Chapter 14

Epistemic Luck in Light of the Virtues

Guy Axtell

The problem of restoring integration and cooperation between man’s beliefs about the world in which he lives and his beliefs about the values and purposes that should direct his conduct is the deepest problem of modern life. It is the problem of any philosophy that is not isolated from that life.—John Dewey

I. Luck, Control, and Virtue Theory

In its broadest context, as part of contemporary virtue theory, virtue epistemology can be understood as one side of an integrated account of the human being as both a knower and a valuer. What attracts me to it, personally, is my sense of the fragmented state of reason in contemporary western culture, a condition that is to some extent an inheritance of modern philosophy. Virtue theory promises a response to what Dewey, quoted above from The Reconstruction of Philosophy, presents as a central problem of modern life and even as the theoretical chasm for philosophy to bridge: the chasm between our beliefs about the world and our beliefs about the values and aims that should direct our actions, both individually and collectively. I use these metaphors of bridge and chasm in connection with Dewey’s passage, because this is an essay about bridges: bridges, firstly, between epistemology and ethics, and secondly, between virtue epistemologists themselves, since they, like others, invariably bring with them to the discussion of knowledge their own divergent interests in explanation.

Our approach will be through an analysis of the various possible meanings of “epistemic luck,” and their impact on epistemology. Luck presents a complex of problems that impact both ethics and epistemology, problems that must be balanced with and related to our conceptions of responsible agency. The presence of luck in our cognitive as in our moral lives shows that the quality of our intellectual character may not be entirely up to us as individuals, and that our motivation and even our ability to desire the truth, like our moral goodness, is fragile. But the fragility of our character makes it a more, not a less important topic for philosophical inquiry. As Claudia Card and Martha Nussbaum have both pointed out, the impact of luck on our lives can also add depth to our understanding of responsibility and increase our sense of the worth of the virtues.[1]

Luck also affects our understanding of the reliability of our belief-forming cognitive processes (hereafter BCP’s). The reliability of these processes is in one sense radically subject to luck, to a ‘natural lottery’ as some would describe it, or to the control not of ourselves, but of nature. To the extent that our intellectual habits and dispositions can be influenced by choices under our control, however, their reliability might be taken as an achievement. But to what extent is this? There is keen debate among epistemologists about whether our intellectual dispositions—those that help shape our BCP’s—are less subject to voluntary control than are the moral dispositions that shape our actions. Views on this matter are bound to impact our conceptions of both ethical and epistemic agency. It is a commonplace to hold that the more externalist an epistemologist’s orientation, the more tolerance they have for luck in our cognitive lives.

Here is a first indication of differences among virtue epistemologists: those inclined towards reliabilist views tend on the whole to be more skeptical concerning issues of responsibility for one’s intellectual character than those inclined towards what has recently come to be called “responsibilist” views.[2] In a recent paper “Moral and Epistemic Virtue,” Julia Driver for instance, who takes a stance she calls objective consequentialism [3] allied with “the externalist extreme” of a thoroughgoing reliabilism, writes that “Luck is a fact of life, and it may be a form of moral and intellectual virtue to humbly accept that there will be such limits to success no matter how well the agent is justified.” Those who think epistemic agents have little control, or only very indirect control over their intellectual dispositions and habits, lean towards reliabilism even if, as Linda Zagzebski has suggested, there may be few necessary connections to be found between externality and lack of control, or between internality (awareness) and control.[4]

The issue of responsibility for character (which would appear to follow from our capacity to control or affect our own traits) is a difficult one, and all the more so when we extend the term “character” to include intellectual habits and dispositions in addition to moral ones. The analogies between ethics and epistemology are undoubtedly on stronger ground when one compares the degree of control we enjoy over ethical and intellectual character traits, than when one attempts to compare our control over our actions with our control over our beliefs. But terms of character such as “disposition” and “habit” are used to cover a wide range of attributes not well demarcated by their openness to conscious or to voluntary control. Moreover, epistemologists often invoke different vocabularies when issues arise about control. Some talk without worry about a “weak doxastic voluntarism,” from which control and hence responsibility flows, while others talk as if sure that epistemic agents are about equally responsible or non-responsible for each of their beliefs (certainly not something we hear said about each of our actions!). Both vocabularies can contribute to conflation of the descriptive and evaluative tasks of epistemology. The former is odd if it assumes one needs to support any kind of direct doxastic voluntarism in order to speak about control over traits of intellectual character. The latter is odd in light of the generally-accepted points that “S believes p” is attributable an agent S (with the possible exception of perceptual beliefs) only on the basis of a prior assumption that the agent S does (or would) accept the proposition that p, and that the “justification” for acts of acceptance is not an all or nothing affair, but rather comes in degrees.

Hilary Kornblith avoids this conflation between the normative and the descriptive by distinguishing between “actions” and “processes,” that is, between actions that we ought to perform in order to make the processes by which we arrive at our beliefs more conducive to truth, from those processes (BCP’s) themselves. A related suggestion, to which I will try to adhere, is that in order to demarcate the proper application of normative concepts of agency like “rational,” “responsible,” “praiseworthy” etc., we should distinction between the acceptance of a proposition, and the state of belief. Whichever distinction is preferred, formulating criteria for evaluating an agent’s acceptance of a proposition involves these normative concepts of rationality and responsibility.

n general I find differences over the issue of control over intellectual habits and disposition to run less deep among virtue epistemologists than among epistemologists at large. Part of the reason for this may be that virtue epistemologists share a broad range of assumptions that bring them together with interest in issues—both the descriptive and evaluative—affecting our understanding of epistemic agency. Moreover, one opinion virtue epistemologists widely hold is the present need for a “dual component” conception of knowledge, sometimes also called a “mixed” externalist account of knowledge.[5] This is one that integrates constraints on an agent’s faculty reliability with constraints on the agent’s responsibility in gathering and processing evidence. Kornblith divides the latter into “internal coherence” constraints (characterized by desire for epistemic integration) and “action-theoretic” constraints (characterized in terms of desire for and effort in attaining the truth). To briefly cite three further instances, Ernest Sosa writes that in his virtue perspectivism, a proposition is evident or known (from the K point of view) to a subject “only if both he is rationally justified in believing it and is in a position to know (from the K point of view) whether it is true.”[6] In her Virtues of the Mind (1996), Linda Zagzebski, uses the term “dual-component” to describe her mixed account of believing from virtue, which includes both a motivational and a success component. And John Greco says that his account is intended to satisfy both the “subjective justification” and the “no accident” conditions on knowledge by drawing on the resources of virtue theory.

Mixed accounts I think present distinct advantages over their competitors, because they insist on the complementarity of two concerns that have driven recent developments in epistemology: what it means to have beliefs formed by reliable cognitive processes, and what it means for agents to be responsible and well-motivated in their acquired cognitive habits and dispositions. But differences persist among virtue epistemologists as elsewhere in philosophy, and can at times be quite pronounced. Ernest Sosa, John Greco and Alvin Goldman describe their virtue epistemologies as specific forms of reliabilism. Lorraine Code formulated the distinction and contrast between reliabilist and responsibilist virtue epistemology in her Epistemic Responsibility (1987), arguing that “the concept ‘responsibility’ can allow emphasis upon the active nature of knowers/believers, whereas the concept ‘reliability’ cannot.”[7] I will draw a distinction as I have elsewhere, between those inclined towards virtue reliabilism and those inclined towards virtue responsibilism.[8] The differences that these labels identify should not be exaggerated, since these two hardly represent anything like the extremes of externalism and internalism of the recent past in epistemology. Indeed the issues of first-person constraints such as “internal access” to the grounds of one’s belief seems not to be the dividing line between virtue reliabilist and responsibilist accounts of justification. Neither thinks of the nature of justification as turning on that issue, so closely associated with the received definitions of internalism and externalism.[9] But I will take the descriptions of an author as a virtue responsibilist or virtue reliabilist to have a useful application in pinpointing differences of interest and attitude, and to often be applicable to particular authors even where a “mixed” account may be explicitly advocated.

I hope to show how an analysis of epistemic luck can lend substantial support to mixed externalist epistemologies. But the analysis of luck that will be offered is unique, and first I want to reflect upon how problems that surround epistemic luck have in fact served to divide reliabilist and responsibilist accounts in the recent past. If we concern ourselves first with the issue of the role of the will and responsibility in propositional acceptance, we find that it is the responsibilists who are most keen to explain the ground for evaluative judgments of epistemic agents: They remind us, for instance, that the Aristotelian account (which allows attributions of responsibility for intellectual habits and dispositions) is not committed to a direct voluntarism (Hookway); that only weak voluntariness is at issue in responsibility (Montmarquet)[10]; and that that there is a range of voluntariness in belief as in action—where “accident and intentional action are two ends of a spectrum of conscious control” (Zagzebski).[11]

Reliabilists for their own part are likely to point out, as does Driver, that “there is a built-in responsibility for action” that does not seem to be paralleled in epistemology, where she believes we are “more forgiving” of defects. But this contrasting emphasis is by no means to reject responsibility for character tout court, nor by itself does it signal any deep division between our two sub-groups within virtue epistemology. Underlying the mixed account of knowledge is what Thomas Nagel would term a “compatibilist account” of cognitive freedom and causality. This is the epistemic analogue of the view in ethics that one may be responsible for what one does, even if what one does depends in important ways on factors not within one’s direct control. In epistemology as in ethics, Nagel tells us, compatabilism “would leave room for the ordinary conditions of responsibility—the absence of coercion, ignorance, or involuntary [actions]” without excluding from the analysis all influence of factors that are a matter of luck.

Let me clarify this last point. In his influential paper “Moral Luck,” Nagel frames the relevant epistemic analogue to his compatibilist position in ethics:

The corresponding position in epistemology would be that knowledge consists of true beliefs formed in certain ways, and that it does not require all aspects of the process to be under the knower’s control, actually or potentially. Both the correctness of these beliefs and the process by which they are arrived at would therefore be importantly subject to luck. The Nobel Prize is not awarded to people who turn out to be wrong, no matter how brilliant their reasoning.[12]

Such a compatibilist stance in epistemology, to continue the analogy, is simply one that would leave room for the ordinary conditions of epistemic responsibility—primarily the absence of coercion—without excluding all influence of factors that are a matter of epistemic luck.

Yet this shared compatabilism aside, we can certainly identify issues where the reliabilist and the responsibilist are more sharply divided. Let us examine these. One such issue is how far the analogy between ethical and epistemic luck can or should be pushed. Zagzebski has commented that the recent interest in epistemic luck “makes the attempt to model epistemic evaluation on moral evaluation easier to do,” because it indicates that “it is much too facile to distinguish evaluation in the two areas on the ground that we control the one but not the other.”[13] Our intellectual habits and dispositions, she holds, are generally not less subject to voluntary control than are their ethical counterparts, although a range of voluntariness must be acknowledged in regard to both. Not everyone would share these responsibilist views, and I use them to illustrate that it is much easier to show how the theoretical differences between virtue reliabilists and responsibilists are manifested than it is to locate and sort out the interests in explanation that drive their divergent epistemological accounts.

In the most widely-discussed example, overt differences are manifested in sharply divergent definitions of the intellectual virtues themselves. Reliabilists define them by their conduciveness to the production of truth beliefs, allowing genetically-endowed faculties and powers(such as those of perception, and memory) to be counted among the virtues; responsibilists tend to take them more restrictively in terms of acquired habits and dispositions—traits internal to agency that are the proper object of praise and blame. Responsibilists may also reject the attempt to define the virtues consequentially by what they do, in favor of viewing them as conceptually prior to and as partly constitutive of the epistemic goal itself.[14]

The differences between reliabilism and responsibilism manifest in another way of which we should take notice: in debate over the possibility and desirability of a unified account of ethical and epistemic virtue. Julia Driver usefully distinguishes between the claim that an account of the virtues is unified, and the claim that the virtues themselves are unified. What sense should we give to the latter, more contentious claim? If it is understood as the robust classic Socratic/Aristotelian claim that a person cannot have the intellectual virtues without the moral virtues (and conversely), then it probably has few supporters.[15] But there are various senses of a unified account of the virtues that many responsibilists, and only responsibilists, want to defend. The strongest of these may be the subsumption thesis that Zagzebski argues for in Virtues of the Mind (1996), where she “subsumes the intellectual virtues under the general category of the moral virtues, or aretai ethikai, roughly as Aristotle understands the latter.”[16]

Less strong, James Montmarquet provides an account of the relationship between the ethical and epistemic virtues that is critical, on the one hand, of Zagzebski’s strong “assimilationist position,” and on the other, of the strong “externalist anti-assimilationist position” which Driver may be thought to hold. The model Montmarquet develops for the relationship between the virtues does not arrange them by content or domain, but rather by the susceptibility of each disposition to direct control. Central to his understanding of control is the agent’s efficacy in exemplifying a particular virtue at will, i.e. where “trying” to exemplify the virtue is sufficient for exemplifying it. This innovative approach issues in a model reflecting a three-fold distinction and “a hierarchy of evaluatively relevant qualities of persons, which largely cuts across the supposed epistemic/moral divide”:

At the lowest level will be those ‘externalist’ traits subject only to our indirect control. At the middle level will be the ‘internalist’ epistemic virtues and perhaps certain broadly moral virtues. At the summit, will be certain narrowly moral traits” [picked out by the fact that we are able both to exemplify them and to exemplify their contraries at will].[17]

A unified account of the virtues may be associated with a still-weaker thesis, but one that like the preceding claims also invites “cutting across” some of the disciplinary boundaries between ethics and epistemology. This thesis is one that associates virtue epistemology with the development of a general theory of value. Montmarquet and Zagzebski have both supported a sense of this thesis in their book-length treatments of virtue epistemology. I have as well, in previous papers that connect the appeal of virtue epistemology with a broader metaphilosophical re-orientation in philosophy. Responsibilists, then, side with Zagzebski’s general claim that “the unification of moral and epistemic evaluation is a welcome advantage.” They also tend to resist the drift of reliabilism towards a view of epistemology as a disparate field that studies various but unrelated ways that beliefs can be justified, and thereby side with Paul Bloomfield’s general claim that “it would be preferable to find ‘justification’ to be helpfully univocal.”[18] By contrast, those of reliabilist orientation tend to be more sanguine about the prospects of unification in any of the senses we have discussed, and so we find Driver arguing “that a unified account of virtue is neither doable nor desirable.”

Let me give just one substantial example of how this debate plays out, using Zagzebski and Driver to exemplify sharply divided responsibilist and reliabilist viewpoints. Zagzebski rejects any sharp distinction between moral and other sorts of evaluation, which she thinks is a residue of the faulty Kantian view that the good will (seen as radically free in its noumenal existence) is the only proper object of moral evaluation. But her reliabilist critic may perceive her (in a sense analogous to Bernard Williams’ critical perspective in Moral Luck (1981)), as actually involved in a neo-Kantian attempt to immunize epistemology from luck by exaggerating the freedom and control of the agent (in this case the epistemic agent). While Zagzebski like Williams criticizes the Kantian attempt to ground moral evaluation in something that is luck-free, she also takes it as a definite advantage of an epistemic theory that it “reduces the component of luck.” Taking the reduction of the impact of luck as an advantage may seem unobjectionable, but in the context of Zagzebski’s approach it reflects her attempt to develop a “motivation-based virtue theory” that places goodness within motivational structure and identifies certain motives as intrinsically good. Seeing Zagzebski’s responsibilist account in light of this project makes it difficult to separate Driver’s complaint that a motivation-based virtue theory “doesn’t provide a criterion,” and leaves ‘mysterious’ what makes a motive good or bad, from Williams’ notable objection addressed to the Kantian, that “the dispositions of morality, however far back they are placed in the direction of motive and intention, are as ‘conditioned’ as anything else.”[19]

To summarize, we have seen that virtue epistemologists often explicitly disagree over the understanding and identification of the virtues, over the issues of responsibility for character, over the strength of the analogy between ethical and epistemic evaluation, over the prospects for a unified account of the virtues, and over the possibility of a univocal sense of epistemic justification. But if these issues, as we said, merely mark spots where the pot boils over—where explicit theoretical differences persist and main lines of division are manifested among virtue epistemologists—then we should now be enticed to look further, and to seek the source of these differences in their divergent interests in explanation. The problem of epistemic luck is an especially promising area to explore in seeking these deeper sources.

II. Forging Bridges: Riggs’ Conjunctive Approach

Is knowledge compatible with epistemic luck? An initial step in answering this fundamental query is to distinguish questions about the extent of epistemic luck from questions about the kinds of epistemic luck thought to have an impact on human cognition. Luck in all its senses appears to involve discontinuities or lack of control either over a process or over its outcome. And as Nagle points out, “epistemological skepticism arises from consideration of the respects in which our beliefs and their relation to reality depend on factors beyond our control.”[20] Epistemic luck is a generic notion, and some kinds of luck may simply be irrelevant to an account of justification. But it is a highly contested issue which types present a relevant challenge to knowledge-claims. Both groups of virtue epistemologists concede that we cannot always avoid error without luck, but they typically see the relevant discontinuity, and hence the kind of luck involved, very differently.

In an innovative recent paper, “What Are the Chances of Being Justified?” Wayne Riggs argues “that both truth-conducivist and responsibilist conceptions of epistemic justification are directed at disallowing a particular kind of chance (true) belief from counting as epistemically sanctioned.”[21] Many of Riggs’ examples are drawn from virtue epistemologists, making his analysis especially pertinent here. “Chance” is selected as the common term characterizing conditions that might preclude instances of true belief from constituting knowledge, and the author then goes on to explicate and define two relevant kinds. A virtue reliabilist (Riggs uses the term “truth conducivist”) might define the justification condition on knowledge in such a way as to preclude the kind of chance that Riggs simply calls epistemic luck: “S’s coming to hold a true belief, P, is a matter of epistemic luck for S to the extent that P is unlikely, given that it was produced by process R.”[22] I will utilize Riggs’ terminology in this section, except that I will refer to this first kind of chance as coincidence, allowing us to preserve a broader and more conventional use for the term “epistemic luck” in the sections that follow. The discounting of coincidence reflects the insistence upon an objective link between process and truth that is characteristic of the truth-conducivist account of justification. The motivation behind reliabilism centrally involves disallowing beliefs that are merely ‘lucky guesses’ or coincidental truths from counting as knowledge. Consideration of Gettier cases, for instance, has strongly pushed towards external constraints on knowledge that can prevent a justified, true belief from counting as an instance of knowledge if its truth is merely a coincidence and is not “linked” with its causal ground in an appropriate (epistemizing) way.

By contrast, the kind of chance that responsibilists want to preclude from counting as knowledge is what Riggs refers to as “epistemic accident.”[23] Citing John Greco and Hilary Kornblith as examples, he takes their accounts of justification as primarily concerned with the evaluation of epistemic agents or their practices. The responsibilist begins from the agent’s motives, and so insists that a belief is justified only where the intentions to have true beliefs and to avoid false one have played appropriate grounding roles in the agent’s belief-forming practices. What the responsibilist means by an appropriate grounding role Riggs takes to be a certain kind of causal role, and so “S’s coming to hold a true belief, p, is an epistemic accident for S if coming to hold p was not (sufficiently) caused in an appropriate way by S’s intention to have true beliefs.”[24]

The responsibilists’ motivation for placing such a constraint precluding epistemic ‘accidents’ from counting as knowledge is partly explained by their intuition that the justification condition demands beliefs held (or propositions accepted) not only in conformance with our epistemic norms, but from or in light of such norms.[25] An agent’s consciously-held belief that satisfies this constraint will be one that is guided by his taking a support relation to hold between reason state r and his acceptance of p. This concern to preclude epistemic accidents, as Riggs points out, also explains the enthusiasm that responsibilists have shown for cases of deviant causal chains, where a reliable BCP has been in operation yet a belief produced by it fails to ‘link up’ with the agent’s intentions or subjective justification in an appropriate (epistemizing) way. Summarizing Riggs’ analysis briefly, coincidence (or epistemic luck for Riggs) is primarily a matter of success despite low-likelihood, given the agent’s actual process, while accident is primarily a matter of success despite intentional inefficacy, given the agent’s intentions.

Riggs’ most challenging claim in assessing the upshot of his analysis is that “The responsibilist and truth-conducivist conceptions of justification define distinct epistemic evaluations, and so are not in any interesting sense rival notions.”[26] He views their divergent constraints on chance as mutually-consistent, deriving from “an identical epistemological concern.” From this view of their consistency, he proposes an intriguing conjunctive formula for the justification condition on knowledge. The necessary and sufficient conditions for knowledge are according to this formula true belief, and a revised third (justification) condition, where the conjunction of the former two conditions “is not a matter of chance” in either of the two senses we have specified. This conjunctive definition is interesting in its own right, and may prove more resistant to the usual roundup of counter-example cases than definitions that do not include both the reliabilist and responsibilist constraints. He proposes a new focus for epistemology in line with his approach, that of working out “an adequate account of epistemically responsible belief, as well as an account of sufficiently truth-conducive belief.” It should not be lost on us that these two points also serve well to describe the mixed account of justification that we have seen is widely advocated among virtue epistemologists.

The compatibility that Riggs asserts to obtain between reliabilism and responsibilism is quite appealing; it must be correct in some sense for a mixed account of justification to be plausible. The motivation for his conjunctive strategy also seems clear enough: reliabilist justification prevents a true belief that will constitute knowledge from being a coincidence, but not from being an accident; and responsibilist justification prevents a true belief that will constitute knowledge from being an accident, but not from being a coincidence. However, at the same time I have reservations about Riggs’ conjunctive strategy for defining knowledge—that is, the simple adding of the one constraint to the other in order to render the account of justification more adequate or ‘complete’. This strategy may be shown to be problematic if it invokes assumptions that make it more rather than less difficult to respond to the challenge of skepticism. The one assumption I would focus on here is that of the incompatibility of luck and knowledge. We should notice, after all, that the analysis he presents puts us nowhere closer to explaining the commonplace belief with which we began, that the reliabilist takes herself to be more amenable to epistemic luck than the responsibilist. This point should help us to see that using stipulative definitions excluding certain forms of chance cannot provide a suitable answer to the entire complex of problems associated with epistemic luck. As discussions in ethics have rightly pointed out, luck is not entirely a conceptual matter, but depends upon the make-up of ourselves, our world, and our relationship to it. If we all rely on luck in our intellectual lives, then we cannot merely stipulate that knowledge excludes it—not, at least, without simply begging the question against the skeptic by assuming from the outset that we know what we think we know.

I do not mean to burden Riggs with holding the incompatibility thesis in its most objectionable form, the claim that knowledge is incompatible with luck simpliciter; he has defined the specific senses to be excluded, and it may be only a limitation of his focus that he does not address other senses of luck which may be compatible with knowledge. But an adequate account must go beyond showing that excluding epistemic coincidences is compatible with excluding epistemic accidents, to a positively stated account both of this compatibility and of epistemic justification. This would be one that explains in positive terms what it means to be “in a position to know” and what it means to be a “well-motivated and responsible” agent. Riggs hints at such a positive account in saying that the constraints against the different kinds of chance “arise from an identical epistemological concern,” and that the parties to the debate share a common conception of what epistemic justification is for (that is, a kind of filter of knowledge claims). Related claims that Riggs makes, that the divergent constraints on justification represent value-charged demands for the intelligibility of knowledge, and that neither can claim logical priority over the other, also hint at arguments for their compatibility. Yet these comments far outstep the arguments Riggs actually provides in his article.

Such limitations to his analysis are not surprising, I want to say, because there is another whole side to the issue of luck we have not yet explored. We begin to engage it in the next section by directing critical attention to the thesis of the incompatibility of knowledge with luck. I will there briefly lay out an argument against the incompatibility thesis, showing its self-defeating nature. Then in Section IV I will further elaborate certain senses of “luck” to which the reliabilist and the responsibilist must each respectively acquiesce once the incompatibility thesis is abandoned. In acknowledging kinds of luck that cannot be excluded but must instead be part of any adequate account of knowledge, I hope to build upon the idea of constraints on justification as value-charged demands for the intelligibility of knowledge, and thereby to clarify the basis for a positive account of the compatibility of reliabilism with responsibilism.

III. Skepticism and the Incompatibility Thesis

In this section I make use of another thought-provoking paper on epistemic luck, “Is Luck Compatible with Knowledge?” where its author Mylan Engel argues that “both internalist and externalist epistemologies lead directly to skepticism when they are coupled with the incompatibility thesis.”[27] Why must both internalist and externalist epistemologies reject the incompatibility thesis? Because the view that luck simpliciter is incompatible with knowledge entails skepticism. Why in turn is this the case? Because each of these theories can in fact be shown to entail a kind of luck. Luck, as Nagle pointed out for us, has always provided the skeptic with a foothold; unrealistically denying luck’s role in knowledge may provide him a ladder![28] For the sake of clarity I will distinguish the kinds of epistemic “chance” asserted in the previous section to be incompatible with knowledge from forms of epistemic “luck” whose nature we have yet to explore.

Engel’s paper utilizes the distinction between internalist and externalist epistemologies rather than the more subtle distinction between responsibilist and reliabilist epistemologies which is our real concern. We will need to follow Engel in this for the present, and in order to underline differences, I will typically mean by the term “externalism” its more ‘pure’ or non-mixed versions unless I specifically refer to a “mixed” version. We can also pick up the metaphors with which we began, by examining these kinds of luck in terms of the ideas of perceived crucial discontinuities or inferential lacuna (“Gaps”), and responding to these, attempts to build theoretical connections (“Bridges”).

Gaps are well-acknowledge in ethics, a prominent example being the Gap between being virtuous and living well, according to Aristotle: uncontrolled happening (moral luck) can step into this Gap, impeding the person of arete from finding her proper fulfillment in eudaimonia. Now it is clear that for internalist accounts in epistemology, there must remain a Gap between justified belief and truth. Internalists maintain that epistemic justifiedness is exclusively a function of the cognizer’s internal states—belief, memory, or perceptual states. No internalist theory can provide a conceptual Bridge or connection between justification and truth, as Gettier-cases purport to show. According to Engel then, “no internalist theory can eliminate the role luck plays in a person’s coming to have a true belief…. So if luck really is incompatible with knowledge, then no internalist epistemology can give rise to knowledge.”[29] Since the experience of those in a demon world is subjectively indistinguishable from our own, the internalist is content to construe demon world inhabitants justified in their beliefs if the quality of their efforts to attain truth and avoid error have been beyond impeachment. The internalist takes beliefs as virtuously held on the basis of evidence available to a situated agent, and when it is revealed that those beliefs are nevertheless false due to the demon’s systematic deception, she can merely acquiesce to a form of epistemic luck, conceding that luck must intervene to turn our justified beliefs into true beliefs, and hence into knowledge.

Externalist theories may appear to be on much stronger ground, but as Engel points out, if this is so the reason cannot be that the externalists need not admit luck into their accounts. On the contrary, “externalist epistemologies with truth-connected theories of justification simply replace one kind of epistemic luck with another, for while it is not a matter of luck when a [process-reliabilist]-justified belief turns out to be true, it is a matter of luck when a belief turns out to be [process-reliabilist]-justified.”[30] Engel terms the kind of luck to which the internalist must acquiesce veritic luck, and that to which the externalist must acquiesce evidential luck. Veritic luck might most simply be understood as luck with respect to the output of our BCP’s, and evidential luck as luck with respect to the empirical or evidential input that our BCP’s must work from and with.

A demon-world case can also be used to illustrate evidential luck. Consider yourself in relationship to your doppelganger in the demon world, and reflect on the fact that the two of you live phenomenologically indistinguishable cognitive lives. Since there is no discernible difference between your worlds, you must conclude that it could have just as easily been you who were the one with the unreliable belief-forming cognitive processes. Given your situated place within this scenario, there is no evidential basis that could possibly serve you as a guide, from which to reach any other conclusion. If this is so, then it is still a matter of luck that, in the context of this scenario, it is you and not she who has the process-reliabilist true beliefs.

According to the foregoing argument, the externalist must acknowledge that it is always a matter of evidential luck when reliably-produced beliefs turn out to meet the subjective conditions that internalists would place upon knowledge. And the internalist must acknowledge that it is always a matter of veritic luck when our beliefs turn out to meet the objective conditions an externalist would place upon knowledge. But is this a cogent line of argumentation? Clearly Engel has allowed the very basis for the characterization of luck to differ in his demon-world scenarios. But why should he not? Perspective matters! If it takes an externalist perspective to point out the kind of luck which internalism entails, it seems valid to allow the same in reverse, by allowing the evidential factors rather than matters of objective fact to be the ones that may vary in the examples. The perspective focussing on the evil-demon’s revelation to me illustrates that it must always be a matter of luck from an external or factual perspective when an internally-justified belief turns out to be true. The perspective focussing upon my relationship with my doppelganger illustrates that it must always be a matter of luck from an internal or evidential perspective that one of us, myself or my double, should in fact be the one possessing sound evidence for normally-functioning BCP’s to utilize as input.

But are these kinds of luck of equally significant import for epistemology? In Gettier and other types of cases which might have been discussed in connection to the kinds of “chance” examined in the previous section, our intuitions tell us that an agent’s justification is undermined by certain facts about his situation that are beyond her ken. In the eyes of their devisers, at any rate, such examples exploit factual defeaters to an agent’s knowledge claims that, once they have done their dirty-work, leave the would-be knower only with what Matthias Steup calls nonepistemizing justification. But is that also the case with the kinds of demon-world scenarios we are now considering? That seems not as clear. We are now encountering forms of epistemic luck that have greatest relevance to the general challenge of philosophical skepticism. This is partly why I separate them here from Riggs’ two forms of chance discussed in the previous section.

Our two forms of luck both have important epistemic implications, even if they differ from the forms of chance we earlier investigated. Engel himself argues that only veritic luck has epistemological import, thereby giving the strong advantage to externalism. I disagree here: Evidential luck, too, is a kind of luck that concerns the crucial relationship between the agent and the known fact; it is not, any more than is veritic luck, a kind that concerns merely the existence of the known fact or the existence or abilities of the person who knows. I will need to say more on this below, but I take input and output luck to be on generally equal footing in this regard (see also footnote 32). It takes the other’s perspective to point out the kind of epistemic luck that each account must admit, and internalists and externalists both think that once pointed out, the kind of luck systematically implied in their adversary’s account represents a deep fault in their approach. Moreover, both perspectives contribute to explicating the pre-theoretical notion of luck as referring to factors affecting that crucial relationship (between the agent and the know fact) that remain beyond our ken, or in other words, beyond human control.

Since internalism and externalism are defined by Engel in the usual mutually-exclusive and exhaustive way (exhaustive except for the option that would radically eliminate justification as a condition on knowledge altogether), we are left with the implication that luck is deeply involved in knowledge on either account. So to return to our main line of argument, if luck simpliciter really were incompatible with knowledge, then both theories would make knowledge impossible, and hence lead directly to skepticism. This argument I believe is cogent, and it leaves us with the following options: 1) embrace skepticism, 2) embrace a non-justificationist account of knowledge, or 3) reject the incompatibility thesis.

If we accept this scenario, then the last option appears most promising. This means that, again following Engel’s lead, we must try to “reconcile the rather strong intuition that epistemic luck [chance] is not compatible with knowledge with the equally evident observation that it must be.”[31] Both accounts have been only partially successful in shielding knowledge from a kind of luck that its adversaries see as undermining its claim to adequacy. Of course, there are certainly kinds of luck that are simply irrelevant to the question of knowledge. But neither of the types of luck Engel discusses appear to be of such a type, and the route of a non-justificationist form of externalism is unappealing. Nor should we assume that the presence in any degree of the one or other kind of epistemic luck is sufficient to undermine knowledge, an assumption which might lead us to embrace skepticism. Consistent with Engel’s third option of rejecting the incompatibility thesis, the point we should stress is rather that on both internalism and externalism, understood as mutually-exclusive accounts, it is necessary for luck to step into a perceived “Gap” in order for justified belief and truth to link-up in an epistemizing way. Both theories, in this sense, remain essentially ‘incomplete’ in their response to skepticism.

We have now argued that the incompatibility thesis is mistaken. A strictly conjunctive account, we should now be able to see, is unrealistic as an attempt at Bridge-building. It requires too much of human cognizers, because it excludes by fiat from our analysis all influence of factors that are a matter of luck. An adequate account of epistemizing justification cannot be stated in Riggs’ negative fashion of merely excluding luck from knowledge, even where that statement is in fact a conjunctive one expressing both internalist and externalist chance-precluding constrains. We have also argued that pure internalism and externalism—theories that state their conditions for knowledge in mutually-exclusive terms, must each by their own respective logic “acquiesce” to an important role for epistemic luck. Since each is premised upon the incompatibility of luck with knowledge, this acknowledgment paralyzes their ability to state conditions of knowledge in a positive manner. The upshot (here in line with Riggs) is that internalism and externalism as so understood are necessarily incomplete accounts, and that neither alone can be an adequate response to the skeptic. Taking this as background, we are now in position to discuss the advantages of the mixed externalist approach to justification, and perhaps outline the demands upon the positive account of epistemic justification we are seeking.

IV. The Gaps Problem: Dealing with the Incompleteness of Epistemic Theories

It is in their acquiescence to different forms of luck that I want to say we find the deepest source of the divergence between externalists and internalists, and by extension, between reliabilists and responsibilists. In the context of our earlier discussion, we said that Gaps represent perceived inferential lacunae or discontinuities in the human cognitive process. Gaps are present to us existentially simply as forms of luck which demand our recognition because they impact our lives. Philosophically, and for better or worse, we tend to perceive these demands differently one to another. If it weren’t obvious to the reader already, the discussion of the previous section serves to show how epistemologist’s routinely view the forms of luck logically implied by their own favored position as epistemologically innocuous, and their adversary’s forms as nothing short of a plague upon their house. This disemblance cannot long be maintained, and the oft-noted “stalemate” between internalism and externalism as mutually exclusive adversaries is easy to recognize. By admitting any Gap, any discontinuity in the theoretical account of human cognition, internalist and externalist epistemologies reveal themselves as incomplete. Since the forms of luck that the internalist and externalist each must acquiesce to are the direct target of their adversaries, it becomes apparent that neither approach, freestanding, has the resources to capture both sets of intuitions.

Conceptualizing this incompleteness in terms of “Gaps” is simply meant to formalize this problem. Gaps correspond to acknowledged forms of luck impacting human cognition and potentially blocking epistemizing justification. Bridges are attempts to respond by way of philosophical reconstructions to the threat posed by Gaps, and to lay out conditions under which justification is indeed “epistemizing”—that is, sufficient to turn true belief into knowledge. To clarify this, we must first try to formalize our conception of the Gaps themselves. Internalist theories, we’ve shown, leave us with a gap between(1)truth and (2) justified belief. We can simply call this 1—2 gap the Veritic Gap to correlate with the form of luck to which internalists must acquiesce. Externalist theories, on the other hand, do provide a conceptual linkage between justified belief and truth, that is to say, a conception of a necessary connection between knowledge and its object. But for them a gap re-opens between belief (or more strictly, propositional acceptance) and (3) good reasons or adequate evidence. This is a token of a discontinuity between the reliabilist's strong conceptual claims and their lack of a supporting theory of evidence.[32] Borrowing a Greek term to balance against the Latin veritas, we can call this 2—3 gap the Zetetic Gap. This comes from zetesis, referring to the quality of an investigator’s effort, or of her method of investigation, and was used by the Greeks in the context of a common inquiry into an unknown truth. Again this is intended to map correlate with the sense of evidential luck discussed above.

This characterization of Gaps follows the analysis of our previous section. But we should add at least one more, that between (3) reasons and (4) (proper) motivation. We can call this 3—4 gap the Enkratic Gap, for the Greek term for continence. Do reasons have to motivate? For instance, if I accept that I have reason to doubt/believe some proposition, must I be motivated to some degree to doubt/believe that proposition? What Christopher Hookway calls (motivational) internalism answers yes; it holds the normative impact of reasons on an agent to be inseparable from motivation, at least in rational agents. For the internalist, epistemic rationality requires us to be properly motivated: we should no even say that evidence provides a person with “reasons,” unless it is integrated in an appropriate way with their motivational tendencies, and “it seems that the existence of a gap between my normative judgments and my motivations signals a kind of irrationality.”[33]

Now if we hold this model of three Gaps between four terms [1—2—3—4] in mind, we can say that the amalgam of views associated with internalism allows it to present us with bridges for both the Zetetic and EnkraticGaps, but at the cost of leaving the Veritic Gap radically open. On the other hand it is often said that externalism presents a great departure from “traditional epistemology” with its predominantly internalist character. In the terms we have used in this paper, we might then say that externalism has the advantage of being able to radically bridge the Veritic Gap, but at the cost of leaving us with suspect resources for addressing the other two.

To conclude our brief discussion in this section, I would point out that each of the Gaps and associated kinds of epistemic luck we have discussed have strong analogues in Nagle’s discussion of moral luck. The Veritic Gap (and with it Engel’s veritic luck) bears resemblance to Nagle’s consequential luck; the Zetetic Gap (and with it Engel’s evidential luck) can be profitably compared with Nagel’s circumstantial luck; and finally, the Enkratic Gap, as exemplified in the possibility of a kind of epistemic akrasia that Hookway examines, betokens largely unexplored comparisons with Nagel’s constitutive luck. What I will call the Gaps Problem is the problem of providing a positive account of knowledge and justification, one that addresses each of the discontinuities under a single unified perspective, without ignoring or negating the influence of factors that are a matter of luck. Thus the parallels suggested between acknowledged forms of moral luck and the forms of epistemic luck we have examined are further indication of the import of the Gaps Problem for epistemology today.

V. Mutual Assistance

We have now described two specific forms of “chance” (section II) that, though sharply contested among reliabilists and responsibilists, are each claimed by some to preclude true beliefs from constituting knowledge; we have also described forms of “luck” (Section III) that must be acknowledged rather than precluded and that, while also impacting epistemology, do so primarily in terms of their relevancy to a general skeptical challenge that virtue epistemologists, whether those we have called virtue reliabilists or responsibilists, must confront together. In section IV we have seen that the presence of veritic and evidential luck indicates an incompleteness in internalist and externalist epistemologies, and most importantly, we have gone on to connect each form of luck with a theoretical “Gap” whose place in any positive account of justification demands acknowledgment. In this final section we will discuss objections to the present analysis of epistemic luck, and clarify the demands on a mixed externalist account of justification in addressing what we have come to call the Gaps problem.

One objection that might be brought against the discussion of epistemic luck in the previous two sections is simply that it seems to bear little direct implication for virtue epistemology. After all, the arguments in those sections lean heavily upon the dichotomy between internalism and externalism, understood as mutually-exclusive and exhaustive accounts of justification. But we had previously been at pains to show how and why virtue epistemologists have been attracted to a “mixed” externalist account, one that is capable of consistently integrating responsibility constraints on justification that ensure the guiding role of good reasons in reflective human knowledge. To hold a mixed account is already to agree that purely internal or purely external constraints cannot provide the right conditions for epistemizing justification. Even if there are serious differences among those who advocate mixed accounts, those we began by characterizing as virtue responsibilists are not really internalists, since they certainly do not think that the conditions on justification are exclusively internal; and those we characterized as virtue reliabilists do not claim, as some externalists might, that we can do without an account of subjective justification for reflective knowledge. So the discussion became skewed when we slipped back into talking in terms of a dichotomy between internalism and externalism that most of us agree has outlived its usefulness. For one who accepts a more compromising mixed account, the problem of epistemic luck need not have either the decisive importance or the divisive character I have suggested it has.

In response, I would firstly have the reader note that I have been concerned with an indirect effect of the problem of epistemic luck on mixed externalist accounts. We have considered how the understanding of epistemic luck in virtue epistemology tends still to be divided along lines that reflect a backdrop of externalist/internalist debate in epistemology over the past three decades. I agree that this distinction is becoming outmoded, and hope that the distinction between reliabilist and responsibilist virtue epistemology will before long also outlive its usefulness. But that should not preclude our notice of real and present tensions among those working in the field.

Secondly, two common claims made by virtue epistemologists of all stripes are that their approach has something unique to offer in the way of overcoming the opposition between internalist and externalist conceptions of justification, and that it provides substantial resources for addressing the challenge of skepticism. I can find no better way to put these two common claims to test than to inquire into the reception and understanding of epistemic luck. A virtue-based account of a mixed type should show that justification does not turn on the issue of “internal accessibility” to the ground of one’s belief, the issue that the distinction between internalism and externalism was built around. While the various strategies of virtue epistemologists who advocate mixed accounts of justification suggest a displacement of the access issue, we have found that the issues have been transformed but not resolved, and that serious disagreements persist between virtue reliabilists and virtue responsibilists.

Thirdly and most importantly, I disagree with Driver when she says that mixed accounts represent inherently unstable compromises, and that “the superficial plausibility of the mixed account is purchased at the cost of significant theoretical advantages” presented by the “pure extremes” of internalism and externalism. On the contrary, I have tried to argue that the advantages reside in a view that takes seriously both reliability and responsibility constraints on justification, because this mitigates the necessary incompleteness of ‘pure’ internalism and externalism and thereby provides a more satisfactory basis for responding to skepticism.

But I would make a partial concession to Driver’s complaint: Mixed accounts certainly are not immune to the Gaps Problem or to more generalized problems concerning epistemic luck; they remain subject, as Driver claims, to many of the same theoretical problems which afflict the internalist and externalist extremes. This concession may indeed be implied by my view that one simply can’t bridge all three of our recognized gaps simultaneously under one perspective (as the notion of adding or conjoining responsibilist to reliabilist conditions on justification would suggest). This is philosophically unrealistic, because it would produce a conception of the human agent with cognitive powers far in abundance of what the cognitive sciences and our own lived experience reveals to us. To say as we did before, that Nagle’s three kinds of ethical luck can be seen to have important epistemic analogues, is to say that a mixed account is one that will seek to understand the epistemic import of each of the three Gaps we have sketched, and to produce a positive account of epistemic justification in light of them.

Acknowledgment of the Gaps Problem motivates rather than prevents the tasks of theoretical Bridging. But my conception is one on which there are serious trade-offs involved in such theoretical endeavors: to emphasize a conceptual connection here is to allow a potentially larger lacuna elsewhere in one’s system of thought. And if this is correct then there are undoubtedly a variety of different ways that epistemic luck can be theoretically addressed. One might say that our theoretical Bridges are makeshift, especially in this era of new empirical studies in the cognitive sciences, and that they will need to be constantly adjusted one to another as we learn more about human cognition and motivation structures, and as we advance our theories of knowledge and evidence in light of them.

It is simply not clear at present how well mixed accounts can handle this task, but we have shown reason to think that mixed accounts are on better footing than non-mixed accounts in this regard. While sometimes lauded for a certain perceived rigor or inner consistency, we should now be able to see that non-mixed approaches place themselves in a position of having to argue, most implausibly, that the type of luck to which their account must acquiesce lies outside the relationship between the knower and the known, and is of no real epistemological significance. To return to Driver’s objections, I cannot resist playing her claim of inherent instability off against Jonathan Dancy’s similar claim. An irony is present because Dancy uses similar reasoning to prescribe a move away from consequentialist/reliabilist conception of the virtues towards a virtue epistemology based upon Aristotle’s account of the moral virtues:

Non-consequentialists also are unwilling to admit that the consequentialists are right about anything, because they feel that consequentialism is like a cancer—once one has let it in at all it will grow until it has taken over completely. The crucial question is whether the two camps are right at least about this, that no compromise is intellectually acceptable. And I think that they are.[34]

Though the conclusions that Driver and Dancy reach are highly antithetical, their ‘inexorable’ logic on this score is strongly analogous, since both argue that compromise solutions are theoretically unstable and intellectually unacceptable.[35] Such stability-objections really amount to little more than predictions concerning the fate of mixed accounts. If mixed accounts prove only superficially plausible as both think, virtue reliabilists and responsibilists can be expected to soon part company and to do their research under separate self-descriptions rather than accepting their shared description as virtue epistemologists. With such predictions I need not agree. If this prediction is not to prove correct, however, there is a burden on the virtue epistemologist to present a theoretically consistent, positive account of epistemic justification, one that reliabilists and responsibilists both substantially agree with and contribute to. At the end of a lengthy paper, however, it may come as some relief to the reader to learn that I do not have such a theoretical account ready to offer!

I would however put forward a practical point in closing. The stability of any research program is often as much a practical matter as a theoretical one. The easiest way to exhibit instability is to lose the valuable benefits of mutual assistance. Reliabilists have focussed on the demands of philosophic naturalism and on conceptions of the supervenience of normative properties; they have been especially attuned to the role of the cognitive sciences in understanding both reflective and non-reflective (including animal) knowledge, and to the social and communal dimensions of epistemic evaluation. Responsibilists have focussed on concerns with active agency in the context of reflective knowledge, and with studies and thick-descriptions of particular epistemic virtues and vices; they have been especially attuned to the interconnections between ethical and epistemological dispositions, and to the dynamics of individual psychology. With a mixed externalist account of justification progressing as a shared project among virtue epistemologists, I am hopeful that the complementarity of these research foci will achieve greater recognition, and that the interests that shape research within virtue epistemology will merge significantly further. Surely this mutual-acknowledgment and integration of interests is practically necessary if virtue epistemologies of a “mixed” externalist character are to confront the stability-objection that we have heard expressed from positions representing internalist and externalist extremes.

Notes