I used to think, at the time of my militant days in analytic philosophy, that one’s professional duties in philosophy implied being able to discuss the views of people that one does not particularly like or sometimes whom one does not like at all. I was wrong: most of the time one writes about subjects that one likes and for one’s friends (even though in print one may not call them by their first names). So I am particularly lucky and grateful to receive this gift from my old friends, on some of the topics which have occupied me most. I am most grateful to Julien Dutant, Davide Fassio and Anne Meylan to have made this possible. I have grouped my replies by theme.

1 On norms of belief and assertion

1.1 To Erik Olsson

I have always been impressed by Olsson’s pragmatic stance in philosophy and by his defence of it in the tradition of Carnap and of Isaac Levi. I always took Rorty to be, in Susan Haack’s phrase, a “vulgar” pragmatist, one who, unlike Peirce, disvalues truth and knowledge, and takes a deflationist view of metaphysical issues. I discussed his views because it seemed to me important to get right on what I was opposing, that is, a complete nihilism about not only the value of truth and knowledge, but also the very idea that such notions made sense.

Rorty accepts that there is a difference between a statement being justified and its being true, but interprets it as the contrast between those beliefs that appear justified in front of our present day audiences and those which will appear justified in front of future audiences. So truth is always justification before an audience, actual or hypothetical, and Rorty in the end defends the “indistinguishability thesis”. Moreover he denies that truth could have any normative import and that it is the goal of inquiry. One should to distinguish here between (i) the question whether beliefs “aim at truth”, and (ii) the question whether our epistemic goal is to have true beliefs. The first concerns the nature of belief, the second concerns the nature of inquiry, and can be formulated as the goal of maximising the number of true beliefs. I interpret (i) in the normative sense, as meaning that there is a norm of belief, which is truth (actually, knowledge, of which truth is a part). According to me (i) takes precedence over (ii). In other words it is because truth is the norm of belief that it can be the goal of inquiry. But the norms of belief and knowledge are one thing, the norms of inquiry another thing. Inquiry has to do with the formation and regulation of knowledge (what Levi calls “the enterprise of knowledge”). Knowledge itself has a specific nature independently of inquiry, which concerns the way it is acquired, maintained and managed. I accept also that truth (knowledge) is the goal of inquiry. Rorty denies both. Olsson glosses Peirce’s view to be much similar to Rorty’s, except that he takes the goal of inquiry to be belief, or settled opinion, but not truth. I shall not dispute his interpretation of Peirce, although one should keep in mind that one never finds in Peirce the theme which is prominent in Rorty (and in Dewey) that truth is essentially social, in the sense of being determined by consensus alone. Olsson’s point, which he takes from the theory of goal setting in Management, agrees on this point with Rorty: the goal of justification is more “directionally complete” than the goal of truth. In other words it allows a better guidance, since justification is more easily achieved than truth. I agree, but I refuse the consequence which Rorty draws—truth plays no normative and regulative role—and unless one makes the confusion between (i) and (ii), it does not show that truth is not the norm of belief in the sense in which the correctness condition for belief is truth. Sometimes indeed the two norms conflict: there are things that one ought to believe on the basis of the best information available, although the same things ought not to be believed because they are actually false. But that does not show that truth is not the main and leading norm, once we understand it in fully normative terms, not in teleological ones, as a goal. Olsson clearly prefers the second.

1.2 To Conor McHugh

I take the norm of belief to be this: a belief is correct only if it is true (setting aside the delicacies of whether one ought to formulate it as a conditional or as a biconditional). This is belief’s correctness condition. The correctness condition is normative, since correctness is normative. But the only if part is not normative. It just states the state of affairs which makes a belief correct, and this condition is the truth of the content. We can call it the satisfaction condition. McHugh makes too much of my use of Thomson’s distinction between external (“e-correctness”, the property which has to be instantiated in order to meet a certain standard) and i-correctness (the property that an agent has to exemplify to reach the standard). He objects that i-correctness is not correctness at all, and that I am wrong to take e-correctness as non-normative and descriptive. This rests on a misunderstanding of my use of this distinction. I agree with him that there cannot be two kinds of correctness. The only correctness condition is e-correctness. I introduced Thomson’s distinction only to reach my own position, and did not endorse it. Neither do I want to say that correctness is not normative, as those writers who reduce the normative load of belief to its mere “direction of fit”. So I do not want to divide correctness. But McHugh is wrong not to recognise that there is a second dimension in the evaluation of belief, which has to do with the way or the mode of operation by which the correctness condition is fulfilled. Ernest Sosa cashes out the distinction in terms of his metaphor of an archer who hits a target, and calls the satisfaction condition the “success condition”. Some writers say that there are two kinds of norms, the “subjective” and the “objective” ones. Others say that there is a difference between the “normative requirements” for an action or a belief, which are fully objective, and the reasons one has to reach these requirements, which are subjective and perspectival. These distinctions are similar, but not equivalent. I agree with John Broome that the objective, or the external, sense of “norm” is the leading one, and that rationality should not be “divided”. It does not make sense to say that belief is subject to two distinct kinds of norms. It does not follow that we cannot assess beliefs along another dimension, distinct from its correctness condition, although related to it. In terms of Sosa’s metaphor, the dimension in question is the aptness of the belief, or the competence which the archer manifests “in the right way” (an archer can manifest this competence but fail to reach the target, or he can hit the target through a bad exercise of his competence, or by sheer luck). One way or another we have to accommodate this distinction, which is common to all attitudes (although Sosa’s aptness condition is only one version of it). Thus one can conjecture or guess more or less accurately, although there is only one way for conjectures or guesses to be correct. The way the correctness condition is reached differs depending on the attitude. I agree with McHugh that it hard to say that one believes better or worse, and I agree with him that believing is not a kind of action or performance (“Well believed!” would be an odd compliment but “well imagined” makes sense). But there is room to say that our justifications for believing are not the same thing as their being correct. The usual way to make the distinction is to say that belief is subject, in addition to a norm of truth, to a secondary norm of evidence. “Secondary” here cannot mean “auxiliary”. Evidence is evidence for truth. Evidence aims at truth. But, as we saw above, evidence and truth, justification and truth can diverge. At this point, we can go in two directions. Either we can say that truth is the basic objective norm for belief, and that evidence is a secondary norm, but we shall encounter the problem of their potential conflict. Or, alternatively, we can integrate the two norms into a single one, which is knowledge. Knowledge entails both truth and evidence. This is my preferred solution, for it unifies the truth norm and the evidential norm, although a lot more should be said to defend the proposal.

Reasons for belief are not their correctness conditions. Indeed we can, psychologically speaking, have different kinds of reasons to believe, including practical ones. But we should not confuse the psychological fact that the causes of our believing are diverse with the normative fact that there is only one “right” kind of reason for belief, which is epistemic: evidence, in so far as it leads to truth. Epistemic reasons form an exclusive normative club with respect to belief: practical reasons are not admitted. Expressivists deny this, and would like to make all reasons psychological and to construe objective reasons as being at best idealisations of our subjective reasons. This is not the sense in which I propose to construe the norm of belief as an ideal of reason. By this I mean that it is an essential feature of our concept of belief. I also want to use the notion of an idealisation to suggest a solution to the guidance problem: how can a norm motivate and guide belief formation? Here I want to say that there are degrees of the application of the norm. This does not mean that the norm itself is a matter of degree. It is not, for it is objective, and absolute. McHugh also wants to defend the fitting attitude conception of correctness. I fully agree, and have committed myself to it (Engel 2013). So in the end there is much less disagreement between us that he seems to assume.

1.3 To Adam Carter

That the norm of belief cannot be divided does not mean that there cannot be other dimensions of evaluation or assessment of belief. If these dimensions are to count as being of the right kind, they have to be epistemic. This seems to contrast with the parallel issue about assertion, which Carter examines here. For assertions are actions, and one can assert things for a variety of reasons. One can assert that P to state that P, but also to suggest that P, with irony, or with the purpose of lying. Or one can, in the cases that Carter describes as “epistemic hypocrisy”, fail to comply with a norm of “epistemic integrity”. These are clearly cases where the asserter is in some sense insincere or dishonest. But are these exceptions to the knowledge norm of assertion (KAA)? And does this show that the “uniqueness hypothesis” is false? I do not think so. The asserter in such situations is, as Carter insists, justified, but nevertheless epistemically criticisable. But that does not seem to me to be a violation of the KAA. In the article discussed by Carter, I warn against the temptation of construing the KAA as implying that the asserter actually knows the content of her assertion. She has to be in position to know in order to satisfy the norm. I would also be prepared to follow here Reynolds (2013) who argues that it is enough that one must assert that P if one has the appearance of knowing that P (although I cannot argue for this here). Lackey’s selfless assertions are cases where our role as asserter conflicts with other roles, but there are not situations where the norm of assertion—to assert what one knows—is violated or suspended. In ADOPTION-2, Nina’s assertion is epistemically criticisable, but her interlocutor, Irene, by discovering this, is entitled to say: “Next time you should know better!” If we look at thing diachronically, we see that the KAA is still in place.

1.4 To Igor Douven

Just as I want to resist Carter’s suggestion that there is not a unique norm of assertion but possibly a number of norms which depend upon contexts, I want to resist Douven’s parallel proposal about conditionals. He defends “inferentialism”, according to which a conditional is true in a given context if and only if its consequent follows from its antecedent plus contextually relevant background knowledge, where “follows” encompasses a variety of inferences holding between antecedent and consequent, mostly inductive and statistical, but not necessarily deductive. Douven’s framework is thus both probabilistic (although he remains propositionalist: he does not subscribe to the view that conditionals do not have truth conditions) and in part pragmatic. I have three worries. First the view may cast too large a net over the territory of conditionals in natural language. How are speakers to recognise the various kinds of inference at play when presented a conditional? Second, it is not clear that conditionals, on this view, obey compositionality (this is a difficulty for inferential role semantics and it may well be one for inferentialism in Douven’s sense). It is not completely sure that inferentialism can preserve propositionalism. Third, what happens to classical inference rules such as modus ponens? Well, as Douven says, there are exceptions. Here my classicist instincts about logic make me suspicious. I should, however, welcome inferentialism about conditionals, since, as Douven formulates it, it is inspired by a suggestion that I myself made about identity (and which others have made about truth): just as one can conceive of identity as a second-order functional property obeying the classical rules but multiply realised in various kinds of objects, one can conceive of conditionals as answering a certain functional role, which is constituted by its being a “connector”, realised in different ways depending on the kinds of conditionals at issue and the type of inferential relation involved in each case. A brilliant idea, although I am not sure how to make it work. It raises the same kind of problems as those that I saw for identity: if the first-order realiser properties of the connector can be so different, depending on kinds of inference and background contextual conditions, how can we be sure that we are still in presence of the conditional relation between antecedent and consequent? If classical rules fail in some realisers how can we say that the same functional role is instantiated?

2 On the value of knowledge

2.1 To Duncan Pritchard

I have a great debt towards Duncan Pritchard’s numerous contributions to epistemology. I am glad that he agrees with me on pragmatic encroachment. He is surprised that in the article he discusses I held that pragmatic encroachment is not relevant to the issue of the value of knowledge or to the value of its subparts. I said this because I take it that pragmatic encroachment is a problem which concerns the nature of knowledge, of evidence and of justification, and which is independent of the question of how we value knowledge. This may be unduly strong, in particular for the parallel claim about the relevance of knowledge to action. But my claim was motivated by two considerations. One can expect that a certain kind of conception of knowledge be in harmony with a certain kind of conception of its value. Thus a virtue epistemologist has clearly the beginnings of the solution of the second problem when he defends his view that knowledge consists in some kind achievement or performance. But a certain view of knowledge and action does not entail a view about its value. Thus an evidentialist can want to define knowledge in terms of the possession of evidence, but hold an instrumentalist conception of its value. Or a “knowledge-first” theorist can hold that knowledge is primitive, and necessary for action, without subscribing to an instrumentalist conception of its value: he can take it to have intrinsic value. The second consideration is that it seemed to me that many discussions of the value of knowledge tend to presuppose an instrumentalist framework, or at least a teleological conception of value. I wanted to emphasise a normative conception, in more deontic terms, which seems to me more in line with the evidentialism that I was defending in this article. I am happy to accept Pritchard’s suggestion that just as a big fly is not big simpliciter, that something has epistemic value does not mean that it is valuable simpliciter. The instrumentalistic conception of value, according to which the value of knowledge consists in its being at the service of a further end, say happiness, leads precisely to the confusion that Pritchard denounces, and I fully concur with him about the independence of the issues about epistemic value (and epistemic normativity) from questions of its instrumental value, and from eudaimonism in general.

2.2 To Christopher Kelp and Mona Simion

Kelp and Simion make the interesting suggestion that one could treat knowledge as a commodity, a certain kind of quantity which is out there and available like water, and that one can account for its value in these terms. Their view seems to me to be particularly relevant for social epistemology and for views which insist on the “extension” of our minds outside the limits of our skin and on the fact that much, perhaps the majority, of our knowledge is “outsourced”. The proposal is clearly externalist, and as they emphasise, it has strong links with the view of the value of knowledge defended by Goldman and Olsson. But although I am myself an externalist about knowledge, I am concerned by the question of how we can accommodate the internalist elements. The ideal position would be a form of “epistemic compatibilism” capable of integrating the two dimensions, but I am not sure which one is the best (Sosa’s view is a strong candidate). But if one cares about the internalist dimension, Kelp and Simion’s suggestion raise two worries. The first is that it is not clear that it can accommodate the individualistic elements. Knowledge is not merely a quantity like water. The internet age can lead us to think that way, but one has to make room for our individual achievements in getting and maintaining knowledge, and I am not sure that the water analogy does the trick; and when it comes to resisting the so-called “extended knowledge” provided by internet, I am fully internalist: what justifies us to believe so much crap? The second is that there might not be so much knowledge around, after all. There are huge things that we don’t know within science (why there is black matter?), plenty of things are hidden from us (galaxies and numbers), and I am often tempted to think of the internet age as a new age of ignorance rather than as the “society of knowledge”. Resistance to the apparent availability of knowledge implies reflection, and reflective second-order knowledge is not so easily available. But this is not really a disagreement with Kelp and Simion, since they say that their view does not imply that knowledge can be valuable or non-valuable in other ways than through being commodious.

2.3 To Jon Elster

I owe a huge debt to Jon Elster who introduced me, in the early eighties, to so many issues: the nature of rationality and choice, motivated belief formation, akrasia and self-deception, and how to apply these to ethics and political theory. Although I fully agree with his diagnosis about the pathologies of social choice and share his criticism of the misuses of formalism and of fallacies committed by economists and other social scientists, I have some doubts about his classification of obscurantism into “hard” and soft”. Elster says that one cannot give a definition of obscurantism by intension, so gives only extensional characterisations. The distinction between “hard” and soft” is inspired by the distinction between hard and soft sciences. Soft obscurantism is the kind of bullshitting and “fashionable nonsense” which pervades many sectors of academia under the names of French Theory, postmodernism, etc. “Hard obscurantism” is the misuse and abuse of statistical models and more generally it is due to the intrinsic indeterminacy of rational choice theory and to its inability to deal with a set of irrational behaviours. The two do not seem to me to illustrate the same kind of epistemic vice. The “soft” version of obscurantism goes deep: it disregards and even scorns truth and reason. In contrast, those whom Elster calls “hard” obscurantists do respect the values of truth and reason, but they misuse them. It is a case of ignorance in matters of method. They are more like the bear in La Fontaine’s fable “The bear and the amateur gardener”: they want to do well, but they fail. As Elster likes to say: “To fail is to fail at something”. Both illustrate a form of stupidity, but the stupidity of the soft obscurantist seems to me deeper and more dangerous than the stupidity of the hard obscurantist. So the hard obscurantist seems to me softer, and the soft obscurantist seems to me harder with regard to epistemic vice.

3 On knowledge, perception and scepticism

3.1 To Paul Egré

Paul Egré gives an ingenious reconstruction of Gettier cases, and of the difference between accidentally true belief and knowledge in terms of the de re/de dicto distinction. This is illuminating, because both reference and knowledge involve a causal relation between mind and world, which is instantiated in de re situations, and not guaranteed when a propositional content or an expression is read de dicto. This seems to fit both the original Gettier cases and to a certain extent the fake barns cases. But does the fact that one can translate these situation in de dicto /de re terms capture our intuitions? A crucial feature of Gettier cases is that they involve a form of epistemic luck, which is not present in the difference between de dicto and de re reference. Neither is it present in the difference between propositions (which may be true or false) and facts (in the sense in which a fact is a state of affairs which obtains). What we want to understand, when we consider Gettier cases, is what more is needed for the causal relation to relate “in the right way” mind and world in order to be knowledge. A purely modal or semantic distinction helps to characterise the difference, but does not give us the beginnings of an explanation. This is why sensitivity or safety does not suffice. We need to explain further, either in terms of reliability, or in terms of virtue, or in some sort of relation connecting warrant and truth. I suspect that it is why theorists of reference appeal to psychology and to epistemic notions in order to explain the semantic relation. Egré himself finds support in empirical work about our intuitions in this respect. Nevertheless, I am not sure that it would give us the full story, which has to be given in epistemological terms.

3.2 To Jérôme Dokic

Most of what I know about the philosophy of perception comes from Jérôme Dokic. I accept, like many philosophers, what he calls the epistemic conception of perception (ECP), according to which perceiving is knowing: if we perceive that something is the case we hereby know that it is the case. Dokic presents us cases of perceptual hysteresis which threaten this view. These are cases where the experience of a given state of affairs is veridical, has a stable content, but where the subject does not know the content. It is presupposed that the content of perception is propositional, and not, for instance, based on non-conceptual content. It is also presupposed that knowledge is safe, i.e that margins of errors are possible. It is also presupposed that the experience is veridical, hence not hallucinatory. I am tempted to answer in the way suggested by Dokic: although at the time when the object disappears the subject is not any more related to it and does not know the content, he has a previous knowledge in t* which allows him to retain his former knowledge at t. Dokic rightly rejects the idea that it could be a species of memory belief. But can’t we say that the subject has in t* a tacit knowledge of what he has seen? Indeed the notion of tacit knowledge has to be explained and cannot be taken from granted, but it might be that if we had more information about visual processing in the possible effects of cognitive penetration in such cases we could maintain ECP. Another option, which Dokic does not consider, would be to use a distinction (mentioned by Williamson) between seeing that P and seeing a situation in which P. A husband might see a situation where his wife kisses lovingly another man, without seeing that she is kissing her lover. He might come to realise later, hence to know, that he has seen his wife in the arms of her lover. Adapting this distinction to perceptual hysteresis cases might allow us to say that although ECP is in trouble, they are not strictly speaking cases where the subject perceives but fails to know. This would not necessarily imply that the content of such seeings is non-conceptual. It may be semi-epistemic. This would probably bring us back to the notion of being in a position to know which Dokic calls the weak form of ECP. So I am not convinced that such a weak ECP is refuted by his examples of perceptual hysteresis.

3.3 To Keith Lehrer

In 1978 in Berkeley a fellow student handed me a copy of Knowledge by Keith Lehrer and said: “If you want to read real analytic philosophy, this is it”. Since then Lehrer has been one my main teachers in epistemology. Among other things he introduced me, long before the topic became fashionable, to the belief/acceptance distinction. His self-portrait is hung on the wall in my study. In his contribution, Lehrer gives a masterly overview of his present treatment of Moore on knowledge and scepticism. I agree with him that for Moore to convince us against the sceptic, he must not only take his purported knowledge that he has hands at face value, and proceed, using closure, to his claim that he can’t be the victim of a sceptical scenario and that there is an external world. I agree with Lehrer that Moore does not give us an explanation of why he knows that he has hands. Many have held that the reasoning is circular, and actually presupposes what is in question: that there is an external world. As I understand it, Lehrer’s explanation involves three elements in which readers will recognise the main pillars of his theory of knowledge: (i) an element pertaining to the nature of the intellectual state or act involved: accepting the Moorean premise that one has hands, and not simply believing it; (ii) a “Humean” element: basing one’s acceptance on the evidence of the senses, through taking sense impressions as exemplars of the meaning of representations (here: of hands); and (iii) an element of self-trust, which has also to be accepted (one accepts that one is trustworthy) combined with a coherentist element: acceptance must not be defeated and has to be part of a coherent system. Lehrer’s view, as well as his reconstruction of Reid, is internalist: the subject bases the reasonableness of his acceptance on the evidence of the senses, and accepts his trustworthiness. One could ask here more about what kind of acceptance is involved, and I am a bit wary about what “acceptance of one’s trustworthiness” means. I would also like to know more about what “reasonable” acceptance means, especially since Lehrer says that reasonableness is “undefined”. Some (including myself in a previous exchange with Lehrer) have been concerned by the potential circularity or bootstrapping which the principle of trustworthiness seems to involve. But Lehrer (2012: p. 166) has replied that he sees the role of acceptance and trust as securing mutual support in the coherentist sense: the acceptance of the principle of trustworthiness is supported by our other acceptances. If this is so, and if it is the way we ought to understand the Moorean argument, then Moore has to accept a lot of things in order to say that he knows that he has two hands. I once thought that a form of prima facie justification or entitlement to the premises would be enough to secure Moore’s argument, but I came to doubt this. I have taken a more externalist line since then. Moore has evidence that he has hands, evidence that he would not have if he were in an evil demon scenario, although this evidence is not accessible to him. He has a reliable belief that he has hands. But in order to know that he has hands, I do not think that he has to accept that he is trustworthy, although he must know (not necessarily reflectively) that he is reliable. We can talk here of acceptance, but not in the sense of taking from granted or assuming. The acceptance needed must be based on one’s knowledge of one’s own reliability. I agree with Lehrer that we have to introduce acceptance as a different state from belief, but acceptance for me supervenes on knowledge, rather than it is a building block or a keystone to justification. In other words, an account of Moore’s knowledge, and a reply to scepticism, has to include both internalist and externalist elements. Like most epistemologists who espouse a form of dualist view I still do not know how to combine the two elements convincingly and how to defend what is sometimes called “the holy grail of epistemology”.