5 Epistemic Normativity
The repudiation of veritism opens an important gap.1 If truth-conduciveness is not the standard by which epistemic policies and practices are to be judged, what is? I suggest that rather than epistemic practice having an exogenous aim, its aims are delineated by and within the practice itself. Epistemic standards are vindicated not by being reliable promoters or indicators of truth or any other external end, but by being products and promoters of responsible epistemic agency. Epistemic norms, then, are norms of responsible epistemic agency. In this chapter I argue that epistemic norms are norms that would emerge from the deliberations of suitably idealized epistemic agents. Drawing on Kant, I urge that such agents are to be construed as legislating members of a realm of epistemic ends. This is an idealized model of real-life communities of inquiry. The idealization eliminates factors that distort deliberations. I argue that legislators must be, in a political sense, free and equal. They must be free to advance any hypothesis they consider viable and equally entitled to be heard. This does not mean construing them as equally smart or construing their views as equally worthy of acceptance. It means only that they are entitled to be heard. The result is a form of epistemic responsibilism. It explains why a system in reflective equilibrium is worthy of acceptance even though there is no guarantee that it is true.
Reflective Endorsement
If epistemic normativity is keyed to epistemic responsibility, what are the criteria for epistemic responsibility? One familiar requirement on responsibility is ‘ought’ implies ‘can’: x is responsible for y only if whether y obtains is under x’s (direct or indirect) control. If x cannot bring about y, or x cannot prevent y, then x is not responsible for the occurrence or nonoccurrence of y.2 Epistemic agents have little if any control over their epistemic environment. They cannot, for example, control whether they live in a Berkeleyan world. So they are not epistemically remiss if the falsity of their beliefs derives from an undetectable mismatch between the evidence and its metaphysical ground. Nor are agents epistemically remiss if, in a more auspicious world, they form a false belief on the basis of evidence that is scrupulously garnered and carefully analyzed but misleading. Truth-conduciveness, then, is not the criterion of epistemic responsibility. What is the alternative? Even in an inauspicious world, agents can exercise epistemic self-discipline (see Weatherson, 2008). They can decide whether on reflection to endorse considerations that present themselves as candidates for acceptance.
According to Kant (1981), someone who acts autonomously makes the laws that bind her, while someone who behaves heteronomously is bound by constraints he neither makes nor endorses. He is not strictly an agent, for he does not act; he merely reacts. He responds to stimuli. Although Kant describes the heteronomous subject as driven by his inclinations or desires, action is a joint product of belief and desire. So a subject is equally heteronomous if he is driven by beliefs that are not under his control. This suggests that it may be fruitful to import Kant’s distinction into the epistemic realm.
Inasmuch as we are cognitive beings with sensory and representational capacities, a variety of things simply strike us as being the case. Deliverances of sensation, emotion, apparent memory, and imagination present themselves as candidates for acceptance. It might seem that this makes us, at base, epistemically heteronomous. We can’t help but see, hear, and feel what we do. But the issue is not whether we are passive in the reception of inputs; it is what we do with those inputs. The heteronomous individual simply takes deliverances at face value. For him, seeing is believing. Or perhaps seeing is not always believing, but whether he believes what he sees is not up to him. He simply finds himself believing something on the basis of his deliverances, being under the sway of whatever belief forming mechanisms happen to be operative. Since his beliefs are considerations that just strike him—since he can neither defend nor criticize them—they are not under his control.
The beliefs he finds himself with might be reliable. Perhaps humans evolved in such a way that what strikes us as dangerous usually is dangerous. If so, when the subject thinks an animal is dangerous, it usually is; when he thinks an animal is harmless, it usually is. Maybe this generalizes. To the extent that his dispositions are reliable, he is objectively secure. Nevertheless, he is in a subjectively precarious position. Reliabilism is a form of veritism that holds that a disposition or process is epistemically creditable just in case it is reliably truth conducive and a true belief is knowledge just in case it is a nonfortuitous product of exercising a reliable disposition or of appropriately using a reliable process. Whether the reliabilist requirement is satisfied is independent of whether the subject thinks it is and independent of her reasons (if any) for thinking it is. Evidentialist veritism holds that a subject’s true belief qualifies as knowledge just in case it is backed by adequate evidence, where the adequacy condition covers not just the amount of evidence but also its being nonfortuitously linked to the belief’s truth-maker. Again it makes no difference whether the agent thinks that the requirement is satisfied. As far as veritism is concerned, heteronomous, justified, nonfortuitously true beliefs, and the processes and dispositions that underwrite them, are epistemically creditable. Still, insofar as the subject’s reflective endorsement or repudiation of his beliefs and the processes that gave rise to them have no bearing on their epistemic status, he is a victim of circumstances.
Perhaps his situation is not so dire as I make it out to be. The believer can register his successes and failures, and credit the dispositions and mechanisms that engender success. He might notice that deliverances of foveal vision tend to be borne out by subsequent experience more frequently than deliverances of peripheral vision. So he assigns foveal deliverances greater credence. But the question then arises for the second-order beliefs, processes, and dispositions he credits. Do they just strike him as correct? Can he defend the way he registers, identifies, records, and tallies successes and failures? Even if his methods are in fact reliable, so long as he has nothing more to say than that it strikes him that this is a good way to proceed—to identify, tally, or measure—he remains vulnerable. Second-order heteronomy is as problematic as first-order heteronomy.3 Moving to yet higher orders is of no help. Regardless of the level, so long as the best that can be said for the principles or processes he ultimately relies on is that they strike him as correct, he is vulnerable. He may be on safe ground, but he has no reason think he is. He is driven to believe by considerations he happens to be struck by. And he is not responsible for what strikes him.
Is the autonomous agent any better off? She too may be struck by things. Her inputs may be exactly the same as those of the heteronomous subject.4 But her relation to them is different. Forming opinions is something she does, not something that happens to her. What she does in forming opinions is, as it were, filter deliverances (such as putatively perceptual inputs, apparent memories, or simply ideas that cross her mind) through a critical sieve, accepting only those she considers worthy of her reflective endorsement. What blocks the regress to ever higher-order heteronomy is that second-order endorsement is the product of agency. The agent decides to accept, reject, or withhold. Epistemic subjects are epistemic agents; they take their beliefs, practices, and policies to be answerable to certain norms because they think that epistemically acceptable beliefs, practices, and policies ought to be answerable to those norms.
To be sure, an epistemic agent does not entertain deliverances one by one and ask herself, ‘Am I buying this?’ Rather, she develops and deploys a variety of methods, mechanisms, heuristics, and habits that enable her to credit or discredit wide swaths of inputs efficiently. Apparent sightings of not unexpected middle-sized objects in good light in the center of her visual field are apt to be readily accepted. Given her prior experiences and other epistemic commitments, having no reason to reject them may constitute reason enough to accept them. But if there is something incongruous about a deliverance—if, for example, she seems to see a camel on Main Street in small town America—she does not simply accept it. She may reject it as a hallucination, an illusion, or a misperception; or she may suspend judgment until she investigates whether a camel is part of a local publicity stunt, political rally, or circus parade. Similarly if she is aware of something untoward about her condition as an observer: if she is half awake or on new medication or intently focusing on something else, she may refuse to credit a surprising deliverance. The critical point is that what she accepts is a product of her reflective endorsement. She is willing to stand behind it because it satisfies her standards. Even if her acceptances are largely automatic, she could, if she chose, withhold acceptance of the content of a deliverance, or of a source of deliverances. What makes the epistemic agent responsible for her opinions is that she takes responsibility for them.
Voluntarism
Can she really choose? Doxastic voluntarists, such as William James (1951), maintain that belief is at least sometimes subject to the will. According to James, if someone thinks his life would go better if he believed that a benevolent God watches over him, it is both possible and epistemically permissible for him to do so. Pretty much everyone acknowledges that the will can indirectly influence belief. An agent can garner evidence, put herself in a position to be influenced, even arrange to be hypnotized in hopes that the belief she desires will result. Sometimes such strategies are effective. But voluntarists like James claim that the influence of the will is direct: A person can believe something simply by deciding to do so. James restricts his voluntarism to realms, like the religious, where empirical evidence does not have much purchase. Such a restriction is imperative if the position is to be at all plausible. It would hard to maintain that a driver could simply decide to believe that she was cruising down a country lane in the south of France when all the evidence points to her being caught in a traffic jam on the Cross Bronx Expressway. But, James maintains, where there is no evidence either way, a person can believe that God exists or believe that God does not exist, as she desires.
Williams disagrees. He argues that because belief is world directed, and belief formation is responsive to evidence, the state that would emerge from such a resolution would not be belief.
If I could acquire a belief at will, I could acquire it whether it was true or not; moreover I would know that I could acquire it whether it was true or not. If in full consciousness I could will to acquire a ‘belief’ irrespective of its truth, it is unclear that before the event I could seriously think of it as a belief, i.e., as something purporting to represent reality. At the very least, there must be a restriction on what is the case after the event; since I could not then, in full consciousness, regard this as a belief of mine, i.e., as something I take to be true, and also know that I acquired it at will. (1978, 148; see also Adler, 2002)
In order to be responsive to the world, Williams maintains, belief must be resistant to the will.
The issue is trickier than Williams recognizes. A standard objection to doxastic voluntarism is that if belief (or belief formation) were voluntary, an agent could form any belief she likes, just by deciding to do so. The manifest impossibility of believing whatever one likes regardless of the evidence seems to settle the case. But voluntary actions are not completely unfettered. Walking is voluntary even though no one can at will walk from San Francisco to Honolulu. Voluntary actions are subject to constraints.
Some are what we might call enterprise-specific constraints. An enterprise both enables and limits the range of a participant’s voluntary actions. Only in chess is a knight fork possible; and how a player can create a knight fork is constrained by the rules of the game. Nevertheless, moving one’s knight so as to create a fork is a voluntary action. The fact that a player cannot perform that action whenever she pleases—while playing football, for example, or while baking brownies—does not undermine the status of the move as a voluntary action. Nor does the fact that a chess player cannot create a fork just by wanting to. The pieces on the board must be distributed in such a way that a player can move her knight into a position that puts two of her opponent’s pieces in jeopardy. This is not always possible. But when it is possible, her decision to move her knight to the relevant position is surely a voluntary act.
How does this bear on belief? As we have seen, to take belief formation to be subject to the unfettered will is a mistake. We cannot make ourselves believe that it is snowing (Feldman, 2008), or that we’re driving through France, or that a benevolent God exists (Pascal, 2007) simply by deciding to do so. The question is whether the doxastic realm is, like a chess game, a realm in which voluntary actions are subject to enterprise-specific constraints (see Shah, 2002). I suggest that it is. Beliefs are supposed to be responsive to evidence. Still, agents can and often do cut corners, jump to conclusions, reason carelessly, and think sloppily. They are subject to criticism for their errors and omissions, for they could do better if they tried. Or so, evidently, we think. There is apparently a range of freedom available to agents in their evidence gathering and assessing. The availability of that freedom to maneuver affords a basis for praise and blame, because how scrupulous epistemic agents are is under their voluntary control. The freedom to maneuver is delimited by the epistemic enterprise. To accept that p is to be willing to use p as a basis of inference or action when one’s ends are cognitive. This is not the same as to be willing to use p whenever one likes. The agent is not free to import into the epistemic realm factors, such as wishes or desires, that reside elsewhere. So the problem with James’s ‘will to believe’ is not, as Williams thinks, that belief is wholly involuntary. It is that James contends that epistemic agents can draw on factors outside the epistemic realm in deciding what to believe.
Nevertheless, a case sketched by Firth (1981) and elaborated by Berker (2013) might seem to raise a serious difficulty for my position. Suppose Jane is a talented chemist with a promising hypothesis. She needs funding to test it and the only available source of funds is a foundation that only supports research by theists. Jane, unfortunately, is an atheist and a terrible actor. She knows that she cannot persuasively fake religious belief. So she has herself hypnotized into believing that Odin exists. (The foundation insists on belief in a god but is ecumenical about which god it is.) She gets the grant, does the research, and makes contributions that advance the understanding of her subject considerably. Despite the good that came of it, one wants to say that her belief in Odin’s existence is epistemically unacceptable. It was illicitly imported from an extra-epistemic realm.
It might seem, however, that I can say no such thing. Jane accepted the contention that Odin exists as a basis for action (the action required to get the grant) when her ends were cognitive (doing the research). So ‘Odin exists’ looks like a felicitous falsehood. If I am committed to this, my position is in trouble. Luckily, I am not. Jane’s false and unwarranted belief is instrumentally relevant to her contributions to science. An epistemic consequentialist should recognize that it is truth conducive, hence presumptively acceptable. But the belief does not figure in Jane’s contributions to science. Talk of Odin appears nowhere in her scholarly work. Nor is it presupposed by that work. Once she has obtained the funds, her belief in Odin’s existence plays no further role. It is not woven into her fabric of epistemic commitments. Felicitous falsehoods, on my theory, are integral to tenable accounts. They are not merely causal antecedents. My position is nonconsequentialist.
An agent’s epistemic behavior is, in a variety of obvious ways, subject to her will. She can decide when to stop gathering evidence, when to stop checking inferences, when to stop investigating other perspectives, and so on. Epistemic acceptance thus is properly subject to ‘ought implies can’.
Still, it might seem that to some extent belief is heteronomous. Given her evidence for p, an agent cannot help but believe it. Here the distinction between conviction and acceptance pays dividends. Perhaps she cannot help but feel that p is so; p, then, is her conviction. But acceptance is a product of autonomous agency. An agent decides to accept p as a basis for cognitively serious inference and action when she reflectively endorses that p. In acceptance, the will has a role to play.
Williams is right that the world impresses itself on epistemic agents. We do not get to simply wish its deliverances away. Nor can we cavalierly refuse to countenance deliverances because we want to. The issue is what we can endorse upon reflection—that is, in light of our other epistemic commitments. We can responsibly reject or suspend judgment about a deliverance if it seems sufficiently implausible. This is what Nozick (1981) calls the Optional Stop Rule. It is always open to epistemic agents to refuse to countenance a conclusion that they consider sufficiently implausible. They can deem that the argument for it contains an undetected flaw. In everyday epistemic practice, we thus have a mechanism for responsibly withholding reflective endorsement. But consigning a deliverance to the realm of misperceptions or hallucinations, or an argument to the realm of those with undetected flaws, has costs. Consigners then face the job of explaining what gave rise to the misperception or hallucination, or where the flaw is located. To bring their views about their recent experience into reflective equilibrium, they need to account for the recalcitrant deliverance.
The role of the will in reflective endorsement is not at the level of asking whether the agent wants to accept a particular contention. An agent cannot reflectively endorse a contention simply because she wants things to be as it says they are. Nor can she reflectively endorse it because, despite its manifest untenability, it would be, as it was in Jane’s case, epistemically fruitful to do so. Rather, the agent, along with other members of her epistemic community, sets standards of evidence and thresholds for acceptance. They determine how much evidence is required for a contention of a given sort to be worthy of reflective endorsement. The agent is not at liberty to decide as she likes whether she sees a camel on Main Street. But she and the other members of the community have a measure of liberty in deciding how much evidence she needs for the contention that she sees the camel on Main Street to be acceptable. That liberty is constrained by requirements of consistency and coherence with other matters they reflectively endorse.
Epistemic endorsement, then, is a product of agency. Acceptance is under an epistemic agent’s voluntary control. But it does not follow that a responsible epistemic agent can accept whatever she pleases. Reflective equilibrium requires that her epistemic commitments be mutually supportive and that they constitute an account that is at least as reasonable as any available alternative in the epistemic circumstances.
Reasons
Reflection does not occur in a vacuum. It involves sensitivity to epistemic ends and means, capacities and limitations. It is imbued with (often tacit) background assumptions and is responsive to epistemic circumstances. A sighting of what looks to be camel on Main Street typically warrants further investigation; a sighting of what looks to be a mid-priced American car on Main Street ordinarily does not. An argument leading to an astonishing conclusion often provokes further inquiry; an argument leading to an expected one normally does not. An agent’s reflective endorsement is a willingness to be bound by an epistemic commitment (an opinion, method, or standard) because, given her background epistemic commitments, she thinks that her epistemic purposes will be served by her being so bound. She considers the items she reflectively endorses trustworthy bases for inference and action when her ends are cognitive.
It might seem that this too leaves the agent vulnerable. She is willing to stand behind the commitments she reflectively endorses. But should she be? The reliable but heteronomous subject is vulnerable because he cannot reflectively endorse his beliefs. It might seem that the autonomous agent, as I have characterized her, has the opposite problem. She can reflectively endorse her opinions, but they might be unreliable. She is perhaps subjectively secure, but she remains objectively at risk. In fact, if her personal standards are sufficiently skewed, her reflective endorsement may be epistemically worthless. Can we evade an untenable subjectivism here?
When an agent’s commitments are ones she reflectively endorses, they are at least prima facie defensible. She can adduce what she takes to be reasons for them. But it is not enough that the agent has something she considers a reason. If her claim that chicken soup alleviates colds were challenged, she could not prevail by saying that she read graffiti to that effect. Reasons, as Korsgaard (1996b) argues, are considerations we give to each other. They are not mere expressions of personal conviction, but considerations our interlocutors should countenance. The agent could give as her reason that she had amassed and evaluated the relevant evidence, that she read about the correlation in the New England Journal of Medicine or the New York Times, that she had been told by her physician that this particular old wives’ tale is true. She might even say that her grandmother told her. Assuming her grandmother is not an expert on contagious diseases, this would be a relatively weak reason. But if members of her epistemic community were inclined to give the grandmotherly voice of experience some epistemic weight, it would still be a reason.
Among the considerations that are within the agent’s epistemic purview are those she can call to mind. She has learned them and has not forgotten them. These are her available reasons. If she recognizes their relevance, she can surely appeal to them to defend her claim. But her purview is wider than that: it includes information she can readily draw on—information obtainable in her epistemic milieu, whether she is currently privy to it or not. These are her accessible reasons. In some contexts, we take available reasons to set the limits of accountability. Phil has an available reason to believe that Jane is old enough to vote, because he knows that Jane is older than Jim and he knows that Jim can vote. In others, we appeal to accessible reasons. Meg has an accessible reason to believe that the course has a prerequisite, since that is clearly stated in the catalog, and she has access to the catalog. Accessibility and availability are matters of degree. Some reasons are readily accessible; others take considerable digging. Some available reasons are easily called to mind; others need to be dredged up. The resources available to an agent’s epistemic community figure in her range of accessible reasons. She can in principle gain access to whatever information or insight is publicly available to the members of her community. She can draw on her intellectual compatriots and, as a member of that community, she has reason to endorse them.
Reasons are considerations that can properly be adduced to support a conclusion. Drawing on Scanlon (1998), I suggest that a consideration c can properly be adduced to support p only if others who are competent to assess the relation between p and c and motivated to know whether p is acceptable could not responsibly reject the claim that c supports p. Reasons thus are public. No purely subjective consideration counts as a reason; for others who are similarly motivated to find out whether p is acceptable could responsibly reject private intimations. However heartfelt, Pat’s feeling that the Red Sox will win is not a reason to think that they will win. It is merely a hunch. Moreover, since nothing epistemically inaccessible qualifies as a reason, a reliable method for establishing that p is not a reason to accept or believe that p until investigators establish that the method is, or is likely to be, reliable. Currently, then, there is no reason to accept the deliverances of Marie’s ESP, even if it is reliable. And a motley collection of evidence does not supply a reason to accept that p, unless its connection to the conclusion is accessible to the agent.
No truth that fails to satisfy accessible standards of relevance counts as a reason. Considerations that enhance the objective probability that p are not reasons if, according to current or foreseeable standards of relevance, they have no bearing on whether p. Even if left-handed people are more likely to develop arthritis than right-handed people, so long as we are unaware of the correlation and are not remiss in being unaware, Ben’s left-handedness affords us no reason to think that he is more at risk for arthritis than his right-handed brother. The correlation is epistemically inert.
Whether a consideration qualifies as a reason depends on the epistemic circumstances. Public standards of evidence and relevance rest on background assumptions about the topic under investigation and what is known or reasonably believed about it. Considerations that at one point in history could not responsibly be rejected may be readily rejected later, when matters are better understood. In 1975, it was widely accepted that stress causes peptic ulcers. A gastroenterologist could not at that time responsibly reject the contention that the high incidence of ulcers among Wall Street traders was due to their stressful jobs. Once the bacterial basis for ulcers was discovered, stress can no longer be adduced as a reason why so many traders have ulcers. A factor that functions as a reason in one cognitive environment may fail to so function in another.
Reasons are keyed to methodology. Among the factors that figure in whether others can responsibly reject the claim that c supports p are views about methods for establishing whether p. Are standardized aptitude tests a good predictor of college success, or can one responsibly reject the contention that a student’s acing the SATs indicates that he will do well? Are focus groups a good way of gauging public opinion, or can one responsibly reject the contention that because a focus group found the candidate’s message compelling, the public at large will do so too? Answering such questions requires validating the methods that connect c and p, and insuring that appropriate standards of rigor are satisfied. A major question is: whose opinions matter? To be exceedingly tolerant, to allow just about everyone to have a say, is to court skepticism. For any interesting thesis, there are bound to be people whose background commitments are such that they could, without violating their own epistemic principles, reject a contention that c is a reason to hold that p. A blind person might reject the contention that the car’s looking blue in daylight is reason to think that it is blue. He has no direct evidence of that. A statistical novice looking at graduate school admissions might conclude that the university discriminates against women. (Cartwright, 1983, 36–38). Being unaware of Simpson’s paradox, he does not understand why, if departments do their own admissions, looking department by department rather than surveying the university as a whole yields the appropriate statistics. A religious fundamentalist might reject the contention that the fossil record affords reason to believe that birds evolved from dinosaurs. He denies that anything could afford genuine reason to believe that evolution occurred. To disqualify such reactions requires a principled way to delineate the class of others whose reactions matter.
More is needed than that the standards be shared. All three rejecters might be deploying standards they share with others. The issue is whether those standards are cognitively acceptable standards, standards whose satisfaction fosters the advancement of understanding. To resolve this, we need to look at the standards and the goods they promote.
In all three cases, we dismiss the rejecters’ opinions because we know better. This is not intellectual arrogance. We know what relevant resources are accessible and how and why to draw on them. We dismiss the statistical novice’s opinion because it rests on a misunderstanding of statistical methods. This indicates that acceptable reasons must be consonant with the proper use of acceptable methods, where the acceptability of the methods turns on their having been validated and endorsed by experts on the subject and the best available ways to justify conclusions about it. But it is not enough if by chance statistics are properly used. The proper use of the methods should be based on an understanding of the methods, their ranges of application, their powers and limitations. This sounds circular, but it is not. Rather, it is a manifestation of holism. For reflective equilibrium, methods, standards, and results must mesh. When one strand of the network of commitments is doubtful, the rest of the network is properly called on for support. We dismiss the blind man’s opinion because it is based on a dearth of direct evidence and a failure to make appropriate use of accessible indirect evidence. This indicates that epistemic agents are expected to be aware of their own limitations and of the resources they can draw on to compensate for them. We dismiss the fundamentalist’s stance because it is dogmatic. It is not sensitive to evidence, since he would retain his position whatever the evidence. This indicates that for something to count as a reason for an empirical claim, it should be responsive to evidence.
The fundamentalist might reply that his opinion is responsive to evidence. Scripture, he maintains, provides incontrovertible evidence that evolution did not occur. He disregards the fossil record because nothing it shows could override scripture. His belief then rests on shaky grounds. He relies on a single source, so his belief is unstable. Should that source be discredited, he has nothing to shore up his belief. He relies on a single, disputed interpretation of the deliverances of that source, so it is fragile.
The scientist is in an epistemically stronger position. Because her opinion is multiply tethered, it is solidly grounded. We would have to be massively wrong about genetics, anatomy, and physiology to discredit the direct biological evidence for evolution; wrong about physics to discredit the evidence provided by carbon dating; and wrong about geology to discredit the evidence of the fossil record. The scientist then can draw on a sophisticated theory of empirical evidence, which explains why and how the sorts of factors biology relies on should be considered trustworthy. A second strength to her position is its fallibilism. Science does not consider any methods or results incontrovertible. Even the best of today’s methods and results might need to be reconsidered on the basis of future findings. So science builds in stabilizers to accommodate its recognition that with the advancement of understanding come refinements in methods and epistemic standards. The conviction that standards and methods neither can nor need ever be revised leaves the dogmatist in an epistemically precarious position.
It may seem paradoxical that commitments we are prepared to revise are epistemically more solidly grounded than ones its adherents consider unrevisable. But that is so. The epistemic value of science lies not in the conclusiveness of its results, but in the self-correcting character of its methods and the self-refining nature of its standards. Because we consider a method good, we consider its deliverances trustworthy. Because it delivers results that satisfy our epistemic standards, we consider it a good method. Because results that satisfy these standards promote our epistemic ends, we consider these standards the proper ones. Our views about the particular theses we take ourselves to have reasons for intertwine with the methods we take to count as generating reasons, and the standards we take reasonable opinions to have to satisfy. The interdependence is dynamic. As our understanding of a topic grows, so does our understanding of the best ways to investigate it, to validate the methods for investigating it, to calibrate the instruments used to investigate it, to identify the standards to which views about it should be held, and hence to characterize what counts as evidence in that area (see Hacking, 1983). Solidity does more than just provide insurance. It also deepens understanding. The various strands tie a solid opinion to different commitments, yielding a wider perspective as to where it fits and how it functions in our overall account.
Still, the publicity requirement may raise a worry. Suppose a brilliant mathematician constructs a valid proof that no one else can grasp.5 Given what I’ve said, it might seem that he has no reason to accept the proof, since his colleagues cannot back him up. The worry is unfounded. If the proof proceeds from publicly accepted axioms via publicly accepted rules, it satisfies the publicity requirement. No mathematician has reason to reject it, since it satisfies the community’s standards for an acceptable proof. Other mathematicians might, and probably should, suspend judgment about its validity until they can follow the proof. But they lack grounds for rejection. Nonetheless, the theorem prover’s position is epistemically precarious. If the proof is so complicated that no other mathematician can follow it, he ought not be confident that it is free of subtle errors. This is why mathematicians check one another’s proofs.
The dialectical interplay needed to achieve reflective equilibrium may take time. In the case we’ve been considering the proof is merely a complicated application of antecedently accepted commitments. The only difficulty is how to bring others to see that it satisfies standards that they all already endorse. Innovations can be more radical. Sometimes it is only with hindsight that we can tell that an initially tenable consideration was acceptable. When it was first proffered, its epistemic status was doubtful. The method of reflective equilibrium accords a presumption in favor of what has already been accepted. But that presumption is defeasible. So a putative proof or finding that skirts, or even violates, currently accepted standards, or is generated by methods that are not currently accepted, has an uphill battle.
But a novelty can prevail. If it is powerful or intriguing enough, it may prompt reconsideration of prevailing standards of acceptability. When first introduced, the Monte Carlo method did not satisfy accepted statistical standards. To assimilate it, statisticians had to rethink the nature of statistical sampling. They had to recognize that the pseudo-random numbers generated by von Neumann’s algorithm were random enough to be reliably unbiased, and had to reconcile themselves to the ineliminable use of computers for running simulations (Metropolis, 1987). The power of the method, and the recognition that, for the purposes for which it was used, its deviations from statistical orthodoxy were unproblematic, made a convincing case for the revision.6
By itself, the requirement that considerations worthy of reflective endorsement be responsive to reasons says nothing about what range of reasons a particular contention or account must be responsive to. That varies with subject matter, available resources, and the relevant community of inquiry’s tolerance for epistemic risk. The point here is only to show that there is a viable conception of a reason and a viable schema for responsiveness to reasons that are independent of truth-conduciveness.
The Realm of Epistemic Ends
We have seen that reasons are public and are keyed to community standards. This raises questions about the nature of the relevant community. Here again it pays to turn to Kant. One formulation of the categorical imperative is that an agent should act only on a maxim that he could accept as a legislating member of a realm of ends. These maxims are not just laws that the members of the realm of ends are subject to, they are laws that members make themselves subject to. In the moral realm, Kant (1981) maintains, agents are legislators who enact the laws that bind them. I suggest that the same holds in the epistemic realm. What gives second-order norms their epistemic authority is that they express standards, rules, or principles that epistemic agents can on reflection endorse. Extending Kant’s idea to epistemology, then, an epistemic agent should accept only considerations that she could advocate and endorse as a legislating member of a realm of epistemic ends. Let us call this the epistemic imperative. Although Kant’s argument for the categorical imperative is transcendental, my argument for its epistemic counterpart is pragmatic. It focuses on the practices of inquiry and seeks to delimit the aspects of those practices that figure in their epistemic esteemability. There is no claim that these aspects are metaphysically necessary.
If agents in a realm of epistemic ends make the laws that bind them, they are not merely subject to the requirements that govern their cognitive behavior; they set those requirements. Since epistemic agents would not make commitments that they consider it inappropriate to be bound by, they think it right that they be subject to the laws of the realm of epistemic ends. They reflectively endorse those laws and the commitments that flow from and are mandated by them. They think that their being bound by a particular constellation of norms fosters their epistemic success.
In making and reflectively endorsing commitments, an agent exercises her autonomy. Because she is an agent, she can both start and end a justificatory path.7 She considers herself justifiably bound by a network of commitments pertaining to methods, rules of inference, standards of acceptability, and the like, because she believes that being so bound will promote her epistemic ends, given her epistemic resources. The ends may, as the veritist maintains, be truths; they may be truths that satisfy further standards (for example, standards of significance or relevance); they may be nontruths (perhaps nonpropositional representations, effective models, informative idealizations, or illuminating fictions) that she takes to exemplify features of the phenomena; they may be more holistic desiderata pertaining to the system of commitments as a whole. The critical point is that agents set their ends. So even if an agent agrees with the veritist that the overarching end is truth, her relation to that end is different from his.
It might seem that reflective endorsement yields only subjective value. The agent sets epistemic ends for herself, makes epistemic commitments for herself, and on reflection is willing to stand behind them. At best this seems to assure a measure of internal consistency, but not much more. But Kant does not say that a maxim is acceptable only if an agent could enact it as the philosopher king of a realm of ends. Satisfying merely personal predilections is not enough. When an agent reflectively endorses an epistemic principle, she considers it reasonable that her cognitively serious actions—asserting, professing, and inferring—accord with that principle. She repudiates the gambler’s fallacy because she recognizes that in committing the fallacy she makes herself vulnerable to Dutch books—sets of commitments that are jointly incoherent in the strong sense that were a person to bet on all of them, he would inevitably lose. She endorses modus ponens because she recognizes that it is truth preserving. She has no reason to think that as an epistemic agent she should be subject to principles that other similarly situated epistemic agents are not subject to. She does not think, for example, that the desirability of avoiding Dutch books is a matter of personal preference, for she recognizes that commitments vulnerable to Dutch books are not cotenable. Because she takes the principles she reflectively endorses to be reasonable and rational in the epistemic circumstances, she thinks that they should be binding on similarly situated epistemic agents. But she recognizes that epistemic agents should be subject only to principles that they consider worthy of their reflective endorsement. By her own lights, then, the only principles that merit her reflective endorsement are those that similarly situated epistemic agents—those who constitute her epistemic community—could also endorse. If she is wrong to think that her compatriots can reflectively endorse p, she is, by her own lights, wrong to endorse p.
Her relation to other epistemic agents pays dividends. Their commitments serve as a check on hers. Confronted with what looks to be a camel on Main Street—or indeed, a car on Main Street—she can ask, ‘Did you see that?’ Arriving at a surprising solution for a complicated calculation, she can ask, ‘What did you get?’ Her interlocutors’ agreement serves as corroboration for the conclusion and for the adequacy of the first-order considerations that led to it. Their disagreement should give her pause.
For all I’ve said so far, the support compatriots provide is consonant with an epistemic individualism where each agent amasses and assesses her own reasons for a conclusion. She thinks that similarly situated agents will share her conclusions, because each has amassed the same evidence on his own and assessed it using the same standards. If her reasons are good reasons for a particular contention, her standards for assessing such things are appropriate, and she has applied them correctly to her reasons, it is no surprise that her similarly scrupulous compatriots agree. According to epistemic individualism, it is each epistemic agent’s standing in the same relation to the evidence, not each agent’s standing in any particular relation to other agents, that accounts for their agreement. I suggest, however, that epistemic interdependence runs deeper. It both underwrites certain judgments and grounds the norms that govern epistemic practice.
Compatriots’ responses can validate or invalidate the methods by which agents form and justify their opinions. For example, by himself, an agent cannot tell that he is color-blind. He can tell whether he is a consistent judge of color. The items that looked red to him yesterday look red to him again today; variations in the apparent colors of enduring objects correlate with variations lighting conditions; one visual deliverance as of color supports another. Even so, some items that consistently appear red to him might be green. People discover that they are color-blind by learning that other people discriminate shades that they cannot tell apart. If an agent cannot appeal to other people’s color perceptions to check his own, he has no way to tell whether his socks match. For all he knows, he might be color-blind. The same holds for the other secondary qualities. On his own, the epistemic agent does not know what he is missing.
Such reliance on others is not restricted to the sensory realm. Without the support of an epistemic community, an individual lacks the resources to discover certain sorts of systematic error, such as confirmation bias. If Life-long Crusoe (who was shipwrecked at birth and reared by wolves) is given to confirmation bias, he weighs evidence that supports his convictions more heavily than evidence that undermines them. He may have no way to discover his bias, particularly if its impact is slight. Nevertheless, many of his beliefs are fated to be unjustified. Suppose, however, that he is not prey to confirmation bias or kindred failings. Then he is, we may assume, as reliable as the rest of us, and his first-order evidence supports his conclusions as strongly as our first-order evidence supports ours. It might seem therefore that he is just as capable of epistemic success as we are. But we have a resource that he lacks. We are members of communities that embed and reinforce certain epistemic values. Confirmation bias involves believing or accepting that p on the basis of considerations that one wrongly takes to be sufficient. If, as is likely, the agent endorses her own basis, it is unlikely that her further scrutiny of that particular basis will disclose their inadequacy. But if she appeals to community standards and subjects her belief to (implicit or explicit) peer review, she can discover the error of her ways. She realizes that whatever she may think of her evidence, it will not pass muster with the relevant community of inquiry. To be sure, the fact that some community or other disagrees neither shows nor automatically gives her reason to suspect that she is wrong in the assessment of her evidence. But if she shares the values of that community, and recognizes that the opinion in question is properly subject to its standards, she is in a position to see that her belief does not measure up. The protection against epistemic failings like confirmation bias derives not from the mere fact that there is a community whose standards bear on her views, but from her taking her views to be rightly subject to the judgment of that community. She endorses the public standards by which her belief is to be judged. There need, of course, be no actual public assessment. Having internalized the standards she can invoke them to assess her own views. This is why the brilliant mathematician can accept his proof.
We assess our beliefs in light of the standards that an epistemic community has designed to (among other things) filter out failings like confirmation bias. So we have reason to think that beliefs that satisfy those standards are not products of such failings. Life-long Crusoe has no such reason. His reasons for confidence in his beliefs are thus epistemically impoverished compared with ours. His epistemic situation is precarious. It is just by luck that he has managed to avoid the pitfalls.8
It is no accident that Kant’s term ‘legislating members’ is plural. Legislators work together to enact laws. To be effective, they must convince their colleagues of the acceptability of the legislation they propose. This requires that the basis for their recommendation be publicly articulable and justifiable to other legislators in light of the evolving commitments and goals that they share. The Kantian realm of ends is not, as some translations have it, a kingdom; it is a commonwealth. So is the realm of epistemic ends. The reasons for acceptable epistemic commitments must be specifiable and justifiable to the other members of the epistemic community. Moreover, since the realm of epistemic ends is supposed to be the arena within which epistemic agents live their cognitive lives, those commitments must mesh. Not only must each be individually acceptable; all must be collectively acceptable. This means that there are strong consistency and coherence constraints on what epistemic legislators can endorse.
It does not follow that all members of an epistemic community are expected to agree about everything. Some commitments—for example, those pertaining to consistency and coherence, are binding on everyone. Others set ranges of permissibility. Any commitment within a given range, if suitably enmeshed with other permissible commitments, is worthy of reflective endorsement. There are, for example, multiple acceptable interpretations of Hamlet. An epistemic agent need neither doubt her interpretation nor assume her compatriot is mistaken if their opinions diverge, so long as both fall within the range of acceptable interpretations. Under some acceptable interpretations, Hamlet is about indecisiveness; under others it is about revenge. But even literary criticism’s tolerance for diversity is not inexhaustible. Under no acceptable interpretation is the play about the fall of the Roman Empire.
Such permissiveness is not restricted to the arts. Philip Kitcher (1990) argues that rather than setting a fixed threshold for epistemic acceptance, science rightly recognizes a range within which either accepting a new theory or retaining one’s commitment to the old one is reasonable. This enables the discipline to balance inertia with creativity. Scientists taking opposing positions need not disagree about what the evidence is, or about how much evidence there is. They may disagree about how to interpret the evidence, about what evidence is relevant, and/or about how much evidence is needed for acceptance. This is why, rather than saying that a responsible epistemic agent should accept only considerations her compatriots would accept, we should follow Scanlon (1998) and say that she should accept only considerations her compatriots could not, by their collective lights, responsibly reject. It is open to them to endorse a consideration if it falls within the purview of an epistemic community and, given the commitments of that community, they could not reasonably reject it.
The epistemic imperative directly vindicates many familiar epistemic virtues. Open-mindedness is one. An epistemic agent cannot reflectively endorse a contention unless she is in a position to think that other members of the epistemic community could endorse it as well. And she is in no position to responsibly think that unless she entertains the alternatives they might consider credible. So she must act open-mindedly in developing the contentions she advocates. Her open-mindedness is not boundless: it need not extend to entertaining hypotheses involving interventions by space aliens, since those hypotheses are not ones that her intellectual compatriots would, or by their own lights should, take seriously. Rigor, responsiveness to evidence, and impartiality receive similar treatment: the scope of the demand a virtue places on the agent is delimited by the epistemic community. Each agent stands in a reciprocal relation to other members of the community. Besides venturing only hypotheses that she considers worthy of their reflective endorsement, she must be in a position to responsibly reflectively endorse, reflectively suspend judgment on, or reflectively repudiate their hypotheses. She needs to be able to think and act impartially, fair-mindedly, and knowledgeably.
Other epistemic virtues are vindicated by what we have come to understand about our epistemic situation. As we increase our understanding of a topic, we understand more about our understanding of that topic. Object-level advances in understanding thus pay metalevel dividends. We devise new methods and design new instruments. We discover hitherto unforeseen limitations and unacknowledged biases that cast doubt on results generated by old methods and instruments. We uncover mistakes and diagnose their sources. Considerations that our predecessors reflectively endorsed no longer merit reflective endorsement. Considerations they would responsibly have rejected we are in a position to endorse. Two hundred years ago, physicists readily endorsed the contention that space is Euclidean and, had it been offered to them, would have adamantly refused to endorse the contention that mass increases with acceleration. Now we know better. Because the history of inquiry shows that advances in understanding often involve correcting the errors of our forebears, intellectual humility is a virtue; intellectual arrogance, a vice. An epistemic agent, seeing that other agents who satisfied the standards of their communities of inquiry turned out to be wrong, should recognize that even though she satisfies the standards of her community she too might be wrong. She cannot claim that her contentions are permanently acceptable, only that they are as reasonable as any available alternative in the current epistemic circumstances. In professing them and proffering them to the epistemic community, that is what she is willing to defend.
The Division of Labor
Who belongs to a realm of epistemic ends? What qualifies someone for membership? Kant might say that the community of epistemic ends consists of all rational agents throughout history. But little is separately justifiable to or reflectively endorsable by all rational agents. What an epistemic agent is in a position to reflectively endorse depends heavily on her cognitive milieu. The contemporary community of particle physicists reflectively endorses much that is counterintuitive or unintelligible to the rest of us. But it would be wrong to dismiss their epistemic achievements because of our ignorance or incompetence.
The resolution to this problem derives from the division of epistemic labor. Earlier I said that compliance with the epistemic imperative does not require vetting each candidate for epistemic acceptance individually. We routinely endorse deliverances from creditable sources—memory, perception, and so on. One such source is testimony. On her own, each epistemic agent is in a position to know very little. To compensate for our individual shortcomings, we rely on the testimony of others. There is a division of cognitive labor both within and across the communities constituting the realm of epistemic ends. The relevant communities need not be scientific. We defer to the trained auto mechanic for an explanation of the ominous rattle in the engine, to the knowledgeable historian for an understanding of the influence of the cotton gin on American slavery, to the experienced parent for insights into how to soothe a teething baby.
Still, there are communities and communities. We might be fairly sanguine about saying that satisfying community standards suffices for epistemic responsibility in particle physics or auto mechanics. But what about the community of psychics? Are we forced to say that if they have managed to contrive, reflectively endorse, and by their own lights satisfy standards for contentions about the paranormal, we should credit their claims? We are not. For outsiders are in a position to assess their standards, methods, and claims, using coarse-grained measures that apply across realms.
One is predictive success. Not every discipline makes predictions; but those that do are subject to assessment on the basis of their track records. For a predictive practice to be epistemically creditable, its predictions should be borne out considerably more often than chance, considerably more often than what one would expect if its distinctive causal commitments were dropped, and should compare favorably with the track records of its rivals.9 The predictions must be definite enough that it is possible to tell whether they are borne out. And it must be determinate whether their being borne out redounds to the credit of the putative predictor. Psychics make predictions. If the standards of their community are satisfied by predictions that are too vague to be tested, are not borne out when tested, or simply replicate with what would otherwise be expected, then satisfying their standards does not make for epistemic credibility. When Professor Trelawny divines that Harry Potter has a mortal enemy, Hermione protests, “Everybody knows that!” (Rowling, 1999, 106). Coming up with that insight is no reason to credit divination. If we are to trust divination, Hermione rightly intimates, it should disclose things that we would not otherwise have known.
Similar points can be made about explanatory success, systematicity, fit with the commitments of neighboring areas of inquiry, and so forth. The familiar list of theoretical and empirical desiderata can be brought to bear to assess the claims of a putatively cognitive enterprise. Not only must the commitments be internally coherent and consistent, they must cohere with and be consistent with other things we reflectively endorse. If the discipline is empirical, claims need to be backed by evidence. If the practice involves prediction, the predictions must (often enough) be borne out. These are entirely familiar requirements. The commonwealth of epistemic ends is a federation. Local communities make fine-grained commitments that are answerable to coarse-grained commitments that are more widely shared. The failure of a local community’s fine-grained commitments to satisfy (or at least approximate) coarse-grained requirements, unless backed by strong reasons to think that the fine-grained commitments need not satisfy them, is a reason to refuse to reflectively endorse their findings.
What justifies treating particle physics as a community in a federation of epistemic ends is that its commitments satisfy widely shared coarse-grained epistemic commitments. Particle physics respects such requirements as that a scientific account be explanatory, simple, perhaps predictive; that it be responsive to evidence that is judiciously and impartially gathered and assessed; that its results be intersubjectively observable and reproducible; that the account cohere, or at least not be incompatible, with accepted constellations in neighboring disciplines; that its findings stand up under peer review; and so on. Just how these requirements are concretized in a given discipline or subdiscipline depends on local factors; and it may be that only the members of a local community can tell whether this has been done. But it is critical that merely meeting local standards is not enough. The constellation of commitments that meets local standards is, as a whole, answerable to more global standards.
This pattern is reminiscent of what we see in law. Local parking ordinances are justified and binding in a community largely because they were duly enacted. But they are subject to broader constraints—those specified by state and federal laws, which in turn are subject to constitutional constraints. Regardless of what a particular city council enacted, a law specifying different parking fines for members of different races who commit the same infraction would be impermissible.
The Political Character of the Epistemic Community
To perform their function, legislating members of a realm of epistemic ends must be, in a suitable—political—sense, free and equal. Agreement among free and equal inquirers enhances the epistemic standing of a claim; coerced agreement does not. If inquirers are free, they can adopt any perspective they like, and examine the issue from there. If they are equal, all inquirers have an equal opportunity and an equal right to venture hypotheses, to raise objections and counterhypotheses, and provide reasons for them. It is not enough that members be free to venture their opinions. Their interlocutors must also be disposed to listen to them and take them seriously (Longino, 1990).
Members of a properly functioning epistemic community see those who disagree with them not as rivals to be bested in argument, but as resources whose perspectives may provide valuable insights (Dewey, 1916). By appreciating one another’s points of view, deliberators enrich their understanding of an issue and the available initially tenable ways of thinking about it. Entertaining a new perspective may convince an agent to adopt it. She may see that her original assessment of the evidence or the options was limited or biased. Alternatively, such deliberation may lead to the adoption of a conclusion that no one initially envisioned. That the evidence admits of multiple interpretations may reveal that none by itself is adequate and point toward a different solution. By accessing the opinions of others and the reasons for those opinions, epistemic agents augment their data base. They gain new information about the situation and ways it might be fruitful to think about it. So rather than seeing those who disagree with us as opponents, we should see them as allies who, by envisioning things differently, extend our epistemic range. Diversity of opinion thus is not an impediment to deliberation but an asset to it.
On this conception, a community of inquiry provides a venue for fruitful disagreement. Not every disagreement is worth taking seriously. There is little to be gained by entertaining the economic views of someone convinced that extraterrestrial forces control the stock market. But when conscientious, competent peers with a suitable level of expertise about a topic disagree, the situation is different.
When parties disagree, it might seem obvious that at least one of them is mistaken. Either he overlooked some relevant evidence or erred in his use of it. These are typically identifiable and corrigible errors. Once the reasons are spelled out, it may be straightforward to see where the error lies. One party miscalculated or ignored base rates or overlooked data or whatever. Not all disagreements are like this. Sometimes the root of a disagreement is not an error; nor is it something that needs to be corrected. Parties to a disagreement might assign different (reasonable) weights to different bits of evidence. Fred might, for example, take a blood test to show that the patient’s hemoglobin count is dangerously low. George, placing greater emphasis on the trajectory, notes that this count is higher than the previous reading. Fred believes that the patient is unlikely to survive. George believes he is on the mend. Both agree about the facts. Their disagreement concerns which fact is more significant. A disagreement about a matter of fact, then, can be rooted in a disagreement about how to weigh the evidence.
Disagreements can also arise when parties consider different factors salient. When an anorexic patient exhibits a dangerous tachycardia after consuming high levels of caffeine, the emergency room physician judges that excessive caffeine consumption is the cause (Ankeny, 2014). He might overlook or downplay her anorexia, concentrating entirely on the fact that she has overdosed on caffeine. The expert in eating disorders judges that anorexia is the cause, since anorexia gives rise to tachycardia in the presence of a suitable stimulant. Each thinks something true. The difference of opinion lies in the link in the causal chain that they focus on. Their choice of focus stems from what each can do to alleviate the problem and prevent its recurrence. Here the difference is within the field of medicine. But similar disagreements can cut across fields. On a rainy night, a car crashes into a tree. The police say that speed caused the crash; the auto engineer says that the unresponsiveness of the steering column was the cause; the highway architect places the blame on the way the road was banked. In such cases, no one is saying anything false or misleading. Given their different interests and areas of expertise, they simply focus on different aspects of the situation.
Jamesian veritists hold that our overarching goal is to believe as many truths as possible and disbelieve as many falsehoods as possible. I have argued that this ought not be our overarching goal. Nevertheless, where we are concerned with what to believe, we face a trade-off between the value of believing truths and the disvalue of believing falsehoods. To a remarkable (and alarming) extent, advocates of Jamesianism ignore the trade-off and focus exclusively on believing truths. But the goal as stated involves a proportion. To settle on a criterion of justification requires weighing the values against each other (see Riggs, 2003). There is no reason to think there is a uniquely best way to assign weights. If the threshold for acceptability is a probability of 0.90 on the evidence, a few false hypotheses will count as justified. If the probability has to be 0.95, even fewer will count. On the other hand, when we raise the standard from 0.90 to 0.95, fewer truths will count as justified. In moving from 0.90 to 0.95, we accept fewer falsehoods and reject more truths. Setting the threshold involves a trade-off. It is widely and rightly conceded that for a hypothesis to count as justified, it must be highly probable given the evidence. That, by itself, does not tell us where to draw the line. Nor does it commit us to the claim that there is a single, optimally correct line to be drawn. A disagreement among experts might be due to differences in their tolerance for epistemic risk. They could conceivably agree about what the evidence is, what weight to attach to the evidence, and which factors are and should be salient, but still disagree because one holds and the other denies that a probability of 0.92 is justification enough.
Where does this leave us? We’ve seen that the opinions of epistemically responsible agents can diverge along a variety of trajectories. Divergence in and of itself does not show that any of the parties to a dispute is mistaken. Disagreements among experts—those who are competent, conscientious, and suitably sensitive to the level of confidence that their evidence supplies—provide information about the community’s current understanding of an issue.
A community of inquiry may have resources to adjudicate some disagreements among its members. It may currently lack resources to adjudicate others. In that case, the disagreement reveals something about the lacunae in its understanding of its subject matter. It may also be in a position to recognize that some legitimate disagreements need no adjudication. Kitcher’s point is that the community may best serve its collective epistemic ends by, at any given time, countenancing a range of conflicting views. The reason is largely prudential. When there is a significant chance that a currently disfavored view is true, or that a currently favored one is false, if the members of the community want to believe what is true, they ought not foreclose inquiry prematurely. Riggs’s point is that multiple ways of balancing the value of believing truths against the disvalue of believing falsehoods are permissible. This has two important consequences. The first is that there may be no general answer to the question what balance should be struck. For some issues, avoiding falsehood may be more important; for others, believing truth might dominate. How risk averse epistemic agents should be, then, depends on what they are risking. The risk in question may be practical, as in risking an epidemic by accepting false views about contagion, or purely theoretical, as in risking an epistemically costly mistake, such as one that is likely to prompt a cascade of false beliefs. Moreover, even in the long run, there may be no objective resolution to some disputes. There is no reason to think that there is a single optimal balance.
The value of disagreement for a community of inquiry is that it can serve as a source of insight into the subject under investigation and the available resources for understanding it. Socrates maintained that knowing that one does not know is epistemically valuable. What may be more valuable is recognizing what one does not know and understanding why one does not know it. Suitably expert disagreement can engender such understanding.
Deweyan Deliberation
If free and equal members of an epistemic community engage in Deweyan deliberation, there are no political impediments to understanding the phenomena. But if consensus results from coercion, collusion, or credibility inflation or deflation, the situation is different. If inquirers cannot investigate an issue as they see fit, if they cannot raise objections or gain a hearing for them, if their views are given undue weight or are deprived of the weight they are due, intersubjective agreement provides small reason to think that a conclusion is acceptable, or that it stands up to serious testing.
Arrangements that silence certain voices or deflate their credibility can undermine the epistemic standing of the commitments that are accepted (see M. Fricker, 2007; McGowan, 2009). By omitting or downplaying the significance of particular perspectives, or by unduly emphasizing other perspectives, prejudice and stubbornness can skew matters so that intersubjective agreement does not supply corroboration. If naysayers are silenced, then the fact that every speaker agrees is not a good reason to think that a contention is acceptable or a method is sound. If the credibility of naysayers’ testimony is deflated, then the fact that the balance of what is taken as evidence weighs heavily in favor of a hypothesis is at best a weak reason to accept it. The negative epistemic consequences do not just undermine the epistemic standing (and self-confidence) of those who have been silenced. They affect the entire epistemic community. Without access to the discredited opinions, community members do not know what they are missing. Moreover, the skewing of the evidence that results from credibility deflation undermines the trustworthiness of what evidence they have, for it artificially inflates the credibility of the voices that are heard. Such a situation is epistemically unjust.
An example brings this out. In the 1940s and early 1950s Barbara McClintock published a series of papers contending that she had discovered sequences of genetic material that change position on the chromosome of corn. She was not taken seriously. As a result, she said, “I stopped publishing detailed reports … when I realized, and acutely, the extent of disinterest and lack of confidence in the conclusions I was drawing from my studies” (McClintock, 1973). Historians of science disagree about whether sexism figured in the scientific community’s disregard of McClintock’s findings. The other candidate is a prejudice against studies done on corn (see Keller, 1983; Comfort, 2001). Whatever the reason, she was effectively silenced. Her discovery was not taken up until the 1960s when François Jacob and Jacques Monod discovered that the same transposition occurs in bacteria. Plainly the understanding of genetics was retarded by the failure to give McClintock’s discoveries their due. Such epistemic injustice deprives the community of inquiry of data and skews the evidence.
‘Free and equal,’ here, is a political characterization of a community of inquiry. That community may be located in a wider setting whose members are politically neither free nor equal. A repressive society might, for example, accord greater freedom of thought to its scientists in the context of their investigations than it generally accords to its citizens. If the constraints the society normally places on its citizens are not operative in the context of inquiry, the epistemic commitments inquirers reflectively endorse are legitimately normative for their inquiry, hence for the nonce binding on their practice.10
Even so, the idea that the members of communities of inquiry are free and equal inquirers is an idealization. I do not claim that actual communities fully accord with it. I do, however, claim that the closer actual communities come to satisfying these requirements, the more epistemically acceptable their criteria are likely to be. This is because we understand how inequalities of power and limitations on orientation can lead to error. By reflecting on the strengths and weaknesses in the ways the actual communities of inquiry formulate their standards and construct their practices, we contrive a conception of what is epistemically acceptable. The requirement of freedom and equality thus emerges from the method of reflective equilibrium. We draw on, correct, extend, and emend our current best theories and practices until we arrive at a system we can on reflection endorse.
The free and equal requirement is merely a political requirement. It is not a claim that all are equally knowledgeable or equally intelligent. Some hypotheses are plainly untenable; some perspectives are obviously skewed; some methods are demonstrably unsound. Once they are given a hearing, they are promptly and rightly dismissed. Nevertheless, the opportunity to venture a hypothesis and the right to have it assessed on its merits must be real. There must be forums where criticisms can be offered, and where there is an expectation that criticisms will be answered (see Longino, 1990). As Mill insists, “The beliefs we have the most warrant for have no safeguard to rest on but a standing invitation to the whole world to prove them unfounded” (1978, 20). The epistemic value of what Habermas (2001) calls uncoerced conversation lies, I suggest, not so much in the particular agreements it generates as in its propensity to uncover and correct errors.
Since ‘ought’ implies ‘can’, the standards the responsible epistemic agent must respect are not those of some idealized, timeless community of rational agents. She could have no idea what those standards are or what it would take to satisfy them. Rather, she must respect the current or reasonably foreseeable standards of a realm of epistemic ends she inhabits. Just what these standards are is not determined a priori. They emerge over the course of inquiry, as the community learns about the subject matter, how to investigate the subject matter, and what sort of understanding of the subject matter is practically possible in the epistemic circumstances, given the resources at hand, and the questions the community wants to answer. Respecting the standards is not the same as being obliged to comply with them. Rather, respect requires that a responsible epistemic agent either comply with the standards of her epistemic community or supply a cogent reason to revise them, reject them, or restrict their range of applicability.
Nevertheless, the extent to which we have to rely on one another might seem to undercut any claim to epistemic autonomy. Often, it seems, we amateurs are in no position to defend our own beliefs, as they depend on fine-grained epistemic commitments that we are not privy to and would probably not understand even if we were. We defer to the community of particle physicists to specify what counts as evidence for the existence of the Higgs boson, what counts as sufficient evidence, and when such evidence is in hand. We have no choice. Nevertheless, epistemic commitments must be defensible. If my defense of the contention that the Higgs boson exists consists in saying, ‘They said so!’, it looks weak.
It is not. Although expertise is distributed across the epistemic community, this does not undermine the responsibility of the individual epistemic agent. She confers epistemic authority on those she counts as experts; and she retains both the right and the responsibility to revoke it if the circumstances warrant. She can and at some point should decide that the auto mechanic does not know what he is talking about (even though she does not know what a catalytic converter does), that the investment counselor’s advice is not credible (even though she is only dimly aware of what a hedge fund is), that the physician who links autism with vaccines is untrustworthy (even though she has no idea why the incidence of autism is rising). The requirement that autonomous epistemic agents think for themselves is not in tension with the recognition of the extent of our epistemic interdependence. Rather, it means that each agent is ultimately responsible for her choice of experts. As a legislating member of a realm of ends, she confers epistemic authority, thinking that other similarly situated epistemic agents would do the same; and she can and should retract that authority when she considers it no longer warranted. If a contention’s defense consists of ‘They say so!’ backed by defensible reasons for thinking that their saying so is trustworthy, it is not a weak defense; it is a strong exercise of epistemic autonomy.
I have argued that epistemic norms are deontological. They are grounded in the epistemic imperative. An epistemic subject ought to accept only what she could reflectively endorse as a legislating member of a realm of epistemic ends—that is, as a responsible citizen/legislator of a united federation of autonomous epistemic agents. To abide by the imperative requires standing not just in a suitable relation to the phenomena she seeks to know or understand, but also in a suitable relation to other members of the epistemic community. On this view epistemic norms do not, nor do they purport to, ensure that epistemic agents will arrive at the truth. Nor do they contend that failure to reach the truth is always a mark of epistemic inadequacy. But by acting in accord with the epistemic imperative, an agent makes the best use of available epistemic resources.
Epistemic standards owe their normative status to their satisfying the epistemic imperative: they are standards that would be framed and reflectively endorsed by the legislating members of a realm of epistemic ends. Although a realm of epistemic ends is modeled on a scientific community, it obviously involves an element of idealization. As the McClintock example illustrates, we should not assume that actual communities of inquiry consist of free and equal inquirers. Rather than describing the norms accepted or acted upon by an actual community, the ideal emerges from a process of adjudication in which we elaborate, extend, and correct our relevant commitments until we arrive at a conception that is in reflective equilibrium with our other epistemic commitments.
The actual communities we respect serve as touchstones. We treat their policies and practices as initially tenable and attempt as far as is feasible to incorporate them into the constructed ideal. The scientific community is one we respect. It regularly deploys felicitous falsehoods—models, idealizations, and thought experiments. The emendations we saw fit to introduce into our ideal correct for bias and prejudice. They do not undermine the permissibility of relying on felicitous falsehoods. So, absent independent reasons to consider reliance on such devices epistemically deleterious, we should recognize that scientific (and other disciplinary) communities are within their epistemic rights to deploy them.
Notes
1. This chapter was made possible through the support of a grant from the Intellectual Humility Project at St. Louis University, the Varieties of Understanding Project at Fordham University, and the John Templeton Foundation. The opinions expressed here are those of the author and do not necessarily reflect the views of the Intellectual Humility Project, the Varieties of Understanding Project, or the John Templeton Foundation.
2. Agents are responsible for actions they perform when drunk or under the influence of recreational drugs. As Aristotle (1985) argues, even if they are out of control when they perform the particular actions in question, because they were in control when they imbibed to the point where they lost control, they are responsible for their drunken behavior. And in general, if x ought to φ, then x is ceteris paribus obliged to refrain from putting herself in a position where she cannot φ. The bearing of this point on my argument is this: if someone irresponsibly neglects to acquire or exercise an epistemic virtue, he can be held responsible for the epistemic defects that result.
3. Such heteronomy is character of what Sosa (2007) calls animal knowledge. It is plausible that animals cannot reflectively endorse or criticize their attitudes, and are subjectively vulnerable in just the way I suggest. On the other hand, they presumably do not go in for, and do not need to go in for, self-reflection. So the vulnerability is not a subjective problem for them.
4. I think that epistemic agents can hone their perceptual, representational, and imaginative capacities. So they are not entirely passive even with respect to inputs. But I will not argue that here.
5. I am grateful to an anonymous reviewer for this example.
6. I am grateful to Michael Ashooh for this example.
7. I am grateful to Jonathan Adler for this point.
8. I am grateful to Alvin Goldman and Sanford Goldberg for pressing me to clarify this point.
9. In the cases Kitcher is concerned with, rival theories might be approximately equal in predictive success, and/or each might best the other across some part of the domain. Then their predictive track records compare favorably with one another. Moreover, since scientific acceptability involves multiple desiderata, there are trade-offs. A measure of predictive success might be sacrificed for an increase in precision or scope. In such cases, the practices as wholes compare favorably with one another.
10. I am grateful to Thomas Kelly for raising this issue.