis in opposition to its contextualist counterpart. We will now consider what this view will involve in more detail. 7. There is a related problem in this regard which is the odd status of one’s knowledge of the denials of anti-sceptical propositions on the contextualist view. After all, one cannot truthfully say of oneself that one has this knowledge, since to do so would be to move to a sceptical conversational context in which what is said is now no longer true. Such knowledge is thus forever tacit, at least as far as self-ascription of knowledge goes. 8. I raise a further problem for the view in Pritchard (2001), for example.
7
3. Basic neo-Mooreanism We noted above that the chief problems relating to the Moorean position are related to its dialectical impropriety. One way in which the neo-Moorean proposal might begin to deal with this problem is by incorporating the kind of view just described which accepts that there is something context-sensitive about the propriety of assertions of ascription sentences but which accounts for that context-sensitivity in an anti-sceptical fashion by treating the problematic assertions of ascription sentences in sceptical contexts as being merely improper, rather than also false (as the contextualist claims). On this view, insofar as an agent has knowledge of a proposition then the assertion of an ascription sentence which ascribes that knowledge to the agent will be true in all contexts (just so long as all non-conversational factors remain constant), even though in some contexts, such as in sceptical contexts, this assertion may well be inappropriate. I specify one way in which this line of argument might run in Pritchard (2005a), where the view is spelt-out along Gricean lines such that the conversational implicatures generated by the assertion of an ascription sentence fluctuate in response to contextual factors such that the very same (true) assertion will in one context generate true conversational implicatures (and so be appropriate) while generating false conversational implicatures (and so be inappropriate) in another. For example, it is commonly agreed that in making an assertion— especially one which ascribes knowledge— one thereby represents oneself as possessing adequate relevant evidence to back-up that assertion. Crucially, however, whether or not one’s evidence does count as ‘relevant’ or ‘adequate’ will be a context-sensitive matter, in that it will depend on the presuppositions of the context in which the assertion is made. Accordingly, two otherwise identical agents who are in exactly the same evidential state could make the same assertion and yet the one assertion be appropriate and the other inappropriate because of the differing truth-values of the conversational implicatures regarding the agent’s evidential position that are generated by these assertions in the different contexts. For instance, it may be— and I will say more about this below— that in sceptical contexts there are constraints in operation which restrict what might count as relevant or adequate evidence that are not in operation in normal contexts. Notice, however, that the claim here is not that one’s evidential state can depend on purely contextual
8
factors— which would result in a kind of contextualist thesis— but rather that the sort of conversational implicatures concerning one’s evidential state that are generated by an assertion can fluctuate with context. Other ways of spelling-out the details of a view of this sort are available, of course, and a position of this kind can be supplemented in all sorts of ways, though this is an issue that is best explored elsewhere.9 What is important for our purposes is that adding this further thesis to the basic Moorean line will inevitably lead to a fundamental change in the view, over and above the obvious immediate change to the basic Moorean stance by allying it to a theoretical claim about assertions of ascription sentences. This is that any account of the assertion of ascription sentences of this sort will have the result that it won’t be appropriate for the Moorean to simply assert her anti-scepticism in response to the sceptical argument. Instead, such assertions on this view will be deemed inappropriate, even if they are in fact true. Although this amendment to the Moorean line is a concession of sorts to both the sceptic and the contextualist— in that it allows that there is something improper about making ‘Moorean’ assertions— the concession is squarely in the spirit of the Moorean approach to scepticism. After all, it is intuitive that there is something amiss about the Moorean assertions, and thus any commonsense treatment of scepticism ought to be able to do justice to this thought. Moreover, by making this concession the Moorean is able to strip both scepticism and contextualism of a great deal of their plausibility, since part of the attraction of these views is precisely the impropriety of Moorean-style anti-sceptical assertions, whereas the neo-Moorean line under consideration here accounts for this fact without thereby conceding that what is asserted is false. With this addition to the Moorean thesis in play, the neo-Moorean can start to deal with the problem of dialectical impropriety facing the Moorean view. For notice that if the neo-Moorean does not claim to be able to properly make her anti-sceptical assertions then this will go some way to diminishing the counterintuitiveness of the position in this regard. It is no wonder that Mooreanism seems dialectically inappropriate if the assertions made by the Moorean are entered in a context in which such assertions are conversationally inappropriate, even if 9. For a useful exchange on this topic, see Brown (2005) and Black (2005). For the main defence of the contextualist line in this regard, see DeRose (2002).
9
true. Disentangling the Moorean line from the distinctive assertions made by the Moorean can thus go some way to evading the charge of dialectical impropriety. In order to make further headway in this direction, however, it is going to be essential to say something about how it is that we are able to know the denials of sceptical hypotheses as Mooreans claim, since without this further story we will still be left with the other aspect of the dialectical impropriety charge, which is that Mooreanism simply leaves us with a second-order scepticism which is almost as problematic as its first-order cousin. Only with this further theoretical story in hand will it be possible for the Moorean to motivate the denial of this premise over the sceptic’s endorsement of it. Usefully for our purposes, however, the resources needed by the Moorean to make this move can be found in the contextualist’s own treatment of this premise. As we noted above, contextualists also deny this premise, albeit only in undemanding everyday contexts where the epistemic standards are low. They tend to do this by appealing to some form of ‘safety’ principle— see, for example, DeRose (1995)— which demands that one’s belief in the target proposition ‘matches’ the truth in near-by possible worlds such that, where the proposition is believed it is true and where it is not true it is not believed. Only near-by possible worlds are relevant in everyday contexts because the epistemic standards are so low. Relative to near-by possible worlds, however, it is possible for one’s belief that, say, one is not a BIV, to match the truth in this way. After all, provided that the actual world is much as we take it to be, then this proposition will be true in all near-by possible worlds and, presumably, one will believe it in all near-by possible worlds. Such a safety-based approach thus rescues our knowledge of the denials of sceptical hypotheses in everyday contexts where the epistemic standards are low. In general, safety principles of this sort have been advanced by a number of figures⎯most notably Ernest Sosa (1999), but also myself (see Pritchard 2002d)⎯with the rationale behind their adoption being that they capture a sense in which one’s knowledge is non-lucky in that it involves the possession of a true belief that could not have easily been false. Interestingly, for contextualists it must be the case that at least sometimes it is sufficient for knowledge that one’s belief is merely safe, since otherwise the contextualist would not be able to account for our knowledge of the denials of sceptical hypotheses (we don’t obviously meet any other epistemic condition, such as an internalist justifica-
10
tion condition, when it comes to these propositions). If this is right, however, then it ought to be possible for the neo-Moorean to make use of this principle to account for how we might have knowledge of these propositions in a context-insensitive way. The guiding thought behind the adoption of this account of what knowledge fundamentally consists in will be that knowledge is always an undemanding concept such that we never need to meet the austere epistemic standards that are demanded by the sceptic in all contexts and demanded by the contextualist in sceptical contexts.10 Notice that the adoption of something like the safety principle in order to account for our knowledge of the denials of sceptical hypotheses commits both contextualism and neo-Mooreanism to an externalist epistemology. After all, what is central to the externalist position is that it allows that agents can have knowledge, at least in some cases, without having to meet any kind of internalist epistemic condition, such as an internalist justification condition, where what makes the condition internalist is that the relevant justifying facts are reflectively accessible to the agent concerned. Although there is of course a great deal of room for manœuvre concerning exactly how one understands this internalist requirement, it ought to be clear that our knowledge of the denials of sceptical hypotheses on the safety-based view won’t meet this requirement on any plausible rendering since, ex hypothesi, the agent concerned will have no good reflectively accessible grounds for thinking that she is not a BIV, and so can’t plausibly be thought to have met any internalist epistemic condition. Since both contextualists and neo-Mooreans are committed to externalism here, however, any problems facing the safety-based externalist view will be problems for both camps, and so will not decide the issue of whether we should be neo-Mooreans or contextualists.11 With this sup10. Black (2002) argues for a neo-Moorean view which makes use of the ‘sensitivity’ principle, a principle which is held by some — most notably Dretske (1970) and Nozick (1981) — to entail the rejection of closure. As Black shows, however, properly understood the principle does not have this result. In Pritchard (2005b, chapter 6) I argue that this should not come as a surprise because once one formulates the safety and sensitivity principles correctly then they come out as making roughly the same epistemic demands. For the sake of simplicity, however, we will focus on the formulation of the neo-Moorean view that employs safety here. 11. DeRose (1995) and Lewis (1996) are explicit in their endorsement of externalism. In contrast, Cohen tries to consistently maintain an internalist construal of the contextualist
11
porting epistemology in place — moreover, a supporting epistemology which, in outline at least, could not be objectionable by contextualist lights since it borrows its essentials from the contextualist account — we have the beginnings of a story of how one might legitimate the Moorean rejection of the sceptic’s first premise over the sceptic’s endorsement of that premise, thereby evading the problem of second-order scepticism left us by the basic Moorean line. After all, there is now an account of knowledge on the table — one that draws on the widespread intuition that knowledge is non-lucky true belief — which can account for our knowledge of the denials of sceptical hypotheses.12 Furthermore, notice that this supporting epistemology complements the other neo-Moorean thesis just offered concerning the context-sensitivity of the assertibility conditions of ascription sentences, since there is now a further prima facie explanation of why Moorean assertions can seem false even though, as this account shows, they are in fact true (provided the relevant external conditions obtain at any rate). After all, we would not normally regard a claim to know as being appropriate if the agent concerned was unable to back-up that assertion with appropriate grounds (grounds which must thus be reflectively accessible). If externalism about knowledge is right, however, then the fact that one is unable to offer adequate reflectively accessible grounds in support of one’s belief that one is not the victim of a sceptical hypothesis thesis. In doing so, however, he gets into a number of fairly fundamental problems, as he himself recognises. See especially, Cohen (2000). 12. There are important issues to be dealt with here, of course. For example, this strategy only works if we do tend to retain our anti-sceptical beliefs in near-by worlds, and one might question this. For my own part, I’m inclined to think that such beliefs are involuntary in that one can’t help but have a general anti-sceptical doxastic response to the world, and thus that where the relevant entailments are known one can’t help holding anti-sceptical beliefs of this sort. See Wittgenstein (1969) for a compelling case for the psychological necessity of an anti-sceptical response to the world. Of course, this approach raises the further question of whether such ‘Humean’ psychological attitudes should be counted as genuine beliefs, but I will set this to one side in what follows. A related issue in this regard is the basis on which our anti-sceptical beliefs are formed. Typically, we can point to a specific ground or process which gave rise to a belief, but this doesn’t seem to be the case when it comes to anti-sceptical beliefs. It is not as if we believe that we’re not BIVs because we see that we have two hands, for example. Indeed, Wittgenstein (1969) argues, persuasively to my mind, that some of our most commonsense everyday beliefs are not the result of any specific process or grounds either, such as the belief that (in normal circumstances) we have two hands. In order to keep matters simple I set this issue to one side here. For more on the ramifications of Wittgenstein (1969) for contemporary epistemology, see Pritchard (2003; 2005e).
12
will not decide the matter of whether or not one has knowledge of this proposition, even though it does decide the matter of whether or not one can properly claim to know this proposition. The contextualist move from the unassertability of a claim to know in this regard to its falsity is thus premature given the prior commitment to an externalist theory of knowledge. We have, then, the beginnings of a neo-Moorean account, one that can deal with the main difficulties facing the Moorean view. Moreover, since the neo-Moorean account draws on key features of the contextualist position in such a way as to also avoid some of the main problems facing contextualism, it follows that we have here an anti-sceptical view with distinct dialectical advantages over its contextualist counterpart. As we will see in the next section, however, this neo-Moorean story does need to be complicated in order to fully deal with the sceptical problem. Nevertheless, these complications do not undermine the dialectical advantage that neo-Mooreanism enjoys relative to contextualism. 4. Contextualism, evidence, and neo-Mooreanism The need to complicate the neo-Moorean story arises because the sceptical argument as it is outlined above does not really capture the underlying form of the problem, which is the evidential lack that the sceptical argument implicitly turns upon. The trouble is, any response to the non-evidential formulation of the sceptical argument is not thereby a response to the evidential formulation. Note, however, that since most anti-sceptical responses in the contemporary literature — including contextualism — are focused upon the non-evidential formulation of the sceptical challenge, this is a problem for all the participants in this debate, and not just a problem for the neo-Moorean. Consider the following formulation of the sceptical argument which focuses on the evidential basis of our knowledge: S1c S’s evidence cannot favour her belief that she is seated (and much else besides) over the (known to be incompatible) sceptical hypothesis that, say, she is a BIV. S2c If S’s evidence cannot favour her belief that she is seated (and much else besides) over the (known to be incompatible)
13
hypothesis that she is a BIV, then S is unable to know that she is currently seated (and much else besides). Hence: SCc S is unable to know that she is currently seated (and much else besides). Note that the premises of this argument are just as compelling as the premises of the earlier sceptical argument, if not more so. For one thing, the motivation for the second premise seems secure. In effect, the principle being appealed to here is an ‘underdetermination’ principle which states that if one lacks evidence which favours the target proposition over known to be incompatible alternatives, then one is unable to know the target proposition.13 Of course, so stated — without qualification — this principle may seem quite controversial, since there are a number of propositions which might plausibly be thought to be known even while lacking adequate evidential support. Notice, however, that this class of ‘ungrounded’ knowledge is bound to be small, since it is hardly plausible that most of our knowledge is ungrounded in this way. If this is right, then it does not matter for our purposes that underdetermination does not hold unrestrictedly since so long as it holds for most empirical propositions that will suffice to get the sceptical argument up-and-running (the sceptic only needs to call into question the epistemic status of a wide class of propositions which we believe in order to wreak the necessary intellectual havoc). So understood, however, this principle seems to be beyond reproach. After all, it would hardly be an intellectually satisfying response to the problem of scepticism to argue that one could have adequate evidential support for those beliefs which require such support even when the evidence in question fails to favour that belief over a known to be incompatible alternative. This would be a very flimsy notion of ‘evidential support’, and hardly one that has much in the way of anti-sceptical appeal. The trouble is, of course, that the first premise is highly intuitive as well, since it does seem to be true that we cannot have evidence which favours our everyday beliefs over (known to be incompatible) sceptical 13. This principle is plausible even without the demand that the alternatives be known to be incompatible with the target proposition, but since the alternatives we are interested are, typically at least, known to be incompatible with the relevant target propositions, we might as well add this clause so that the principle is understood in its strongest form.
14
alternatives, because, ex hypothesi, the sceptical alternatives are phenomenologically indistinguishable from everyday scenarios. Indeed, it is just an evidential intuition of this sort that guided the first premise of the non-evidential formulation of the sceptical argument outlined above. The problem is that neither the contextualist nor the neo-Moorean anti-sceptical strategy seems to contain within it any thesis that could be usefully brought to bear on this argument, since neither approach deals with the issue of the evidential basis of scepticism. The challenge facing both the neo-Moorean and the contextualist is thus to extend their anti-sceptical views such that they can offer some way of dealing with this argument, and this is going to mean opting for one of the following horns of a dilemma. Either they must (i) show that we can have evidential support for our beliefs in everyday propositions which favours this belief over the known to be incompatible sceptical alternatives; or else (ii) show that adequate evidential support for our beliefs in everyday propositions is consistent with an inability to rule-out the incompatible sceptical alternatives.14 Although most contextualists have ignored this formulation of the sceptical problem, there is one contextualist account in the literature that explicitly focuses on evidential scepticism, which is due to Ram Neta (2002; 2003).15 Essentially, Neta’s ingenious idea is that we need a contextualist account of knowledge because knowledge entails evidential support and the notion of evidence must be understood in a contextualist manner. The contextualist treatment of knowledge thus falls out of the contextualist treatment of evidence, where the latter deals directly with the evidential formulation of the sceptical problem. In particular, the claim is that the extent of one’s evidence varies from context to context, such that one might have very strong evidence in one context and yet very weak evidence in another, even though all that has changed has been the context. In this way, Neta argues that we can account for why the evidence we cite in favour of our perceptual beliefs in normal contexts is sufficient to favour those beliefs over sceptical alternatives, even though this is not the case in sceptical contexts. 14. For a sustained discussion of the relationship between the closure-based sceptical argument and the ‘evidential’ sceptical argument just formulated, see Pritchard (2005d). 15. Cohen (e.g., 2000) is also sensitive to the underlying evidential problem facing contextualist treatments of scepticism, though even by his own lights the approach he adopts in this regard — which essentially opts for the second horn of the dilemma just proposed — is not that compelling.
15
Neta’s approach to the problem of evidential scepticism thus opts for the first horn of the dilemma proposed above, in that he tries to account for how, at least in everyday contexts, our beliefs in everyday propositions could be regarded as supported by evidence which favours these beliefs over sceptical alternatives. Neta motivates this claim by noting that in such contexts we have a practice of allowing factive evidence — i.e., evidence which entails what it is evidence for. This is certainly true. If asked, for example, how I know that my colleague is at work today (perhaps by my head of department, who is speaking to me on the telephone), I might naturally say that I know this because I can see that he is here — where seeing that something is the case entails that it is the case. In general, Neta’s idea is that in normal contexts we do not restrict what counts as evidence to the purely phenomenal — that is, we do not restrict it to what merely seems to be the case, which is clearly not factive — and this allows factive evidence to be available. If this is right, then in everyday contexts one can possess evidence which favours one’s everyday beliefs over sceptical alternatives, since factive evidence in favour of one’s belief that, say, one was presently seated (that one could see that one was presently seated, for example), would be evidence which favoured this everyday belief over a sceptical alternative, such as the BIV hypothesis. The contextualist element to Neta’s proposal comes in via his concession that in sceptical contexts what counts as evidence becomes restricted in such a way that one’s perceptual evidence does not extend beyond the purely phenomenal. Accordingly, in sceptical contexts the evidential sceptical argument will go through, since one’s evidence will not favour one’s everyday beliefs over sceptical alternatives, and thus one will lack knowledge of both everyday and anti-sceptical propositions. We thus get a proposal which mirrors the general contextualist line on knowledge, most notably in that it poses no problem for closure, since we either know both everyday and anti-sceptical propositions or we know neither of them. The most contentious aspect of Neta’s proposal is, of course, his claim that in everyday contexts we can have evidence which favours our everyday beliefs over sceptical alternatives. As Neta notes, however, this feature of the proposal does have some independent support from other quarters. A number of philosophers — among them John McDowell (e.g., 1982, passim) and Timothy Williamson (e.g., 2000) — have noted that the standard arguments that are usually cited in favour of
16
treating evidence as purely phenomenal are actually quite suspect. In essence, the source of their complaint with the phenomenal conception of evidence is that it employs a ‘highest common factor’ style reasoning which equates the agent’s evidence in the ‘good’ case where she is not being deceived with what her evidence would be in a parallel (i.e., phenomenologically indistinguishable) ‘bad’ case in which widespread deception is taking place. It is not clear, however, why those generally impressed by epistemic externalism — contextualists and neo-Mooreans alike — should be persuaded by a line of argument of this sort. Why should what is phenomenally available to the subject in reflection determine what counts as one’s evidence? More precisely, why shouldn’t the externalist allow that the nature of one’s evidence is determined by facts that extend beyond the realm of what is reflectively accessible to the subject? Indeed, if the externalist offered a ‘reflective access’ requirement on the facts that determine one’s evidence, then it would be puzzling why he didn’t also incorporate such a requirement into his account of other epistemic notions like knowledge and justification. The phenomenal conception of evidence, while plausible by internalist lights, is thus highly contentious by externalist ones.16 We thus have good grounds to agree with Neta that the necessary evidential support is in principle available for our everyday beliefs. The problem for Neta’s account, however, is that if we can indeed make sense of the idea of there being the necessary ‘favouring’ evidence for our everyday beliefs, then it is far from clear why we need to also endorse any form of contextualism. That is, since this thesis alone will suffice to meet the evidential problem, what would be the advantage of endorsing a contextualist thesis on top of this claim? Indeed, it seems that the neo-Moorean can exploit the resources brought to bear on this problem by Neta without joining him in endorsing contextualism, since the principled denial of the first premise of the evidential sceptical argument is just what the neo-Moorean is looking for in order to respond to this sceptical problem in a broadly Moorean fashion. Note that this move on the part of the neo-Moorean mirrors that made earlier on behalf of the Moorean as regards the standard contextualist view, in 16. Moreover, there are compelling arguments which purport to show that the content of one’s perceptual experience could be different even in two phenomenologically indistinguishable cases, and since one’s perceptual experience surely forms part of one’s perceptual evidence, this too would lend support to a general resistance to the phenomenal account of evidence. See, for example, McDowell (1994, passim).
17
that the claim is that since neo-Mooreanism is here only making one potentially counterintuitive claim — as opposed to the contextualist’s two potentially counterintuitive claims — so neo-Mooreanism is at a dialectical advantage relative to contextualism. Moreover, an ‘evidential’ neo-Mooreanism of this sort would also avoid another key problem which faces Neta’s evidential contextualism. For what is troubling about such a view, like its non-evidentialist contextualist counterpart, is that by allowing the high-standards conception of evidence evidential contextualism seems to concede the intuitive force of the sceptical argument. That is, epistemically high standards contexts are, by their nature, epistemically privileged contexts, and thus to allow that in such contexts the only evidence we can appeal to is phenomenal evidence generates the tempting — and scepticismfriendly — thought that the only evidence we really have is phenomenal evidence, even though we might — loosely, as it turns out — appeal to non-phenomenal factive evidence in everyday contexts. What is needed by proponents of a view of this sort is thus some account of why our everyday evidential practices are legitimate even despite the fact that raising the standards employed in these contexts would destroy the knowledge gained via these evidential practices. The problem for such a proposal, however, is that any account of this sort would tend to undermine the position by making it plausible to suppose that we can reject the restrictions imposed by the phenomenal account of evidence in all conversational contexts. That is, the choice seems to be between allowing contextualism about evidence, but then conceding the epistemically privileged status of the sceptical context; or else disputing the epistemic hegemony of sceptical standards, but at the expense of undermining the evidential contextualism. Clearly the alternative neo-Moorean view in this respect does not face this problem, and so is again at a dialectical advantage relative to its contextualist counterpart.17 17. There is also a further problem for this view. The interesting case for the contextualist about knowledge is that in which we have two counterpart agents where the only difference between the situations that the two agents find themselves in is a conversational difference, but where the first agent speaks truly by saying ‘I know that P’, while the second agent speaks falsely. A natural way of understanding such a pair of cases is by saying that the evidence which the two counterparts have is exactly the same, it’s just that the strength of evidence one needs in order to know is different in the two contexts because of the different conversational standards in play. In this way, one isolates epistemic differences from mere conversational differences. This is important because that two otherwise counterpart agents
18
The response from the evidential contextualist to this line of argument will undoubtedly be to claim that neo-Mooreanism is unable to offer anything in the way of diagnosis when it comes to the problem of scepticism — it cannot, for example, explain why we find scepticism so intuitive. Notice, however, that the account of the context-sensitivity of the propriety conditions for knowledge ascriptions outlined above which we allied to neo-Mooreanism can be brought forward to deal with this issue. As stated above, this line of argument explained the apparent falsehood of certain assertions — like claims to know in sceptical contexts — in terms of how such assertions in these contexts generate false conversational implicatures about the agent’s evidential position. If this is right, then the appeal of scepticism — in both its standard and evidential guises — can be straightforwardly explained, since it arises out of a failure to recognise that the unassertability of these assertions is not a good indication of their falsity, since they would still be unassertable even if true. Indeed, notice that the contextualist account offered by Neta can be exploited by the neo-Moorean in this regard. Neta’s idea is that there is something about the raising of sceptical error-possibilities that restricts what counts as evidence, such that in sceptical contexts where these error-possibilities are at issue one’s evidence does not extend beyond the phenomenal. In essence, the idea is that one’s evidence is restricted to what evidence one would have if the error-possibilities under discussion were to obtain. If a sceptical error-possibility — like the BIV hypothesis — were to obtain, then clearly one’s evidence would not extend beyond the phenomenal because it would bear next to no could have identical beliefs with a different epistemic status where there is an epistemic difference between their two situations is hardly contentious, and lends no support at all for contextualism. Clearly, however, an evidential contextualist like Neta cannot help himself to this move, since by his lights the evidence possessed by the two counterpart agents is not the same. If this is so, however, then it appears to follow that more than just conversational differences are in play as regards the situations that the two counterparts find themselves in. Instead, the two agents are in different epistemic situations, and now it becomes unclear what the motivation for evidential contextualism is supposed to be. That is, what is it that holds fast between these two supposedly epistemically equivalent contexts? And if there is an epistemic difference between them, then why explain the different truth-values of assertions of the relevant knowledge ascribing sentences in each case in contextualist terms, rather than simply in epistemic terms? This line of argument is suggested by a remark made by Williamson (2000, 181) about a possible evidential contextualism, though he does not have Neta’s view specifically in mind.
19
correlation with events in the world. Thus, one’s evidence is for this reason restricted to the phenomenal in sceptical contexts. This sort of restriction on what counts as evidence is more naturally thought of, however, as a restriction on what can be legitimately cited as evidence in that context, where the latter has no immediate contextualist ramifications. That is, it is more natural to suppose that what we can legitimately cite as evidence is context-sensitive rather than the stronger claim that evidence itself is context-sensitive, especially if one has already signed up to an externalist epistemology. After all, if one were in a dialectical exchange with a sceptic where sceptical error-possibilities are at issue it does not seem all that counterintuitive to concede that it would not be legitimate to cite factive — and thus non-phenomenal — reasons in favour of one’s beliefs, since this would clearly be to fail to co-operate with the sceptic in this context. Moreover, as we noted above, when we talk of the evidence that one needs to legitimate an assertion we are clearly talking of reflectively accessible evidence — evidence that one can cite. In this context, however, a claim to know will generate the conversational implicature that one has evidence to back-up this assertion which excludes sceptical possibilities, and one does not have reflectively accessible evidence of this sort. Accordingly, this assertion — which is, note, a Moorean (but not a neo-Moorean) assertion — will be illegitimate since it will generate false conversational implicatures. Nevertheless, this does not mean that what is asserted is thereby false — which is what Neta is obliged to say — since for the externalist one’s evidence is not determined by what is reflectively accessible to one; still less is it determined by incidental features about the context in which one claims to posses that knowledge. Pulling the various threads of the neo-Moorean position together thus enables us to offer a diagnosis of why the Moorean assertions are so problematic — both dialectically and conversationally — since they purport to engage with the sceptic in the sceptical context, even though any engagement of this sort is bound to result in false conversational implicatures being generated. Note, however, that this diagnosis of the attraction of scepticism does not concede anything of substance to the sceptic, since it still remains the case that the sceptic has given us no reason for thinking that what the Moorean says is false.18
18. I discuss Neta’s view at more length in Pritchard (2004).
20
5. Concluding remarks Of course, there is much more that could be usefully said to re-enforce the neo-Moorean stance. It would be helpful, for example, if one could motivate neo-Mooreanism not just in relation to contextualism, but also in relation to other anti-sceptical theories, like the arguments for non-closure offered by Dretske (1970) and Nozick (1981), or the new contrastivist proposal put forward by Jonathan Schaffer (2005).19 Moreover, one would like to see a more positive and nuanced account of how the neo-Moorean can employ a non-phenomenal account of evidence.20 It would, however, take me too far afield to make further headway in this direction here. Instead, I will simply flag one issue which I think the neo-Moorean would be wise to explore in more depth. This issue concerns whether allowing that we are unable to properly argue head-on with the sceptic by making anti-sceptical assertions doesn’t concede something substantial to the sceptic after all. By this I don’t mean to withdraw or qualify any of the claims made above, but rather to draw attention to the fact that perhaps the problem of scepticism is not exhausted by the two forms of sceptical argument that we have discussed here. Accordingly, while it might be true that the neo-Moorean can offer a fairly straightforward response to these two forms of scepticism, there may be other versions of the sceptical challenge that are not so simply dealt with. The additional sceptical problem that I have in mind here is the Pyrrhonian sceptical challenge of antiquity, since what is distinctive of this form of scepticism is its focus on legitimate assertion rather than on knowledge possession per se, such that the mark of the knowing agent was specifically an ability to assert what was known without fear of having that assertion conversationally undermined. If this gets the essentials of the Pyrrhonian sceptical challenge right, and if such a challenge is to be regarded as posing a serious problem, then the concession made by the Moorean that we are unable to properly make certain assertions may be more troubling than is at first apparent.21 19. I explicitly discuss how the neo-Moorean should respond to arguments for non-closure in a number of places. See, especially, Pritchard (2005b, chapters 2 & 3). I respond to contrastivism in Pritchard (forthcoming a). 20. I expand on this issue in Pritchard (forthcoming b). 21. For further discussion of this point, see Pritchard (2005b, chapter 8). I discuss a related point in Pritchard (2005c).
21
It should also be noted, however, that even if neo-Mooreanism does turn out to be seriously troubled by the Pyrrhonian challenge, so understood, that offers no real comfort to the contextualist. This is because the contextualist is in an even worse position in this respect, since not only can the knowledge not be properly claimed, it isn’t even possessed in sceptical contexts! Yet again, then, we have seen that neo-Mooreanism is at a dialectical advantage relative to contextualism.22
REFERENCES Baldwin, T. (1992). ‘G. E. Moore’, A Companion to Epistemology, (eds.) J. Dancy & E. Sosa, Blackwell, Oxford. Black, T. (2002). ‘A Moorean Response to Brain-In-A-Vat Scepticism’, Australasian Journal of Philosophy 80, 148–63. — (2005). ‘Classic Invariantism, Relevance, and Warranted Assertibility Manœuvres’, The Philosophical Quarterly 55. Brown, J. (2005). ‘Adapt or Die⎯The Death of Invariantism?’, The Philosophical Quarterly 55. Cohen, S. (2000). ‘Contextualism and Skepticism’, Philosophical Issues 10, 94–107. DeRose, K. (1995). ‘Solving the Skeptical Problem’, Philosophical Review 104, 1–52. — (2002). ‘Knowledge, Assertion, and Context’, Philosophical Review 111, 167–203. Dretske, F. (1970). ‘Epistemic Operators’, The Journal of Philosophy 67, 1007–23. Lewis, D. (1996). ‘Elusive Knowledge’, Australasian Journal of Philosophy 74, 549–67. McDowell, J. (1982). ‘Criteria, Defeasibility, and Knowledge’, Proceedings of the British Academy 68, 455–79. — (1994). Mind and World, Harvard University Press, Cambridge, Massachusetts. 22. An earlier version of this paper was presented at a conference on ‘Contextualism’ at the Vrije Universiteit Amsterdam, the Netherlands, in October 2004. I am grateful to the audience on this occasion for their perceptive comments, with special thanks to my commentator, Finn Spicer, and to the organiser of this conference, Martijn Blaauw. Versions of this paper have also been presented in Bled (Slovenia), the University of York, and the University of Birmingham.
22
Moore, G. E. (1925). ‘A Defence of Common Sense’, Contemporary British Philosophy (2nd series), (ed.) J. H. Muirhead, Allen and Unwin, London; reprinted in his Philosophical Papers, Allen & Unwin, London (1959). — (1939). ‘Proof of an External World’, Proceedings of the British Academy 25, 273–300; reprinted in his Philosophical Papers, Allen & Unwin, London (1959). Neta, R. (2002). ‘S Knows that P’, Noûs 36, 663–81. — (2003). ‘Contextualism and the Problem of the External World’, Philosophy and Phenomenological Research 66, 1–31. Nozick, R. (1981). Philosophical Explanations, Oxford University Press, Oxford. Pritchard, D. H. (2001). ‘Contextualism, Scepticism, and the Problem of Epistemic Descent’, Dialectica 55, 327–49. — (2002a). ‘McKinsey Paradoxes, Radical Scepticism, and the Transmission of Knowledge across Known Entailments’, Synthese 130, 279–302. — (2002b). ‘Radical Scepticism, Epistemological Externalism and Closure’, Theoria 69, 129–61. — (2002c). ‘Recent Work on Radical Skepticism’, American Philosophical Quarterly 39, 215–57. — (2002d). ‘Resurrecting the Moorean Response to the Sceptic’, International Journal of Philosophical Studies 10, 283–307. — (2003). ‘McDowell on Reasons, Externalism and Scepticism’, European Journal of Philosophy 11, 273–94. — (2004). ‘Neo-Mooreanism, Contextualism, and the Evidential Basis of Scepticism’, Contextualism, (ed.) D. Suster, Rutgers University Press, Rutgers, New York. — (2005a). ‘Contextualism, Scepticism and Warranted Assertibility Manœuvres’, Knowledge and Skepticism, (eds.) J. Keim-Campbell, M. O’Rourke & H. Silverstein, MIT Press, Cambridge, Massachusetts. — (2005b). Epistemic Luck, Oxford University Press, Oxford. — (2005c). ‘Scepticism, Epistemic Luck and Epistemic Angst’, Australasian Journal of Philosophy 83. — (2005d). ‘The Structure of Sceptical Arguments’, The Philosophical Quarterly 55, 37–52. — (2005e). ‘Wittgenstein’s On Certainty and Contemporary Anti-Scepticism’, Investigating On Certainty: Essays on Wittgenstein’s Last Work, (eds.) D. Moyal-Sharrock & W. H. Brenner, Palgrave Macmillan, London. — (forthcoming a). ‘Contrastivism, Evidence, and Scepticism’, Social Epistemology 20. — (forthcoming b). ‘How to be a Neo-Moorean’, Internalism and Externalism
23
in Semantics and Epistemology, (ed.) S. Goldberg, Oxford University Press, Oxford. Schaffer, J. (2005). ‘Contrastive Knowledge’, Oxford Studies in Epistemology, (eds.) J. Hawthorne & T. Gendler, Oxford University Press, Oxford. Sosa, E. (1999). ‘How to Defeat Opposition to Moore’, Philosophical Perspectives 13, 141–54. Unger, P. (1971). ‘A Defense of Skepticism’, Philosophical Review 80, 198–219. — (1975). Ignorance: A Case for Scepticism, Clarendon Press, Oxford. — (1984). Philosophical Relativity, Blackwell, Oxford. Williamson, T. (2000). ‘Scepticism and Evidence’, Philosophy and Phenomenological Research 60, 613–28. Wittgenstein, L. (1969). On Certainty, (eds.) G. E. M. Anscombe & G. H. von Wright, (tr.) D. Paul & G. E. M. Anscombe, Blackwell, Oxford. Wright, C. (1991). ‘Scepticism and Dreaming: Imploding the Demon’, Mind 397, 87–115.
24
Grazer Philosophische Studien 69 (2005), 25–49.
LIVING WITHOUT CLOSURE Krista LAWLOR Stanford University Summary Epistemic closure, the idea that knowledge is closed under known implication, plays a central role in current discussions of skepticism and the semantics of knowledge reports. Contextualists in particular rely heavily on the truth of epistemic closure in staking out their distinctive response to the so-called “skeptical paradox.” I argue that contextualists should re-think their commitment to closure. Closure principles strong enough to force the skeptical paradox on us are too strong, and closure principles weak enough to express unobjectionable epistemic principles are too weak to generate the skeptical paradox. I briefly consider how the contextualist might live without (strong) closure.
0. Introduction Epistemic closure is the idea that knowledge is closed under known implication. Here is one likely formulation: Necessarily, if S knows that p, and knows that p implies q, and believes that q on the basis of believing both that p and that p implies q, then S knows that q. Many philosophers believe that some such principle is true. Contextualists in particular, rely heavily on the truth of epistemic closure in staking out their distinctive response to skepticism. But should the contextualist accept epistemic closure? Should we? One can hear the strains of complaint already: “Not again! Haven’t we settled this long ago? Epistemic closure is obviously true — even if it does take work to specify a reasonable articulation of it, that work is largely about finding a formulation immune to small quibbles.”
I hear the complaint, and I understand. I do not mean here to bring to light some small quibble against closure. I think epistemic closure is importantly not true of us — at least when it takes forms strong enough to generate philosophical puzzles. Characterizing our epistemic lives and the challenges of skepticism absolutely demands seeing how closure, in these strong forms, fails to be true of us. On a more modest note, given the role that epistemic closure has played in shaping recent discussions of skepticism and the semantics of knowledge reports, it is at least well worth our while to consider closure carefully. And, given closure’s central role in defining contextualism, the contextualist in particular might do well to consider whether life without closure is possible. 1. Contextualism and the skeptical paradox Let us start by considering the role that epistemic closure plays in the contextualist’s response to skepticism. The contextualist says that knowledge reports have a context-sensitive semantics — a report’s truth or falsity depends on the context of utterance.1 Contextualists say this fact lets us make peace with the skeptic: in ordinary contexts, knowledge reports can be true, while in skeptical contexts, our knowledge reports turn out false.Contextualists thus claim we can give skepticism its due while preserving the truth of many commonsense knowledge attributions. The contextualist’s position is nicely illustrated by their distinctive resolution of the so-called “skeptical paradox” (Cohen 1988). The paradox arises when we reflect upon how natural it is to claim knowledge of many mundane propositions, and yet how compelling it is to admit skeptical doubts. Naturally, I claim to know that I have all ten fingers, while once you mention it, it seems I must admit that for all 1. To be precise: contextualists say that “knows” always contributes the same relation, Knows: <S, p, ea>, but this relation involves a parameter for the epistemic standard, ea, that one’s claim must meet, and the value of this parameter is settled by the context of utterance — specifically, in light of facts concerning the ascriber. For helpful review see MacFarlane (2005). Contextualism is to be distinguished, then, from Relevant Alternatives theory (Dretske 1981), which holds that the truth-value of a knowledge claim depends on what the speaker or hearer take to be relevant, but not on features of the context in which the knowledge claim is uttered.
26
I know I might be a brain-in-a-vat, a creature without hands or fingers. But how can this be? The paradox then, is this: Let S stand for a skeptical proposition, for example, that I am a brain-in-a-vat. Let M stand for a mundane proposition, for example, that I have all ten fingers. I find it natural to claim knowledge that I have all ten fingers: (1) I know M But it seems that if I know this, then I can also know that the relevant skeptical proposition, S, is false: (2) If I know M, then I know that not-S That is, if I know I have all ten fingers, then I know that I am not a brain-in-a-vat (such creatures are fingerless, after all). But it seems I have to admit that for all I know I am a brain-in-avat, so: (3) I don’t know that not-S Each claim seems true, although taken together these claims are inconsistent. What should we say about the paradox? We can imagine the skeptic eager to exploit the paradox, arguing that if (3) is true, then (1) is false. Many non-skeptical approaches to the paradox also try to identify a false statement. For instance, G.E. Moore famously argues that claim (1) is at least as plausible as (3) so, if we are facing a puzzle of which statement to hold on to, we might keep (1) and reject (3) — that is, we can know that the skeptical proposition is false. Relevant alternatives theorists argue that (2) is false — knowing M does not require knowing the falsity of all logical contraries to M, but only the relevant alternatives to M, and S is not a relevant alternative to M (Dretske 1981). Recent contextualists favor an irenic solution to the paradox. They argue that each statement of the paradox is true, only in different contexts of utterance. Taking account of the sensitivity of knowledge claims to contextually determined evidential standards, we more carefully state the claims that we are in fact committed to in this way:
27
(1)* (2)* (3)*
I know M, relative to ordinary evidential standards. If I know M, then I know not-S, relative to ordinary evidential standards. I don’t know not-S, relative to high evidential standards.
All of which claims are true. Explaining the appearance of paradox, contextualists argue that we confuse propositions. Specifically, we are confused into thinking that with (3) one really says (3)** I do not know not-S, relative to ordinary evidential standards. which is a false claim. Fortunately, (3)** is not what one is committed to. (The contextualist might go on to explain that our confusion stems in part from the fact that in the context set by claim (1), standards of evidence are set low, and so it seems that with (3) one says (3)**, but in fact a change of evidential standards has occurred with the mere mention of the skeptical proposition, so that one really says (3)*.2 Contextualism has come to be identified with this peacemaking resolution. And a tidy resolution it is: the paradox is only apparent, since the contextualist shows how in fact there is no inconsistency in the content of our knowledge attributions. Granted, the skeptic is correct that ordinary knowledge-reports are false in skeptical contexts, but we ordinary speakers are correct that ordinary knowledge-reports are true in ordinary contexts. Thus in a single stroke the contextualist promises to save commonsense commitments and also explain the tug of skeptical propositions — the propositions exert their pull because in certain contexts they are true. The skeptical paradox is widely acknowledged to set a standard for weighing alternative accounts of the semantics of knowledge attributions. And contextualism’s tidy solution is also widely held to be a significant advantage over other semantic accounts.3 The skeptical paradox turns centrally on epistemic closure. Actually, epistemic closure, in some suitably restricted form, plays a role twice 2. Lewis (1996) suggests mere mention of a skeptical hypothesis is enough to change the contexts, while other contextualists want more demanding requirements. 3. Note that for some, the contextualist’s peacemaking strategy seems concessive (Sosa 2000). For others, contextualism doesn’t get the semantics for knowledge attribution right (Hawthorne 2004 and MacFarlane 2005).
28
over in the skeptical paradox. First, (1) and (2) only imply not-(3) if some form of closure is true. Second, attempts to resolve the paradox by denying premise (2) run afoul of closure: denying (2), one claims that one can know M, the mundane proposition, while not knowing that not-S, the falsity of the skeptical proposition; but how is it possible for one to know one has all ten fingers, while not knowing that one is not a (fingerless) brain-in-a-vat? Especially if the latter is a known consequence of the former? Closure dictates against this possibility. In sum, it is crucial to contextualism that epistemic closure should be true. Without epistemic closure, there is no skeptical paradox. And without a skeptical paradox, there is no distinctive contextualist resolution. Moreover, even many opponents of contextualism assume that the skeptical paradox provides a test of adequacy for candidate theories of the semantics of knowledge attribution. Given the centrality of epistemic closure to recent discussions of knowledge attribution, then, we do well to consider — is epistemic closure true? 2. Why it seems we cannot live without closure More carefully, the question is, why think human knowers are accurately described by some form of epistemic closure principle? Some will find the question easily answered. These philosophers believe closure, in some suitably qualified form, is obviously true. Feldman (1995), considering closure for epistemic justification, writes: I believe that some version of the closure principle, restricted to known consequences, is surely true. Indeed the idea that no version of this principle is true strikes me, and many other philosophers, as one of the least plausible ideas to come down the philosophical pike in recent years. (1995, 487)
Whether or not closure seems obvious to many, it is still legitimate to ask, why think epistemic closure is true? Several distinct claims have been made in favor of closure: first, it has been suggested that closure principles express norms of epistemic life, which if not entirely descriptively accurate, are prescriptive guides we can be seen to live by (Vogel 1990). Second, it has seemed to some that there are significant costs if closure should fail to be true of us — costs so great, and so obviously not borne by us, that we can safely assume closure principles must be true of us. Finally, and no doubt more per-
29
vasively, there is the thought that closure principles merely state in perspicuous terms epistemic principles that can be expressed more informally and unobjectionably. The idea that closure principles are simply perspicuous expressions of otherwise completely acceptable epistemic principles is, I believe, by far the most compelling reason for thinking closure to be true. It will be my focus in what follows. Regarding the other two claims in favor of closure, this much can be said: it seems that epistemic closure, at least as it figures in recent accounts of the skeptical paradox, is a descriptive and not a prescriptive principle. The costs of rejecting epistemic closure, then, must be reckoned as necessarily borne if closure is in fact not true of us. (Just what these costs are is a good question. Assessing the costs of living without closure will be part of our undertaking.) 3. Closure and deduction Let us turn then to the idea that closure principles, in some suitably qualified form, merely state in perspicuous terms epistemic principles that can be expressed more informally and unobjectionably. One particularly compelling version of this idea is that epistemic closure principles merely encode claims about the epistemic value of deduction. Timothy Williamson (2000) identifies a principle he calls “intuitive closure”: … knowing p1, … pn, competently deducing q, and thereby coming to believe q is in general a way of coming to know q.
Williamson goes on to remark: We should in any case be very reluctant to reject intuitive closure, for it is intuitive. If we reject it, in what circumstances can we gain knowledge by deduction? (2000, 117–118)
In a similar spirit John Hawthorne (2004) suggests the following formulation in terms of an explicit closure principle: Single Premise Closure (SPC): Necessarily, if S knows p, competently deduces q from p, and thereby comes to believe q, while retaining knowledge of p throughout, then S knows q.
Hawthorne sums up his defense of SPC (and a related principle governing the multiple premise case) thus:
30
Our closure principles are perfectly general principles concerning how knowledge can be gained by deductive inference from prior knowledge. (2004, 34)
It is safe to say that almost everyone who thinks that closure surely applies to human knowers has something like this idea in mind. The exact expression of the idea varies — for instance, sometimes the defender of closure will say that closure just is the claim that deduction is a good epistemic method, or that knowledge can be extended by deduction, or that deduction transmits knowledge. (Defenders of justification closure focus on parallel claims to the effect that justification can be extended by deduction.) I think we have reason to doubt whether closure principles express such unobjectionable claims. To take a specific case for more careful discussion, let us consider whether “intuitive closure” really is the principle that knowledge can be gained by deductive inference from prior knowledge. (It is one case only, but we can easily imagine how results might generalize to alternative formulations.) The formulation I will focus on, restricted to the single premise case, is this: Intuitive Closure (IC): Necessarily, if S knows p, competently deduces q from p, and thereby comes to believe q, then S knows q. To begin, here is one question that clarifies the power of IC: Can we conclude anything about what a person, S, knows on the basis of our knowledge of what S already knows and our knowledge that S is governed by IC? The answer is, No. We must add that we know that S actually has performed a specific deduction and has a particular belief as a result. Then IC tells us that we can be assured of S’s knowledge of the conclusion. We are assured by IC of S’s knowledge of the conclusion, as opposed to S’s mere belief (we had to already know about S’s belief), and our assurance comes entirely from the fact that deduction is a good epistemic method. Having IC in our toolbox, then, seems not to give us much that we didn’t already have with a general claim to the effect that deduction is a good epistemic method. For the purposes of epistemic logic, or for concluding things about what S knows on the basis of S’s prior knowledge, IC seems not to add much that we did not already have with such
31
a claim about deduction. This fact can lead us to think that IC is nothing over and above such a claim.4 We must exert care here, however. IC does not just say that knowledge can be gained through deduction. It says that knowledge is gained, necessarily, when deduction is applied to known premises. That is a stronger claim. The claim that deduction can extend knowledge is not equivalent to the claim that it necessarily does extend knowledge. This is no small quibble. For creature like us, the difference makes a difference. To bring out the difference between IC and the claim that knowledge can be gained through deduction, it will be helpful to consider a case. Imagine Edward believes that some form of determinism is true, that events in the natural world all have causes. He believes that for effects to be produced, some causal mechanism must be at work, no matter how difficult it might be to know the mechanism. Edward also takes homeopathic medicines. He believes that a homeopathic cure, one that is similar to the nature of the illness, will relieve his suffering. The cold cure he takes has a concentration of 1 part per 100200. Edward knows this (it is an advertised virtue of the medicine that it reaches this level of dilution). Edward took chemistry just recently, so he also knows that there is a limit to the dilution that any substance can undergo, after which it is unlikely that even one molecule of the substance remains. And he knows that this limit is set by Avogadro’s number, roughly 1 in 1024. Now, if Edward were to stop and think about it, he could easily infer that it is not likely that there remains a single molecule of the active ingredient in the cold cure he just purchased. He could further infer that this fact precludes the active ingredient from exerting any causal force on his system — precludes it from being active at all. Let us suppose that one day Edward finds himself ruminating along these very lines. Consequently, he arrives through deduction at the belief that this cold medicine will very likely not work. Edward reasons competently, and retains knowledge of his premises throughout. If IC were true of Edward, then having arrived through deduction at the belief that this cold medicine will very likely not work, he would necessarily know as much.5 But does he? 4. In fact, Hales (1995) says IC is nothing more than such a claim about deduction being a good method. I think that’s not right, because IC needs some revision if it is to state this unobjectionable claim. I undertake the revision below. 5. Perhaps his reasoning looks to involve multiple premises. But let us not argue over
32
Here is one scenario on which it seems reasonable to think that Edward doesn’t know the conclusion. Edward was raised by believers in homeopathy, and he has the strong conviction that he has seen illnesses cured by homeopathic treatments. His mother used just this cold medicine to treat him as a child, and he seems to remember feeling better after he took it. Up until now, he has been quite certain in the belief that this cold medicine will very likely work. That is, after all, why he bought it. As we might say, Edward has many antecedent beliefs that support this belief. Now, when Edward ruminates about what he knows from chemistry class and deduces the belief that this cold medicine will very likely not work, his earlier conviction in the medicine’s efficacy is shaken. We can even imagine that he (instantly) loses the belief that the medicine will work. But still, his antecedent beliefs in the power of homeopathy leave him concerned about the conclusion of his inference — they leave him questioning whether this cold medicine will very likely not work. Could it be that it really is so worthless? That his mother and everyone he has listened to over the years, avowing homeopathy’s efficacy, is wrong? That all the cases of illnesses cured with this very medicine were only apparently cases of cures? Given the strength of his earlier convictions — which we can imagine were very strong — Edward just doesn’t feel certain in his new belief that the medicine will fail, despite what he knows about chemistry. In light of his lack of certainty, one might argue, Edward doesn’t know that this cold medicine will very likely not work. To avoid misunderstandings about our case, note that Edward will soon come to know that this cold medicine will very likely not work, after he goes through the process of questioning his antecedent beliefs in the efficacy of homeopathy and all his memories of the cold cure seeming to work, and so forth. As he comes to understand how all those beliefs could be false, his certainty in his conclusion increases to the point where he knows it. What we want to focus on, however, is not what Edward comes to know after a period of reflection, but what Edward knows immediately upon his having made his deduction. Were closure true, Edward knows the conclusion that this cold medicine will very likely not work necessarily upon having made his inference. But the case suggests otherwise — it suggests that Edward just doesn’t feel this — the final inference Edward makes, from this medicine likely has no active ingredients to this medicine will very likely not work is all that we need consider.
33
certain in his new belief that the medicine will fail, despite what he knows about chemistry. Consequently, Edward doesn’t know that this cold medicine will very likely not work. But how, says the defender of closure, can Edward not know this? He has deduced it from premises he knows to be true! Before giving a general answer, we should first settle on some basic epistemological terrain: Knowledge, it seems, requires something like justified true belief. Sources of one’s justification are varied: in many central cases one’s justification must be based on evidence that one grasps, while in other cases one’s justification is based on how one arrives at the belief. Often these varied sources will each have something to contribute to one’s total justification for a belief. Moreover, one’s justification for a belief may increase, for instance, when one gains more evidence. Whether one’s justification must reach a level of conclusiveness, so that what one knows with certainty, is a matter of dispute (Audi 1995, 218). However, we can imagine that there is some level of justification below which one’s belief does not count as knowledge. Now the general answer to the question, How can Edward not know the conclusion of his inference, when it is competently derived from known premises? The general answer is that being justified sufficiently to count as knowing is a complex matter. Edward’s justification for believing the conclusion of his inference, that this cold medicine will very likely not work, involves several factors. (In sketching Edward’s case, I have spoken of Edward’s feeling of certainty in his conclusion, but we might do better to speak solely of his justification for his beliefs.) One factor in his justification for his belief is the fact that he has deduced it from known premises. Another factor is the fact that he has antecedent beliefs that seem to him to provide evidence against the proposition in question. These antecedent beliefs undermine Edward’s justification for his conclusion. Until these antecedent beliefs themselves are thrown over, the conclusion he has drawn is one he is not sufficiently justified in for his belief to count as knowledge. Edward’s case is just one instance of a wider phenomenon. We do not simply settle for the conclusions of our inferences. We sometimes have to catch up with ourselves. And we can imagine that in Edward’s case, he will catch up with himself. Being a good epistemic agent, Edward feels uncertain about his new belief that the medicine will fail, despite what he knows about chemistry. To remove his uncertainty, he feels
34
compelled to consider his pool of conflicting antecedent beliefs, and to consider how each might in fact be false. When this de-credentialing of his conflicting beliefs is accomplished, he will be sufficiently justified in his belief for it to count as knowledge. Now we can imagine some responses to the case that the defender of closure might want to make. (i) One response is that if Edward does not know the conclusion, then the case must be one in which Edward’s beliefs in the efficacy of the medicine also undermine his justification in his premises to the point where they, too, no longer count as known. Consequently, the case is one in which IC is not supposed to apply, and so presents no counter-example.6 It is not clear, though, why Edward’s knowledge of his premises must also be undermined. Our case, I think, is one where the presence of antecedent beliefs precludes knowledge of a conclusion, but not because they undermine knowledge of the premises. Edward’s antecedent beliefs (about homeopathy) speak directly against the conclusion that this cold medicine will likely not work just in that they support its negation. As a result, he is under pressure to see if he can de-credential them. These troublesome antecedent convictions needn’t speak directly against his knowledge of his premises (about chemistry) to exert this pressure: the antecedent convictions that formerly supported Edward’s belief that this cold medicine will work do not have any immediate bearing on the value of Avogadro’s number or similar premises from chemistry. Edward can be less than sufficiently justified in his conclusion, but still his knowledge of the relevant chemistry is quite good. To speak metaphorically, it is the relative independence of his two belief pools that creates this possibility. The pool of beliefs regarding chemical facts and the pool of beliefs regarding homeopathy are largely independent of each other, and only meet in respect of Edward’s belief about the efficacy of the medicine. Absent further reason to suppose that Edward’s antecedent beliefs in the efficacy of the medicine must also undermine his justification in 6. Hawthorne (2004, note 83) for instance, makes this argument in defense of his own version of closure, SPC: “What of the case where one has other beliefs that imply not-q? … I doubt that the materials for a counterexample are available here. If those other beliefs destroy knowledge of the premises, then the reasoning will not be from known premises, in which case the principle will not be relevant. If they do not, then I do not see why the presence of those beliefs should preclude knowledge of the conclusion.”
35
his premises, the response looks question-begging — it looks to involve an appeal to the idea that closure simply must be true. (ii) The defender of closure might have something further to say here. The defender might say that if Edward’s antecedent beliefs conflict with his conclusion, then they must also conflict with his premises. That is a matter of logic. His belief pools cannot be completely independent, in this sense. So Edward does not know his premises, if he does not know his conclusion. Of course, the problem with this response is that it certainly involves an appeal to closure — the idea that logical relations determine what is known and not known, without further ado, is precisely what is under discussion when we consider whether closure is true of us. Let us consider some other, more compelling responses that a defender of closure might make. (iii) The defender of closure might argue that, although Edward knows the conclusion that this cold medicine will very likely not work, he does not know that he knows it. His uncertainty borne of his antecedent beliefs stands in the way of his second-order knowledge, but not his first-order knowledge, which is ensured by being a product of competent deduction. This response gains its initial appeal, I think, from imprecision in the notion of certainty. Is Edward’s uncertainty an uncertainty about his conclusion, or about whether he really knows his conclusion? It can feel hard to say. That is why I prefer to reformulate the discussion of the case in terms of justification. The problem with this response is then clearer: the problem is we have little reason to think that Edward’s antecedent beliefs in homeopathy actually undermine a second-order belief of Edward’s about whether he knows his conclusion. It is simply more plausible that Edward’s antecedent beliefs in the efficacy of homeopathy undermine his justification for his belief in the inefficacy of homeopathy. (iv) The defender of closure might try to defend externalism about knowledge: Edward knows the conclusion that this cold medicine will very likely not work, simply because his belief was produced by a reliable belief-forming mechanism, namely, deduction. The fact that Edward has conflicting antecedent beliefs does nothing to take away from this fact. One thing to note about the response is that defenders of closure do not typically stake their defense on externalism’s being true. In fact, no
36
contentious commitments about knowledge are assumed to be necessary to the defense of closure. Another thing to note is this: although the case is one on which it is fairly easy to imagine that Edward’s antecedent beliefs in homeopathy were not formed reliably, we might imagine that they were. (Reliably formed beliefs might still be false, after all.) Imagine Edward arrived at his beliefs by carefully considering cases of homeopathic cure, testing the medicine on himself with some success, and so on. The point is, it is possible to imagine that a reliably formed belief conflicts with still other reliably formed beliefs. In which case, until the conflict is resolved, it will not do to count the belief as known simply because one reliable method produced it. (v) Another response: Edward’s case is an odd case — the better description of it is that he knows the conclusion that this cold medicine will very likely not work, but he has a hard time “really believing it.” So whatever oddness there is in the case can attributed to Edward’s psychology, to the belief and not to its epistemic status as known. Let us try to flesh out this response a little further: Imagine that Edward has a hard time “really believing” the conclusion that this cold medicine will very likely not work. Among other things, we can imagine that this means the belief will not control Edward’s actions in the direct way it would were he to simply believe that this cold medicine will very likely not work. So for instance, he will not advocate its use, but neither will he proselytize his old friends against the cold medicine. He will not buy more of the medicine, but then, neither will he throw away in disgust what he has already purchased. He will say things under his breath like, “That stuff really can’t work, can it?” as though trying to convince himself. Given his behavior, does it seem that Edward knows that the cold medicine will very likely not work? It seems odd to describe him as knowing this, since if he knew it, then wouldn’t he simply throw the pack away, and stop muttering such things in a fretful tone of voice? It seems Edward does not know that this cold medicine will very likely not work, not yet anyway. Now of course one might try to re-introduce the distinction, saying Edward really knows the proposition in question, even if he still isn’t acting in accord with that knowledge. But here it seems the defense of closure is stretched to an uncomfortable limit. Knowledge makes action rational, and in light of actions that tempt us to describe a person as not really believing, we also find it natural to say the person doesn’t really know.
37
Barring further responses, we can conclude that Edward’s case illustrates the following: deduction may be used well, and provide its customary justification for a conclusion q, but if one already has beliefs that imply not-q, one’s justification in believing q may not yet rise to the level necessary for one’s belief to count as knowledge. Now the defender of closure might want to reply: You said you weren’t going to quibble. If Edward had no contradictory beliefs, the presence of which might undermine his confidence in the conclusion of his inferences, then his deduction will bring him knowledge necessarily upon his making it. So we simply need to add another condition to our antecedent, arriving at the following refinement of IC: Necessarily, if S knows p, competently deduces q from p, and thereby comes to believe q, and S has no contradictory beliefs that support not-q, the presence of which might undermine S’s justification in the conclusion of S’s inferences, then S knows q. The refinement means that Edward’s case is not one to which we might expect closure to apply. Consequently, it provides no counterexample. I think this is a reasonable move on the part of a person defending closure. But of course it is not just the presence of antecedent conflicting beliefs that raise a problem. We might imagine a slightly different case, in which Edward’s knowledge of their former presence and strength, even though he does not now maintain them, is enough to shake his confidence in matters concerning the efficacy of cold medicines; so his justification for believing q does not rise to the level necessary for his belief to count as knowledge. Again, complaining about quibbles, the defender of closure might build a condition to rule out such a possibility. But then, again, another possibility presents itself — perhaps Edward doubts his conclusion, because he could be so misled about cold cures before, and he is disturbed that it never pressed itself upon him until today to wonder about homeopathy. Maybe he’s not as rational as he thought after all? Once he’s doubting his rational powers, he cannot be justified in his conclusion.7 And so on, and so on. 7. One could try to respond by saying that closure requires competent deduction, and deduction that is competent will involve not doubting one’s rational powers. Perhaps compe-
38
Edward’s is just one case — we can imagine still others that illustrate distinctive routes by which one might fail to be justified in the conclusion of one’s inferences. What is the defender of closure to do? I think it reasonable for the defender of closure to halt the process of ever-refining IC, by reminding us of what we are about. Recall, we are trying to articulate a principle to the effect that knowledge can be gained through deduction from known premises. What we want, then, is a principle that tells us about the justificatory power of deduction as such, whatever else might be the case about further avenues by which justification may be enhanced or undermined. To arrive at such a principle, we can refine IC by adding a general-purpose condition that sets aside all other considerations about one’s justification in a single go: (IC*) Necessarily, if S knows p, competently deduces q from p, and thereby comes to believe q, and nothing stands in the way of S’s belief in q having the level of justification sufficient for knowing q, then S knows q. With IC* we have a principle that does all the defender of closure wants. IC* is certainly a true claim: if q is true, and if S believes q on the basis of having deduced q, and the justification that deduction confers is justification enough, then S knows q. Moreover, IC* is intuitive and quibble-proof — it seems impossible to imagine counter-examples. More importantly, IC* has the form of a closure principle: it states how knowledge is closed under deduction, given suitable restrictions. Of course, the restrictions are fairly strong, but that is to be expected. That is, the unobjectionable claim that knowledge can be gained through deduction may be framed in terms of a closure principle, because some fairly encompassing restrictions are in place. Let us take stock. We have seen that principles like “intuitive closure” (IC) stand in need of revision, if they are to express what are really unobjectionable epistemic claims. But we have also seen that we can make the necessary revisions, and express such unobjectionable claims as that knowledge can be gained though deduction in terms of a closure principle (IC*), provided that the principle is restricted appropriately. tent deduction involves more than simply arriving correctly at a conclusion, but also knowing what one is about in doing so. However, it also seems one might know what one is about in the relevant sense, and still wonder whether one’s deductive powers are to be trusted.
39
Now, one might wonder, where does this leave us? Can’t the defender of closure rest her case here? After all, we could not make sense of much of our epistemic lives if we held that deduction doesn’t extend knowledge: Life without IC* is clearly unthinkable. I agree, of course. But our job is not done. For we have also seen that life without IC is clearly thinkable. So the question is, which principle has played a role in shaping recent discussions of skepticism and the semantics of knowledge reports? I will argue below that it is IC (and like principles) that has played a key role, and not the unobjectionable IC*. 4. Generating paradox: strong and weak closure Let us start by clarifying the relative strength of our two closure principles. Note that IC* is a weak principle. That is to be expected, if IC* simply gives formal expression to the unobjectionable claim that knowledge can be gained by deduction. What IC* focuses our attention on is this: Deduction does extend knowledge, necessarily, provided that whatever justification one has solely in virtue of having performed the deduction is all the justification one needs. So in cases where one needs more justification for one’s belief than can be provided simply by having performed a deduction, principle IC* will not apply. (Understanding the strength of IC* will be crucial in what follows.) Now, keeping in view this fact about IC*, we can ask about its role in philosophical puzzles of skepticism. In particular, the question is, does IC* express a principle that is strong enough to generate the skeptical paradox? Let us recall how the skeptical paradox gets generated. Recall that instances of the skeptical paradox all involve knowing M, some mundane proposition and deducing not-S, the negation of a skeptical proposition. For example: I know M, that I have all ten fingers, from which proposition I deduce not-S, that I am not a fingerless brain-in-a-vat. Paradox arises because it seems I cannot know this. Couldn’t it really be that I am after all a brain-in-a-vat? It seems I must acknowledge that it is possible that I might be a brain-in-a-vat. And so I cannot know that I’m not. In other words, the paradox is generated by the fact that one has antecedent reason for thinking that S might be true, so one cannot know its negation.
40
Let us focus on this fact about how the paradox is generated: the paradox is generated only because something stands in the way of my knowing the conclusion of my inference. (In our example, something stands in the way of my simply knowing, on having made the deduction, that I am not a fingerless brain-in-a-vat, namely, the acknowledgement that for all I know I might be.) Now recall, in such a case, where something stands in the way of my knowing the conclusion of my inference, IC* does not apply. As we noted above, IC* is silent about cases where something stands in the way of one’s belief in the conclusion of one’s inference having the level of justification sufficient for knowledge. So IC* cannot generate the skeptical paradox. IC* is too weak a principle to do the job. IC* is too weak to generate the skeptical paradox. This fact is interesting in itself — after all, many discussions of the paradox proceed as though any sort of closure principle would do. The real interest of this fact, however, is this. Although we haven’t considered all possible formulations of closure, the argument is certainly suggestive of the following moral: closure principles strong enough to force the skeptical paradox on us are stronger than unobjectionable epistemic principles, such as the principle that deduction can extend knowledge. And closure principles weak enough to simply state in formal terms what are otherwise unobjectionable epistemic principles are not strong enough to generate the skeptical paradox. What I want to undertake, in the following sections, is an assessment of where this moral leaves us if it is true. (Our consideration of IC* and IC is good grounds for thinking this general moral is true, although we haven’t considered all possible formulations of closure.) Specifically, let’s suppose that life without strong closure is possible. Where would that leave us? One immediate implication is that we no longer face the skeptical paradox. This in itself has some importance for recent discussions of skepticism and the semantics of knowledge reports. In sorting out what life without strong closure might be like, I’ll consider three main questions in turn: (1) whether we bear extraordinary costs on having strong closure principles fail to be true of us; (2) whether we might defeat skepticism, if only weak closure is true of human knowers; and finally (3) where does all this leave the contextualist?
41
5. The costs of life without strong closure Let us call “strong closure” any closure principle that can generate the skeptical paradox, and “weak closure” any principle that cannot. IC is a strong closure principle, while IC* is a weak closure principle. The question is, do we bear extraordinary costs, on having strong closure principles fail to be true of us? Some have argued along these lines in defense of strong closure principles. Now, as I noted above, if closure principles are descriptive, then if a closure principle is false of us, the costs of its failing to be true, whatever they may be, simply must be borne. So the better way to understand what defenders of strong closure have in mind is this: arguments for and against closure are vexed, and difficult to assess; so if we can show that some other, more obvious principles must be false if a candidate closure principle is rejected as false, then we’ll have good reason for thinking that the candidate closure principle is true — we’ll have good reason for thinking that purported arguments against the candidate closure principle must have gone off the rails somewhere. John Hawthorne, in one of the more thoughtful defenses of closure, argues that denying strong closure has very serious costs. Recall Hawthorne’s formulation of strong closure: Single Premise Closure (SPC): Necessarily, if S knows p, competently deduces q from p, and thereby comes to believe q, while retaining knowledge of p throughout, then S knows q.8 Hawthorne argues that if we deny SPC, we must also reject the following principles: Addition Closure (AC): Necessarily, if S knows p and competently deduces (p or q) from p, thereby coming to believe (p or q), while retaining knowledge of p throughout, then S knows (p or q). and
8. The difference between SPC and IC is very small: SPC includes an explicit demand that the subject retain knowledge of the premise of her reasoning throughout her reasoning, while this is a tacit demand of IC. So the two principles are very much the same.
42
Equivalence Principle (EP): Necessarily, if S knows a priori that p and q are equivalent, and S knows p, then S is in a position to know q (assuming appropriate deductive competence and opportunity). And, Hawthorne claims, both AC and EP are intuitively plausible — they are even more secure than strong closure.9 If rejecting SPC means rejecting AC and EP, Hawthorne argues, we must assume SPC is true of us, whatever our former reasoning has told us. Hawthorne’s argument is as follows: Deniers of closure, such as Dretske and Nozick, will want to grant: (6) (7)
I have a hand. Either I have a hand or I am not a brain in a vat.
But (7) is a priori equivalent to (8)
It is not the case that I lack hands and am a brain in a vat.
Now, (8) is the kind of proposition that Dretske and Nozick mean to deny that one knows precisely by denying closure. However, we derived (8) by means of EP alone. Similarly, deniers of closure will want to grant (in the relevant setting): (9) That is a zebra. But AC allows one to deduce from (9) (10) That is a zebra or that is a thing that is not a cleverly disguised mule. and (10) is a priori equivalent to
9. This is open to question; EP is certainly a very strong principle, as I suggest below.
43
(11) That is not a cleverly disguised mule. So again, (11) is the kind of proposition that Dretske and Nozick hope to deny one knows precisely by denying closure. However, we derived (11) by AC and EP. In sum, Hawthorne argues, the proposition that Dretske and Nozick mean to deny by denying closure can still be derived through use of AC and EP. Hawthorne concludes Since Dretske and Nozick deny that I can know claims like (8) and (11), they must reject AC [or EP]. But can we really live with the denial of th[ese] principle[s]? I doubt it (…) In relinquishing SPC, we are thus forced to relinquish certain other principles —Addition Closure and (…) Equivalence — that are very compelling. A denial of closure thus ramifies into costs that are extremely high. (2004, 40)
Hawthorne’s argument strategy is intriguing. Moreover, he is to be lauded for attempting to produce a substantive and independent consideration in favor of strong closure, instead of simply insisting on its intuitive plausibility. However, I believe that in the end his argument doesn’t succeed (although it might be telling against Dretske and Nozick, in their specific attempts to resist certain propositions through a denial of closure). In order for Hawthorne’s argument to secure the claim he suggests, namely, that denying SPC will force us to deny AC and EP, we need to see that AC or EP or both together imply SPC. Unless AC or EP imply SPC, it doesn’t follow that denying SPC requires the denial of either AC or EP or both. Now Hawthorne’s cases do not demonstrate that AC or EP or both together imply SPC. His cases show that a proposition implied by SPC is also implied by AC and EP together; and this does not show that AC and EP imply SPC. SPC might be false, then, while AC and EP are true. So far, then, Hawthorne’s argument is inconclusive. One might wonder, though, even if the denier of closure is not forced to deny AC or EP, might she not want to deny them in any case? Let’s consider AC and EP in turn. It seems that AC is actually an instance, or special case, of SPC, with (p or q) being the conclusion of a single premise inference from p. So, although one might not be forced to deny AC on denying SPC (since AC does not imply SPC), it is likely one will want to deny AC if one denies SPC. Consider EP next. It seems that EP too, bears a close relation to SPC. Namely, it seems that if one
44
conforms to EP and thereby comes to know q, the epistemic work in knowing q is done in fact by one’s conforming to SPC: specifically, if from p and p and q are equivalent, one can know q, it seems this is because one sees that q must be true if p is, which is the consideration that leads one to believe q in conforming to SPC. Just as in the case of AC, then, the connection to SPC is very tight. So, although one might not be forced to deny EP on denying SPC (since EP does not imply SPC), it is likely one will want to deny EP if one denies SPC. What matters for our purposes is this: Given the close ties between AC, EP and SPC, I think the defender of closure cannot rest her argument on AC and EP’s being intuitively obvious. AC and EP might seem intuitively obvious for the reasons that strong closure has seemed so intuitive — namely, because we fail to recognize that the principles in question are really quite strong, and that weaker counterparts do a better job expressing our actual epistemic judgments. The denial of AC and EP may very well seem natural, then, once we have absorbed the reasons for denying SPC. Now recall, the central reason for denying strong closure principles like IC, SPC and their ilk, is that they leave no room to acknowledge the fact that as epistemic agents we often have to catch up with ourselves, when we make inferences. Since this is true, there is reason to think that AC and EP could also use weakening to account for this fact. For instance, we might better express the intuitions behind AC and EP in this way: (AC*): Necessarily, if S knows p and competently deduces (p or q) from p, thereby coming to believe (p or q), while retaining knowledge of p throughout, and nothing stands in the way of S’s belief in (p or q) having the level of justification sufficient for knowing (p or q), then S knows (p or q). and (EP*): Necessarily, if S knows a priori that p and q are equivalent, S knows p, and nothing stands in the way of S’s belief in q having the level of justification sufficient for knowing q, then S is in a position to know q (assuming appropriate deductive competence and opportunity).
45
These weaker principles might be maintained even if we deny strong closure (in the form of SPC). At most we will be committed, in upholding AC* and EC* to a weakened version of SPC, namely SPC*: Single Premise Closure* (SPC*): Necessarily, if S knows p, competently deduces q from p, and thereby comes to believe q, while retaining knowledge of p throughout, and nothing stands in the way of S’s belief in q having the level of justification sufficient for knowing q, then S knows q.10 Before concluding we should return to Williamson’s concern — he gestured at the idea that if we reject closure we shall have a hard time saying in what circumstances we can gain knowledge by deduction. Surely it would be a significant cost to us if, in rejecting closure, we could not say under what circumstances we gain knowledge by deduction! I think the right reply here is to remember the distinction between strong and weak closure, and remember that we are rejecting strong closure principles only. If we have weak closure principles such as IC* or SPC* to fall back on, we can say in what circumstances we gain knowledge by deduction — namely, circumstances where one knows the premise(s), competently deduces the conclusion, and nothing antecedent stands in the way of one’s belief in the conclusion having the level of justification sufficient for knowing the conclusion. Moreover, should some antecedent factors stand in the way of one’s belief in the conclusion having the level of justification sufficient for knowledge, resolving those difficulties will permit one to gain knowledge through one’s deduction. It remains to be seen, then, whether there are any significant costs in denying that closure, in its strong forms, applies to us. So far as the costs we’ve considered, it seems safe to deny strong closure. 10. I should note that Hawthorne identifies a second cost of denying closure: Many suppose that knowledge is a norm of assertion — that to assert properly, one should know what one asserts — and, Hawthorne claims, “the denial of closure interacts disastrously with the thesis that knowledge is a norm of assertion” (2004, 39). So much the worse for the idea that knowledge is a norm of assertion, I would say: after all, it seems one can properly assert what one sincerely believes. Even if a case can be made for taking knowledge to be a norm of assertion, whether denying closure interacts disastrously with the norm will be open to dispute, since there are many competing norms of assertion, in light of which one may properly resist asserting things one knows. For this reason, identifying undisputed data from our linguistic practice will prove very difficult. And Hawthorne’s strategy is to find undisputed data that all can agree upon, so as to adjudicate the disputed issue regarding closure.
46
6. Weak closure and skepticism If life without strong closure is possible, perhaps that leaves us in an especially happy position? There are those who want to use the falsity of closure against the skeptic. Relevant Alternatives theorists, for instance, hold that closure is false, and that the falsity of closure is the key to defeating skepticism. If our moral is true, that strong closure principles — principles strong enough to generate skeptical paradox — are false, then perhaps we have an answer to skepticism? I am not so hopeful. This much is true of course: we do not face logical inconsistency when we resist the skeptic’s reasoning in the skeptical paradox, if strong closure is false. But it is a further question whether we thereby completely escape skeptical argument. I do not think we do. Strong closure is important for only one kind of skeptical argument, namely, the skeptical paradox. But there are many kinds of skeptical argument, and there is wide agreement that these other kinds of skeptical argument do not rely on closure principles. For instance, it has been pointed out that Cartesian skeptical challenges raised by indiscernible counter-possibilities are best understood as relying on underdetermination principles (Brueckner 1994). It has also been noted that still other Cartesian skeptical arguments center on doubts about the reliability of one’s cognitive equipment (Klein 2000). Neither of these challenges deploy epistemic closure principles. So the falsity of strong closure is not the key to answering wholesale traditional skeptical arguments. Consequently, while strong closure is false, its falsity does not win the day for us against the skeptic. 7. The contextualist’s life without (strong) closure Where does all this leave the contextualist? One of the claims of contextualism is that it handles the skeptical paradox neatly. If we no longer face a paradox, this claimed advantage lapses. That of course is not in itself an argument against contextualism. All it means is that a purported advantage of the view is not really an advantage. Contextualism might still be independently motivated. It is true, however, that if we no longer face a skeptical paradox, we are free to consider alternative semantics for “knows” that do not provide solutions to the paradox. Contextualism will have to win out
47
over these alternatives on other scores, especially on its capacity to make sense of features of our linguistic practices of knowledge attribution. Whether contextualism fares better on this score is open to debate (Hawthorne 2004, MacFarlane 2005). Finally, contextualists will have to break certain habits they have acquired in presenting their view as a solution to the skeptical paradox. It seems we sometimes face the epistemic predicament of knowing a proposition, p, and knowing that it has another proposition, q, as a logical implication, and yet not yet knowing that q. And we don’t always immediately know what sort of epistemic work needs to be done. Must we de-credential conflicting beliefs? Must we question the premises from which we made our deduction? Or both at once, until one or another course of action looks to be epistemically sound? Often, it isn’t transparent to one, as a rational being, what further work needs to be done in order to secure knowledge. Contextualists must not say, as was their wont when discussion the skeptical paradox, that we only seem to find ourselves in such predicaments. The task rather is to explain how such predicaments arise, and what we might do when faced with them, not to explain them away. And it is an open question whether contextualism in particular allows us to saying anything distinctive about such epistemic predicaments.
REFERENCES Audi, R. (1995). ‘Deductive Closure, Defeasibility and Scepticism: A Reply to Feldman’, Philosophical Quarterly 45, 494–499. Brueckner, A. (1994). ‘The Structure of the Skeptical Argument’, Philosophy and Phenomenological Research 54, 827–835. Cohen, S. (1988). ‘How to be a Fallibilist’, Philosophical Perspectives 2: Epistemology, (ed.) J. Tomberlin, 581–605. — (2000). ‘Contextualism and Skepticism’, Skepticism, (ed.) S. Villanueva, Blackwell, Boston. Dretske, F. (1981). ‘The Pragmatic Dimension of Knowledge’, Philosophical Studies 40, 363–78. DeRose, K. (1995). ‘Solving the Skeptical Problem’, Philosophical Review 104, 1–52. Feldman, R. (1995). ‘In Defence of Closure’, Philosophical Quarterly 45, 487–494.
48
Hales, S. (1995). ‘Epistemic Closure Principles’, Southern Journal of Philosophy 33, 185–201 Hawthorne, J. (2004). Knowledge and Lotteries, Clarendon Press, Oxford. Klein, P. (2000). ‘Contextualism and the Real Nature of Academic Skepticism’, Skepticism, (ed.) S. Villanueva, Blackwell, Boston. Lewis, D. (1996). ‘Elusive Knowledge’, Australian Journal of Philosophy 74, 549–567. MacFarlane, J. (2005). ‘The Assessment Sensitivity of Knowledge Attributions’, Oxford Studies in Epistemology, (eds.) T. Gendler & J. Hawthorne, Oxford University Press, Oxford. Stroud, B. (198). The Significance of Philosophical Scepticism, Clarendon Press, Oxford. Vogel, J. (1990). ‘Are There Counterexamples to the Closure Principle?’, Doubting: Contemporary Perspectives on Skepticism, (eds.) M. Roth & G. Ross, Kluwer, Dordrecht. Williamson, T. (2000). Knowledge and its Limits, Oxford University Press, Oxford.
49
This page intentionally left blank
Grazer Philosophische Studien 69 (2005), 51–69.
CONTESTING CONTEXTUALISM Patrick RYSIEW University of British Columbia, Canada Summary According to Keith DeRose, the invariantist’s attempt to account for the data which inspire contextualism fares no better, in the end, than the “desperate and lame” maneuvers of “the crazed theory of ‘bachelor’”, whereby S’s being unmarried is not among the truth conditions of ‘S is a bachelor’, but merely an implicature generated by an assertion thereof. Here, I outline the invariantist account I have previously proposed. I then argue that the prospects for sophisticated invariantism — either as a general approach, or in the specific form I have recommended — are not nearly as dim as DeRose suggests.
1. No one denies that whether you know that p depends on many features of you and your situation. Of course, people disagree over which such features are important — whether knowing depends on what else you believe, on objective features of your situation over and above whether p is true, on certain counterfactuals holding of you, on your evidence for (/against) p, on the etiology of your belief, and so on. The natural thought, though, is that once such factors are fixed, it’s either true that one knows or it isn’t — changes in such things as what the speaker has in mind in attributing knowledge to you don’t affect the truth conditions of what he says, and so don’t affect whether he speaks truly in saying that you know. But contextualists deny precisely this. In Keith DeRose’s words, “the truth conditions of sentences of the form ‘S knows that p’ or ‘S does not know that p’ vary in interesting ways depending on the context in which they are uttered” (DeRose 1992, 914), and ‘context’ here means none other than such things as the interests, expectations, and so forth of knowledge attributors (e.g., DeRose 1999, 190–191, Cohen 1999, 57).
Now, various theorists have argued that contextualism faces certain theoretical difficulties;1 others have claimed that the data that contextualists cite in supporting their view — including, and especially, the fact that there is a manifest flexibility in our willingness to attribute knowledge — are in fact compatible with thinking that the truth conditions of knowledge sentences don’t shift with changes in context. As to the latter claim, however, the ‘invariantist’ needs to do more than simply “‘chalk it all up to pragmatics’— to claim that the varying epistemic standards that govern our use of ‘knows’ are only varying standards for when it would be appropriate to say we know” (DeRose 2002, 194), and just leave it at that. Granted: since our willingness to utter a sentence, and our intuitions as to whether it’s true, may be due as much to pragmatic as to semantic factors, any direct inference from our verbal behavior to semantic claims would be a non sequitur. But by the same token, simply to say that the context-sensitivity of knowledge attributions is explained by pragmatics is not to show how that’s so. But while (nonsceptical2) invariantists have tended to rest content with “a ‘bare warranted assertability maneuver’” (DeRose 1999, 199; cf. 2002, 176), this is no longer the case. Elsewhere (Rysiew 2001), for example, I have presented an account of how pragmatic phenomena of a familiar Gricean sort, together with an invariantistic conception of the truth conditions of knowledge sentences, can be used to account for the flexibility in our willingness to attribute knowledge; and others have recently offered such accounts as well.3 Here, in Section 2, I briefly outline the account I have proposed. In Section 3 I argue, following Brown (forthcoming), that the reasons DeRose offers for being sceptical about the prospects of sophisticated invariantism as a general approach are open to question. In Section 4 I argue that DeRose’s reasons for finding the specific form of sophisticated invariantism I 1. For example, many complain that contextualism does not really help to address scepticism (see Feldman 1999 and 2001, Klein 2000, Kornblith 2000, and Sosa 2000); Hofweber (1999), Rysiew (2001), and Schiffer (1996) argue that the contextualist’s error theory — our getting mixed up about contexts — is in one way or another problematic; Stanley (2004) argues that ‘knows’ doesn’t behave like context-sensitive terms with which we’re familiar; and Cappelen and Lepore (2003) argue that contextualists rely on illegitimate context shifting arguments. 2. In the view of many, nonsceptical invariantism is the only variety of invariantism worth wanting. (Cf. Cohen 1999, 83 and DeRose 1999, 202.) 3. These include Bach (forthcoming), Brown (forthcoming), and Pritchard (forthcoming). (Importantly, Bach’s proposal does not rely on any form of ‘WAM’.)
52
have offered unsatisfactory are not decisive. If these arguments are on the right track, then sophisticated invariantism — whether as a general approach, or along the lines of the account I have proposed — remains a viable alternative to contextualism. 2. The key idea of sophisticated invariantism, and what separates it from invariantism simpliciter, is that it attempts to explain the data which inspire contextualism by appealing to a distinction between the literal contents of knowledge sentences and the information generated by (and inferable from) specific utterances thereof. How one states the particular sophisticated invariantist view I favor depends on how neutral one wishes to remain about the truth conditions of knowledge sentences. In neutral terms, the basic idea is this: the proposition literally expressed by the sentence ‘S knows that p’ is (unsurprisingly) that S knows that p; and, minimally, that S knows that p entails that S believes that p, p, and that S is in a good epistemic position with respect to p. It is a familiar idea, though, that what a given sentence means need not be what the speaker means in uttering it — i.e., what he intends to communicate. To capture this, I say that an utterance of ‘S knows that p’ pragmatically imparts (or conveys) a proposition of the form, ‘S’s epistemic position with respect to p is good enough …’, where the ellipsis is completed according to the interests, purposes, assumptions, etc., of the speaker. So, for instance, if I say, “I know who’s leading the NBA in scoring”, what I say — i.e., the proposition literally expressed — is true just in case the requirements for knowing are fulfilled. But, in a given case, I might say such a thing, not in order to remark upon my knowing (/not knowing) per se, but to get across the information that, for instance, my epistemic position is good enough that you can take my word on the matter. (I am able to do so partly in virtue of the fact that ‘S knows that p’ entails that S is in a good epistemic position with respect to p, and partly in virtue of the fact that language-users count on others to make their conversational contributions conform to the Cooperative Principle4 — see below.) 4. Or Bach & Harnish’s (1979) ‘communicative presumption’, or Sperber & Wilson’s (1986) ‘principle of relevance’. I tend to favor a Gricean framework, but a ‘relevance theoretic’ one, say, would serve just as well.
53
Obviously, given that whether my epistemic position with respect to p is good enough in the operative sense is going to depend upon our shifting interests, purposes, etc. — on whether we’re just chatting casually, whether we’re about to place a large bet, and so on –, there may be contextual variation in the information pragmatically imparted by the relevant utterance. Add to this the fact that our ‘semantic intuitions’— our intuitions about the truth conditions of sentences — are generally insensitive to the semantic / pragmatic distinction (we often read things into what is literally said, even by ourselves, without realizing we’re doing so), and we have all we need to explain why it can seem that the truth conditions of knowledge-attributing sentences depend on context when in fact they do not. Now, in my paper, I go beyond this neutral framework and adopt, for the sake of illustration, a particular elaboration of it. Specifically, I adopt a relevant alternatives (RA) view, whereby the goodness of epistemic position required for knowing that p is cashed out in terms of the subject’s ability to rule out the relevant not-p alternatives. Further, I suggest that the relevant alternatives are fixed by what we, qua normal humans, take to be the likely counter-possibilities to what the subject is said to know. As the standard of relevance is not unreasonably high, we preserve our judgments as to subjects’ (/our) possessing knowledge in a wide variety of situations. And as the standard is fairly invariantistic, the present view gives us a fairly invariantistic RA semantics for knowledge sentences: the mere mentioning of a certain not-p alternative, or the mere fact that the speaker has certain not-p possibilities in mind when he says, “S knows that p,” doesn’t affect what that sentence means (what it literally expresses). What such things do or can affect, however, is just what information a given utterance of a knowledge sentence is liable to pragmatically impart. So, that I’m just a (bodiless) brain in a vat (a BIV) isn’t likely to be a relevant alternative to my having hands (since the BIV possibility is something normal people hardly ever consider); but if that possibility has been raised, and I counter with, “No, I know that I have hands,” I’m liable to be taken to be implying (i.e., to mean) that I can somehow rule that possibility out. And so it usually goes, on this view: while the truth of the sentence requires only that the relevant alternatives be ruled out, a knowledge attributor typically means to get across the more specific information that the subject’s epistemic position is good enough that he’s able to rule out the salient alternatives — i.e., those alternatives that are actually ‘in
54
the air’. And since, patently, there will be variation in what the salient alternatives (to p) are, we can expect to find a corresponding flexibility in our willingness to attribute knowledge. As I say in the paper (2001, 487, 501), dissatisfaction with either the whole RA approach to knowledge or with my specific version thereof shouldn’t be taken to cast doubt on the general framework being employed. (The reader is free to use his/her own preferred account of knowledge in fleshing-out the framework sketched above.) The idea is just to give the reader a sense of how that framework might be deployed, given a specific understanding of what’s required for knowledge. It’s not required that that specific understanding be the one I’ve chosen to operate with. Nor is this ecumenicalism unmotivated. For all the general framework requires are some quite uncontroversial features of rational communication operating on only a partial, but in itself irreproachable, account of the truth condition of the relevant sentences. That is, given only a minimal view of the truth conditions of knowledge sentences — viz., that ‘S knows that p’ implies that S believes that p, p, and that S is in a good epistemic position with respect to p — , on the assumption that our linguistic exchanges are governed by something like Grice’s Cooperative Principle (CP), we have every reason to expect that there will be a flexibility in our willingness to attribute knowledge along the lines I have suggested. Specifically, it’s because speakers strive to conform, and are known to so strive, to the maxim of relation (“Be relevant”) that, in uttering a sentence of the form, ‘S knows that p’, the speaker is naturally taken to intend / mean that S’s epistemic position with respect to p is ‘good enough’ given the epistemic standards that are operative in the context in question. For it is only if speakers are understood to be intending to communicate information about how the subject fares vis-à-vis the contextually operative standard(s) — hence, i.a., about whether the subject can or cannot rule out any contextually salient alternatives — that they can be seen as striving to be maximally relevantly informative (hence, as conforming to CP) (2001, 491–2).5 Now, it is being assumed here that information can be pragmatically generated by a given utterance without requiring for its generation the complete semantics of the sentence uttered. After all, ordinary speak5. The injunction to make one’s conversational contribution ‘maximally relevantly informative’ is Harnish’s (1976) suggested encapsulation of CP.
55
ers seem rather adept at communicating information by extra-semantic means; but even those whose job it is to provide truth conditional analyses of certain familiar sentence types can seldom agree on what they are. Further, there seem to be clear cases in which the pragmatic generation of certain information depends upon semantic features of the relevant sentences while leaving the exact truth conditional analysis thereof underdetermined. Thus, you don’t need to have successfully Chisholmed sentences of the form ‘O looks F [to S]’ in order to infer, from the fact that someone says, “The table looks black,” that they have some doubt about whether it really is black — all you need to know is that, since x’s looking F doesn’t entail that x is F, if the speaker does think that the table is black, that’s what he should say, since that would be more informative. Similarly, you don’t have to have anything worth calling a theory of belief in order to find it natural to say “I believe that p” in order to indicate that you’re not absolutely certain as to p — since your believing that p would already be obviously inferable from you’re saying “p”, and since ‘S believes that p’ doesn’t imply that p, you can make explicit use of the term in order to emphasize the contrast with the categorical, “p”, thereby flagging your lack of certainty.6 And so too in the present case: you don’t have to have a complete and satisfactory theory of knowledge in order to think that speakers might utter knowledge sentences in order to get across quite specific information about how people (themselves or others) measure up to the operative epistemic standards. All you need is to see is that if I didn’t think that S did so measure up, it would be odd, indeed uncooperative, of me to say, “S knows that p”; for whatever exactly knowledge is, we know that S’s knowing that p entails that S is in a good epistemic position with respect to p. “And why”, the hearer may well ask, “would he say something that means [semantically implies] that if he didn’t in fact think that S measured up to the epistemic standards that are in play, and so was able to put to rest any doubts or rule out any not-p possibilities that had just been raised?”7 6. Even if some of our pretheoretic intuitions surrounding the semantics of the expressions in question end up being explained away, some such inputs to truth-conditional analyses of the relevant expressions are necessary (otherwise, we won’t know where to begin); and the semantic facts being assumed in the examples just given — that neither ‘x looks F’ nor ‘S believes that x is F’ implies that x is F — seem as widespread and uncontroversial as any. 7. Cf. Grice’s suggestion as to a non-conversational analogue of the maxim of relation:
56
Notice a second quite general point that can be gleaned from the examples cited just above. Those examples might be seen as relying on DeRose’s ‘assert the stronger’ rule: [T]here’s a very general conversational rule to the effect that when you’re in a position to assert either of two things, then, other things being equal, if you assert either of them, you should assert the stronger. (2002, 175; 1999, 197)
However, it is less clear whether the examples illustrate this rule if we take the ‘strength’ in question to be such that ‘p’ is stronger than ‘q’ just in case it “impl[ies] but [is] not implied by” ‘q’ (ibid.). For instance, if what’s at issue is whether x is F, intuitively, “x is F” is ‘stronger’ than either “S believes that x is F” or “x looks F”. — Certainly, someone’s saying either of the latter would be less informative than the simple, “x is F”.8 But it seems false that x’s being F implies(-but-isn’t-implied-by) the proposition that x appears F [to S], or that S believes that x is F. So, if it is ‘assert the stronger’ which is the conversational rule at work in examples like those just given, the ‘strength’ in question is something like the overall informativeness of one’s conversational contribution, given what’s at issue.9 Better, then, to see the examples just given as illustrating Grice’s maxim of quantity — “Make your [conversational] contribution as informative as is required (for the current purposes of the exchange)” (1989, 26) — , and say that uttering a sentence which expresses a proposition that is weaker (in DeRose’s sense) than something else one is in a position to say is not the only way of being less informative than one might have been.10 Either way, the important point “I expect a partner’s contribution to be appropriate to the immediate needs at each stage of the transaction” (1989, 28). 8. Once again, assuming that what the conversants are interested in is x and its features, and not facts about S. If the latter were what they were interested in, being told that S believes that x is F, e.g., might well be more informative than being told that x is F. 9. It was because I was thinking of ‘strength’ in this way that it seemed to me that both “I know that p” and “I know that not-p” were stronger than “It’s possible that p” (assuming that what’s at issue is whether p): either of the former evinces a commitment to something that entails a univocal answer as to whether p, whereas, whatever its exact truth-conditional content, ‘It’s possible that p’ does not; hence, from a speaker’s saying the latter, one can infer that the speaker doesn’t (take himself to) know either that p or that not-p. — Such was my thinking. Even so, it is of course a mistake to suggest that one might execute a ‘WAM’ of both ‘I don’t know that p’ and ‘I don’t know that not-p’ using DeRose’s ‘assert the stronger’ rule and his stated notion of strength (see my 2001, 510, n. 30, and DeRose 2002, 198–199, n. 40). 10. Thanks to Kent Bach for discussion on this point.
57
is that, just as you needn’t have a full and satisfactory account of the truth conditions of the relevant sentence in order to be able to use it to get across, or to glean from its use, information that’s plausibly not part of its truth conditional content, you don’t have to restrict yourself to strictly logical relations among the propositional contents of the sentences people utter, or might have uttered, in order to explain why certain information is pragmatically generated by their saying what they do — just as relevant are such things as which questions the parties are concerned to resolve. These are the sorts of facts about linguistic communication which it is important to keep in mind in assessing attempts to explain certain linguistic behavior by appeal to pragmatic factors, as well as in evaluating DeRose’s reasons for being sceptical about the viability of sophisticated invariantism; it’s to those reasons that we now turn. 3. With a view to developing a general account of why some ‘warranted assertability maneuvers’ (WAMs) succeed and others fail, DeRose gives the following examples of each: (i) ‘lame’ WAM (1999, 197; 2002, 174): A “crazed philosopher of language” gives an account whereby S’s being unmarried is not among the truth conditions of the sentence, ‘S is a bachelor’; rather S’s being unmarried is simply among the warranted assertability conditions of such a sentence. (ii) credible WAM (1999, 196–7; 2002, 174–5): If someone knows that p, it seems wrong, even false, for him to say (merely), “It’s possible that p”; this might tempt us to think that ‘It’s possible that p’ is true iff S ‘doesn’t know either way’ (DKEW) — i.e., iff (1) S doesn’t know that p and (2) doesn’t know that not-p. However, a defender of the ‘doesn’t know otherwise’ (DKO) account, who thinks that (2) alone is a genuine truth condition of the sentence, could argue that the appearance of (1)’s being a truth-condition derives from the fact that ‘It’s possible that p’ is weaker than ‘I know that p’, and therefore that, from the fact that someone asserts the former, we can infer that he doesn’t know that p.
58
What makes (ii) a credible WAM and (i) a miserable failure? According to DeRose, the reasons for according (ii), but not (i), some legitimacy, are as follows. First, (ii) is motivated by genuinely conflicting intuitions: where someone knows that p, it would seem wrong for him to say either “It’s possible that p” or “It’s not possible that p”; so we have good reason to think that some apparent falsehood is going to have to be explained away here (ibid.; 2002, 192). But there is a conspicuous lack of genuinely conflicting intuitions in the case of (i): “in the problem cases for the crazed theory of bachelor … there’s no such pressure to have to explain away any appearances: It seems false to say of married men that they are bachelors, and it seems true to say of them that they are not bachelors” (1999, 198; 2002, 192). Further, as it stands, (i) is just a bare warranted assertability maneuver, whereas the second WAM, utilizing “a very general conversational rule” (‘assert the stronger’— see above), explains the apparent falsity in terms of the generation of a false implicature. Here is DeRose’s illustration, wherein the speaker knows that p, but has said only, “It’s possible that p”: To assert that weak possibility statement is unwarranted and generates the false implicature that the speaker doesn’t know that P by the following Gricean reasoning, which is based on the assumption that the speaker is following the “assert the stronger” rule: “If he knew that P, he would have been in a position to assert something stronger than ‘It’s possible that Pind’, and thus would have asserted some stronger thing instead. But he did assert ‘It’s possible that Pind’, not anything stronger. So he must not know that P.” We can easily mistake the falsehood of the implicature generated by such an assertion of “It’s possible that Pind” and the resulting unwarrantedness of the assertion for the falsehood of the assertion itself. (2002, 175; cf. 1999, 197)
DeRose is of course right that WAMing shouldn’t be allowed to be carried too far, so that, “[w]henever your theory seems to be wrong because it is omitting a certain truth-condition … you can simply claim that assertions of the sentences in question generate implicatures to the effect that the condition in question holds” (2002, 173). But, on the face of it, the sophisticated invariantist view outlined above satisfies DeRose’s criteria for a successful WAM. Thus, while it can seem wrong to claim that I know that I’m not a BIV, for example, it can seem no less wrong to say that I don’t know that I’m not a BIV. After all, I know that I have hands, and BIVs are handless (cf. DeRose
59
1995, 2).11 In terms of this same example, I explain the apparent falsity of “I know that I’m not a BIV” by appeal to the falsity of what would be conveyed by an utterance thereof — viz., that I can rule out the possibility that I’m a BIV; and the conversational rules appealed to — Grice’s CP and, specifically, the maxim of relation — are as general as one could like. But matters are not so straightforward. For, if DeRose is right, when we look closer we find reason to doubt whether an invariantist WAM such as the one just described can really succeed. Thus, it’s a central idea of the present account that speakers can utter knowledge sentences in order to convey information that plausibly is not part of those sentences’ truth conditional contents. And since it is the information actually (pragmatically) conveyed that guides speakers in their knowledge-attributing (/-denying) practices, speakers will sometimes resist uttering a (literal) truth because they rightly sense that in doing so they’d be committing themselves to something false.12 So, once again, I suggest that ‘I know I’m not a BIV’ is one such sentence; it is, I claim, true (given that my being a BIV is not a relevant alternative), but an utterance thereof will standardly convey that the speaker can rule out the possibility that he’s a BIV, and speakers (understandably) might not want to commit themselves to that. In fact, speakers might go so far as to say something that’s literally false — e.g., “I don’t know that I’m not a BIV” — in order to communicate a salient truth (viz., that they can’t rule out the possibility that they are BIVs). Nor is this surprising if speakers are, as seems plausible, concerned with communicating (though not necessarily uttering only literal) truths. DeRose, however, finds this sort of claim incredible: [I]nvariantists do not begin with a good candidate for WAMing, and they have to explain away as misleading intuitions of truth as well as intuitions of falsehood. For in the “low standards” contexts, it seems appropriate and 11. See the discussion of the (apparent) reality of conflicting intuitions in the cases in question, below. 12. As discussed below, the claim is not that speakers would necessarily have transparent knowledge of the matter, in just these terms. Like many others, I think that ordinary speakers (including us, most of the time) simply don’t distinguish between information that’s semantically encoded in sentences and that which is pragmatically generated by various utterances thereof. And why should they (we)? What they (we) care about, first and foremost, is the overall communicative content of utterances, not the relative contributions thereto of semantic and pragmatic factors.
60
it seems true to say that certain subjects know and it would seem wrong and false to deny that they know, while in the “high standards” context, it seems appropriate and true to say that similarly situated subjects don’t know and it seems inappropriate and false to say they do know. Thus, whichever set of appearances the invariantist seeks to discredit — whether she says we’re mistaken about the “high” or the “low” contexts — she’ll have to explain away both an appearance of falsity and (much more problematically) an appearance of truth. (2002, 193) [But,] except where we engage in special practices of misdirection, like irony or hyperbole, don’t we want to avoid falsehood both in what we implicate and (especially!) in what we actually say? (2002, 192–3; 1999, 199)
Thus, as DeRose sees it, the data in question are simply unfriendly to an invariantistic handling: when we restrict our judgments (as we should) to their proper contexts, we see that our intuitions in the cases in question do not in fact conflict; and once those judgments are properly understood, we see that any invariantist account of the data is going to face the unenviable task of having to explain why we would utter a literal falsehood in order to communicate some truth. As Brown (forthcoming) argues, however, on neither of these counts is DeRose’s argument persuasive. First, if we consider the sorts of cases the proper handling of which is at issue, it seems that every party to the current debate is going to have to explain away an appearance of some truth. As DeRose says of the independently plausible but (apparently) mutually inconsistent propositions which together constitute the sceptical “argument from ignorance”, for instance, “something plausible has to go” (1995, 2). Of course, the contextualist denies that conflicts of intuitions such as the one generated by the argument from ignorance, or more generally by comparing what it seems “appropriate and true” to say in high standard cases with what it seems “appropriate and true” to say in low standards cases, are genuine. —According to him, the appearance of conflict stems from our conflating contexts; so no apparent truths need actually be denied. But such cases draw our interest and seem to us to be of theoretical import precisely because it seems to us to be true that we have conflicting intuitions; and, as Brown says (forthcoming: Section 2), that that isn’t in fact the case — that our intuitions are, in fact, mutually consistent — is among the contextualist’s central claims. So if having to explain away an apparent truth is in itself a sign that one is headed down the wrong theoretical path, that’s bad news for everyone.
61
Granted, the contextualist doesn’t have us deliberately uttering falsehoods so as to communicate truths; rather, on his view we simply fail to see that certain things — e.g., that certain of our judgments really are in conflict — are false. But, as will be clear from our discussion to this point, this response rests on the misunderstanding of the sort of sophisticated invariantist view being defended here. For it is essential to that view that speakers often conflate what is semantically expressed by a sentence with what an utterance thereof pragmatically imparts: it is among the sophisticated invariantist’s central claims that our pretheoretic intuitions as to what we’re ‘saying’ are insensitive to the semantic/pragmatic distinction (Rysiew 2001, Sections 3–4, Bach 1994 & forthcoming, Brown forthcoming, Section 3).13 To the extent that this is so, whether speakers see themselves as ‘saying’ something true/ false will depend on the perceived truth value of the information that is merely pragmatically conveyed — if they think it’s true, they won’t see themselves as expressing falsehoods, deliberately or otherwise. Further, as Brown points out (ibid.), there are cases in which certain theorists have found it plausible to think that a literally false sentence can seem true because of the information an utterance thereof would communicate — those theorists may be wrong about this, of course; but the idea is “part of a standard and well-entrenched approach in the philosophy of language.” Given that this is so, intuitions of the sort DeRose is invoking here cannot, in themselves, be used to arbitrate between sophisticated invariantism and contextualism. To suppose that they can begs the question.14
13. It this because of this that I have been hesitant (2001, 510 n. 32; cf. nn. 26, 29) to use ‘implicature’ in stating my view, preferring to speak more neutrally of information that is merely ‘(pragmatically) imparted’ or ‘conveyed’. In implicatures, properly so-called, one means what one (literally) says but also something else. But in less than fully literal speech — e.g., in Bach’s (1994) implicitures, one of the cases Brown discusses — what one means is not what one (literally) says, but the more specific information that is generated by one’s utterance. 14. And not just against sophisticated invariantists. It begs the question against the many philosophers and linguists — for instance, Bach (1994), Sperber & Wilson (1986), Recanati (1989), Carston (1988), and Gibbs & Moise (1997) — who have argued that in very many cases, the literal truth conditions of the uttered sentence fails to correspond to what the speaker intends to communicate. In their view, irony, figurative speech, and so forth, are best viewed as limiting cases on a continuum of less than fully literal speech.
62
4. In addition to being pessimistic about the prospects for sophisticated invariantism as a general approach, DeRose finds my particular brand thereof unsatisfactory. He writes: … While, as Rysiew points out, his account is “underwritten” by the very general conversational rule, “be relevant,” his is not the type of account I have urged is to be desired. On Rysiew’s relevant-alternativist theory of knowledge attributions, an assertion of “S knows that P” carries two separate but related meanings … Where the contextually salient alternatives diverge significantly from the relevant alternatives, the injunction to “be relevant” gets the listener to fasten on the second, pragmatic meaning, since the semantic content of “S knows / doesn’t know that P” is clearly irrelevant to the purposes at hand. Thus, on Rysiew’s treatment (490) of my “high-standards” Bank Case, I utter the falsehood that I “don’t know” that the bank is open on Saturday, but all is well, for my wife can be counted on to ignore the falsehood I strictly said, since mere “knowing” / “not knowing” is not what is relevant to our there unusually heightened concerns, and focus instead on the second, pragmatic meaning that my assertion carries — that it’s not the case that I have a true belief and can rule out all of the salient alternatives to P. I am highly suspicious of accounts that help themselves to two such meanings in the way Rysiew’s does, disliking not only the loss of economy in explanation, but also worrying that we will not be able to combat all manner of absurd theories about the truth-conditions of various sentence if defenders of those theories are able to posit separate pragmatic meanings that do the work of accounting for usage in the troublesome cases, allowing their account of the truth-conditions to sit safely off in the corner, above the fray … (2002, 198, n. 17)
As should be clear from the summary given above, however, that assertions of knowledge sentences “carry two meanings” is not a bare stipulation of the present account. The leading idea of that account, rather, is simply that we need to distinguish between the content of a given sentence and what one might convey in uttering it on a given occasion; and that, with some help from general conversational rules (as well as parties’ knowledge of the context, their general background knowledge, and so forth), uttering a given sentence (in context) can, and often does, serve to communicate information over and above, or even quite other than, what the sentence itself expresses. — Not, of course, that this means that the latter information (in the present case, that the
63
subject can/can’t rule out the salient alternatives) is something that the sentence ‘carries around’, as it were. On the contrary, the whole point of appreciating pragmatic phenomena for what they are is to avoid such things as theorists’ introducing extra meanings by hand. Then again, if the role of the meaning of knowledge sentences on the present account was to trigger, thanks solely to the irrelevance of that semantic content in the cases in question, the hearer’s inference to the information that is said to be (merely) pragmatically generated, that would be a problem. For then that content could be just about anything, and its connection to the information I allege to be pragmatically conveyed would be mysterious — the latter would be in no interesting way “generated by the semantic meaning together with general principles” (ibid.). However, the present account does assign knowledge sentences’ semantic content an essential role in leading the hearer to the information I claim is pragmatically conveyed, and not just in virtue of its (sometimes) irrelevance. Let me clarify how this is so. In outlining the general framework above (Section 2), I indicated how a minimal but still substantive account of knowledge sentences’ truth conditions, interacting with familiar Gricean principles (especially, the maxim of relation), might explain the generation of the information which I claim is imparted by actual attributions/denials of knowledge; and in my paper I suggested that, supposing a particular invariantistic elaboration of knowledge sentences’ truth conditions to be correct, we can predict the pragmatic generation of the sort of data that have seemed to some to favor contextualism. Granted, neither the discussion of Section 2, nor that of the hearers’ inferences in the cases in question that I’ve offered — e.g., in the high standards bank case (2001, 490–2) — requires the specific invariantistic RA semantics I described above.15 But this doesn’t show that the generation of information that is said to be merely pragmatically conveyed does not essentially rely on the meaning of knowledge sentences. For recall (Section 2) that such information can depend for its generation on (presumed) semantic features of the sentences involved while leaving the correct truth conditional analysis of those sentences underdetermined. And, in the 15. Much less does it require that hearers have a complete and explicit grasp of the distinction between relevant and merely salient alternatives, as opposed to some (admittedly fallible, though on the whole reliable) judgments as to the sorts of cognitive-epistemic abilities that are and aren’t required for knowing.
64
present case, if ‘S knows that p’ didn’t semantically imply (at least) that S is in a good epistemic position with respect to p, the inference I allege — from a speaker’s saying “I [/don’t] know that p” to his [/not] being (or, his representing himself as believing that he is [/not]) in a good enough epistemic position to be able to rule out the salient not-p alternatives — would not go through. For example, in the reconstruction just mentioned, I portray the hearer as relying on the fact that “‘to say “I don’t know …” is to indicate that one isn’t sure’” (2001, 491); and the account of why firstperson knowledge claims might be used to express one’s confidence (492) — hence, why their denials might be used to indicate that one isn’t sure — requires that the subject’s fulfilling the conditions for knowing is seen to be both implied by ‘S knows that p’ and presumed by a person’s asserting, simply, that p. Further, though I do not say this in the paper, in interpreting her husband’s utterance the hearer in this example is plausibly seen as relying as well, even if only implicitly, on the literal content of ‘I don’t know the bank will be open’, inasmuch as she has no reason to think either that the speaker does not take himself to have a true belief (there is no indication that he doubts the bank will be open on Saturday), or that he takes himself not to be in a good epistemic position with respect to the bank’s being open, period (he’s been to the bank on a very recent Saturday, for example, and under other circumstances would be credited with knowing). Together with the assumption that he is striving to conform to the maxim of relation, these points make it natural to suppose that when the husband says, “I don’t know …”, he means, and is taken to mean, that his epistemic position isn’t so good that he can rule out the possibility, which his wife has just explicitly raised, that the bank has very recently changed its hours. — Or so it seems to me; and I could just be wrong about this. But I don’t wish to say that my discussion of these matters is decisive, or that it couldn’t be improved upon.16 The present point, rather, is that the specific semantic content of knowledge sentences plays an essential role on the account being defended here, even if its playing that role does not require that the RA semantics described above be correct. What the eliminability of the latter shows is just that a particular invariantistic account of what 16. For example, Brown (forthcoming, Section 4) argues that I under-rate the role of practical importance (as against salience) in our knowledge-attributing behavior, and offers her own reconstruction of (e.g.) the high standards bank case.
65
constitutes the goodness of epistemic position required for knowing is not singled out by the phenomena in question. That is why, as I have emphasized, the general sophisticated invariantist framework I’ve described is largely neutral as between differing accounts of the exact truth conditions of knowledge sentences. The foregoing would, of course, constitute a problem for the present view if it was intended to show that an invariantistic RA semantics is the correct account of the truth conditions of knowledge sentences. But it is not; the task, rather, is to show that that account, indeed any (nonsceptical) invariantistic semantics, could be correct: just as DeRose’s WAM is intended to show that the DKEW theorist’s proposed extra truth condition on ‘It’s possible that p’ can be explained away (assuming that DKO is correct, and using CP), the present account is intended to show that if an invariantist semantics such as RA is true of knowledge sentences, that semantics, together with CP, can account for the data which have been taken to favor contextualism. That said, though, if we do distinguish between the truth conditions of a given type of sentence and the information generated by particular tokenings thereof, how are we going to go about evaluating this or that account of what the former are? Indeed, how are we going to decide whether to adopt a contextualist or a non-contextualist stance towards those truth conditions? Well, of course, there are no fast and easy answers to these questions. But this is hardly news. We start off with some intuitions as to the truth conditions of a certain class of sentences; in the face of one or another reason — some conflicting intuitions, e.g. — for thinking that some other account of those truth conditions, or some other view of their sensitivity to context, might be correct, we have to decide which account which makes best overall sense of all of the relevant data — hence, which intuitions should in the end be directly reflected in our final account of the truth conditions, and which should be explained away. The process is seldom quickly decided; but it is, unfortunately, all too familiar, and there is no reason to think that any view which issues from it is going to inevitably be ad hoc. Finally, it’s worth pointing out that the view of the context-sensitivity of knowledge attributions I have been defending here may actually be supported by some of DeRose’s own observations (cf. Rysiew 2001, 492):17 As DeRose and others have suggested, in asserting that 17. This is not to say that what follows is its sole support. What follows requires first-
66
p one represents oneself as knowing that p (see DeRose 2002, 179ff.). DeRose (2002) uses this fact as a basis for a new kind of argument for contextualism. Arguably, though, it points away from contextualism and towards sophisticated invariantism. For if asserting that p already involves representing oneself as knowing that p, this gives us reason to suppose that, when speakers who are striving to conform to the CP do actually say, “I know that p”, they might do so in order to get across information that wouldn’t already be obviously inferable from their saying, simply, “p” (including, that they take themselves to know that p). Whereas, the flexibility in our knowledge-attributing practices is most naturally seen as supporting contextualism when we take speakers to be guided solely by what the sentences they utter mean — no more, and no less. 5. As DeRose says, “[i]t is clear and uncontroversial that knowledge attributions are in some way governed by varying epistemic standards” (2002, 187). What is controversial is whether those varying standards bear upon the truth conditions of knowledge attributions. Here, I have argued that we need not be as pessimistic as is DeRose about the viability of sophisticated invariantism, either as a general approach, or in the form I have recommended. How, then, are we to decide between contextualism and sophisticated invariantism? The question is one of deciding which account of the context-sensitivity of knowledge attributions, the contextualist’s or the sophisticated invariantist’s, is favored by the balance of considerations. The goal here has been to suggest that the balance doesn’t obviously tip the contextualist’s way.18
person attributions; the general rationale for the approach (Section 2) does not. 18. Thanks to Kent Bach (who also suggested the title), Martijn Blaauw, Jessica Brown, Adam Leite, and Mohan Matthen for their thoughtful and helpful comments on earlier drafts of this paper.
67
REFERENCES Bach, K. (1994). ‘Conversational Impliciture’, Mind and Language 9, 124– 162. — (forthcoming). ‘The Emperor’s New “Knows”’, Contextualism in Philosophy: On Epistemology, Language and Truth, (ed.) Gerhard Preyer and Georg Peter, Oxford University Press, Oxford. Bach, K. and R. M. Harnish (1979). Linguistic Communication and Speech Acts, MIT Press, Cambridge, Massachusetts. Brown, J. (forthcoming). ‘Contextualism and Warranted Assertibility Manoeuvres’, Philosophical Studies. Cappelen, H. and E. Lepore (2003). ‘Context Shifting Arguments’, Philosophical Perspectives 17: Language and Philosophical Linguistics, 25–50. Carston, R. (1988). ‘Implicature, Explicature, and Truth-theoretic Semantics’, Mental Representations: The Interface Between Language and Reality, (ed.) R. M. Kempson, 155–181, Cambridge University Press, Cambridge. Cohen, S. (1999). ‘Contextualism, Skepticism, and The Structure of Reasons’, Philosophical Perspectives 13: Epistemology, (ed.) James E. Tomberlin, 57–89, Ridgeview, Atascadero CA. DeRose, K. (1995). ‘Solving the Sceptical Problem’, The Philosophical Review 104.1, 1–52. — (1999). ‘Contextualism: An Explanation and Defense’, The Blackwell Guide to Epistemology, (eds.) J. Greco and E. Sosa, 185–203, Blackwell, Cambridge, MA. — (2002). ‘Assertion, Knowledge and Context’, The Philosophical Review 111.2, 167–203. Feldman, R. (1999). ‘Contextualism and Skepticism’, Philosophical Perspectives 1: Epistemology, 91–114. — (2001). ‘Skeptical Problems, Contextualist Solutions’, Philosophical Studies 103, 61–85. Gibbs, R. W., and J. F. Moise (1997). ‘Pragmatics in Understanding What is Said’, Cognition 62, 51–74. Grice, H. P. (1989). Studies in the Way of Words. Harvard University Press, Cambridge, MA. Harnish, R. (1976). ‘Logical Form and Implicature’, An Integrated Theory of Linguistic Abilities, (eds.) T. Bever, J. Katz, T. Langendoen, Crowell. Reprinted in S. Davis (ed.), Pragmatics: A Reader, Oxford University Press, 1991. Hofweber, T. (1999). ‘Contextualism and the Meaning-Intention Problem’,
68
Cognition, Agency and Rationality, (eds.) K. Korta, E. Sosa, and X. Arrazola, 93–104, Kluwer, Dortrecht, Boston, and London. Klein, P. (2000). ‘Contextualism and the Real Nature of Academic Skepticism’, Philosophical Issues 10, 108–16. Kornblith, H. (2000). ‘The Contextualist Evasion of Epistemology’, Philosophical Issues 10, 24–32. Lewis, D. (1979). ‘Scorekeeping in a Language Game’, Journal of Philosophical Logic 8, 339–359. Pritchard, D. (forthcoming). ‘Contextualism, Scepticism, and Warranted Assertibility Manoeuvres’, Knowledge and Skepticism, (ed.) J. Keim-Campbell, M. O’Rourke & H. Silverstein, MIT Press, Cambridge, MA. Recanati, F. (1989). ‘The Pragmatics of What is Said’, Mind & Language 4.4, 295–329. Rysiew, P. (2001). ‘The Context-Sensitivity of Knowledge Attributions’, Noûs 35.4, 477–514. Schiffer, S. (1996). ‘Contextualist Solutions to Scepticism’, Proceedings of the Aristotelian Society 96, 317–333. Sosa, E. (2000). ‘Skepticism and Contextualism’, Philosophical Issues 10, 1–18. Sperber, D., and D. Wilson (1986). Relevance: From Communication to Cognition, Blackwell, Cambridge, MA. Stanley, J. (2004). ‘On the Linguistic Basis for Contextualism’, Philosophical Studies 119, 119–146.
69
This page intentionally left blank
Grazer Philosophische Studien 69 (2005), 71–99.
COMPARING CONTEXTUALISM AND INVARIANTISM ON THE CORRECTNESS OF CONTEXTUALIST INTUITIONS Jessica BROWN University of Bristol Summary Contextualism is motivated by cases in which the intuitive correctness of a range of phenomena, including knowledge attributions, assertions and reasoning, depends on the attributor’s context. Contextualists offer a charitable understanding of these intuitions, interpreting them as reflecting the truth value of the knowledge attributions and the appropriateness of the relevant assertions and reasoning. Here, I investigate a range of different invariantist accounts and examine the extent to which they too can offer a charitable account of the contextualist data
0. Introduction Contextualism is motivated by cases in which the intuitive correctness of a range of phenomena, including knowledge attributions, assertions and reasoning, depends on the attributor’s context. Contextualists take such cases to show that the truth conditions of knowledge attributions depend on the attributor’s context. For instance, they suggest that, in the high context of such cases, the denial of knowledge to the subject seems correct because it is literally true. Invariantists deny that the truth conditions of knowledge attributions depend on the attributor’s context. So, it may seem that invariantists need to treat our intuitions about contextualist cases as misguided. For instance, if invariantists hold that, in the high context of the contextualist cases, the denial of knowledge is literally false then, on their view, our intuition that the denial is correct seems misguided. Prima facie, if invariantism treats our intuitions about contextualist cases as misguided, this may seem to count against the view. DeRose argues that it is objectionable for an account of ‘knows’ to treat as
misguided our intuitions about simple applications of the term, suggesting that ‘what speakers know best about the piece of language at issue is how and when to make simple positive and negative applications of the term in question’ (DeRose forthcoming, 24). Some suggest this prima facie consideration may be counterbalanced by considering our intuitions about more complex uses of ‘knows’. They suggest that invariantism provides a more charitable treatment than contextualism of metalinguistic judgements about the truth-value of knowledge attributions, comparative judgements about whether various knowledge attributions and denials conflict, and judgements about the correct use of ‘knows’ in speech and belief reports (e.g., Hawthorne 2004, §2.4–2.7; Williamson forthcoming and Wright forthcoming).1 Here I focus just on the extent to which invariantism can offer a charitable treatment of intuitions about simple applications of ‘knows’ and related intuitions concerning assertion and practical reasoning. If an invariantist account need not treat these intuitions about contextualist cases as misguided, then the prima facie concern that invariantism is uncharitable does not arise. In discussing this issue I will focus solely on so-called ‘classic invariantism’, which denies that the truth-value of attributions of knowledge depends on whether, in either the attributor’s or the subject’s context, the issue is important and error has been raised. Subject-sensitive invariantism also denies that the truth-value of knowledge attributions depends on the attributor’s context, but argues that it does depend on whether, in the subject’s context, the issue is important and error has been raised (e.g., Hawthorne 2004; Stanley forthcoming). In virtue of this subject-sensitivity, subject-sensitive invariantism can offer a charitable explanation of the contextualist’s intuitions at least in cases where the attributor is the subject. For instance, in a first-person contextualist case, the subject-sensitive invariantist can argue that, in the high context, it seems correct for the attributor to deny that she herself knows since the denial of knowledge would be true. Since the classic invariantist denies that the truth-value of knowledge attributions depends on whether, in either the subject’s or attributor’s context, the issue is important and error has been raised, it seems to face a tougher challenge in accommodating contextualist intuitions than subject-sensitive invariantism. In focusing on classic invariantism, then, we are making the task as 1. For counter-arguments see Cohen (forthcoming) and DeRose (forthcoming).
72
hard as possible for the invariantist. From now on, I will use the term ‘invariantism’ to refer to classic invariantism alone. 1. Intuitions about contextualist cases In Cohen’s airport case, Mary and her companion John are wondering whether the New York flight stops in Chicago. Mary asks someone, Smith, whether he knows whether the flight stops there. Smith replies, ‘Yes, I know the New York flight stops in Chicago; my itinerary says so’. Suppose that Mary and John are in a low context (hereafter ‘LOW’), in which it’s not important whether the New York flight stops in Chicago, and no one has raised any error possibilities. In LOW, Mary would attribute knowledge that the New York flight stops in Chicago to Smith. Further, her attribution of knowledge to Smith seems correct. By contrast, it would seem incorrect if she were to deny that Smith knows. Further, it seems correct for her to assert that the New York flight stops in Chicago and to use that proposition in practical reasoning. For instance, she might reason that since the New York flight stops in Chicago, the journey will be longer than if the flight were non-stop and she should buy an extra magazine to read. Now suppose that instead Mary and John are in a high context (hereafter ‘HIGH’) in which it’s very important that the plane stops in Chicago and John has raised the possibility of itinerary error (perhaps, Mary and John are heart surgeons due to perform an emergency operation in Chicago). In response to Mary’s query, Smith again responds ‘Yes, I know the plane is stopping in Chicago; my itinerary says so’. In this case, Mary would fail to attribute knowledge to Smith and her failure to do so seems correct. Instead, she would deny that Smith knows and this denial seems correct. Further, it would seem incorrect if Mary were to assert that the New York flight stops in Chicago, or to practically reason from this proposition. We may summarise the intuitions about the case as follows, letting ‘c’ stand for the proposition that the New York flight stops in Chicago: Attribution intuition: in LOW, Mary would attribute knowledge that c to Smith and this seems correct; in HIGH, Mary would fail to attribute knowledge that c to Smith and this failure seems correct. Denial intuition: in HIGH, Mary would deny that Smith knows that
73
c and this denial seems correct; in LOW, Mary would not deny that Smith knows that c and her failure to deny seems correct. Assertion intuition: in LOW, it seems correct for Mary to assert that c; in HIGH, it seems incorrect for Mary to assert that c. Practical reasoning intuition: in LOW, it seems correct for Mary to practically reason from c; in HIGH, it seems incorrect for Mary to practically reason from c.2 Contextualism provides a unified account of these data. The contextualist explains the attribution and denial intuitions by claiming that they reflect the truth-value of the relevant attributions and denials.3 On the contextualist view, in LOW, Mary’s attribution, ‘Smith knows that c’, is true and so she would attribute knowledge to Smith and would not deny that he knows; further, her attribution and failure to deny seem correct. Further, the contextualist holds that, in HIGH, Mary’s denial, ‘Smith does not know that c’, is true. As a result, she would fail to attribute knowledge to Smith and would instead deny that he knows; further, her failure to attribute and her denial seem correct. On the contextualist view, were Mary to self-attribute knowledge that c, the truth-value of her self-attributions would follow the same pattern as her attributions to Smith: in LOW it would be true for Mary to say ‘I know that c’ but, in HIGH, it would be false for her to say ‘I know that c’. As we will see, the contextualist may use the truth-value of Mary’s self-attributions to explain the assertion and practical reasoning intuitions by appeal to certain principles linking knowledge and appropriate assertion and practical reasoning. According to the ‘knowledge rule for assertion’, or KR for short, one should assert that p only if one knows that p. Slote (1979), Unger (1975), Williamson (2000) and DeRose (2002) all endorse the knowledge rule.4 2. For discussion of the attribution and denial intuitions, see Cohen (1988) and DeRose (1992); for the assertion intuition, see DeRose (2002) and Hawthorne (2004); for the practical reasoning intuition see Hawthorne (2004) and Williamson (forthcoming). Although most contextualists and invariantists accept that the cases are characterised by these intuitions, some disagree. For instance, Pritchard (2005) rejects the denial intuition. 3. For a concern about the extent to which the truth-value of the relevant attributions/ denials explains the intuitions, see Spicer (forthcoming). 4. DeRose (2002) also endorses the claim that that the knowledge rule is the only rule governing when a subject is well-enough positioned to assert that p, and uses this in an argument for contextualism. Brown (forthcoming a) argues against this claim and against the argument for contextualism based on it.
74
KR may be supported by noting that it explains the paradoxical nature of Moorean statements of the form ‘p but I don’t know that p’, and the fact that an assertion that p may be legitimately challenged by saying, ‘How do you know that p?’. The contextualist may explain the assertion intuition by using KR in a contextualist form: KR: one should assert that p only if ‘I know that p’ is true in one’s context. According to contextualism, in HIGH it would be false for Mary to say ‘I know that c’. Thus, by KR, it is inappropriate for her to assert that c. By contrast, in LOW, it would be true for Mary to say ‘I know that c’, and so she meets the necessary condition for appropriately asserting p specified in KR. (Such an explanation of the assertion intuition is offered in DeRose 2002.) The contextualist may use the truth-value of Mary’s self-attributions of knowledge to explain the practical reasoning intuition by exploiting a principle, KPR, which links practical reasoning and knowledge. KPR: one should use p as a premise in practical reasoning if and only if ‘I know that p’ is true in one’s context. KPR has been defended by Hawthorne and Williamson who argue that our intuitions about when a subject knows that p and when it’s appropriate for her to practically reason from p go together. For instance, before the lottery draw is announced it seems false for me to claim that I know my ticket is a loser. Further, it seems incorrect for me to rely on the proposition that my ticket is a loser in practical reasoning, say in the inference that since my ticket is a loser I should sell it for a penny. After the draw has been announced and my ticket is not the winning ticket, I do know that my ticket is a loser and it also seems appropriate for me to rely on that proposition in practical reasoning, say in the inference that since my ticket is a loser I should sell it for a penny (Hawthorne 2004, 30; see also Williamson forthcoming). The contextualist can exploit KPR to explain the practical reasoning intuition. In LOW, given that it would be true for Mary to say ‘I know that c’, by KPR it is appropriate for her to practically reason from c. In HIGH, since it would be false for Mary to say ‘I know that c’, by KPR it would
75
be inappropriate for her to practically reason from c. (Hawthorne 2004 argues that although contextualists can use KR and KPR to explain the assertion and practical reasoning intuitions, contextualists are committed to certain counterintuitive claims concerning practical reasoning and knowledge. I am setting aside consideration of such complex claims here and just examining how contextualists and invariantists deal with our simple intuitions about contextualist cases.5) We have seen that, on the contextualist account, the attribution, denial, assertion and practical reasoning intuitions are correct. The attribution and denial intuitions reflect the truth-values of the relevant knowledge attributions; the assertion and practical reasoning intuitions reflect whether, given KR and KPR, it is appropriate for Mary to assert c or practically reason from c. Invariantism denies that the truth conditions of knowledge attributions depend on the attributor’s context. So, it may seem that the invariantist must treat our intuitions about contextualist cases as misguided. For instance, suppose that the invariantist holds that, in HIGH, Smith knows that c. On that view, Mary’s denial that Smith knows is literally false. I will investigate whether invariantism must treat our intuitions about contextualist cases as misguided by examining a range of different non-sceptical invariantist views. In doing so, I will assume the principles KPR and KR. Some invariantists would prefer to offer a treatment of the contextualist intuitions while rejecting these principles (e.g., Pritchard and Rysiew reject KR6). However, these principles do receive intuitive support and are accepted by some invariantists, e.g., Williamson. Further, it is interesting to see if invariantists can offer a charitable treatment of the data while accepting the key assumptions of KR and KPR used in the contextualist treatment of the data.
5. Hawthorne argues that the contextualist is committed to the truth of such counterintuitive claims as 6) People often flat-out assert things that they do not know to be true but are not thereby subject to criticism; and 7) You should rely on propositions you don’t know to be true in your practical reasoning (2004, 88). 6. Pritchard accepts the Gricean principle that one’s assertions should be supported by adequate evidence (2005, 75, note 9); Rysiew endorses the claim that one should assert p only if one has a justified true belief that p (2001, 492).
76
2. Pure error theory A pure error theory explains the contextualist data by appeal to mistaken judgements about knowledge, and without bringing in any further factors. A number of different mechanisms have been suggested by which HIGH may lead us and Mary to falsely suppose that she and Smith lack knowledge that c. Perhaps the salience of error in HIGH leads us and Mary to overestimate the likelihood of error via the availability heuristic (Hawthorne 2004), and/or it may be that we use convenient but highly fallible rules of thumb in determining whether someone knows, rules which lead us astray in HIGH (Williamson forthcoming). Suppose that whether Mary is in LOW or HIGH, it is true for Mary to say ‘Smith knows that c’ and ‘I know that c’, however Mary’s and our judgements do not reflect these epistemic facts. Focus first on Mary’s judgements about her own and Smith’s epistemic positions. Suppose that when Mary is in LOW, she judges truly that she and Smith know that c; however, when Mary is in HIGH, she judges falsely that she and Smith do not know that c. On this supposition, she would attribute knowledge to herself and Smith in LOW, but not in HIGH; instead, in HIGH, she would deny that she or Smith knows that c. Now focus on our judgements about Mary’s and Smith’s epistemic positions. Suppose that our judgements follow the same pattern as Mary’s; suppose that when we consider LOW, we judge truly that Mary and Smith know that c, but that when we consider HIGH, we falsely judge that neither Mary nor Smith know that c. As a result, in LOW it seems correct for Mary to attribute knowledge to herself and Smith, but in HIGH, it seems incorrect for Mary to do so; instead, in HIGH, but not LOW, it seems correct for Mary to deny that either she or Smith know. The pure error theorist could exploit our judgements about whether Mary knows combined with KR and KPR to explain the data concerning assertion and practical reasoning. By KR, one should assert that p only if one knows that p. If, in HIGH, we judge that Mary does not know that c, then we will also judge that it is incorrect for Mary to assert that c. If, in LOW, we judge that Mary knows that c, then we will also judge that she meets KR’s necessary condition for appropriately asserting that c. By KPR, one should use p as a premise in practical reasoning if and only if one knows that p. If, in HIGH, we judge that Mary does not know that c, then we will also judge that it is incorrect for Mary to practically reason from c. If, in LOW, we judge that Mary knows that
77
c, then we will also judge that it is appropriate for Mary to practically reason from c. Notice that although the pure error theory explains the attribution, denial, assertion and practical reasoning intuitions, it treats them all as misguided. On the theory, these intuitions are based on Mary’s and our false judgements about Smith’s and Mary’s epistemic state. For instance, when considering HIGH, we judge that it is incorrect for Mary to attribute knowledge to Smith since we falsely judge that Smith does not know that c. Similarly, considering HIGH, we judge that it is incorrect for Mary to assert c, or practically reason from c, for we mistakenly judge that Mary does not know that c. According to a pure error theory, when Mary is in HIGH, it is correct for her to say ‘Smith knows that c’ in the sense that it is literally true, and incorrect for her to say ‘Smith does not know that c’ in the sense that it is literally false. Assuming that the theory accepts KPR, it seems committed to saying that when Mary is in HIGH, since she knows that c, it is appropriate in the sense of KPR for her to practically reason from c. Further, if it accepts KR, then if, in HIGH, Mary knows that c, then she meets KR’s necessary condition for appropriately asserting c. So while the pure error theory explains the attribution, denial, assertion and practical reasoning intuitions, it treats those intuitions as in fact incorrect.7 This motivates an examination of other invariantist accounts to see if they also treat our intuitions about contextualist cases as misguided. 3. Bach and belief removal A different way to explain the contextualist intuitions is offered by the belief-removal model. At the heart of this view is the claim that HIGH leads the attributor to lose belief and so knowledge in the relevant proposition. On one possible development of the view, HIGH undermines belief by making error salient (for a discussion, see Hawthorne 2004, 169–170). On Bach’s different development, HIGH provides the attributor with a practical reason to doubt the relevant proposition (Bach forthcoming). Here, I consider Bach’s development of the view. 7. Note that a view which combines an error theory with further considerations need not treat all of our intuitions about contextualist cases as misguided. See the discussion of Bach’s view and Williamson’s view, §3 and §5.
78
Bach rejects the contextualist claim that the truth conditions of a knowledge attribution depend on the attributor’s context. He defends a non-sceptical invariantist view on which, whether Mary is in LOW or HIGH, her attribution, ‘Smith knows that c’ is true. Bach argues that, in HIGH but not LOW, the high stakes give Mary reservations about the truth of c; she doubts that c. Given the high stakes, she has ‘good practical reason’ not to take Smith’s word for whether the plane stops in Chicago.8 Mary’s special practical interest ‘raises the threshold for believing c’, in other words she has practical reason to raise the level of epistemic justification required for belief; at these raised standards for belief, the itinerary evidence is not sufficient for belief. Bach argues that since Mary doubts that c it would be inappropriate for her to attribute knowledge of c to Smith. In general, an attribution of the form ‘Smith knows that p’ is true only if ‘p’ is true. So, if Mary were to assert ‘Smith knows that c’, Mary would commit herself to the truth of c. Since, in HIGH, she doubts that c, it would be inappropriate for her to attribute knowledge to Smith. In this way, Bach explains the attribution intuition. By itself, the claim that, when Mary is in HIGH, Mary doubts that c does not explain why Mary goes so far as to deny that Smith knows that c. Mary’s doubt about c may make inappropriate the attribution of knowledge to Smith but does not warrant the claim that Smith does not know that c. To explain the denial intuition, Bach further supposes that Mary’s doubt that c leads her to conclude that her evidence doesn’t epistemically justify her belief that c. His idea seems to be that Mary, finding herself doubting that c despite her evidence for c, would conclude that her evidence doesn’t epistemically justify the belief that c. Since Smith has the same evidence as Mary, Mary would also judge that Smith’s evidence does not justify his belief that c. Thus, Mary would go so far as to deny that Smith knows that c. For Bach, Mary’s judgement that neither her nor Smith’s evidence justifies the belief that c is mistaken.9 8. Discussing the high standards version of the bank case, Bach says that DeRose’s wife ‘has good practical reason not to take Keith’s word for whether the bank is open on Saturday’ (forthcoming, 25). 9. As Bach puts it, ‘in the high-standards versions of the airport and bank cases, a special practical interest gives the attributor reservations about the truth of the proposition in question and raises the bar for attributing knowledge to someone else. This practical element is epistemically irrelevant, even if it can affect whether one can make the knowledge attribution’ (forthcoming, 23).
79
In LOW and HIGH, Mary has the same evidence and has no positive reason to doubt that evidence. As an invariantist, Bach rejects the claim that the truth conditions of attributions of knowledge, or other epistemic states, such as justification, depend on the attributor’s context. So, he holds that Mary has the same level of justification for c in LOW and HIGH. Since the evidence is sufficient for knowledge in LOW, it is also sufficient for knowledge in HIGH. Bach’s idea, then, is not that in HIGH Mary has less justification for c than in LOW. Rather, he holds that in HIGH, but not LOW, Mary has a practical reason to doubt c which she confuses for the different claim that she lacks epistemic justification for believing that c. There are, then, two elements to Bach’s account. Like the pure error theorist, Bach holds that HIGH leads Mary to make a mistaken judgement about whether she and Smith know the relevant proposition. However, the account also contains an additional element, that HIGH leads Mary to lose belief in and so knowledge of the relevant proposition. This belief-removal element allows Bach, unlike the pure error theorist, to treat many of our intuitions about contextualist cases as correct. Although he holds that when Mary is in HIGH, it would be true for her to say ‘Smith knows that c’, it would nonetheless be inappropriate. For, in attributing knowledge to Smith Mary would be committing herself to something which, in HIGH, she doubts, namely c. Further, when Mary is in HIGH, Mary’s failure to attribute knowledge to herself and her denial that she knows reflect the truth-values of the attribution and denial. If HIGH leads Mary to doubt that c, then she fails to meet the belief requirement for knowledge that c. So, it would be literally false for Mary to self-attribute knowledge and it is literally true for Mary to deny that she knows. So, Bach treats as correct the intuitions that, in HIGH, it is inappropriate for Mary to attribute knowledge to herself or Smith, and appropriate for her to deny that she knows. Now consider the assertion and practical reasoning intuitions. If in HIGH, but not LOW, the stakes lead Mary to doubt that c and so fail to know that c, then Bach can exploit KPR and KR to argue that, in HIGH, it not only seems incorrect for Mary to practically reason from c and assert c, but it would in fact be incorrect for her to do so. By KPR and KR, it is appropriate for Mary to assert that c and practically reason from c only if she knows that c. Since, in HIGH, Mary fails to believe and so know that c, it would be inappropriate for her to assert that c or practically reason from c. Note that, in LOW, there is no similar
80
mechanism to dislodge Mary’s belief that c, so, in LOW, Mary knows that c. As a result, it follows from KPR that, in LOW, it is appropriate for Mary to practically reason from c. Further, she meets KR’s necessary condition for appropriately asserting c. It seems, then, that the belief-removal element in Bach’s account allows him, unlike the pure error theorist, to treat the attribution, assertion and practical reasoning intuitions as correct. Further, it allows him to treat as correct the first-person version of the denial intuition. Although Mary’s lack of belief explains why her first-person denial of knowledge is correct, it does not help explain her third-person denial of knowledge to Smith. That Mary lacks belief, and so knowledge, that c does not explain her asserting that Smith does not know that c. Instead, Bach explains the denial intuition by appeal to the second element in his account, his idea that Mary’s loss of belief leads her to falsely judge that she and Smith lack justification for and so knowledge that c. Note that Bach’s explanation of Mary’s denial that Smith knows treats this third-person denial as incorrect. On his view, in LOW and HIGH, Smith does have justified belief, and knowledge, that c. So, Mary’s denial that Smith knows is false. It seems, then, that Bach’s view provides a charitable treatment of all the intuitions about the airport case except the intuition that, in HIGH, it is correct for Mary to deny that Smith knows. Since, with the exception of the third-person version of the denial intuition, Bach’s account can treat all our intuitions about contextualist cases as correct and not misguided, it may seem a real rival to the contextualist’s account. However, unfortunately, it faces several objections. The account is built around the idea that, when Mary is in HIGH, Mary’s practical interest leads her to lose belief that, and so knowledge that, c. This directly explains why, in HIGH, it seems correct for Mary to fail to attribute knowledge that c, to assert c or use c in practical reasoning. Further, the loss of belief leads Mary to falsely believe that she and Smith do not have justification or knowledge that c. This false judgment explains why Mary denies that Smith knows, although this denial is literally false. However, it is not obvious that in all contextualist cases, the attributor loses the relevant belief in HIGH. For instance, consider a variant of the airport case in which in HIGH, it’s pretty important that Mary and John get to Chicago for a business meeting, but not a matter of life or death. Further, so far no error possibilities have been explicitly raised in their conversation. They ask whether anyone knows whether
81
the New York flight is stopping at Chicago. Smith replies, ‘Yes, I know the flight stops there; my itinerary says so’. The following conversation then takes place: John: Great. The plane’s going to Chicago, let’s take it. Mary: Smith’s itinerary may say that it’s going to Chicago, but Smith doesn’t know that. The itinerary might be mistaken; perhaps there’s been a late change of schedule. We better check. John: Well, it’s not very likely that the itinerary’s incorrect. We’ve certainly no reason to think there’s been a late change. How often have you come across an incorrect itinerary? Of course, the plane’s going to Chicago. Mary: OK, I know it’s unlikely that the itinerary’s wrong. I believe the plane’s going to Chicago too. But that’s not the point. Just imagine what would happen if we took the flight and it doesn’t stop at Chicago. We’d miss the meeting. We can’t rule out the possibility of an itinerary error. So we don’t know the plane will stop and we better check. Here both Mary and John make it clear that they believe that the New York flight stops in Chicago. Nevertheless Mary claims that they do not know that it does and her denial seems intuitively correct. But given that Mary and John do believe that the flight stops in Chicago, one cannot explain why it seems correct for Mary to deny that they know that the flight stops there by appeal to Bach’s account. This example highlights a general worry facing belief-removal accounts.10 Whether HIGH leads one to doubt the relevant proposition is a matter of psychological fact. So, suppose instead that as a matter of psychological fact, HIGH does not lead Mary to doubt that c. It still seems inappropriate for Mary in HIGH to attribute knowledge that c to herself or Smith, to assert that c or to practically reason from c. But, if HIGH does not lead Mary to lose belief, it’s not clear how the 10. Compare Hawthorne’s objection to a range of contextualist and invariantist accounts which explain the contextualist data by appeal to the salience of error in HIGH. Hawthorne argues that such accounts, whether in the form of a classic invariantist belief-removal model, a subject-sensitive invariantist model or a contextualist model cannot explain why it is wrong, in HIGH, for Mary to assert that p and practically reason from p if Mary continues to dogmatically believe that p because, as a matter of psychological fact, the counter-possibilities are not salient to her (2004, 173–4).
82
belief-removal account can treat these intuitions charitably. The beliefremoval model is non-sceptical and holds that, in LOW, Mary knows that c. Further, it is invariantist and holds that the attributor’s context does not affect the truth conditions of attributions of knowledge and other epistemic states. Given this invariantism, if in LOW Mary knows that c, in HIGH, if Mary believes that c, she knows that c. If in HIGH Mary continues to believe and so know that c then the belief-removal model has no account of why, in HIGH, it is inappropriate for her to attribute knowledge to herself or Smith. Further, if in HIGH, Mary knows that c then, assuming KPR, it is appropriate for her to practically reason from c, and she meets KR’s necessary condition for appropriately asserting that c. So if, in HIGH, Mary continues to believe that c, then the belief-removal model lacks an account of why it is wrong for Mary in HIGH to assert that c, practically reason from c, or attribute knowledge that c. Note that the belief-removal account cannot overcome this difficulty even if it adds some mechanism independent of belief-removal by which HIGH leads Mary to falsely judge that she and Smith do not know that c. Suppose that HIGH does not lead Mary to lose belief that c, but on grounds independent of belief-removal, it leads her to judge falsely that she and Smith do not know that c. This false judgment would explain why Mary fails to attribute knowledge that c, denies knowledge that c, and fails to assert c and practically reason from c. However, so long as Mary still believes that c then on this invariantist view, she knows it. But, then, it would be literally true for her to self-attribute knowledge that c, appropriate in the sense of KPR for her to use it in practical reasoning, and she meets KR’s condition for appropriately asserting that c. So, this modified view would still treat the attribution, assertion and practical reasoning intuitions as misguided. Bach might try to overcome the problem of cases in which HIGH does not lead the attributor to lose belief in a different way. He could argue that even if, in HIGH, Mary continues to believe that c, her doxastic practices are criticisable. A praiseworthy agent would doubt that c given the stakes and so such an agent ought not to self-attribute knowledge that c, assert that c, or practically reason from c. Thus, he may say, even if Mary continues to believe that c, she is blameworthy for selfattributing knowledge that c, asserting c and practically reasoning from c. However, the fact that a different, and perhaps more praiseworthy agent who does not believe that c ought not to self-attribute knowledge,
83
assert c or practically reason from c does not show that Mary ought not to do so. For there is a relevant difference between them: Mary believes and so knows that c. As a result, her self-attribution is true; by KPR, it is appropriate for her to practically reason from c; and, further she meets KR’s necessary condition for appropriately asserting that c. A different reply might start from the thought that, given the stakes, Mary herself ought not to believe that c and so ought not to do things whose appropriate performance requires that she believe that c. However, on Bach’s view, the relevant ought appears to be related solely to practical reasoning, and not epistemic justification. Since the beliefremoval model is invariantist, it rejects the claim that whether a subject can be truly ascribed epistemic properties depends on how high the stakes are. So this view should hold that if, in HIGH, Mary continues to believe that c, then her belief has the same level of epistemic justification as it does in LOW. Given that, in LOW, Mary’s belief that c counts as knowledge, so it does in HIGH. So, even if in HIGH Mary has a practical reason not to believe that c, she still has epistemic justification to believe that c and knows that c. So, in HIGH, if she continues to believe that c, she does know that c. As a result, were she to selfattribute knowledge, her attribution would be literally true. Further, since KR and KPR tie appropriate assertion and practical reasoning to Mary’s epistemic state, not considerations of practical reasoning, if in HIGH Mary knows that c it is appropriate in the sense of KPR for her to practically reason from c and she meets KR’s necessary condition for appropriately asserting that c. It seems, then, that whether the belief-removal account can treat the intuitions about contextualist cases charitably depends on whether HIGH leads the attributor to lose belief. Whether HIGH does undermine belief is a matter of brute contingent psychological fact. We have seen that it is possible to construct contextualist cases in which HIGH does not undermine belief. For such cases, the belief-removal account cannot offer a charitable explanation of the data. 4. Warranted assertibility manoeuvres A warranted assertibility manoeuvre, or WAM for short, makes no use of the idea that, when Mary is in HIGH, she doubts that c and so lacks knowledge that c. Rather, a WAM tries to explain the contextualist data
84
compatibly with accepting that, whether Mary is in HIGH or LOW, it would be true for her to say ‘Smith knows that c’ and ‘I know that c’. At the core of a WAM is the idea that the intuitions about contextualist cases can be explained by appeal to the truth-value of the propositions pragmatically conveyed by knowledge attributions, rather than the literal truth-value of those attributions. There are a number of different ways of filling out the basic strategy of a WAM. Here I will sketch Rysiew’s detailed example of such an account (see also Brown forthcoming b). According to the contextualist, the strength of epistemic position required for knowledge depends on context. By contrast, on Rysiew’s invariantist position, there is a context-invariant level of epistemic position required for knowledge although, in some contexts, an attribution of knowledge may pragmatically convey that the subject has a stronger epistemic position. More specifically, he argues that a subject knows that p iff she can rule out the relevant alternatives to p, i.e., those alternatives which ‘we (normal) humans take to be the likely counterpossibilities to what the subject is said to know’ (Rysiew 2001, 488). However, even an irrelevant alternative may be salient in a conversation, say because it has been mentioned (Rysiew 2001, 488). In such a context, Rysiew claims that, via Grice’s rule of relevance, an attribution of knowledge pragmatically conveys that the subject can rule out the salient alternative(s). For instance, Rysiew argues that in both the low and high contexts of the airport case, it’s literally true for Mary to say ‘Smith knows that c’ or ‘I know that c’. For she and Smith can rule out the relevant alternatives to c. The possibility of an itinerary error is not a relevant alternative in either LOW or HIGH. However, in HIGH, the possibility of itinerary error is salient in the conversation so an attribution of knowledge would pragmatically imply that the subject of the attribution can rule out this possibility. Since neither Mary nor Smith can do so, the attribution is conversationally inappropriate, although literally true. As a result, Mary fails to attribute knowledge to either herself or to Smith. Instead, she makes the false assertion that she and Smith do not know, pragmatically conveying the truth that neither she nor Smith can rule out the possibility of itinerary error. DeRose has raised doubts about the viability of such a WAM. Since I have discussed these arguments elsewhere, I merely summarise my conclusions here (see Brown forthcoming a and forthcoming b). DeRose (2002) presents an argument for contextualism and against an
85
invariantist WAM from ‘the knowledge account of assertion’, which combines KR with the claim that KR is the only rule governing how well-positioned a subject must be to warrantedly assert p. Brown (forthcoming a) argues against DeRose’s claim that KR is the only such rule; here, we are merely granting KR. Earlier (1998) DeRose argued that a plausible WAM should exploit general conversational rules and should arise from a conflict of intuitions. As we have seen, Rysiew’s WAM does exploit a general conversational principle, namely Grice’s rule of relevance. Further, it arises from a conflict of intuitions. For instance, in the airport case, if we focus on the fact that Smith’s itinerary says that the flight stops in Chicago, it seems plausible that Smith knows, but if we focus on the importance of the issue to Mary and John and the salience of error, then it seems implausible to claim that Smith knows (Brown forthcoming b). The most serious question DeRose raises about the viability of a WAM is his suggestion that a WAM cannot explain why, in HIGH, it seems correct to deny knowledge. According to the invariantist view suggested, the denial of knowledge is literally false but conveys a truth. However, DeRose argues that a literally false claim cannot seem correct even if it conveys a truth: ‘don’t we want to avoid falsehood both in what we implicate and (especially!) in what we actually say? So, it would seem that it would be unwarranted to assert a falsehood, even if doing so generates a true implicature’ (1998, 200). In response, Rysiew points out that speakers are often poor at distinguishing what’s literally said and what’s pragmatically conveyed. If speakers mistake what’s conveyed for what’s said then, contrary to DeRose, a falsehood which conveys a truth may seem correct. However, even if this is granted, it raises a further question. Under what circumstances are speakers likely to confuse the semantics and pragmatics of an utterance in such a way that they would take a literally false utterance which conveys a truth to be correct? Rysiew suggests that this is especially likely to happen when the utterance in question nearly universally conveys the relevant pragmatic claim (Rysiew 2000, 496). However, this explanation is problematic. The term ‘and’ nearly universally conveys ‘and then’ although that’s no part of its literal meaning. As a result, on Rysiew’s account, we should expect that speakers would confusedly read the conveyed temporal order into the literal meaning of a sentence containing ‘and’. If speakers did suffer this confusion, then in the situation in which Jack and Jill got married and then fell in love, we would expect speakers
86
to take the following utterance to be correct, even though it is literally false: ‘Jack and Jill did not fall in love and get married’. However, in the imagined circumstances, the utterance would not seem correct. Rather, it would seem appropriate to say ‘Jack and Jill fell in love and got married, but not in that order’.11 It seems, then, that important work remains to be done to defend the explanation a WAM offers of the denial intuition. Setting that issue aside for now, let us examine the extent to which a WAM can treat as correct the contextualist’s intuitions. We have seen that a WAM explains how, in HIGH, the attribution of knowledge seems incorrect even though it is literally true by arguing that it conveys a falsehood. This explanation can treat the attribution intuition as correct, as long as we assume that the attribution intuition is not only concerned with whether the attribution is literally true. If the content of the intuitive judgement that the attribution is incorrect amounts to the claim that the attribution is literally false then, on a WAM, that intuition would be incorrect since, on this view, the attribution is literally true. However, it seems plausible that our judgements about whether a given assertion is correct concern not only whether it is literally true, but also whether it is conversationally relevant, e.g., whether the assertion would be misleading. So, it seems more appropriate to treat the content of the intuitive judgement that the attribution is incorrect as the claim that it is either false or conversationally inappropriate. On this understanding, the attribution intuition is correct, for the attribution of knowledge is true but would be misleading since it conveys a falsehood. Notice that, if the content of our intuitive judgement that a knowledge attribution or denial is correct amounts to the claim that it is neither false nor conversationally inappropriate, then a WAM treats the denial intuition uncharitably.12 A WAM explains the denial intuition by claiming that, in HIGH, the denial of knowledge seems correct although it is literally false because it conveys a truth. So, on this account, the judgement that, in HIGH, the denial is correct is misguided: even though the denial pragmatically conveys a truth, it is still inappropriate in the sense that it is literally false. 11. See Halliday (forthcoming). 12. The only way to avoid this consequence would be to argue, implausibly, that the content of our intuitive judgement that, in HIGH, it is correct to deny knowledge merely concerns whether the denial conveys a truth, and is not at all concerned with whether the denial is literally true.
87
Now consider how a WAM might be extended to explain the assertion and practical reasoning intuitions. It may be suggested that, in HIGH, a simple assertion seems incorrect for the same reason as an attribution of knowledge: although literally true, it conveys a falsehood. For instance, in the high context of the airport case, the possibility of itinerary error is salient. As a result, Rysiew argued that an attribution of knowledge that the flight stops in Chicago would pragmatically convey the falsehood that the subject can rule out the possibility of itinerary error. Similarly, it may be suggested that the mere assertion that the flight stops in Chicago would pragmatically convey that the subject can rule out the possibility of itinerary error. If a WAM can be extended in this way, then it would treat the assertion intuition as correct: it would be incorrect for Mary to assert that c since although the assertion is literally true in HIGH it would convey a falsehood. The claim that, in HIGH, the simple assertion of c by Mary would convey that she can rule out the possibility of itinerary error seems plausible. However, someone who defends this claim about the pragmatic implications of simple assertions should relate this claim to a general theoretical account of pragmatic implications and how they are generated.13 However, Rysiew’s main defence of his account of the pragmatic implications of knowledge attributions does not carry across to the case of simple assertion. Rysiew defends his claim that ‘S knows that p’ conveys that S can rule out the salient alternatives by endorsing Malcolm’s view that ‘“I know” is used in contrast with someone’s (perhaps one’s own) previous, present or potential, disbelief, or doubt or insecure belief’ (Malcolm 1986, 212). Even if Malcolm’s view applies to knowledge claims14, it is not plausible for simple assertions. Simple assertions are not typically used in contrast to disbelief or doubt. For instance, I may simply say to my colleague, ‘It’s 12 o’clock. I must go to my lecture’, without any previous doubt or disbelief about what time it is.15 13. Pritchard defends this claim by appeal to an account of assertion which is a rival to KR, the claim that the assertion that p conveys that the speaker has adequate evidence for p (2005, 80). See also Brown (forthcoming b) for an account on which the assertion that p would be true but misleading in HIGH. 14. Halliday (forthcoming) casts doubt even on the view for the case of knowledge claims. 15. Rysiew further supports the account by appeal to an account of what’s conveyed by an assertion which is a rival to the knowledge rule assumed here. On KR, the assertion that
88
If this first strategy of explaining the assertion intuition fails, then a WAM may employ a different strategy. We have seen that Rysiew’s WAM suggests that Mary’s denial of knowledge seems appropriate because, although it’s literally false it conveys a truth, and we and Mary confuse the truth of what’s conveyed for the truth of what’s said. In other words, we and she take it that she does not know that c. By KR, it is appropriate for Mary to assert that c only if she knows that c. So, if we and she falsely judge that she does not know that c, we and she would also judge that it is inappropriate for her to assert that p.16 Similarly, by KPR, it is appropriate for Mary to use c in practical reasoning only if she knows that c. So, if we and she judge that she does not know that c, we and she would also judge that it is not appropriate for her to use c in practical reasoning. In this way, a WAM could be used to explain the intuitions concerning assertion and practical reasoning. Notice, though, that without further supplement, this explanation treats the assertion and practical reasoning intuitions as incorrect. On this view, in HIGH it is true though conversationally inappropriate for Mary to say ‘Smith knows that c’, or ‘I know that c’. Given that and assuming KPR, it would be correct for Mary to practically reason from c, at least in the sense of KPR. Further, Mary meets the necessary condition for warranted assertion of c formulated in KR, namely that she knows that c. Without challenging KR and KPR, a WAM could avoid treating the assertion and practical reasoning intuitions as incorrect only by bringing in further elements, for instance further rules concerning practical reasoning. (In the next section, we see that Williamson brings in such extra rules to provide a charitable treatment of the assertion and practical reasoning intuitions.) It seems, then, a WAM would treat many of the contextualist intup conveys that the speaker knows that p. On Rysiew’s rival view, a sincere assertion that p conveys that the speaker has a justified true belief that p. He argues that it follows from this account that ‘I know that p’ must convey more than that the speaker has a justified true belief that p. Instead, he suggests that it conveys that the speaker is confident or certain, that she can counter salient doubts about p (Rysiew 2001, 492). For otherwise, he claims, saying ‘I know that p’ would violate Grice’s maxim of manner; it would be a less perspicuous way of conveying just what the simple ‘p’ conveys. This second argument cannot be applied to argue that a simple assertion typically conveys that the speaker can counter salient doubts. 16. It is not obvious that Rysiew would accept this explanation of the assertion intuition since he holds that the sincere assertion that p conveys that the speaker has a justified true belief that p, which is a weaker condition than the condition that the speaker knows that p (Rysiew 2001, 492).
89
itions as misguided. Although a WAM can explain and treat as correct the attribution intuition, it would treat as incorrect the denial intuition and the practical reasoning intuition. Further, depending on which analysis it offers of the assertion intuition, it may treat that intuition as incorrect also. Given this result, it seems that we should search for a different invariantist response to contextualist cases, one which can treat more of our intuitions about such cases as correct. 5. Williamson Williamson (forthcoming) rejects the idea that we should explain the contextualist data as a result of the Gricean principle of relevance. Instead, he focuses on the conditions for assertion and practical reasoning. He argues that, in some cases, appropriate assertion and practical reasoning require second-order knowledge. However, he claims that such second-order knowledge is lacking in contextualist cases. Williamson concentrates on first-person contextualist cases; however we will see that his account can be extended to third-person cases. To set up the account, let us start by focussing on Mary and her self-attribution of knowledge. Williamson endorses a non-sceptical invariantist view on which, in both LOW and HIGH, Mary knows that c via Smith’s testimony. This claim combined with the fact that Williamson also endorses both KR and KPR might seem to make it difficult for Williamson to explain the assertion and practical reasoning intuitions. By KPR, that a subject knows that p is the condition for her to appropriately use p in practical reasoning. By KR, a subject is warranted in asserting p only if she knows that p. So, on the view that, in both the LOW and HIGH contexts, Mary knows that c, in the sense of KPR and KR, it is appropriate for Mary to assert c and use c in practical reasoning. Williamson hopes to explain why it nonetheless seems incorrect for Mary to assert c or use c in practical reasoning by defending additional principles according to which in high stakes situations, assertion and practical reasoning require second-order knowledge and arguing that Mary lacks such knowledge (forthcoming, 230–33). In support of the latter claim, Williamson argues that although Mary knows that c, she only just counts as knowing; she is an example of a borderline case of knowledge. According to Williamson’s anti-luminosity argument, for any borderline case of
90
a condition C, one does not know that C obtains, i.e. luminosity fails. So, although in both LOW and HIGH, Mary knows that c, she does not know that she knows that c (forthcoming, 232). Further, Williamson claims that, when the stakes are high, a subject is blameworthy for asserting p or practically reasoning from p if she does not know that she knows that p. His general idea is that when the stakes are high, the question of whether the subject has warrant for asserting p and practically reasoning from p becomes pressing. In high stakes cases, we are not merely interested in whether the subject meets the condition for appropriate assertion and practical reasoning, but also in whether she knows that she does, in other words whether she knows that she knows that p. For instance, he argues that when the stakes are high, a subject is blameworthy for using p in practical reasoning if she knows p but does not know that she knows that p. Similarly, when the stakes are high, a subject is blameworthy for asserting p if she knows p but does not know that she does. As a result, in HIGH but not LOW, it seems incorrect for Mary to assert c or practically reason from c. Mary’s lack of second-order knowledge that c also explains why, in HIGH, it seems correct for her to fail to attribute knowledge to herself. By KR, it is appropriate for Mary to assert ‘I know that c’ only if she knows that she knows that c. But, she lacks such second-order knowledge. The explanation of why, in HIGH, it seems correct for Mary to fail to attribute knowledge to herself can be extended to explain why, in HIGH, it seems correct for Mary to fail to attribute knowledge to Smith. Like Mary, Smith counts as a borderline case of knowing that c. The conclusion of Williamson’s anti-luminosity argument is that, for any borderline case of a condition C, one does not know that C obtains. Thus, Mary does not know that Smith knows that c. By KR, it would be appropriate for Mary to say ‘Smith knows that c’, only if Mary knows that Smith knows that c. But Mary lacks that second-order knowledge about Smith. Although Mary’s lack of second-order knowledge explains why, in HIGH, it seems correct for Mary to fail to attribute knowledge that c, assert c and practically reason from c, it does not explain why it seems correct for her to deny that she and Smith know that c. The inappropriateness of attributing knowledge to herself or to Smith does not show that it is appropriate to deny that either she or Smith know that c. Instead, Williamson explains the denial intuition by appeal to psychological bias. He argues that, when Mary is in HIGH, the salience of error leads Mary
91
to judge falsely that she and Smith do not know that c. As a result, Mary denies that she and Smith know that c. Further, when we consider Mary in HIGH, the salience of error leads us to judge falsely that neither she nor Smith knows that c. As a result, it seems correct for Mary to deny that she and Smith know that c (forthcoming, 234–5). As we have seen, a failure of second-order knowledge is at the core of Williamson’s account of the assertion, practical reasoning and attribution intuitions. There may be some contextualist cases in which the subject not only has first-order knowledge but also has second-order knowledge. Williamson suggests that for such cases, he may apply his explanation of the assertion, practical reasoning and attribution intuitions at a higher level of knowledge. For finite minds like ours, knowledge does not iterate indefinitely. A subject who knows that p and knows that she knows that p, will fail to know at some higher level. When the stakes are high enough, one may be blameworthy for asserting p, practically reasoning from p, or self-attributing knowledge that p if, although one knows that one knows that p, one lacks some higher level of knowledge (forthcoming, 234).17 Having seen how Williamson explains the contextualist data, let us assess how charitable his treatment is. For reasons which are by now familiar, Williamson’s treatment of the denial intuition is uncharitable. On his view, Mary goes so far as to deny that Smith knows since she falsely believes that Smith lacks knowledge. It is less clear whether Williamson treats the other intuitions charitably since his account effectively introduces several different notions of correctness for assertion and practical reasoning. The extent to which Williamson’s account is charitable turns out to depend on which of these notions of correctness we take our intuitions to be concerned with. Williamson endorses KR and KPR according to which knowledge that p is required for appropriate assertion and practical reasoning. Let us use the term ‘appropriate*’ for the notion of appropriateness at play in these two rules. As we have seen, Williamson introduces a further dimension of assessment of assertion and practical reasoning, whether the assertion or reasoning is blame free. He holds that, in high stakes cases, a subject is blameworthy for asserting that p or practically reasoning from p unless she knows that she knows that p. An assertion or piece of reasoning might be appropri17. For some concerns about Williamson’s approach to contextualist cases in which the attributor has second-order knowledge, see Brown (forthcoming c).
92
ate* while also blameworthy, if the subject knows that p, but does not know that she does. It turns out that the distinction between the notions of being appropriate* and being blame free makes no difference to whether Williamson offers a charitable treatment of the intuition that, in HIGH, it is incorrect for Mary to attribute knowledge. Were Mary to attribute knowledge, her attribution would be inappropriate* and blameworthy. According to KR, it is appropriate* for Mary to assert that she (or Smith) knows that c only if she knows that she (or Smith) knows that c. Since she lacks such second-order knowledge, the attribution would be inappropriate*. According to Williamson’s further principle, if the stakes are high, a subject is blameworthy for asserting p if she doesn’t know that she knows that p. So, given the stakes, Mary’s assertion that she (or Smith) knows that c would be blameworthy unless she knows that she knows that she (or Smith) knows that c. Since Mary lacks second-order knowledge that she (or Smith) knows that c, she lacks the third-order knowledge required for her attribution of knowledge to be blame free. By contrast, the distinction between being appropriate* and blame free is important for determining whether Williamson offers a charitable treatment of the assertion and practical reasoning intuitions. When Mary is in HIGH, it seems incorrect for her to assert that c or practically reason from c. Williamson attempts to explain this intuition by arguing that although, in HIGH, it would be appropriate* for Mary to assert that c or practically reason from c (since she knows that c), she would be blameworthy for doing so (since she does not know that she knows that c). Whether this explanation offers a charitable explanation of the intuitions depends on the content of those intuitions and, in particular, in what sense(s) we judge that it would be incorrect for Mary in HIGH to assert that c or practically reason from c. If the content of our judgement is that it is inappropriate* for Mary to assert that c or practically reason from c, then the judgement is false. For, on Williamson’s account, it is appropriate* for Mary to do these things. Alternatively, the content of our judgement may be that it is blameworthy for Mary to assert that c or practically reason from c. On this suggestion, our intuition is correct for, on Williamson’s account, Mary would be blameworthy for asserting that c, or practically reasoning from c. Instead of focussing on just one of these notions of correctness, our intuitive judgement may be concerned with both the notions of being appropriate* and blame free. On this view, the judgement that it would be incorrect for Mary
93
to assert that c, or practically reason from c, should be interpreted as the claim that it is either inappropriate* or blameworthy for her to do so. Notice that this last view need not be committed to the claim that subjects explicitly distinguish the two notions, but rather that their judgements of correctness factor in both notions, even if they do not clearly distinguish them. On this third interpretation, Williamson can treat the practical reasoning and assertion intuitions as correct: although it would be appropriate* for Mary to assert that c, or practically reason from c, it would be blameworthy for her to do so. A last suggestion might be that our intuitions of incorrectness are vague and do not clearly focus on any one notion of correctness, or any specific combination of notions. On this view, our intuition that, in HIGH, it is incorrect for Mary to assert or practically reason from c, would turn out to lack a determinate truth-value. We have seen that Williamson can offer a charitable treatment of the intuition that, in HIGH, it is incorrect for Mary to attribute knowledge that c, assert that c, or practically reason from c. He can do so by defending either of two claims, namely that the content of the intuitive judgement of incorrectness is concerned only with the notion of blameworthiness, or that it factors in both the notions of blameworthiness and appropriateness as suggested in the third interpretation above. Of course, in assessing how charitable Williamson’s account is it is not only important to see whether he can offer a charitable treatment of our intuitions concerning HIGH, but also whether he can offer such a treatment of our intuitions concerning LOW. It seems easy for him to offer a charitable treatment of the intuition that, in LOW, it is correct for Mary to assert that c or practically reason from c. In LOW, Mary’s assertion and practical reasoning are not only appropriate* (since she knows that c), but also blame free. For, it is only in high stakes cases that one is blameworthy for asserting that c or practically reasoning from c without knowing that one knows that c. So, under any of the first three interpretations offered above of our intuition that it is correct for Mary in LOW to assert that c, or practically reason from c, the intuition turns out to be correct.18 18. It may be argued that the relevant intuition is correct even on the last interpretation on which our intuitive judgements are vague. Whether those judgements are precisified to the notion of being appropriate*, being blame free, or, neither inappropriate nor blame free, they count as correct. If the judgement is correct on any precisification, then perhaps it can be treated as correct tout court.
94
The varying potential interpretations of the content of our intuitions about Mary do affect whether Williamson can offer a charitable treatment of the intuition that, in LOW, it seems correct for Mary to attribute knowledge. On Williamson’s view, in LOW and HIGH, Mary and Smith know that c, but Mary lacks second-order knowledge of this fact, she fails to know that she and Smith know that c. As a result, by KR, in LOW, it is inappropriate* for Mary to attribute knowledge to herself or to Smith, even though the attributions are literally true. This might initially seem to make it difficult for Williamson to offer a charitable treatment of the intuition that, in LOW, it is correct for Mary to attribute knowledge. Williamson may reply to this worry by pointing out that, on his view, it is not always a great crime if KR is broken. He appeals to this idea in defending KR against the objection that we do not always take great pains to verify a proposition before asserting it, e.g., in gossip or casual conversation (2000, 258–59). Williamson argues that this does not show that some assertions are not governed by KR, but rather that it is not always a serious matter if one breaks this rule. As he puts it, ‘when we are relaxed in applying the rule, we feel entitled to assert p whenever we are not confident that we do not know p’ (2000, 259). Similarly, in LOW where not much is at stake, we feel that it is acceptable for Mary to attribute knowledge even though she is breaking KR in doing so. In this way, Williamson can explain why it seems intuitively correct to attribute knowledge in LOW. Even if Williamson can explain why it seems correct for Mary to attribute knowledge in LOW, it is a further question whether he can treat this intuition as correct. We earlier distinguished three different ways of interpreting the content of our intuitive judgements about contextualist cases. On the first interpretation, the content of our judgement of intuitive correctness is that Mary’s attribution is appropriate*. On this interpretation, our intuitive judgement is incorrect for Mary’s attribution of knowledge breaks KR. On the second interpretation, the content of our judgement of intuitive correctness is that Mary’s attribution is blame free. On this second interpretation, our judgement is correct for, given the low stakes, she’s not blameworthy for breaking KR. On the third interpretation, the content of the relevant judgement is that Mary’s attribution is neither inappropriate* nor blameworthy. Notice that, on this third reading, the intuition is incorrect, for although Mary isn’t blameworthy for breaking KR, she is nonetheless breaking it. It turns
95
out, then, that there is an asymmetry between the intuitions concerning LOW and HIGH. Earlier, we saw that Williamson could provide a charitable treatment of the HIGH intuitions on either the interpretation that the content of our intuitive judgements of correctness solely concerns the notion of being blameworthy, or the interpretation that it concerns the notion of being neither blameworthy nor inappropriate. However, we have now seen that a charitable treatment of all the intuitions concerning LOW requires the interpretation on which the content of our intuitive judgements of correctness solely concerns the notion of being blameworthy.19 We have seen, then, that on a suitable interpretation of our intuitive judgements concerning contextualist cases, Williamson can offer a charitable treatment of all these judgements with the exception of the denial intuition, namely the intuition that, in HIGH, it is correct for the attributor to deny knowledge. Since Williamson holds that the judgement that, in HIGH, it is correct to deny knowledge is the result of a false judgement, he treats the denial intuition as misguided. It seems interesting that Williamson is not alone in treating the denial intuition as misguided. Rather, it is a common feature of all the invariantist accounts examined that they provide an uncharitable treatment of the denial intuition. This is not accidental. The accounts examined are all nonsceptical invariantist accounts. On a non-sceptical invariantist view, in LOW, the subject knows the relevant proposition. Given invariantism if, in HIGH, the subject still believes that proposition, her belief constitutes knowledge, and so the denial of knowledge is literally false. As a result, so long as it is part of the content of the HIGH judgement that the denial is correct, that the denial is not false, then the denial intuition is misguided. On this understanding of the denial intuition, the only way for a non-sceptical invariantist account to avoid treating the denial intuition as incorrect would be by arguing that HIGH leads the subject of that denial to lose the relevant belief, and so knowledge. 19. Notice that Williamson’s account can explain what is wrong with a dogmatist who, in HIGH, asserts that c, practically reasons from c, or attributes knowledge that c. Suppose that despite the high stakes in HIGH, Mary continues to believe that she and Smith know that c, and that she asserts that c, practically reasons from c, and attributes knowledge that c to herself and Smith. Given Williamson’s account of practical reasoning and assertion, since Mary does not know that she knows that c, it follows that, in HIGH, she would be blameworthy if she asserted that c or practically reasoned from c, and it would be inappropriate* and blameworthy if she attributed knowledge that c to herself and Smith.
96
Even, then, that would only enable us to treat as correct first-person denials of knowledge where the subject is the attributor, but not thirdperson denials in which the subject is distinct from the attributor. For even if the attributor’s being in HIGH undermines her, the attributor’s, belief, it needn’t undermine the belief of distinct subjects to whom she may attribute knowledge, subjects who themselves need not be in a high stakes context. It seems, then, that no non-sceptical invariantist account can treat as correct the denial intuition in both first and thirdperson versions. 6. Conclusion Contextualists support their account of the intuitions concerning contextualist cases by claiming that it provides a more charitable treatment of these intuitions than the invariantist account. Initially, it may seem difficult to see how the invariantist could offer a charitable treatment of the data since she denies that the truth conditions of knowledge attributions depend on the attributor’s context. However, of the variety of invariantist accounts considered here, it turns out that only the pure error theory offers an uncharitable treatment of all the contextualist intuitions. A WAM can charitably treat at least the attribution and perhaps the assertion intuitions, although it provides an uncharitable treatment of the denial and practical reasoning intuitions. Further, both the belief-removal account and Williamson’s luminosity account can provide a charitable treatment of all the contextualist intuitions with the exception of the denial intuition. So it is far from obvious that charity considerations clearly support contextualism over invariantism.20
20. Thanks to helpful comments from participants at the Amsterdam conference on Epistemological Contextualism.
97
REFERENCES Bach, K. (1994). ‘Conversational Impliciture’, Mind and Language 9, 124– 62. — (forthcoming). ‘The Emperor’s New ‘Knows’’, Contextualism in Philosophy: On Epistemology, Language and Truth, (eds.) Gerhard Preyer and Georg Peter, Oxford University Press, Oxford. (Page references from the manuscript.) Brown, J. (forthcoming a). ‘Adapt or Die: the Death of Invariantism?’, Philosophical Quarterly. — (forthcoming b). ‘Contextualism and Warranted Assertibility Manoeuvres’, Philosophical Studies. — (forthcoming c). ‘Williamson on Luminosity and Contextualism’, Philosophical Quarterly. Cohen, S. (1988). ‘How to be a Fallibilist’, Philosophical Perspectives 2, 91–123. — (1999). ‘Contextualism, Scepticism, and the Structure of Reasons’, Philosophical Perspectives 13, 57–89. — (2000). ‘Contextualism and Scepticism’, Philosophical Issues 10, 94– 107. — (forthcoming). ‘Knowledge, Speaker and Subject’, Philosophical Quarterly. DeRose, K. (1992). ‘Contextualism and Knowledge Attributions’, Philosophy and Phenomenological Research 52, 913–29. — (1995). ‘Solving the Sceptical Problem’, Philosophical Review 104, 1–52. Page references from Scepticism, (eds.) K. DeRose and T. Warfield (1999), 183–219, Oxford University Press, Oxford. — (1998). ‘Contextualism: an Explanation and Defense’, The Blackwell Guide to Epistemology, (eds.) J. Greco and E. Sosa, 187–206, Blackwell, Oxford. — (2002). ‘Assertion, Knowledge and Context’, Philosophical Review 111, 167–203. — (forthcoming). ‘The Ordinary Language Basis for Contextualism’, Philosophical Quarterly. Halliday, D. (forthcoming). ‘What Explains Our Intuitions About Knowledge Ascriptions?’, Proceedings of the Aristotelian Society. Hawthorne, J. (2004). Knowledge and Lotteries, Clarendon Press, Oxford. Malcolm, N. (1986). Wittgenstein: Nothing is Hidden, Blackwell Publishers, Cambridge, MA.
98
Pritchard, D. (2005). Epistemic Luck, Oxford University Press, Oxford. (Page references from the pre-publication manuscript.) Rysiew, P. (2001). ‘The Context-Sensitivity of Knowledge Attributions’, Noûs 35, 477–514. Schaffer, J. (2005). ‘Scepticism, Contextualism and Discrimination’, Philosophy and Phenomenological Research 69, 138–155. Slote, M. (1979). ‘Assertion and Belief’, Papers on Language and Logic, (ed.) J. Dancy, Keele University Library, Keele. Stanley, J. (forthcoming). ‘Context, Interest-Relativity and Knowledge’, Philosophical Studies. Spicer, F. (forthcoming). ‘Epistemic Intuitions and Epistemic Contextualism’, Philosophy and Phenomenological Research. Unger, P. (1975). Ignorance, Oxford University Press, Oxford. Williamson, T. (2000). Knowledge and Its Limits, Oxford University Press, Oxford. — (forthcoming). ‘Contextualism, Subject-Sensitivity Invariantism, and Knowledge of Knowledge’, Philosophical Quarterly. Wright, C. (forthcoming). ‘On the Contextualist Way with Scepticism’, Philosophical Quarterly.
99
This page intentionally left blank
Grazer Philosophische Studien 69 (2005), 101–125.
SOME WORRIES FOR WOULD-BE WAMMERS Adam LEITE Indiana University, Bloomington Summary DeRose appeals to ordinary English usage to support his contextualist semantics for “know”-attributions. A common objection holds that though the relevant assertions are both appropriate and seemingly true, their seeming truth arises merely from their appropriateness. This Warranted Assertability Maneuver (WAM) aims to provide a stand-alone objection by providing a reason not to take the ordinary language data at face-value. However, there is no plausible model or mechanism for the pragmatic phenomena WAMmers must postulate. Given what the WAM requires, it is doubtful it could work out in detail.
1. Introduction According to contextualists, the truth-conditions of sentences of the form “S knows that p”1 are determined in part by the intentions, purposes, interests and practical circumstances relevant in the conversational context. Two broad lines of evidence have been mustered for this proposal. First, it is claimed to cohere well with the knowledge account of assertion and to provide a unified solution to epistemological puzzles relating to skepticism and lottery-type cases. Second, it is claimed that ordinary English usage supports the contextualist semantics (DeRose 1992; 2002; forthcoming). My concern in this paper is with the latter claim; in particular, I will focus on one important example, Keith DeRose’s “Bank Cases” (DeRose 1992, 913; cf., Cohen 1999, 58–59). My question is whether a Warranted Assertability Maneuver (“WAM”) provides certain noncontextualists (moderate invariantists) with a satisfying response to this appeal to ordinary usage. 1. More precisely: of such sentences on occasions of utterance. I ignore this refinement in this discussion, as nothing turns on it.
DeRose’s Bank Cases consist of a pair of cases in which the speaker of a first-person “know”-attributing or denying sentence is (a) equally confident of the truth of a given proposition p, (b) is in the same epistemic position2 regarding p, and (c) regards himself as being in the same epistemic position. Yet because of a difference in the costs of error, the speaker asserts, “I know that p” in one case (the low stakes case) while asserting, “I don’t know that p” in the other (the high stakes case). In particular, in the low-stakes case, the speaker asserts “I know that the bank will be open on Saturday”, while in the high-stakes case — in which the speaker stands to lose his house if his paycheck is not deposited before Monday morning — the speaker asserts “I do not know that the bank will be open on Saturday.” That (a)–(c) are held constant in both cases is essential. If they were not, then the example would no longer provide any evidence for contextualism, since the speaker’s behavior would be adequately explained by the shift in non-conversational factors. DeRose muddies his example by having a conversational participant offer additional reasons for doubt in the high-stakes case. My interest is in a version of the example aimed at showing that purely practical factors make a crucial difference. Accordingly, I will focus on a “high stakes” case where there is no variation in the doubts raised. (Assume, again, that requirements (a)–(c) are met). Speaker A: Shall we stop at the bank to deposit our paychecks? B (Spouse): Nah, we can go tomorrow. A: But are they open on Saturdays? Lots of banks aren’t. B: I was there two weekends ago and checked their hours. They’re open until noon. [This much is identical in both cases. The highstakes case continues as follows.] A: Are you sure? If our paychecks aren’t deposited by Monday morning, our mortgage check will bounce, and we’ll lose our house! B: You’re right. I don’t know they’ll be open. The contextualist makes four claims about this case.
2. By “epistemic position” I mean one’s condition regarding the epistemic factors, whatever they are, that in addition to truth and belief are required for knowledge.
102
1. Speaker B’s assertion of “I don’t know” is conversationally appropriate, in that it is perfectly reasonable, in accordance with the rules of orderly discourse and linguistic usage, and not an odd, forced, or obviously nonliteral use of this sentence. For convenience, I’ll use the term “appropriate” to capture all of this.3 2. If B were to assert “I know that the bank will be open”, the assertion would be inappropriate. 3. What B said in asserting “I don’t know” seems to be true.4 4. If B were to assert “I know that the bank will be open”, what he thereby asserted would seem to be false. According to the contextualist, this case contrasts with the “low stakes” case on all four counts. The contextualist then argues that in the absence of reasons not to do so, we should take the appearances as a good guide: the truth-conditions for “I know that the bank will be open” and its denial differ in the two cases. The contextualist is guided here by the presumption that if competent and properly informed speakers take an assertion to be both appropriate and true, then all else equal, the assertion is true (DeRose forthcoming, sect. 1). Given claims (1)–(4), the bank cases thus prima facie favor the contextualist semantics; the burden of proof lies with the objector. It is crucial to the contextualist’s brief that in the high stakes case the assertion, “I don’t know …,” is appropriate and seemingly true. If the claim were rather that the assertion, “I know the bank will be open”, would be inappropriate and seemingly false, it would be less plausible that the truth-conditions shift. For if it were inappropriate and seemingly false to assert both “I know” and “I don’t know”, then it would be equally plausible that the truth-conditions had remained constant but contextual factors had made it inappropriate to say anything about whether one does or doesn’t know in the high-stakes case. Moderate Invariantism5 is a family of non-contextualist views which 3. The relevant notion of appropriateness goes beyond politeness and the like; the speaker is not just being politely concessive or conciliatory. 4. I will talk indifferently of sentences (on occasions of utterance) as being true and of what is expressed as being true. Nothing turns on this issue for my purposes here. When I say that a sentence is true, I mean that what is literally expressed by the assertive utterance of it on that occasion is a truth. For convenience, I will likewise use the term ‘assertion’ sometimes to refer to the asserted sentence, sometimes to what is literally expressed. 5. Moderate because non-skeptical; invariantist because non-contextualist.
103
accept the following two claims: 1. There is some “low-stakes” case in which the speaker knows and the knowledge claim is appropriate. I will assume that the low stakes bank case is such a case. 2. Neither the truth-values nor the truth-conditions of “know”-attributing sentences shift with anyone’s practical circumstances. Moderate Invariantists have two primary lines of response to the Bank Cases.6 The first denies that the high-stakes assertion of “I don’t know …” is really appropriate and seemingly true. It grants that in very similar cases — cases in which the speaker gets worried and loses his confidence or reappraises the strength of his evidence — it would be appropriate and seemingly true for the speaker to say this. But it claims that when specifications (a)–(c) are met, it would be odd for the speaker to declare “I don’t know …” (This isn’t to claim that it would then be appropriate to assert “I know …”; it might not be appropriate to assert either.) Ordinary cases in which it looks appropriate to assert “I don’t know p” as a result of high stakes will then get explained away as arising from shifts in the speaker’s degree of confidence or appraisal of his own epistemic position (Leite 2004, 346–7; Bach forthcoming). The second, more concessive, response grants that the assertion of “I don’t know …” is both appropriate and seemingly true. But it denies that these concessions reveal anything about the sentence’s truth-conditions, on the grounds that its seeming truth arises merely from the fact that it is appropriately asserted. This is the Warranted Assertability Maneuver (WAM).7 It aims to provide a stand-alone objection to the contextualist’s argument from ordinary language by providing a reason not to take the ordinary language data at face value. 6. One could reject claim (2) and still deny contextualism by claiming that it is the practical circumstances of the subject of the attribution or denial that matter, not those relevant in the speaker’s context. This route is taken by “Subject Sensitive Invariantism” (SSI) (Stanley forthcoming; Hawthorne (2004); cf. Kaplan (forthcoming), though Kaplan is perhaps best read as making no claim about truth-conditions at all). SSI strikes me as implausible, since it makes incorrect predictions about the appropriateness-conditions of third-person “know”ascribing sentences. My concern here, however, is the responses available to the Moderate Invariantist. 7. For important criticisms, see DeRose (1999; 2002). For a well-developed attempted WAM, see Rysiew (2001).
104
My aim in this paper is to evaluate the prospects for this latter response. As I will argue, there is a straightforward reason for pessimism: once one clearly recognizes what it requires, it is extremely hard to see how it could work out in detail.8 2. The WAMmer’s burdens The WAMmer’s project has two components, an explanatory component and a justificatory component. The explanatory component requires a story which, if true, would explain the contextualist’s data on the assumption that the moderate invariantist semantics is correct. The justificatory component provides an argument in favor of the explanatory story. I begin with the explanatory component. 2.1 The WAMmer’s explanatory tasks The WAMmer takes on two main explanatory burdens. The first is to explain why the assertion, “I don’t know …,” is appropriate in the high stakes case even though false. The second is to explain why it seems to ordinary speakers that in asserting this sentence, the speaker has said something true. To meet the first burden, it is not sufficient to explain why it would be inappropriate to assert, “I know …”; in general, to explain why it is inappropriate to assert Ø is not (yet) to explain why it is appropriate to assert not-Ø, since it might instead be inappropriate to assert either. To meet the second burden, it is not sufficient to meet the first, since ordinary speakers can take assertions to be appropriate without taking them to be true. The second burden imposes constraints on how the first is met; the account of why the assertion is appropriate must square with the account of why ordinary speakers are confused. To meet these burdens, the WAMmer first distinguishes what is literally expressed by a given assertive utterance of a given sentence and 8. My main concern is the use of a WAM as a stand-alone objection to the contextualist’s argument from ordinary language. If one accepts independent semantical arguments against the contextualist semantics, one might still attempt a WAM to explain the contextualist’s examples. My argument below suggests that this won’t work unless the WAM appeals to conventional implicatures. Moreover, a WAM isn’t the only option here, since there is also the possibility of the less concessive response sketched above. Anticontextualists shouldn’t assume that one way or another, a WAM must work out.
105
what is thereby pragmatically imparted, that is, what is communicated by the act of utterance even though it is not part of what is literally expressed nor entailed by what is literally expressed. To take a standard example, when asked to evaluate a student, I might reply, “His papers are well formatted and he shows up for class.” My utterance literally expresses that his papers are well formatted and that he shows up for class. What is communicated, however, is that I don’t think much of his philosophical abilities. There has been hot debate over how this distinction is to be drawn, what role pragmatic factors play in fixing what is literally expressed, and the like. For my purposes here we can prescind from these details.9 The WAMmer’s next step is to postulate two things: first, an account of what is literally expressed when one asserts “S knows/doesn’t know that p” in the circumstances of the contextualist’s example; second, an account of what is thereby pragmatically imparted. The details of what is literally expressed will depend upon which particular moderate invariantist view the WAMmer selects, but whichever it is, the assertion of “I don’t know …” in the high stakes case will literally express a falsehood. According to the postulated pragmatic theory, however, something conversationally appropriate will thereby be imparted, and the WAMmer postulates, moreover, that according to the norms determining conversational appropriateness, the appropriateness of what is thus communicated outweighs the prima facie inappropriateness arising from the falsehood of what is literally expressed. If true, these postulations would thus explain why it would be appropriate in the given conversational circumstances to assert “I don’t know …”. Finally, the WAMmer proposes to explain the seeming truth of this assertion by claiming that (we) ordinary speakers confuse what is pragmatically imparted by the utterance with what is literally expressed. If (we) ordinary speakers are confused in this way, and if we recognize what is thus pragmatically imparted to be true, then it is no wonder that the uttered sentence would seem true.10 The suggestion is not that this confusion explains why the assertion of “I don’t know …” is appropriate. Admittedly, an assertion can 9. The phrases “pragmatically impart” and “pragmatically communicate” suggest communicative success. But as I mean to use them, it is possible for something to be pragmatically imparted by a given utterance even if the hearer fails to cotton on. 10. The preceding three paragraphs broadly follow Rysiew (2001).
106
be appropriate or reasonable when, because of semantic confusion, the speaker mistakenly thinks it true and relevant. In such cases, the assertion is excusable and the speaker blameless. But this is not what the contextualist had in mind. He meant that someone who is fully aware of the relevant facts and clear about the semantics would judge that the speaker had acted in a fully correct way, in accordance with the relevant norms, not merely done the best he could given his misunderstanding. So the WAMmer rejects the contextualist’s data if he claims that the assertion in the high stakes case is appropriate simply because the speaker is confused about what is literally expressed by his utterance. For this reason, the WAMmer’s account must allow that the assertion would be appropriate even if the speaker were fully aware of its falsehood. The WAMmer does not discharge his explanatory obligations if he simply says, “The speaker thought he was literally expressing something true and relevant, and that’s why his assertion was appropriate.” This point can appear to spell trouble for the WAMmer. His first explanatory burden requires him to explain why the assertion is appropriate — in a way that does not amount to mere excusability — when made by an ordinary speaker. His response to the second explanatory burden claims that ordinary speakers are in error about what is literally asserted in these cases. But if ordinary speakers are confused in this way, then aren’t their assertions in these cases at best excusable — and so not appropriate in the way the contextualist requires? The WAMmer’s way with the second explanatory task thus appears incompatible with the data which motivate the first. To avoid this problem, the WAMmer must hold that an assertion can both arise from semantic confusion and be conversationally appropriate in the richer sense. This requires that semantic confusion be compatible with correctness according to the relevant norms. How could this be? Suppose that a speaker is confused about the truth-conditions of an uttered sentence (“X”) but intends to communicate something true and relevant (Ø). Suppose, moreover, that the norms governing appropriate usage involve or generate a general and standard link between sentences like “X” and contents like Ø, so that the default use of “X” in these circumstances would be to communicate what he intends to communicate. In such conditions, mistakenly thinking that he is literally asserting Ø is fully compatible with appropriately asserting “X” — so long as his intention in doing so is to communicate Ø. Hence, the WAMmer sees
107
the situation like this: an ordinary speaker’s assertion of “I don’t know …” in the high stakes case is not a misuse of the language, because the speaker abides by the relevant norms in uttering what she does. As long as her utterance aims at communication of what is thereby standardly imparted according to the relevant norms, her assertion is appropriate in the relevant sense; she’s just wrong about what she is literally expressing.11 The WAMmer therefore postulates a general, default rule governing the use of “I don’t know …” such that its utterance in the high stakes case would communicate what, according to the WAMmer, is thereby pragmatically communicated. The relation between the use of such sentences and the communicated contents will then be such that despite the falsity of what is literally expressed: a) The assertion (made by an ordinary speaker) is appropriate in a way that goes beyond excusability resulting from semantic confusion, b) The assertion would be appropriate even if the speaker, being clear about the semantics, recognized it to be false, and c) It is plausible that ordinary speakers could be confused about the truth-conditions of what is literally expressed. 2.2 The WAMmer’s justificatory task So far the WAMmer has offered a series of postulations which, if true, would explain the contextualist’s data while rejecting contextualism. Such postulations count for little, however; they can easily be made whenever one wants to reject a putative truth-condition while accepting the linguistic data offered in its defense. The WAMmer’s justificatory task, then, is to vindicate postulation of the relevant pragmatic phenomena. As we will see, this is where trouble arises. 2.2.1 Conventional implicature The simplest proposal is this. (1) As a matter of convention, asserting “I know/don’t know that p” entitles a hearer to make certain inferences which are not licensed by the asserted sentence’s semantic content. Thus an assertion of “I don’t know that p” would be appropriate, even if false, 11. Ordinary speakers will be unclear about their own minds as a result of this semantic confusion. For instance, they will misunderstand their communicative intentions, stating that in cases like the high stakes bank case, they intend to communicate that they do not know that the bank will be open. Worries about the WAMmer’s strategy might be raised on this score, but I leave them to one side.
108
if what is thereby conventionally imparted is true and conversationally relevant. (2) The conventions function as defaults, so ordinary speakers might well confuse what is imparted by such assertions with their literal semantic content. But a speaker who was fully clear about their semantics would nonetheless regard them as appropriate. This proposal puts the WAM in the category which Grice termed “conventional implicature”. It runs into a fundamental problem. Because the ordinary language data presumptively favor the contextualist’s account of the truth-conditions of the relevant sentences, it is question-begging for the WAMmer to claim simply, “This is how we use these expressions: ‘know’-attributions are conventionally used to convey Ø, even though Ø is not part of the semantic content of what is asserted nor entailed by it.” The WAMmer consequently needs a principled argument for regarding the content in question as only conventionally associated with the sentences in question, not part of nor entailed by their semantic content. Merely claiming that something is, as a matter of convention, pragmatically imparted does not make it so. It can be easy to miss this when considering particular proposals. Suppose the WAMmer suggests that “I know that p” is standardly used, as a matter of convention, to indicate (in part) that one is sure enough, or has good enough evidence, to reasonably act on the assumption that p given the risks. If that were so, then it would be reasonable to suggest that an assertion of “I don’t know that p” standardly indicates that one is not sure enough, or that one does not have good enough evidence, to act on p in one’s practical situation. Accordingly, what is literally expressed by the assertion in the high stakes bank case would be false, but what would be conveyed or indicated is that one isn’t sure enough, or lacks good enough evidence, to put off the deposit until the next day. The proposal is both initially plausible and meets the WAMmer’s explanatory needs. However, because of the way the contextualist takes the truth-conditions of “I know that p” to shift with the practical circumstances, the contextualist holds that if one’s utterance of “I know that p” is true, then (as a matter of entailment) one must be sure enough that p, or have good enough evidence that p, to warrant acting on p in one’s circumstances. Accordingly, the contextualist holds that what is literally expressed by an utterance of “I don’t know that p” is true if one fails to meet this condition. The WAMmer has therefore gotten nowhere unless he provides an independent argument against the contextualist’s account of the truth-conditions. This is a problem with
109
the general strategy, not with the particulars of this proposal.12 The main test to distinguish conventional implicature from truthconditional content is detachability: conventional implicatures are generated by the use of particular expressions, so an inference is based upon a conventional implicature rather than semantic entailment if it is not generated by other ways of formulating the same truth-conditional content. Unfortunately, this test doesn’t help the WAMmer. Since the dispute concerns precisely the truth-conditional content of the relevant expressions, there is no neutral way to test for detachability.13 2.2.2 Conversational implicature and related phenomena An independent argument for the WAMmer’s postulated pragmatic effect can be provided if the effect is a conversational implicature or related phenomenon (such as so-called explicature or conversational impliciture). Such phenomena are aspects of the speaker’s intended meaning, not semantically entailed by what’s literally expressed nor purely conventional, but recoverable from the fact that the speaker uttered what s/he did, given the assumption that s/he intended thereby to make an appropriate conversational contribution. They come in two basic varieties: particularized and generalized (Horn 2004, 5; Grice [1967] 1989, 37–8). A particularized conversational implicature or related effect is one which is induced only in a special, quite particular context, while a generalized implicature or related effect is one which is generated in the absence of a special, marked context. Generalized conversational implicatures thus function as defaults. This suits them especially well for the WAMmer’s purposes, since a default conversational implicature or related effect is most likely to generate just the sort of semantic confusion the WAMmer postulates.14
12. A similar problem arises for the proposal that in saying “I know …”, one is typically attempting to communicate that one’s hearers can take one’s word on the matter. 13. One could construct a WAM utilizing considerations relating to the speech-act conventionally performed (in certain cases) by declaring “I know”. In the interests of space, I leave this proposal aside. It works well in first-person cases, but does not generalize to cover contextualist-friendly second and third-person cases. 14. As Levinson comments, “those implicatures that are both derived from observing the maxims and are generalized have a special importance for linguistic theory. For it is these in particular which will be hard to distinguish from the semantic content of linguistic expressions, because such implicatures will be routinely associated with the relevant expressions in all ordinary contexts” (1984, 127; see also Levinson 2000).
110
Postulation of a conversational implicature or related effect is justified by means of a conception of conversation and communication as a rational, cooperative enterprise guided by principles such as Grice’s general “Conversational Principle”: (CP) Make your conversational contribution such as is required, at the stage at which it occurs, by the accepted purpose or direction of the talk exchange. (Grice [1967] 1989, 26) To vindicate the claim that an assertive utterance of “U” in given circumstances pragmatically imparts Ø, one shows how a fully rational hearer in those circumstances could “work out” — from (1) what is (according to the semantic theory) literally said, (2) the fact that the speaker uttered those words in those circumstances, and (3) the assumption that the speaker is abiding by the CP— that the speaker meant to communicate Ø. So, for instance, a fully rational conversational participant who heard me respond to a request to evaluate a student by saying, “His formatting is excellent and he shows up for class,” would reason: “There is no reason to suppose that he [the speaker] is not observing the principles governing conversation. He means to communicate something that is relevant to the request for an evaluation of student’s philosophical abilities. If he thought the student was good he would say so, since there is no reason for him to offer faint praise when he thinks more is warranted. He knew that I would recognize this. So he must mean to communicate that the student isn’t very good.” According to the Gricean, the claim that the student’s lack of ability is pragmatically imparted is vindicated precisely by the fact that this component of the communication can be worked out in this way. Two points should be emphasized. First, the claim is just that if a plausible line of reasoning of this sort can be produced by the theorist, that vindicates postulating the pragmatic phenomenon in question. The claim need not be that ordinary speakers always or ever go through any such reasoning explicitly, nor need it be that such reasoning models a psychologically real process by which actual hearers comprehend speakers’ intended meanings. Second, in cases where the semantics are in dispute, a theorist is not entitled to postulate the pragmatic effect in question unless s/he can provide reasoning (of the relevant sort) by which a rational hearer
111
could in principle arrive at what is putatively pragmatically conveyed.15 As Grice put it, (…) the final test for the presence of a conversational implicature [has] to be, so far as I [can] see, a derivation of it. One has to produce an account of how it could have arisen and why it is there. And I am very much opposed to any kind of sloppy use of this philosophical tool, in which one does not fulfill this condition. (1981, 187; in Neale 1992, 527, Neale’s insertions)
This claim is shared by all theorists working within the broadly Gricean framework. It has an important consequence for the anticontextualist WAMmer. The attempted WAM stands or falls with the WAMmer’s ability to provide the relevant reasoning in the high-stakes example. There are four crucial constraints on the WAMmer’s attempt to do so. First, the relevant speech situation is one in which (a) both the speaker and the hearer are neither confused about the semantics nor in error about any relevant facts, and (b) it is common knowledge between them that this is so. Here’s why. Suppose that the hearer is semantically confused in just the way that the WAMmer takes ordinary speakers to be. Then, the hearer in the high-stakes case would take the speaker to have literally expressed something true, conversationally relevant, and otherwise in accordance with the principles governing conversation – exactly what, according to the WAMmer, was pragmatically imparted. But then the relevant pragmatic reasoning wouldn’t be triggered at all, so the postulated pragmatic effect won’t have been vindicated. Likewise, if the hearer takes the speaker to be semantically confused in the way that the WAMmer holds ordinary English speakers to be, then the hearer would reason as follows: “The speaker is confused about the truth-conditions of the sentence he uttered; given what he takes them to be, he thinks he has literally expressed something true and relevant. So that’s why he uttered what he did: he took himself to be fully satisfying the conversational principles. So I don’t need to look for any further pragmatic communicative intent in order to explain his utterance.” On this line of reasoning, the speaker is interpreted as making an excusable false utterance. In neither case, then, does the hearer engage in reasoning 15. This point holds even if one has an independent semantical argument against the contextualist semantics, since appealing to conversational implicature isn’t the only option here.
112
that would justify the WAMmer’s pragmatic claim: he does not “work out” the pragmatic communication that the WAMmer hypothesizes. Second, the WAMmer’s account of the implicature must at least render it intelligible that reflective and thoughtful speakers could mistake what is pragmatically communicated for what is literally expressed. To see the issue here, consider an example. When a speaker evaluates a philosophy student by saying, “His papers are well-formatted and he shows up for class,” it is conversationally implicated that the student isn’t any good at philosophy. But no halfway intelligent speaker would think that the truth-conditions of the uttered sentence involve the student’s being no good at philosophy. In this case, then, it would be exceedingly implausible to postulate such confusion as part of an explanatory theory. This is an example of a particularized implicature; the pragmatic effect in this case is generated only via special-case reasoning arising in response to an apparent violation of the conversational maxims. An implicature generated in this way would not justify the WAMmer’s postulation of a default, standardized linkage between the relevant expressions and the relevant pragmatically-imparted content, and so it would not justify the postulation of widespread semantic confusion. Third, given the justificatory purpose of sketching the hearer’s reasoning, the reasoning cannot appeal to ancillary knowledge of what speakers usually use “know”-attributions or denials to impart. It might seem that it could. For instance, suppose that speakers generally use “I know …” to impart that they are very confident and use “I don’t know …” to impart that they are not very confident, though these things are not part of what is literally expressed by the utterance of these sentences. It would then seem that the hearer should be able to reason as follows: “He said that he didn’t know that the bank would be open. This isn’t true; he does know it. But speakers generally use sentences of the form ‘I don’t know …’ to impart that they are not very confident. So, given that he meant to communicate something true and relevant, and given that he knows that this is how such sentences are standardly used (and knows that I know this too) he must have meant to communicate that he wasn’t very confident that the bank would be open.” However, sketching the hearer’s reasoning is supposed to support or establish a claim about what utterances of the relevant sentences are generally used to communicate in addition to what they literally express, so to appeal to such considerations in the sketch of the hearer’s reasoning would assume the very thing that the WAMmer needs to establish. Consequently, the
113
hearer’s reasoning must proceed only from considerations about the truth-conditions of what was literally expressed, the words the speaker uttered (and the fact that the speaker uttered them), the circumstances, and the conversational principles; it cannot make use of ancillary information about what is standardly pragmatically imparted by use of the relevant expressions.16 The fourth constraint concerns the speaker’s attitude to the asserted sentence. By the description of the case, the speaker (1) believes that the bank will be open, (2) is in an evidential position which (according to the moderate invariantist) is adequate for knowledge and knows what his evidence is, and (3) knows what the truth-conditions of “I know the bank will be open” are. So given the moderate invariantist’s account of the truth-conditions, the speaker will take himself to be literally expressing a falsehood when he utters, “I don’t know the bank will be open tomorrow.” Moreover, the hearer will recognize this fact about the speaker, since the hearer is aware of (1)–(3). Consequently, the postulated reasoning must begin with the recognition that the speaker has intentionally asserted a sentence which he takes to literally express a falsehood. 2.2.3 The problem It is extremely hard to see how the hearer’s reasoning could go. In outline, it looks clear enough. The hearer should reason as follows. “What he just literally expressed — that he doesn’t know that the bank will be open tomorrow — is something he takes to be false (and he expects me to realize this). But since he is operating in accordance with the conversational principles, he meant to communicate something true and relevant (and he knows I will recognize this). What he must have meant to communicate is that …,” where the ellipsis is filled in with something conversationally appropriate, such as: “he isn’t absolutely certain that the bank will be open tomorrow” or “given our circum16. Several of Rysiew’s proposals fall afoul of this constraint. For instance, his account of the hearer’s reasoning from the maxim of Relation has the hearer appeal to ancillary information that “to say, ‘I don’t know…’ is to indicate that one isn’t sure” (2001, 491). The speaker in the high-stakes case might be sure enough for knowledge, so the “indication” in question will not be a semantic entailment, but rather a pragmatic effect. Cf. his attempt to explain the pragmatic effects of first-person ‘know’-ascriptions in terms of the epistemic commitments assumed in assertion (where he has the hearer appeal to the ancillary non-semantic claim that ‘I know’ adds an element of confidence or certainty to the bare assertion of p) (492).
114
stances his epistemic position with regard to whether the bank will be open tomorrow isn’t good enough (without further checking) to warrant putting off the deposit until tomorrow”. But how is the hearer to get from the thought, “He must have meant to communicate something else,” to a conclusion about what the speaker must have meant to communicate? As noted previously, he cannot rely upon ancillary knowledge of what such sentences are standardly used to communicate. He must rely only upon the circumstances, the conversational principles, his knowledge of the (moderate invariantist) semantics of the uttered sentence, the fact that the speaker uttered the sentence while believing it to be false, and the assumptions that the speaker isn’t confused about what he has literally expressed or mistaken about the relevant facts and aims to abide by the conversational principles. These are meager resources – especially given that what the speaker has done is very strange. It isn’t obvious that anything would be pragmatically conveyed in these circumstances. An analogy may help bring out the difficulty. Imagine that a speaker is drinking something. The question of whether it is hot is conversationally salient. The speaker believes it to be at least warm, and the hearer recognizes that he believes this. The speaker announces, “It’s not warm. I’d better heat it up.” On the assumption that he isn’t making a slip, semantically confused or in error about the temperature of his drink, the natural conclusion is that he is not acting in accordance with the conversational principles at all. Whatever he’s doing, it’s completely bizarre. But that’s exactly what the situation is like, as the WAMmer is conceiving it. The contextualist takes there to be a scale of epistemic standards, some more stringent than the one which must be met for the truth of a “know”-attributing sentence according to the moderate invariantist semantics. The WAMmer doesn’t dispute that. The contextualist holds that an epistemic standard more stringent than that picked out by “know” on the moderate invariantist semantics is conversationally salient in the high stakes bank case. The WAMmer agrees. Call the epistemic position which meets that more stringent standard “absolute certainty” (just to give it a label). The conversationally salient question is whether the speaker has attained the epistemic position absolute certainty — a position better than that required for knowledge. The speaker believes that he knows, and the hearer recognizes that he believes this. The speaker announces, “I don’t know. I’d better check.” On the assumptions that he isn’t making a slip, semantically confused or in error about his epistemic position and that the moderate invari-
115
antist semantics for “know” are correct, the natural conclusion is that he is not acting in accordance with the conversational principles at all. Whatever he is doing, it is completely bizarre. In the analogy, the hearer can only take the speaker to intend to communicate that his drink is not hot. There’s no other way to make any sense of his behavior. Likewise, in the bank case we can’t make any sense of the speaker’s behavior unless we take him to mean to communicate that he isn’t absolutely certain. In both cases, however, there is something unintelligible in the speaker’s behavior; neither speaker is communicating in a fully rational way. This fact importantly limits the explanatory potential of the hearer’s ability to divine the speaker’s communicative intentions. In the drink case, the hearer’s discovery of the speaker’s communicative intentions does not ground any sort of general, default pragmatic phenomenon. Nor does it make it at all intelligible that competent speakers could generally be confused about the truth-conditions of “It’s not warm” — not even when it is uttered in these sorts of conditions. The same conclusion appears to follow in the case of “knows”, if the moderate invariantist semantics is correct. There seems to be no significant difference between the cases.17 3. Some proposals I now want to examine the most promising possible responses to this problem. So far as I can see, none fully meets the WAMmer’s needs. 3.1 Classical Gricean proposals Grice’s Conversational Principle encapsulates a set of general maxims, divided into four groups, governing rational conversational interchange (Grice [1967] 1989, 26–7): Quality: Try to make your contribution one that is true. 1. Do not say what you believe to be false. 2. Do not say that for which you lack evidence.
17. This consideration presents difficulties for Rysiew’s appeal to the maxim of Relation (2001, 491–2).
116
Quantity: 1. Make your contribution as informative as is required (for the current purposes of the exchange). 2. Do not make your contribution more informative than is required. Relation: Be relevant. Manner: Be perspicuous. 1. Avoid obscurity of expression. 2. Avoid ambiguity. 3. Be brief. 4. Be orderly. I will focus upon possible implicatures arising from the maxims of Quantity, Relation, and Quality, in that order.18 (I will ignore the maxims of manner, as they seem irrelevant in this context.) 3.1.1 Quantity The maxims of Quantity generate important scalar effects, as when a speaker reports that someone is “nice-looking”, thereby communicating that he isn’t drop-dead gorgeous. Since both the contextualist and the WAMmer postulate a scale of epistemic positions, it is natural to expect scalar effects in the high-stakes bank case. Unlike standard scalar cases, however, this case would involve the speaker’s falsely asserting that a lower point on the scale is not reached in order to communicate that a higher point is not reached. There is no accepted Quantity-based model for scalar effects of this sort. One can say, “He’s not cute [stress]” to convey that he is gorgeous. But one can’t use “He’s not cute” (where this is taken by both parties to be false) to convey that he is not gorgeous. Assertion of a literal falsehood can generate Quantity-based pragmatic effects in certain non-scalar cases. For instance, the assertion “He lost a finger in the accident” ordinarily communicates that he did not lose a thumb. This can be explained via the maxims of Quantity and the lexicalization of “thumb” (Horn 2004, 16). This pragmatic effect generates phenomena somewhat analogous to what the WAMmer needs. For instance, the assertion “I didn’t lose a finger [stress] in the accident”, made by someone who is known to have lost a digit, communicates 18. Grice’s list has been challenged, but I am aware of no proposed revisions which substantially affect the argument of this section.
117
that he or she lost a thumb. (Cf., “rectangle” and “square”.) Here, what is literally expressed is false, but the assertion is appropriate because what is thereby communicated is true and relevant. These cases provide a poor model for the high-stakes bank example. In these cases, a term X refers to a general class, only one member of which has a lexicalized term Y referring to it. By saying (in appropriate circumstances) not-X, one implicates that Y. One thus specifies, picking out the member of the class referred to by the more specific lexicalized term. But in the high-stakes bank case, the speaker does not (falsely) negate a more inclusive term in order to implicitly apply a more specific lexicalized term. I am aware of no other ways in which asserting a literal falsehood can generate anything like the relevant sort of pragmatic effects via the maxims of Quantity. 3.1.2 Relation The conversational issue in the high stakes bank case is whether the deposit can reasonably be put off until the following day. Considerations of relevance suggest that whatever the speaker intends to communicate by uttering, “I don’t know …”, it is something which will help resolve this issue. So it would seem that the ideal hearer should be able to reconstruct the intended communication as follows. “He intends to make a relevant contribution, and he knows that I recognize his intention to do so. He has said something about his epistemic position regarding whether the bank will be open tomorrow. But he takes what he has literally asserted to be false, and he knows that I will recognize this. So he must mean to communicate something else about his epistemic position. It would be relevant to our situation if he weren’t in a good enough epistemic position to be comfortable putting off the deposit until tomorrow. So that must be what he intended to convey.” There are several significant problems with this proposal. First, by themselves considerations of relevance do not enable the hearer to determine what the speaker meant to communicate. If all the speaker said was, “I don’t know …” it would be equally possible to interpret him as intending to communicate that he does not merely know, but rather is in a better position than the one required for knowledge — and hence that it would be best to put off the deposit. The mere utterance of “I don’t know …” in the stipulated circumstances is pragmatically underdeterminate, so far as the maxim of Relation goes.
118
Second, further contextual factors (the speaker’s other linguistic behavior — tone, stress, and other utterances) enable the hearer to narrow things down. But this just highlights that the utterance of “I don’t know” was completely unnecessary. The speaker could just as well have uttered, “Fitzmallin guzampt fishcake. I guess I’d better check.” In fact, the speaker could simply have said “We’d better check,” and the hearer, reasoning that the speaker meant to convey something relevant, could have figured out what the speaker meant to convey. Consequently, both the semantic content of the utterance and the fact that it was uttered are completely idle in the account. The point is not that they are redundant; other contextual factors are doing all the explanatory work. So vindicating the postulated pragmatic effect via the maxim of Relation doesn’t render the speaker’s utterance linguistically appropriate in this case. The hearer can figure out what the speaker meant to communicate, but a residual mystery remains: why did he choose those words to do so? It looks like a case of linguistic misuse. Third, this account does not provide any reason to expect a generalized, default implicature. For this reason, it doesn’t make it intelligible how ordinary speakers could be confused about the truth-conditions of utterances of the relevance sentences. Nor does it provide any reason to postulate a further pragmatic element of the appropriate sort in ordinary speakers’ usage. 3.1.3 Quality Here’s a more promising proposal. For any position on the scale of epistemic positions, if you don’t possess it, then you also don’t possess any stronger position. (This is a logical relation, a matter of entailment.) According to the moderate invariantist, some position on this scale lower than the highest possible is adequate for knowledge. Suppose, however, that the position required for being warranted in the high-stakes case in putting off the deposit until Saturday is stronger than the minimum position required for knowledge. Then not knowing (because one’s epistemic position is inadequate for knowledge) will entail that one isn’t in the stronger position either. Accordingly, “I don’t know that the bank will be open” will logically entail (or close enough19) something both true and conversationally relevant: that one 19. Calling this an entailment is convenient shorthand, though a bit of a cheat. Given the WAMmer’s semantics, “I don’t know that p” entails: “P is false or I don’t believe that p or
119
is not in the epistemic position required for being warranted in putting off the deposit. The proposal, then, is that the hearer exploits this fact in order to work out the implicature. The reasoning would go roughly like this. “He literally expressed that he does not know the bank’ll be open. But he should take this to be false, given that he believes that the bank will be open and believes that his epistemic position is adequate for knowledge. But given everything else he believes, he would take I don’t know that the bank will be open to entail I’m not in the minimum epistemic position required for being warranted in putting off the deposit until Saturday, and he would expect me to recognize this. Call that entailed proposition “~Ø”. ~Ø is both true and directly relevant. There is no natural and quick way for him to literally express it. So ~Ø must be what he meant to communicate, via my recognition that (1) he would take what he literally expressed to be false, (2) ~Ø is directly entailed by what he literally expressed, (3) it is true and relevant, and (4) he has no straightforward and reasonably quick way to express it.” On this proposal, the implicature is generated by the interplay of the maxim of Quality, a principle of least effort (given the readily available linguistic resources), and the scalar structure of epistemic positions.20 The proposal is that the assertion, “I don’t know …,” in the high-stakes case is a kind of over-statement: the speaker is saying something logically stronger (and false) in order to easily communicate something closely related and logically weaker. This proposal meets the WAMmer’s explanatory needs. If general conversational principles dictate that absent a straightforward way to formulate what one intends to communicate, one should make the most easily formulated (even if false) assertion that obviously entails what one intends to communicate, then the relevant implicatures will be generated by utterances of the relevant sentences across a wide range of standard conditions. In particular, assertion of “I don’t know that p”, my epistemic position regarding p isn’t adequate for knowledge.” By uttering “I don’t know that p”, the speaker does not communicate that he does not believe that p is true; if he had wanted to do that, he should have said either “Not-p” or “I don’t believe that p” (or something to that effect). So given that the speaker has already indicated that he does believe that p is true (by asserting p), the hearer can conclude that the speaker means to focus attention more particularly on the proposition that his epistemic position is not adequate for knowledge. This proposition entails that his epistemic position is not any better than the minimum position required for knowledge. 20. The principle of least effort figures centrally in Horn’s recent model (2004, 12–17)).
120
whether true or false, will communicate that the speaker’s epistemic position isn’t good enough to warrant acting on p whenever the required epistemic position is stronger than the minimum position required for knowledge. It would be no wonder, then, if ordinary speakers are semantically confused. Still, this proposal faces three significant objections. First, it is not obvious that implicatures are generated in the proposed way; there are no paradigmatic cases of this sort nor any obvious models upon which to draw. What is needed is an ordinary example in which a speaker who is not semantically confused knowingly asserts a falsehood in order to communicate something entailed by it, something which there is no obvious and straightforward way to express. A possible model is provided by “loose” constructions, such as “He has a square face.” It could be proposed that an utterance of this sentence literally expresses something false but communicates something true and obviously entailed which we cannot easily formulate (“His face is more like a square than like any other easily identified shape”, perhaps). However, it is hardly obvious that what is literally expressed here is something false, nor that if it is, this proposal provides the best treatment of loose constructions. I defer here to linguists and philosophers of language. Second, it is doubtful that the proposal really applies to the case at hand. The proposal requires that there be no equally simple and straightforward way to express what the speaker wishes to communicate. But that is not true in the high stakes bank case. The speaker could simply say, “I’m not certain that the bank will be open,” or, perhaps, “It’s not certain that the bank will be open.” So it is not at all obvious that the WAMmer can really justify postulating the necessary pragmatic phenomena.21 Finally, any proposal which exploits scalar relations between epistemic positions faces a further explanatory challenge. Quantity-based scalar implicatures are widespread. Assuming scalar relations between epistemic positions, one would expect that asserting “I know that p” could communicate — via the Maxim of Quantity— that one has not 21. This worry is especially significant in light of Horn’s conjecture (1972) that because of a redundancy constraint, if use of a given lexical item generates a certain generalized conversational implicature, then there will be no lexical item that directly encodes that implicature. (Horn, On the Semantic Properties of the Logical Operators in English. Mimeo. Indiana University Linguistics Club. Cited in Levinson 1984).
121
attained any stronger epistemic position. So on the WAMmer’s semantics, it should be appropriate to utter “I know that the bank will be open” in the high stakes case; one would thereby assert something true and communicate something relevant. The WAMmer needs to explain why things don’t work this way instead. 3.2 Conversational impliciture Consider this example. A parent heartlessly comments to a young child screaming from a scraped knee: (1) You aren’t going to die. The uttered sentence is false but appropriate, since it successfully communicates that the child will not die from the scrape. This as an example of what Kent Bach (1994) calls “conversational impliciture.” In conversational impliciture, “the speaker is not being fully explicit. Rather, he intends the hearer to read something into the utterance, to regard it as if it contained certain conceptual material that is not in fact there” (126). The conversational impliciture is what the speaker literally meant, and is reconstructible through supplementation of the literal linguistic content of the utterance. It can seem that (1) provides the model the WAMmer needs. However, there are two significant reasons why it doesn’t. First, conversational implicitures of the relevant type are detachable (Bach 1994, 137). However, if the WAM exploits entailments arising from scalar relations between epistemic positions, then it must be modeled on a non-detachable pragmatic effect, since these entailment relations hold regardless of the particular words used. Second, implicitures of the relevant type involve “what might be called ‘lexical’ strengthening, in that what is being communicated could have been made fully explicit by the insertion of additional lexical material” (Bach 1994, 134); the hearer supplies “missing portions of what is otherwise being expressed explicitly” (154). But in the high stakes bank case no additional lexical material would convert the utterance into an explicit expression of only the claim that the speaker’s epistemic position isn’t good enough to warrant putting off the deposit. Moreover, in cases of conversational impliciture ordinary speakers easily recognize that the uttered sentence is “strictly speaking” false.
122
According to the contextualist, this is not so in the high stakes bank case. The WAMmer agrees. Accordingly, the model leaves it unclear how speakers could be confused in the relevant way about the truthconditions of “know”-attributions. 3.3 Back to ordinary speakers At this point, the WAMmer might suggest something like this.22 “Perhaps the trip through the Gricean apparatus was a mistake: we don’t need to explain how an ideal hearer, operating on the assumptions that the speaker is obeying the Conversational Principle and not semantically confused, could “work out” what the speaker intends to communicate; we just need to understand how actual hearers do it. And we don’t need any special explanation for why ordinary speakers are semantically confused; their grasp of the semantics operates tacitly in the production of their linguistic behavior, so there is no particular reason to think they’d get it right without a lot of work. Here, then, in outline, is how ordinary hearers do it: guided by their tacit understanding of the semantics, they look around for the most relevant proposition “in the ballpark”. In the high stakes bank case, the “ballpark” is that the speaker meant to communicate something relevant about his epistemic position. The most relevant thing he might plausibly have meant (given his other behavior) is that his epistemic position isn’t adequate to warrant putting off the deposit. So that’s what the hearer concludes he meant. And since the hearer isn’t clear about the semantics at the conscious level, the hearer will naturally be confused about whether this is what is explicitly expressed by his utterance.” This is a last ditch attempt. The trouble with it is simply this: it is completely neutral on the semantics. The contextualist could tell the same story. 4. Conclusion The WAMmer’s project has run into a fundamental difficulty. None of the proposals we’ve considered provide a fully satisfactory fix. So I suggest that we start over. Is the speaker’s assertion really appropriate in the high stakes case? Consider the conversation again, keeping in 22. Inspired, perhaps, by Relevance theorists.
123
mind that the second speaker is — and remains — as confident that the bank will be open as one ordinarily can be about such things and that he was there just two weeks ago and saw no sign that they would be changing their hours (nor does he have any special reason to doubt his memory, etc.). It helps to read it aloud. Speaker A: Shall we stop at the bank to deposit our paychecks? B (Spouse): Nah, we can go tomorrow. A: But are they open on Saturdays? Lots of banks aren’t. B: I was there two weekends ago and checked their hours. They’re open until noon. A: Are you sure? If our paychecks aren’t deposited by Monday morning, our mortgage check will bounce, and we’ll lose our house! B: You’re right. I don’t know they’ll be open. Is it really linguistically appropriate for Speaker B to say “I don’t know they’ll be open,” given that neither his appraisal of his evidence nor his confidence shifts and he isn’t just being conciliatory? To my ear, it isn’t; it is decidedly odd. This isn’t to say that it would be appropriate for him to claim outright that he does know, either. More naturally, he’ll say, “You are right. Let’s not risk it”, or something to that effect. I am therefore inclined to think that moderate invariantists should deny — not explain — the contextualist’s putative ordinary language data.*
REFERENCES Bach, K. (1994). ‘Conversational Impliciture’, Mind and Language 9, 124– 162. — (forthcoming). ‘The Emperor’s New ‘Knows’’, Contextualism in Philosophy, (ed.) G. Preyer and G. Peter, Oxford University Press, Oxford. Cohen, S. (1999). ‘Contextualism, Skepticism, and the Structure of Reasons’, Philosophical Perspectives 13, 58–89. DeRose, K. (1992). ‘Contextualism and Knowledge Attributions,’ Philosophy and Phenomenological Research 52, 913–929.
* I would like to thank Kent Bach, Steven Gross, Ram Neta, and Patrick Rysiew for comments and conversation.
124
— (1999). ‘Contextualism: An Explanation and Defense’, The Blackwell Guide to Epistemology, (ed.) E. Sosa and J. Greco, 185–203, Blackwell Publishing, Oxford. — (2002). ‘Assertion, Knowledge, and Context’, The Philosophical Review 114, 167–203. — (forthcoming). ‘The Ordinary Language Basis for Contextualism and the New Invariantism’, The Philosophical Quarterly. Grice, H.P. (1981). ‘Presupposition and Conversational Implicature’, Radical Pragmatics, (ed.) P. Cole, 183–198, Academic Press, New York. — (1989) [1967]. ‘Logic and Conversation’, Studies in the Way of Words, Harvard University Press, Cambridge, MA. Hawthorne, J. (2004). Knowledge and Lotteries, Clarendon Press, Oxford. Horn, L. R. (2004). ‘Implicature’, The Handbook of Pragmatics, (ed.) L. Horn and G. Ward, 3–28, Blackwell Publishing, Oxford. Kaplan, M. (forthcoming). ‘Deciding What You Know’. Leite, A. (2004). ‘Is Fallibility an Epistemological Shortcoming?’, The Philosophical Quarterly 54, 232–51. Levinson, S. C. (1984). Pragmatics, Cambridge University Press, New York, NY. — (2000). Presumptive Meanings: The Theory of Generalized Conversational Implicature, MIT Press, Cambridge, MA. Neale, S. (1992). ‘Paul Grice and the Philosophy of Language’, Linguistics and Philosophy 15, 509–559. Rysiew, P. (2001). ‘The Context-Sensitivity of Knowledge Attributions,’ Noûs 35, 477–514. Stanley, J. (forthcoming). Knowledge and Interests, Oxford University Press, Oxford.
125
This page intentionally left blank
Grazer Philosophische Studien 69 (2005), 127–146.
CHALLENGING CONTEXTUALISM Martijn BLAAUW University of Aarhus Summary In order to explain such puzzling cases as the Bank Case and the Airport Case, semantic contextualists defend two theses. First, that the truth-conditions of knowledge sentences fluctuate in accordance with features of the conversational context. Second, that this fluctuation can be explained by the fact that ‘knows’ is an indexical. In this paper, I challenge both theses. In particular, I argue (i) that it isn’t obvious that ‘knows’ is an indexical at all, and (ii) that contrastivism can do the same work as contextualism is supposed to do, without being linguistically implausible.
0. Introduction Contextualists such as DeRose and Cohen argue that the truth-conditions of knowledge ascribing and knowledge denying sentences (or ‘knowledge sentences’ for short) fluctuate in accordance with features of the conversational context in which the knowledge ascription or knowledge denial is made. And they explain this fluctuation by arguing that ‘knows’ is an indexical: ‘knows’ has an invariant character but a variable content. DeRose and Cohen thus defend a very specific type of contextualism, to which I will refer as semantic contextualism. Semantic contextualism, then, consists of two theses. First, the thesis that the truth-conditions of knowledge sentences fluctuate in accordance with features of the conversational context of the attributer of knowledge. I call this the Fluctuation Thesis. Second, the thesis that this fluctuation should be explained by maintaining that ‘knows’ is an indexical. I call this the Indexicality Thesis. In this paper, I focus on both theses and argue for the following two claims. First, that the Indexicality Thesis is implausible from a linguistic point of view — it isn’t obvious that ‘knows’ is an indexical
at all. Second, that the Fluctuation Thesis offers an explanation of the puzzle-cases that is rivalled — there is another explanation to consider that does the same work contextualism is designed to do without being linguistically implausible. All this does not show that contextualism is false. But it does show that the contextualist needs to do more work to make her theory acceptable. 1. Contextualism and fluctuation Semantic contextualists (or just: contextualists) typically have two reasons to defend their two theses. First and foremost, defending the two theses enables them to explain away some puzzling cases of knowledge attributions and knowledge denials such as DeRose’s Bank Case, or Stewart Cohen’s Airport Case. In the second place, it enables them to respond to the sceptical paradox.1 In what follows, I will focus solely on the first motivation. The puzzling cases that are the motivation for contextualism usually are pairs of cases with — in the one case — a knowledge attribution, and — in the other case — a knowledge denial. In both cases, the subject, the proposition, and the time that figure in the respective knowledge sentences remain constant. Moreover, there does not seem to be any significant epistemic difference between the two cases. Now what is puzzling about the cases is that we have the intuition that both the knowledge attribution and the knowledge denial are true. As an example of such a case, consider the Bank Case that has been put forward by Keith DeRose: Bank Case A. My wife and I are driving home on a Friday afternoon. We plan to stop at the bank on the way home to deposit our paychecks. But as we drive past the bank, we notice that the lines inside are very long, as they often are on Friday afternoons. Although we generally like to deposit our paychecks as soon as possible, it is not especially important in this case that they be deposited right away, so I suggest that we drive straight home and deposit our paychecks on Saturday morning. My wife says, 1. For more on this, see DeRose (1995). At the Stirling contextualism conference (March 2004), DeRose acknowledged — contrary to what most people thought — that contextualism is primarily motivated by its alleged ability to adequately respond to such puzzling cases as the Bank Cases.
128
“Maybe the bank won’t be open tomorrow. Lots of banks are closed on Saturdays.” I reply, “No, I know it’ll be open. I was just there two weeks ago on Saturday. It’s open until noon.” Bank Case B. My wife and I drive past the bank on a Friday afternoon, as in Case A, and notice the long lines. I again suggest that we deposit our paychecks on Saturday morning, explaining that I was at the bank on Saturday morning only two weeks ago and discovered that it was open until noon. But in this case, we have just written a very large and very important check. If our paychecks are not deposited into our checking account before Monday morning, the important check we wrote will bounce, leaving us in a very bad situation. And, of course, the bank is not open on Sunday. My wife reminds me of these facts. She then says, “Banks do change their hours. Do you know the bank will be open tomorrow?” Remaining as confident as I was before that the bank will be open then, still, I reply, “Well, no. I’d better go in and make sure.” (DeRose 1992, 913)
According to DeRose, in these and similar cases we have the intuition that the knowledge attributing sentence (in Bank Case A) and the knowledge denying sentence (in Bank Case B) are true. That is, we are strongly inclined to think that both in Bank Case A and in Bank Case B, DeRose is truthfully attributing knowledge to or denying knowledge of himself. And this is the case, even when DeRose is in no better epistemic position to know in Bank Case A than he is in Bank Case B; in Bank Case A he doesn’t have more justification or more warrant than he has in Bank Case B. Given the fact that it is puzzling how a knowledge attribution and its denial can both intuitively be true while all the relevant parameters remain the same, we would like to see an explanation of how this is possible. As a first step towards explaining this, contextualists argue that the truth-conditions of knowledge sentences vary in certain ways in accordance with the conversational context in which they are uttered. In some (very strict) contexts, ‘S knows that p’ requires for its truth that S be in a very strong epistemic position, while in other (more liberal) contexts, ‘S knows that p’ requires for its truth only that S meet some lower epistemic standards. It is possible, then, that one speaker truthfully says ‘S knows that p’ while another speaker truthfully says ‘S doesn’t know that p’, though both speakers are talking about the same person S and the same proposition p at the same time. Here is how DeRose himself puts the point:
129
[Contextualism] is a theory according to which the truth-conditions of sentences of the form ‘S knows that p’ or ‘S does not know that p’ vary in certain ways according to the context in which the sentences are uttered. (DeRose 1992, 914)
In sum, in explaining the Bank Cases, the contextualist adheres to the Fluctuation Thesis: The truth-conditions of knowledge sentences fluctuate in accordance with features of the conversational context of the attributer of knowledge. 2. Contextualism and indexicality What feature of ‘knows’ makes that the truth-conditions of knowledge sentences are indeed sensitive to changes in the conversational context? An explanation of the Fluctuation Thesis that has been put forward by DeRose and Cohen is that ‘knows’ is an indexical in the fashion Kaplan thought of indexicals: it has an invariant character but a shifty content.2 To illustrate, according to Stewart Cohen: [Contextualism] … construes ‘knowledge’ as an indexical. As such, one speaker may attribute knowledge to a subject while another speaker denies knowledge of that same subject, without contradiction. (Cohen 1988, 97)
In order to strengthen our grip on contextualism, we need to get more precise about what it means to say that ‘knows’ is an indexical in the Kaplan-sense of the term. Kaplan distinguished between the ‘character’ of an indexical and the ‘content’ of an indexical. Consider the indexical ‘I’, for instance. The constant character of ‘I’ is that ‘I’ always refers to the person who utters ‘I’, whereas the shifty content of ‘I’ is just the person who uses ‘I’. Contextualists claim that ‘knows’, like ‘I’, has a constant character but a shifty content. Here, Keith DeRose says: While ‘knows’ is being used with the same character, it is not being used with the same content. Or so the contextualist will claim. (DeRose 1992, 921)
2. Compare Kaplan (1989).
130
This naturally presents two questions. First, what is the character of ‘knows’? And, second, what is the content of ‘knows’. As to the character of ‘knows’, here is what DeRose has to say: (…) the character of ‘S knows that p’ is, roughly, that S has a true belief and is in a good enough epistemic position with respect to p; this remains constant from attribution to attribution. (DeRose 1992, 922)
And as to the content of ‘knows’, DeRose says, But how good is good enough? This is what varies with context. What the context fixes in determining the ‘content’ of a knowledge attribution is how good an epistemic position S must be in to count as knowing that p. (DeRose 1992, 922)
In what follows, I will call this type of contextualism semantic contextualism, since at its core we can clearly find a semantic thesis about ‘knows’. In sum, in explaining what feature of ‘knows’ makes that the truth-conditions of knowledge sentences are sensitive to changes in the conversational context, semantic contextualists adhere to the Indexicality Thesis: ‘Knows’ has an invariant character but a variable content. Now it is important to note that there are types of indexicality and, accordingly, different indexical models for ‘knows’. Cohen (1999), for instance, takes as his indexical model for ‘knows’ words like ‘tall’ and ‘flat’. DeRose (1992; 1999), however, takes as his indexical model for ‘knows’ words like ‘I’ and ‘here’. So although both Cohen and DeRose defend that ‘knows’ is an indexical, what they mean by this is subtly different. We have seen that the contextualist defends two theses: the Fluctuation Thesis and the Indexicality Thesis. Now how do these two theses explain the puzzling Bank Cases? The central idea is that the knowledge attribution (in Bank Case A) and the knowledge denial (in Bank Case B) are not contradictory after all. What DeRose attributes to himself in Bank Case A is that he knows — in a sense of ‘knows’ that involves weak epistemic standards — that the bank will be open tomorrow. And what DeRose attributes to himself in Bank Case B is that he does not know — in a sense of ‘knows’ that involves strong epistemic standards — that the bank will be open tomorrow. But it is perfectly
131
coherent to say that ‘S knowsweak that p’ and that ‘S does not knowstrong that p’. And, crucially, this is the case even if both the subject, the proposition, and the time remain constant in both knowledge sentences. So in this respect, ‘knows’ is similar to ‘tall’ and ‘here’ (or so the contextualist will claim at any rate). There is no such thing as being ‘tall’ simpliciter; one is always ‘tall’ relative to a particular reference class. Accordingly, so long as different classes of reference are in play, a tallness attribution and a tallness denial need not be contradictory. And words like ‘here’ do not have a fixed content. Accordingly, so long as different contents are in play, a ‘here’-attribution and a ‘here’-denial need not be contradictory. My purpose in the remainder of this paper is to challenge both the Indexicality Thesis and the Fluctuation Thesis. I will, first, argue that the thesis that ‘knows’ is an indexical is extremely implausible if we take into account some relevant linguistic data (§3). I will, secondly, argue that the thesis that the truth-conditions of knowledge sentences fluctuate in accordance with features of the conversational context isn’t the only possible explanation of the Bank Case (and of other similar cases). Specifically, I will show that epistemological contrastivism (e.g., the thesis that the knowledge relation doesn’t hold between two relata but between three relata — a subject, a proposition, and a set of contrastive propositions) can explain the Bank Cases without being linguistically implausible (§4–6). 3. Testing Indexicality How plausible is it to think that ‘knows’ is an indexical? In particular, can we still defend the Indexicality Thesis if we take into account some relevant linguistic data? Given that DeRose and Cohen take different indexical models for ‘knows’, this question translates into: (i) do the linguistic data support that ‘knows’ is an indexical in the sense that ‘here’ is an indexical? and (ii) do the linguistic data support that ‘knows’ is an indexical in the sense that ‘tall’ is an indexical? The purpose of this section is to argue that both questions should be answered negatively. In order to do so, I will try and cast some doubt on both types of indexical model for ‘knows’ by providing a couple of experiments. Roughly, the idea is to find various specific characteristics of indexicals like ‘tall’ and ‘here’, and to investigate whether ‘knows’
132
shares these characteristics. If so, this strengthens the contextualist case. If not, this weakens the contextualist case: she needs to do more work to make the claim that ‘knows’ functions like an indexical — of either type — acceptable.3 TEST (A) One feature of indexicals like ‘tall’ is that some things can just be too tall, or not tall enough, for the purposes at hand. The following examples illustrate this: (1) John is not tall enough to be a basketball player. (2) Jill is too tall for her bed. In both cases, the degree of tallness exhibited just doesn’t fit the degree of tallness required. If ‘knows’ is an indexical like ‘tall’, we may expect that it displays this kind of sensitivity to what is required as well. Yet it is hard to see how ‘knows’ displays this kind of sensitivity: (3) ?John does not know enough that he has hands. (4) ?Jill too knows (/knows too much) that he has hands. I take it that (3) and (4) do not sound right. How can one know not enough that something is the case, or how can one know too much that something is the case?4 TEST (B) A further feature of indexicals like ‘tall’ is that they can be modified to various degrees5: (5) John is quite tall. (6) John is very tall. (7) John is modestly tall.
3. We can find a simililar strategy in Jason Stanley’s (2004). Especially relevant here is his (2004, section 2). 4. I don’t mean to deny, of course, that one can know too much. Imagine the criminal whispering to his accomplice ‘We should kill John — he knows way too much’. But I think that such a sentence simply means that John knows too much propositions (and not that John knows one proposition ‘too much’). 5. See Stanley (2004, 124ff).
133
If ‘knows’ is an indexical like ‘tall’, we should be able to modify it as well. But modification doesn’t seem to work for ‘knows’. Compare these sentences: (8) ?John quite knows that he has hands. (9) ?John very knows that he has hands. (10) ?John modestly knows that he has hands. I take it that (8), (9), and (10) do not sound right. Therefore I think that ‘knows’ fails the modification test. TEST (C) Another feature of indexicals like ‘tall’ is that they can be used in a comparative way.6 For ‘tall’ we can say ‘taller than’, for ‘flat’ we can say ‘flatter than’. If ‘knows’ is an indexical like ‘tall’, we should be able to use it in a comparative way as well. But there is no such comparative use of ‘knows’. Even on a charitable construal, sentences like (11) do not sound acceptable: (11) ?John knows that he has hands more than Jill does. Of course, we can say that John has more by way of evidence or justification for his belief that he has hands than Jill has for her belief that she has hands. But I don’t think this implies that John then more knows that he has hands than Jill does. And, of course, we can say that John knows more about his hands than Jill does (knows more propositions, that is). John might know, for instance, that the thenar eminence is the area between his thumb and wrist that consists of muscles. But that’s something completely different from what we are considering here. Therefore I think that ‘knows’ fails the comparatives test. TEST (D) Yet another feature of indexicals like ‘tall’ is that they can be related to expected characteristics. So we can say (12) John is tall for a four-year old. (13) This house is small for its price. (14) This track is flat for a dirt track.
6. Also see Stanley (2004, 124ff) for more on this.
134
In all of these cases, something is reasonably expected: that a four-year old has a certain length; that an expensive house has a certain surface area; that a dirt track has a certain roughness. If ‘knows’ is an indexical like ‘tall’, it should be possible to relate it to expected characteristics as well. But although we can say ‘Jill knows a lot for a four-year old’, we can not say: (15) ?Jill knows that she has hands for a four-year old. — and that’s what we would expect if ‘knows’ were an indexical like ‘tall’. Therefore I think that ‘knows’ fails the expected characteristics test. Putting the preceding tests together, the thought arises that ‘knows’ cannot be treated as an indexical modelled after ‘tall’. TEST (E) Turning to ‘here’ as a different type of indexical model for ‘knows’, I think that this model doesn’t provide anything more by way of plausibility for the contextualist claim than ‘tall’ was supposed to provide. One feature of indexicals like ‘here’ (and ‘I’) is that they can be used to mislead. Here is an example of the misleading capacity of such indexicals that was introduced by Weatherson (2002): Jack’s and Rick’s. Jack leaves the following message on Jill’s answering machine late one Saturday night. “Hi Jill, it’s Jack. I’m at Rick’s. This place is wild. There’s lots of cute girls here, but I’m just thinking about you. See you soon.” In the background loud music is playing, as if Jack were at a nightclub, indeed as if Jack were at Rick’s, so Jill reasonably concludes that Jack was at Rick’s when he sent the message, and hence that ‘here’ refers to Rick’s. In fact Jack was home alone, but wanted to hide this fact, so he turned the stereo up to full volume while leaving the message. Despite the fact that a reasonable and attentive member of the target audience inferred on the basis of contextual clues left by Jack that the context was Rick’s, it was not, it was Jack’s house, and ‘here’ in Jack’s message referred to his house. (Weatherson 2002, 309)
In this example, Jack produces evidence to the effect that ‘here’ refers to ‘Rick’s’, while ‘here’ actually refers to ‘Jack’s’. Now, if ‘knows’ were an indexical, we would suspect that it can be used to mislead in a similar way as ‘here’ can be used to mislead. The most probable way in which contextualists will construe misleading
135
with ‘knows’ is by pretending one is in a different epistemic context from the epistemic context one is actually in.7 In a case of context-pretending, one pretends that one is in a particular context (for instance, a high-standards context) whereas one actually is in another context (for instance, a low-standards context). To illustrate, suppose that Jill calls John at the office to ask him whether he knows where her credit card is. John has lost this credit card yesterday, but is afraid to admit this to Jill. Therefore, he says, ‘Actually, honey, I’m just reading this book by Peter Unger called “Ignorance”, and I don’t know where your credit card is.’ In this case, John tries to mislead Jill into believing that he does not know where Jill’s credit card is. This is a strange way to mislead someone. But to make matters worse for the contextualist, this way of misleading is structurally different from the type of misleading involving ‘here’ that was portrayed above. In Jack’s and Rick’s, Jack tries to mislead Jill into thinking that ‘here’ refers to ‘Rick’s’ although it actually refers to ‘Jack’s’. So what we are looking for is a case in which one person is trying to mislead another person into thinking that ‘knows’ refers to ‘a high-standards context’ whereas it actually refers to ‘a low-standards context’. However, in the credit card example, John isn’t trying to mislead Jill into thinking that ‘knows’ refers to ‘a high-standards context’ nor to ‘a low-standards context’. What he is trying to do is mislead Jill into thinking that he does not know where her credit card is. So are there cases of misleading with ‘knows’ that are structurally similar to cases of misleading with ‘here’? I must admit that I cannot think of any. Think of you and I as being involved in a game. You are to fool me into believing that ‘knows’ refers to conditions c. I am to fool you into believing that ‘knows’ refers to conditions c*. Would you know what to do? I wouldn’t.8 Moreover, would you know why to do it? I wouldn’t. 7. Thanks to Jonathan Schaffer for discussion. 8. There are other ways to mislead involving ‘knows’, but they do not even seem approximately close to the type of misleading involved in Jack’s and Rick’s. Such standard ways of misleading involving ‘knows’ seem to be: Misleading A. Professor is lecturing on Descartes. Student looks as if he understands everything, and nods vigorously to express his agreement with the Professor. Student just seems to know what the Professor is talking about. Comment: Student is misleading Professor in thinking that Student already knew everything the Professor said.[see next page]
136
TEST (F) So neither ‘tall’ nor ‘here’ can be coherently seen as an indexical model for ‘knows’. Accordingly, it seems that the Indexicality Thesis as defended by DeRose and Cohen should be viewed with suspicion. The semantic contextualist might try to save the Indexicality Thesis by pointing out that there is a third type of indexicality (beyond the types associated with ‘tall’ and ‘here’) on the precedent of verbs such as ‘to like’. As far as I know, no contextualist has ever linked ‘knows’ to ‘to like’. But if ‘knows’ cannot be grouped in the class of ‘tall’ and ‘here’, linking ‘knows’ to ‘to like’ might be a way out for the contextualist. So might ‘knows’ not be grouped in that class of verbs? I think not. In the first place, treating ‘knows’ on the precedent of ‘to like’ runs into similar problems as treating ‘knows’ on the precedent of ‘tall’. All of the following sentences sound correct: (16) I don’t like Jill enough to marry her. (17) I like Jill a lot. (18) I like Jill more than I like Julie. However, the corresponding sentences in which ‘knows’ figures do not sound correct: (19) I don’t know enough that it rains. (20) I know a lot that it rains. (21) I know that it rains more than I know that today is Saturday. This counts against treating ‘knows’ on the precedent of ‘to like’. A second argument against treating ‘knows’ in a similar vein as ‘to like’ is the argument from meta-linguistic negation.9 Meta-linguistic negation is negation not of content but of the appropriateness of the utterance: Misleading B. Professor is lecturing on Descartes. He asks, ‘What is Descartes’ famous principle?’ Nobody answers. Professor asks Student1 — no response. Professor asks Student2 — no response. Finally, he just gives the answer himself ‘Cogito, ergo sum!’ Student3 nods vigorously, smiling knowingly. Comment: Student3 is misleading Professor in thinking that Student3 knew the answer (knew, that is, that Descartes’ famous principle is ‘Cogito, ergo sum’). In both Misleading A and B, there is some amount of misleading. But, again, they are structurally different from the type of misleading involved in Jack’s and Rick’s. 9. Thanks to Brian Weatherson for discussion on this subject.
137
(22) I don’t like strawberries, I love them! (23) I don’t like swimming pools, I hate them! (23) is a normal, everyday use of negations. (22), though, is what can be called a meta-linguistic negation (MLN). What is denied in (22) is not that one likes strawberries (for one does like them) but that it would be inappropriate to say that one likes strawberries since one actually loves them. This suggests, among other things, that there is a something of a scale from liking something to loving something. Moreover, MLN’s typically are followed by a rectifying clause. So we say ‘I don’t like strawberries, I love them!’ where ‘I love them’ is the rectifying clause. Moreover still, the rectifying clause is always a superlative; we wouldn’t consider ‘I don’t like swimming pools, I hate them!’ to be a MLN. Now is there such a use for ‘knows’? It seems not: (24) ?I don’t know that I have hands, I absolutely know it. sounds, to say the least, weird. I conclude that ‘knows’ can not be treated as an indexical verb such as ‘to like’ either. Putting the previous points together, there is a lot of evidence that ‘knows’ cannot be coherently treated as an indexical. Hence the contextualist needs to do more work to make plausible the claim that ‘knows’ functions like an indexical. Until that has happened, contextualism seems linguistically implausible. 4. Contrastivism As we saw in the previous section, the Indexicality Thesis is not very plausible. In this section, I turn to the Fluctuation Thesis. The core job of this thesis was to explain away such puzzling cases of knowledge attributions and knowledge denials such as the Bank Cases (§1). Over the years, various alternative explanations of these cases have been proposed, the most prominent of these being the so-called ‘warranted assertability manoeuvre’.10 In the remainder of this paper, I argue that there is a rival explanation of the phenomena the contextualist sets out to explain that 10. For more on warranted assertibility maneuvers, see DeRose (1999; 2002), Rysiew (2001), and Blaauw (2003).
138
neither involves accepting contextualism nor involves defending some version of a warranted assertability manoeuvre. I will argue that if we accept a contrastivist account of knowledge, then the Bank Case will pose no problem. Arguing for this requires three stages of development. First, we need to know what contrastivism is (§4). Second, we need to show that contrastivism is linguistically plausible (§5). Third, we need to show how contrastivism explains the Bank Cases (§6). So, first, what is contrastivism? Contrastivism is a view with respect to propositional knowledge. In a nutshell, the contrastivist disputes the standard view on which the knowledge relation holds between two relata: a subject S and a proposition p. Instead, the contrastivist argues that the knowledge relation holds between three relata: a subject S, a proposition p, and a set of contrastive propositions Q.12 So on the contrastivist view there is no such thing as knowing that p, though there is such a thing as knowing that p as opposed to a contrast class Q. Knowing a proposition is always knowing that proposition against the background of a set of contrastive propositions. Thus construed, contrastivism is similar both to the relevant alternatives theory and to contextualism. It is similar to the relevant alternatives theory in that contrastivism requires that for S to know that p, S must be able to eliminate certain error-possibilities (or relevant alternatives, or contrasts). It is similar to contextualism in that the context of the attributer determines what the error-possibilities are that S needs to eliminate in order to know. Contrastivism also is crucially different from both the relevant alternatives theory and from contextualism. It differs from the relevant alternatives theory in that on the contrastivist view, the error-possibilities that S must eliminate in order to know are incorporated into the knowledge relation, resulting in a three-place knowledge relation between a subject, a proposition and a set of (eliminated) error-possibilities.13 It differs from 11. For some key papers on contrastivism, see Schaffer (2004; 2005; manuscript). Also see Morton & Karjalainen (2003), Rieber (1998), and Johnsen (2001). 12. Of course, not every view according to which the knowledge relation holds between three relata is a contrastivist view. Hetherington (2001), for instance, defends that knowledge is gradable, and thus could be interpreted as defending the position that the knowledge relation is a three-place relation between a subject, a proposition and a degree. But he is clearly not a contrastivist. 13. This makes that contrastivists are not compelled to deny the closure principle for knowledge, as relevant alternatives theorists are compelled to do.
139
contextualism in that on the contrastivist view, the truth-conditions of knowledge sentences remain constant throughout contexts, resulting in an invariantist perspective on the nature of the knowledge relation. In sum, when it comes to the knowledge relation, contrastivists defend ternicity with respect to its adicity and invariantism with respect to its nature.14 5. Is contrastivism linguistically plausible? Since contrastivism is in the first instance a linguistic thesis, a natural question to ask is — is contrastivism linguistically plausible? One reason to think not is that we hardly ever see or hear contrasts in knowledge attributive sentences at all. Hence, an objector might conclude, contrastivism is on the wrong track since it seems to invent a third relatum. The present section will answer this objection. In order to do so, I first distinguish between six different knowledge attributions: binary knowledge attributions (Jack knows that it rains); stressed knowledge attributions (Jack knows that it rains); wh-knowledge attributions (Jack knows where the umbrella is); contrastive knowledge attributions (Jack knows that it rains rather than snows), direct object knowledge attributions (Jack knows the meteorologist); and skill-knowledge attributions (Jack knows how to unfold his umbrella). These six knowledge attributions naturally divide into two categories. First, the category that is in some sense concerned with propositions — to this category belong binary, stressed, and wh-knowledge attributions. Secondly, the category that is not concerned with propositions — to this category belong direct object knowledge attributions and skill-knowledge attributions.15 Since I am only concerned with propositional knowledge, I will for present purposes concentrate on instances of knowledge attributions of the first category. Now the objection to contrastivism can be put as saying that though it may be true that we sometimes make contrastive knowledge attributions, 14. It is an important question what the nature of the contrastivized knowledge relation is. Since the answer to this question does not matter for present purposes, I will not go into it here. 15. Though note that Stanley and Williamson (2001) have argued that knowledge how reduces to knowledge that. Since, for present purposes, not much seems to depend on this, I will simply assume that knowledge how and knowledge that are different.
140
it is certainly not the case that we always make contrastive knowledge attributions. Most of our knowledge attributions are, or so the objector argues, of the binary form. In response, I will in the remainder of his section first show that all types of propositional knowledge attributions but the binary one naturally present contrasts, and, secondly, explain why those contrasts aren’t explicit. That all types of propositional knowledge attributions but the binary one present contrasts already puts some pressure on holding that binary knowledge attributions will follow their lead. Thirdly, I will turn explicitly to binary knowledge attributions to show that they present contrasts as well and, finally, explain why the contrasts aren’t explicit in those attributions.16 Consider contrastive knowledge attributions first. Suppose that Jack says ‘I know that it rains rather than snows’. This knowledge attribution is clearly contrastive. The contrast is explicitly present in the knowledge attribution so there can be no question of why we don’t see it (for we do see it!). Turning to wh-knowledge attributions, take a look at these typical wh-knowledge constructions: (25) Jack knows where the umbrella is. (26) Jack knows what time it is. (27) Jack knows when his birthday is. I make two remarks with respect to these sentences. As a first remark, I would like to note that wh-knowledge attributions are probably the attributions that we use most. We hardly ever say things like ‘Jack knows that it rains,’ or ‘Jack knows that it rains rather than snows,’ but we do frequently say things like ‘Jack knows what the weather is like,’ or ‘Jack knows what time it is’. As a second remark, I would like to note that there is a strong connection between wh-knowledge and wh-questions.17 If the question is ‘Where is the umbrella?’ and if Jack truly answers ‘It is in the cupboard,’ then we might say ‘Jack knows (the answer to the question) where the umbrella is’. If the question is ‘What is the time?’ and if Jack truly answers ‘It is three o’clock,’ then we might say ‘Jack knows (the answer to the question) what time it is’. Wh-knowledge attributions are made in the context of a question. If we add that questions naturally present alternatives, then it follows 16. Schaffer (2005, section 3) makes a similar point. 17. This connection is also drawn out by Morton and Karjalainen (2003).
141
that wh-knowledge attributions will naturally present alternatives as well, namely those alternatives that are significant in the context of the question. So ‘Jack knows (the answer to the question) where the umbrella is’ presents a set of alternative umbrella locations from which Jack is invited to select the correct one. And ‘Jack knows what time it is’ presents a set of alternative times from which Jack is invited to select the correct one. So when it comes to wh-knowledge attributions, the contrasts are not visible since it will be clear from the context of the attribution what they are. To repeat them would be redundant. Turning to stressed knowledge attributions, compare: (28) Jack knows that the umbrella is in the cupboard. (29) Jack knows that the umbrella is in the cupboard. (30) Jack knows that the umbrella is in the cupboard. The first of these stressed sentences, stresses ‘umbrella’. If we say this sentence out loud, then we may end up wondering: the umbrella rather than what? The second of these stressed sentences stresses ‘in’. If we say this sentence out loud, then we may end up wondering: in rather than what? The third of these stressed sentences stresses ‘cupboard’. If we say this sentence out loud, then we may end up wondering: the cupboard rather than what? Stressed knowledge attributions naturally present contrasts. And they will not be explicit since it will be obvious what they are, due to the context of the attribution. To repeat them would be redundant. As a small digression, I would like to draw out the connection between stressed knowledge attributions and wh-knowledge attributions. Take a closer look, for instance, at (29). When would anyone say such a thing? Probably in a context in which it is questioned whether Jack knows where the umbrella is positioned with respect to the cupboard. So if Jill says ‘Jack knows that the umbrella is in the cupboard,’ then the contrasts she has in mind are other locations where the umbrella might be positioned with respect to the cupboard. Interestingly, we see from this that wh-knowledge attributions and stressed knowledge attributions are connected. If a word in a knowledge attribution is stressed, then there is (or we can imagine) a context in which there is a debate about where, why, when, which, what, etc. Stressed knowledge attributions and wh-knowledge attributions are deeply connected: stressed knowledge attributions invite us to consider a set of contrasts through wh-questions.
142
So all knowledge attributions but the binary one present contrastive sets. I now turn, finally, to binary knowledge attributions. The contrastivist argues that there are no binary attributions: strictly speaking, every binary knowledge attribution is an implicit contrastive knowledge attribution. If this is correct, however, then why isn’t the contrast class explicit? Doesn’t the contrastivist simply invent the third relatum? I have two answers to this objection. My first answer is that binary knowledge attributions are typically instances of stressed knowledge attributions. Whenever we make a binary knowledge attribution — which is quite rare anyway — we have the tendency to stress particular words. And we have already seen why stressed knowledge attributions can do without explicit mention of the contrasts and how the contrasts are reached (through wh-questions). On this view, binary (/stressed) knowledge attributions are simplified versions of contrastive knowledge attributions, and they are simplified because repeating the contrasts would be redundant. In some cases, however, the contrasts aren’t explicit not because stress already provided them, but because the knowledge sentence is followed by an explanation of why one thinks to know (or fail to know) the proposition in question. My second answer to the question why the contrast class isn’t explicit, therefore, is that the explanation of why one thinks to know (or fail to know) provides it. The Bank Cases are an illustration of this mechanism, as I will show in the following section. 6. Contrastivism and the Bank Cases Can contrastivism explain the Bank Cases? I think it can: contrastivism provides an invariantist way of dealing with the Bank Cases that respects our intuitions that the knowledge attribution in Bank Case A is true and that the knowledge denial in Bank Case B is true as well. The crucial element in the Bank Cases is that we are dealing with a knowledge attribution (in Bank Case A) and a knowledge denial (in Bank Case B) that intuitively are both true, even though they concern the same subject, the same time, and the same proposition. Moreover, there is no epistemic difference between the two cases. The Bank Cases thus present us with a puzzling situation. The contextualist explains this puzzling situation by arguing that Bank Case A involves a low-
143
standards context, whereas Bank Case B involves a high-standards context. In Bank Case A, DeRose does not need to be in a very strong epistemic position in order to count as knowing that the bank will be open. In Bank Case B, however, DeRose must be in a very strong epistemic position in order to count as knowing that the bank will be open. So according to the contextualist, the word ‘knows’ has different contents in the two respective Bank Cases. This explains why both the knowledge attribution and its denial can be true: they don’t contradict. On contrastivism, however, the knowledge attribution in Bank Case A: (31)
I know that the bank will be open tomorrow.
is made with a set of contrasts in mind, as is the case with the knowledge denial in Bank Case B: (32)
I don’t know that the bank will be open tomorrow.
So what are the contrasts? Starting with (31), DeRose himself seems to provide them. When DeRose’s wife says ‘Maybe the bank won’t be open tomorrow. Lots of banks are closed on Saturdays’, DeRose responds ‘No, I know it’ll be open. I was just there two weeks ago on Saturday. It’s open until noon.’ In responding in this way, he explains why he thinks to know that the bank will be open tomorrow. So what he is actually claiming is: (31c) I know that the bank will be open tomorrow morning rather than being closed tomorrow morning in accordance with its old operating hours. Turning to (32), what could be the contrasts here? DeRose again seems to provide them. When DeRose’s wife says ‘Banks do change their hours. Do you know the bank will be open tomorrow?’, DeRose responds by saying ‘Well, no, I’d better go in and make sure.’ In responding in this way, he explains why he thinks that he fails to know that the bank will be open tomorrow. So what he is actually claiming is: (32c) I don’t know that the bank will be open tomorrow morning
144
rather than being closed tomorrow morning in accordance with its new operating hours.18 When we convert (31) and (32) into (31c) and (32c) we respect the everyday use of ‘knows’ without having to succumb to a contextualist account of ‘knows’, nor to a WAM for that matter. For (31c) and (32c) can both be true without contradiction because the contrasts are different.19 The key to explaining the Bank Cases is to acknowledge that the binary knowledge attributions that are made are implicit contrastive knowledge attributions. Put differently, due to the difference in contrasts, the proposition that is known or not known in the two respective Bank Cases is different. 7. Conclusion In this paper, I have argued for two claims. First, that the linguistic data do not speak in favour of contextualism at all. Various tests can be construed that show that ‘knows’ neither fits the ‘here’-model for indexicality nor the ‘tall’-model for indexicality. Moreover, ‘knows’ doesn’t fit the ‘to like’-model for indexicality either. This does not constitute a knock-out argument against contextualism. But it does make one wonder whether the contextualist has anything to say to make the Indexicality Thesis more plausible. Secondly, I have argued that there is another explanation of the Bank Cases that is not implausible from a linguistic point of view. On this contrastivist explanation, the binary knowledge attributions that are made in the two Bank Cases are actually implicit contrastive knowledge attributions. Once we make explicit the contrasts involved in the knowledge attributions, the knowledge attributions no longer contradict. Thus, we can both know that the bank will be open tomorrow and fail to know that the bank will be open tomorrow. The contrasts make all the difference.20 18. Thanks to Jonathan Schaffer for helpful comments here. 19. There are similar results for Cohen’s aiport case. Demonstrating this, however, I leave as an excersice to the reader. 20. Versions of this paper have been presented at conferences and seminars in Amsterdam, Stirling, and Aarhus. I thank the audiences on those occassions for their perceptive comments. For useful feedback on earlier versions of this paper, I would especially like to thank Duncan Pritchard, Jonathan Schaffer, and René van Woudenberg.
145
REFERENCES Blaauw, M. (2003). ‘Wamming Away at Contextualism’, SATS: Nordic Journal of Philosophy 4, 88–98. Cohen, S. (1988). ‘How to be a Fallibilist’, Philosophical Perspectives 2, 91–123. — (1999). ‘Contextualism, Scepticism, and the Structure of Reasons’, Philosophical Perspectives 13: Epistemology, (ed.) J. Tomberlin, 57–89. DeRose, K. (1992). ‘Contextualism and Knowledge Attributions’, Philosophy and Phenomenological Research 52, 913–929. — (1995). ‘Solving the Sceptical Problem’, Philosophical Review 104, 1–52. — (1999). ‘Contextualism: An Explanation and Defense’, The Blackwell Guide to Epistemology, (eds.). E. Sosa & J. Greco, 187–206, Blackwell, Oxford. — (2002). ‘Assertion, Knowledge, and Context’, Philosophical Review 111: 167–203. Hetherington, S. (2001). Good Knowledge, Bad Knowledge, Oxford University Press, Oxford. Johnsen, B. (2001). ‘Contextualist Swords, Sceptical Plowshares’, Philosophy and Phenomenological Research 62, 385–406. Kaplan, D. (1989). ‘Demonstratives’, Themes from Kaplan, (eds.) J. Almog, J. Perry, and H. Wettstein, Oxford University Press, Oxford. Morton, A. and Karjalainen, A. (2003). ‘Contrastive Knowledge’, Philosophical Explorations 6, 74–89. Rieber, S. (1998). ‘Scepticism and Contrastive Explanation’, Noûs 32, 189– 204. Rysiew, P. (2001). ‘The Context-Sensitivity of Knowledge Attributions’, Noûs 35, 477–514. Schaffer, J. (2004). ‘From Contextualism to Contrastivism in Epistemology’, Philosophical Studies 119, 73–103. — (2005). ‘Contrastive Knowledge’, Oxford Studies in Epistemology 1, (eds.) T. Gendler & J. Hawthorne, Oxford University Press, Oxford. — (manuscript). ‘Knowing the Answer’. Stanley, J. and Williamson, T. (2001). ‘Knowing How’, Journal of Philosophy 98, 411–44. Stanley, J. (2004). ‘On The Linguistic Basis for Contextualism’, Philosophical Studies 119, 119–146. Weatherson, B. (2002). ‘Misleading Indexicals’, Analysis 62, 308–10.
146
Grazer Philosophische Studien 69 (2005), 147–164.
CONTEXTUALISM AND THE MANY SENSES OF KNOWLEDGE René van WOUDENBERG Vrije Universiteit Amsterdam Summary Contextualists explain certain intuitions regarding knowledge ascriptions by means of the thesis that ‘knowledge’ behaves like an indexical. This explanation denies what Peter Unger has called invariantism, i.e., the idea that knowledge ascriptions have truth value independent of the context in which they are issued. This paper aims to provide an invariantist explanation of the contextualist’s intuitions, the core of which is that ‘knowledge’ has many different senses.
0. Contextualism’s crucial contention Contextualism, as developed by Stewart Cohen and Keith DeRose, is a theory according to which the truth-value of sentences in which knowledge is ascribed to, or denied of, a certain subject is relative to the context of the utterer of the sentence. So sentences of the form “S knows that P” and “S doesn’t know that P” (henceforth referred to as K-sentences), aren’t true or false simpliciter, but are only true or false relative to the context of the utterer. As DeRose says: In some contexts [of utterance, RvW], “S knows that P” requires for its truth that S have a true belief that P and also be in a very strong epistemic position with respect to P, while in other contexts, the very same sentence may require for its truth, in addition to S’s having a true belief that P, only that S meet some lower epistemic standards. Thus, the contextualist will allow that one speaker can truthfully say “S knows that P”, while another speaker, in another context where higher standards are in place, can truthfully say “S doesn’t know that P”, though both speakers are talking about the same S and the same P at the same time. (DeRose 1999, 187–8)1 1. It should be noted that DeRose wavers as to what it is that is context relative. Most
The same point can also be put as follows. The contextualist allows that “S knows that P” can both be true and false at the same time, provided there are two contexts, a low standard one (in which it is true) and a high standard one (in which it is false). Cohen and DeRose have proposed that the contextual relativity of the truth-value of K-sentences is due to the fact that the word ‘know’ and its cognates are context relative in much the same way as attributive terms2 like ‘flat’ and ‘bald’, and indexicals like ‘I’ and ‘here’ are. In this paper I will argue against their proposal. I will furthermore argue that the sort of cases they have marshalled in order to provide contextualism with some initial credibility (or perhaps I should say: the sort of cases that trigger contextualist intuitions), can be explained in a better way, one that has no contextualist implications. I will go some way towards arguing that the word ‘knowledge’ and its cognates such as ‘know’, have various senses or meanings. In effect I will be suggesting that the contextualist’s intuitions can best be explained by reference to the phenomenon of polysemy. 1. Contextualism’s motivation I start out my argument by reminding us of two important motivations behind contextualism. First, there is the ambition to solve the sceptical problem, and, second, there is the conviction that the sceptical problem can be solved in the slip-stream of an explanation of a surprising feature of our intuitions about the truth-value of K-sentences. Let me expand on both. The sceptical problem contextualism means to solve is the problem of how to respond to the following closure-based sceptical argument: (1) I don’t know that I am not a brain-in-a-vat (or BIV for short). (2) If I don’t know that I am not a BIV, then I cannot know anything about the external world. (3) Therefore, I cannot know anything about the external world. (cf. DeRose 1995, 1) often he says it is the truth-value of K-sentences. But sometimes he also says that it is knowledge, and other times that it is the meaning of ‘knowledge’ that is context relative. I will be working with the formulation that contextualism is a thesis about the truth-value of K-sentences. 2. Searle reports that old-fashioned grammarians called ‘tall’, ‘hot’, and other such rela-
148
The conclusion of this argument, (3), is surely most unwelcome and no one, or virtually no one, will accept it. After all, we do think that we know a fair bit about the external world. Since the conclusion does seem to follow from the premises, the only way to escape it is to deny at least one of the premises. But when we look at the premises, they seem so obviously true that we wouldn’t want to deny either of them. We therefore face the problem of how to respond to the sceptical argument. This is the sceptical problem. Contextualists think they can solve it in the slip-stream of an explanation they have proposed for certain intuitions about K-sentences, for instance certain intuitions about the Bank Cases that DeRose has presented. Bank Case A. My wife and I are driving home on a Friday afternoon. We plan to stop at the bank on the way home to deposit our paychecks. But as we drive past the bank, we notice that the lines inside are very long, as they often are on Friday afternoons. Although we generally like to deposit our paychecks as soon as possible, it is not especially important in this case that they be deposited right away, so I suggest that we drive straight home and deposit our paychecks on Saturday morning. My wife says, ‘Maybe the bank won’t be open tomorrow. Lots of banks are closed on Saturdays.’ I reply, ‘No, I know it’ll be open. I was just there two weeks ago on Saturday. It’s open until noon.’ Bank Case B. My wife and I drive past the bank on a Friday afternoon, as in Case A, and notice the long lines. I again suggest we deposit our paychecks on Saturday morning, explaining that I was at the bank on Saturday only two weeks ago and discovered it was open until noon. But in this case, we have just written a very large and very important check. If our paychecks are not deposited into our checking account before Monday morning, the important check we wrote will bounce, leaving us in a very bad situation. And, of course, the bank is not open on Sunday. My wife reminds me of these facts. She then says, ‘Banks do change their hours. Do you know the bank will be open tomorrow?’ Remaining as confident as I was before that the bank will be open then, still, I reply, ‘Well, no. I’d better go in and make sure.’ (De Rose 1992, 913)
Here we have two cases with, in each, a K-sentence—the first contains a knowledge ascription, the second a knowledge denial. The important tive terms ‘attributive terms’ (Searle 1979, 79). Even though neither DeRose nor Cohen uses this term, for ease of reference I will adopt it.
149
difference between these cases is that in case A not so much is at stake, whereas in B a lot is. DeRose’s intuitions tell him that in both cases the K-sentence is true. In Case A the sentence that he knows that the bank will be open on Saturday is true. In Case B the sentence that he doesn’t know that the bank will be open on Saturday is true. I call this a surprising feature of the contextualist’s intuitions. For isn’t it surprising to have the intuition that both “S knows that p” and “S doesn’t know that p” are true? Aren’t the contextualist’s intuitions simply contradictory? 2. Contextually varying standards for knowledge ascriptions The contextualist now proceeds to explain these seemingly contradictory intuitions, that is, he tries to show that they are not contradictory after all. In order to do just this, he frames an answer to the question “When, generally speaking, are K-sentences true?” An answer to this question, the contextualist seems to imply, will explain the seemingly conflicting intuitions and, in its slip-stream, solve the sceptical problem. So when, generally speaking, are K-sentences true? We have seen the essence of the contextualist’s answer in the quotation in section 0: that depends on the conversational context of the utterer. For an utterer’s K-sentence to be true in an ordinary context, much less is required than in a context where higher standards are in place. In some contexts what is required for the truth of a knowledge ascription is that S believes that P and also is in a very strong epistemic position with respect to P. In other contexts the same knowledge ascription, i.e., the ascription of knowledge of the same proposition P to the same S, requires for its truth next to S’s having the true belief that p, only that S meets some lower epistemic standard. So, DeRose’s intuitions that in the one case “I know that the bank is open on Saturday” is true whereas in the other its negation is true, are explained by invoking contexts of ascription. The contextualist then continues to argue that this explanation of what I have called the surprising features of the contextualist’s intuitions, helps to forestall the skeptical conclusion that no one knows anything about the external world. The reasoning is that when the ascriptor is in a non-sceptical context, i.e., in a context where the stakes are not so high, then for the subject S to know that P, it isn’t needed that S is in a very strong epistemic position with respect to P. The reasoning is that
150
in certain ordinary contexts an ascriptor may truthfully ascribe knowledge to S even when S’s evidence does not exclude certain skeptical possibilities. In a sceptical context, however, i.e., in a context where the standards are very high, an ascriptor may truthfully ascribe knowledge to S only if S is in a very strong epistemic position vis-à-vis P. And what does that mean? It means that S is able to rule out certain sceptical possibilities. Since S cannot rule out such possibilities, knowledge ascriptions are false in contexts where the standards are high. Nevertheless, in non-skeptical contexts the same knowledge ascriptions may be true. In this way the general skeptical conclusion is evaded. Applied to the bank cases this gives the following result. In case A, in order for DeRose to be able to truthfully ascribe knowledge to himself, he only needs to meet a low epistemic standard, perhaps the standard of “having some reason for thinking that the bank is open on Saturdays”. In case B, however, in order for DeRose to truthfully ascribe knowledge to himself, he must meet a much stronger standard, perhaps the standard of “ruling out the possibility that the bank has changed its hours”. And so long DeRose hasn’t done that, the K-sentence “I don’t know that the bank will be open on Saturday” will be true, which means he has no knowledge of the bank’s opening hours. The contextualist’s explanation of the surprising features of his intuition also explains why the skeptical argument seems so compelling. The explanation is that the sceptical argument unthinkingly implies or assumes that the standards for knowledge ascriptions are the same in all contexts of attribution and that the standards for knowledge ascriptions are always as high as they are in sceptical contexts; i.e., it assumes that all knowledge ascriptions require for their truth that S is in a very strong epistemic position with respect to P. 3. No disquotation for the truth of K-sentence and no invariantism A contextualist, then, thinks he can say without contradiction or embarrassment that the sentence “S knows that P” can be true and false at the same time. He can say this without contradiction so long as one instance of “S knows that p” is uttered in a low standards context (where it is true), and the other in a high standards one (where it is false). He can also say this without embarrassment, because he is able to provide an explanation of the fact how K-sentences have this remarkable capability
151
of being true and false at the same time. This means that the contextualist in effect denies disquotation for K-sentences. He denies that the sentence “S knows that p” is true only if S knows that p, or only if S is in the knowledge state vis-à-vis p. As the contextualist sees it, whether or not a K-sentence is true depends on the contextual standards of the ascriptor. It doesn’t depend on the knowledge state of S, or the obtainment of the knowledge relation. Had it been thus dependent, disquotation would hold, and “S knows that p” would be true if and only if S knows that p. But it is not, says the contextualist, and hence disquotation doesn’t hold. Another way of saying essentially the same thing is that contextualism denies what Peter Unger has called ‘Invariantism’ (see DeRose 1999, 188): the claim that there is a single and invariant set of standards that determines whether or not a particular K-sentence is true, regardless of the context in which the K-sentence is uttered. 4. ‘Knows’ behaves like either an attributive term or an indexical The contextualist, then, holds that the truth-value of K-sentences depends on the context of the utterer. But how, exactly, should we understand this remarkable capability of K-sentences? What quality of K-sentences is responsible for their contextual dependency? The contextualist’s answer is that the word ‘knows’ linguistically behaves either like attributive terms such as ‘flat’ and ‘short’ (DeRose 1992, 925; DeRose 1995, 30), or like indexicals such as ‘I’ and ‘here’ (Cohen 1998, 517). The characteristic linguistic behaviour of such words is that sentences that contain them are not true or false simpliciter, but only true or false in (or given) a context. Some examples may illustrate this. Consider the following two sentences: (a) Sam is short. (b) He is in pain. The truth-value of these sentences depends on contextual factors, in the following way. If (a) is uttered in the context of a discussion about the length of basketball players it may be true, whereas if it is uttered in the context of a discussion about liliputs it may be false. The truth-value of (b) depends on contextual factors in a similar way. If ‘he’ refers to you
152
it may be true, but if it refers to me it may be false. In both cases, the context that is relevant for determining the truth-value is the context of the ascriptor. (For ease of reference, let us call sentences that contain attributive terms (i.e., adjectives that allow of degrees), A-sentences, and sentences that contain an indexical, I-sentences.) Since, as contextualists hold, ‘knows’ linguistically behaves like ‘flat’ and ‘I’, the truth-value of K-sentences also depends on contextual factors—factors immanent in the ascriptor’s context. For instance, if in a certain context someone says “S knows that P”, his K-sentence will be true if his conversational context determines that the K-sentence requires for its truth that S is in an average epistemic position vis-à-vis P; the K-sentence will be false, however, if the conversational context determines that the K-sentence requires for its truth that S be in a very strong epistemic position vis-à-vis P. DeRose’s bank case A is supposed to be a case where the K-sentence “The bank is open on Saturday until noon” is true; its truth does not require that S is in a very strong epistemic position with respect to that proposition. In case B, however, the K-sentence is false because the context of the ascriptor determines that for the K-sentence to be true, it is required for its truth that S is in a very strong epistemic situation vis-à-vis P—a requirement that, on DeRose’s view, isn’t satisfied. The contextualist’s basic idea, then, is that whether or not ‘knows’ applies to a subject, depends on the context of the knowledge ascriptor, just as whether or not ‘is tall’ applies to a subject depends on the context of the tallness ascriptor, and just as whether or not ‘here’ refers to Amsterdam depends on the location ascriptor. The contextualist, then, claims that ‘knows’ linguistically behaves either like attributive terms or like indexicals. He further claims that this fact explains both his intuitions about the bank cases and our intuition that in sceptical contexts we cannot ascribe knowledge to anybody, whereas in everyday contexts it is fully legitimate to do so. 5. Problems for the claim that ‘knows’ behaves like either an attributive term or an indexical The problem with this explanation, however, is that ‘know’ neither behaves like an attributive term, nor like an indexical.3 3. Related arguments are offered in Stanley (2004). See also Blaauw (2005).
153
As to the first point, words like ‘tall’, ‘small’, ‘heavy’ and ‘flat’, as I have indicated already, allow of degrees. There are degrees of tallness, smallness, etc. This fact accounts for the intelligibility of such comparative phrases as ‘A is taller than B’, ‘B is smaller than A’, ‘This is flatter than that’, ‘this is heavier than that’, etc. If ‘knows’ is to behave in much the same way as these adjectives, then it should allow of degrees as well. In that case, we should be able to say that one can know that P to various degrees. We should then be able to make sense of such expressions as “S knows more that P than S*”. But we can’t. S either knows he has two hands, or he doesn’t. Of course, S may have more reason to believe that he has two hands than S* has, but this does not mean that S* knows less that he has hands than S knows that he has hands. And again, sometimes we use the expression that S ‘knows more’ or ‘has more knowledge’ than S*. These locutions, however, never specify a proposition that is the object of knowledge. They specify, or suggest, only a subject matter, as in ‘S knows more about air pollution than S*.’ This indicates that while ‘knowing about (something)’ allows of degrees, ‘knowing that’ does not – and the contextualists have their eyes on the latter. One might be tempted to retort that there is a locution that suggests degrees of ‘knowing that’, viz., the locution ‘knowing better’, as in “Mary knows much better than I do, that Jack is wealthy”. Such expressions, however, should be taken to mean that Mary has had more opportunity to see for herself that Jack is wealthy. She doesn’t know ‘more’ that Jack is healthy than I do. ‘Knows’, then, doesn’t allow of degrees.4 This is of crucial importance in evaluating the contextualist’s case. For it is precisely the claim that adjectives like ‘small’ and ‘tall’ allow of degrees, that makes the truth-value of A-sentences context dependent. Is ‘knows’, then, an indexical, and are K-sentences I-sentences? One shared characteristic of indexicals like ‘I’, ‘here’ and ‘now’ is that while their meaning remains the same in every context in which they are uttered––namely the present speaker/writer, the present place, and the present time respectively—what they refer to is dependent on the context of the utterer. Put differently, the character of an indexical is 4. I can forego the fact that notions that allow for degrees differ among each other. Some notions that know degrees, have an ‘absolute maximum’, others don’t. For example, there doesn’t seem to be an absolute maximum of ‘greatness’, but there seems to be an absolute maximum of ‘flatness’. Although this difference is interesting in its own right, we can neglect it for present purposes.
154
context invariant, but its content is not (Kaplan 1989). ‘I’ uttered by me refers to me, but uttered by you to an entirely different thing: you. ‘Here’ uttered by me now refers to my office in Amsterdam, but may refer to a wholly different place when uttered by you. So tokens of the same indexical type have the same meaning but different referents, depending on the utterer. This is what is responsible for the context dependency of the truth conditions of I-sentences. Now, if ‘know’ is to behave like an indexical, it should display similar characteristics. For example, it should in various contexts have the same character, but different contents; or, as I prefer to put it, it should have the same meaning, but different referents. But this does not seem to be the case. Consider “S knows that P,” uttered by Alice and Mary respectively in very different contexts. The contextualist assumes that ‘knows’ retains the same character (meaning), but refers to one thing in Alice’s utterance and to another in Mary’s. My present point is that I see no reason to think that this is correct. I see no reason why, assuming ‘knows’ is used univocally, ‘knows’ refers to one epistemic position of S vis-à-vis P in Alice’s sentence but to another epistemic situation of S vis-à-vis P in Mary’s sentence. I simply can’t see that this is correct. And we have not been offered a compelling argument for the thesis that ‘knows’ behaves like an indexical. We have only been offered the argument that that thesis, if true, would nicely explain what I referred to as the surprising feature of the contextualist’s intuitions about, for example, the bank cases.5 But that is not an independent argument for the indexicality of ‘knows’. In the rest of this paper I will be working my way to the conclusion that, contrary to what contextualists (and in fact many if not most epistemologists now writing) assume, ‘knows’, as used in propositional contexts, has not one univocal meaning, one character that stays the same in varying contexts. ‘Knows’, I will be arguing, has many different meanings or senses, and this provides a better explanation of the contextualist intuitions about, for example, the bank cases.
5. Stephen Schiffer has offered the following sort of argument against ‘knows’ being an indexical. No one is ever really led into thinking that ‘I’, uttered by himself, refers to someone else. But the sceptical argument really can make one think that one doesn’t know anything. The semantics that the contextualists propose for the word ‘knows’ fails to explain why this should be the case. See Schiffer (1996).
155
6. The polysemy of knowledge: a first approach In order to make the idea that ‘knows’ has various meanings more vivid, I want to return to the bank cases. (But first I should like to draw attention to the fact that there is something artificial about the conversation reported in case B. Mrs. DeRose is reported as saying, “Banks do change their hours. Do you know the bank will be open tomorrow?” I don’t think it is natural to suppose that in a real life situation Mrs. DeRose would use those words. It is far more likely that she would have said, “Banks do change their hours. Are you sure the bank will be open tomorrow?” But let us stipulate that the conversation, albeit perhaps somewhat artificial, actually took place.) Then the point of case B is that the K-sentence “Keith knows that the bank is open on Saturdays till noon” is false. But why is it false? According to DeRose, because the conversational context determines that the truth of that K-sentence requires S’s being in a rather strong epistemic position vis-à-vis the proposition the bank is open on Saturdays till noon. And Keith isn’t in that position. Hence, the K-sentence is false. Now I myself don’t have any strong intuitions about the truth-value of this K-sentence. I would not want to say that it is false, nor that it is true. My response to the K-sentence involved is: whether it is true or false depends on what you take ‘knowledge’ to be. If you think that knowledge requires certainty, then you will think the K-sentence is false, since Keith isn’t certain that the bank is open on Saturdays till noon. But if you don’t think that knowledge requires certainty, then you might think that the K-sentence is true. If you think that knowledge requires appropriate evidence, then you may be unsure about whether the K-sentence it true; if you think that Keith’s evidence is appropriate, then you could also think that the K-sentence is true, but if you think it is not, then not. Or suppose you think that knowledge requires reliably formed belief. Then, if you think that Keith’s belief about the bank’s opening hours on Saturday is reliably formed, then you will think the K-sentence is true; but if you think it isn’t, then not. So, what I am saying is that the truth-value of K-sentences depends on what one takes ‘knowledge’ to require. This, of course, is no more than a first suggestion. In order to strengthen it, I should now like to offer two lines of argument.
156
7. The polysemy of knowledge: argument 1 My approach to K-sentences is that in different contexts, ascriptors use ‘knows’ in different senses or meanings. Hence, my claim is that, depending on the meaning of ‘knows’ used in the conversational context in which it is used, different properties can be ascribed to S. And, accordingly, my suggestion is that the contextualist’s intuitions about, for example, the bank cases should not be explained by an appeal to standards that can be differentially satisfied, nor by appealing to attributor contexts that determine how strong S’s epistemic position must be with respect to P, in order for the relevant K-sentence to be true. They should be explained by reference to the polysemy of ‘knows’. My first argument for the polysemy of ‘knows’ is that it seems to accord with a rather natural interpretation of actual and imaginary cases of knowledge ascriptions. I should like to make this clear by considering various cases of knowledge ascriptions. [A] Ann: “Do you know Jack’s telephone number?” William: “I believe it is 444.6676, but I’m not certain of it.” In the context of this discussion, William takes knowledge to require certainty. After all, since he feels unsure, he only ascribes belief to himself, not knowledge. In other contexts, however, ‘knows’ is used in such a way that certainty isn’t required for knowledge. Compare [B] Quizmaster: “When was the battle of Hastings?” Jack on television: “eh, eh, well, …” Quizmaster: “You are running out of time.” Jack: “eh, eh, … 1066???” [The zoom goes off immediately] Quizmaster: “Correct.” Jack’s mom watching television on the sofa with a friend: “I told you Jack would know the answer!” Here Jack’s mom takes knowledge to require truth, but definitely not certainty; nor does she take it to require having evidence, or being justified. She could even be taken to equate knowledge with ‘true belief’. Sometimes, however, ‘knows’ is used in a way that does require what Jack’s mom apparently does not require, viz., having a belief that is based on adequate evidence. Compare the following scenario (due to Stroud 1984):
157
[C] Not by his own will, Sam is locked into a room with some 10 television screens. Screen 1 shows a football match, screen 2 a weather report, screen 3 a presidential speech, screen 4 a royal wedding, etc. You and I are monitoring Sam and see him watching the weather report, that says: There was bright sunshine today. I say: “Now Sam knows there was bright sunshine today”. You: “No he does not, for Sam has no clue whatsoever whether or not the weather report he watched was in fact today’s weather report, nor whether or not the report was about the area he happens to be in (he was, after all, kidnapped). Sam is simply clueless, he lacks the appropriate evidence.” In this conversation you clearly take knowledge to require that one has appropriate evidence and ‘has’ it not in the way one can ‘have’ a coin in one’s pocket without knowing it, but has it in a reflective way, such that one can cite it, or such that one can reflexively access it. In effect you have, of course, articulated an internalist intuition. But now consider the following case: [D] Jane is standing in front of an apple tree. She forms the belief that there is an apple tree in front of her. I ask, “Does Jane know there is an apple tree in front of her?” You answer: “Sure, Jane’s perceptual apparatus is in perfectly good working order and her recognitional skills are excellent—she can tell an apple tree from a pear tree”. In this case, you obviously take knowledge to require a true belief that is produced by reliable mechanisms and skills—you don’t take it to require that Jane possesses evidence to which she has reflective access. Here you articulate an externalist intuition. In some situations that look like [D], however, something else is needed for knowledge. [E] Jane is standing in front of an apple tree. She forms the belief that there is an apple tree in front of her. I ask, “Does Jane know there is an apple tree in front of her?” You say: “Well, Jane is unable to rule out the possibility that she is dreaming that she sees a tree; she is furthermore unable to rule out the possibility that she is being deceived by a Cartesian demon; nor can she exclude the possibility that she is a BIV. Since, for all she knows, any of the scenario’s may obtain, Jane doesn’t know there is a tree in front of her.”
158
In this conversation you have used ‘know’ in a way that differs from the way you used it in [D]. Whereas in [D] you took knowledge to require a belief that is formed by de facto reliable mechanisms and skills, in [E] you take it to require the exclusion of certain skeptical possibilities. Many more of such cases can be offered. What I claim for these cases is (1) that the K-sentences that occur in them are nothing out of the ordinary––that those are the sort of sentences that are entirely natural to use in that sort of situations, (2) that this fact strongly suggests that the word ‘knows’ is used to ascribe rather different properties to subjects, and (3) that what this come to is that ‘knows’ has different meanings. 8. The polysemy of knowledge: argument 2 My second argument derives from the on-going and seemingly neverending discussion among epistemologists about the proper analysis of knowledge. The question that has framed and structured this discussion is very often put as follows: “What needs to be added to true belief in order to get knowledge?”, or “What is it that bridges the gap between mere true belief and knowledge?” (e.g., Plantinga 1993). A really bewildering array of answers has been offered. Here is a small sampling: knowledge is true belief [a] that is held with certainty. (Unger 1974) [b] that is such that the subject has been able to eliminate all sceptical alternatives to it.6 [c] that the subject has adequate evidence (reasons, grounds …) for. [d] that is based on adequate evidence. (Conee & Feldman 1985) [e] that is formed by a sufficiently reliable belief-forming process. (Goldman 1979; Kornblith 2002) [f ] that is formed by the proper functioning of one’s cognitive faculties. (Plantinga 1993) [g] that is formed by the exercise of an intellectual virtue. (Sosa 1991; Greco 2000) [h] that is rational in one of its many senses. (Foley 1987) 6. This is sometimes called the ‘Cartesian’ position, even though Descartes didn’t hold it. See BonJour (2002, 38–43).
159
that is justified and satisfies an anti-Gettier condition.7 that is such that the subject has cognitive access to the evidence. (BonJour 1985; BonJour 1998) [k] that the subject cannot be blamed for having. (Ginet 1975) [l] that is formed responsibly. (Chisholm 1977) [m] that coheres with a large body of one’s other beliefs. (BonJour 1985; Lehrer 1990) [i] [j]
This is a truly impressive list of answers, and there is no space here to discuss any of them in any detail, nor to discuss the interrelations between them.8 I shall have to confine myself to the remark that some of the answers on the list are radically different from some of the others. For example, [j] requires for knowledge that the subject has cognitive access to the evidence, which according to [e] isn’t even necessary. Answer [a] requires that for true belief to be knowledge, it need be held with certainty, but is silent about evidence, cognitive access, blamelessness, responsibility or coherence. And this raises the question why, if so many brilliant philosophers disagree so radically about what knowledge is and what it takes to have it, we shouldn’t conclude that there simply isn’t just one thing or state that is called ‘knowledge’. Given deep and continuing disagreements, shouldn’t we conclude that the word ‘knowledge’ simply isn’t used to refer to just one state of mind––‘the knowledge state of mind’? Shouldn’t we conclude that the word ‘knowledge’ refers to various different states of mind, and that it thus has a number of different senses? Before I am going to endorse this suggestion, I should like to acknowledge that there are continuing disagreements where it is out of doubt that there is one unique correct answer to the question under discussion. Examples would be the questions of whether all living organisms are genetically related, or whether there is extraterrestrial 7. And there has been no shortage of proposals about how to avoid Gettier problems. See Shope (1983). 8. For a very thorough discussion of this, see Alston (2005). My discussion in this section owes much to Alston (1993) and Alston (2005). Alston’s argument is that the continuing disagreements over ‘justification’ indicate that epistemologists have been chasing a phantom; there is not one such thing as ‘justification’. What there is are various different worthwhile epistemic states, epistemica desiderata, as he calls them. But Alston doesn’t draw any conclusions from this as to what knowledge is. But I do think it has such implications, and that is the point of this section.
160
life. So from continuing disagreements about an issue one can’t simply conclude that there is no one unique true answer to it. Still, such disagreements call for an explanation. Disagreement about the genetic relatedness of all organisms can, perhaps, be explained by the fact that the available evidence underdetermines the theory at hand and that those who endorse that theory have extra-evidential, for instance philosophical, reasons for it. Is there an explanation of that sort available for the abiding disagreements about knowledge? Can they be explained by the fact that each answer is underdetermined by the available data? Well, what are the data? I should think that the data are in a large measure our intuitions about cases, our data are the intuitions we have about whether or not the Nogot/Havit case, the Sheep in the Field case, the case of the Inflexible Mountain Climber, the case of the Aging Forest Ranger etc. present us with instances of knowledge. That is, those cases trigger our intuitions about whether or not we should apply the word ‘knowledge’ to the subject in each of the cases. But the fact is that our intuitions vary enormously. There are many cases where one epistemologist’s intuitions tell him that ‘knowledge’ can be applied, whereas another’s intuitions forbid him to do so. I take this fact to be evidence for the claim that ‘knowledge’ has many meanings; about as many as there are analyses of knowledge. This conclusion can be reinforced in the following way. Sometimes we don’t have an analysis available of a certain word, and still can zero-in on the object(s) it stands for. One way to do this is to work with paradigm cases and take the whole class of cases to be those that are sufficiently similar to the paradigms. Numerous words and concepts are successfully used on the basis of paradigms, for example, the word ‘cat’ and other natural kind terms. Even if one doesn’t have an analysis of the word ‘cat’ (that is, even if one doesn’t know what the necessary and sufficient condition are for the application of the word ‘cat’), one still can successfully apply the word ‘cat’ to a great number of animals on the basis of paradigm cases. Can’t we approach ‘knowledge’ in the same way, that is, on the basis of paradigm cases? If we succeed in this, the thesis that ‘knowledge’ has many senses becomes much less plausible. But if we can’t, it becomes more plausible. Now, the point I should like to make is that epistemologists even disagree over what the paradigm cases of knowledge are. For example, BonJour discusses imaginary cases of true belief formed by a highly reliable clairvoyant and he holds that they don’t constitute knowledge. But a hard-nosed
161
reliabilist as D.M. Armstrong takes them to be clear cases of knowledge. Where Armstrong’s intuitions urge him to apply ‘knowledge’ to the clairvoyant’s state of mind, BonJour’s intuitions forbid him to do so. My explanation for this is that Armstrong and BonJour use ‘knowledge’ in different senses. The same explanation could be given for diverging intuitions about other cases. So, I have been arguing for two related theses. (1) There isn’t one meaning of the word ‘knowledge’; knowledge is polysemous. (2) There isn’t just one type or state of mind that is the knowledge state; there are various different states that we refer to by means of the word ‘knowledge’. Where does this leave us with respect to contextualist intuitions about, for example, the bank cases? If the theses just mentioned are correct, then we have an alternative explanation of the contextualist’s intuitions. The contextualist himself explained them by defending the thesis that the standards of knowledge ascriptions vary with the context of the ascriptor. That explanation was based on the idea that ‘knows’ behaves like either an attributive term or an indexical. But this idea turned out to be very problematic. My explanation of the contextualist’s intuitions is that the word ‘knows’ is used in different senses. And to this I add that there may very well be a correlation between contexts and senses in which the word ‘knows’ is used. In bank case A, the context is such that ‘knows’ is used in a sense that differs from the sense in which it is used in case B. 9. Polysemy and the return of disquotation and invariantism If what I have said so far is correct, we can have disquotation back, while at the same time acknowledging the relevance of conversational contexts. I have acknowledged the relevance of conversational contexts in that I have suggested that the word ‘knows’ may be used in various meanings depending on the conversational context. In one context, a K-sentence may be used to ascribe to S that his true belief was reliably formed, whereas in another context the same K-sentence may be used to ascribe to S a belief held with certainty, and in yet another context a belief that is appropriately based on evidence, etc. But then we have disquotation back. For we can no longer talk about K-sentences without reserve. Whenever a K-sentence is used, we need
162
to make clear to ourselves in what meaning ‘knows’ is used in that Ksentence. Suppose, referring to the list in the previous section, we use indices to mark the various senses of ‘knows’. Then the K-sentence “S knows that p” must be taken either as “S knowsa that p”, or “S knowsb that p”, or “S knowsc that p”, and so forth. The truth-value of any of these Ks-sentences (the subscript s is for ‘specified’) is independent of the context of the ascriptor. “S knowsa that p” will be true iff S knowsa that p. This means we have disquotation back. If what I have said so far is correct, we also have back invariantism, i.e., the thesis that the truth conditions of K-sentences don’t vary with the context of the ascriptor—the position contextualists mean to oppose. To be sure, if what I have argued is correct, there is not just one standard that determines the truth-value of K-sentences. There are as many standards as there are meanings of ‘knowledge’. But, and this is the crucial point, each standard is invariant. The truth-value of Kindex sentences doesn’t vary with the context of the attributor.9 REFERENCES Alston, W. P. (1993). ‘Epistemic Desiderata’, Philosophy and Phenomenological Research 53, 527–551. — (2005). Beyond “Justification”. Dimensions of Epistemic Evaluation, Cornell University Press, Ithaca, NY. Blaauw, M. J. (2005). ‘How Intuitive is Contextualism?’, Grazer Philosophische Studien 69 (this issue). BonJour, L. (1985). The Structure of Empirical Knowledge, Harvard University Press, Cambridge, MA. — (1998). In Defense of Pure Reason, Cambridge University Press, Cambridge. — (2002). Epistemology. Classic Problems and Contemporary Responses, Rowman & Littlefield, Lanham. Chisholm, R. (1977). The Theory of Knowledge. Second edition, Prentice-Hall, Englewood Cliffs. Cohen, S. (1998). ‘Contextualist Solutions to Epistemological Problems: Scepticism, Gettier, and the Lottery’, Australasian Journal of Philosophy 76, 289–306. 9. For comments on earlier drafts of this paper, I want to thank Martijn Blaauw and, especially, Filip Buekens.
163
Conee, E & Feldman, R. (1985). ‘Evidentialism’, Philosophical Studies 48, 15–34. DeRose, K. (1992). ‘Contextualism and Knowledge Attributions’, Philosophy and Phenomenological Research 52, 913–929. — (1995). ‘Solving the Skeptical Problem’, Philosophical Review 104, 1–52. — (1999). ‘Contextualism: an Explanation and Defense’, The Blackwell Guide to Epistemology, (eds.) J. Greco & E. Sosa, Blackwell, Oxford. Foley, R. (1987). The Theory of Epistemic Rationality, Harvard University Press, Cambridge, MA. Ginet, C. (1975). Knowledge, Perception, and Memory, Reidel, Dordrecht. Goldman, A. (1979). ‘What is Justified Belief’, Justification and Knowledge. New Studies in Epistemology, (ed.) G. Pappas, Reidel, Dordrecht. Greco, J. (2000). Putting Skeptics in their Place, Cambridge University Press, Cambridge. Kaplan, D. (1989). ‘Demonstratives: An Essay on the Semantics, Logic, Metaphysics, and Epistemology of Demonstratives and Other Indexicals’, Themes from Kaplan, (eds.) Joseph Almog, John Perry & Howard Wettstein, Oxford University Press, Oxford. Kornblith, H. (2002). Knowledge and its Place in Nature, Oxford University Press, Oxford. Lehrer, K. (1990). Theory of Knowledge, Routledge, London. Plantinga, A. (1993). Warrant and Proper Function, Oxford University Press, Oxford. Schiffer, S. (1996). ‘Contextualist Solutions to Scepticism’, Proceedings of the Aristotelian Society 96, 317–333. Searle, J. (1979). Expression and Meaning. Studies in the Theory of Speech Acts, Cambridge University Press, Cambridge. Shope, R. (1983). The Analysis of Knowing, Princeton University Press, Princeton, NJ. Sosa, E. (1991). Knowledge in Perspective, Cambridge University Press, Cambridge. Stanley, J. (2004). ‘On the Linguistic Basis for Contextualism’, Philosophical Studies 108, 19–146. Stroud, B. (1984). The Significance of Philosophical Scepticism, Clarendon Press, Oxford. Unger, P. (1974). ‘An Argument for Skepticism’, Philosophical Exchange 1, 131–155.
164
Grazer Philosophische Studien 69 (2005), 165–182.
AVOIDING THE DOGMATIC COMMITMENTS OF CONTEXTUALISM Tim BLACK & Peter MURPHY California State University, Northridge & University of Tennessee Summary Epistemological contextualists maintain that the truth-conditions of sentences of the form ‘S knows that P’ vary according to the context in which they’re uttered, where this variation is due to the semantics of ‘knows’. Among the linguistic data that have been offered in support of contextualism are several everyday cases. We argue that these cases fail to support contextualism and that they instead support epistemological invariantism — the thesis that the truth-conditions of ‘S knows that P’ do not vary according to the context of their utterance.
According to the thesis of epistemological contextualism, the truth-conditions of sentences of the form ‘S knows that P’ and ‘S does not know that P’ vary according to the context in which they are uttered, where this variation is due to the semantics of ‘knows’. Among the linguistic data that have been offered in support of epistemological contextualism are cases that are ordinary in the sense that they involve a consideration neither of skeptical hypotheses nor of skeptical arguments. Both Stewart Cohen and Keith DeRose, contextualism’s two most prominent proponents, provide such cases. In a recent paper, DeRose goes so far as to claim that such cases provide the best grounds for accepting contextualism (see DeRose forthcoming, 1). In what follows, we argue that these cases do not support contextualism. In fact, they point in the direction of epistemological invariantism — the thesis that sentences of the form ‘S knows that P’ and ‘S does not know that P’ do not vary according to the context in which they are uttered.
1. The cases Let’s begin with the cases. Here, first, is DeRose’s bank case: [O]ne character (myself, as it happens), claims to know that a bank is open on Saturday morning in the “low standards” case. This belief is true, and is based on quite solid grounds: I was at the bank just two weeks ago on a Saturday, and found that it was open until noon on Saturday. Given the practical concerns involved — my wife and I are deciding whether to deposit our paychecks on Friday, or wait until Saturday morning, where no disaster will ensue if we waste a trip to the bank on Saturday only to find it closed — almost any speaker in my situation would claim to know the bank is open on Saturdays. And, supposing “nothing funny” is going on (there has not been a recent rash of banks cutting their Saturday hours in the area, etc.), almost all of us would judge such a claim to know to be true. But in the “high standards” case, disaster, not just disappointment, would ensue if we waited until Saturday only to find we were too late: We have just written a very large and very important check, and will be left in a catastrophically bad situation if the check bounces, as it will if we do not deposit our paychecks before Monday. (And, of course, the bank is not open on Sunday.) Given all this, my wife seems reasonable in not being satisfied with my grounds, and, after reminding me of how much is at stake, in raising, as she does, the possibility that the bank may have changed hours in the last couple of weeks. This possibility seems well worth worrying about, given the high stakes we are dealing with. Here I seem quite reasonable in admitting to her that I “don’t know” that the bank is open on Saturdays, and in endeavoring to “make sure.” Almost everyone will accept this as a reasonable admission, and it will seem true to almost everyone. (DeRose forthcoming, 5–6)1
Next is Cohen’s airport case: Mary and John are at the L.A. airport contemplating taking a certain flight to New York. They want to know whether the flight has a layover in Chicago. They overhear someone ask a passenger Smith if he knows whether the flight stops in Chicago. Smith looks at the flight itinerary he got from the travel agent and responds, “Yes I know — it does stop in Chicago.” It turns out that Mary and John have a very important business contact they 1. DeRose notes that he first offered the case in his (1992, 913). For more on the history of these cases, see the next footnote.
166
have to make at the Chicago airport. Mary says, “How reliable is that itinerary? It could contain a misprint. They could have changed the schedule at the last minute.” Mary and John agree that Smith doesn’t really know that the plane will stop in Chicago. They decide to check with the airline agent.” (Cohen 1999, 58)2
Each of these cases involves two conversational contexts. In one context, someone utters a sentence of the form ‘S knows that P’, and we are urged to take this as true. In the other context, however, someone utters a sentence of the form ‘S does not know that P’, that is, an utterance according to which the same person, S, in just the same circumstances at just the same time fails to know just the same proposition, P. And we are urged to take this utterance too as true. Consider Cohen’s case.3 The first context is one in which Smith says, “Yes I know — it does stop in Chicago.” The second is one in which Mary says, “Smith doesn’t really know that the plane will stop in Chicago.” Both utterances are about Smith’s belief that the plane stops in Chicago: Smith says that he knows that the plane stops in Chicago; and Mary says that he doesn’t know. There are three views about the standards that govern knowledge attributions (or attributions of a lack of knowledge) such as these. According to the high standard view, one high standard governs both attributions. Smith fails to meet this standard, and so when he says that he knows, he says something false. According to the low standard view, one low standard governs both attributions. Smith meets this standard, and so when Mary says that Smith doesn’t know, she says something false. According to the third view, contextualism, different standards govern the two attributions. The standard that governs Smith’s self-attribution is the low standard, and he truly attributes knowledge to himself. However, a higher standard is active in the second context, and Smith fails to meet that standard. Thus, according to contextualism, Mary says something true when she says that Smith fails to know. Cohen is primarily concerned to show that the low standard view is inadequate. To do this, he focuses on the disagreement between con2. Cohen reports in Cohen (1999, 83, fn. 3) that he first presented his case at a 1990 meeting of the APA. He also notes that Dretske (1981) presents similar cases. 3. DeRose points out that there are extra complications with the first case. These complications arise from the fact that the ascriptions are first-person in nature. See DeRose (forthcoming, section 5).
167
textualists and proponents of the low standard view — he focuses, that is, on whether Mary says something true in the second context. Cohen argues that, in that context, the operative epistemic standard is high, not low. He says: If we say that contrary to what both Mary and John presuppose, the weaker standard is correct, then we would have to say that their use of the word ‘know’ is incorrect. But then it is hard to see how Mary and John should describe their situation. Certainly they are being prudent in refusing to rely on the itinerary. They have a very important meeting in Chicago. Yet if Smith knows on the basis of the itinerary that the flight stops in Chicago, what should they have said? “Okay, Smith knows that the flight stops in Chicago, but still we need to check further.” To my ear, it is hard to make sense of that claim. Moreover, if what is printed in the itinerary is a good enough reason for Smith to know, then it is a good enough reason for John and Mary to know. Thus John and Mary should have said, “Okay, we know the plane stops in Chicago, but still, we need to check further.” Again it is hard to make sense of such a claim. (Cohen 1999, 58–59)
There are two arguments here. Both employ a claim that Cohen takes to be indisputable, namely, the claim that John and Mary need to check further. The first argument focuses on John and Mary’s third-person attribution (1) “Smith knows that the flight stops in Chicago, but still, we need to check further”, while the second argument focuses on their first-person attribution (2) “We know that the flight stops in Chicago, but still, we need to check further.” Cohen maintains that it is difficult to make sense either of (1) or of (2). Still, in each case, he takes the second conjunct to be indisputable. The problem, therefore, must lie with the first conjuncts — in the first case, with the claim that Smith knows that the flight stops in Chicago; and in the second case, with the claim that we (Mary and John) know that the flight stops in Chicago. Cohen’s arguments turn on two claims: that it is difficult to make sense of either (1) or (2); and that the second conjunct of each is true.
168
We focus on the former claim. Now, either of two reasons might explain why it is difficult to make sense of (1) and (2): each of those conjunctions could be false (i.e., semantically deficient) — in which case, given that their second conjuncts are true, their first conjuncts will be false — or they could simply be pragmatically deficient. Now, according to the low standard view, both Smith and John and Mary know that the plane stops in Chicago. Thus, proponents of the low standard view may not maintain that (1) and (2) are false (given that they agree with Cohen that John and Mary need to check further). Those who favor the low standard view must therefore maintain that (1) and (2) are pragmatically deficient. If it turns out, however, that (1) and (2) are not pragmatically deficient, then we must abandon the low standard view. In the next section, we examine Cohen’s argument against diagnosing (1) and (2) as pragmatically deficient. 2. Cohen’s argument against a diagnosis of pragmatic deficiency Cohen argues that we should diagnose neither (1) nor (2) as pragmatically deficient. Since his argument involving (1) is identical to the one involving (2), we focus here on (2). Cohen utilizes a simple test in order to determine whether suitable conjunctions are semantically deficient or pragmatically deficient. When faced with an odd-sounding conjunction, p & q, we should consider the conditional p ⊃ ~q. If this conditional is true, then the conjunction sounds odd because it is false. On the other hand, the conjunction, p & q, is only pragmatically deficient if the conditional, p ⊃ ~q, is false. To determine whether that conditional is false, we see whether we can cancel the implication from p to ~q. So, for example, there are circumstances in which my saying that Mario is a good soccer player suggests that he is not a great soccer player. In those circumstances, the conjunction Mario is a good soccer player, and he is a great soccer player will sound odd. Is this due to a pragmatic deficiency, or to a semantic one? If it is a pragmatic deficiency, then the implication from Mario is a good soccer player to Mario is not a great soccer player will be cancelable. And it is cancelable by saying, for example, “Mario is a good soccer player — in fact, he is a great soccer player.” It follows that when it sounds odd to say, “Mario is a good soccer player, and he’s a great soccer player,” it does so because that conjunction is pragmatically, rather than semantically, deficient.
169
Employing this simple test, then, Cohen maintains that (2) is not pragmatically deficient. Consider the implication from the first conjunct of (2) — we know that the flight stops in Chicago — to the negation of its second conjunct — we need not check further. We cannot cancel this implication, for, as we have already noted, it sounds odd to say, “We know that the flight stops in Chicago, but we do need to check further.” Thus, whenever the conjunction We know that the flight stops in Chicago, but we need to check further sounds odd, it does so because that conjunction is semantically, rather than pragmatically, deficient. This suggests, therefore, that we ought to abandon the low standard view. For there is no way to maintain, as proponents of the low standard view must, that (1) and (2) are pragmatically deficient. This argument succeeds, however, only if it is true that we know that the flight stops in Chicago semantically entails we need not check further. That there is such an entailment is a dogmatic thesis. For if one’s knowing that P entails that one should seek no further evidence for P, then a form of dogmatism is true, a dogmatism according to which knowers might need to revise their beliefs in accordance with incoming further evidence, but they are under no obligation to seek further evidence. Knowers are therefore permitted to be apathetic dogmatists — they are permitted to be indifferent toward the active pursuit of new evidence.4 3. Epistemic directives and other cases Is apathetic dogmatism defensible? To answer this question, we need to take a closer look at claims of the form ‘We ought to seek more evidence that P’. We call claims of this and similar forms epistemic directives. Epistemic directives are statements that assign a deontic status — e.g., obligatory, permissible, impermissible — to someone’s relationship to some body of evidence — e.g., that they seek that evidence, or ignore it, or request it, or demand it. For example, each of the following is an epistemic directive: It is obligatory that S seek more evidence that P; it is obligatory that S ignore evidence that P; it is permissible that S seek more evidence that P; it is permissible that S ignore evidence 4. For more on apathetic and other forms of dogmatism, see Murphy (manuscript). Harman (1973) credits Kripke with first formulating the basic paradox of dogmatism.
170
that P; it is impermissible that S to seek more evidence that P; and it is impermissible that S ignore evidence that P. Now, to retrace our steps: If, as Cohen argues, ‘S knows that P’ cannot felicitously be conjoined with ‘S ought to seek more evidence that P’, then there is a semantic entailment, and not just a pragmatic implicature, from ‘S knows that P’ to ‘It is not the case that S ought to seek more evidence that P’. But by the usual rules of deontic implication, ‘It is not the case that S ought to ij’ entails ‘It is permissible that S refrain from ijing’. If this is correct, then Cohen is committed to the claim that ‘S knows that P’ entails ‘It is permissible that S refrain from seeking further evidence that P’. This means that Cohen is committed to dogmatism, the thesis that it is permissible not to seek further evidence that some proposition is true. But such dogmatism, as we will now argue, is mistaken. Given that our arguments are sound, we should reject Cohen’s argument against the low standard view, for that argument is based on dogmatism. Is there, as Cohen alleges, always something infelicitous about claims of the form ‘S knows that P, but S ought to acquire more evidence that P’? Consider a case in which I learn by testimony that someone knows something — I read in a trustworthy text, for example, that Einstein knows that E = mc2. Skepticism about testimony aside, I know that Einstein knows that E = mc2. Still, it might be that I ought to acquire more evidence that E = mc2. Perhaps, for example, I will be asked for evidence that E = mc2 on an upcoming physics exam, and the professor will not accept my testimonial evidence that I read in a trustworthy text that E = mc2 is true and that Einstein knows that E = mc2. In light of this, it seems perfectly acceptable to say, (3) “Einstein knows that E = mc2, but I ought to acquire more evidence that E = mc2.” There is nothing infelicitous about this utterance. While (3) conjoins a claim that someone else knows something with a claim that I ought to acquire more evidence, there are other counterexamples that are not like this. Consider a second case, one that is similar to (2) in that it conjoins two first-person claims. Under the assumption of fallibilism, understood as the view that the smallest degree of evidence sufficient for meeting the evidence condition on knowing that P need not entail that P is true, there is such a thing as inductive knowledge.
171
Suppose, then, that a scientist has inductive evidence for her belief that P, and that, on the basis of that evidence, knows that P. However, if she acquires further evidence that P, she increases her chances of receiving a research grant. She says (4) “I know that P, but I ought to acquire more evidence that P.” Here again, there seems to be nothing infelicitous about this utterance. Even if fallibilism is false, one might nevertheless have good reason to acquire further evidence that P when one knows that P. For example, a mathematician might have a deductive proof that P, a proof on whose basis she knows that P. As is frequently the case, mathematicians seek multiple proofs, and so our mathematician might seek further evidence in the form of additional proofs that P. Suppose moreover that she wishes to have her findings published in a journal that requires multiple proofs. She says, (5) “I have a proof that P, but I ought to acquire more evidence that P.” Again, her utterance seems perfectly felicitous. These three cases suggest that the second conjuncts of (1)–(5), each of which is of the form ‘S ought to acquire more evidence that P’, are elliptical. For in each case it makes sense to ask, for what end ought S to acquire further evidence that P? Is it for the sake of winning a research award? Or for the sake of doing well on an exam? The thought here is quite intuitive: When we seek further evidence, we often do so in order to achieve certain goals, some of which are epistemic, and some of which are not. This certainly applies in John and Mary’s situation. They ought to check further in order to increase the probability that they will achieve their important goal of meeting their business contact. Suppose for a moment that John and Mary know, on the basis of Smith’s testimony, that the plane stops in Chicago. This knowledge would be inductive rather than deductive. So, whatever justifies their belief fails to entail that their belief is true. This means that John and Mary can acquire more evidence for their belief that the plane stops in Chicago. Putting this in terms of probability — of some sort, at least — we can say that
172
initially, John and Mary’s belief is probable, given their total evidence at that time, to some degree d. Yet d falls short of a probability of 1, which means that John and Mary’s acquiring additional evidence will increase the degree to which their belief is probable, given their new total evidence, and therefore will help them to achieve their goal.5 On this reading, then, (1) and (2) are elliptical for (1*) “Smith knows that the flight stops in Chicago, but still, we need to check further for the sake of meeting our business contact.” (2*) “We know that the flight stops in Chicago, but still, we need to check further for the sake of meeting our business contact.” At this point, someone might object that if knowledge entails truth, and one knows that P, then it follows that one has epistemic access to the truth as to whether P. So why check further? Why acquire further evidence? This objection goes too far, however. For consider mere true belief. If I merely truly believe that P, then I have epistemic access (in one sense) to the truth as to whether P. It is obvious, however, that those who merely truly believe that P have good reason to acquire further evidence that P. For one thing, they should acquire further evidence that P for the sake of knowing that P. We conclude that apathetic dogmatism is not defensible, and that the possibility remains open that (1) and (2) are pragmatically, rather than semantically, deficient. Still, someone might suggest that the second conjuncts of (1*) and (2*) should have the following form: We ought to acquire more evidence that P for the sake of knowing that P. This response, however, begs the question against those who favor the low standard view. For if John and Mary ought to acquire more evidence for P in order to know that P, then they don’t already know that P. Once we determine that contextualism is correct, we may then maintain that the second conjuncts of (1*) and (2*) are elliptical for We ought to acquire more evidence that P for the sake of knowing that P. Until then, however, we should not beg the question against proponents of the low standard view. 5. In fact, even in cases where someone has deductive knowledge of some proposition, as with a mathematical proof that P, it seems to us that there is a real sense in which the person can acquire yet more evidence for believing that P—for example, the person might construct a second independent proof that P.
173
Third, someone might object that we have provided counterexamples only against a certain form of apathetic dogmatism, which we might call Strong Apathetic Dogmatism, or (SAD) S’s knowing that P entails that it is not the case that S ought to seek further evidence that P. One who objects in this way might go on to maintain that Cohen’s view needs only a weaker form of dogmatism, against which our counterexamples are ineffective. Call this weaker form Weak Apathetic Dogmatism, or (WAD) S’s knowing that P entails that it is not the case that S ought to seek further evidence that P for the sake of ensuring (or making sure) that P is true. Given this distinction, someone might argue that whether SAD or WAD applies in a particular case depends on whether, in that case, S’s goal can be achieved only if P is true. In particular, WAD applies only in cases in which S’s goal can be achieved only if P is true, while SAD applies in cases in which S’s goal can be achieved even when P is false. Now, WAD does not apply, an objector might argue, in the cases surrounding (3), (4), and (5). For, in the case of (3), the goal is to do well on the exam, and that goal can be achieved even if P is false — whether one will do well on the exam depends on whether one provides certain kinds of evidence, not on whether P is true. In the case of (4), the goal is to get a research grant, and that goal can be achieved even if P is false — whether one will get the grant depends on whether one provides sufficient evidence, not on whether P is true. In the case of (5), the goal is to get one’s findings published in a particular journal, and that goal can be achieved even if P is false — whether one’s findings will be published in that journal depends on whether one provides another proof, not on whether P is true. Thus, we cannot use the cases surrounding (3), (4), and (5) to show that WAD is defective. We can use those cases only against SAD. Moreover, in Cohen’s Airport Case, Mary and John’s goal is to meet their business contact in Chicago today, and they can achieve that goal only if P is true, that is, only if it is true that the flight will stop in Chicago today. Cohen might therefore urge that (2) is elliptical for
174
(2**) We know the flight stops in Chicago, but still, we ought to check further for the sake of making sure that it is true that the flight stops in Chicago. In order to hold that (2**) is false, Cohen need rely only on WAD. And Cohen might maintain that since none of our counterexamples tell against WAD, we have done nothing to show that the conjunction in the Airport Case is not semantically deficient.6 We reject the idea that WAD does not apply in the cases surrounding (3), (4), and (5), and thus the idea that we cannot use those cases to show that WAD is defective. For, just as in (2), the truth of P is important in each of (3), (4) and (5). In (2), Mary and John’s principal aim is to seek evidence that will help them to make sure that their belief (that the plane stops in Chicago) is true. Yet in spite of the fact that this is their principal aim, we describe the end of further inquiry in practical terms — they should check further for the sake of meeting their business contact in Chicago. Now, each of the other cases that we provide — those surrounding (3), (4), and (5) — share both of these features with the Airport Case. First, in each of the cases surrounding (3), (4), and (5), what’s principally important for the epistemic agent as the end of further inquiry is the acquisition of evidence that will, when added to the evidence she already possesses, make it more likely that her belief is true, thus helping her to make sure (or to become more sure) that her belief is true. Moreover, just as in the Airport Case, in each of the cases surrounding (3), (4), and (5), the acquisition of this evidence helps the subject to achieve her practical end. There is, then, no distinction along these lines to be drawn between (2) on the one hand and (3), (4), and (5) on the other. We conclude, then, that WAD does apply in the cases surrounding (3), (4), and (5), and we may use those cases as we have in order to show that WAD is defective. We have now seen that we should avoid apathetic dogmatism, and that we should consequently leave open the possibility that (1) and (2) are pragmatically deficient. In the next section, we explain how epistemological invariantism helps us both to provide a pragmatic explanation of the infelicity of (1) and (2), and to avoid the dogmatism of contextualism. 6. Thanks to Doug Portmore for voicing this concern and for formulating both (SAD) and (WAD).
175
4. How to avoid contextualism’s dogmatism To avoid contextualism’s dogmatism, we should reject the claim that if it sounds odd to say, “S knows that P, but S needs to check further,” then the implication from S knows that P to S need not check further is a semantic entailment and not simply a pragmatic implicature. Once we see that there are counterexamples to this conditional — that is, cases in which it sounds odd to say, “S knows that P, but S needs to check further,” and in which we can cancel the implication from S knows that P to S need not check further — we open the door to a rejection of Cohen’s argument against the low standard view. We also revive the possibility of a pragmatic explanation of the impropriety of “S knows that P, but S needs to check further.” How is the invariantist to explain the pragmatic impropriety of that assertion? Here’s our suggestion: The conjunction of ‘S knows that P’ and ‘S needs to check further’ sounds odd because its second conjunct pragmatically implicates that S needs to check further for the sake of knowing that P.7 The conjunction therefore both says that S knows that P and pragmatically implicates that S needs to check further in order to know that P. Moreover, the second of these claims suggests that S does not know that P. For if S needs to check further in order to know that P, then S does not already know that P. But this claim and the claim made by the conjunction’s first conjunct are contraries — it cannot be true both that S knows that P and that S does not know that P. According to our suggestion, then, the assertion, “S knows that P, but S needs to check further” sounds odd because it is pragmatically deficient — in making that assertion, we say that S knows that P, and we pragmatically implicate the contrary claim that S does not know that P. Moreover, we suggest that we can cancel the implication from S knows that P to S need not check further. To do so, we need only to assert that S does need to check further, but for some sake other than knowing that P, for example, for the sake of receiving a research grant, or for the sake of meeting a business contact. We might say, “S knows that P, but S needs to check further — not, mind you, for the sake of knowing that P, since she already knows that P, but for the sake of receiving a research grant.” This opens the door to a rejection of Cohen’s 7. Perhaps it does so because knowledge — or, in general, some epistemic state—is mentioned so prominently in the first conjunct.
176
argument against the low standard view, and it breaths life back into epistemological invariantism. We should also address DeRose’s Bank Case. DeRose maintains that in the (alleged) high-standards context of that case, he seems quite reasonable in admitting to his wife that he doesn’t know that the bank is open on Saturdays, and in endeavoring to make sure that it is open. In the high-standards context, then, DeRose’s utterance, “I don’t know that the bank is open on Saturdays, and I need to make sure that it is,” would be perfectly felicitous. We grant that in this same context if DeRose were to utter “I know that the bank is open on Saturday, but I need to make sure that it is,” his utterance would sound odd. But, again, we claim that the utterance would be pragmatically, not semantically, deficient. Moreover, the second conjunct of this utterance is sufficiently different from the second conjuncts of Cohen’s examples — each of which has the form ‘S needs to check further’— to warrant attention. For it seems that ‘S needs to make sure that P’ is not elliptical in the way that ‘S needs to check further’ is. In fact, in the case of ‘S needs to make sure that P’, the end of further checking — namely, to make sure that P— seems to be written directly into the conjunct. So how can the epistemological invariantist explain the pragmatic impropriety of DeRose’s conjunction? Here’s our suggestion: The conjunction of ‘S knows that P’ and ‘S needs to make sure that P’ sounds odd because its second conjunct pragmatically implicates, rather than semantically entails, that S does not know that P. Here again, then, this claim and the claim made by the conjunction’s first conjunct are contraries — it cannot be true both that S knows that P and that S does not know that P. Thus, according to our suggestion, the assertion, “S knows that P, but S needs to make sure that P” sounds odd because it is pragmatically deficient. The natural and obvious worry, though, is this: It seems to be an epistemological maxim that S knows that P only if S is sure that P (that is, only if S believes that P sufficiently confidently). And given that S needs to make sure that P, it follows from this alleged maxim that S does not know that P. This suggests, contrary to our proposal, that the assertion, “S knows that P, but S needs to make sure that P” is semantically deficient. But is it indeed true that S knows that P only if S is sure that P? It seems that there are plenty of counterexamples to this alleged maxim. It might be, for example, that Ashley knows the answer to a question on
177
her history exam — say, that Washington surrendered Fort Necessity in 1754 — even though she does not confidently believe that Washington surrendered Fort Necessity in 1754. She knows that Washington surrendered the fort, but she nevertheless isn’t sure whether he did. Suppose, however, that Ashley’s lack of confidence generates in her the belief that Washington might not have surrendered Fort Necessity in 1754. Will her having this belief keep her from knowing that he surrendered it? We think not. We are not averse to maintaining that Ashley can know that Washington surrendered Fort Necessity in 1754 even though she believes that he might not have. Here’s why: Either Ashley has evidential grounds for her belief that Washington might not have surrendered Fort Necessity in 1754, or she doesn’t. Suppose that she doesn’t — her belief is simply a product of her doubts and based on no evidence whatsoever. In this case, we don’t see why we should be forced to say that Ashley’s believing that Washington might not have surrendered Fort Necessity in 1754 is incompatible with her knowing that he did surrender it. This case therefore seems to reveal no incompatibility between Ashley’s knowing that Washington surrendered Fort Necessity in 1754 and her not being sure whether he did. Suppose, on the other hand, that Ashley does have evidential grounds for her belief that Washington might not have surrendered Fort Necessity in 1754. In this case, Ashley’s evidence might play a defeating role, in which case her evidential grounds for the belief that Washington surrendered the fort might no longer be sufficient for knowledge. Ashley, then, fails to know that Washington surrendered Fort Necessity in 1754. It is false both that she knows that Washington surrendered Fort Necessity in 1754 and that she is sure that he did. Thus, this case too reveals no incompatibility between, on one hand, Ashley’s knowing that Washington surrendered Fort Necessity in 1754 and, on the other, her not being sure whether he did. There is another, more general and perhaps more convincing way of calling into question the alleged maxim. To do this, we need first to modify Cohen’s airport case: In that case, it seems that Mary, for one reason or another, is not sure whether the plane stops in Chicago. Thus, given the alleged maxim, Mary does not know that the plane stops in Chicago. Moreover, it ought to seem to Mary that Smith doesn’t know that the plane stops in Chicago. After all, (i) Mary (recognizes that she) doesn’t know, and (ii) she recognizes that Smith has no evidence that she doesn’t have.
178
But is this enough to make it seem to Mary that Smith doesn’t know? That is, will (i) and (ii) do the trick by themselves? It seems that they won’t, and that we should supplement (i) and (ii) with the following: Mary believes that (iii) evidence is all that matters epistemically. Suppose, though, that Mary is epistemologically informed and believes that (iii) is false. Suppose, that is, that she believes that our alleged maxim is indeed a maxim. Suppose, furthermore, that Mary recognizes that Smith has sufficient evidential grounds for his belief, as well as that Smith is free from the uncertainties that plague her (and from any other relevant uncertainties). This means, according to the alleged maxim, that Smith’s epistemic position vis-à-vis the proposition that the plane stops in Chicago is better than Mary’s. May she therefore consult Smith in order to allay her doubts or, more generally, in order to improve her epistemic position? It seems that she may not (see Cohen 2000, 95–97).8 This point, as it turns out, is significant, for it’s the centerpiece in an argument against the alleged maxim. Here’s the argument: (6) Suppose that both evidence and confidence matter epistemically (i.e., S knows that P only if S has sufficient evidential grounds and believes that P sufficiently confidently). 8. Of his Airport case, Cohen says, “When Smith says, “I know…”, what he says is true given the weaker standard operating in that context. When Mary and John say “Smith does not know …”, what they say is true given the stricter standard operating in their context” (Cohen 2000, 97). Yet he also maintains that “Mary and John agree that Smith doesn’t really know that the plane will stop in Chicago on the basis of the itinerary. They decide to check with the airline agent” (Cohen 2000, 95). In this case, then, even though, according to Cohen, Smith is in a better epistemic position than Mary and John — he knows, but they do not — Mary and John may not consult him regarding whether the plane will stop in Chicago; Mary and John must consult the airline agent instead. (Compare DeRose, 2004, 348: Suppose that Thelma is being interrogated by the police in a context in which the epistemic standards have been raised. In this context, Thelma claims not to know that John was at the office on the day in question. Meanwhile, Thelma’s friend, Louise, who has the same evidence as Thelma for John’s being at the office, but who is in a low-standards conversational context, claims to know that John was at the office. “And suppose,” DeRose says, “that Thelma is somehow aware of all this about Louise’s context. Still, in Thelma’s “high-standards context,” if Thelma is counting herself as a non-knower, then, when she is considering Louise as a potential informant, she will likewise describe Louise as a non-knower, regardless of Louise’s conversational context” (DeRose 2004, 348). In this case, too — one that is much more similar to the case I construct, since Thelma knows so much about how things stand epistemically with Louise — Thelma may not consult Louise even though (she recognizes that) Louise is in a better epistemic position than she is.)
179
(7) There is a context, C2, in which S and S* have identical (and individually sufficient) evidential grounds for believing that P. Moreover, in C2, S* believes that P sufficiently confidently (and knows that P), but S does not believe that P sufficiently confidently (and hence does not know that P). (8) Thus, S*’s epistemic position vis-à-vis P is better than S’s. (9) If S*’s epistemic position vis-à-vis P is better than S’s, then S may consult S* regarding P. (10) Thus, in C2, S may consult S* regarding P. (11) But S may not consult S* regarding P in C2. (12) Thus, it is not the case that both evidence and confidence matter epistemically. (13) It is clear that evidence matters epistemically. (14) Thus, confidence does not matter epistemically. This suggests that our alleged maxim is false, and therefore that it cannot stand in the way of a pragmatic explanation of the infelicity of “S knows that P, but S needs to make sure that P.” Still, if we are to provide a pragmatic explanation of that infelicity, we need to be able to cancel the implication from S knows that P to S need not make sure that P. We suggest that in order to cancel that implication, we need only to assert that S does need to make sure that P, perhaps for the sake of making herself confident that P to a satisfying degree. Here again, let’s allow that Mary has inductive knowledge, on the basis of Smith’s testimony, that the plane stops in Chicago. Given this, whatever justifies her belief fails to entail that it’s true. So, Mary can acquire more evidence for her belief that the plane stops in Chicago. Moreover, it might be that Mary acquires enough evidence to know that P long before she acquires enough evidence to make herself confident to a satisfying degree.9 To cancel the relevant implication, then, we might say, for example, “S knows that P, but S needs to make sure that P in order to make herself confident that P to a satisfying degree.” Once again, this opens the door to a rejection of arguments against the low standard view, and it revitalizes epistemological invariantism. 9. This allows us to say, for example, that while the standards for knowledge are invariant across contexts, the standards for being confident to a satisfying degree vary from context to context, perhaps even on the basis of contextual features highlighted by contextualists, e.g., “speaker intentions, listener expectations, presuppositions of the conversation, salience relations, etc.” (Cohen 1999, 61).
180
5. Conclusion An examination of the linguistic data that have been offered in support of epistemological contextualism by its most prominent proponents reveals that contextualists are committed to apathetic dogmatism, the claim that knowers may refrain from seeking further evidence. We have shown, however, that apathetic dogmatism is indefensible. Furthermore, given that dogmatism is an essential element of contextualists’ argument for the claim that certain odd-sounding conjunctions are semantically deficient, we have reason to set those arguments aside. This opens the door to a pragmatic explanation of the fact that those conjunctions sound odd, and this is just the sort of explanation that epistemological invariantists need to provide. We have also supplied, on behalf of invariantists, pragmatic explanations of the infelicity of certain conjunctions, explanations that also make it clear just how we can cancel certain relevant implications. Thus, since invariantists can provide the pragmatic explanations that they must provide, and since contextualists’ semantic explanations seem to force them to adopt a gravely problematic form of dogmatism, we conclude that invariantism has the upper hand on contextualism.
REFERENCES Cohen, S. (1991). ‘Scepticism, Relevance, and Relativity’, Dretske and his Critics, (ed.) B. McLaughlin, 17–37, Basil Blackwell, Oxford. — (1999). ‘Contextualism, Skepticism, and the Structure of Reasons’, Philosophical Perspectives 13, 57–89. — (2000). ‘Contextualism and Skepticism’, Philosophical Issues 10, 94– 107. DeRose, K. (1992). ‘Contextualism and Knowledge Attributions’, Philosophy and Phenomenological Research 52, 913–929. — (1999). ‘Contextualism: An Explanation and Defense’, The Blackwell Guide to Epistemology, (eds.) John Greco and Ernest Sosa, 187–205, Blackwell, Malden, MA. — (2000). ‘Now You Know It, Now You Don’t’, Proceedings of the Twentieth World Congress of Philosophy, Vol. V, Epistemology, 91–106. — (2004). ‘The Problem with Subject-Sensitive Invariantism’, Philosophy and Phenomenological Research 68, 346–350.
181
— (forthcoming). ‘The Ordinary Language Basis for Contextualism and the New Invariantism,’ The Philosophical Quarterly, online at http:// pantheon.yale.edu/~kd47/OLB.pdf. Dretske, F. (1981). ‘The Pragmatic Dimension of Knowledge’, Philosophical Studies 40, 363–378. Reprinted in Dretske (2000), 48–63. — (2000). Perception, Knowledge and Belief: Selected Essays, Cambridge University Press, Cambridge. Harman, G. (1973). Thought, Princeton University Press, Princeton, NJ. Murphy, P. (manuscript). ‘New Paradoxes of Dogmatism’.
182
Grazer Philosophische Studien 69 (2005), 183–205.
A CONTEXTUALIST SOLUTION TO THE PROBLEM OF EASY KNOWLEDGE Ram NETA University of North Carolina Summary Many philosophers hold some verion of the doctrine of “basic knowledge”. According to this doctrine, it’s possible for S to know that p, even if S doesn’t know the source of her knowledge that p to be reliable or trustworthy. Stewart Cohen has recently argued that this doctrine confronts the problem of easy knowledge. I defend basic knowledge against this criticism, by providing a contextualist solution to the problem of easy knowledge.
I. Basic knowledge and the problem of easy knowledge Many philosophers claim that situations of the following kind are possible: S knows that p, and S obtains this knowledge by means of source X, but S doesn’t know that X is reliable or trustworthy.1 For example, a child might know that his mother is home, and he might know this by seeing his mother at home, but all of this can be true even if he doesn’t know that his vision is reliable or trustworthy. Again, I might know by means of testimony and memory that I was born on a certain date, even if I don’t know that the testimony or the memory on which this knowledge is based is reliable or trustworthy. More exotically, an amnesiac might see a tree in front of her, and so come to know that there is a tree in front of her, even if she doesn’t recall any experiences on the basis of which she could come to know that her visual perception is at all reliable or trustworthy. 1. Virtually all externalists believe this, and so do many internalists. The view is criticized by Sellars (1963) and BonJour (1978), but those criticisms are rebutted in Alston (1983).
I will follow Cohen (2002) in using the phrase “basic knowledge” to denote all such knowledge — i.e., all knowledge possessed without knowledge of the reliability or trustworthiness of its source. What I’ll call the “doctrine of basic knowledge” is simply the view that we can possess basic knowledge, i.e. we can possess knowledge without also knowing its source to be reliable or trustworthy. In other words: (The Doctrine of Basic Knowledge) For some subject S, some source x, and some proposition p: S knows that p, by means of source x, and S doesn’t know that x is reliable or trustworthy. The doctrine of basic knowledge is widely (but not universally) accepted among epistemologists. Recently, different versions of the doctrine of basic knowledge have been subject to a particular objection — roughly, that they do not have the resources to explain why it’s impossible to obtain knowledge by means of certain kinds of inferences, which I will henceforth call “easy knowledge” inferences. (I’ll characterize these inferences below.) Cohen (2002) refers to this objection, in its most general form, as “the problem of easy knowledge”.2 In this paper, I defend the doctrine of basic knowledge (the various versions of which I will henceforth refer to as “basic knowledge theories”) from this objection. More specifically, I offer a contextualist solution to the problem of easy knowledge. Many philosophers have already proposed solutions to the problem of easy knowledge.3 Almost all of these proposed solutions reject a presupposition of the problem, namely, that we generally cannot gain knowledge that p by means of an easy knowledge inference to the conclusion that p. But I think that we should not simply reject this presupposition. As I argue elsewhere, if we simply reject this presupposition, then we will not be able to distinguish, in a principled way, easy knowledge inferences from certain other inferences that are not (on anybody’s view) warrant-transmitting.4 My contextualist solution to the problem of easy knowledge does not give up the doctrine of basic knowledge, 2. Fumerton (1995, 175-6) and Vogel (2000) both level this line of criticism against reliabilism. 3. See, for instance, Van Cleve (2003), Davies (2003), Klein (2004), Pryor (2004), Markie (2005), and Bergmann (2004). 4. I argue for this in Neta (forthcoming b).
184
and does not simply reject the presupposition that the easy knowledge inferences cannot transmit warrant from premises to conclusion. Before setting out the problem of easy knowledge, we must first get clearer about what the doctrine of basic knowledge says. In stating the doctrine of basic knowledge, I followed common practice in talking about “sources” of knowledge. But what is a “source” of knowledge? Some philosophers use the term “source” to refer to the cognitive faculties that produce knowledge, or to the evidence on which knowledge is based. But what, if anything, is the genus of which cognitive faculties and evidence are two species? Different proponents of basic knowledge theories might offer different answers to this question. For the purposes of stating the doctrine of basic knowledge in order to articulate the problem of easy knowledge, we can lay down the following general constraint on what we will count as a “source” of knowledge: If X is a “source” of S’s knowledge that p, then all of the following conditions obtain: (i) S believes that p, and (ii) X is a reason that S has for believing that p, and (iii) S believes that p for, at least, that very reason (X)5, and (iv) X is a token of a type T1, and p a token of type T2, such that, in general, tokens of T1 are reasons for people to believe propositions of type T2, and (v) S can know that (i)–(iv) are all true, without X being a source of that knowledge. I’ll summarize conditions (i)–(iii) by saying that X is “a reason why” S believes that p. I can then say that X is a source of S’s knowledge that p only if X is a reason why S believes that p, and things like X are reasons for people to believe things like p, and finally S can know all of this without X being a source of that knowledge. When I discuss the problem of easy knowledge below, it will become clear why it is that, for the purposes of explaining the doctrine of basic knowledge in a way that will make clear why it seems to be subject to the problem of easy knowledge, we should understand a “source” of knowledge as something that satisfies the five conditions listed above. 5. S’s believing that p for that very reason is compatible with S’s believing that p for other reasons as well — our believings are sometimes overdetermined.
185
To illustrate: If Fred’s testimony is a source of Jones’s knowledge that Mabel will be late to the party, then (P1) Fred’s testimony is a reason why Jones believes that Mabel will be late to the party. And Jones can know that P1 is true, without Fred’s testimony being Jones’s reason to believe that P1 is true. (Presumably, if Jones believes that P1 is true, that is because Jones can recognize, upon reflection, what reasons move him to believe that Mabel will be late to the party.) Again: If the smell of the stew is a source of Smith’s knowledge that there’s too much paprika in the stew, then (P2) The smell of the stew is a reason why Smith believes that there’s too much paprika in the stew. And Smith can know that P2 is true, without the smell of the stew being Smith’s reason to believe that P2 is true. (Once again, if Smith believes that P2 is true, that is because Smith can recognize, upon reflection, what reasons move him to believe that the stew contains too much paprika.) But contrast: If reasoning is a source of my knowledge that 23 + 57 = 80, then (P3) My reasoning is a reason why I believe that 23 + 57 = 80. And I can know that P3 is true, without reasoning being my reason to believe that P3 is true. Since, in the standard case, I will not know that P3 is true without reasoning being a reason why I believe that P3 is true, it follows that, in the standard case, reasoning does not satisfy the preceding characterization of a “source” of knowledge. For the purposes of this problem, we must understand sources to be individuated more finely. When we state the problem of easy knowledge below, it will become clear that, if an epistemologist does not think of sources as satisfying the complicated constraint that I’ve just laid down, then her view will not be even apparently subject to the problem of easy knowledge. We should therefore understand the problem of easy knowledge as a problem that is posed for basic knowledge theories that conceive of sources as satisfying the complicated constraint above.
186
Now that we’ve clarified the doctrine of basic knowledge, what is the problem of easy knowledge? Cohen presents it as having two versions, which I consider in sections II and III below. II. The first problem of easy knowledge: one-premise deductions against defeaters The first problem of easy knowledge is generated by the following plausible epistemic closure principle: One-Premise Closure: If S knows that p, and S competently deduces q from p without losing her knowledge that p, then S knows that q.6 Notice the qualification that S retain her knowledge that p while deducing that q. This is important. If, while performing the deductive inference, S comes to lose confidence in p, or comes to lose her justification for believing that p, she will not be able to earn knowledge that q on the basis of the deduction. This point is distinct from the point that S’s deduction must be “competent”. S could retain her knowledge that p while making the deduction, but if S bungles the deduction in some way then, even if she somehow still ends up at the conclusion that q, the conclusion doesn’t count as knowledge. Bungled deductions do not transmit knowledge from premise to conclusion — not even if the conclusion reached happens to be one that actually follows from the premise.7 Now, how does One-Premise Closure pose a problem for basic knowledge theories? Let‘s consider a case that Cohen describes: ‘Suppose my son wants to buy a red table for his room. We go in the store and I say, “That table is red. I’ll buy it for you.” … he worries, “Daddy, 6. Here I state the version of Closure similar to one recently defended by Hawthorne (2004). (Also see Williamson (2000, 117)) It differs from the version of Closure put forward in Cohen (2002), but I think that the problems of easy knowledge can be more clearly stated using Hawthorne’s version of Closure than Cohen’s own version. 7. Harman (1973, 157) questions whether there is any such psychological act as deductive inference. He points out that, when we construct a deductive argument, we typically do not perform any psychological act of deductive inference. He is right about that. But it’s also true that, when we follow a deductive argument that’s presented to us, we typically do perform a psychological act of deductive inference.
187
what if it’s white with red lights shining on it?” I reply, “Don’t worry — you see … it is red, so it’s not white but illuminated by red lights.” Surely he should not be satisfied with this response. Moreover I don’t think it would help to add, “Now I’m not claiming that there are no red lights shining on the table, all I’m claiming is that the table is not white with red lights shining on it.’8
Now, is Cohen right to claim that the reasoning of the father in his example (whom I’ll henceforth call “Stewart”) is bad reasoning — more specifically, that such reasoning does not provide Stewart with knowledge that the table is not white with red lights shining on it? To answer this question, let’s consider how Stewart could have achieved his knowledge that the table is red. Suppose that Stewart achieved knowledge that the table is red by means of the following deductive inference: The table is not white with red lights shining on it. If the table is not white with red lights shining on it, then it is red. The table is red. If Stewart achieved his knowledge that the table is red by means of this deductive inference, and he achieved his knowledge of the premises of the inference in some other way altogether, then the problem with the reasoning described in Cohen’s example is that such reasoning cannot provide Stewart with knowledge that the table is not white with red lights shining on it, since Stewart had knowledge of that fact already. (And if he lost that bit of knowledge while making the deduction that Cohen describes, then the deduction won’t give him knowledge of the conclusion, since it rests on a premise that he doesn’t know to be true.) More generally, if Stewart achieved his knowledge that the table is red on the basis of his knowledge that the table is not white with red lights shining on it, then of course the reasoning described in Cohen’s example cannot be the way that Stewart acquires knowledge that the table is not white with red lights shining on it. If, however, Stewart achieved his knowledge that the table is red in some other way — for instance, simply by virtue of seeing the table looking red in normal circumstances — then again, it seems that Stewart cannot acquire, merely by means of the reasoning described, knowledge 8. Cohen (2002).
188
that the table is not white with red lights shining on it. For, if all he has to go on is that the table looks red, then he cannot achieve knowledge that the table is not white with red lights shining on it. But if Stewart has basic knowledge that the table is red, and OnePremise Closure is true, and he can competently deduce the proposition “the table is not white with red lights shining on it” from the proposition “the table is red” without losing his knowledge of the latter, then this reasoning can provide him with knowledge that the table is not white with red lights shining on it. So something is wrong here: it seems that (a) Stewart’s reasoning in Cohen’s example is actually good, knowledge-transmitting reasoning, or (b) One-Premise Closure is not true, or (c) Stewart cannot competently deduce the proposition “the table is not white with red lights shining on it” from the proposition “the table is red” without losing his knowledge of the latter, or (d) Stewart cannot have basic knowledge that the table is red. Since the first three of these three options are extremely implausible, we are under considerable pressure to adopt the fourth option, and so deny that Stewart can have basic knowledge that the table is red. But this example is perfectly representative: if Stewart in Cohen’s example cannot have basic knowledge that the table is red, then none of us can ever have basic knowledge of any particular contingent fact about the external world. So the first problem of easy knowledge that besets basic knowledge theories is illustrated by this example: If Stewart has basic knowledge that the table is red, then it seems that he should be able to know by deduction, all too easily, that the table is not white with red lights shining on it. Now, in this example, a source of Stewart’s basic knowledge that the table is red is this: the table looks red. So our example suggests a general problem of the following form: if S has basic knowledge that p, and One-Premise Closure is true, then it follows that S can, by means of a simple and obvious one-step deduction, come to know — all too easily, it seems — that particular defeaters do not obtain (e.g., that it merely appears to her as if p). Why is it that, in the relevant range of cases, S cannot gain such knowledge so easily? The thought seems to be this: in the relevant cases, the reason why S believes that p — i.e., the source of S’s knowledge that p — cannot serve as a sufficient reason for S to believe that certain defeaters don’t obtain. So, if we accept One-Premise Closure, it seems that we’re driven to the following conclusion: if S knows that p, and r is a reason why S believes that p, then S must also
189
know that r is a reliable or trustworthy reason. If S doesn’t also know that r is reliable or trustworthy, then S will be able to gain knowledge by deduction from p — and all too easily — that certain defeaters of r don’t obtain. If r does not supply S with such knowledge all by itself, then it seems S cannot acquire such knowledge merely by deduction from things (like p) that she knows solely on the basis of r. And if r does supply S with such knowledge, then S’s knowledge that p is not basic. In other words, if we accept One-Premise Closure, then we must reject any doctrine of basic knowledge that conceives of the sources of our knowledge as satisfying constraints (i), (ii), and (iii) listed above in our account of what’s involved in something’s being a “source” of knowledge. In the next section, we’ll see why constraints (iv) and (v) are important also. III. The Second Problem of Easy Knowledge: Bootstrapping Inferences Against Counterpossibilities The second problem of easy knowledge is generated by assuming the following epistemic closure principle: Ampliative Closure: At least some ampliative arguments are such that, if S knows all of the premises to be true, and she competently infers the conclusion from the premises without losing her knowledge that the premises are true, then she knows the conclusion to be true. Ampliative Closure makes an existence claim, but it doesn’t specify which ampliative arguments preserve knowledge in this way — that is a substantive and contentious question within confirmation theory. Ampliative Closure says only that there are some such ampliative arguments. Now, how does this pose a problem for basic knowledge doctrines? Consider an ampliative argument which exemplifies the property mentioned in Ampliative Closure — someone can achieve knowledge of its conclusion by competently inferring it from the premises. Let’s suppose that the following inductive argument has this property:
190
(1a) (1b) (2a) (2b) (3a) (3b) …
At t1, it looks to me as if the table is red. At t1, the table is red. At t2, it looks to me as if the cat is on the mat. At t2, the cat is on the mat. At t3, it looks to me as if it’s snowing. At t3, it’s snowing.
(conclusion) The way things look to me is (generally) the way they are. In this inference, let’s suppose, each (a) premise states a source of the knowledge expressed by the (b) premise that immediately follows it. And let’s suppose that someone knows all of these premises to be true. More specifically, let’s suppose that someone — call her “Nancy” — has basic knowledge of each of the (b) premises. (If, for whatever reason, you are inclined to think that this cannot be reasonably supposed of the (b) premises I’ve chosen, then imagine other (b) premises of which it can be reasonably supposed. If you accept the doctrine of basic knowledge, then you are committed to believing that there are some such premises.) Since each (a) premise states a source of the basic knowledge expressed by the corresponding (b) premise, Nancy is able to know the (a) premises to be true without reliance on those sources themselves. (In the case I’ve chosen, Nancy can — I assume — introspectively know the (a) premises to be true.) Thus, Nancy knows (the truth of) all of the premises of this inductive argument. And the inductive argument for the conclusion is a straightforward case of induction by enumeration. Of course, some inductions do not give Nancy knowledge of their conclusion because those inductions rely on a small or unrepresentative data set. But this need not be a problem for the inductive argument above, since we can expand our data set to include everything that we know by means of the source in question (here, the source would be the looks of things). What Nancy has here, then, is a seemingly good inductive argument from known premises to a conclusion. If she competently goes through this reasoning without losing her knowledge of the premises, can doing so give her knowledge of its conclusion? It seems not. If Nancy is wondering why she should trust her visual experiences, and she then goes through the reasoning above, she should not regard this reasoning as
191
supplying her with knowledge that she should trust her visual experience. But how could that be? It could be that induction by enumeration generally doesn’t give us knowledge of its conclusion. But if this is so, then what sort of ampliative inference does give us knowledge of its conclusion? Whatever form that is, couldn’t Nancy duplicate the problematic reasoning by means of that form of inference? Whatever the answer to this last question, it needn’t detain us. For even if the basic knowledge theorist isn’t willing to grant that induction can ever give us knowledge of its conclusion, she will still confront a version of the Bootstrapping problem so long as she accepts the following closure principle. Ampliative Closure concerning epistemic reasons: At least some ampliative arguments are such that, if S knows all of the premises to be true, and she competently infers the conclusion from the premises without losing her knowledge that the premises are true, then she has good epistemic reason to believe the conclusion to be true. Even if we don’t accept the original version of Ampliative Closure, we should at least accept Ampliative Closure concerning epistemic reasons. But how does this principle apply to our example above? Can the inductive argument stated above ever give Nancy good reason to believe its conclusion? Again, it seems clear that it cannot do so. If Nancy is wondering what good reason she has to trust her visual experiences, and she goes through this reasoning, she should not regard this reasoning as supplying her with extra reason to trust, or to believe that she should trust, her visual experience.9 Her reasons for believing that her visual experience is trustworthy are not improved by virtue of her going through this reasoning. So it seems that we have to deny that Nancy knows all of the premises of the induction to be true. But then which of the premises does she not know to be true? If she doesn’t know the (a) premises, then she doesn’t know how things look to her, and that is absurd. Of course she knows how things look to her! Even if she doesn’t know everything there is to 9. Some philosophers would deny this claim. They would say that such inductive arguments can — at least in some cases — give us good reason to believe their conclusions. See, for instance, Alston (1986; 1993), Sosa (1997), Van Cleve (2003), and Bergmann (2004). I argue against this view in my (forthcoming b).
192
know about how things look to her, and even if she is sometimes wrong about how things look to her, still she does (or at least we can coherently and plausibly stipulate that she does) know a great deal about how things look to her. At least, she knows enough about how things look to her that she can use such knowledge in an inductive argument of the form just sketched. Is it that Nancy doesn’t know the (b) premises then? If Nancy has any basic knowledge of visible things around her, then there must be some premises which are such that she can have basic knowledge, supplied by vision, that they are true. Whatever those premises are, let them be the (b) premises of our bootstrapping argument. In that case, Nancy will know the (b) premises of the argument to be true. If Nancy has any basic knowledge of visible things around her, then either she doesn’t know the (a) premises of the inductive argument, or the inductive argument somehow gives her an additional reason to believe its conclusion. Both of these last two options are absurd, and so we are under pressure to deny that Nancy can have basic knowledge.10 The problem of easy knowledge that is illustrated by this example can be spelled out more generally as follows. Suppose that S has basic knowledge that p1, p2, p3, etc. Also, the source of S’s basic knowledge that p1 is r1, the source of S’s basic knowledge that p2 is r2, the source of S’s basic knowledge that p3 is r3, and so on, where the various p’s constitute a projectible type of proposition and the various r’s constitute a projectible type of reason to believe a proposition. Now, since r1 is the source of S’s basic knowledge that p1, then, given our account of what’s involved in something’s being a “source” of knowledge, S can know, without relying on r1, that r1 is the reason why she believes that p1. Furthemore, S can know, without relying on r2, that r2 is the reason why she believes that p2, and again she can know, without relying on r3, that r3 is the reason why she believes that p3, and so on. In that case, S will know all of the premises of an inductive argument of the following form:
10. The arguments provided by Fumerton (1995) and Vogel (2000) are similar to the argument of this section.
193
r1 is my reason for believing p1. p1. r2 is my reason for believing p2. p2. r3 is my reason for believing p3. p3. … r’s are generally reliable reasons for believing p’s. S will know all of the even-numbered premises by means of the respective r’s. And S will know all of the odd-numbered premises in some other way than by means of those same respective r’s. Thus, merely by having basic knowledge of the p’s, and also knowledge of which r’s supply her with reason to believe which p’s, S will have good reason to believe, by some ampliative inference, that r’s are generally reliable or trustworthy. But it seems clear that S cannot have good reason to believe that r’s are reliable or trustworthy merely by means of any such ampliative inference. If S knows that r1 is a reason why she believes p1, and she knows that r2 is a reason why she believes p2, and so on, then how can it be rational for her to gain confidence in the reliability or the trustworthiness of the r’s by means of this inference from those very beliefs, in conjunction with her belief of the p’s? If a “source” of knowledge satisfies conditions (i)–(v) listed above in our account of “sources”, then — it seems clear — such ampliative inferences cannot suffice for S to have good reason to believe their conclusion. IV. Cohen’s proposed solution to the problems of easy knowledge After posing these problems for Basic Knowledge theories, Cohen proposes the following solution: allow that there is some basic knowledge (what Cohen, following Ernest Sosa, calls “animal knowledge”), and that this basic knowledge can be inferentially expanded in some ways, but it cannot be inferentially expanded by the kinds of inference that are used in the One-Premise Closure and Bootrapping reasoning above. Cohen thus denies that certain plausible general principles of knowledge-transmission hold for basic knowledge.
194
Cohen candidly admits that this solution is ad hoc. It is ad hoc because it is unable to explain, by appeal to any already agreed-upon principles, why it is that, for any particular piece of basic knowledge, we can inferentially expand it in some ways but not in others. To illustrate: if Stewart has basic knowledge that the table is red, then he can know by deduction that the table is not green. So why is it that he can know by deduction that the table is not green, but he cannot know by deduction that the table is not white with red lights shining on it? Why is it that Stewart can inferentially expand his knowledge that the table is red into knowledge that the table is not green, but he cannot inferentially expand his knowledge that the table is red into knowledge that the table is not white with red lights shining on it? Again, if Nancy has basic knowledge of all of the (b) premises of the induction above (“the cat is on the mat”, “it’s snowing”, “the table is red”), then she should be able to know by induction (given normal background knowledge) that if the cat jumps into the snow and then onto the table, the part of the table’s surface that’s underneath the cat’s paws will become wet. But how could Nancy acquire knowledge of this fact by induction, if she couldn’t acquire knowledge of the reliability of her visual faculties by induction? Once again, what’s the difference between the two cases? Cohen suggests that, while basic knowledge can be inferentially expanded in some ways, it cannot be inferentially expanded to reach conclusions that are not, as Nozick (1981) would say, “sensitive”. That is, basic knowledge cannot be inferentially expanded to produce knowledge that p, for any p which is such that we would still believe it (and by the same method) if it were not true. This proposal might help to explain why Stewart can inferentially expand his basic knowledge that the table is red into knowledge that the table is not green, but cannot inferentially expand his knowledge that the table is red into knowledge that the table is not white with red lights shining on it. But this view leaves us with two unanswered questions. First, suppose that Stewart competently performs the following inference: The table is red. Therefore, it’s not the case that I falsely believe that the table is red.
195
Here again, the conclusion is insensitive. But can’t Stewart gain knowledge of this conclusion by means of competently performing this inference? The answer to this question is not a clear “no”. It seems not as clearly wrong to say that Stewart gains knowledge of the conclusion above by competently performing the inference above, as it does to say that Stewart gains knowledge of the conclusion “the table is not white with red lights shining on it” by competently performing the inference from “the table is red”. Since Cohen’s view can’t explain this difference, it leaves us with an unanswered question. And there is a second unanswered question: even if sensitivity is relevant in the way that Cohen thinks it is, still, why is that difference — the difference between sensitive and insensitive conclusions — one that should make a difference to the possibility of inferentially expanding one’s basic knowledge? Cohen’s solution, again, doesn’t answer this question. So the gap in Cohen’s proposed solution should teach us this: we want a solution to the problem of easy knowledge to explain why it is that, for any particular piece of basic knowledge, it seems that we can inferentially expand it in some ways but not in others — even when the form of inference is the same across these cases. Such a solution should not only accurately predict which cases fall on which side of that distinction, but it should also explain to us why those cases fall where they do. V. Rule of context-shifting We can develop a satisfying solution to the problem of easy knowledge if we approach these problems from a view of knowledge and justification that I’ve defended elsewhere.11 For present purposes, I will simply state — without arguing for — a rule of context-shifting that forms part of my account. I begin with some terminology. Consider any two hypotheses — h1 and h2 — that can be about anything whatsoever. Let’s say that h1 and h2 are “introspectively indistinguishable for S” just in case: (a) If h1 were true, then there would be a certain probability distribution over the mental states that S could be in (i.e., there is a certain 11. See Neta (2002; 2003; 2004; forthcoming c).
196
probability that S would be in mental state M1, there is a certain probability that S would be in mental state M2, etc.), and (b) If h2 were true, then there could be a different probability distribution over the mental states that S could be in (i.e., there is either a different probability that S would be in mental state M1, or a different probability that S would be in mental state M2, etc.), and (c) For any two mental states m1 and m2 which are such that the probability of S being in one of them rather than the other would differ depending upon whether h1 or h2 were true, the difference between m1 and m2 is not introspectively available to S.12 To say that the difference between two mental states is not “introspectively available” to S is just to say that there is no difference in what it’s introspectively like for S to be in one rather than the other mental state. As far as S’s introspection reveals, the two mental states are indistinguishable. In order to clarify this definition of “introspective indistinguishability for S”, I’ll offer some examples of pairs of hypotheses that are introspectively indistinguishable for me: (A1) I am now drinking Coke. (A2) It am now drinking something that looks and tastes exactly like Coke. (B1) The creature before me is experiencing pain. (B2) The creature before me is behaving exactly as if she’s experiencing pain. (C1) I have hands. (C2) I am a brain in a vat being neurochemically stimulated to have the experiences that I’m now having, and that lead me to believe that I have hands. (D1) There are 3 teaspoons of sugar in the coffee that I’m now drinking. (D2) There are 3.1 teaspoons of sugar in the coffee that I’m now drinking. 12. If we assume determinism, then we can formulate this definition of introspective indistinguishability much more simply, as follows: h1 and h2 are introspectively indistinguishable for S just in case If h1 were the case, then S would be in m1, If h2 were the case, then S would be in m2, and The difference between m1 and m2 is not introspectively available to S.
197
Each of the four pairs above contains two hypotheses that are introspectively indistinguishable for me. The D-pair cases might be qualitatively different, but this qualitative difference is — I assume — so slight that it falls below the threshold of introspective discriminability. Of course, which discrimination tasks we are capable of performing depends in part upon the circumstances in which those tasks are presented to us. (For instance, while it might be impossible to discriminate m1 from m2 when they’re considered by themselves, it might be possible to discriminate m1 from m2 when they’re considered along with m3, which is more similar to m2 than to m1.) But when I say that a difference between m1 and m2 is “introspectively unavailable” to S, I mean to imply that there are no circumstances in which S can perform the task of introspectively discriminating m1 from m2. Now, let’s say that a hypothesis H is “an uneliminated counterpossibility” with respect to S’s knowing that p at t just in case (i) H implies that S doesn’t know that p at t and (ii) H and “S knows that p at t” are introspectively indistinguishable for S. (In other words, what it’s introspectively like for S when H is true is the same as what it’s introspectively like for S when “S knows that p at t” is true, and yet H is incompatible with “S knows that p at t”.) An appraiser X “raises” an uneliminated counterpossibility with respect to S’s knowing that p at t just in case X (seriously and sincerely) treats that counterpossibility as a reason to regard S’s epistemic state as being worse than it would be were it not for that uneliminated counterpossibility. Now using the terminology just introduced we can state the following rule: (R) When X raises a hypothesis H that is an uneliminated counterpossibility with respect to S’s knowing that p at t, X moves into a context of ascription13 in which she can truthfully affirm, and cannot truthfully deny, that S’s total evidence at t includes all and only the evidence that S would have at t whether or not H is true.14 13. It is somewhat misleading to say that X “moves” into such a context of appraisal, since X may already have been in such a context. Maybe it is more accurate to say that, by raising the counterpossibility, X sees to it that she is in such a context. One’s listeners may refuse to enter into one’s context of appraisal by refusing to treat H as relevant to the appraisal of S’s epistemic state. One can expect this to happen when one raises skeptical hypotheses in a courtroom or a scientific laboratory, for instance. 14. Some epistemological internalists may object that if H and “S knows that p at t” are
198
Since one’s beliefs are justified by all and only one’s evidence for them, it follows from (R) that ascriptions of justification are also semantically context-sensitive. I have elsewhere argued15 that (R), in conjunction with a particular analysis of knowledge, provides us with an elegant, unified, and intuitively plausible solution to the Gettier problem, the lottery paradox, the dogmatism paradox, skeptical problems of underdetermination, the skeptical closure puzzle, and the problem of distinguishing epistemic from non-epistemic justification. Furthermore, an independently wellmotivated account of the function of evidence ascriptions16 predicts that such ascriptions should be semantically context-sensitive in accordance with some rule like (R). It seems to me that the convergence of all of these apparently independent considerations provides a compelling case for (R). But I do not have the space to rehearse all of these independent considerations here. I mention them now solely in order to indicate the variety of grounds on which I’ve elsewhere defended (R). In this paper, I will add to those grounds by appealing to (R) in order to solve the problem of easy knowledge. When we raise uneliminated counterpossibilities to S’s knowing that p at t, we thereby (in accordance with rule (R)) move into a context of appraisal in which we cannot truthfully say, and can truthfully deny, that S has evidence for p (rather than q) at t. In that new context, S’s belief that p cannot be truthfully said to be well justified at all, and so cannot truthfully be said to be knowledge. Thus, by raising skeptical counterpossibilities to S’s knowing that p, we make it true to say that S’s belief that p does not count as knowledge. VI. Solving closure and bootstrapping Let’s now show how we can use the accounts of knowledge and justification developed above to solve the problems of easy knowledge. introspectively indistinguishable for S, then what evidence S has cannot depend upon whether or not H is true — and this is true independently of any context of epistemic appraisal. This objection rests upon the assumption that the only facts that qualify as someone’s evidence for p at t are facts that the agent can know by introspection at t. But, as Alvin Goldman (1999) argues, this assumption founders on “the problem of stored beliefs”. 15. See Neta (2002; 2003; 2004). 16. See Neta (forthcoming a).
199
Closure: If Stewart has basic knowledge that the table is red, then he would be able to achieve knowledge (by deduction) that the table is not white. So it is possible to achieve, by means of deduction, knowledge of a sort that is very similar to what we’ve been calling “easy knowledge”. Indeed, there seem to be clear and unproblematic cases of this. You show me a table that is obviously red. I thereby have basic knowledge that it is red. But from this I deduce, and so come to know, that it’s not green. Equally, I can deduce, and so come to know, that it’s not white. It seems that I have thereby achieved something like “easy knowledge” that the table is neither green nor white. But if this is so, then why does there seem to be something wrong Stewart’s claiming to know that the table is not white with red lights on it? To answer this question, note that, according to (R), when we bring up the hypothesis that the table is white with red lights shining on it, we move into a context in which we can truthfully claim that Stewart does not know that the table is red (and can truthfully deny that Stewart knows the table to be red).17 So, if, in this context, we consider Stewart’s claim that the table is red, we will naturally (and, in this context, correctly) regard Stewart as making a claim that he doesn’t know to be true. That’s why, once we consider the hypothesis that the table is white with red lights shining on it, it now seems as if Stewart’s inference is epistemically worthless, even though it didn’t seem that way before we considered that hypothesis. The hypothesis that the table is white with red lights shining on it is an uneliminated counterpossibility to Stewart’s knowing that the table is red, and so it is just the kind of hypothesis the raising of which would tend to generate a context-shift, according to (R). In contrast, the hypothesis that the table is green is not an uneliminated counterpossibility to Stewart’s knowing that the table is red. That’s why raising the hypothesis that the table is green (for instance) does not also generate a context shift. Bootstrapping: Recall the example we considered above.
17. This is also predicted by a version of Jonathan Schaffer’s contrastivist view. See Schaffer (2004).
200
(1a) (1b) (2a) (2b) (3a) (3b) …
At t1, it looks to me as if the table is red. At t1, the table is red. At t2, it looks to me as if the cat is on the mat. At t2, the cat is on the mat. At t3, it looks to me as if it’s snowing. At t3, it’s snowing.
(conclusion) The way things look to me is (generally) the way they are. We’ll call this inference the “Bootstrapping” inference. Here again, there seem to be obvious and unproblematic cases of acquiring something very similar to “easy knowledge” by ampliative inference from premises of which we have basic knowledge. Consider the following. (1ac) (1bc) (2ac) (2bc) (3ac) (3bc) …
At t1, you’re saying that the table is red. At t1, the table is red. At t2, you’re saying that the cat is on the mat. At t2, the cat is on the mat. At t3, you’re saying that it’s snowing. At t3, it’s snowing.
(conclusion) You are a reliable reporter of the way things are. It seems that, by means of this inductive inference, you can move from basic knowledge of all the premises to easy knowledge of the conclusion. But if this is so, then why does there seem to be something wrong with the Bootstrapping inference? Once again, it helps to appeal to contextualism. According to the contextualist view we’ve been considering, when we bring up the hypothesis that the way things look to you is not the way they are, we move into a context in which we can truthfully claim that you do not have basic knowledge of the (b) premises of our induction. So, if, in this context, we consider the conclusion of the induction, we will naturally regard you as ignorant of the truth-values of the (b) premises of the induction. That’s why, once we consider the conclusion of the
201
induction, it seems as if you are being epistemically irresponsible in making the induction, even though it didn’t seem that way before we considered that conclusion. The conclusion is a hypothesis that, in some sense, you have not ruled out, and so it is just the kind of hypothesis the raising of which would tend to generate a context-shift, according to standard contextualist views. In contrast, the hypothesis that the way you say things are is the way they really are is easily ruled out by the same sorts of considerations that could lead us to believe the (a) and (b) premises of the Bootstrapping induction. That’s why considering the latter induction does not generate a context-shift, and the induction seems perfectly in order. Notice that (R) can help to explain why easy knowledge, as Cohen understands it, strikes us as unattainable while things that are very similar to easy knowledge seem easily attainable. From his basic knowledge that the table is red, Stewart should easily be able to gain knowledge by deduction that the table is not green. By parity, it seems that he should easily be able to gain knowledge by deduction that the table is not white. So what’s wrong with his gaining knowledge by deduction that the table is not white with red lights shining on it? The difference here is this: the hypothesis that the table is green is not an uneliminated counterpossibility to S’s knowing that the table is red, since the two hypotheses are not introspectively indistinguishable. Neither is the hypothesis that the table is white such an uneliminated counterpossibility, since again, the hypothesis that the table is red and the hypothesis that the table is white are not introspectively indistinguishable. But the hypothesis that the table is white with red lights shining on it is an uneliminated counterpossibility to S’s knowing that the table is red, so by raising the former hypothesis, we move into a context of appraisal in which we can no longer truthfully affirm, and can truthfully deny, the latter.18 Again, from our basic knowledge of all of the (b) premises of the testimony induction above, the inductive agent should easily be able to 18. In personal communication, Jonathan Schaffer has suggested that I make this point by saying that each of the hypotheses that we express by saying “the table is red”, “the table is green”, and “the table is white” includes a tacit domain restriction to the case in which the lighting is non-deceptive. So really the basic knowledge that I’m allowing Stewart can have is knowledge that the table is red (given that the lighting is non-deceptive). Some readers may find this a helpful way of understanding the point I’m making, but I’m reluctant to commit myself to this view about what’s included in the hypotheses themselves. For an explanation of my reluctance, see Neta (forthcoming c).
202
gain knowledge by induction that the way you say things are is (usually, at least) the way they are. So what’s wrong with her gaining knowledge by induction that the way things look to her is (usually, at least) the way they are? The difference, again, is this: the hypothesis that the the way you say things are is not the way they are is not an uneliminated counterpossibility to S’s having basic knowledge of the (b) premises. But the hypothesis that the way things look to S is not the way they are is an uneliminated counterpossibility to S’s having basic knowledge of the (b) premises. So by raising the second hypothesis (which we do when we consider the conclusion of the ampliative inference), we move into a context in which we cannot truthfully say, and can truthfully deny, that S has knowledge of the (b) premises of the induction. VII. Conclusion Sometimes, we have basic knowledge. This view has been thought to give rise to intractable problems of easy knowledge. But, as I have argued above, the problems of easy knowledge are not intractable once we accept the independently well-motivated view that ascriptions of evidence (and so of justification and of knowledge) are semantically context-sensitive. We can offer a non-skeptical solution to the problem of easy knowledge without having simply to grant that easy knowledge inferences transmit warrant from their premises to their conclusions, and without having to deny the Closure principles.19
REFERENCES Alston, W. (1983). ‘What’s Wrong with Immediate Knowledge?’, Synthese 55, 73–95. — (1986). ‘Epistemic Circularity’, Philosophy and Phenomenological Research 47, 1–30. — (1993). The Reliability of Sense Perception, Cornell University Press, Ithaca, NY. 19. I am grateful to Michael Bergmann, Martijn Blaauw, Matthew Chrisman, Jonathan Cohen, Adam Leite, Bill Lycan, Eric Marcus, Ted Parent, Jonathan Schaffer, James Shelley, Nicholas Silins, Jay Rosenberg, and especially to Stewart Cohen for helpful discussion.
203
Bergmann, M. (2004). ‘Epistemic Circularity: Malignant and Benign’, Philosophy and Phenomenological Research. BonJour, L. (1978). ‘Can Empirical Knowledge Have a Foundation?’, American Philosophical Quarterly 15, 1–13. Cohen, S. (2002). ‘Basic Knowledge and the Problem of Easy Knowledge’, Philosophy and Phenomenological Research 65, 309–29. Craig, E. (1990). Knowledge and the State of Nature, Clarendon Press, Oxford. Davies, M. (2003). ‘Solving the Problem of Armchair Knowledge’, Hempel Lectures, Princeton University. Fogelin, R. (1994). Pyrrhonian Reflections on Knowledge and Justification, Princeton University Press, Princeton, NJ. Fumerton, R. (1995). Metaepistemology and Skepticism, Rowman and Littlefield, Lanham, MD. Goldman, A. (1999). ‘Internalism Exposed’, Journal of Philosophy 96, 271–93. Harman, G. (1973). Thought, Princeton University Press, Princeton, NJ. Hawthorne, J. (2004). Knowledge and Lotteries, Oxford University Press, Oxford. Klein, P. (2004). ‘Closure Matters: Academic Skepticism and Easy Knowledge’, Philosophical Issues 14, 165–184. Markie, P. (2005). ‘Easy Knowledge’, Philosophy and Phenomenological Research. McDowell, J. (1982). ‘Criteria, Defeasibility and Knowledge’, Proceedings of the British Academy 68, 455–79. Neta, R. (2002). ‘S knows that P’, Noûs 36, 663–681. — (2003). ‘Contextualism and the Problem of the External World’, Philosophy and Phenomenological Research 66, 1–31. — (2004). ‘Perceptual Evidence and the New Dogmatism’, Philosophical Studies 119, 199–214. — (forthcoming a). ‘Epistemology Factualized: New Contractarian Foundations for Epistemology’, Synthese. — (forthcoming b). ‘Why Moore was not a Fallibilist’ in Themes from G.E. Moore: New Essays in Epistemology and Ethics, (eds.) Susana I. Nuccetelli and Gary Seay, Oxford University Press, Oxford. — (forthcoming c). ‘Undermining the Case for Contrastivism’, Social Epistemology. Nozick, R. (1981). Philosophical Explanations, Belknap Press, Cambridge, MA. Pryor, J. (2004). ‘What’s Wrong with Moore’s Argument?’, Philosophical
204
Issues 14, 349–378. Schaffer, J. (2004). ‘From Contextualism to Contrastivism’, Philosophical Studies 119, 73–103. Sellars, W. (1963). ‘Empiricism and the Philosophy of Mind’ in Science, Perception and Reality, Ridgeview, Atascadero, CA. Sosa, E. (1997). ‘Reflective Knowledge in the Best Circles’, Journal of Philosophy 94, 410–30. Van Cleve, J. (2003). ‘Is Knowledge Easy — Or Impossible? Externalism as the Only Alternative to Skepticism’ in The Skeptics (ed.) S. Luper, Ashgate, Aldershot. Vogel, J. (2000). ‘Reliabilism Levelled’, Journal of Philosophy 97, 602–23. Williamson, T. (2000). Knowledge and its Limits, Oxford University Press, Oxford. Zagzebski, L. (1999). ‘What is Knowledge?’, in The Blackwell Guide to Epistemology, (eds.) J. Greco and E. Sosa, Blackwell, Malden, MA and Oxford.
205
This page intentionally left blank
Grazer Philosophische Studien 69 (2005), 207–228.
A CONTEXTUALIST SOLUTION TO THE GETTIER PROBLEM Igor DOUVEN Erasmus University Rotterdam Summary According to the deontological view on justification, being justified in believing some proposition is a matter of having done one’s epistemic duty with respect to that proposition. The present paper argues that, given a proper articulation of the deontological view, it is defensible that knowledge is justified true belief, pace virtually all epistemologists since Gettier. One important claim to be argued for is that once it is appreciated that it depends on contextual factors whether a person has done her epistemic duty with respect to a given proposition, many so-called Gettier cases, which are supposed to be cases of justified true belief that are not cases of knowledge, will be seen to be not really cases of justified belief after all. A second important claim is that the remaining alleged Gettier cases can be qualified as cases of knowledge. This requires that we countenance a notion of epistemic luck, but the requisite kind of luck is of a quite benign nature.
1. Introduction Until not so very long ago, it was commonplace among epistemologists that knowledge is justified true belief. This changed abruptly and famously with the publication of Gettier’s (1963), in which he argued that justification is not enough to turn a true belief into knowledge, a conclusion that has since found general acceptance in philosophy. In his paper, Gettier proceeded by describing two cases in which, he claimed, a person justifiedly believes a true proposition yet does not know it. In both of these cases, it is now commonly diagnosed, it is only through a ‘felicitous coincidence’ (Klein 1971, 61) that the belief in question is both true and justified. This made it natural to think that at least in broad outline the solution to Gettier’s problem — or the Gettier problem, as it soon came to be called — was to add a sort of Anti-Luck Condition
as a further necessary condition to the three already referred to by the classical, tripartite analysis of knowledge, a condition excluding that knowledge could result from lucky coincidences. However, spelling out this condition in detail has proved to be remarkably hard. In the literature concrete proposals have typically been followed by the presentation of new (what are now commonly called) Gettier cases in which whatever condition or conditions were put forward in the proposal were satisfied by a person that apparently or at least allegedly still failed to know. In the following I want to argue that epistemologists may well have been too quick to abandon the classical analysis of knowledge. In particular I want to propose a notion of justification that, when added to true belief, does yield knowledge. The notion to be presented can be regarded as a contextualized version of the well-known deontic conception of justification. It will be seen that, given this notion, first, many of the examples classified in the literature as Gettier cases are not really cases of justified true belief at all, and second, those alleged Gettier cases that are cases of justified true belief can be categorized as cases of knowledge. Although the second claim requires that we countenance the compatibility of knowledge with a certain kind of epistemic luck, that kind of luck is, as will be explained, deserved epistemic luck, which makes the ascription of knowledge in the latter Gettier cases intuitively acceptable. I begin, in section 2, by describing the intuitions behind my contextualized version of the deontic theory of justification. In section 3, this theory will be made more precise and it will be argued, among others, why it is not open to the customary objection that the process of belief acquisition is not suĜciently under our control for our beliefs to be open to evaluation in deontic terms (such as praise and blame). I will then, in section 4, argue that my theory provides a satisfactory solution to the Gettier problem. Section 5 deals with the notion of epistemic luck that in some measure is still involved in that solution. There I will argue that, as already intimated, the notion of epistemic luck involved is of a quite benign, and therefore for our purposes acceptable, sort. 2. The deontic conception of justification contextualized Advocates of the deontic conception of justification all share the intuition that justification is ultimately to be analyzed in terms of
208
epistemic responsibility, of fulfilling one’s epistemic duty, of nonculpably believing a proposition, or in terms closely related to the foregoing. In fact, even some opponents of the view seem to share the intuition; for instance, Alston, who thinks the deontological conception of justification is ultimately untenable (see section 3), admits that justification is ‘most naturally understood [in deontic terms]’ (1988, 257). Various proposals have been made to turn the intuition into a full-blown theory of justification.1 However, it seems to me that, all their dissimilarities notwithstanding, these theories tend to suěer from a common defect, to wit, that they are insuĜciently sensitive to certain contextual aspects of justification, particularly what may be considered to be contextually determined economic aspects of justification. While the theory to be presented in this section is in agreement with the extant deontic theories of justification in that it holds that a person is justified to believe that H only if that person has tried her best to ascertain the truth-value of H, it adds to this that we cannot in general say what, given some proposition, counts as having tried one’s best to ascertain the truth-value of that proposition. We even cannot say this given that a proposition and a person have been fixed. We will have to consider a proposition and a person within a particular context or situation. Only then can we say what it is, or should be, worth to the person to be right about the proposition and hence whether she has suĜciently invested in finding out whether or not the proposition is true — which on the account to be oěered is what having done one’s epistemic duty (roughly) amounts to.2 The suggestion that economic considerations have an epistemological bearing is not at all new, as witness, for instance, Rescher’s (1976) discussion of Carnap’s (1962, 211) principle of total evidence. Roughly speaking, this principle requires that the degree of confirmation of a proposition be evaluated on the basis of all the evidence available at the time of evaluation. This was proposed as a fundamental principle of inductive logic, and not as part of a defense or an explication of the deontic conception of justification. Still it suggests a necessary condi1. See, for instance, Ginet (1975), BonJour (1985), and Chisholm (1989). 2. I should note that the contextualism involved in my theory has nothing to do with the currently popular brand of contextualism that is advocated in, among others, Cohen (1988), DeRose (1992; 1995), and Lewis (1996). In DeRose’s (1999, 190) apt terminology, the contextualism to be defended here is subject contextualism, whereas the popular version is attributor contextualism.
209
tion for justified belief that at first sight seems very natural from a deontic point of view: A person has fulfilled her epistemic duty with respect to a proposition, and thus is justified in believing that proposition, only if she has considered all the available evidence for that proposition. On closer inspection the condition must be judged too strong, however, and perhaps even, as Rescher (1976, 75 f) puts it, ‘neurotic’ and ‘unworkable’. For, as he rightly remarks (ibid.), there tends to be no theoretical limit to what relevant information is — in principle — available, but in general there are very real practical limits to how far we can go in gathering and taking into account relevant information. As an alternative, Rescher proposes a principle that gives an interesting economic twist to Carnap’s principle. On Rescher’s (1976, 76) proposal we should not require that one consider the total evidence, ‘but rather the maximal volume of relevance-endowed evidence that one can obtain relative to the available recourses …’. Of course there is considerable vagueness in this principle (how to determine the maximal volume of evidence to be obtained relative to the available recourses?), and, like Carnap’s original principle, Rescher’s is primarily meant as a principle of inductive logic, concerned with probabilities and not (directly) with justified beliefs. Yet it is possible to turn it into a fairly precise theory of justification, or so I hope to show in this and the next section. Before beginning with an outline of the theory of justification, I should emphasize that no pretense is made that the theory is complete; for instance, I shall have nothing to say about the possibility of a priori justification. What I do aim to do is elaborate the theory to such an extent that it becomes applicable to the sort of propositions that are typically encountered in Gettier cases. A crucial assumption of my theory is that propositions have epistemic import, a quantity that, it is further assumed, is measurable on an interval or ratio scale. The epistemic import of a proposition is always relative to a person and a context; how great it is depends on how important the person finds it to ascertain the proposition’s truth-value in the situation she is in (so a proposition and its negation always have the same epistemic import for a person in a context). It should be stressed that the epistemic import does not indicate how important it is to the person that in the given context the proposition is true (or false); it measures the importance, according to the person, of her knowing whether or not the proposition is true (an importance that may, but need not be, related to the importance of the truth or falsity of the proposition to the
210
same person in the same situation). I shall not endeavor to chart all the factors that possibly could play a role in determining epistemic import and limit myself to giving some examples of such factors. A person’s purposes constitute an obvious factor. For instance, if you are purposed to go to the opera tonight, then it is important to know whether there are still tickets for the performance available (it is, in this case, also important that there are still tickets available). Another possible factor is curiosity. You may simply be very curious to know whether or not Mozart was poisoned, even if knowledge of what actually happened to him will or would have no consequences at all for you. Social factors are also obvious candidates. From reporters, police oĜcers, prosecutors, judges, teachers, and so on, we expect that in their various capacities they assign great epistemic import to certain propositions; in part, that is what they get paid for. And from parents we expect, on pain of subtle social sanctions, that they assign great epistemic import to such questions as Is our child behaving well during school?, Is it making its homework regularly and properly?, What is it doing after school?, and so on.3 The fundamental idea of the analysis of justification to be proposed is that the question whether a given body of evidence justifies a given proposition — or, as I shall also say, is adequate for that proposition — cannot be answered without taking into consideration (i) the epistemic import of that proposition (for a person in a particular context), and (ii) the expenditure of eěort and/or time and/or money it takes to obtain better evidence — in a sense to be defined — for that proposition (for the given person in the given situation). To introduce the intuition underlying this idea, it is best first to consider, in the abstract, the extreme cases. Suppose you attach a very great epistemic import to a certain proposition in your present situation and although you are already in the possession of evidence which makes you as good as certain of the truth of the proposition, it is extremely easy for you to obtain further evidence relevant to that proposition. Then we do not want to say you have tried your best to find out the truth-value of that proposition if you are not prepared to gather that further evidence, regardless of the eěort it took you to gather the evidence you already 3. As I intend them to be understood, the expressions ‘assign great (or whatever) epistemic import to the proposition that P’ and ‘assign great epistemic import to the question whether (or not) P ’ (or ‘… to the question “P?”’) have the same meaning.
211
possess; in that case the latter evidence is, as I will say, not adequate. At the other extreme, we can consider a situation in which you assign hardly any epistemic import to a given proposition, in which you have evidence that makes you nearly certain of that proposition, and in which obtaining further relevant evidence would demand a great expenditure of eěort on your part. In such a case we definitely do not want to say that you have failed to fulfil your epistemic duty if you are not prepared to make an investment in order to obtain further evidence; in this case your present evidence is adequate. There are many gradations in between these extremes. In the next section, the theory of justification will be made more precise. It will there be seen that we can quite generally say under what conditions someone has tried suĜciently hard to find out the truth-value of a proposition and hence under what conditions a person is justified in believing a proposition. Preliminary to that, let me try to make clear by dint of a homely example that the conception of justification we are concerned with is quite natural and is in line with what I expect to be perfectly recognizable behavior. When I leave home I always lock the garage. Before I do so, I always briefly inspect the garage so as to make sure I am not locking up Timmy, my cat, that loves to sniě around in the garage. When I intend to be back shortly, I no more than cast a quick glance in the garage. If I do not see my cat, that is suĜcient to convince me of the truth of the proposition that he is not in the garage. When I will be away for more than a few hours, however, I inspect the garage more thoroughly. In that case the question whether Timmy is in the garage has a greater epistemic import for me. Just inspecting the garage briefly will make me as good as certain that Timmy is not there. But it is utterly simple to become even more certain of that proposition; it is very little trouble to see if, for instance, Timmy is hiding behind a pile of old newspapers. It seems unlikely to me that, if he was, I would not still see him if I just gave the garage a quick check. Nevertheless, I prefer not to take the chance and therefore check more carefully. It seems natural to say, at least to me, that while in both cases a quick glance in the garage convinces me that Timmy is not in it, in the latter case justification for that proposition is harder to obtain. If you do not have a cat, or in any event do not recognize the justdescribed behavior, it may help to think of checking whether the windows and doors are closed before you leave on a vacation and, by contrast, before you go out to run some quick errands, for instance.
212
Presumably you will in the former situation more easily tend to doublecheck than in the latter, even though also in the former your first check will normally convince you that everything is locked. The reason — it seems plausible to assume — is that when you leave the house alone for a longer period of time the question whether the windows and doors are closed is of more importance than when you will be back soon (however important it may already be in the latter case). In the next section I will try to go beyond mere intuitions and make the concepts just hinted at precise in order to come to a more or less formal theory of deontic justification. 3. Contextual deontic justification specified I shall assume that there is a value t close but unequal to 1 such that a person counts as believing a proposition H if and only if H has a probability exceeding t relative to that person’s background knowledge (cf. Foley 1993, Ch. 4). Furthermore, I assume the probability analysis of evidence according to which E is evidence for H exactly if p(H p E) > p(H). Say that E is t-evidence for H precisely if it is evidence for H and, in addition, p(H p E) > t. We define some further notions. Let p() be the function representing a person S’s degrees of belief at time t, and let this be a probability function (so S is assumed to be probabilistically coherent). Definition 3.1 E is evidentially superior to E′ for H for S at t exactly if: 1. E ≡ E′ ∧ E″ for some E″ such that E′ ¡ E″ and 2. p(H p E) p(H p E′).4 The intuitive idea is that E comprises E′ and that coming to know E will or would make S either more or less certain of H than if she only comes or came to know E′. In general there will be a cost to gathering further evidence that will at least in part depend on the situation one is in (in some situations one may be in a much better position to obtain a certain further piece of evidence than in others). Assuming that E ≡ E′ ∧ E″, let c(E, E′,S,t) 4. Notice that from this clause it follows that E ¡ E″.
213
be the cost for S at t of obtaining evidence E given that she already possesses evidence E′. We may also assume that, in general, coming to have superior evidence for a given hypothesis will be, or at least should be, valuable to a person (to whatever precise extent). Assuming that E is evidentially superior to E′ for H for S at t, let d(E, E′,H, S, t) be the highest price S is, or should be (see the comments below), willing to pay at t for extending her evidence E′ to E given the epistemic import H has for her at the time. We can now define that Definition 3.2 E is adequate evidence for H for S at t exactly if: 1. E is t-evidence for H for S at t, and 2. there is no E′ such that: (a) E′ is evidentially superior to E for H at t, and (b) c(E′,E,S,t) ≤ d(E′,E,H,S,t). And, finally Definition 3.3 H is justifiedly credible to S at t exactly if S at t has evidence that is adequate for H for S at t. Note that if H is justifiedly credible to S at t, that does not mean that S at t could not obtain further evidence relevant to H (relevant in the sense that it would make S either more or less certain about H than she is). Rather it means that there is no such evidence S at t could obtain at a reasonable price; further evidence, if any can be obtained at all by S at t, is too expensive — too expensive, given the epistemic import H at t has for S, that is. The theory requires a number of (further) comments, which follow in random order. 1. If we want to maintain that justified credibility is closed under entailment, or at any rate some qualified version thereof (such as obvious or known entailment), as I think we should (cf. Douven 2002a, Sect. 2), then definition 3.3 cannot stand unadorned, for as it stands it is vulnerable to Kyburg’s (1961) lottery paradox and Makinson’s (1965) preface paradox; see Douven (2002a) and Douven and UĜnk (2003) for some suggestions of how to refine the definition. For the present discussion, the required refinements would only introduce needless complications, however.
214
2. It is worth noticing that a change of context can alter the justificational status of a proposition. Suppose, for instance, that you are in Amsterdam and plan to travel to Paris today. Further suppose that according to the schedule you have there is a train going from Amsterdam to Paris at 8 PM. So you believe that there is a train from Amsterdam to Paris leaving at 8 PM. However, the schedule is six months old, and you could easily call the Amsterdam station to check whether there indeed is a train to Paris at 8 PM. Thereby you would obviously obtain superior evidence for your belief. Still, given that you intend to fly to Paris, the proposition that there is a train at 8 PM has relatively little epistemic import for you so that your schedule gives you adequate evidence for it. Now suppose that quite suddenly a snowstorm comes up, and that therefore all flights are cancelled (trains are still riding, however). Then all of a sudden the proposition that there is a train to Paris at 8 PM may get great epistemic import for you. Given that it is so easy for you to obtain better evidence for it, you may well no longer be justified in believing it. Of course, it is not hard to imagine situations in which the opposite happens and an unjustified belief becomes justified, simply because the proposition that is the content of the belief loses much of its epistemic import. 3. The notions of probability, costs, and prices as they figure in the previous definitions are in principle open to a variety of interpretations. I do not here want to commit myself to one specific interpretation for each of the notions, but I do want to indicate what I believe are more and less plausible ways to conceive of them in the context of our theory. One possibility would be to interpret probabilities, costs, and prices all in an entirely subjective fashion; probabilities would thus be subjective probabilities, the cost related to an extension of evidence in a given context by a given person would simply be what according to that person’s subjective judgement the cost is, and it would be up to everyone to determine what is the highest price to be paid for such an extension (given a particular context). On the other hand, we could interpret all notions purely objectively. Probabilities would then be objective probabilities or chances, and costs and maximal prices would be thought of as being somehow determined in a way that is independent of our judgements. But neither of these extreme interpretations seems to go very well with the theory of justification we presented. In eěect, it seems to me that interpreting even one of the notions either purely
215
subjectively or purely objectively would yield an unsatisfactory theory of justification. As to objective interpretations, note that it is certainly not generally possible to find out objective (conditional) probabilities (provided there are any objective probabilities) nor is it clear how we could find out costs and prices that are determined in some objective, human-independent fashion (supposing it makes sense to assume the existence of such objective costs and prices at all). And if it is not generally possible for a person to find out such objective probabilities/costs/prices, then it will not in general be possible for that person to determine whether she has adequate evidence for a proposition and, consequently, whether she is justified in believing that proposition. I find this strong kind of externalism quite unpalatable.5 Interpreting probabilities/costs/prices purely subjectively seems to have a different drawback. Suppose the slightest indication that H might be the case has me assign a very high probability to H, but subsequently there is nothing that could make me assign a diěerent probability to H. Then according to the above theory I would have adequate evidence for H (whatever epistemic import that proposition has for me). Quite clearly, it would intuitively be wrong if obtaining justification for a proposition were that easy. It can be readily seen that a subjective interpretation of costs and/or prices would be objectionable for the same reason. Fortunately, in our choice of a suitable interpretation we are not restricted to either a purely objective or a purely subjective interpretation of probabilities, costs, and prices. We could opt for some ‘morethan-merely-subjective’ interpretation of these notions. For the notion of probability, several such interpretations have been proposed. Kyburg (1990) defends a so-called evidential interpretation of probability. Roughly, on this view probabilities are degrees of belief based on, or informed by, the available (relevant) statistical data. And on Gillies’s (1991; 2000) intersubjective interpretation of probability, probabilities are degrees of belief that — ideally — are shared by communities that have common goals and that freely exchange information. A corresponding social interpretation of costs and prices would seem most appropriate for our theory. We are not able to determine quite generally 5. According to Plantinga (1993), externalism is even incompatible with a deontological conception of justification, but this is debatable; see Bergmann (2000).
216
what the cost of mounting a staircase for a person is; that depends on the person’s physical shape as well as on certain properties of the staircase. But given that someone is in a certain particular shape and given that the staircase has certain particular properties, we will in general be able to come to a fairly unanimous verdict about that cost (e.g., for someone in good physical shape we will not easily think it too much trouble to mount a ‘normal’ staircase). Likewise, there in general seems to be consensus about how much someone should be prepared to spend on getting better evidence for a proposition that has a certain epistemic import for her. A parent who asks his or her child how it is doing in school can on the basis of the child’s answer come to the conviction that it is doing well, but since — as we may assume — the question has great epistemic import for him/her, we will think the parent should be willing to inquire further (e.g., most of us will think that the parent should from time to time have a look at the child’s homework, visit the school’s parents’ evenings, and so on), even though we do not feel he/she should quit his/her job so as to be able to peek in through the classroom’s window all day. Admittedly, intersubjective probabilities, socially determined costs and prices, will rarely be exactly determined. One should consider, however, that in the past century philosophers have become accustomed to working with idealizations. 4. There may seem to be another objection to the eěect that justification can, on the above account, be too easy to come by, this one running as follows: ‘By assigning probability 1 to some proposition we become automatically and permanently justified in it. After all, once a proposition has unit probability, it will have unit probability ever after. No possible evidence could alter the proposition’s probability, and hence one’s evidence for it cannot but be adequate’. Let us in response to this first notice that, depending on the interpretation of the notion of probability, it may not be all that easy to assign probability 1 to a proposition. But secondly and more importantly, while it is correct that a proposition that has probability 1 will have that probability henceforth if we suppose that Bayes’s rule is the only rational update rule (as Bayesians do), this supposition is open to doubt; elsewhere I have argued that the commonly adduced argument for it fails (Douven 1999). And in the literature probabilistic inference rules or systems of such rules have been proposed that do not guarantee that a proposition that has unit probability at a given point in time will have unit probability whatever new
217
evidence the future may bring (see, e.g., Williamson 1998 and Kuipers 2000, Ch. 3). The probabilistic version of the so-called Inference to the Best Explanation proposed in Douven (1999) also has the feature that a proposition that now has probability 1 may at a later time come to have a probability smaller than 1, and it may be possible to make a case for this rule of inference (see Douven 2002b; 2005a; 2005b). Here I will not suppose any particular rule of inference, but just want to note that there is currently insuĜcient ground for the objection. 5. Finally, I should point out that the deontic theory of justification as presented here escapes the standard objection against such theories. It appears that equating justified belief in a proposition with its being the case that the person cannot be rightly reproached for holding the belief, as deontologists want to have it, requires that believing a proposition is something that is under voluntary control. If one cannot help but believe some proposition, then how can one be blamed for believing it? And critics of the deontic conception have tried to show that believing is not (or at least not suĜciently) under our control (see, e.g., Alston 1988 and Plantinga 1993a). Given our conception of belief as according a probability greater than a value t close to 1 to a proposition, these critics’ point comes to the claim that we often cannot help but assign a high probability to a proposition. We can grant this point6 and still maintain that the deontic conception of justification is viable. For, even if we cannot but assign a high probability to a given proposition, that does not mean that there is no question of whether we have done our duty with respect to that proposition. A proposition may have a high probability and yet our evidence for it may not be adequate; there may, at a reasonable price, be evidence available superior to the evidence we currently have. We may or may not be willing to pay the price for the evidence, but if we are not, then we certainly will not have done our epistemic duty with respect to that proposition and hence our belief in it will not be justified. Even if believing is not (always) under our control, gathering further evidence for what we believe is, at least whenever such evidence is available at a reasonable price; and that is what matters.7
6. I am not convinced that we must grant it; much may also depend here on the interpretation of probability assumed. 7. See in the same vein Swinburne (2001, 186).
218
4. Gettier cases reconsidered The crucial question now of course is how the theory just presented fares with respect to the Gettier problem. If the foregoing is at least basically a correct theory of justification, then it can be remarked right away that virtually all examples of Gettier cases that have been presented in the literature are to an important extent underspecified. In such examples it is typically stipulated that a person has a justified true belief, but since we are mostly left in the dark about the epistemic import the person attaches to that belief as well as about the possibilities to gather evidence for that belief superior to the evidence the person already possesses, it is impossible to ascertain whether the true belief is indeed justified (on our understanding of this term). Still, sometimes the context of the example does allow us to conclude something about those, in our theory central, factors. This is for example the case in Russell’s (1948, 154) Gettier case avant la lettre, the famous case of the stopped clock. A stopped clock indicates the correct time twice a day. In Russell’s case you happen to look at the clock at one of those moments and thus come to have a true belief about what time it is. Since clocks tend to indicate the right time, your belief about what time it is, is justified, at least that is the conclusion standardly drawn from this case. But is it really? We are told absolutely nothing about the epistemic import the question what time it is has for you in the context of this example, but let us suppose it has at least some epistemic import for you. (If it hasn’t, then why look at the clock at all?) The next question then is: How hard is it, in the given context, to obtain better evidence for your belief? Well, how long does it take to verify that a clock is running? Provided it is a ‘normal’ clock, that is, one with a seconds hand, the answer is: Only slightly longer than one second (it of course depends on your feeling for time; if that is very precise, it takes no longer than one second). It therefore seems to me that, although a quick glance at the clock may convince us that it is now (say) 4 o’clock, such a glance is insuĜcient to justify that conviction — assuming that the proposition that it is now 4 o’clock has at least some epistemic import for us; more is needed to obtain adequate evidence. It thus seems that, on the proposed understanding of the notion of justification, the case of the stopped clock does not constitute a counterexample to an analysis of knowledge that equates knowledge with justified true belief: a quick glance is, on our understanding of justification, not enough to justify
219
the belief that it is now such-and-such time — and anything more than a quick glance will, we may suppose, not lead to that belief. Or consider the following paradigm Gettier case. Nogot, one of your colleagues, since a few weeks comes to work driving a Ford; you have seen the car’s registration book in his hands; and you know that he had a long-time wish to buy a Ford. It is now asserted in this case that on the basis of the cited data you are justified in believing that Nogot owns a Ford. If so, then it can be plausibly held that you are also justified in believing the weaker proposition that one of your colleagues owns a Ford. It now so happens that just when Nogot was about to buy a Ford a few weeks back, his aunt oěered him the use of her Ford, an oěer Nogot happily accepted. Thus while Nogot drives a Ford, and even has the car’s registration book (with, of course, his aunt’s name in it), he does not himself own a Ford. However, by chance another of your colleagues, Havit, does own a Ford, unbeknownst to you. It is commonly concluded from this that, on a classical analysis of knowledge, you know that one of your colleagues owns a Ford. But is this conclusion correct? We may assume that the evidence in the example makes it very likely that Nogot is the owner of a Ford, and we may also assume that on the basis of that evidence you believe he is the owner of a Ford. Again, however, we are kept dangling about the epistemic import the question whether Nogot owns a Ford has to you. Given the theory of justification presented in section 3, it is not at all clear that, if the question has any importance to you at all, you are justified in believing that Nogot owns a Ford. After all, it would seem to be quite easy to obtain superior evidence in this case, for instance by simply asking Nogot whether he is the owner of the car he drives since a few weeks. Surely that would not be an impertinent question, and unless Nogot has the intention to mislead you (and the supposition that he does is not part of the example), he may be assumed to answer it truthfully and tell you that he is not the owner of a Ford, thereby giving you evidence superior to the evidence you had for the proposition that Nogot owns a Ford. And if you are not justified in believing that Nogot owns a Ford, then, since the justification for your belief that one of your colleagues owns a Ford is in the example supposed to derive from the alleged justification for the belief that Nogot owns a Ford, neither are you justified in believing that one of your colleagues owns a Ford. So again we do not have a manifest counterexample against the classical analysis of knowledge.
220
Goldman’s (1976) fake barn example is another case in point. In this story, a certain person, Henry, sees, while driving in his car, a barn and thereupon comes to have the true belief that he sees a barn. However, unknown to Henry, the district he has entered is full of papier-mâché facsimiles of barns. These facsimiles look from the road exactly like barns, but are really just façades, without back walls or interiors, quite incapable of being used as barns. Having just entered the district, Henry has not encountered any facsimiles; the object he sees is a genuine barn. But if the object on that site were a facsimile, Henry would mistake it for a barn. (p. 773)
According to Goldman, this constitutes another Gettier case, that is, a case of justified true belief that is not knowledge. On my theory it is not so clear, however, that we should be willing to admit that Henry does not know that he sees a barn. For all we can tell from the story, it has very little, if any, epistemic import for Henry whether the barn is real or not. Henry is just naming various of the things he sees around him for the edification of his son, who is sitting next to him in the car. Presumably little harm will be done should he mistake a barn façade for a barn; his son would thereby come to know what a barn looks like anyway (more or less as he would by looking at a picture of a barn). Therefore, a quick glance from the car window may suĜce to provide Herny with adequate evidence for his belief, that is, evidence that suĜces to turn his true belief into knowledge. If, on the other hand, Henry were the district’s Barn Inspector General (say) and presently in the duty of a barn count, he might, on the account of justification that was proposed, well have failed to know that it is a barn he sees. For someone in that capacity may be expected to go to the (not very big, as we may imagine) trouble of getting out of the car and taking a closer look at the barn; if Henry is not willing to do that, then, on our account, he will likely fail to be justified in his belief. While this is only a small sampling of Gettier cases to be found in the literature, I submit that considerations similar to those just brought forward would reverse our verdicts in other such cases that many have 8. Some quick hints: In Lehrer and Paxson’s (1966, 228) Grabit case, the person referred to as ‘I’ did not know that Tom Grabit removed the book if he or she was hired by Tom’s employer to find out whether there is anything to the rumor that Tom is a cleptomaniac. And
221
thought to be relatively uncontroversial.8 Nevertheless, there will be cases in which a person has a true belief for which she has justification according to my theory of justification and in which the truth of the belief and the person’s having justification for that belief coincide as a matter of mere luck. Suppose, for instance, that in the previously mentioned Ford case Nogot does have the intention to deceive you and not only lies to you about the ownership but also has a forged registration book with his name in it, et cetera. Then it can certainly happen that, even if you go through great pains in trying to find out the truth about Nogot’s ownership, you come away with the false belief that he owns the Ford. If, then, in this case, too, you have a colleague who owns a Ford, your belief that one of your colleagues owns a Ford is not only true but may also be justified. It thus seems that, while the theory of justification presented in this section may reduce the Gettier counterexamples to the tripartite analysis, it does not entirely eradicate them. My proposal now is to bite the bullet and accept that these remaining cases are really cases of knowledge. The person or persons having a justified true belief in these cases has or have been so lucky as to acquire knowledge. In the introduction it was said that Gettier’s paper convinced virtually all epistemologists that any adequate theory of knowledge would have to encompass some sort of Anti-Luck Condition. If this general conviction is correct, my theory obviously cannot be correct; at a minimum, it in that case is seriously incomplete. However, in the next section I will argue that it is currently not obvious why an epistemology should implicate such an Anti-Luck Condition, and certainly not one that is so general as to forbid any kind of epistemic luck. 5. Deserved epistemic luck How objectionable is it if a theory of knowledge must leave room for epistemic luck? Undoubtedly most epistemologists will regard it to be a defect of the theory. Still, to the best of my knowledge, no one so far has been able to explicate what is supposed to be so problematic about the idea that knowledge (sometimes/partially) can be due to the kind in Harman’s (1973, 143 f) assassination case, Jill may well know that the political leader has been killed, given that the question whether or not the leader has been killed does not seem to have real epistemic import for her.
222
of luck encountered in the Gettier cases. In fact, most epistemologists have not even seen a reason to attempt to do this; they simply take for granted that there can be no place for epistemic luck in a viable epistemology. Zagzebski (1999) may be the only exception in this respect. By way of answer to the question why epistemic luck is inadmissable she cites — with approval —Aristotle: ‘To leave the greatest and noblest of things to chance would hardly be right’. Now it seems to me that this answer is not in the least as plausible as Zagzebski appears to suppose. Consider, for example, that in modern art principles of chance have been explicitly admitted. In twentieth century music we encounter socalled aleatoric techniques, which let for instance chance mechanisms (partly) determine the structure of a piece of music (such as in Boulez’s third sonata for piano or in Stockhausen’s Klavierstück No. XI). And techniques as were employed by Jackson Pollock give maximal room to chance processes in visual art. Thus at least in modern art we explicitly allow chance processes to play a role in the creation of beauty, which plainly qualifies as a great and noble thing if anything does. In a similar vein one might think of love, doubtless another great good. Perhaps by now psychologists know enough about the workings of the human mind to be able to determine — on the basis of a country wide screening of adolescents, say — with a reasonable degree of certainty for any young person what other person or persons would be most suited for him or her to engage with in a romantic relationship. But even if we assume that choosing one’s partner on the basis of a psychologist’s advice can help to avoid a significant amount of marital misery, I presume that most of us will abhor this Brave New World kind of selection process and will stick to their preference for the ordinary, ‘old-fashioned’, ways of meeting someone to share one’s life with, even if chance plays an ineluctable role in them. One may insist that, even if it is not entirely clear why, intuitively an epistemology that can make do without a notion of epistemic luck is preferable to one that cannot, at least ceteris paribus. However, while I have no intention to advertise the fact that my theory of knowledge must admit an element of epistemic luck as its most attractive feature, I do want to note two things in response to the objection. First, it cannot be excluded that there simply is no epistemology which is capable of dealing with, among others, the Gettier problem without having to appeal to some form of epistemic luck and that is not for diěerent
223
reasons more problematic than mine. Without going into any details, let me in this connection point to Morillo’s (1984) paper, where it is argued that all so-called externalist theories of knowledge, such as those put forward by Goldman (1976) and Dretske (1981), are committed to the compatibility of epistemic luck with knowledge.9 Secondly, and more importantly, in the following I will argue by means of an analogy that the kind of epistemic luck present in my theory is of a a quite benign nature, and in eěect is such that we should not have any qualms about granting it. Suppose an athlete is training frantically in order to prepare herself for the 100 meters sprints at the next Olympics, and suppose that by the time the games begin she is in excellent shape and full of confidence. If she does well on the 100 meters, we presumably will not want to say that she has been lucky: she worked hard for it. Clearly, however, she is not a machine; in spite of all the hours she trained, she may have an oě-day at the day of the finals and fail dismally. If she does have an oě-day, and as a result finishes in a time of 12 seconds, say, we might say that she has had bad luck. But, of course, even if she has an oě-day, and finishes in 12 seconds, she might still end first; all her opponents might even do worse (the athlete in lane 1 trips over her shoelaces, the athlete in lane 2 is disqualified because of a double false start, and so on). If our athlete wins gold in this way, we will certainly want to say that she has been lucky, but it is only because she trained so hard for it that we will want to say the luck was deserved. Given the odd things that happened to her opponents in the race, she might have won gold even if she had not prepared herself so well for the Olympics. But although in that case we would also say that she had been lucky, we presumably would not think of the luck as being deserved. The analogy now is as follows. Even if you attach great epistemic import to a question and have done the best you could do to be able to exclude that the belief which answers your question is false, it may happen that the evidence you possess is misleading and the belief is false. If, in the same case, your evidence does not mislead you, and points to the truth, you have knowledge. But in that case we will not say you have been lucky; just as the athlete has, by training hard, done her best to put up a tremendous performance, you have done your best to gain knowledge. It may happen, however, that your evidence for the belief 9. See in the same vein Gjelsvik (1991).
224
is adequate but misleading yet the belief is true. As intimated above, I propose that in this and similar cases we bite the bullet and grant that you have knowledge. Like the athlete who in spite of all the training she has done has an oě-day yet wins the race, you have been lucky — but you deserved to be lucky, because you have done what duty required you to do, relative to the epistemic import the question has for you, to find out the correct answer to that question. I should add, incidentally, that the same cannot be said of the sort of epistemic luck that according to Morillo is involved in externalist theories of knowledge. Whether the external connections (whatever their precise nature) between our beliefs and the world that, on an externalist view, are necessary for knowledge hold, obviously depends to one part on our habitat. And, as Morillo argues, that may change in quite unpredictable ways and to an extent which would make our sensory organs entirely unreliable indicators of what is going on around us, and thus no longer capable of ensuring the requisite belief–world connections. She further argues that it is in general not in our hands whether our habitat changes in any such way. If all this is true, then, if our habitat does not undergo that kind of change, we will be just lucky in the same way in which an athlete is just lucky who does not seriously prepare for the race and yet wins it because of the bad performances of her opponents. 6. Conclusion If something along the lines of the contextualized deontic account of justification is correct, then a first conclusion is that many of the examples presented as Gettier cases in the literature are badly underspecified. As a result, it is in many of these cases not possible to determine whether the belief in question is justified. In others, however, even from the sparse information they provide it seems safe to conclude that the belief in question is not justified. So, many of the alleged Gettier cases cannot be claimed to be cases of justified true belief that are not cases of knowledge. Nevertheless, there do remain some cases in which the truth and the justification of a belief (for a given person) coincide by luck only. Still, since the luck in these cases is deserved, in the sense explained in the previous section, it is acceptable to say that these cases do not constitute counterexamples against the classical analysis of knowledge
225
either. In sum, if the theory of justification presented in the paper is tenable, then knowledge is justified true belief after all.10
REFERENCES Alston, W. (1988). ‘The Deontological Conception Of Epistemic Justification’, Philosophical Perspectives 2, 257–299. Bergmann, M. (2000). ‘Deontology and Defeat’, Philosophy and Phenomenological Research 60, 87–102. BonJour, L. (1985). The Structure of Empirical Knowledge, Harvard University Press, Cambridge MA. Carnap, R. (1962). Logical Foundations of Probability, Chicago University Press, Chicago. Chisholm, R. (1989). Theory of Knowledge (3rd ed.), Prentice Hall, New York. Cohen, S. (1988). ‘How to be a Fallibilist’, Philosophical Perspectives 2, 91–123. DeRose, K. (1992). ‘Contextualism and Knowledge Attributions’, Philosophy and Phenomenological Research 52, 913–929. — (1995). ‘Solving the Sceptical Puzzle’, Philosophical Review 104, 1–52. — (1999). ‘Contextualism: An Explanation and Defense’, The Blackwell Guide to Epistemology, (eds.) J. Greco and E. Sosa, 187–205, Blackwell, Oxford. Douven, I. (1999) ‘Inference to the Best Explanation Made Coherent’, Philosophy of Science (Proceedings) 66, S424–S435. — (2002a). ‘A New Solution to the Paradoxes of Rational Acceptability’, British Journal for the Philosophy of Science 53, 391–410. — (2002b). ‘Testing Inference to the Best Explanation’, Synthese 130, 355– 377. — (2005a). ‘Empirical Equivalence, Explanatory Force, and the Inference to the Best Theory’, Logics of Scientific Cognition: Essays in Debate with Theo Kuipers, (eds.) R. Festa, A. Aliseda, and J. Peijnenburg, Rodopi, Atlanta (in press). — (2005b). ‘Evidence, Explanation, and the Empirical Status of Scientific Realism’, Erkenntnis (in press). 10. I am greatly indebted to Martijn Blaauw and Cornelis van Putten for very helpful comments on an earlier version of this paper.
226
Douven, I. and J. UĜnk (2003). ‘The Preface Paradox Revisited’, Erkenntnis 59, 389– 420. Dretske, F. (1981). Knowledge and the Flow of Information, MIT Press, Cambridge MA. Foley, R. (1993). Working Without a Net, Oxford University Press, Oxford. Gettier, E. (1963). ‘Is Justified True Belief Knowledge?’, Analysis 23, 121–123. Gillies, D. (1991). ‘Intersubjective Probability and Confirmation Theory’, British Journal for the Philosophy of Science 42, 513–533. — (2000). Philosophical Theories of Probability, Routledge, London. Ginet, C. (1975). Knowledge, Perception, and Memory, Reidel, Dordrecht. Gjelsvik, O. (1991). ‘Dretske on Knowledge and Content’, Synthese 86, 425441. Goldman, A. (1976). ‘Discrimination and Perceptual Knowledge’, Journal of Philosophy 73, 771–791. Harman, G. (1973). Thought, Princeton University Press, Princeton. Kemeny, J. (1955). ‘Fair Bets and Inductive Probabilities’, Journal of Symbolic Logic 20, 263–273. Klein, P. (1971). ‘A Proposed Definition of Propositional Knowledge’, Journal of Philosophy 68, 471–482. (Reprinted in Epistemology: An Anthology, (eds.) E. Sosa and J. Kim, 60–66, Blackwell, Oxford; the page reference is to the reprint.) Kuipers, T. (2000). From Instrumentalism to Constructive Realism, Kluwer, Dordrecht. Kyburg, H. (1961). Probability and the Logic of Rational Belief, Wesleyan University Press, Middletown CT. — (1990). Science and Reason, Oxford University Press, Oxford. Lehrer, K. and T. Paxson (1966). ‘Knowledge: Undefeated Justified True Belief’, Journal of Philosophy 66, 225–237. Lewis, D. (1996). ‘Elusive Knowledge’, Australasian Journal of Philosophy 74, 549– 567. Makinson, D. (1965). ‘The Paradox of the Preface’, Analysis 25, 205–207. Morillo, C. (1984). ‘Epistemic Luck, Naturalistic Epistemology And The Ecology Of Knowledge Or What The Frog Should Have Told Dretske’, Philosophical Studies 46, 109–130. Plantinga, A. (1993). Warrant: The Current Debate, Oxford University Press, Oxford. Rescher, N. (1976). ‘Peirce and the Economy of Research’, Philosophy of Science 43, 71–98. Russell, B. (1948). Human Knowledge: Its Scope and Limits, Allen and Unwin, London.
227
Swinburne, R. (2001). Epistemic Justification, Oxford University Press, Oxford. Williamson, T. (1998). ‘Conditionalizing on Knowledge’, British Journal for the Philosophy of Science 49, 89–121. Zagzebski, L. (1999). ‘What Is Knowledge’, The Blackwell Guide to Epistemology, (eds.) J. Greco and E. Sosa, 92–16, Blackwell, Oxford.
228
Grazer Philosophische Studien 69 (2005), 229–245.
VARIETIES OF CONTEXTUALISM: STANDARDS AND DESCRIPTIONS Peter BAUMANN University of Aberdeen Summary Most contextualists agree that contexts differ with respect to relevant epistemic standards. In this paper, I discuss the idea that the difference between more modest and stricter standards should be explained in terms of the closeness or remoteness of relevant possible worlds. I argue that there are serious problems with this version of contextualism. In the second part of the paper, I argue for another form of contextualism that has little to do with standards and a lot with the well-known problem of the reference class. This paper also illustrates the fact that contextualism comes in many varieties.
Only a couple of years ago, one could still easily make sense of general questions like “What do you think about epistemological contextualism?” In the meantime, so many different positions have been developed under the heading of “contextualism” that one is tempted to reply “It depends on what you mean by ‘contextualism’”. The variety has become so great that what is a serious objection to one form of contextualism might be even welcome support for another form of contextualism. Therefore, one should look at them one by one. One distinction is that between attributor and subject contextualism (cf. DeRose 1992, 918ff.; DeRose 1999, 190f.; Cohen 1987; Cohen 1988). In the meantime, it has become more popular to refer to the latter under the heading of “subject sensitive invariantism” (cf., e.g., Hawthorne 2004, ch.4) and to reserve the term “contextualism” for the former. I will go with this and take contextualism to be, broadly speaking, the following thesis: The truth-value of knowledge ascriptions of the form “S knows that p” (and of related forms, of course) may (but need not) change with the speaker’s context (or the thinker’s context, for that matter). That is, it may change from speaker to speaker or between different contexts one
and the same speaker finds herself in. To choose what is perhaps the most overused example in this context: In one (an ordinary) context it might be true to say or think that Jack knows that he has hands, but in another (sceptical) context it might not be true.1 But what is a context and how should we individuate contexts? Most contextualists seem to agree that contexts differ with respect to the relevant epistemic standards. In an ordinary context, epistemic standards are modest enough to allow for Jack to know that he has hands; in a sceptical context, however, standards are so strict that he won’t even know that. So, contexts are determined by the epistemic standards which count as the relevant ones in a given situation. It has become so much standard among contextualists to make contexts depend on epistemic standards that one could also speak of “standard”-contextualism in both senses of the word. In what follows I will take a look at a very prominent way to explain the difference between more modest and stricter standards, and will also say why I’m not happy with it. After that, I will try to make a constructive alternative proposal and defend another form of contextualism — a version that has little to do with standards and deserves much more attention than it has received so far. But first to standards. 1. Standards According to authors like DeRose (1995) and others (for a slightly different view cf. Cohen 1998, 289–291 or Cohen 1999, 57–60) S’s knowledge that p requires that S can rule out or that S’s evidence eliminates alternate possibilities incompatible with the truth of “p”. To rule out a possibility, S needs to have evidence to the effect that the possibility in question is not actual. An uneliminated alternate possibility is one in which the subject would have exactly the same experience and evidence as they actually have (cf. Lewis 1996). For lack of a better term, I will say that S “eliminates” a possibility just in case S can either rule it out or S’s evidence eliminates it. Now, which alternate possibilities does the subject need to eliminate? Well, according to the philosophers just referred to, that depends on the epistemic standards 1. I will take the stylistic freedom to switch from the metalinguistic level to the objectlevel whenever nothing hinges on it (cf. Lewis 1996, 567).
230
relevant in the context of the attributor (speaker or thinker). According to laxer standards, Jack has to eliminate fewer and less remote possibilities in order to make “Jack knows that there is a hot dog in front of him” true. For instance, he just has to eliminate the possibility that there is nothing but a fake hot dog, made out of wax, in front of him. According to stricter standards, Jack has to eliminate more and more remote possibilities in order to make “Jack knows there is a hot dog in front of him” true. For instance, he would have to be able to eliminate the possibility that he is a Cartesian dreamer dreaming of some hot dog in front of him when there is none. Given stricter standards, Jack has to eliminate all possibilities he has to eliminate under laxer standards plus some more remote possibilities.2 Given laxer standards, it is true to say that Jack knows that there is a hot dog in front of him, at least if we assume that Jack can tell wax dogs from real hot dogs. However, given stricter standards, this might not be true. One of the main points of contextualism is that there is no single “true” standard or context. There is rather a plurality of contexts and in some it is true whereas in others it is false to say that S knows that p. In this sense — but in this harmless sense only — contextualism is a form of relativism. There are several general objections one could make against the version of contextualism just sketched (and I really only want to talk about the basic idea here and neglect finer differences between different versions of it). First, one might complain about a certain circularity in talk about “eliminating possibilities”. Suppose we claim that Jack knows there is a hot dog in front of him and that he can eliminate the possibility that it is only a wax imitation of a hot dog. If “eliminating the possibility that it is an imitation” means something like “knowing that it is no imitation”, then we are explaining “knowledge that there is a hot dog” in terms of “knowledge that it is not something else”. We are using the explanandum in the explanans and that is not good. What if “eliminating the possibility that q” does not mean or imply “knowing that not q”? Well, what could it mean then? Something like “having good reasons to believe that not q”? But even good reasons can be 2. It seems to be a (usually implicit) assumption of many contextualists that if S can eliminate a more remote possibility, then they can ipso facto eliminate a closer possibility. It is not obvious at all whether this is the case. Is it not conceivable that I can eliminate the possibility that I am a robot from Mars (remote possibility) but that I cannot, for some reason, eliminate the possibility that I was born on the 5th of May rather than, say, the 4th of May at 11.45 pm?
231
misleading. Jack has excellent reasons to believe there is no wax hot dog in front of him; however, unbeknownst to him, he is confronted with the latest high tech wax imitation of a hot dog and cannot tell it from a real hot dog. He therefore doesn’t know there is a hot dog in front of him. In other words, it seems we are facing a dilemma: Either we are reading “eliminating” in a sense that is too weak (e.g., as “having good reasons”); or we are reading it in a strong enough sense but then the situation is even worse because we smuggle the explanandum into the explanans. Hence, one might conclude, all this stuff about “eliminating possibilities” does not lead anywhere (on “ruling out” see also Dretske 1970 and 1981b). The short reply to this objection is that we need not be interested at all in a reductive explanation of the concept of knowledge here (who has ever managed to reductively define concepts like that?). Apart from that, the objection only seems to work for the internalist interpretation of “eliminating a possibility” as “ruling it out”. The objection does not seem to work for, say, Lewis’ more externalist talk about “eliminating possibilities”. Hence, the objection above is, one might reply, without force. I think one can reformulate the objection but there is a much more serious problem with the contextualist story about eliminating possibilities. It concerns the underlying assumption that one can rank possibilities or possible worlds, for that matter, in terms of their remoteness from the actual world (compare, for instance, DeRose 1995, 36–37 and, as background, Nozick 1981, 172ff. on sensitivity. See also, in general: Williams 2004). Our contextualist wants to say that in a stricter context the subject has to eliminate all the possibilities she has to eliminate in laxer contexts plus more remote possibilities. It is false to say “Jack knows that he has hands” if he can only eliminate the possibility that he is a brain in a vat but not the possibility that he has stumps instead of hands.3 If there is something wrong with the idea of a ranking of possible worlds in terms of remoteness, then this version of contextualism collapses, as we will see. So, what could be wrong with the idea of a ranking? Take the usual question whether Jack knows that he has hands. Let us assume that Jack has (two) hands. According to the kind of 3. This does not contradict the last footnote: The fact that knowledge requires that one can eliminate both the more remote and the closer possibility does not imply that one can indeed eliminate both.
232
contextualism under consideration, it is true to say “Jack knows that he has hands” if the attributor finds herself in an ordinary context in which the standards only demand that the subject be able to eliminate close possibilities (like the one that he’s been in a car accident and lost both hands, and similar horror scenarios). However, it is false to say “Jack knows that he has hands” if the attributor finds herself in a sceptical context in which the standards also demand that the subject be able to eliminate remote possibilities like the one that he is just a handless brain in a vat who is merely imagining that he has got hands. The whole question I want to raise is this: Why should one believe that a brain in a vat-world is a remote possible world (cf. also Neta 2003, 16, fn. 51 who raises this question en passant)? Why believe a world in which Jack was involved in an accident leading to the loss of his hands is closer to the actual world (or to what we take to be the actual world)? Usually, the closeness or remoteness of possible worlds is spelled out in terms of degrees of similarity between the worlds considered (cf. Lewis 1973, 48–52, 66–67; Lewis 1986, 20–27). Let us call a world in which Jack is involved in an accident an “accident-world” and a world in which he is just a brain in a vat a “vat-world”. Why should one think that an accident-world is more similar to actuality than a vat-world is? Compare an accident-world with very different laws of nature with a vat-world with our laws of nature. Which one is more similar to the actual world? Ask a physicist and he’ll probably tell you (if he is willing at all to engage in that kind of speculation) that the vat-world is more similar. What matters to him is not so much whether he or Jack or we all are envatted but what the correct physics for the world under consideration is. Then ask an epistemologist and he’ll probably give you the opposite answer. What matters to him is not so much the laws of nature but what his or Jack’s or our epistemic situation is.4 4. The physicist’s perspective is a third person perspective whereas the epistemological perspective is a first person one: It is about “my” epistemic situation. I won’t pursue things in terms of this aspect of the difference here. By the way: It seems obvious that there are many epistemologists who would and do in fact respond to the questions above in the way I indicated. Physicists seem less interested than some epistemologists in ranking possible worlds but as far as I can see they would also rather tend to respond in the way indicated. Just ask one! In the end, however, that does not matter too much for the argument. What matters is that nothing having to do with the characteristics of the different possible worlds gives us a reason to answer the above questions about closeness one way or another.
233
So, ironically it all seems to depend on the context. Epistemologically, the accident-world is closer but ontologically the vat-world is closer to the actual world. A case like this suggests that there is no remoteness-ranking of worlds as such. Remoteness or closeness of possible worlds is relative to a context in which different standards of importance (physics, epistemology) are the ones that count. There is no contextindependent hierarchy of contexts or standards in terms of strictness or laxness (cf. also Williams 1996 and Heller 1995, 505–507).5 The irony is that the contextualist story of people like DeRose and others is blocked by an unwelcome context-dependency.6 If there is a plurality of equally acceptable but mutually incompatible rankings of possible worlds in terms of remoteness and closeness, then we might, for instance, have to deal with two rankings like the following ones: one ranking according to which the possibility that I am a brain in a vat (or a robot from Mars) is less remote than the possibility that I have either 1 or 3 hands, and another, second ranking which gives the reverse order. Given such a plurality of rankings, the contextualist story against the sceptic seems to collapse: We cannot “secure” an ordinary context from sceptical threats anymore because one could always see the latter as very close to home (too close for a contextualist response to the sceptic). Apart from that, and more generally, the contextualist account of how the truth of knowledge claims varies with context shifts would lose its plausibility if spelled out in terms of rather arbitrary remoteness-rankings. 5. DeRose (1992, 922, ftn. 18) points out (following Unger 1986) that we have different epistemic standards for different aspects of beliefs: for the degree of confidence, for the degree of non-accidentality of the belief, etc. In some contexts the standards for the necessary degree of confidence are very strict and the standards for non-accidentality are rather relaxed whereas in other contexts it is the other way around. Hence, we should not expect a unique hierarchy of contexts. Whatever one would have to say about this point, it is obvious that it is quite different from the point made in the text above which deals with just one aspect or dimension, namely the closeness or remoteness of possible worlds. 6. All this should not come as a surprise if one explains closeness in terms of similarity of worlds, like Lewis. The notion of similarity is notoriously vague, multi-dimensional and context dependent (cf. Fine 1975, 451–458; Jackson 1977, 4–8; Slote 1978, 20–25 and Heller 1999, 116; see also Goodman 1972 and Tversky 1977). Lewis himself accepts the context dependency of the notion of similarity (cf. Lewis 1973, 91–95) and thus of the closeness of possible worlds (cf. Lewis 1973, 50–52, 66–67). Jackson, Slote and others offer alternative accounts of counterfactuals but I doubt these accounts can make the above problem go away: It is a very general one.
234
One way to respond to this would be to propose to take the second context-dependency on board and thus “deepen” and “radicalize” contextualism. What we then get is a “two-step”-contextualism. First, whether it is true to say that S knows that p depends on how far out the possibilities are that S has to eliminate, according to the attributor. Second, what is far out or really close depends on standards of relevance used by the attributor (not the subject). There are two dimensions of context-dependency and the latter one is more fundamental than the former one (the one DeRose and others have in mind). In the end, it still all depends on the attributor’s standards of relevance. I am sceptical about this kind of reply. It rescues the DeRose-style of contextualism by transforming it into something different. This move still seems to make things a bit arbitrary: It all depends on what an attributor happens to think is relevant. You might think this is relevant and I might think that is relevant, and that seems to be the end of the discussion before it has even really started. Apart from that, this doesn’t seem to offer any basis for a somehow convincing reaction to scepticism. Contextualism of this kind would be much less attractive for those contextualists who also expect a reply to the sceptical challenge from contextualism. As mentioned above: The contextualist couldn’t secure an ordinary context from sceptical threats anymore. One might try to avoid all this and just say that what matters here is epistemological closeness or remoteness, not ontological or any other kinds of ranking. There is a certain plausibility to this: Aren’t we talking about epistemology here?7 But there are further problems if one chooses this kind of reply. What determines epistemological closeness or remoteness? Intuitively it seems pretty clear that a world in which I am a brain in a vat is epistemically much more remote from what I take to be the actual world than a possible world which differs from the actual world (or what I take to be the actual world) only insofar as in that world I know certain facts about monkeys which I don’t know in the actual world. Epistemological closeness or remoteness seems to depend on how much I know in the relevant worlds and what the difference as to the amount of my knowledge is in these worlds: I know some things, it seems, in the actual world, a bit more in the close 7. It is interesting to see that Lewis chooses ontological criteria of remoteness; whether that is compatible with standard contextualism (given the remarks above) must be left open here (cf. Lewis 1973, 75–76; Lewis 1979, 472).
235
possible world (where I know all that stuff about monkeys) but I know nothing about the external world in the vat-world. “Knowing quite a bit” is much closer to “Knowing a bit more than that” than to “Knowing nothing”. One might want to add considerations of “quality” to such ideas of quantity: It also matters how important the relevant pieces of knowledge are. So, it would be a weighed sum of the quantity and quality of knowledge that determines the epistemological closeness or remoteness of possible worlds. There are a lot of problems with this: How to weigh the quantitative against the qualitative aspect? What determines both aspects anyway? That is, how can one spell out in a reasonable way all that talk about importance of pieces of knowledge? And how could one possibly quantify how much someone knows? I don’t want to go into any of these problems. Whether one can deal with them or not doesn’t concern me here. I only want to offer a rough picture of this version of the standard story and point at a much more simple problem with it: the threat of a vicious regress. According to our proposal, the truth-value of “Jack knows that he has hands” depends on how far out, epistemologically, the possibilities are that he has to eliminate (according to the attributor). How far out a given possible world is, depends on the difference as to what the subject knows in both worlds. In short, the truth-value of “Jack knows that he has hands” depends on what Jack knows in the actual world and on what he knows in all other relevant possible worlds. Since we are contextualists here, we should — when we are taking things strictly — stay at the meta-linguistic level and say this: The truth-value of “Jack knows that he has hands” depends on the truth-values of all sentences attributing knowledge to Jack as evaluated as true or false in the actual world and on the truth-values of all such sentences as evaluated as true or false in all the other relevant possible worlds. The truth-value of “Jack knows that he has hands” thus depends on the truth-value of many, many other knowledge sentences. It seems obvious that the truthvalue of those other sentences will, again, depend on the truth-value of still other sentences. Hence we have an infinite regress (the alternatives of circularity or an arbitrary stop somewhere are not attractive). This is bad not so much because the explanans uses the explanandum (we don’t have to try to define the concept of knowledge) but rather because it makes truth-values of knowledge sentences dependent on each other and thus “hanging up” in the air. We could no longer understand what the truth conditions of some particular knowledge-sentences are because
236
we’re always referred to the truth conditions of still other knowledgesentences. One might, finally, try to explain epistemological closeness or remoteness of worlds not in terms of “knowledge” but rather in terms of other epistemological concepts, like justification. I won’t go into this but just remark that I have doubts that this will help a lot. Even if one can avoid the regress I would guess that the account would be too weak to explain ‘knowledge’ (see the problem mentioned in a similar context above). The upshot of all this is that standard-contextualism à la DeRose and others faces serious difficulties. More generally, it seems very hard to imagine how one could come up with an interesting hierarchy of contexts that is in itself neither context dependent nor arbitrary. Sure, in some respects so-called “sceptical” contexts or standards are more remote than at least some of the so-called “ordinary” contexts or standards but there are other respects in which it is just the other way around. As Michael Williams has pointed out (1996; 2004), if a historian starts having doubts about the reality of the past she is not replacing laxer by stricter standards but rather changing the topic. Rather than a hierarchy of contexts and standards we should expect a diversity and plurality of them without a hierarchical order. If all this is true, then both the positive (standard) contextualist account of knowledge and the contextualist reply to the sceptic become quite implausible. All this does, of course, not imply that the notion of epistemic standards should not play an important role for contextualists. It just cannot be used for the tasks some contextualists want to use it for. But there certainly are different standards corresponding to different contexts. Mary is a meteorologist and it might be true to say of her in a lay-context that she knows that it will rain later today but not true in a professional context of her weather forecast lab: She hasn’t run the typical tests yet. Apart from that, there is a wide variety of kinds of things that could be called “standards”. In some contexts, a true knowledge ascription presupposes that the subject meets certain standards of justification, etc. In other contexts, however, not a lot more than true belief might be sufficient for a true knowledge ascription (cf. Ernst 2002). Sometimes knowledge is compatible with a fixed “epistemic position”, that is, an epistemic position that cannot — within the confines of the situation at hand — get better or worse (e.g., epistemic lottery situations); sometimes knowledge is not compatible with a fixed epistemic position (cf. Baumann 2004).
237
So, what we can learn from all this is not that contextualists should completely forget about the idea of standards. It is rather that they should put much (much!) less theoretical weight on it. Those variants of contextualism that work with remoteness-rankings of possible worlds run into insurmountable problems. And: There are other, rather neglected, aspects of the context-sensitivity of knowledge ascriptions that have little or nothing to do with epistemic standards. To that and the more constructive part of this paper I turn now. 2. Descriptions I don’t think one will ever be able to define the concept of knowledge. But even so, there is a very useful (partial) explanation of “knowledge”: Knowledge requires a true belief that has been acquired in a reliable way. Usually, “reliability” is taken in an externalist sense (cf. Goldman 1986, 103; Dretske 1981a, ch.4) but I want to take it in a much broader sense, including internalist ways of belief acquisition (good reasoning while being aware of the rules of correct reasoning, for instance, is a reliable way of belief acquisition). Let us call the way the subject acquires her belief her “method” of belief acquisition. Again, I want to use that word in a very broad sense: It should not imply that the subject uses the method intentionally or consciously (perception would be an example here). There is a problem that has caused a lot of headaches for reliabilists: the so-called generality problem (cf. Feldman 1985; Alston 1995; Conee and Feldman 1998; Pollock 1984; Bach 1985; Brandom 1998; Beebe 2004; Hudson 2004). There is usually more than one way to specify the method of belief acquisition. Joe just acquired the belief that there is a desk in front of him but what exactly was the method? Perception, seeing, seeing-with-his-glasses-on, looking-at-furniture-with-his-glasseson-while-suffering-from-a-remarkable-sleep-deficit? Depending on the specification of the method we can get different degrees of reliability, and thus, even different truth-values for my claim that Joe knows there is a desk in front of him. Since there is no way to choose among these different specifications, reliabilism turns out to be an empty and therefore useless theory. So much for a short description of the generality problem. I think it is a specific version of a much more fundamental problem, namely the
238
reference class problem. What is the probability that an individual a is F (that Jack will live to the age of 72 years)? It depends on the relevant reference class a belongs to (the class of smokers, city dwellers, bikers, smoking city bikers, etc). Different reference classes give us different probabilities and the problem is to pick the one relevant reference class. There is widespread pessimism concerning the possibility of a solution of the reference class problem. The generality problem is, I said, just one aspect of this much broader problem. Even if we hold the method fixed (assuming that we have solved the generality problem), there will still be a reference class problem for reliabilism (cf. Wallis 1994). It doesn’t only concern the individuation of the method but also the determination of the reliability of a given method. Does that mean that the prospects for reliabilism are even gloomier than the generality problem suggested? No, I think there is a different and quite constructive lesson to be learnt, and contextualism delivers it; in a sense, a particular version of contextualism is the “solution” of the generality problem and related problems (cf. also Heller 1995 and Cohen 1988, 115 who shortly mentions this issue). But let me explain. I will focus on one phenomenon which one can call “the time-sensitivity of reliability” (there is a parallel spatial case but I won’t go into that here). Consider the following example. Jack is the only witness in some court-case. The question is whether asking Jack about what happened is a reliable method of finding out the truth. This depends, of course, on how reliable Jack is. Suppose that Jack is reliable during the whole week in which the trial takes place. This is compatible with the fact that Jack suffers from a single momentary blackout between 5 to 10 am and 10 am on Wednesday. Only Jack notices that he suffers from a blackout. He happens to give correct answers to questions during his blackout. A single short blackout during the whole week does not put his general reliability as a witness during the whole week into doubt. Who is reliable at every moment in time? Suppose further that Jack is being interrogated just between 5 to 10 and 10 on Wednesday. With regard to this time-span of five minutes, Jack is, of course, not reliable. Suppose further that the trial takes place during summer and that Jack is a periodical drunkard who consumes alcohol only during the cold seasons (and then in abundance). If Jack is drunk, then he is completely unreliable. The situation then is as follows:
239
Jack • is no reliable witness between 5 to 10 am and 10 am on Wednesday; • he is a reliable witness during the whole week; • he is no reliable witness with regard to the whole year. Similar things are true, mutatis mutandis, for the reliability of finding something out by relying on Jack’s testimony. We can ad libitum consider further time-intervals with regard to which Jack might or might not be a reliable witness. The important general point is that reliability (e.g., of Jack’s testimonies and of reliance on his testimonies) is or can be time-sensitive. There can be and often is a great variation of probability with chosen time spans (like in the example just given). In that case, there is no reliability tout court but only reliability with respect to some time or time interval (and unqualified talk about reliability loses its meaning). Reliability is time-sensitive if the corresponding probability distribution over time is not constantly yielding the same value (or very similar values) but rather markedly different values for different times or time intervals. A constant probability distribution might be the exception. Hence, we can assume that normally or very often probability (and thus, reliability) varies with time. A classical answer to this kind of problem says that one should choose the smallest reference class as the relevant one (cf. Reichenbach 1994, 383; on different but ultimately similar lines cf. Hempel 1965, 53–79; 397–403, and Salmon 1966, 90–92; cf. also Beebe 2004, 181). The problem with this and other, similar, proposals is not just that sometimes there is no smallest reference class (is there a shortest time interval?). More seriously, even if we have identified a smallest reference class the class will often be too small to allow for statistical information. It might be very small or even just consist of the one element we are considering. Furthermore, it is very hard to even make sense of probabilistic statements concerning individual cases or very small samples (but this is, of course, not uncontroversial and leads to more general questions concerning the interpretation of probability). There are more problems. Take our Jack-example. Why should the shortest time-interval be the relevant one in the first place? Let us consider the proposal that the 5 minutes during which Jack is indeed questioned by the judge constitute the relevant time span. Why should
240
that be so? One might say “Well, because that is when Jack gave his answers!” But what does that have to do with Jack’s reliability? And, by the way, weren’t his answers all correct? The point here is a bit subtle. We are interested in the 5 minutes of the interrogation because we are interested in the content and truth-value of Jack’s answers but not because we are interested in Jack’s reliability as a witness. That is a different issue. Consider this analogy. My notebook is reliable (in a non-epistemic sense). But just now it happens to freeze. Does that mean it is not reliable? Or not reliable at this moment? No, it just means that even a reliable notebook freezes from time to time. What would we say if we changed the example a bit and assumed that Jack was fine (no blackout or anything like that) during the time of questioning but never fine (say, always drunk or hungover) at other times? I don’t think we would be forced to consider him reliable; many would rather want to say he is not reliable, even if his answers happened to be correct, and even if he was reliable during the questioning period. Apart from that, one attributor might be quite “liberal” and just require that the witness be reliable during his testimony whereas another attributor might “raise” the bar and require that the witness ought to be reliable during the whole day, week, month, etc. There is no reason why the latter should be considered “wrong” in any sense. Even if we assume that the time during which Jack is being questioned is the relevant one here: What determines his reliability at that time, between 5 to 10 and 10? In order to answer that question, we need to determine the relevant reference class, again: Is it the class of all summer interviews? Or of interviews done when his stomach feels in this particular way? We are getting into a regress. In other words, there seems no way to get rid of the indeterminacy of the relevant reference class. One might propose to determine the relevant reference class at the time between 5 to 10 and 10 by using a possible worlds approach: Jack’s reliability during the relevant time depends on how easily he could have gotten things wrong. But what determines the closeness or remoteness of possible worlds in which Jack gives false answers? This leads back to the objections made above against the idea of a remoteness-ranking of possible worlds. What does all this imply? It seems that there just is no single right or relevant temporal reference class. We, the attributors of knowledge and reliability, must pick one — whether we are clearly aware of this or
241
not. In the end, it all depends on our practical and theoretical interests. The prosecution might have an interest to use Jack in other trials as a witness, too; hence, it will (tend to) judge Jack’s reliability according to a longer time-interval. The defence might be rather interested to interrogate Jack in this particular phase of the trial (at the beginning of the Wednesday session); it will thus (tend to) judge Jack’s reliability according to a shorter time interval (e.g., the interval between 5 to 10 am and 10 am). Now, given what I have said above, different interests and different criteria of relevance yield different views about the corresponding relevant probabilities and, thus, different views of Jack’s reliability and of the reliability of the judge’s method of interrogating Jack. And all those different descriptions are equally legitimate and adequate. This does, of course, not exclude that we very often agree amongst each other in our practical interests or on the criteria of relevance. Sometimes, there are even established and institutionalized rules (e.g., for what counts as good evidence or as trustworthy witnesses in court). And sometimes, there are such rules because without them we would not agree with each other. We can thus say the following, given that knowledge presupposes the reliability of the method used: The truth-value of “The judge knows (by Jack’s testimony) that the car was parked at the crime scene” varies with the attributor’s view or description of the relevant temporal framework. Analogous points can be made about space. And finally, there is also the initial generality problem concerning the individuation of the method used. So, with respect to all these three aspects we should, I think, give a contextualist analysis of knowledge sentences. In this sense, contextualism “solves” the generality problem and the underlying reference class problem for “knowledge”: It doesn’t make the point go away, it just tells us that it really isn’t a big problem after all. This brand of contextualism is, of course, different from what I have called “standard-contextualism”: It doesn’t have to do with standards but rather with descriptions (of spatio-temporal frameworks and of methods of belief acquisition). One could call it “description-contextualism”. The latter form of contextualism does not seem to suffer from the problems of the former that I discussed above.8 However, I don’t want 8. One more (potential) relative advantage: Description-contextualism is not or at least not to the same degree susceptible to “warranted assertability maneuvers” (compare, for instance, DeRose 1999, 195–203). It is hard to see how one could “wam” description-contextualism.
242
to give up all forms of standard-contextualism but rather reject only those forms of it that are close enough to DeRose’s and related ones. I think that all this also shows that contextualism comes in more than one variety. It has at least two dimensions: standards and descriptions (perhaps there are even more than these two). A context dependency in one dimension does not imply anything about context dependencies in the other dimension; in that sense both aspects are independent from each other. To cut a long story short: There is more than one legitimate version of contextualism.9
REFERENCES Alston, W. P. (1995). ‘How to Think about Reliabilism’, Philosophical Topics 23, 1–29. Bach, K. (1985). ‘A Rationale for Reliabilism’, The Monist 68, 246–263. Baumann, P. (2004). ‘Lotteries and Contexts’, Erkenntnis 61, 415–428. Beebe, J. R. (2004). ‘The Generality Problem, Statistical Relevance and the Tri-Level Hypothesis’, Noûs 38, 177–195. Brandom, R. (1998). ‘Insights and Blindspots of Reliabilism’, The Monist 81, 371–392. Cohen, S. (1987). ‘Knowledge, Context, and Social Standards’, Synthese 73, 3–26. — (1988). ‘How to Be a Fallibilist’, Philosophical Perspectives 2, 91–123. — (1998). ‘Contextualist Solutions to Epistemological Problems: Scepticism, Gettier, and the Lottery’, Australasian Journal of Philosophy 76, 289–306. — (1999). ‘Contextualism, Skepticism, and the Structure of Reasons’, Philosophical Perspectives 13, 57–89. Conee, E. and Feldman, R. (1998). ‘The Generality Problem for Reliabilism’, Philosophical Studies 89, 1–29. DeRose, K. (1992). ‘Contextualism and Knowledge Attributions’, Philosophy and Phenomenological Research 52, 913–929. 9. I would like to thank Martijn Blaauw, René van Woudenberg, an anonymous referee, and an audience at the Vrije Universiteit Amsterdam, the Netherlands, for comments and discussions. For discussion and comments on former versions of this paper I am grateful to Berit Brogaard, Duncan Pritchard, and Barry Smith. I would also like to thank audiences in Göttingen, Tübingen, Erfurt, Santiago de Compostela, and Dublin.
243
DeRose, K. (1995). ‘Solving the Skeptical Problem’, Philosophical Review 104, 1–52. — (1999). ‘Contextualism: An Explanation and Defense’, The Blackwell Guide to Epistemology, (eds.) J. Greco and E. Sosa , 187–205, Basil Blackwell, Oxford. Dretske, F. I. (1970). ‘Epistemic Operators’, Journal of Philosophy 67, 1007–1023. — (1981a). Knowledge and the Flow of Information, MIT Press, Cambridge/ MA. — (1981b). ‘The Pragmatic Dimension of Knowledge’, Philosophical Studies 40, 363–378. Ernst, G. (2002). Das Problem des Wissens, mentis, Paderborn. Feldman, R. (1985). ‘Reliability and Justification’, The Monist 68, 159–174. Fine, K. (1975). ‘Critical Notice’ [of Lewis 1973], Mind 84, 451–458. Goldman, A. I. (1986). Epistemology and Cognition, Harvard University Press, Cambridge/MA. Goodman, N. (1972). ‘Seven Strictures on Similarity’, Problems and Projects, 437–447, Bobbs-Merrill, New York. Hawthorne, J. (2004). Knowledge and Lotteries, Oxford University Press, Oxford. Heller, M. 1995, ‘The Simple Solution to the Problem of Generality’, Noûs 29, 501–515. — (1999). ‘The Proper Role for Contextualism in an Anti-Luck Epistemology’, Philosophical Perspectives 13, 115–130. Hempel, C. G. (1965). Aspects of Scientific Explanation and other Essays in the Philosophy of Science, Free Press/ Collier, New York/ London. Hudson, R. G. (2004). ‘The Generality Problem’, Southern Journal of Philosophy 42, 193–211. Jackson, F. (1977). ‘A Causal Theory of Counterfactuals’, Australasian Journal of Philosophy 55, 3–21. Lewis, D. (1973). Counterfactuals, Basil Blackwell, Oxford. — (1979). ‘Counterfactual Dependence and Time’s Arrow’, Noûs 13, 455– 476. — (1986). On the Plurality of Worlds, Basil Blackwell, Oxford. — (1996). ‘Elusive Knowledge’, Australasian Journal of Philosophy 74, 549–567. Neta, R. (2003). ‘Contextualism and the Problem of the External World’, Philosophy and Phenomenological Research 66, 1–31. Pollock, J. (1984). ‘Reliability and Justified Belief ’, Canadian Journal of Philosophy 14, 103–114.
244
Reichenbach, H. (1994). Wahrscheinlichkeitslehre — eine Untersuchung über die logischen und mathematischen Grundlagen der Wahrscheinlichkeitsrechnung, Vieweg, Braunschweig. Salmon, W. C. (1966). The Foundations of Scientific Inference, University of Pittsburgh Press, Pittsburgh, PA. Slote, M. A. (1978). ‘Time in Counterfactuals’, Philosophical Review 1987, 3–27. Tversky, A. (1977). ‘Features of Similarity’, Psychological Review 84, 327– 352. Unger, P. (1986). ‘The Cone Model of Knowledge’, Philosophical Topics 14, 125–178. Wallis, C. (1994). ‘Truth-Ratios, Process, Task, and Knowledge’, Synthese 98, 243–269. Williams, M. (1996). Unnatural Doubts. Epistemological Realism and the Basis of Scepticism, Princeton University Press, Princeton. — (2004). ‘Knowledge, Reflection and Sceptical Hypotheses’, Erkenntnis 61, 315–343.
245
This page intentionally left blank
Grazer Philosophische Studien 69 (2005), 247–253.
CONTEXTUALISM BETWEEN SCEPTICISM AND COMMON-SENSE Jesper KALLESTRUP University of Aarhus, Denmark Summary This paper examines two recent objections against contextualism. The first is that contextualists are unable to assert their own position, and the second is that contextualists are forced to side with common-sense against scepticism. It is argued that once we get clear on the commitments of contextualism, neither objection succeeds in what it aims to show.
I Sceptical arguments are paradoxes that purport to derive an incredible conclusion from prima facie credible premises. To give a satisfactory response is thus not merely to deny a premise or reject a rule of inference; rather it is to explain in addition how the premises and the reasoning can seem so plausible yet yield a conclusion so implausible. Consider what has become the paradigm sceptical argument (PSA): (P1) ¬K¬SH (P2) ¬K¬SH → ¬KP (C1) ¬KP — where P is some ordinary proposition inconsistent with the sceptical hypothesis SH, and K is the knowledge-operator. There are three dominant anti-sceptical responses. The sensitivitytheorist, e.g., Nozick (1981), accepts (P1) but rejects (P2) and (C1): a necessary condition on K¬SH and KP is that B¬SH and BP be sensitive, but SH l→ ¬B¬SH is false, while ¬P l→ ¬BP is true; B is the belief-operator. Hence
(Closure) (KP; K(P → Q)) → KQ, on which (P2) relies, must be jettisoned. The safety-theorist, e.g., Sosa (2000), however, accepts (P2), but rejects (P1) and (C1): a necessary condition on K¬SH and KP is that B¬SH and BP be safe, and both B¬SH l→ ¬SH and BP l→ P are true. A safety-based theory of knowledge meets the sceptic head-on by allowing that P is knowable without being able to rule out far-off ¬P-possibilities. According to the safety-theorist, the intuition supporting the truth of (P1) is explained away by the fact that sensitivity resembles, but is distinct from, safety. Counterfactuals don’t validly contrapose. But of course, pace Sosa (op. cit.), a similar move is available to the sensitivity-theorist. The intuition supporting the falsity of (P1) is explained away by the fact that sensitivity resembles, but is distinct from, safety. At first blush neither response is entirely suitable. There is a firm intuition behind (P2), which the sensitivity-theorist has not addressed. And if the safety-theorist concedes the equally firm intuition supporting (P1), then, by (Closure), there ought also to be a firm, albeit false, unexplained intuition behind (C1). The third response is the contextualist attempt to explain all the conflicting intuitions without actually explaining any of them away. According to contextualism, the truth-conditions of knowledge-ascribing/denying sentences vary with shifts in contextual parameters.1 In a sceptical context S the standards that one’s evidence has to meet in order for one’s true belief to count as knowledge are raised so high that (P1) is true. And since (Closure) is an intra-contextual principle, (C1) is true in S. But when the standards are low as they are in ordinary contexts O, (P1) and (C1) are both false. So, there are only equivocal ways of inferring the truth of (C1) in O from the truth of (P1) in S. Accordingly, the primary motivation behind contextualism, as in DeRose (1995) and Cohen (2000), is two-fold: to safeguard ordinary claims to knowledge and to explain the persuasiveness of (PSA). II Let’s briefly review the commitments of scepticism and common-sense. The sceptic claims not only ¬KSP, but also KS¬KOP, hence ¬KOP given 1. Since the objections that I am about to discuss apply to both subject- and ascriber-
248
that knowledge is factive; the subscripts S and O indicate the sceptical and ordinary contexts respectively. In fact, if we ignore relativism about truth, then we know in any context that if we know P in some context, then P is true in all contexts. So, the conjunction of ¬KSP and KSKOP is inconsistent: KSKOP and KS(KOP → P) together entail KSP by (Closure). Common-sense, however, claims not only KOP, but also KOKSP, hence KSP. The safety-theory is a way of encoding common-sense such that raising the standards by making SH contextually salient makes no difference to what is known. The contextualist thinks the sceptic is right that ¬KSP, but wrong that KS¬KOP, but also that common-sense is right that KOP, but wrong that KOKSP. All this is knowable in the context of contextualism C, so the contextualist endorses KC¬KSP and KCKOP. And given that the contextualist wants to hold ¬KSP, she must also accept ¬KSKOP. Once standards go up one loses knowledge whether one knows when standards are low. The contextualist thus agrees with both scepticism and common-sense about certain knowledge-claims, but also disagrees with both views about certain other such claims. However, the upshot of the two objections to contextualism that I will discuss in the following section is that they collapse into scepticism and common-sense respectively. I suggest that they misfire by paying insufficient attention to the contextualist’s commitments. III The first objection, in Williamson (2001), is that contextualists are unable to assert their own view. In C it is known that it is known in O that P. Knowledge is factive, so P is true in O. But contextualism about knowledge need not entail relativism about truth, so P is true in C as long as P does not contain ‘know’. Now it is not known in S that P. But C = S since both are epistemological contexts. So, it is not known in C that P. So, in C contextualists are committed to ‘P but it is not known that P’, which is Moore-paradoxical.2 Consider: contextualism, I shall leave it open whether these parameters pertain to the subject, or the ascriber, of knowledge. 2. A similar problem arises for contrastivism (Schaffer 2004) according to which the sentence ‘S knows that P’ expresses the three-place relation S knows that P rather than Q,
249
(P1*) (C1*) (P2*) (C2*) (C3*)
K OP P ¬KSP ¬KCP P & ¬KCP
Contextualist Assumption Factivity Contextualist Assumption C=S (C1*), (C2*)
A little reflection shows that contextualists will reject the assumption C = S. They hold both KCKOP and ¬KSKOP. But then, by substitution, if C = S, one would come to know and not know the same proposition in the same context. If C = S, the contextualist would have to side with the sceptic against common-sense. From her perspective, the consoling thought that KOP would be unavailable after having realised that ¬KSP. One may reply that (C1*) and (P2*) are bad enough as they commit the contextualist to asserting ‘P & ¬KSP’. But suppose we are in S. Then P is not known, and so although ‘P & ¬KSP’ is true in S, one is not in a position to assert this sentence, if as Williamson (2000) has argued, it is independently plausible that one should assert only what is known. If instead we are in C, then P is known, but asserting ‘P & ¬KSP’ in C is not Moore-paradoxical, if C ≠ S. By contextualist lights, asserting P, or even KP, in one context is compatible with asserting ¬KP in a different context. All I have said is that C = S conflicts with crucial contextualist commitments. Obviously C ≠ S needs to be independently justified. I shall not attempt to do that here. On the other hand, assuming C = S in an argument against contextualism calls for more justification than just remarking that both contexts are epistemological. IV The second objection, in Wright (2005), has the exact opposite effect, namely that contextualists are forced to side with common-sense against scepticism. The contextualist holds that it is known in C that it is known in O that P. But knowledge is factive, so it is known in C that if it is where Q is a contextually generated contrast proposition. Suppose that S knows that P rather than R, where R is some ordinary contrast proposition. Knowledge is factive, so P is true. But S doesn’t know that P rather than SH. So S is in a position to assert (in any context) ‘P but I do not know that P rather than SH’, which to my ear at least sounds odd.
250
known in O that P, then P is true in O. Contextualism about knowledge need not entail relativism about truth, so if P is true in O, P is true in C provided that P is knowledge-free. So, by (Closure), P is known in C. Consider: (P1**) KC(KOP → P) (P2**) KCKOP (C1**) KCP
Factivity Contextualist Assumption (Closure)
Contextualism, in Wright’s view, is supposed to be the claim that scepticism and common-sense are both right relative to S and O respectively, and so is meant to sustain the idea that in C neither the sceptic’s claim ¬KP nor the common-sense claim KP carries stronger weight. But (C1**) supposedly demonstrates that this ‘intended even-handedness’ cannot be achieved. The contextualist must picture the sceptic as cognitively blind to knowable truths. But this gets the thrust of contextualism wrong: it is the view that knowledge is context-dependent, and not the view that views about knowledge — scepticism and common-sense — are context-dependent. Common-sense and scepticism embody invariantist conceptions of knowledge that are at odds with contextualism not only in O and S, but also in C. Given that these conceptions issue in incompatible knowledge-claims, the contextualist must take a stand. Agnosticism is unavailable — it can be that neither KCP nor KC¬P, but it must be that either KCP or ¬KCP. As contextualism aims to protect KP in non-sceptical contexts, it is no surprise that P is known in C if C S. Just as we lose knowledge in S that we know P in O when we lose knowledge in S that P, we gain knowledge in C that P when we gain knowledge in C that we know P in O. What drives contextualism is the rather less ambitious thought that scepticism and common-sense are both right about certain knowledge-claims that pertain to conflicting intuitions in (PSA). By making these fairly balanced concessions, the contextualist believes she can do a better job in resolving this conflict than the safety- and sensitivity-theorists.3 3. Note that if C = S, then (C1**) entails KSP, which we assume is false. In fact there is, as Wright (op. cit.) points out, another way of showing that the assumption C = S leads to a contradiction. The contextualist maintains both KCKOP and KC¬KSP. KCKOO entails KCP. But if C = S, then KC¬KSP entails ¬KCP by substitution and factivity.
251
V The foregoing suggests that, on the one hand, it is wrong to portray contextualism as conceding everything to scepticism. The contextualist accepts ¬KSP and ¬KSKOP, but she disputes ¬KOP and therefore also KS¬KOP. So, the loss of knowledge in S is not cross-contextual as the sceptic urges; it is knowledge easily gained in O. On the other hand, it is equally wrong to portray contextualism as conceding everything to common-sense. The contextualist accepts KOP, but she disputes KSP and therefore also KOKSP. So, the gain of knowledge in O is not crosscontextual as common-sense demands; it is knowledge easily lost in S. So, we have to surrender assumptions integral to both scepticism and common-sense. Contextualism should thus not be construed as attempting to fully accommodate these views in their respective contexts. The more modest motivation is to protect ordinary claims to knowledge while being able to lift the explanatory burden imposed by (PSA). Once we get clear on the underlying conceptions of knowledge, we can see how ¬KSP and KOP can both be true. What the contextualist will say when the sceptic moves from ¬KSP to ¬KOP is that his argument needs SH, but making SH sufficiently salient is to illegitimately shift the standards for knowledge. There is no non-question-begging way of showing that one cannot have knowledge in O by pointing out that one cannot have knowledge by standards that are much higher than those in O. Similarly, what the contextualist will say when the proponent of common-sense moves from KOP to KSP is that such an argument illicitly assumes that in order to know P in S, one need only have knowledge by standards much lower than those in S. Of course neither of these points amount to any positive defence of the contextualist conception of knowledge. That calls for independent support. All I have been trying to establish is that putative objections should first carefully reflect on the scope and intent of contextualism.4
4. Thanks to members of the NAMICONA epistemology group, especially Martijn Blaauw.
252
REFERENCES Cohen, S. (2000). ‘Contextualism and Skepticism’, Philosophical Issues 10, 94–108. DeRose, K. (1995). ‘Solving the Skeptical Problem’, The Philosophical Review 104, 17–52. Nozick, R. (1981). Philosophical Explanations, Oxford University Press, Oxford. Schaffer, J. (2004). ‘From Contextualism to Contrastivism’, Philosophical Studies 31, 73–103. Sosa, E. (2000). ‘Skepticism and Contextualism’, Philosophical Issues 10, 1–19. Williamson, T. (2000). Knowledge and its Limits, Oxford University Press, Oxford. — (2001). ‘Comments on Michael Williams’ Contextualism, Externalism and Epistemic Standards’, Philosophical Studies 103, 25–33. Wright, C. (2005). ‘Contextualism and Scepticism: Factivity, Even-handedness, and the Surreptitous Raising of the Standards’, The Philosophical Quarterly 55, No. 219, 236–262.
253
This page intentionally left blank
Graduate Faculty Philosophy Journal The Graduate Faculty Philosophy Journal, published biannualy in association with the Department of Philosophy, New School for Social Research, is a forum for historical and contemporary issues in philosophy. The Journal also publishes an annual special issue dedicated to investigating the historical development of concepts and problems in the history of philosophy and their ongoing relevance to issues in contemporary philosophy.
Vol. 26, No. 1
Philosophy of Dialogue Hannah Arendt/Reiner Schürmann Memorial Symposium Charles Kahn The Philosophical Significance of the Dialogue Form for Plato Vittorio Hösle Self-transcendence of Philosophical Dialogues in Plato and Cicero Caryl Emerson The Dialogic Word in Literary Art: What Sort of Fiction is This? Dmitri Nikulin Dialogue vs. Discourse: On the Possibility of Disagreement in Human Communication Rainer Forst Between Discourse and Dialogue: The Normativiy of Morality Georgia Warnke Interpretive Democracy
Jean-Luc Marion Thomas Tam Drucilla Cornell
Additional Contributions Phenomenon and Event The Death of Art: Bataille, Malraux, Hegel Who Bears the Right to Die
Reflections on Jacques Derrida by Richard J. Bernstein and Agnes Heller
All communications should be addressed to the Editor, Graduate Faculty Philosophy Journal, Department of Philosophy, New School for Social Research, 65 Fifth Avenue, New York, NY 10003. The Journal is biannual. Domestic rates: Individuals: $25.00/year; Students: $15.00/year; Institutions: $50.00/year.
An International Review of Philosophy and the Human Sciences EDITOR: DR ULLRICH HAASE
Volume 36, No.2, May 2005 Husserl and Derrida Articles Presence and Origin: On the Possibility of the Static-Genetic Distinction, by Michael K. Shim Husserl, Derrida and Genetic Phenomenology, by Gary Banham Derrida’s Intentional Skepticism: a Husserlian Response, by Pol Vandevelde Klein and Derrida on the Historicity of Meaning and the Meaning of Historicity in Husserl’s Crisis-Texts, by Burt C. Hopkins Husserl, Freud, A Suivre: Derrida on Time, by Joanna Hodge Derrida’s ‘New Thinking’ A Response to Leonard Lawlor’s Derrida and Husserl, by James R. Mensch ‘Verstellung’ (Misplacement): Completions of Immanence, by Leonard Lawlor The JBSP publishes papers on phenomenology and existential philosophy as well as contributions from other fields of philosophy. Papers from workers in the Humanities and human sciences interested in the philosophy of their subject will be welcome. All papers and books for review to be sent to the Editor: Dr Ullrich Haase, Department of Politics and Philosophy, The Manchester Metropolitan University, Manchester M15 6LL, England. Subscription and advertisement enquiries to be sent to the publishers: Jackson Publishing and Distribution, 3 Gibsons Road, Heaton Moor, Stockport, Cheshire, SK4 4JX, England.
Satz und Sinn.
Bemerkungen zur Sprachphilosophie Wittgensteins. Volker A. Munz: Amsterdam/New York, NY 2005. 302 pp. (Studien zur österreichischen Philosophie 39) ISBN: 90-420-1716-3
€ 60,-/ US $ 84.-
Der vorliegende Band liefert eine umfassende Analyse zum Verhältnis von Sprache und Realität in der Philosophie Ludwig Wittgensteins. Die Untersuchungen konzentrieren sich dabei auf die im Tractatus-Logico-philosophicus entwickelte Idee einer strukturellen Identität zwischen Satz, Gedanke und Sachverhalt, auf die Forderung nach einer phänomenologischen Sprache als Ausdruck unserer unmittelbaren Erfahrungen sowie auf die zentralen Begriffe der grammatischen Regel und ihrer sinnvollen Anwendung. Das Buch versteht sich vor allem als ein Beitrag zur Frage der Beziehung zwischen einem Satz und seinem Sinn, wobei es versucht, besonders die Probleme und Motive zu rekonstruieren, welche im Zusammenhang zu Wittgensteins gewandeltem Sprachverständnis stehen. In diesem Band werden erstmalig auch Auszüge bisher unveröffentlichter Mitschriften von Wittgenstein-Vorlesungen publiziert. Die Aufzeichnungen stammen von Yorick Smythies, einem engen Freund und Schüler Wittgensteins und sind größtenteils aus den Jahren 1937 bis 1939. Darüber hinaus enthält der Text ebenfalls unveröffentlichtes Material aus dem philosophischen Nachlass von Rush Rhees.
USA/Canada: 906 Madison Avenue, UNION, NJ 07083, USA. Call toll-free (USA only)1-800-225-3998, Tel. 908 206 1166, Fax 908-206-0820 All other countries: Tijnmuiden 7, 1046 AK Amsterdam, The Netherlands. Tel. ++ 31 (0)20 611 48 21, Fax ++ 31 (0)20 447 29 79
[email protected] www.rodopi.nl Please note that the exchange rate is subject to fluctuations
Poverty
A Philosophical Approach Paulette Dieterlen Amsterdam/New York, NY 2005. VII, 176 pp. (Rodopi Philosophical Studies 6) ISBN: 90-420-1975-1
€ 40,-/ US $ 56.-
In Poverty: a philosophical approach, the author studies various philosophical issues concerning poverty in the Program for Education, Health and Food (PROGRESA) that was in effect in Mexico, from 1997 to 2002, and shows how theoretical discussion is necessary to clarify some ideas concerning the application of a social policy. Poverty is one of the main problems concerning economics, political philosophy, and ethics. It is an ethical problem because of its relationship with self-esteem. Since poverty is intimately related to social policies, the philosophy of poverty must consider the distribution criteria used to attend to people in situations of extreme poverty. This would involve attention to their needs, preferences, capabilities and “well-being” rights. The book considers social policies applied to poverty, and their occasional abuse of utilitarian instruments. Many are implemented without considering cultural differences, including varying patterns of conduct in diverse communities. Equality also matters. Since poverty and inequality are not the same, the study of the latter allows us to target groups found in the lowest levels of “the playing field”.
USA/Canada: 906 Madison Avenue, UNION, NJ 07083, USA. Call toll-free (USA only)1-800-225-3998, Tel. 908 206 1166, Fax 908-206-0820 All other countries: Tijnmuiden 7, 1046 AK Amsterdam, The Netherlands. Tel. ++ 31 (0)20 611 48 21, Fax ++ 31 (0)20 447 29 79
[email protected] www.rodopi.nl Please note that the exchange rate is subject to fluctuations
Contemporary Pragmatism Edited by John R. Shook and Paulo Ghiraldelli, Jr. Amsterdam/New York, NY 2004. IV, 196 pp. (Contemporary Pragmatism 1:2) ISBN: 90-420-1844-5
€ 40,-/US $ 54.-
Contents Bjørn RAMBERG: Naturalizing Idealizations: Pragmatism and the Interpretivist Strategy Vincent COLAPIETRO: Doing – and Undoing – the Done Thing: Dewey and Bourdieu on Habituation, Agency, and Transformation Melvin L. ROGERS: Rorty’s Straussianism; Or, Irony Against Democracy Symposium on Pragmatism and Radical Orthodoxy Mary DOAK: A Pragmatism Without Plurality? John Milbank’s ‘Pragmatic’ New Christendom Brad Elliott STONE: Making Religious Practices Intelligible: A Prophetic Pragmatic Interpretation of Radical Orthodoxy Jacob Lynn GOODSON: Theology After Epistemology: Milbank between Rorty and Taylor on Truth Review Essay David SEIPLE: Pragmatist Representationalism and the Aesthetics of Moral Intelligence Book Reviews
USA/Canada: 906 Madison Avenue, UNION, NJ 07083, USA Call toll-free (USA only)1-800-225-3998, Tel. 908 206 1166, Fax 908-206-0820 All other countries: Tijnmuiden 7, 1046 AK Amsterdam, The Netherlands. Tel. ++ 31 (0)20 611 48 21, Fax ++ 31 (0)20 447 29 79
[email protected] www.rodopi.nl Please note that the exchange rate is subject to fluctuations
Philosophie, Gesellschaft und Bildung in Zeiten der Globalisierung Herausgegeben von Hermann-Josef Scheidgen, Norbert Hintersteiner und Yoshiro Nakamura Amsterdam/New York, NY 2005. 348 pp. (Studien zur Interkulturellen Philosophie 15) ISBN: 90-420-1785-6
€ 70,-/US $ 98.-
Der Begriff „Globalisierung“ wird zunehmend nicht nur in den Disziplinen der Ökonomie und der Kommunikationswissenschaft diskutiert, sondern auch in den Gesellschafts- und Erziehungswissenschaften sowie insbesondere in der Interkulturellen Philosophie. Der Diskurs über Globalisierung verläuft dabei teilweise analog demjenigen über Inter- und Multikulturalität. Die Beiträge dieses Bandes geben ein breites Spektrum wieder. Sie reichen von der Konstatierung der Globalisierung als einer Rahmenbedingung, zu der man sich als Wissenschaftler reflexiv oder reaktiv zu verhalten habe, über die Analyse diverser Teilaspekte und über visionäre Einforderungen utopischer Globalisierungsauslegungen bis zur Negierung ihrer erkenntnistheoretischen Bedeutung. Das Gemeinsame dieser Vielfältigkeit ist die Bestimmung des Globalen als eines Szenarios der Begegnung.
USA/Canada: 906 Madison Avenue, Union, NJ 07083, USA. Fax: (908) 206-0820 Call toll-free: 1-800-225-3998 (USA only)
All other countries: Tijnmuiden 7, 1046 AK Amsterdam, The Netherlands. Tel. ++ 31 (0)20 611 48 21, Fax ++ 31 (0)20 447 29 79
[email protected] www.rodopi.nl Please note that the exchange rate is subject to fluctuations
Plato’s Cratylus:
Argument, Form, and Structure Michael W. Riley Amsterdam/New York, NY 2005. XV, 151 pp. (Value Inquiry Book Series 168) ISBN: 90 420 1875 5
€ 34,-/US $ 48.-
This book explains how the Cratylus, Plato’s apparently meandering and comical dialogue on the correctness of names, makes serious philosophical progress by its notorious etymological digressions. While still a wild ride through a Heraclitean flood of etymologies which threatens to swamp language altogether, the Cratylus emerges as an astonishingly organized evaluation of the power of words. “Among the varied theories and approaches rivaling one another and enlivening Platonic scholarship today, Riley’s entry must be regarded as a significant event. … In the light of Riley’s study, the Cratylus will gain added stature as evidence for the larger identity of Platonic thought.” Thomas G. Rosenmeyer, University of California, Berkeley Michael Riley is Professor of Classics and Tutor, formerly Director, of the Integral Liberal Arts Program at Saint Mary's College in Moraga, California. He has lectured on Homer, Hesiod, and Plato. His other interests include Herman Melville and Mark Twain.
USA/Canada: 906 Madison Avenue, UNION, NJ 07083, USA. Call toll-free (USA only)1-800-225-3998, Tel. 908 206 1166, Fax 908-206-0820 All other countries: Tijnmuiden 7, 1046 AK Amsterdam, The Netherlands. Tel. ++ 31 (0)20 611 48 21, Fax ++ 31 (0)20 447 29 79
[email protected] www.rodopi.nl Please note that the exchange rate is subject to fluctuations