Epistemic akrasia and epistemic virtue

Christopher Hookway


1. Introduction

            Ordinary practical akrasia, incontinence, or weakness of the will occurs when someone consciously or deliberately makes a choice that she sincerely believes is wrong. She decides to continue smoking in spite of sincerely acknowledging that it would be best, all things considered, to give up; or she reaches for a cigarette while fully aware of her resolve to smoke no more.  Such phenomena are a distressingly common feature of most lives, yet philosophical theories of practical rationality, action and evaluation can have considerable trouble escaping from the conclusion that they are simply impossible. This makes study of practical akrasia an especially useful technique for uncovering important and heretofore unaccounted-for complexities in the structure of practical reasoning.

            This paper is concerned with whether a parallel situation arises in the study of theoretical rationality. Are there also cases of epistemic, doxastic or theoretical akrasia? In the most full-blooded form, such cases would occur if someone consciously accepts some proposition while also accepting that it is epistemically wrong to do so � perhaps she thinks that there is strong reason to accept its negation. This paper discusses why such phenomena can seem so problematic (section 4) but argues that, once we look closely at the structure of theoretical reasoning and inquiry, we can make sense of forms of irrational belief which are closely analogous to akratic action (section 5), and, indeed, there are many philosophically interesting examples of this (section 6).

            Although these issues are of intrinsic interest, they are examined here for the sake of the light they cast upon some general issues about epistemic evaluation.  Inquiries and deliberations are activities with distinctively epistemic goals: they are directed at solving problems of fact, at finding things out. Our epistemic normative standards are reflected in the ways in which we carry out these activities: whether our inquiries and deliberations take us to the truth will depend in part upon how skilfully we control their progress and upon the adequacy of the standards we employ in doing so. Making sense of epistemic akrasia will require us to understand some important complexities in the structure of inquiries and deliberations, and to see what sorts of capacities we must possess if we are to take responsibility for how well they are carried out. I shall conclude that if epistemic akrasia is, indeed, possible, then the normative regulation of deliberation and inquiry requires the agent to possess states or traits that are best seen as virtues. They are states, probably states of character, which are manifested in the ways in which we organize and regulate our activities, and which enable us to carry out those activities well or rationally. I hope to argue from the possibility of epistemic akrasia to the need to take epistemic virtues seriously in our theory of epistemic evaluation. [1]

            Section 2 explores some cases of ordinary practical akrasia, pointing out some features of practical deliberation and inquiry, and emphasizing the most important features which should carry over into plausible examples of epistemic akrasia. Section 3 then introduces a first weak form of epistemic akrasia: if activities can be carried out akratically and inquiries (and deliberations) are activities then, it seems, the latter can also be carried out akratically. This provides background for the ensuing discussion of full-blooded epistemic akrasia..


2. Ordinary akrasia: some examples

Philosophers� characterisations of akrasia, of ordinary incontinence, are surprisingly varied. Some describe the akratic as performing one action while acknowledging that there is better reason to perform another; others as performing an action while acknowledging that the reason for doing so is inadequate. Some identify the focus of akrasia as choice against one�s own best judgment (Wiggins 1987: 240), others as action (de Sousa 1987: 199, Rorty 1981: 175, Davidson 1970: 21, 22), yet others as intention (Williams 1990: 120). These different formulations need not be inconsistent: if all are failures of �continence�, if all exhibit lack of the same virtue, then they can be taken as different examples of a related set of phenomena. All involve a failure of sincere value judgment or commitment to have an appropriate influence upon the processes of deliberation and action. And the challenge they raise concerns how something can indeed be a sincere value judgment of mine if it is not manifested in my deliberations and actions.

            A preliminary sorting of some of these phenomena can be obtained by taking seriously the fine detail of practical deliberation. Consider an example.

Imagine someone who believes that it would be good to contribute to alleviating the suffering of famine victims in Ethiopa. On reflection he decides that it would be right, all things considered, were he to do so. The story could then be developed in three ways. Having formed a general resolution to help, his attempts to formulate a more specific intention - perhaps to write a large cheque to Oxfam or to some other agency - all somehow fail. The general resolve never turns into a more specific intention. Or having decided to contribute to Oxfam, and having decided upon the amount, he finds that he is always diverted by other concerns and the cheque never gets written. Or perhaps the need to pay for his holiday gains urgency and he readily (too readily) decides to postpone his donation for yet another month. And in all these cases, recognition of what he has done can be a cause for embarrassment and surprise. Somehow an evaluation he takes to be genuine, and an intention he takes to be real and sincere, fail to translate themselves into action. (Example borrowed and adapted from Velleman (1989: 138)).


Notice some stages in what is described here:

1.      The agent attaches value to the relief of suffering in Ethopia (G1); he judges that action to this end would be good.

2.      He decides that it would be right or best, all things considered for him to act in order to contribute to G1.

3.      He forms a general resolution so to act.

4.      He judges that the best means for achieving his goal, all things considered (or a fully satisfactory means for doing this) is to write a large cheque to Oxfam.

5.      He forms a specific intention to write a large cheque to Oxfam (G2).

6.      He retains that intention in the face of irrelevant competing claims upon his resources.

7.      He acts on that intention.

The three upshots we described above all involved this process being cut off before the final stage was reached. In the first case (there were two variants) it is cut off after stage three or stage four, before a more specific intention has been formed. In the other two cases, the more specific intention was formed but it failed to issue in action: in the second case, because it was abandoned for evidently poor reasons; and in the third case, through what is more straightforwardly a failure of will. These different possibilities (and they may not be exhaustive) all seem to be forms of akrasia. The agent fails to act and deliberate as is required by sincere evaluative commitments.

            The failings fall into two very broad classes. Let us take it that when I make a decision or form an intention, or indeed when I possess any intention at all, I acquire a distinctive commitment. Unlike some other goals or desires, a commitment is not just an end that will be weighed in the balance with others when the need for action arises. Unless a commitment is actually abandoned (often for good reasons), it possesses a kind of authority which prevents our treating it as simply one among a set of possible ends. The first class of evaluative failings concerns the relations between my evaluations - including evaluations all things considered - and my commitments. I may fail to decide or intend to do what I judge would it would be best to do; or I may decide or intend to do what I judge that I should not do. If the process described above were to halt at stage (2) or at stage (4), it would exemplify this pattern. The second class of failings concerns the fate of my commitments. They may just fade away for no good reason; or I may allow myself to abandon them for what I know to be bad reasons. Rationality requires me to ensure that my commitments respect my evaluations. It also requires me to be true to my commitments: abandoning them when there is good reason to do so; but sticking by them when there is no good reason to abandon them.  The other cases all fit this second pattern. According to David Wiggins, the second form of these provides the prime focus of continence: continence is an executive virtue that enables us to sustain our commitments. And Richard Holton has explained the ordinary notion of weakness of will as fitting the second pattern too (1999). When we turn to cases of epistemic akrasia, we must ask whether there are cases falling into each of these classes.

Akrasia, then, involves conflicts between our evaluations and our commitments, or between our commitments and acts falling under them. The most interesting and problematic cases occur when both of the conflicting elements are fully conscious or readily available to consciousness. When we behave akratically, we are often aware that this is what we are doing; this is not a form of behaviour that has, somehow, to be kept secret even from ourselves. The incontinent smoker is fully conscious of her resolve to give up her habit just as she reaches out for another cigarette. Indeed, as she acts, she may be fully confident that she will subsequently feel regret, shame or even guilt about what she is doing. In this it differs from self-deception: when the jealous man self-deceptively believes that his partner is being unfaithful to him, he cannot be, at the same time, aware that he is utterly unreasonable so to believe. Epistemic akrasia should inherit this openness to consciousness from its practical analogue: a simple and extreme case would be one where someone consciously accepts a proposition while �judging� that he is wrong to do so, or even that he has much stronger reason for believing its negation. For reasons we shall discuss below, the existence of epistemic akrasia is much less evident that is the existence of ordinary practical cases. Perhaps it always involves a degree of self-deception - there may be a continuum of cases differing in the degree to which the cognitive operations in question are �open to view�.

            Why do these sorts of phenomena seem problematic? In the case of self-deception, the difficulties typically stem from the fact that the �self� is involved both as deceiver and as person deceived. This seems to require a mass of beliefs, desires and projects which must both be insulated from each other but sufficiently integrated to counted as states of the same person. The challenge is to find a way of thinking about this distinctive kind of evaluation. The problems presented by akrasia are different, and depend crucially upon the fact that some of the elements involved are evaluations and commitments. They seem to be cases where I value A more highly than B, yet my choices and preferences, in situations where both A and B are possible, appear to betray a preference for B. The relative strength of the evaluations is not reflected in the strengths of my motivations when it becomes time to act. Many views of evaluative beliefs tie the relative strengths of my evaluations to the relative strengths of my desires, of my motivational states. Akrasia makes that problematic. Turning to the epistemic case: we need to understand how I can be sincere in my judgement that there is insufficient reason to believe p, indeed much better reason to believe not-p, when, faced with the need to form an opinion, I acquire the belief that p.  Accepting reasons involves making evaluations: evaluations are typically reflected in the ways in which I act and form beliefs; what can the relative strength of sincere evaluations be if it is not reflected in the choices that I make?

            Akrasia raises issues about motivation.  Generally we are motivated to act in accordance with our evaluations and to conform to our commitments unless we acquire good reason not to do so. The akratic appears to lack this motivation. Issues are thus raised about how this motivation works in the cases where we do act as, we know, we ought; and about how that motivation can be lacking in other cases. Cases of akrasia appear to challenge the suggestion that merely understanding and accepting a proposition that expresses a commitment is sufficient to motivate me to act. One merit of appeals to virtue is that they promise to explain how we are motivated to act on our evaluations and commitments.

            So when we turn to epistemic cases: we must distinguish phenomena that concern ineffective evaluations from those which concern wavering commitments; we must accommodate the possibility that the conflicting elements can somehow both be consciously present and effective at the same time or within the same process of deliberation; and we must make sense of how the ineffective evaluations and commitments can be genuine, and their avowals can be sincere. We shall also expect to face some problems about �epistemic� motivation.


3. Akratic inquiry

In this section, we consider some less problematic forms of akrasia whose content is clearly epistemic but which are distinct from the full-blooded cases we described in section (1). Deliberation and inquiry are themselves activities: we raise questions, make observations, conduct experiments, check proofs, consult colleagues, rehearse arguments to check for fallacies and so on. These activities are straightforwardly intentional. Whether my beliefs are justified will depend upon how well I carry out these activities; my view of how strongly the evidence supports some proposition will itself depend upon how carefully I have checked, double checked, consulted other people and so on. My reasons for collecting new evidence will be practical reasons, reasons for carrying out a distinctive activity. That the goal of the activity is an epistemic one does not undermine this fact.

If inquiries (and deliberations) are activities, then, like other activities, they can be carried out akratically. In that case, it is unproblematic that belief can be akratic: it can be produced or sustained by inquiry or deliberation that is akratic in the ordinary practical sense. I know it is best to make careful checks before accepting scurrilous gossip about a friend; but it does not follow that I will always do so. Aware that my intuitive probability judgments, like everyone else�s, are often extremely unreliable, I may formally decide never to trust them. However, when I respond to a run of reds by betting a large sum of money on black, I may persuade myself that this judgment is so obvious that the check is unnecessary. And, in doing this, I may be aware that I am failing to conform to important epistemic commitments. Although these are examples of akratically formed belief, they need not involve full-blooded epistemic akrasia: incontinence may prevent my even forming the conflicting judgements that full-blooded akrasia would require.

Epistemic akrasia can display both of the forms described in the last section: my commitments can fail to conform to my evaluations; and my commitments can fail to be reflected in how I conduct inquiries and deliberations. I can judge that the available evidence is insufficient to support some belief I hold, or believe that the methods used to acquire it were unreliable, yet still fail to form a resolution to examine the matter further. Perhaps I do not decide to phone the airline to check that my plane is not delayed despite my being aware that delays are common and there would be serious practical implications were I to miss a later connection. And, as in the gambling example of the last paragraph, I can fail to conform to my firm epistemic commitments in particular cases. In each case, it seems no harder to envisage each of the conflicting elements being open to consciousness than they are in prototypical practical cases.

It is easy to see how such lapses could be explained: wishful thinking may be involved, or a desire to reduce the anxiety of having to face the tasks involved in adjusting my travel plans. [2] So long as wishful thinking involves failing to conform to epistemic standards we take ourselves to be committed to, it may indeed be a form of epistemic incontinence. That may be the case here: we have failed to take account of all the evidence that was accessible or available. And this is explained by reference to the fact we end up without beliefs that we would prefer not to have � we would prefer there to be no delay and thus we prefer not to have to deal with the possibility. But this differs from the full-blooded epistemic akrasia we described above. The latter requires us to believe where we judge that belief is inappropriate. Some cases of wishful thinking may take this form, but it may also arise when we believe what we ought to judge to be ill supported evidentially and thus normatively inappropriate. Indeed some cases of wishful thinking appear to depend upon our being ignorant of (or deceived about) the warrant our belief possesses. Full-blooded epistemic akrasia, if it is to be found, should lack this dependence upon (culpable?) ignorance or (self?) deception. It seems important that at least some akrasia be distinct from self-deception. Moreover if practical akrasia normally involves a background of self-deceived belief, then putative epistemic akrasia may simply collapse into self-deception. We now turn to a consideration of how far this is the case. Can full-blooded epistemic akrasia be understood as a special form of the akrasia of inquiry?

Before doing this, however, it will be useful to make two observations about the structure of inquiry and deliberation and the norms that govern these activities.  One point emerges from our discussion of ordinary practical akrasia and has already been alluded to here: there will be an important interplay of evaluations, evaluations all things considered, general resolutions and commitments, and more specific commitments. An adequate account of epistemic activities must take note of the special kinds of evaluations and commitments that they involve. My practice may be affected by evaluations and commitments that are not, properly speaking, epistemic. I may attach great value to sharing the religious beliefs of those with whom I must live. Or I may acquire a moral commitment never to think of anyone as wholly evil. Each of these may require me to shut my eyes to the weight of evidence or to the reliability of the methods that I employ. �All things considered�, I may judge, I should hold a belief which is poorly supported by evidence or which was formed in an unreliable way. If I remain agnostic in spite of my values, or if I conclude that some individual is, indeed, truly evil, this may be a failing, but it is not an epistemic one.  On the other hand, if I succeed in retaining my faith, or if I succeed in identifying a germ of humanity in Adolf Hitler, this may fail to conform to my epistemic standards but, �all things considered� it need not be a failing. In a broad sense, the resulting belief need not be �normatively inappropriate�. Indeed we may even imagine cases of (non-epistemic) akrasia which occur because someone cannot help being too assiduous in apportioning belief to evidence: in spite of her believing that evidential considerations are not decisive in connection with religious belief, she finds herself unable to avoid a kind of scientistic atheism; or in spite of believing that one should be loyal and trusting of one�s friends, she finds herself making a dispassionate assessment of the evidence when one of them is accused of a crime.

This makes it difficult to specify exactly what would be required for a case of full-blooded epistemic akrasia. I suggest that we adopt the following, probably oversimplified, picture. We shall restrict attention to inquiries and deliberations that are governed by an overarching commitment to solving a problem or assessing a belief relying solely upon considerations that are relevant to truth.  These are activities that have goals that are fully cognitive. Then we can define epistemic akrasia as a distinctive form of irrationality which is internal to these �fully cognitive� inquiries: we employ means whose use is, we are fully aware, inconsistent with the values and commitments which apply to fully cognitive inquiries of this kind, or which emerge, rationally, within this particular inquiry. [3]

Now for the second point about the structure of inquiry, which concerns the role in it of questions and questioning.  An inquiry is an attempt to solve a problem or, most commonly, to answer a question: it succeeds when we arrive at a solution or answer which meets the commitments that govern the inquiry; in the case of fully cognitive inquiries, when we arrive at an answer which is true. If inquiry is at all reflective, then we shall face further problems, pose further questions, as it proceeds: we can question the methods employed, the methods used to evaluate those methods; we can ask whether further information is available (or even already possessed) which is relevant to the success of the inquiry; we can put questions to others in order to get their opinions of methods or in order to elicit information which they possess; and so on. The success of inquiry depends upon whether we ask the right questions. Things can go wrong if we fail to raise important and relevant issues; and it can also go wrong if we ask too many questions.  The overcautious are likely to ask too many questions; the credulous generally ask too few. Our mastery of fundamental epistemic norms is manifested in the questions we raise and, just as important, in the questions we don�t raise. The norms thus often have a negative character: they are reflected as much in facts about what does not occur to us as in the rules we formulate and reflectively follow. [4]


4. Why does epistemic akrasia seem problematic?

            There is one big difference between full-blooded akrasia and some forms of practical akrasia.  Even if forming a belief is an action, beliefs, unlike actions, are not datable events. Rather they are enduring states of people. Beliefs are more like resolutions and intentions than they are like actions: when I accept a proposition, I acquire a commitment to plan my actions and deliberations on the assumption that this proposition is true. Epistemic akrasia will thus involve conflict between commitments and evaluations or between more general and more specific commitments. For ease of reference, we shall distinguish the conflicting elements as (a) the belief and (b) the normative component. The latter may be a cognitive commitment or an evaluation. The aim of this section is to investigate why full-blooded epistemic akrasia can seem problematic, and to identify the questionable assumptions on which this appearance rests. [5]

            To avoid misunderstanding, I should distinguish the case that interests me here from some others. I am not concerned with cases where the initial belief is held unconsciously and thus does not feature in reflective deliberation. Nor am I concerned with cases where, although both components are accessible to consciousness, only one is actually �accessed�. Nor am I concerned with cases in which it is experienced as an irrationally strong tendency to believe which the agent succeeds in resisting and from which she feels constantly alienated. Nor am I concerned with a case where the mother believes that it is right, in the circumstances, to maintain her son�s innocence in the face of the evidence. A genuine case of full-blooded akrasia would have the following components: Both the belief and the normative commitment are �present� to the agent: in some manner, she is aware of each.

        She has a genuine commitment to each component.

        She is aware of the conflict between their different demands.

        She is aware that she is committed to eliminating this conflict in a different way from that which she actually employs.

It is best to work with an example. Consider a mother who believes that her son is innocent of some particularly heinous crime of which he has been accused. For epistemic akrasia to be possible, she must intend her belief to be fixed by the balance of the evidence � her inquiry is fully cognitive � and her state must have a normative component which renders her belief inconsistent with this intention. This may consist in accepting one of the following: [6]

        The evidence supporting her son�s innocence is slight.

        The evidence available to her is too limited to support a judgment either way.

        There are strong or conclusive reasons to believe in her son�s guilt.

Each component of her cognitive position is present to consciousness. She can be aware of her commitment to both without immediately losing her attachment to either, in spite of the fact that she is aware that they are epistemically inconsistent. A common worry is that this is not possible, that (for example) her confident belief in her son�s innocence will be sufficient to cast doubt on the sincerity of her endorsement of the other propositions, or her normative judgment will inevitably undermine the original belief. [7] The remainder of this section will explore some considerations that support this natural view.

So long as the activation of these conflicting beliefs are temporally distinct, there need not be a problem. Whenever she is in her son�s presence, one set of sentiments ensures that she trusts his avowals of innocence and is sceptical of the evidence that supported her earlier normative belief. But when closeted in her lawyer�s office, another set of concerns ensures that these avowals seem shallow and insincere once she confronts the weight of evidence that counts against him. Each belief fills its explanatory role while the influence of the other is temporally suppressed. While this may be irrationality, it is not full-blooded epistemic akrasia. The latter requires both beliefs to be accessible � and indeed accessed � at much the same time as part of a single process of inquiry or deliberation. The agent must be aware of the force of the one, even as she acts upon the other. [8]

We shall start with something utterly uncontroversial. We often appeal to people�s beliefs in order to explain their outward behaviour and other features of their mental lives. If I know that someone believes that the library closes at six o�clock, I can understand why he rushes towards the library when he notices that it is already five fifty. I shall also understand why he spends time wondering whether he can get from his office to the library in less than ten minutes; and also the irritation he feels when he decides that he cannot, or his surprise on seeing that it is still open at seven. I can also explain why he says �six o�clock� when asked when he thinks the library will close. Supplemented with other information about an agent�s desires and attitudes, beliefs can be used to explain how the agent behaves, the course of his reasoning or deliberation, his emotional and other affective states, and his avowals. Such explanations work only if the agent would have acted, reasoned or felt differently, had his belief been different.

            There is something unsatisfactory about the idea of a �genuine� belief or commitment that can only be manifested in an agent�s avowals, that is insulated from all the other kinds of manifestation that I mentioned. If this was possible, then full-blooded akrasia would be wholly unproblematic: someone could avow that it was epistemically wrong to believe some proposition, while belief in that very proposition would feature in the best explanations of her external behaviour, her deliberations and her feelings. I shall simply assume that if someone avows belief in a proposition when she has absolutely no inclination to use that proposition in planning her conduct or in evaluating her beliefs, and who also has no inclination to feel embarrassment, shame or surprise at this fact, then her avowal, even if sincere, is false. Thus a further necessary condition for full-blooded epistemic akrasia is that neither of the conflicting states should be insulated from exercising all of its causal propensities. If they are genuine mental states of the agent, they can contribute to explanations of her behaviour and of other features of her mental life. We can formulate this as a relatively weak principle:

P1. That X believes that p cannot be made true solely by the fact that X candidly asserts or endorses either the proposition that p or the proposition that she believes that p.

How should this lead to doubts about the possibility of full-blooded akrasia? Problems might arise if the causal explanatory propensities associated with the two conflicting states ensured that one state could exercise its causal propensities only if the other did not.  As is suggested by their role in explanation, suppose that ascriptions of commitments to people support subjunctive conditionals. It is plainly impossible that  the following two conditionals be true in a situation where the antecedent of each was satisfied, where the agent both believed that p and held normative commitment N.

Normally our first order beliefs and our beliefs about what it is rational to believe are in harmony: we believe what we think we ought to believe. And it seems plausible that it would make little sense to think of someone as a subject of beliefs if their first order beliefs and their beliefs about what it is rational to believe were never in harmony, or, indeed, if it were not generally the case that they were in harmony. In cases of full-blooded akrasia, these connections are broken. If our subject both believes in her son�s innocence and believes that the weight of reasons supports his guilt, We need to understand how the mother�s belief in her son�s innocence, and her commitment to the irrationality of such a belief, can simultaneously possess an appropriate role in cognition without giving rise to any conflicting conditionals of the kind we have just described.

Let us make one further assumption:

P2: If X is a reflective rational agent, then, in general, she can control her actions and opinions by reflecting on what there is reason to believe or do.

This requires that our reflective beliefs, including beliefs about what we believe and beliefs about what we ought to believe, stand in reasonably stable causal relations with our other beliefs and our practical and theoretical decisions. By and large, we can suppose, our beliefs and actions reflect our (assessment of our) reasons.  When we decide that one of our beliefs is irrationally held, then, ceteris paribus, we lose confidence in it. When we decide that, all things considered, we would be rational to perform some action, we are likely to do so. One aspect of this may be that our deliberations are sensitive to the demands of reason, and our actions and beliefs are sensitive to the routes taken by our reflections and deliberations.

Let us return to the example of the mother and the son and, provisionally, make a surprisingly common assumption: our beliefs (in conjunction with desires and other attitudes) are primarily manifested in our action, rather than in patterns of deliberation and feeling.  Suppose now that the mother has the goal of preserving the reputation of her family. She has sufficient reason to adopt this goal and has reasonable views about the relative priority of her different goals.  Suppose she also believes:


        If her son is innocent, the reputation of the family is best preserved by declaring her belief in his innocence and doing all she can to secure his acquittal.

        If her son is guilty, the reputation of the family is best preserved by denouncing his immorality, disinheriting him and announcing that she no longer sees him as a son of hers.

The occasion has now arisen when she must decide whether to stand by her son or denounce him. When she works out how to act on this occasion, what would we expect her to do? 

We can get a sense of paradox as follows.  If we take account of her belief his innocence and also recall that her goal is to preserve the reputation of the family, then would expect her to stand by her son. But, given that she believes that it is rational to believe in his guilt, she may also believe that it would be rational for her to denounce him. In line with P2, she may well approach her decision reflectively, wondering which action there is better reason to perform. And she may accept that if there is reason to believe in his guilt and also reason to believe that if he is guilty, he should be denounced, then she has good reason to denounce him. Thus, even if she believes in his innocence, it can still follow that if she is reflective in planning her actions, then she will act as she should if she believes in his guilt. Reflection can always short cut the expected effects of her akratic belief upon her behaviour.  It begins to look as if we have conflicting subjective conditionals of the problematic kind:

        If she believes that her son is innocent, she will defend his reputation.

        If she believes it is rational to believe that her son is guilty, she will denounce him.

If she is generally reflective, and thus generally acts as if she believed in her son�s guilt, it is hard to see what her belief in his innocence can consist in. The pattern in behaviour naturally associated with belief in guilt is present, albeit produced by the role in inference of the apparently distinct belief that there is good reason to believe in his guilt. If, on the other hand, she behaves in accordance with her belief in the son�s innocence, the belief comprising the normative component appears to be explanatory inert, in which case it is hard to see what makes it that she has this belief. We may then begin to wonder whether the first order belief and the belief about what it would be rational to believe are genuinely �distinct existences�. And in that case, epistemic akrasia seems highly problematic. [9]

            A simple example of this kind is insufficient to establish that epistemic akrasia is genuinely problematic. However it does give rise to a significant challenge. If full-blooded epistemic akrasia is possible, then we must give an account of how the belief and then normative commitment can each be operative in the agent�s cognitive life at the same time. We must be able to trace their manifestations back to the beliefs that explain them. If the mother denounces her son, what makes it the case that she does this because she thinks it rational to believe in his guilt rather than because she believes in his guilt. If she continues to defend him, what is there apart from her avowal to show that she retains her commitment to the wrongness of her belief.

            What do we learn from this example? Full-blooded epistemic akrasia requires that both the belief and the conflicting normative commitment should be operative at the same time. (The akratic smoker is aware of acting on her desire for a cigarette while fully conscious of her commitment to stopping smoking.). We need an account of how these beliefs and commitments can be �manifested� or �operative� that allows that what counts as a manifestation of the one does not count against the reality of the other. The example suggested that if we restrict our attention to manifestations in external behaviour, it might be difficult to give such an account. The public manifestations of believing that it is rational to believe that p may be no different from the public manifestations of believing that it is rational to believe that p.  Hence we shall only understand how epistemic akrasia is possible by looking at a wider range of manifestations of beliefs and commitments. At the beginning of this section, I noted that as well as contributing to the explanation of our behaviour and avowals, beliefs and commitments could be invoked to explain both the routes taken by our reasoning and deliberation and our feelings and emotions. And when we described the two ways in the which the mother could arrive at different answers to the question how she should treat her son, we paid attention to the different ways in which her deliberations could go. We shall see in the next section that attention to the process of deliberation and to feelings are both required if we are to make sense of the distinctive roles of our first order beliefs and our normative commitments. This will enable us to see how epistemic akrasia is possible.          


5. How epistemic akrasia is possible.

One lesson of the example we have been using is that we should not fix the functional role of (conscious) beliefs by reference to broad patterns in belief, desire and behaviour. How the mother will act depends upon how she reflects, upon the routes taken by her deliberations and inquiries. Our conscious beliefs provide premises for reasoning, and similar patterns of beliefs can produce different conclusions and different actions according to how they are deployed in processes of reasoning and reflection.  Not only must we attend to the ways in which beliefs guide deliberation, but (a second lesson) we must take account of how our beliefs are activated, of when we take note of them and admit them to processes of deliberation. If she is extremely reflective, then, as we have seen, her beliefs about what it is rational to believe may determine her action and her first order belief may not be �activated�; if her reflections take different directions, then the first order belief may influence her choice while the second order belief does not. Hence we must take account of how and when beliefs are activated in deliberation.

            Let us work with a someone simplified picture of how beliefs enter our deliberations. One way in which our beliefs are activated, summoned to play part in our deliberations, is through our posing questions to which these beliefs provide our answers. This parallels the way in which our beliefs can be called upon in co-operative inquiry: someone asks the question and we give our answer in a form appropriate to the current state of the conversation and inquiry. In the co-operative case, a piece of information possessed by one of the co-operating agents may fail to influence the upshot of the inquiry if the question required to elicit that piece of information is never asked, if it never becomes salient.  Co-operative inquiry can fail because one participant fails to ask the right questions of the others. And it can fail because another participant fails to point out that some relevant question has not been raised. Perhaps a parallel phenomenon is found in solitary deliberation. If the question whether p does not become salient, then my belief that p may not engage with my deliberations. In an earlier paper I expressed this point by saying that self-questioning provides a process through with which propositions can be elicited from our store of information (Hookway 1997). [10]

So our subject�s two beliefs may be elicited through her (or someone else) raising the questions:

Is it the case that p?

Is there better reason to believe that p than not-p?

It seems evident that these are different questions: we can see that in many cases, they have different correct answers. Our concern is with the possible relations between the answers I take them to have, and with the effects upon how I act of which of these questions I ask.

            The following two things seems clear:

1.      I can raise and address the second question even if I have no settled answer to the first. Indeed the second can become salient simply because I am currently agnostic about the first matter.

2.      Cases where I raise the first question can be divided into two sorts. (a) I may just candidly offer the answer I happen to have stored away. (b) I may be prompted to ask the second question as a way of checking my answer to the first one, or in order to replace my current agnosticism about the first question with a firm opinion.

The current proposal is that we can compare different routes that reflection can take by comparing the different questions that are raised (reflectively or unreflectively) as it proceeds. These questions can be used to set the targets of our deliberation, to identify the practical and theoretical problems we aim to solve; they can be raised as reflective comments upon, or challenges to, the progress of these deliberations, and they can also be used to elicit or activate beliefs or items of information that are already possessed by the inquirer. The information I possess, my current beliefs, will have an impact upon the progress of the deliberation, according to this over-simplified picture, only if a question is raised to which the belief provides my answer.

            To summarise this part of the discussion. Reflection, including both practical and theoretical deliberation, is an activity that can be controlled through the exercise of normative standards. Where it ends up will depend upon what we attend to, upon what we notice, upon which of our beliefs are activated in the course of our reflections. That a belief may fail to be salient � fail to be activated � when it is relevant to matters under discussion may be a failure of rationality: the belief may be accessible, even if it is not accessed.  One element in becoming rational is learning to � being trained to � ask the right questions. We must ask the right critical questions if we are to expose errors in our reasoning; and we must also ask the questions that lead us to access our beliefs when they are relevant to our deliberative concerns.

            This suggests one necessary condition for the intelligibility full-blooded epistemic akrasia:

It must be possible to raise the question p without at the same  time raising the question whether belief in p meets some prescribed normative standard: these are different questions.

Another necessary condition is:

It is possible for someone�s candid answers to these questions to be different.

It would be possible for this to be the case yet, either due to self-deception or to the fact that the times in which the different questions are raised are significantly different, the agent does not (perhaps even cannot) notice that this is the case. A further necessary condition for the possibility of epistemic akrasia is that this conflict is available to � perhaps �is noticed by� the agent.  The agent can be aware that her candid answers to these two questions are in tension; and this can incur within a process of deliberation for which, according to the agent, both questions are relevant.

            Although these are necessary conditions for epistemic akrasia, they are by no means sufficient. Recognizing such tensions can promote their removal: the mother may be led to reassess her candid assurance of her son�s innocence; or she may re-examine the grounds of the normative judgment confident that they will be found to contain errors; or her confidence in both judgments may be dramatically reduced. Akrasia requires that she acts on the basis of her judgment of her son�s innocence while, at the same time, continuing to endorse the normative claim that all the evidence confirms his guilt. The challenges that this presents are twofold: we must arrive at a satisfying description of the phenomena, one that makes it plausible that they should occur; and, in the light of the previous section, we should explain how both commitments continue to be operative in what is going on, and also why the agent has been guided by the one which, she holds, should not exercise authority over the other. The remainder of this section will address the first of these tasks.

            In our discussion of the example of the mother and son earlier in this section, we considered two deliberative routes, one leading through beliefs in her son�s innocence to her public defence of her honour, and the other leading through beliefs about the rationality of belief in his guilt to an act of public denunciation. In the case we are considering, her deliberation could involve rehearsing each of these argumentative routes, perhaps successively examining first one and then the other, oscillating back and forth between the two possibilities, each powerless to silence the other. How she acts may depend upon the stage of this oscillation where she decides to terminate her deliberations and act on the belief which is, at that moment, most salient. It is influenced by which questions she chooses to ask last: the irrationality displayed is thus a form of practical irrationality. Whichever wins out, she can respond emotionally. Given her initial commitment to a fully cognitive inquiry into the matter, she will be saddened if her judgments of rationality win through and she has to accept the likelihood of her son�s guilt. But if her belief in the son�s innocence wins, she will experience a variety of reactive emotions, perhaps shame or guilt. In some cases, the second argument will continue to reassert itself and she will understand how her inability to sustain her commitment to a fully cognitive inquiry has led her to act ways which she will regret. If it does not, she may continue to be aware that her confident belief may depend upon irrelevant determinants of when her deliberations should be brought to a close. Perhaps the success of the belief will cause the normative commitment to fade, but, once again, there is no reason why she cannot be aware that this is so.

            This is, of course, an oversimplified description of a particularly vivid instance of epistemic akrasia. It embodies the following elements. First, we can allow that each commitment may be manifested unproblematically in other deliberations in which its rival has no role. Second victory in the deliberation we are concerned with here results from processes that are in tension with the mother�s commitment to a fully cognitive inquiry. Third the losing commitment need not simply disappear but can continue to be manifested in the reactive emotions that accompany the mother�s reflections upon her beliefs and her actions. And fourth, it acknowledges that we can make sense of these phenomena only by attending to the conduct of activities such as deliberation and inquiry. [11]


6. Conflicts of Intuition: some more examples of akrasia

An important step in the argument of the last section was the recognition that the reactive emotions of the mother can attest to the presence in her deliberations of commitments which were somehow silenced or defeated when she decided how to act. This is not the only way in which affective states have a role in the regulation of deliberations, in which they can they can register the presence of standards of evaluation which are not consciously articulated or acknowledged. In an earlier paper, I argued that we can make sense of the epistemic role of states of doubt when we notice that they generally involve a motivational component in the form of an anxiety about the agent�s grasp of the proposition doubted (Hookway 1998). Something similar is likely to be involved in the case that we described. One piece of evidence that the defeated normative standards are operative at the moment of the mother�s decision is that she will feel anxiety as she acts on her belief in her son�s innocence. This affective acknowledgements of the completing claim betrays sensitivity to the irrationality of what she is doing, a sense that her actions may go awry through her own fault. [12] Describing it as �anxiety� explains its �motivational� role: it can stimulate us to reflection, or to abandoning inferences and judgments that are contrary to such intuitive responses. Thus epistemic anxiety can be an immediate unreflective acknowledgement that the deliberator is in danger of going wrong.

            Such anxieties can, of course, be weak and they are not always rational. But it is plausible that what philosophers often describe as �intuitions�, - the intuition that an argument is not a good one, that a concept does not apply to a particular situation, that a sentence is syntactically out of order � are immediate affective embodiments of norms that we follows but which we cannot explicitly formulate. We express anxiety about accepting these claims without fully knowing why or how. Rationality involves trusting, or listening to, our �intuitions�. A distinctive form of akrasia � quite a full-blooded one � can come from the motivated refusal to listen to our intuitions. The intuition provides our only conscious access to a normative commitment by signalling when we are contravening it. Thus we can be aware that our beliefs or inquiries are in conflict with commitments that we cannot formulate or acknowledge. This can illuminate some familiar phenomena of irrationality.

            Familiar psychological studies of reasoning suggest that humans are naturally inclined to accept a variety of inferences that are evidently irrational: familiar examples concern probabilities or inferences that turn on the understanding of conditionals.  Getting it right, in such cases, can be hard work, requiring us resist kinds of inference that are ingrained by habit or built into our cognitive architecture. Those of us who have read the textbooks know that we are likely to go wrong in these cases, and struggle to do better. This does not prevent inferences that we know intellectually to be flawed from feeling compelling. The temptations to think that a run of reds raises the probability of the next spin being red, or to conclude that Linda is more likely to be a feminist bank teller than to be a bank teller, do not simply fade away. Experience may warn us of the likelihood of error in such cases: although we are naturally drawn into the inference, we encounter the inchoate warning that something is wrong and we should attend more carefully to what we are doing. There are two intuitions here: that the inference is a good one; and that such intuitions are not always to be trusted. If we are committed to fully cognitive deliberation, we know which one should be trusted. Rationality requires us to manage these intuitions, to understand their sources and form our beliefs as we should. In each case, we can be more strongly motivated to go the wrong way, to go with an argument that seems intuitively right when we intuitively know that this intuition is untrustworthy. The normative component of this form of akrasia is thus an affective commitment to normative requirements rather than a full belief. But this does not present regret, shame and anxiety being present just as in the case described in the last section.


7. Epistemic virtues

In section one, I suggested that our examination of these phenomena of epistemic akrasia lends support to the view that effective cognition depended upon possessions of states such as virtues, enduring states of character with a role in regulating inquiries and ensuring their success. The examples we have considered have illustrated some ways in which cognition can go wrong; and effective responsible inquiry depends upon the mastery of epistemic norms which can prevent it going wrong in these ways. The discussion has relied upon a particular picture of epistemic evaluation. Its primary role lies in the regulation of activities of inquiry and deliberation. These activities require work: we often have to make an effort to carry them out effectively. We actively address problems and try to formulate and answer questions; we compare our different beliefs and hypotheses, trying to bring to bear opinions about the weight of evidence upon propositions we accept or those about which we are agnostic.  Success will depend upon our doing the right amount of work: addressing issues that need to be addressed without wasting time and energy over irrelevant matters. It also depends upon our efforts having their expected effects. If we spend time collecting and assessing the evidence that supports some proposition, then our results should be reflected in the degree of support we give the proposition. It is easy to see how indolence, partiality, prejudice and a range of other interfering factors can prevent our making our efforts appropriate or prevent them from being effective. Our epistemic values should equip us to carry out a managerial task: employing our epistemic resources and exercising our epistemic efforts in ways that contribute to enable us to solve problems and reach true answers to significant questions. 

The present paper has identified some of the tasks that effective regulation of deliberation faces, some of the potential conflicts we must be able to deal with. We shall now list and elaborate some of these.

        When normative assessments of evidence and argument conflict with confidently held beliefs, a judgment must be made concerning which should have authority over the other.

        Our commitment to fully cognitive inquiry must contend with (often unconscious) inclinations to favour one solution over another by focusing on questions that send our inquiries down potentially distorting routes. This can lead us to abandon the commitment or it can prevent our seeing that it is not fully effective.

        We must be able to weigh the force of apparently incommensurable evaluations, for example: formulated commitments against contrary �intuitive� anxieties.

        We must be able to weigh the force of conflicting evaluative intuitions. In these cases we may not be reflectively aware of the normative standards that are reflected in the intuitions. We may not even be evident whether they really conflict.

        We must be able to deal with the fact that our reflective normative judgements may simply be powerless to adjust our beliefs as rationality requires. This may be due to acquired habits, to properties of our cognitive architecture, to brain damage, to laziness and inattention, to emotional attachments and to a range of other causes.


These kinds of phenomena draw attention to two respects in which our deliberations are not under our control. First: much depends upon whether we raise the right reflective questions. Our mastery of norms is reflected in the questions we don�t raise as well as in the questions we do raise: we are sensitive to irrelevance as well as to relevance. As we have seen, such norms have a negative character: it would be hopeless if we have to consider every possible question and explicitly apply rules in order to decide that it was irrelevant to our deliberations. Second: we cannot control whether the answers we reach to our reflective critical questions will have their intended to effect. The recognition that some belief is poorly grounded may simply be impotent to shake confidence in the belief unless appropriate mechanisms are in place. And the factors that shape our choice of questions, like the factors that influence the effects of normative judgments upon their doxastic objects, are not open to introspection. Much of the time, we don�t know what is going on or why.

            If this is right, then effective epistemic agency depends upon the possession of confidence in our intuitive judgments about which questions should be asked and about the relative weights of apparently incommensurable evaluations. Unless we possess this confidence, we will constantly face questions of how to proceed that we are powerless to answer. This is because intuitions (for example momentary states of anxiety) often provide our best access to normative standards which are ours, but which we are unable to articulate. Unless we can trust these intuitions � and our intuitions about when they can be trusted � those normative standards cannot influence our cognitive activities. Effective epistemic agency also requires that this confidence is not misplaced. Trusting our intuitive judgments ensures that we do ask the right questions and that our comparisons of different evaluations do not impede our cognitive projects. The confidence provides the internalist dimension, and its not being misplaced constitutes an externalist dimension, of epistemic evaluation.

            If we possess this confidence, and it is not misplaced, then our deliberations and inquiries will be broadly �continent�. It is natural to think of this as the possession of a virtue: a state of character which ensures that we take heed of acknowledged reasons and maintain our rational commitments. It may be best to think of this as based upon a cluster of capacities and skills. Some may be innate, others the result of training and education, yet others the product of conscious thought and planning. They are unified by their common role in the evaluative practice that regulates inquiry and deliberation.  I am not committed to claiming that there is a single mechanism that does the whole job; nor that if any members of the cluster are present, then all must be; nor that they need be unified by a common location in the accounts of the mind produced by cognitive psychologists. [13]

            If �continence� is a virtue then, like courage, it is what has been called an executive virtue (Pears 1978, and see Wiggins 1987) or a �virtue of will power� (Roberts, 1984: 230). Such virtues are  exercised for the sake of some further end: success in inquiry or victory in battle. As Roberts puts it, they are concerned with �self-management� or self-mastery, with the ways in which we cope with the different motivational pressures we face and plan our actions in the light of these.  Thus courage would be an example: it is a virtue that can be deployed in the service of any of a variety of ethical (and epistemic) outlooks. While I may be benevolent or temperate for its own sake, I cannot act merely for the sake of courage. The function of such virtues is to enable us to overcome the obstacles to achieving the further goal � to escape from the inclination to avoid the risk of being maimed or killed in battle, or to escape the risks of lapsing into wishful thinking or procrastination and thus thwarting our deliberative aims in the case of continence. Irrationality, in general, is a threat to our cognitive aims, and the cluster of dispositions and capacities provide the knowledge and motivation to do the work that is required to avoid it having this effect. It is plausible to describe continence as a vehicle of self-control. It helps to ensure that my commitments and values shape the development of my opinions. And it ensures that my commitments endure unless it becomes rational to reassess them.

A true Aristotelian would be uneasy about linking virtue to self-control. The latter appears to involve an ability to recognize contrary impulses and inclinations and the power to resist them, to face them down. For Aristotle, this is a response to a difficulty that the truly virtuous person would not face. If we are truly virtuous, then contrary impulses and inclinations do not interfere with the comfortably virtuous life: continence is a way to cope with the fact that we are not ideally virtuous. 

However it is still appropriate to describe these states as virtues. Continence is an enduring state of a person which enable him or her to deliberate well. It does this by exploiting a body of normative standards and capacities for evaluation. The standards of evaluation employed are generally not (and probably cannot be) formulated as precise formal rules.  In many cases, application of rules to particular cases will involve weighing apparently incommensurable values and arriving at an intuitive judgment the bases of which are not fully explicit. Finally a state such as continence has a role in motivation: it enables the agent to inquire well, to adjust beliefs automatically in the light of normative considerations when she judges that it is appropriate to do so. These, I take it, are all marks of a state being a virtue.

            A brief illustration of the point about judgment may help here. In listing the capacities we require for effective inquiry and deliberation, we mentioned the ability to weigh apparently incommensurable epistemic values. Suppose it is one of our epistemic duties to subject testimony to suitable scrutiny before accepting it; we should avoid gullibility so far as is possible. This requirement has a prima facie character: we should respect it so long as more pressing epistemic requirements do not conflict with it. Where obtaining false testimony carries few risks; or where it is important that our co-operative investigation advances quickly; or where the source of testimony is a trusted colleague and her testimony deals with a topic where there has been no reason to doubt her reliability: in all these cases, it would be best if the question of reliability was not made the subject of any reflection or investigation. Deciding whether this is case where the �duty� calls for epistemic action involves comparing the weight of different epistemic desiderata where we lack any overarching rule that tells us how to decide. When we bear in mind the complexity of any particular case, and the need to apply our duties to particular cases, it is easy to see that we rely upon experience and judgment. Even if we decide that this is a case where the informant�s reliability should be assessed, there will be many ways of conforming to this duty. We might recall her past record, collect the opinions of others, test her in areas where we already possess reliable information and so on. How scrupulous we should be is a matter of judgement, again a way of weighing a variety of epistemic and non-epistemic goods. Suppose we then obtain different sorts of evidence that have a bearing on her reliability, some positive, some negative. Once again we must weigh them. And once again we are unlikely to have formal rules that can guide us in doing so. Rules and duties must be applied to complex cases in the course of well regulated inquiries and deliberations. And this cannot be governed by further explicit rules, on pain of a regress. (cf Larmore, 1987, chapter one)

It is easy then to see how our examples of epistemic akrasia involve failures of virtue. In some cases, the agent�s judgment is deficient: she attaches too little weight to normative considerations that she is committed to taking more seriously than she does. In others, the failures are ones of motivation: her anxiety as she acts on her belief in her son�s innocence is a sign that a normative a commitment that she endorsed, and that she is still committed to giving authority over her beliefs, has failed to have its due effect upon her current cognitive state. [14]



Broadie, Sarah. 1991. Ethics with Aristotle. Oxford: Oxford University Press.

Davidson, Donald. 1980. Essays on Actions and Events. Oxford: Oxford University Press.

Davidson Donald. 1985. �Deception and Division�, Actions and Events, ed E. Lepore and B. McLaughlin, Oxford: Blackwell: 138-48.

De Sousa, Ronald. 1987. The Rationality of Emotion. Cambridge: MIT Press.

Graham, Keith, 1974. �The Limits of Irrationality�, Inquiry, vol. 17: 315-26.

Harman, Gilbert. 1999. �Moral Philosophy Meets Social Psychology: Virtue Ethics and the Fundamental Attribution Error�, Proceedings of the Aristotelian Society, XCIX: 315-31.

Heil, John. 1984. �Doxastic Incontinence�, Mind, XCIII: 56-70.

Holton, Richard. 1999. �Intention and Weakness of Will� Journal of Philosophy, vol. XCIV: 241-62.

Hookway, Christopher, 1997. �Analyticity, linguistic rules and epistemic evaluation� in J. Preston (ed.) Thought and Language, Cambridge: Cambridge University Press. 197-218.  (This volume was also published as Philosophy, supplement 42.)

   ------- 1998 �Doubt: affective states and the regulation of inquiry�, Cheryl Misak (ed.) Pragmatism, Canadian Journal of Philosophy, Supplementary Volume 24: 203-26. Reprinted as chapter ten of Hookway (2000).

   ------ 1999 �Epistemic Norms and Theoretical Deliberation�, in Ratio, XII: 380-97. This volume is also published as Jonathan Dancy (ed) Normativity. Oxford: Blackwell, 2000.

   ----- 2000. Truth, Rationality and Pragmatism: Themes from Peirce. Oxford: Oxford University Press.

Johnston, Mark. 1988. �Self-Deception and the Nature of Mind�, in Perspectives on Self-Deception, (eds) B. McLaughlin and A. Rorty, Berkeley and Los Angeles: University of California Press: 63-92.

Larmore, Charles. 1987. Patterns of Moral Complexity. Cambridge: Cambridge University Press.

Lear, Jonathan. 1988. Aristotle: The Desire to Understand. Cambridge: Cambridge University Press.

Pears, David. 1978. �Aristotle�s Analysis of Courage�. Midwest Studies in Philosophy, volume 3.

Roberts, Robert. 1984. �Will Power and the Virtues�,  Philosophical Review, XCIII: 227-247.

Rorty, Amelie. 1983. �Akratic Believers�. American Philosophical Quarterly, 20: 175-8.

Scanlon, T. M. 1999. What We Owe Each Other. Cambridge: Harvard University Press.

Velleman, David.1989. Practical Reflection. Cambridge: Harvard University Press.

Wiggins, David. 1987. Needs, Values, Truth. Oxford: Blackwell.

Williams, B. 1973. Problems of the Self. Cambridge: Cambridge University Press.

Williams, S. 1990. �Belief,  Desire and the Praxis of Reasoning�. Proceedings of the Aristotelian Society, XC: 119-42.

Zagzebski, L. 1996.  Virtues of the Mind. Cambridge: Cambridge University Press.



[1] This argument could provide part of a defence of virtue epistemology. It would need to be supplemented, inter alia, by a demonstration that the standards that regulate inquiries and deliberations are the most fundamental epistemic norms. This contrasts with the more common epistemological view that the evaluation of states such as beliefs as justified or as knowledge is the core of epistemology. The second stage would consist in showing either (a) that the latter concepts can be explained by reference to notions of good inquiry or theoretical deliberation or (b) that the concepts of knowledge and justified belief are less central, less important for understanding the pursuit of truth, than is often supposed. The current paper is not concerned with this further step. Even without this second stage, the argument should establish that epistemic virtues are more than just useful bodies of habits that enable us to achieve more automatically and easily what would otherwise place great demands on our attention and our reflective powers. Without such virtues, I shall argue, rational deliberation and inquiry would be impossible.

[2] Examples of this kind illustrate what Amelie Rorty (1983) has called �akrasia of inquiry�. Someone may succumb to ways of thinking she knows to be intellectually flawed, for example being unfair to the ideas of others or accepting something on the authority of someone she knows not to be trusted. (Zagzebski 1996: 154-5). Section II of Rorty�s paper provides a useful taxonomy of kinds of �doxastic akrasia�: most of them seem to fit the pattern just described.  What I have called full-blooded epistemic akrasia is discussed more fully by Heil (1984).

[3] This strategy can be generalised to a wider range of cases.  Even if an inquiry has goals that are not fully cognitive, it may comprise sub-inquiries which are taken to be fully cognitive. Even someone who wishes to form religious beliefs on grounds that are not fully cognitive can display epistemic akrasia when he investigates just what the prevailing local beliefs are.

[4] This topic is explored more fully in Hookway (1999).

[5] One of the earliest attempts to undermine the common assumption that epistemic akrasia is impossible is in Graham (1974).

[6] John Heil expresses this by saying that the akratic believes two propositions that are �epistemically incompatible�: accepting one of them would constitute a reason for not accepting the other. (1984: 63)

[7] This suspicion is clearly expressed by Lear�s comment that the incontinent is �a stranger to himself� and that �it is in his actions, not in his assertions, that he may discover who he is� (1988:186).

[8] This diagnosis is similar to that of T. M. Scanlon who holds that that the source of doubt about the possibility of epistemic akrasia �is the idea that judging P to be supported by the best evidence is so immediately connected with believing P that there is no room for slippage of the kind that can occur between judgment and action.� (1999: 35) Scanlon�s rebuttal of this claim identifies cases where someone is guided in believing a proposition by a consideration which, at the time, seems to be a good reason for belief although she is aware that other considerations establish that it is not a good reason at all. (ibid: 36).

[9] These considerations seem to support Linda Zagzebski�s contention that �Intellectual akrasia involves self-deception more than does moral akrasia because there is probably a stronger link between believing and believing justified than doing and believing right.� (1996: 154)

[10] We should guard against a misunderstanding here. I am not claiming that in general beliefs influence behaviour only when they are explicitly activated by self-directed questions.  Dispositional beliefs can ground habits of inference or influence the ways in which we describe our experience of find things salience. The claim is: some beliefs, on some occasions, play a role in the formation of behaviour through being elicited to serve as premises in conscious inference.  It is only beliefs that are can be invoked in this way that can be avowed and made a matter of conscious reflection. Full-blooded akrasia can occur when beliefs of this kind coincide with co-existing beliefs of the same kind that express a negative evaluation of their status. Phenomena that are close cousins of such akrasia may occur when just one of the beliefs in which question is available for conscious elicitation.

[11] Some philosophers interested in these phenomena think we should divide the self or introduce homunculi. That does not seem to be required under the current proposal. Rather: it is unproblematic that the questions to which these different propositions are answers need not be asked together if they are asked at all. In which case it is unproblematic that the different �beliefs� can be activated at different times and in different ways by asking appropriate questions.

[12] As Mark Johnston has noted, some forms of irrationality (for example wishful thinking and self-deception) may stem from a reluctance to face up to anxieties about ways in which more careful investigation may thwart our aims or force us to confront unpleasant truths. He sees self-deception as often involving a kind of intellectual cowardice. It is possible that we should view continence as a special application of courage to the case of inquiry and deliberation � or see both as different forms of a single more abstract virtue. But I cannot pursue that issue here.

[13] Pace Harman 1999.

[14] In producing the final version of this paper, I have greatly benefited from the discussion at the conference in Santa Barbara. Wayne Riggs was commentator on that occasion, and his reactions, both at the conference and subsequently, have led to many improvements. I am also grateful to Lucy Burroughs, Tobies Grimaltos, Stephen Makin, David Owens, Jenny Saul, and Leif Wenar, for very helpful discussions of drafts of the paper; and to audiences at talks on this material delivered at the Universities of Bristol and Manchester.