WHAT IS KNOWLEDGE?

Quassim Cassam

1

What would a good answer to this question - call it (WK) - look like? What I’m going to call the standard analytic approach (SA) says that: A) The way to answer WK is to analyse the concept of knowledge. B) To analyse the concept of knowledge is to come up with non-circular necessary and sufficient conditions for someone to know that something is the case. Is the standard analytic approach to WK the right approach? If not, what would be a better way of doing things? These are the questions I’m going to tackle here. I want to look at some criticisms of SA and consider the prospects for a different, non-standard analytic approach (NA) to WK. Here is one objection to SA: the concept of knowledge can’t be analysed, at least if analysis is understood in the way that (B) understands it.[i] (B) assumes a reductive conception of analysis, according to which analysing a concept is a matter of breaking it down into more basic concepts. Let’s say that a concept C1 is more basic than another concept C2 just if one can grasp C1 without grasping C2 but one can’t grasp C2 without grasping C1. Proponents of SA tend to assume that concepts like truth, belief, and justification are in this sense more basic than the concept knows and that that is why they can be used to specify non-circular necessary and sufficient conditions for knowing. If it turns out that such conditions can’t be given, and therefore that the concept of knowledge can’t be analysed, the net result of combining (A) and (B) will be to make WK unanswerable. If this question is one that we are capable of answering then there must be some other way of answering it. This objection to SA raises the following questions: 1) Is it true that the concept of knowledge can’t be reductively analysed? 2) How should WK be tackled if not by giving a reductive analysis of the concept of knowledge?
The first of these questions will be the focus of part 2. I will focus, in particular, on some of Williamson’s arguments for what I am going to call the unanalysability hypothesis (UH), the hypothesis that the concept of knowledge can’t analysed in more basic terms. I’m going to suggest that these arguments are less than conclusive. (2) is worth asking even if one isn’t convinced that the answer to (1) is ‘yes’. It might be that the concept of knowledge can be reductively analysed but that analysing it in this way isn’t the best way of tackling WK. I will consider this possibility in part 3. The upshot is that there are different reasons for rejecting (A). One might do so because one thinks that it requires us to do something can’t be done or simply because one is convinced that there are better ways of tackling WK. The alternative approach that I want to consider – NA- is still broadly ‘analytic’ in its orientation. It agrees that the key to answering WK is to analyse the concept of knowledge but doesn’t think of conceptual analysis in the way that SA thinks of it. So it rejects (B).[ii] What would it be to analyse a concept if not to come up with non-circular necessary and sufficient conditions for its application? The usual answer to this question is that the aim of an analysis should be to provide us with a reflective understanding of a concept, and that the way to achieve that is to elucidate the concept rather than, in the traditional sense, to give an analysis of it.[iii] So what is it to elucidate a concept? One idea is that it is matter of tracing links between it and other concepts that need not be any more basic. This is how Strawson sees things in this passage: Let us imagine…. the model of an elaborate network, a system, of connected items, concepts, such that the function of each item, each concept, could…. be properly understood only by grasping its connections with the others, its place in the system – perhaps better still, the picture of a set of interlocking systems of such a kind. If this becomes our model, then there will be no reason to be worried if, in the process of tracing connections from one point to another of the network, we find ourselves returning to, or passing through, our starting-point. We might find, for example, that we could not fully elucidate the concept of knowledge without reference to the concept of sense perception; and that we could not explain all the features of the concept of sense perception without reference to the concept of knowledge. But this might be an unworrying and unsurprising fact (1992: 19).
What Strawson is setting out in this passage is a picture of non-reductive conceptual analysis. And one way of understanding NA is to understand it as endorsing (A) with the proviso that analysis is understood along Strawsonian lines rather than along the lines of (B). The problem with all this talk of non-reductive conceptual analysis, of the project of elucidating the concept the concept of knowledge, is that it is vague and metaphorical. It’s all very well talking about the project of tracing connections between concepts but what does this mean in practice? What is the precise nature of the links that the non-reductive story describes and what are the results of elucidating the concept of knowledge? In other words, what is the actual answer to WK that Strawson is proposing? We have the suggestion that the concepts of knowledge and of sense perception are closely related but it’s not clear in what sense this is so and how important it is. We might think, for example, that knowledge and perception are connected because knowledge is what perception gives us but does that cast any light on what knowledge is? And where do other sources of knowledge – testimony, reasoning, etc. – fit into the overall story? These are some of the questions that I will be addressing in part 4, where I will outline a version of NA that builds on two ideas: one is that elucidating the concept of knowledge is, at least in part, a matter of getting a grip on the notion of a way of knowing. The other is that ways of knowing are what we appeal to when we want to explain how someone knows, that is, when we want to answer the question ‘How does X know?’. Only some answers to questions of this form are good answers. NA says that understanding what counts as a good answer is the key to understanding what knowledge fundamentally is. Perception is important in this connection because of the efficacy of perceptual explanations of much of our knowledge of the world around us. But before spelling out these thoughts let’s focus on SA and, in particular, on the response to SA that says that the concept of knowledge can’t be analysed. 2 In Knowledge and its Limits, Williamson defends UH. If he succeeds in making it plausible that ‘the concept knows cannot be analysed into more basic concepts' (33) then SA is in trouble.[iv] Analysing the concept of knowledge into more basic concepts can’t be the best way of tackling WK if the concept of knowledge can’t be analysed into more basic concepts. But how good are Williamson’s arguments in support of UH? There are three arguments that we need to consider. The first is what I am going to call the Distinct Concepts Argument (DCA). This argument assumes that every standard analysis of the concept of knowledge equates it with a conjunctive concept like justified true belief. The aim of DCA is then to show that every standard analysis of knows is ‘incorrect as a claim of concept identity, for the analysing concept is distinct from the concept to be analysed’ (34). Then there is the Inductive Argument. This says that ‘experience confirms inductively…. that no analysis of the concept knows of the standard kind is correct’ (30). Finally, there is the False Expectations Argument. The point here is that one should not expect the concept knows to have a non-trivial analysis in more basic terms. Few concepts have such analyses, and there is no special reason to expect knows to be one of them. Is DCA any good? This argument relies on the notion of a mental concept, so let’s start by briefly considering this notion. Although Williamson doesn’t attempt a formal definition, he does say at one point that the concept true is not mental because ‘it makes no reference to a subject’ (30). So a concept won’t count as mental unless it refers to a subject. This is obviously a long way from constituting a definition of the notion of a mental concept, but Williamson’s idea is presumably that we have an intuitive grasp of what mental concepts are, and that this is enough for the purposes of DCA. Now consider the case of a concept C which is the conjunction of the concepts C1,…, Cn. Williamson’s proposal is that 'C is mental if and only if each Ci is mental' (29). On this account, believes truly is not a mental concept of a state since true isn't a mental concept. By the same token, has a justified true belief is not a mental concept. These concepts are not mental because they have 'irredundant non-mental constituents, in particular the concept true' (30). Having accepted that believes truly and has a justified true belief aren't mental concepts, let’s also accept, at least for the sake of argument, that knows is a mental concept. What follows from this? What follows is that the concept knows can't be the same concept as the concept believes truly or the concept has a justified true belief. The point is that if C is a mental concept and D is not a mental concept, then they can't be the same concept. But, as Williamson sees things, every standard analysis of the concept of knowledge takes it that this concept is the very same concept as some conjunctive concept like has a justified true belief. So every standard analysis of the concept knows is incorrect. Crucially, it doesn't matter for the purposes of this argument which particular conjunctive concept the concept of knowledge is equated with, as long as it has the concept true as a constituent. For example, suppose that instead of equating the concept of knowledge with the concept has a justified true belief one equates it with the concept has a reliably caused true belief. Williamson's argument would still go through since 'it applies to any of the concepts with which the concept knows is equated by conjunctive analyses of the standard kind' (30). As long as the analysing concept is not mental, it can't be the same as the concept being analysed, and this is the crux of DCA. Here, then, is a breakdown of the main components of the Distinct Concepts Argument: (a) Every standard analysis of the concept knows equates it with some conjunctive concept which has the concept true as a non-redundant constituent. (b) The concept true is not a mental concept. (c) Any concept with a non-redundant non-mental constituent is not a mental concept. (d) So the conjunctive concepts with which the concept knows is equated by analyses of the standard kind are not mental concepts. (e) The concept knows is a mental concept. (f) A mental concept can't be the very same concept as a non-mental concept. (g) So the mental concept knows can't be the same concept as any of the conjunctive concepts with which it is equated by standard analyses. (h) So every standard analysis of the concept knows is incorrect. To get a sense of what might be wrong with DCA consider the following parallel line of reasoning: let us say that a marital status concept is one that says something about an individual’s marital status. So, for example, married, single, bachelor, separated and divorced are all marital status concepts. Where C is the conjunction of the concepts C1..., Cn, let us stipulate that C is a marital status concept if and only if each Ci is a marital status concept. On this account, unmarried man isn’t a marital status concept, since man isn’t a marital status concept. Bachelor is a marital status concept. So bachelor and unmarried man can’t be the same concept. Something has clearly gone wrong here, because bachelor and unmarried man are identical if any concepts are. The point is this: the sense in which unmarried man isn’t a marital status concept is that it isn’t what might be called a pure marital status concept. It isn’t a pure marital status concept because one of its constituents, the concept man, isn’t a marital status concept. To put it another way, to describe someone as an unmarried man is to say something about his sex as well as his marital status. But if this is why unmarried man isn’t a marital status concept, then bachelor isn’t a marital status concept either; to describe someone as a bachelor is, after all, also to say something about his sex as well as his marital status. So there is no longer any basis for the claim that bachelor and unmarried man can’t be the same concept. This is where the parallel with DCA breaks down. Williamson thinks that knows and has a justified true belief can’t be the same concept because knows is a purely mental concept whereas concepts like has a justified true belief aren’t ‘purely mental’ (30). On this reading of DCA both (d) and (e) need to be slightly modified. Premise (d) should be read as claiming that the conjunctive concepts with which knows is equated by standard analyses aren’t purely mental because they have at least one non-mental constituent. In contrast, (e) now needs to be read as the claim that the concept knows is purely mental. The argument still goes through but is only as compelling as the case for accepting this version of (e). What is the argument for (e)? Williamson’s primary concern isn’t to defend the thesis that the concept of knowledge is mental or purely mental. His main claim is that knowing is a state of mind. This is a metaphysical rather than a conceptual thesis, and he doesn’t argue for the metaphysical thesis from first principles. He thinks that ‘our initial presumption should be that knowing is a mental state’ (22), and then tries to disarm a range of arguments against this presumption. He also concedes that it doesn’t follow from the fact that knowing is a mental state that the concept knows is mental in his sense. He nevertheless argues that someone who concedes that knowing is a mental state ought to concede that the concept knows is mental, that is, purely mental. Let’s call the presumption that knowing is a mental state Williamson’s Presumption (WP). Strictly speaking, WP is not just the presumption that knowing is a state of mind. It is the presumption that it is ‘merely a state of mind’ (21), that is, that ‘there is a mental state being in which is necessary and sufficient for knowing p’. Presumably, it is only because knowing is ‘merely’ a state of mind that the concept of knowing can plausibly be regarded as ‘purely’ mental. So everything depends on whether we should accept the existence of an initial presumption to the effect that knowing is merely mental. Williamson claims that ‘prior to philosophical theory-building, we learn the concept of the mental by examples’ (22). Our paradigms include not just mental states such as pleasure and pain but also non-factive propositional attitudes such as believing and desiring, that is, attitudes that one can have to falsehoods. In contrast, knowing is factive since one can only know that p if p is true. So how is it that factive propositional attitudes are mental given that they are different from non-factive attitudes and also from mental states which aren’t attitudes at all? Williamson’s answer is that ‘factive attitudes have so many similarities to non-factive attitudes that we should expect them to constitute mental states too’ (22). Indeed, he maintains that there are no pre-theoretical grounds for omitting factive propositional attitudes from the list of paradigmatic mental states. It ‘is built into the natural understanding of the procedure by which the concept of the mental is acquired’ (22) that the mental includes knowing and other factive attitudes. What are the similarities between factive and non-factive attitudes? If attitudes are states of mind, then factive and non-factive attitudes are states of mind. But this is not enough for Williamson’s purposes. He needs to show that knowing is sufficiently similar to believing and other non-factive attitudes to sustain the presumption that knowing is merely a state of mind. This is where the idea that knowing is factive might appear to be in conflict with the idea that it is merely a state of mind. As Williamson’s own discussion illustrates, it takes a good deal of sophisticated argument to weaken the prejudice that a factive attitude can’t be merely a state of mind, and this is difficult to reconcile with the suggestion that we have a pre-theoretical commitment to the idea that knowing is merely mental. Perhaps we don’t have a pre-theoretical commitment either way, the concept of the ‘merely mental’ being a philosophical construct rather than an everyday notion. There is also a question about the suggestion that WP is built into the procedure by which the concept of the mental is acquired. The procedure that Williamson has in mind is that of learning the concept of the mental by examples, but is this procedure sufficiently well-defined to sustain the suggestion that WP is built into it? Prior to theory-building, what we acquire by example are concepts of particular types of mental state rather than the concept of the mental as such. It’s arguable that the procedures by means of which we acquire the concept of the mental leave it open whether knowing is mental in the bland sense that there is a mental state being in which is merely necessary for knowing or in the ‘unexpected’ (21) sense that there is a mental state being in which is necessary and sufficient for knowing. To acquire the concept of the mental as such is to abstract from the differences between different types of mental state, and this already involves taking on theoretical commitments which might properly be described as ‘philosophical’. If this is right, then it is doubtful whether we have any conception of the mental as such, prior to some philosophical theory-building. This is not an argument for the falsity of (e). It is an argument for the view that (e) hasn’t been shown to be true. And that’s not the only thing that is wrong with DCA. Its first premise is also dubious: it is false that standard analyses of the concept of knowledge equate it with some conjunctive concept that has the concept true as a non-redundant constituent. Indeed, it’s hard to think of anyone in the tradition that Williamson is discussing for whom concept-identity has really been an issue. The crucial question for SA isn’t whether the concept knows and, say, the concept has a justified true belief are identical but whether having a justified true belief that A is necessary and sufficient for knowing that A. One can think that a given conjunctive concept provides necessary and sufficient conditions for knowing without thinking that that concept is ‘identical’ with the concept knows, whatever that means. The best way of showing that a given concept can be analysed is to analyse it. Ever since Gettier refuted the justified-true-belief analysis of knowledge in 1963 philosophers have been trying to come up with a better analysis. The problem, according to Williamson, is that each successive analysis has been overturned by new counterexamples. This is the basis of the Inductive Argument for UH. This argument claims that UH is confirmed inductively by the long history of failed attempts to provide correct necessary and sufficient conditions for knowing. There are two things that SA can say in reply to this. The first is that fifty years isn’t a long time in philosophy, certainly not long enough to justify Williamson’s pessimism about the prospects for a reductive analysis of the concept of knowledge. The second is that it needs to be argued and not just assumed that every existing analysis is a failure. Since it is obviously unreasonable to expect anyone to demonstrate the inadequacy of every analysis that has ever been proposed it’s tempting to look for a feature that all currently available analyses have in common and that would justify a blanket rejection of them. This is where DCA comes into its own. It purports to identify just such a feature: the presence of the concept true in the analysandum of that every existing analysis of the concept of knowledge. But DCA doesn’t work so it can’t be used to justify the premise of the Inductive Argument. Indeed, if DCA or any other such relatively a priori argument for UH were successful then the Inductive Argument would be superfluous. That leaves the False Expectations Argument, which says that there is no special reason to expect a reductive analysis of knows. Given that truth and belief are necessary for knowledge, ‘we might expect to reach a necessary and sufficient condition by adding whatever knowing has which believing truly may lack’ (32). This expectation is based on a fallacy, Williamson claims. For example, ‘although being coloured is a necessary but insufficient condition for being red, we cannot state a necessary and sufficient condition for being red by conjoining being coloured with other properties specified without reference to red. Neither the equation ‘Red = coloured + X’ nor the equation ‘Knowledge = true belief + X’ need have a non-circular solution’ (3). One question about this argument is whether the analogy with red is appropriate. Since Locke introduced the distinction between simple and complex ideas and insisted that simple ideas can’t be broken down those who have gone in for reductive conceptual analyses have been careful to argue that only complex concepts are analysable. From this perspective red is the paradigm of a simple concept. Its unanalysability should therefore come as no surprise but it doesn’t follow that the concept knows can’t be analysed. More cautiously, it does not follow that this concept can’t be analysed if it is complex rather than simple. If knows is simple, or if there isn’t a viable simple/ complex distinction, then the False Expectations Argument goes through. Yet Williamson doesn’t establish the simplicity of knows or the unsustainability of the distinction between simple and complex concepts. As things stand, therefore, the False Expectations Argument is as inconclusive as all his other arguments for UH. None of this is to say that the concept of knowledge can be given a reductive analysis. The question is whether it has been shown that it can’t be, and hence that the answer to (1) is ‘yes’. Perhaps there are better arguments against SA than the ones that Williamson gives but the discussion so far suggests that SA is still in the running. In that case, perhaps it would be better for critics of SA to change the focus of their attack. Instead of pressing the point that the concept knows can’t be analysed, and that SA is a non-starter for this reason, a different line of attack would be to concentrate on whether giving a reductive analysis of the concept of knowledge is the best way of tackling WK even if, as I have been arguing, the possibility of such an analysis hasn’t been ruled out. This is the point of (2), and it is to this question that I now turn. 3 WK is an example of what might be called a ‘what’ question, a question of the form ‘what is X?’. There are many such questions that are of interest both to philosophers and to non-philosophers. Is there anything useful that can be said, in general terms, about the best way of dealing with such questions? Perhaps not, given their sheer variety. But maybe it would help to fix ideas to compare WK with another ‘what’ question that looks as though it is at least in the same ball park as WK, namely, ‘what is depression?’ (WD). Though not everyone will agree that WD is in the same ball park as WK the comparison doesn’t look completely absurd, especially if one is sympathetic to the idea that knowing is a state of mind. In any case, just to get a sense of the different ways of dealing with a ‘what’ question let’s consider how one might go about tackling WD. What would a good response to WD look like? People who ask this question are generally interested in such things as the incidence, symptoms, and causes of depression. These are the issues that a helpful answer to WD might therefore be expected to address. One might also expect a good answer to WD to mention the different types of depression and the range of possible treatments. What one would not expect is an analysis of the concept of depression in more basic terms, a statement of non-circular necessary and sufficient conditions for being depressed. Even if the concept of depression can be analysed it just doesn’t look as though an analysis is especially relevant if the aim is to say something helpful in response to WD. This is not to deny that for practical purposes clinicians need something like a definition of depression or at least criteria for diagnosing it.[v] Yet the standard definition – the one given in the American Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders – doesn’t provide non-circular necessary and sufficient conditions for depression. Instead, it provides two lists of symptoms and stipulates that a patient with major depression must experience at least five of the nine symptoms just about every day for at least two weeks. The emphasis is on necessary conditions rather than necessary and sufficient conditions, and the “definition” is circular: one of the key criteria for major depression is that the patient is in a persistent depressed mood. How does this help with WK? At the very least it helps by showing that there are questions of the form ‘what is X?’ that don’t call for an analysis of the concept of an X. Saying that a question of this form doesn’t call for an analysis of a concept of an X is different from saying that this concept can’t be analysed. I haven’t said that the concept of depression can’t be analysed, only that it needn’t be analysed for the purposes of answering WD and that the standard clinical definition of depression doesn’t in fact amount to a reductive analysis. This suggests that there is at least nothing wrong in principle with the idea that WK doesn’t call for a reductive analysis of the concept of knowledge even if such an analysis hasn’t been shown to be impossible. It is true that defenders of SA are unlikely to be impressed by any of this. There are at least three things they can say in defence of their approach: i. The analogy between WK and WD is no good because WD is not usually understood as a philosophical question. When WK is read in the way that philosophers tend to read it the challenge is not just to say what knowledge is but to say what it is in a special way. This is what Michael Williams is getting at in his comment that ‘when we ask “What is knowledge?” philosophically, we mean “Don’t just tell us ancillary facts about knowledge: tell us what it is essentially”’ (2001: 13). It is when WK is asked in this spirit that it calls for an analysis of the concept of knowledge. In contrast, much of what is usually said in response to WD consists in the specification of ancillary facts. ii. If the concept of knowledge can’t be analysed then it’s fair enough that one should be looking for a different approach to WK. But what if we can find non-circular necessary and sufficient conditions for knowing? Wouldn’t this be the best possible response to WK? How could a different response possibly be any better? So critics of SA had better concentrate on showing that the concept of knowledge can’t be analysed. There is no future in the idea that there are better ways of dealing with WK if the concept of knowledge can be analysed. iii. What, exactly, is the alternative to reductive conceptual analysis in relation to WK? At least when it comes to WD we have some idea of what the alternative looks like. Crudely, we can say what depression is by specifying its functional role, its inputs and outputs, causes and symptoms. Can WK be given a functional response? If so, what would a functional account of knowledge look like? These are some of the challenges to which I now want to respond on behalf of NA. Taking them in reverse order, the aim will be to develop a non-reductive response to WK, to show how it can be at least as illuminating as any reductive response, and to rebut the charge that the non-reductive alternative to SA only succeeds in telling us ancillary facts about knowledge. The non-reductive approach to WK that I want to flesh out has what might be thought of as a broadly functional orientation but parallels with the functional response to WD shouldn’t be exaggerated. The key isn’t the function of knowledge but the explanation of knowledge. The proposal is that we can elucidate the concept of knowledge in something like Strawson’s sense, and thereby work towards an answer to WK, by looking at what it takes to explain how someone knows, that is, to answer the question ‘How does S know?’. The significance of this question for WK might not be immediately apparent but it will hopefully become clearer below. 4 In his paper ‘Other Minds’, Austin remarks that when we make an assertion such as ‘There is a goldfinch in the garden’ or ‘He is angry’ we imply that we know it. Hence: On making such an assertion…. we are directly exposed to questions such as (1) ‘Do you know there is?’ ‘Do you know he is?’ and (2) ‘How do you know?’ If in answer to the first question we reply ‘Yes’, we may be asked the second question, and even the first question alone is commonly taken as an invitation to state not merely whether but also how we know (1979: 77).
It is no accident that such questions are normally appropriate. It is something like a conceptual truth that someone who says or implies that he knows that P is exposed to the question ‘How do you know?’, and this is something that any serious attempt to elucidate the concept of knowledge had better take into account.[vi] Perhaps ‘I am in pain’ is not exposed to this question but such assertions raise special questions that I don’t want to go into here.[vii] Here are three issues that now need to be addressed: (α) Can one still count as knowing that P if one doesn’t know how one knows that P? (β) What would count as a satisfactory answer to the question ‘How do you know?’ (γ) What light does any of this cast on WK?
On (α), it’s true that people sometimes say ‘I don’t know how I know; I just know’. We allow that the knower might not know the answer to ‘How do you know that P?’ but it is much harder to accept that there doesn’t have to be an answer, known or unknown. Even our willingness to tolerate cases in which the knower doesn’t know how he knows has limits. In the primary sense of ‘knows’ the knower must know how he knows even if he might have trouble articulating his second-order knowledge.[viii] Why does it seem so compelling that when someone knows that P there must be an answer to the question ‘How does he know that P?’. The thought is that if someone knows then there must be something in virtue of which he knows. That can then form the basis of a satisfactory response to ‘How does he know?’. For example, if the concept of knowledge can be analysed – say as justified true belief - then it might be tempting to say that someone who knows that P does so in virtue of having a justified true belief that P. But, apart from worries about whether the concept of knowledge can be analysed, it’s also worth pointing out that ‘by having a justified true belief that P’ would not normally be taken to be a good answer to ‘How does he know that P?’. The reason is that it doesn’t explain how he knows even if (setting aside Gettier complications) it entails that he knows. This brings us to (β). Suppose, to take another one of Austin’s examples, I assert that there is a bittern at the bottom of the garden and am asked how I know. One answer would be ‘I can see it’. Another answer would be ‘I can hear it’. Seeing that there is a bittern at the bottom of the garden or, if one prefers, simply seeing a bittern at the bottom of the garden is a way of knowing that there is a bittern there. In general, Φ-ing that P is a way of knowing that P just if it is possible satisfactorily to explain how S knows that P by pointing out that S Φs that P.[ix] On this explanatory account of ways of knowing seeing that P is clearly a way of knowing that P. Saying that I can see a bittern at the bottom of the garden can explain how I know that there is one there. Someone might question whether I see what I think I see but once it’s agreed that I see a bittern then nothing further needs to be done to explain my knowledge of its presence. Ways of knowing needn’t be perceptual. I can know that P by reading that P or by being told that P. Further, as Austin points out, questions of the form ‘How does S know?’ don’t always elicit answers of the form ‘S Φs’. ‘From its booming noise’ and ‘I was brought up in the fens’ might be given as answers to ‘How do you know there is a bittern in the garden?’. Strictly speaking, however, the question to which the former is a response is ‘How can you tell it’s a bittern?’, while the question to which ‘I was brought up in the fens’ is a response is ‘How do you know about bitterns?’. Neither is a satisfactory answer to ‘How do you know here and now that there is a bittern at the bottom of the garden?’. Being brought up is the fens is not a way of knowing that there is a bittern at the bottom of the garden; it can’t be said that I was brought up in the fens and thereby know that there is a bittern there. Let’s agree, then, that a satisfactory response to ‘How do you know?’ will need to identify one’s way of knowing and that ways of knowing are usually expressed by sentences of the form ‘S Φs’ (where ‘Φ’ stands for a verb). There are some further points about ways of knowing that are worth making: I. Ways of knowing needn’t be propositional attitudes. I can know that there is a bittern at the bottom of the garden by seeing it but this kind of seeing isn’t propositional. II. For Φ-ing that P to explain one’s knowledge that P it is neither necessary nor sufficient that ‘S Φs that P’ entails ‘S know that P’. Not sufficient: ‘S regrets that P’ entails ‘S knows that P’ but saying that S regrets that P doesn’t explain how S knows; regretting that P isn’t a way of knowing that P.[x] Not necessary: ‘S read that P’ can be a good answer to ‘How does S know that P?’ despite not entailing that S knows that P. III. Most ways of knowing are ways of coming to know.[xi] Seeing that P, hearing that P, reading that P are all ways of coming to know that P. A possible exception is remembering that P. How do I know that I went on safari last year? I remember. Remembering going on safari is a way of knowing that I went on safari but we might be reluctant to describe it as a way of coming to know that I went on safari. There is an obvious question that is raised by the discussion so far: given the sheer variety of ways of knowing, of acceptable responses to ‘How do you know?’, what is their unifying principle? What do they all have in common that makes them ways of knowing? The idea that what ways of knowing that P have in common is that they all entail that one knows that P has already been ruled out so where do we go from here? One possibility is that there is nothing further to be said. In practice we have no trouble distinguishing between acceptable and unacceptable answers to ‘How do you know?’ but no further explanation can be given as to why we accept the explanations that we accept and reject the ones that we reject. There are good and bad explanations of a person’s knowledge but our explanations cannot themselves be explained; they have no deeper rationale or unifying principle. I will call someone who argues in this way a minimalist. Minimalism is hard to swallow. As we have seen, acceptable answers to ‘How do you know that P?’ include ‘I perceive that P’ and ‘I read that P’. Unacceptable answers include ‘I guessed that P’ and ‘I imagine that P’. Is it really plausible that there is nothing further to be said about why perceiving that P is a way of knowing that P whereas guessing that P is not? It’s surely not irrelevant, for example, that we regard perception as reliable, as delivering a high ratio of true beliefs, whereas there is no temptation to suppose that the same is true of imagination.[xii] I will come back to this. In the meantime, there is another proposal to consider. This is the proposal that ‘by Φ-ing’ is an acceptable answer to ‘How does X know that P?’ only if Φ-ing is a way of coming to know that P. It might seem that this has already been ruled out by the safari example but that’s not quite right. The discussion of that example assumed that ‘by remembering’ can be a good answer to ‘How do you know that P?’ and that remembering that P is not a way of coming to know that P. Each of these assumptions might be questioned. Perhaps it is never correct to say ‘I remember’ in response to ‘How do you know?’. Alternatively, one might argue that it is sometimes correct to say this but only because remembering that P can be a way of coming to know that P. In what sense can remembering that P be a way of coming to know that P? Suppose that P is the proposition ‘I am now on safari’. What is true is that I do not come to know that I am now on safari by remembering that I am. But what if P is ‘I was on safari’? In that case, I can come to know that P by remembering that I was. Memory, like testimony, is the source of one’s knowledge of many propositions of the form ‘I was F’, just as perception is the source of one’s knowledge of many propositions of the form ‘I am F’. Perception, memory and testimony are all capable of yielding knowledge of the appropriate propositions, and that is why seeing that P, remembering that P and reading that P all count as ways of knowing that P. This brings us, finally, to (γ): what light does the explanatory account of ways of knowing cast on WK? Here is one suggestion: once we have an idea of the sorts of things that can yield the knowledge that P we can then proceed to give an account on this basis of what it is to know that P. In effect, this will be an account of knowing in terms of ways of knowing. Snowdon considers something like this possibility in a discussion of what he sees as the necessary link between knowledge and perception. He says that evidence of such a link ‘comes from our treating it as totally unproblematic that someone’s knowledge that P can be explained by saying that they saw that P’ (1998: 301). This then leads to the suggestion that ‘our fundamental understanding of knowledge is as what is yielded by perception in certain circumstances’ (ibid.).[xiii] This is a partly functional response to WK.[xiv] To say that knowledge is what perception gives us is to give an account of knowledge in terms of its inputs. The account is also non-reductive since it doesn’t say that the concept of perception is more basic than that of knowledge. Just as our fundamental understanding of knowledge is terms of its inputs so our fundamental understanding of perception is in terms of its outputs, the key output being knowledge: ‘perceiving an object is, in its nature, a way to get knowledge about the object’ (Snowdon 1998: 300). One problem with this response to WK is that it neglects non-perceptual sources of knowledge. However, such sources can easily be accommodated by saying that knowledge is to be understood as that which is yielded by perception, memory, testimony, introspection, calculation, and so on. To put it another way, to know that P is to be in a state that one can get into in any number of different ways, for example, by seeing that P, hearing that P, reading that P, calculating that P, and so on. This is an explanatory conception of knowledge, to go with the explanatory conception of ways of knowing: our fundamental understanding of (propositional) knowledge is that it is something whose possession by an individual can properly be explained by reference to any one of an open-ended list of ways of knowing or, if one prefers, ways of coming to know.[xv] Non-circular necessary and sufficient conditions for knowing are not to the point. We now have a version of NA, that is, a non-standard analytic response to WK. It’s an analytic response because it focuses on the concept of knowledge. It’s a non-standard analytic response because it doesn’t try to give a reductive analysis of the concept of knows; instead, it seeks to elucidate this concept by relating it to other concepts that are no more basic. Specifically, it elucidates the concept of knowledge by relating it to the concept of a way of knowing and to concepts of specific ways of knowing. It’s hard to see why this non-reductive approach to WK should be seen as less helpful or illuminating than the standard reductive approach or as only telling us ancillary facts about knowledge. It is not an ancillary fact about the knowledge that P that it can be acquired by seeing that P. Clearly, there are lots of things about the world around us that can’t be known in this way but we might still think that perception is a basic source of our knowledge of many empirical propositions. In any case, the present version of NA takes care not to ignore non-perceptual ways of knowing. The biggest challenge facing this approach to WK is to the suggestion, or implication, that it can do without a reductive analysis of the concept of knowledge and that it therefore has no need for SA. To put it at its most abstract, the worry is that the concept of knowledge is prior to that of a way of knowing and that any attempt to elucidate the former by reference to the latter is doomed. For example, suppose we say that seeing that P is a way of knowing that P whereas wishfully thinking that P is not a way of knowing that P. We might try to explain this difference by saying that only seeing delivers a sufficiently high ratio of true beliefs but why assume that reliability is relevant to ways of knowing? Surely this assumption can only be justified by a prior analysis of the concept of knowledge along reliabilist lines. If, as simple reliabilism says, knowledge is true belief caused by a reliable process then it is not hard to figure out why wishfully thinking that P isn’t a way of knowing that P. But if simple reliabilism is correct then it already provides a standard analytic response to WK without reference to ways of knowing. So it seems that NA collapses into SA. There are several things that are wrong with this line of argument. To start with, it is false that a reliability condition on knowledge can only be justified on the basis of a reductive analysis of the concept knows. From the fact that reliability is necessary for knowledge it doesn’t follow that a reductive analysis of the concept of knowledge is possible.[xvi] It is also debatable whether we can explain why seeing counts a way of knowing solely in terms of reliability. Imagine that I form beliefs about what is going on in distant parts on the world on the basis of what my crystal ball tells me and that my crystal ball is as reliable as ordinary seeing. On a particular occasion I assert that the American President is in Iowa. My answer to ‘How do you know?’ is ‘I can see in my crystal ball that he is in Iowa’. Is this an acceptable answer? If not, then seeing in my crystal ball that P is not a way of knowing that P. Yet seeing (in the ordinary sense) that P is a way of knowing that P. Since (ex hypothesi) there is no difference in reliability between ordinary seeing and crystal ball gazing it can’t be maintained that it is sufficient for ordinary seeing to be a way of knowing that it delivers a high ratio of true beliefs; one can imagine crystal ball gazing or clairvoyance doing that. In that case, what does explain the fact that seeing is a way of knowing? To answer this question we can borrow some insights from virtue epistemology. Virtue epistemologists like Goldman try to give an account of the nature of justified belief. Their idea is that a justified belief is one that is obtained through the exercise of intellectual virtues but they do not propose a definition of intellectual virtue. Instead, Goldman ‘posits a set of examples of virtues and vices as opposed to a mere abstract characterization’ (1992: 158). Exemplary intellectual virtues include ‘belief formation based on sight, hearing, memory, reasoning in certain “approved” ways, and so forth’ (ibid.). Why do these count as intellectual virtues? Reliability is one factor but it is also important that they are ways of obtaining knowledge; they are belief-forming processes that would be ‘accepted as answers to the question ‘”How does X know”’ (1992: 162). Novel or unusual belief-forming processes are then evaluated as virtuous as long as they are sufficiently similar to the exemplary virtues. NA should think of ways of knowing somewhat in the way that Goldman thinks of intellectual virtues. They are ways of obtaining knowledge and they wouldn’t count as ways of obtaining knowledge if they didn’t deliver a high ratio of true beliefs. But in figuring out what counts as a way of knowing we don’t start with a blank slate and then work up to a list of ways of knowing on the basis of considerations like reliability. The position is rather that we start with a list of exemplary ways of knowing, exemplary responses to “How does X know?” such as perceiving, and work up from there to the identification of further ways of knowing and ultimately to a more abstract characterization of the notion of a way of knowing.[xvii] On this account, the status of exemplary ways of knowing such as perceiving is not something that can be explained in more basic terms. If our fundamental understanding of knowledge is as what perception gives us then there is no question of perceiving that P failing to be a way of knowing that P.[xviii] And if this sounds like a sophisticated form of minimalism then so be it. When it comes to explaining why certain explanations of our knowledge are good ones there is only so far we can go.[xix]

REFERENCES
Austin, J. L. (1979), ‘Other Minds’, in Philosophical Papers , 3rd Edition (Oxford: Oxford University Press).
Ayers, M. R. (1991), Locke (London: Routledge).
Cassam, Q. (2007), ‘Ways of Knowing’, Proceedings of the Aristotelian Society, CVII.
Goldman, A. (1992), ‘Epistemic Folkways and Scientific Epistemology’, in Liaisons: Philosophy Meets the Cognitive and Social Sciences (Cambridge, Mass.: The MIT Press).
Hampshire, S. (1979), ‘Some Difficulties in Knowing’, in T. Honderich and M. Burnyeat (eds.) Philosophy As it Is (Harmondsworth: Penguin).
Kornblith, H. (2002), Knowledge and its Place in Nature (Oxford: Oxford University Press: 2002).
Snowdon, P. F. (1998), ‘Strawson on the Concept of Perception’, in L. Hahn (ed.) The Philosophy of P. F. Strawson (Chicago and Lasalle: Open Court).
Strawson, P. F. (1992), Analysis and Metaphysics: An Introduction to Philosophy (Oxford: Oxford University Press).
Stroud, B. (2000), ‘Scepticism and the Possibility of Knowledge’, in Understanding Human Knowledge (Oxford: Oxford University Press).
Unger, P. (1975), Ignorance: A Case for Scepticism (Oxford; Clarendon Press).
Williams, M. (2001), Problems of Knowledge: A Critical Introduction to Epistemology (Oxford: Oxford University Press).
Williamson, T. (2000), Knowledge and its Limits (Oxford: Oxford University Press).

-----------------------
[i] See Williamson 2000: 27-33 for a defence of the view that the concept of knowledge can’t be analysed into more basic concepts.
[ii] There are also non-analytic, naturalistic alternatives to SA that argue that we should focus directly on ‘knowledge itself’ (Kornblith 2002:1) rather than on the concept of knowledge but I won’t be looking at such views here.
[iii] As Williamson remarks, it doesn’t follow from the fact that the concept knows cannot be analysed into more basic concepts that ‘no reflective understanding of it is possible’ (2000: 33).
[iv] All references in this form are to page numbers in Williamson 2000.
[v] 'Qbmv…ˆš¢£¦¨­²³¼É÷ˆ ‰ Š 4S‚¡«¬¸ÆÇ(
;
<
I
U c Î è é
'
øðèðàèØÍÂغ²ªðªØ¢Øš’ðèºèºèð躊è¢|¢èŠètktkth |¶H*[vi]mH sH h |¶mH sH jhŠz0J