2. EXAMINING KNOWLEDGE

We readily ascribe knowledge to animals as well as to adult humans. This leads some philosophers to think that there is no essential difference between knowledge-ascriptions in the two cases. This conclusion is too hasty. Clearly, human knowledge is more extensive and in various ways more sophisticated than animal knowledge (though animals can do things we can't). But is human knowledge, at bottom, more of the sort of knowledge that animals have, or do the differences amount to a difference in kind? This could be so, even if there is a generic concept of knowledge that applies both to animals and human beings.

We can ask many kinds of questions about knowledge. We might be curious about what is known about various matters at different times and places, and how that knowledge was obtained: a matter for historians of science, amongst others. Alternatively, we might want to investigate the capacities and processes that produce and sustain the individual human being's beliefs about the world around him, and how those processes sometimes lead a person astray: a matter for psychology or cognitive science. Looking beyond the individual, we might study the workings of the social institutions that make possible large-scale investigative projects of the kind we see in modern scientific research: a matter for sociologists, or ethnomethodologists. These are all empirical questions about what various people (or animals) know and how they know it.Footnote ¹ But there are also questions about knowledge as such. How does knowledge differ from mere belief or opinion? Does knowledge depend on justification, whatever that is? Is knowledge as such a suitable object of theoretical investigation at all? My questions are of this kind. They concern the nature of knowledge and so belong to the metaphysics of epistemology.Footnote ² It is not obvious that such questions are straightforwardly empirical, even if empirical findings are relevant to them.

So much for the object of my investigation. Turning to my second, methodological question, what is the most important thing to get clear about, if we are not to go astray concerning the nature of knowledge? There seem to be three possibilities. One is knowledge itself: the phenomenon of knowledge, we might say. But many philosophers think that we should start farther back and look into the concept of knowledge. Some – Austin for example – start by examining the word “know.” How we take these objects of interest to relate to one another depends on what other philosophical views we hold.

The suggestion that we should focus on the phenomenon of knowledge, conceptual questions being of minor importance, implies that knowledge is a natural phenomenon. From this standpoint, understanding the nature of knowledge does not differ essentially from understanding the nature and properties of any other natural kind: water, say. But how could knowledge not be a natural phenomenon? The most significant way concerns the dependence of knowledge itself (mature human knowledge, remember) on possession of the concept of knowledge. Many social “phenomena” are like that. No one would be offside if the concept of offside did not figure in the rules of football. So we had better begin with the concept of knowledge. Is it more like water or more like offside?

With respect to concept and word, it seems obvious that there are concepts that one could not have without the words to express them. Concepts that figure in the rules of games are like that. But so are scientific concepts. The Newtonian concepts of gravitation, mass and acceleration are grasped through understanding the dynamical laws in which they figure. If the concept of knowledge (mature human knowledge) is like this, there will not be much difference between examining the concept and examining the word. As it happens, I don't think that the concept of knowledge is exactly like this. Nevertheless, the best way to get a grip on the concept of knowledge is to examine our practices of claiming and attributing knowledge. To what extent these practices require possession of explicit epistemic vocabulary is a further question.

3. A NATURAL PHENOMENON?

The view that the proper object of epistemological investigation is knowledge itself has been forcefully defended by Hilary Kornblith. Kornblith relates his view of knowledge to his broader philosophical naturalism. He writes:

Physicists are interested in concepts like that of electron only in the sense that they use such concepts to understand the nature of matter. Concepts may need to be modified or replaced. But conceptual innovation is part of replacing existing theories by better theories. In the natural sciences, concepts are never the object of investigation. Epistemologists should take the same view. The concepts they need to understand knowledge will emerge in the course of their inquiries. Kornblith concludes that, “for those who wish to understand the nature of knowledge, there may be no reason at all to care about our concept of knowledge.”Footnote ⁴

Kornblith can't mean that there is no reason to care about the concept of knowledge. The very idea that knowledge is a natural phenomenon takes for granted a view of the nature of knowledge that is more the result of reflection on the concept of knowledge than of scientific investigation. Like all naturalistic epistemologists, Kornblith is an externalist-reliabilist. In the broadest terms, he equates knowledge with true belief produced and sustained by suitably reliable cognitive processes. So a more precise way of stating his view is that the concept of knowledge is of little further interest, once the basic conceptual question has been settled in favor of an austere version of reliabilism. Given reliabilism, the interesting questions concern the processes on which knowledge depends.

At this level of abstraction, and waiving questions about how human belief might differ from animal representation, we can put human knowledge and animal knowledge in the same box. But the question of specific differences remains on the table. Is austere reliabilism a plausible account of mature human knowledge? If not, do its inadequacies compromise the view that such knowledge is a natural phenomenon? These conceptual issues are far from settled.

4. ‘KNOWLEDGE’ AS A NORMATIVE CONCEPT

If knowledge is a natural phenomenon, the concept of knowledge must be a natural-kind concept. But it is far from obvious that the concept of knowledge should be thought of this way.

Our current understanding of natural-kind concepts is prefigured in Locke's account of our complex ideas of substances.Footnote ⁵ As Locke points out, though our conception of a substance will involve observable characteristics, it will mostly involve the substance's “powers”: its dispositions to act or be acted on by other substances, or to respond to changes in environmental conditions. Gold is soluble in aqua regia, malleable, fusible, etc. In a more contemporary idiom, natural-kind concepts are law-cluster concepts. The application of a natural-kind concept supports robustly counterfactual inferences: if this lump of gold were placed in aqua regia, it would dissolve. The law-clusters that such concepts involve are stable and empirically discovered. Our knowledge of them is indefinitely expandable. In this way, they are independent of us: given phenomena. But there is a further element in the concept of a natural kind: the idea of some deeper property that is the basis of that substance's powers, and which therefore explains why such stable law clusters are found. Locke calls this deeper property of a substance its “real essence,” it contrasting with the observable properties that (initially) define its concept. When known, real essences provide theoretical principles of classification that can over-ride observable similarities, as in the case of whales and fish.

Locke contrasts ideas of substances with what he calls ideas of mixed modes. Such ideas or concepts determine what are, in a sense, conventional rather than natural kinds. Locke's prime examples are legal concepts, such as murder. Other examples can be drawn from games or from economic activity: offside in (association) football, or money. Such concepts contrast with natural-kind concepts across the board. The modality involved in natural-kind concepts is alethic: physical (or more broadly natural) necessity: what must, may or cannot happen. The modality involved in legal, game-related and economic concepts is deontic: proprieties regarding what must, may or may not be done, which can of course be violated. Murder is unlawful killing. To avoid being called offside, a player must be positioned so that there are two opposing players between him and the goal when the ball is kicked towards him, provided that he is not in his own team's half where there is no offside. Money is legal tender. (When informal, as with cigarettes in prisons, money reflects local community norms that are not officially recognized.)

Things like murder, offside and money have no real essence: they are what we say they are. Since norms can be modified, we are responsible both to and for them. This does not mean that such norms are arbitrary: the law relating to murder may be reformed for reasons of policy or morality; the laws of football are intended to make for an interesting and competitive game, given the normal range of human physical capacities. Without money, a sophisticated market economy cannot exist. Institutions are adapted to natural conditions and social ends. Finally, the existence of such institutional phenomena depends on their being recognized, thus on those who participate in them possessing the relevant concepts. Of course, crimes such as murder are open to empirical investigation and explanation: are murder rates rising or declining, and if so why? Even so, murder is an institutional not a natural phenomenon.

Let us focus on the sort of concepts that we find in the law. Is the concept of knowledge – mature human knowledge – a natural-kind concept, or is it more like a legal concept? I'd say it's more like a legal concept. The familiar idea that knowledge is justified true belief is naturally understood in deontic terms. To be justified in believing that p is to be entitled (in the light of epistemic standards) to use one's commitment to p in inferences (theoretical and practical), to inform others that p, and so on. Epistemic permissions are earned by fulfilling epistemic obligations. None of this is true of the knowledge that we ascribe to animals or small children. I think that this is why talk of animals' having knowledge is much more natural than talk of their having justified true beliefs, even though philosophers often equate knowledge with justified true belief.

In arguing that the concept of knowledge is of no great interest, Kornblith turns a critical eye on Edward Craig's “genealogical” account of the concept of knowledge. Instead of trying to formulate necessary-and-sufficient conditions for knowing, Craig suggests we begin by asking

Craig's suggestion is that human beings, by virtue of their need to share information, need to be able to identify reliable informants. This is the need to which the concept of knowledge answers. In Craig's words, “To put it briefly and roughly, the concept of knowledge is used to flag approved sources of information.”Footnote ⁷ However, this is putting it very roughly. As Craig goes on to argue, informants are not mere sources of information. They are sources of a special kind. Informants tell us things and “there is a distinction between a person's telling me something and my being able to tell something from observation of him.”Footnote ⁸

As Kornblith notes, Craig motivates his strategy by invoking a number of contrasts:

Kornblith is dubious about all this. As he says, the point at issue is whether the category of “knowledge” owes its existence to human activity, as does the classification of tables into coffee tables, dining room tables, and so on. This classification does not aim to cut nature at the joints, but flows directly from the different uses that we find for tables. But why think that epistemic categories are like this? The point that knowledge plays an important social role cuts no ice. Gold plays important social roles, but gold is a natural substance for all that. Indeed gold's ability to play these social roles is partially explained by its natural properties, such as its rarity and resistance to corrosion. Kornblith concludes that it is not at all implausible “to think that knowledge, like gold, plays an important social role, yet its ability to play this social role is explained by deeper features that it has … independently of the social role it plays.”Footnote ¹⁰

This misses the point. The question at issue is not whether knowledge is like gold but whether it is like murder or money; not whether knowledge plays a social role but whether its existence depends on the social role it plays, whether it exists only in so far as its existence is recognized.

Kornblith thinks that knowledge is not in this way recognition- or concept-dependent. Animals have knowledge, though few if any have the concept of knowledge. Their capacity for knowledge features essentially in explanations of animal behavior offered by cognitive ethology. In that field, therefore, the concept of knowledge plays the explanatory role that other natural-kind concepts play in other natural sciences. However, by Kornblith's own account, the conception of knowledge in play in cognitive ethology is only that of reliably produced true belief. The question on the table is whether human knowledge has special features that make it essentially different from the knowledge we ascribe to animals, even though some notion of reliably produced true belief, though perhaps not quite the same notion, may be important in both cases.

As Kornblith notes, the simple claim that the function of knowledge-attributions is to keep track of approved sources of information gives only a very rough account of Craig's idea. Craig's just-so story begins with a “state of nature,” in which the practice of identifying and keeping track of good informants is closely tied to highly specific aims and situations. Those of us on the ground want to know whether a lion is approaching. Fred the lookout, stationed up a tree, is better placed to settle the question. In this restricted context, everyone knows what is at stake, how reliable the information needs to be, and who is best placed to acquire and transmit it. But as social life becomes more complex, information needs to be provided to recipients who are distant in time and space, whose interests in acquiring it will typically not be known. Standards for knowing must be suitable for information that is widely disseminated. The concept of knowledge as we now have it emerges from this process of “objectivisation”.Footnote ¹¹ Humans in the state of nature are only proto-knowers with a concept of proto-knowledge.

These complications matter. Craig's view of the relation between possessing the concept of knowledge and possessing a word for “know” is not altogether clear, though he often writes as if keeping track of good informants involves labeling them. However, in the highly constrained situations typical of information-exchange in the state of nature, identifying and keeping track could be a matter of know-how, with the consequence that explicit epistemic vocabulary would be neither necessary nor of much functional significance. Accordingly, if we focus solely on the primitive concept operative in the state of nature, as in effect Kornblith does, we might easily form the impression that there is no great distance between mature human knowledge and animal knowledge. But whether the concept of knowledge that emerges through objectivization would give the same impression is another matter entirely. Kornblith adds considerable plausibility to his criticisms of Craig by focusing on Craig's entry-point idea of tracking approved information sources, ignoring its subsequent development.

Finally and most importantly, Kornblith offers an unduly unsympathetic account of Craig's methodology. Though Craig himself does not articulate his meta-theoretical framework, his “function first” approach to the concept of knowledge is essentially pragmatic and expressivist. The way that Craig recommends approaching the concept of knowledge has already yielded dividends in accounts of other concepts – normative, semantic and modal – that have often been thought to be metaphysically problematic from a naturalistic standpoint. The sense of puzzlement comes from thinking of the semantics of such vocabularies in referential terms. Taking this approach, we will ask what features of the world such concepts refer to? Some kind of reductive naturalization will then seem to be required, if we are to avoid populating the world with properties that science does not recognize. (In epistemology, one of the attractions of austere reliabilism is that it responds to a related concern.) But from an expressivist standpoint, such worries are misplaced. Natural-kind concepts earn their living in the dimension of description and explanation. Normative, semantic and modal concepts are in a different line of work. Even if Craig gets ahead of himself by announcing at the outset that knowledge is not a given phenomenon, his strategy has the potential to vindicate this claim.

5. EXPRESSIVE AND DESCRIPTIVE

The idea that freestanding claims to or attributions of knowledge have a distinctive expressive-performative function is not new. Austin remarks that saying “I know” resembles saying “I promise.”Footnote ¹² In making an explicit knowledge-claim, I invite you to take my word for something. Ayer makes a similar suggestion: in saying “I know …,” I “vouch for the truth” of what I say.Footnote ¹³ If I lack the right to be sure, I am not entitled to invite you to take my word. So to invite you to take my word is implicitly to claim the right to be sure. No doubt in claiming a right, I also undertake the obligation to vindicate my claim, should the need arise. Craig is very much in this tradition, but with a complementary emphasis on third-person knowledge-attribution rather than first-person knowledge-claims. In claiming knowledge I invite your approval; in attributing knowledge I give you mine.

Austin sometimes seems to suggest that epistemic vocabulary is expressive or performative rather than descriptive. This is a temptation to avoid. Statements about who knows what occur as the antecedents of conditionals. The occurrence of “Tom knows” in “If Tom knows, he'll tell everyone” ascribes nothing to Tom. Knowledge-talk must have conceptual content as well as an expressive-function. The way forward is to see how the functional characteristics and significance of a certain vocabulary-item – expressive-performative function certainly, but perhaps also the utility of having a vocabulary-item with that function – constrains its conceptual content, understood as determined by its inferential engagements and epistemic character. So understood, conceptual content will enable pragmatic significance. The constraining and enabling relations are reciprocal.

By way of illustration, consider deflationary accounts of truth. Such accounts show how the truth-predicate's disquotational property – in effect, a rule licensing the inference from p to ‘p is true’ and vice-versa – underwrite truth-talk's function as a generalizing device. In this way, the function of truth-talk is revealed as essentially expressive, allowing us to express commitment to propositions that we cannot simply list: “Trust John: everything he tells you will be true.”

Expressivist accounts of alethic modals proceed along the same lines. According to Ryle, to represent a conditional as necessary is to issue a “season inference-ticket.” Sellars takes a similar view.Footnote ¹⁴ The expressive function of modal operators is enabled by their material-inferential characteristics. Thus “Necessarily, if p then q” entails “Not possibly p and not-q”: a license, ceteris paribus, to infer q from p just is a license to ignore certain (apparent) possibilities. Or since season inference-tickets are always available for use, whether or not used, alethic modals support subjunctive conditionals.Footnote ¹⁵ Different flavors of alethic modality may share common logical features but still be distinguished by their epistemic characters. Where mathematical necessities are established by proof, physical necessities are subject to revision in the light of empirical evidence. These would be different flavors of necessity even if both conformed to S5.

Craig's rather original proposal is to treat the concept of knowledge in a similar fashion. There is no reason to dismiss his approach out of hand.

6. EPISTEMIC AUTHORITY

Whereas Craig's approach to explicating the concept of knowledge takes the form of a just-so story, Robert Brandom offers a suggestion along Craig's lines, but without the genealogical trappings and with a more pronounced deontic flavor. Thus:

In this pragmatic recasting of the justified-true-belief analysis, knowledge emerges as a “complex, hybrid deontic status.” The attributor acknowledges commitment on the claimant's authority.Footnote ¹⁷ But although Brandom dispenses with Craig's fictional genealogy, he too faces questions that arose for Craig, namely, the functional significance of explicit epistemic vocabulary, and how possession of such vocabulary affects epistemic concepts (if it does). There seems no reason why a practice of epistemic deference – taking someone to be a knower – should depend essentially on having the ability to talk about knowledge. Now it is plausible that assertion itself is governed by epistemic norms; and in this spirit Brandom holds that assertion is essentially connected with the willingness to justify one's implied entitlements by reasons.Footnote ¹⁸ But this doesn't change anything: a practice of reason-giving can exist without the ability to talk about reasons. I warn you that a lion is approaching; you give me a quizzical look: I point to Fred who is signaling from his treetop post. As Kornblith says, sensitivity to reasons does not require the ability to talk about reasons. Similar consideration apply to the so-called “Knowledge Rule,” according to which one should only assert things that one knows to be so. Such a rule could be respected in practice even by speakers without the vocabulary to formulate it. Indeed, in the simple situations encountered in Craig's state of nature, Brandom's rule and the Knowledge Rule would be hard to tell apart.

Let us turn, then, to a direct examination of epistemic authority. Brandom treats knowledge as a hybrid, involving elements of both commitment and entitlement. But I want to claim that entitlement, when it reflects epistemic authority amounting to knowledge, is a different kind of hybrid in that it is appraised in two dimensions: truth-reliability and epistemic responsibility. Authority is compromised by a shortcoming in either.

Some preliminary words about reliability. The rationale for this requirement is straightforward. As Craig notes, when we are seeking information we cannot readily obtain for ourselves, our knowledge of a potential informant's situation will often be incomplete. We therefore want an informant who can be counted on to provide correct information across a range of possible situations. In Craig's example, Fred is up a tree and I am on the ground, putting Fred in a better position than I am to settle the question of whether a lion is approaching. To receive a timely warning, I need an informant with the capacity and opportunity to spot lions more reliably than I can.Footnote ¹⁹ But as we have noted, at least in primitive cases of assertion and communication, the reliability constraint could be observed by people without epistemic vocabulary and thus without any sophisticated capacity for self-conscious reflection on the constraint governing the practice of epistemic deference. Human knowledge may be different from animal knowledge, but so far it doesn't look that different.

The distinctive character of mature human knowledge only emerges clearly with an examination of the idea of epistemic responsibility, or rather ideas of epistemic responsibility, for epistemic authority involves epistemic responsibility in three senses of “responsibility.”

The first and most fundamental sense is what I claimed at the outset to be the feature of human knowledge that makes it special: accountability. To be an epistemic subject just is to be accountable (responsible) for what one believes, in the way that to be a responsible agent is to be accountable for what one does. This involves being accountable to ourselves and others. In the case of animals and small children, we appraise their performances, but we do not hold them to account, either for what they believe or what they do. In the case of small children, we begin at quite an early age to treat them as if they were subjects and agents, as part of the process of acculturation that will eventually make them such: they are not responsible for what they think or do. But we cut them (or ought to cut them) a good deal of slack. We hesitate to apply to animals or very small children the deontic notions we apply to human adults. If the dog continues to sit at the base of the tree, having failed to notice that the squirrel has escaped on to the roof, we can say that the dog is making a mistake. But it would be odd to ask whether or not the dog is entitled to think that the squirrel is still around. It is not clear what would count as an answer.

Do animals hold each other accountable? In some social animals there is behavior that can be described as a primitive form of sanctioning.Footnote ²⁰ But the kind of accountability that goes with being an epistemic subject (remember again that we are talking about mature human knowledge) is importantly different. According to the norm of accountability governing mature human knowledge, human beings cannot properly be held to account unless they can account for themselves. This is why small children are not fully accountable. One is not an epistemic subject without some degree of epistemic self-conscious. Participants in a norm-governed practice must in some sense understand the governing norms. In Kantian terms, they must not merely act in accordance with rules, but for the sake of them. To be an epistemic subject, it is not enough to be a reliable signaler of events in the passing scene. By virtue of being accountable, a subject must possess some degree of understanding of his or her reliability, and in particular of his or her limitations: of how it is possible to go wrong. But while making accountability a condition of subjecthood requires breaking with austere reliabilism, it does not entail denying that reliability is essential for knowledge.

Mention of norms governing epistemic practices brings us to the second idea of epistemic responsibility: doing one's due diligence. To fail to be responsible in this sense is to be irresponsible. Epistemic irresponsibility defeats epistemic authority, even if an irresponsibly held belief happens to be true. The requirement to avoid epistemic irresponsibility also has a fairly obvious instrumental rationale: to promote reliability. However, it would be wrong to think that epistemic responsibility's only rationale is instrumental. I shall come back to this point.

The two kinds of responsibility – accountability and subjection to norms of epistemic conduct – are closely related. We are accountable for our beliefs by virtue of being responsible to epistemic norms, including norms of proper procedure. As with the reliability constraint, subjects cannot be held to account for violating norms that they do not understand. Note, however, that this understanding may be largely implicit and practical. It will be manifested by the normative attitudes that are taken in practices of criticism and justification, e.g., the objections or questions that are taken to impugn entitlement and the responses that are required to vindicate it. I am not suggesting epistemologists with a professional interest in making epistemic norms explicit are the only genuine subjects.

Because knowledge requires both responsibility and reliability, epistemically responsible believing confers only what is sometimes called “personal justification.” It does not guarantee knowledge. I can be as careful as anyone could demand and still end up following an epistemic procedure which is – in the circumstances – unreliable. If you suspect that this has happened, you will not take me at my word. To be knowledgeable, I have to satisfy the reliability constraint, not just do my best to satisfy it. The grounds of epistemic authority are not purely procedural.

The possibility of non-culpable failures of authority brings us to the third idea of epistemic responsibility. We must distinguish accountability as a general condition from liability – being subject to sanction – in a particular case. Here, I am inclined to think that action and belief diverge. In the case of action, the liability structure is default and challenge: we are presumed liable (thus praiseworthy or culpable) unless we have an excuse that cancels or mitigates our (degree of) responsibility. In the epistemic case, liability is strict. Violation of the reliability constraint renders a person subject to epistemic sanction in the form of loss of authority even when he or she has done nothing wrong: that is to say even when the violation is non-culpable.Footnote ²¹ Strict liability is a consequence of the reliability constraint. In law, strict liability applies (where it applies at all) to undertakings that are inherently risky, encouraging participants to exercise due care and attention. Assertion is like that. We are not always (or even often?) cognizant of all the ways we might go wrong. This does not mean we are never entitled to claim knowledge. But it does mean that others are entitled to deny us authority when unreliability, culpable or otherwise, is reasonably suspected.

7. ACCOUNTABILITY AND REFLECTION

Accountability entails epistemic self-understanding. In some way, subjects must recognize the norms by which they are bound. Humans are self-consciously norm-bound in ways that animals are not. However, it is all too easy to exaggerate what this element of epistemic self-consciousness involves, equating the need for epistemic self-consciousness with the claim that in all cases justified believing depends on critically examining any presupposition that is relevant to a belief's positive epistemic status. This view is not only psychologically implausible: it is excessively demanding, thus apt to trigger a recoil to pure reliabilism.

Laurence Bonjour once defended this view in the course of developing what he took to be counter-examples to purely reliabilist accounts of knowledge. Here is one.

Given Norman's de facto reliability, pure reliabilism would credit Norman with knowledge. BonJour argues that this is the wrong result:

Belief in the absence of critical reflection on epistemically relevant presuppositions is always and everywhere epistemically irresponsible.

BonJour's argument is an object lesson in the perils of arguing from contrived examples. He means the argument to show something about epistemic responsibility in general: that there is a standing obligation to reflect critically on the epistemic status of one's beliefs. But Norman is a more than a bit peculiar. It is interesting that BonJour describes his belief as not merely “unjustified” but “irrational.” Rightly so: Norman seems not so much epistemically careless as borderline deranged. The conviction that the President is in New York comes to him out of nowhere. Presumably, this is the first time such a thing has happened to him, for if previous hunches had worked out, he would have had reason to trust this one. But Norman doesn't have a hunch: he simply believes that the President is in New York. To put it mildly, it is very unusual for a person to have a firm belief, to have no idea how he came by it – not even to be able to say things like “I was there,” “I read about it somewhere,” or “I learned it in school” – and yet to be completely unperturbed by or even curious about this strange occurrence. Apparently, BonJour thinks that it doesn't matter whether Norman even suspects that his belief, if true, would indicate clairvoyance: the thought doesn't have to cross his mind. But maybe it should have. Then he would have had to ask himself about the likelihood of his possessing such powers, since there are very good reasons for thinking that no one has them.

Perhaps we are supposed to imagine that Norman is unaware of these reasons. But this won't help. Maybe he should have known about them: on any plausible account of epistemic responsibility, there is such a thing as culpable ignorance. And even when ignorance of relevant facts isn't culpable, it is still an epistemic shortcoming. As I have already argued, failings with respect to reliability compromise epistemic authority even when non-culpable. For all BonJour has shown, Norman's lack of justification has to do with his failure to engage in critical reflection in an epistemic situation that cries out for it. BonJour thus fails to establish the general conclusion he draws. The most that follows from his example is that there is an obligation to reflect critically on the epistemic status of a belief when there is reason to do so.

As Kornblith points out, the claim that there is an unqualified obligation to reflect critically on one's beliefs sets an unreasonably – perhaps impossibly – high bar for justification. If in order to responsibly believe that-p I have to reflect on my reasons for believing that p, these reasons will themselves have to be responsibly believed. A regress looms, and with it the unpalatable choice between internalist foundationalism, coherentism and infinitism, all of which face well-known difficulties. However, pursuing this important objection would take us too far into a discussion of the traditional skeptical problem of the regress of justification. I shall therefore set it aside. Kornblith raises a very interesting and less familiar objection: namely, that there is a regress lurking in the idea of critical reflection itself.

Before getting to this objection, it will be worth considering an example that Kornblith offers to explain why the mistaken view that there is a general obligation to reflect critically on one's beliefs might nevertheless seem attractive. At home, listening to news on the radio, he hears the newscaster say that government policies have led to a rise in unemployment. But he is busy making dinner and gives no thought to whether the radio station or the particular reporter is a reliable source of information on economic issues. He doesn't even wonder whether the reporter's claim fits in with what he takes himself to know about government policy or the causes of unemployment. He just takes the report on board, thoughtlessly as we might say. According to Kornblith, it is tempting to conclude that he is not really justified in the belief he forms because of his failure to reflect on its epistemic status.

This example avoids some of the defects of BonJour's Norman case. BonJour wants a case in which Norman has a belief but no idea why he holds it. Since this is rarely (if ever) the case with beliefs acquired in normal ways, he appeals to a belief formed in an abnormal way: clairvoyance. However, most of us think that clairvoyance doesn't exist, which may color our reaction to the example. But while Kornblith's example is superior for being mundane, it is no more effective. It fails to makes even a prima facie case for the view that there is a general obligation to reflect critically on our beliefs.

Kornblith's argument illustrates the perils of arguing from examples that are under-described. In Kornblith's example, his failure to reflect on whether the station or reporter is a reliable source of information leads to an unjustified belief. But was there any call for such reflection? It is hard to say: the example is just too sketchy.

Let us start with questions about the station and the reporter. Most of us choose the stations we listen to, and in the case of news stations reliability is an important factor in the choice. If this is the case, Kornblith doesn't need to ask questions because he has already answered them, at least to his own satisfaction. Suppose, then, that the station got changed without his knowledge. A change of station is something one can generally recognize: news programs on different stations have distinctive styles. If Kornblith fails to notice the change of station, then his belief may well be unjustified. But the reason for this is not merely that he failed to reflect. It is that, being otherwise occupied, he missed a warning signal that could and perhaps should have given him pause. Like Norman, he failed to reflect when there was reason to do so.

Moving on to the reporter's claim, what the reporter offers is not naturally described as “information.” Rather, it is an explanatory claim about the cause of current unemployment. Explanatory claims concerning economic matters are notoriously controversial, not least because the theoretical preferences of economists are far from free of broader political convictions. This is why they belong in editorial segments. Suppose that the newscaster had merely reported what Norman believes: that the President was in New York City. This is a simple piece of information that just about any news station can be counted on to get right. But in this case, I am not even slightly tempted to think that Kornblith is unjustified in his belief. It is not a mere failure to reflect that accounts for the sense that Kornblith has failed in his epistemic duty: it is his casual acceptance of a controversial claim. Again, Kornblith's example points in the same direction as BonJour's Norman case: there is indeed an obligation to subject one's beliefs to scrutiny, but only in appropriate circumstances.

Another defect that Kornblith's example shares with BonJour's is a cavalier treatment of how one acquires a genuine belief: that is, a stable conviction. Kornblith says that he doesn't reflect on whether the newscaster's claim is plausible in the light of what he (Kornblith) takes himself to know about current economic affairs. No doubt we all do this sort of thing when we hear something that we already disposed to find reasonable. But suppose that the newscaster had said something utterly outrageous: outrageous, that is, in the light of Kornblith's background knowledge (or what he takes to be such). If Kornblith were really listening, even though primarily engaged in other activities, the claim itself would have grabbed his attention. If on the other hand he were not paying much attention, it is unlikely that he would have acquired a stable belief: in one ear and out the other.

With these points in mind, we can turn to the regress that Kornblith finds lurking in the very idea of reflection. If Kornblith reflects on his belief that current levels of unemployment are the result of the government's economic policies, he forms beliefs about his beliefs: about the reliability of the station, how far these beliefs give him reason to believe the claim in question, and so on. But as Kornblith says, his reliability in making these judgments is not the issue. Rather, on Bonjour's view, even if Kornblith is perfectly reliable in forming such second-order beliefs, they will be irresponsibly held, hence unjustified, if they have not been subjected to reflective scrutiny. An infinite regress results.Footnote ²⁴

Now in point of fact, BonJour realizes that there is a problem along these lines. Bonjour's coherentist view of justification is holistic: particular beliefs are justified by virtue of belonging to a coherent system of beliefs. Revisions to the system, whether by addition or subtraction, are justified or not according to whether they increase or decrease global coherence. But Bonjour's coherentism is internalist: it is not enough that a person's system be coherent; the person must recognize its coherence and manage his beliefs in the light of that recognition. BonJour candidly admits that his coherentist epistemology is structurally more complicated than foundationalist alternatives. So if he were willing to go externalist, foundationalism would be the structural option of choice. Coherentism is to be preferred because it is the better option for an internalist. Since the hallmark of internalism, as BonJour understands it, is the requirement that a person be aware of the factors by virtue of which his beliefs are justified, his coherentism entails that justified believing depends on a person's being aware of what beliefs belong to his system and how they hang together. Must these second-order beliefs, to be justified, pass the test of coherence? If the answer is “Yes,” as an unrestricted demand for critical reflection implies, we will need third-order beliefs about our second-order beliefs and how they hang together with themselves and with our first-order beliefs. This is Kornblith's regress, as it arises for BonJour's coherentist conception of justification.

BonJour blocks the regress by appealing to what he calls “the Doxastic Presumption.”Footnote ²⁵ According to this epistemic principle, for the purposes of critically assessing a first-order belief (asking whether it could be adopted without compromising coherence), a person is entitled to take his sense of the contents and current coherence of his belief system for granted. At one time, it seemed to me that this move turned BonJour's supposedly coherentist view into a variant form of internalist foundationalism by virtue of assigning a privileged epistemic status to second-order beliefs. I now think that this isn't quite right. What the Doxastic Presumption amounts to is a restricted waiver of the demand for critical scrutiny. The claim is not that second-order beliefs are privileged, because intrinsically credible in the manner of the basic beliefs of classical foundationalism. It is rather that, for the purposes of critically assessing first-order beliefs, we may permissibly – i.e., responsibly – take our reliability in forming second-order beliefs for granted.Footnote ²⁶

BonJour's own response to the reflective regress shows the unreasonableness of his unrestricted demand for critical reflection. No plausible account of the role of critical reflection in epistemically responsible belief-management can incorporate such a demand. BonJour is not wrong to issue a waiver. His error lies in the particular waiver he issues, which applies only to second-order beliefs.

8. CLAIMS AND QUESTIONS

The correct form of the waiver has already emerged in our examination of the examples offered by BonJour and Kornblith. Epistemically responsible belief-management requires only that a belief be subjected to critical scrutiny, or justified by citing reasons, when there is reason to question either it or its epistemic status.

In his groundbreaking paper, “The Ascription of Responsibilities and Rights,”Footnote ²⁷ H. L. A. Hart insists on two fundamental points. One is the important ascriptive – or as I have put it expressive-performative – function of freestanding claims involving deontic vocabulary. The other is the prevalence in the law of what he called “defeasible concepts.” Hart gives the example of contract. For a valid contract to exist, certain positive conditions must be met: an offer by one party that is accepted by the other, a memorandum, and something of value that each party brings to the exchange (“consideration”). But even when these conditions are met, there are various grounds on which one of the parties can challenge the validity of the contract: for example, he entered into the contract under duress, or that the other party engaged in fraudulent misrepresentation. If the court upholds such charges, the validity of the contract is defeated. I want to suggest that epistemic authority is just such a defeasible concept. Provided that certain conditions are met, a person's assertions and beliefs count as responsibly held. They need no critical scrutiny, thus no defence by reasons, unless there is reason to question them. The structure of accountability with respect to epistemic authority is default and query.

Epistemic authority is thus analogous, not only to validity of contracts, but to the presumption of innocence in criminal law. By contrast, BonJour's demand for unrestricted reflective scrutiny treats belief as automatically under suspicion, so that anyone who fails to establish his right to hold the belief in question is ipso facto epistemically irresponsible. This has the effect of building into the requirements for epistemic responsibility a severe asymmetry with respect to the right to believe versus the right to question. In effect, there is an absolute right – indeed, in terms of each person's management of his own beliefs, an obligation – to question any and every potential belief's epistemic status. The right to believe is always to be established, whereas the right to demand reasons, thus the obligation to seek them, is a standing condition. Candidate beliefs are like immigrants or asylum seekers: they have to apply for a work-permit or the right to remain. Otherwise they are illegal or undocumented and should be refused entry or deported. This threatens to make responsible believing, hence knowledge, impossible.

The crucial feature of a default and query structure is that it involves no such asymmetry, because claims (beliefs) and epistemic queries are equally subject to epistemic norms. As in law, questioners need to have standing; and queries, like law suits, need grounds. Queries do not need to be addressed unless they meet standards of reasonableness and relevance. If they fail to meet such standards, they may not even be fully intelligible, since it will be left unclear what sort of defence is being demanded. Queries must be contextually appropriate, on pain of not being really queries at all.

Following Brandom, I used to call this a “default and challenge structure.” However, as Austin reminds us, epistemic queries – “How do you know …?” or “Why do you believe…?” can be asked in two ways.Footnote ²⁸ Asked pointedly, the implication is that perhaps you don't know or shouldn't believe: you lack appropriate authority. Pointed queries are challenges. Grounds for them will be a reasonable suspicion that a subject has defective credentials with respect to his reliability or responsibility in forming and holding the belief in question: as Austin says, “some definite lack.”Footnote ²⁹ But not all queries are challenges. There are also explanatory questions. You may concede that I do know but for various reasons be curious as to how I came by the knowledge you are happy to credit me with. Obviously, the legitimacy of such questions cannot depend on grounds for suspicion. But this does not free them from normative constraint. For an explanatory question to make sense, there must be something to explain. If your reasons for conceding that I know that p is that you already know how I know, this condition will not be met. But if I can't explain how I know something, you may have grounds for suspicion sufficient to motivate and sustain a challenge. So although the grounds for challenges and explanatory questions are different, the consequences of failing to provide a satisfactory answer may be the same: loss of authority.

In the case of contract law, a contract, though subject to defeat, is presumed valid if certain positive conditions have been met. If epistemic authority is such a defeasible conceptual, what are the positive conditions by virtue of which authority is a default condition? There are two kinds: standing and situational.

The standing conditions are implied by the very idea of accountability. Just as only competent adults can enter into contracts, only epistemic subjects can possess epistemic authority, and to be an epistemic subject just is to be accountable for one's beliefs. Two things follow. Obviously, to be epistemically accountable, a person must have beliefs to be accountable for. But the content of a person's beliefs are determined by the functional role that they play in the perception-inference-action nexus that is a person's way of coping with the world. If a person completely lost touch with the world, moving or speaking randomly, there would be no way of saying what he or she was thinking: indeed, no grounds for attributing beliefs at all. Having reasonably reliable recognitional and inferential capacities, operative in the pursuit of intelligible goals, is a precondition for having any beliefs to be accountable for. Simply to be an epistemic subject, then, is to possess a wide range of default entitlements.

Specific entitlements are situationally variable. They are most clearly visible in a Basic Communicative Context: a context involving the communication of information that (relative to the mutually recognized background knowledge and cognitive capacities of the interlocutors) is not notably exotic or surprising, either in itself or with respect to the claimant's access to it. In such a context, queries will not generally arise. Of course, not all contexts are like this. Some are contexts of investigation or argument. In such contexts, the focus of interest will be claims or beliefs to which objections have been raised, or which face recognized problems, or for which there is simply nothing to be said (so far as some parties to the conversation can see). In the last case, we have mere suggestions.

Let us return to the idea that justified believing always demands epistemic reflection. This idea implies that, since epistemic responsibility always depends on something that one ought to do, entitlement is always to be earned. The model I have been defending has no such implication. Perceptually based beliefs just come to us, and on the whole and for the most part we are entitled to rely on them without more ado. Of course, sometimes we need to take steps to get a better look. But there is no standing obligation to actively investigate the situation or our capacities in order to determine our situational reliability. Such obligations come into play through situation-specific red flags or specifically relevant background knowledge. They have to be triggered.

Having reliable recognitional capacities is what Sellars calls an “ought-to-be” condition. To be authoritative with respect to certain reports, I need to be reliable and to have no reason to distrust my reliability. There is nothing that I need to do, unless something in my situation calls for caution. If my recognitional abilities are not reliable enough for the matter in hand, then if it is worth my while I should take steps to refine them. Otherwise, I can just forego epistemic authority in certain cases, perhaps deferring to others.

When I introduced the idea of accountability, I said that beings that cannot account for themselves cannot be held to account. This implies that epistemic subjects must not just possess reliable capacities for perception, memory and inference: they must know something about their reliability. However, this epistemic self-understanding may be mostly exhibited in practice rather than “thematized.” It is important not to misunderstand why this is so.

Returning to Kornblith's newscast example, Kornblith remarks that one reason he might be thought irresponsible in simply taking the news report on board is that he fails to reflect on the reliability of either the station or the particular reporter. Behind this idea (which of course Kornblith only presents for explanatory purposes, without endorsing it himself) is the idea that knowledge of a source's reliability would at least enhance one's justification for believing information from that source. Some would say that without knowledge of a source's reliability one would have no justification for believing the “information” it provides. Now a radio station, it might be said, is a derivative source of information: its reliability can be checked by seeking information from other sources. But some sources, such as perception, memory and other cognitive faculties, are basic in that they are inevitably involving in any process of checking. There is therefore no way of checking the reliability of our most basic faculties without making use of those faculties themselves. It is has often been argued that attempting to validate the reliability of our basic faculties involves an incurable epistemic circularity. (This problem lies at the heart of what Ernest Sosa calls “the Pyrrhonian Problematic.”Footnote ³⁰) It might be thought that the view I am defending invites this skeptical challenge.

Since I am not trying to refute the skeptic but to understand the practice of knowledge claiming and attribution that we actually have, I cannot go deeply into skeptical matters. However, it is important to see that, on my view, the function of reliability-knowledge, with respect to basic faculties, is not primarily justificational. In general, what we know about the reliability of our basic faculties is no more certain than the knowledge that arises from their exercise and may often be less so. Further, reliability, as it figures in epistemic authority, is itself a defeasible concept. In properly brought-up adult humans, and with respect to a wide range of beliefs and situations, reliability is a default condition. However, it is essential that human beings recognize their cognitive limitations, because such knowledge is the basis on which legitimate queries are recognized and answered.

Our knowledge of our limitations and biases is indefinitely expandable. As Kornblith emphasizes, cognitive science has unearthed previously unrecognized influences on belief. The lesson to learn is that there never was a problem of epistemic circularity. Investigating our basic faculties, using those very faculties, leads not to self-congratulation but to a more precise and accurate understanding of their strengths and weaknesses. Kornblith is quite right to argue that we do not always need to satisfy ourselves as to the reliability of an information-source before responsibly relying on it. As an austere externalist, he concludes that reliability is never essential for knowledge at first-order. My claim is that such knowledge is an essential component in mature human knowledge, but that its function is not primarily justificational. Rather, it is an essential element in the contextually appropriate critical reflection through which we correct and improve our beliefs.

There is a general moral here. Skeptical concerns – whether avowed or subterranean – have a distorting influence on our understanding or ordinary epistemic practices. BonJour's unreasonably demanding conception of epistemic responsibility is not really derived from examples like the case of Norman, which only points towards the default and query conception just sketched. BonJour himself agrees that everyday justification is local: beliefs are justified against the background of a larger belief system that is taken (responsibly?) to be more or less in order. BonJour's real interest is in general skepticism, which challenges us to explain how anything whatsoever that we believe is justified. In the context of responding to this skeptical challenge, nothing can be taken for granted, so everything is open to scrutiny. My suspicion is that BonJour's interest in skepticism leads him to distort the conception of epistemic responsibility that is operative in real life, Unfortunately, the regress generating potential of BonJour's excessively demanding conception of epistemic responsibility makes that conception itself a potent source of skeptical worries. There is a symbiotic relationship between disease and cure. As a result, once we see that the excessively demanding conception of epistemic responsibility is an imposition rather than a discovery, the threat of skepticism recedes.

9. CONCLUSION

By way of conclusion, let me take up two issues left hanging at earlier stages of the discussion.

The first issue concerns the function of explicit epistemic vocabulary. Now the standard way of expressing a claim to knowledge is not to say “I know that …” but simply to make an unqualified assertion. If I am not certain (sufficiently certain, for present intents and purposes) I should warn you. If I don't, you are default-entitled to take me as representing myself as knowledgeable, which I shouldn't do if I am guessing or speculating. Explicit epistemic vocabulary comes into play when we seek or give assurance: i.e., when epistemic authority is in question and may need to be vindicated and explicitly claimed.

If this is right, there is considerable plausibility to the Knowledge Rule: that one should only assert what one knows to be the case. However, given the defeasibility structure of epistemic accountability, there is no conflict between the Knowledge Rule and Brandom's alternative rule that assertion involves a commitment to support what one asserts by reasons. In making an epistemically unqualified assertion, thereby representing myself as knowledgeable, I undertake the commitment to provide reasons should the need arise.

In my discussion of Craig and Brandom, I argued that a restricted practice of epistemic deference could exist in the absence of explicit epistemic vocabulary. The accountability structure would still be default and query, as we see in the case of Fred the lookout. But as the range of human undertakings, theoretical and practical, expands, the need to attribute knowledge becomes detached from situations in which the appropriateness of such attributions is immediately ascertainable. This is the process Craig calls “objectivisation.” It makes more sophisticated and self-conscious practices of knowledge-claiming and attribution involving explicit epistemic vocabulary indispensable.

The second issue concerns reliability. The reliability constraint makes it reasonable to think of animal and human knowledge as falling under a generic concept of knowledge as true belief, reliably produced and sustained. In the animal case, reliability can be understood with respect to the animal's functioning effectively in its natural environment. However, in the human case, the reliability constraint has a distinctive character, by virtue of its interactions with the deontic considerations involved in the idea of epistemic responsibility. Reliability itself becomes reliability for particular purposes. This is particularly evident in sophisticated forms of inquiry. In particle physics, the standard for “detecting” a particle has moved from three to five sigma, the standard in effect when the discovery of the Higgs boson was announced. Meeting this standard means that the probability of obtaining the data actually obtained on the null hypothesis (that the Higgs boson does not exist) is roughly one in three and a half million. This is a very high standard, but a reasonable one given that “discoveries” at three sigma – itself a high standard – have sometimes turned out to be statistical blips. Reliability is a norm that we are not only responsible to but, in certain applications, responsible for. We are accountable all the way down. That is what is special about human knowledge.