4 The Fabric of Understanding
In the previous chapter, I argued that disciplinary understanding—indeed, any sort of systematic understanding—is objectual. Here I argue that it follows that accounts must be assessed holistically. Rather than asking whether each component sentence expresses a fact, we should ask whether the account as a whole is in reflective equilibrium. If so, I will urge, it is acceptable in the epistemic circumstances. After explicating reflective equilibrium I show how adopting it as a standard of acceptability enables us to appreciate the way individually weak considerations can weave together to make a strong case. I then consider objections. Although my theory is not a pure coherence theory, coherence plays a major role. I discuss familiar worries about coherence theories and explain why they do not undermine my position.
Reflective Equilibrium
The move to holism raises questions about justification. After several hundred years of theorizing, it is fairly easy to see why empirical evidence for a hypothesis is truth conducive. Relevant empirical evidence increases the probability that the hypothesis is true. If an account is construed as a long conjunction whose conjuncts are statements of its commitments, it might seem that the same holds of it. The total evidence increases the probability that the long conjunction is true, thereby justifying the conjunction. The difficulty is that under such a construal, even the best accounts turn out to be unjustified. If p and q are evidentially independent, and each has a probability of less than 1, the probability of the conjunction p & q is less than the probability of each conjunct. The longer the conjunction, the lower the probability. Comprehensiveness is considered an epistemic virtue, but if justification consists in probability raising through empirical evidence, it is not. Holism ought not then contend that the justification of the wholes derives from the justification of the parts. Or at least it ought not contend that the probabilistic justification of the parts increases the probability that the whole is justified as well.
The challenge, then, is to explain how systematic interconnections give rise to justification—how the fact that considerations dovetail affords reason to accept them. In previous works I argued that a system of cognitive commitments in reflective equilibrium is epistemically tenable (Elgin, 1996, 2011). This is a procedural account of epistemic tenability. An account is tenable just in case it is, or is rationally reconstructable as, a result of a process of adjudication that brings a collection of initially tenable commitments into reflective equilibrium. It is as reasonable as any available alternative in light of those initially tenable commitments, and the commitments that comprise the account are in general more reasonable in light of one another than they would be alone. I do not claim that reflective equilibrium shows an account to be true or even to probably be true. I claim only that a system of epistemic commitments in reflective equilibrium is the best we can do in the epistemic circumstances. But that is enough to make it acceptable right now. This needs to be spelled out.
Inquiry, as Quine insists, always begins in medias res. We start with opinions, values, methods, and standards that we consider relevant and that we are inclined to credit. Although we recognize that they are less than wholly satisfactory as they stand, they comprise our current best take on the matter under investigation. These commitments can include hunches, rules of thumb, superstitions, and old wives’ tales as well as confirmed generalizations, proven theorems, firmly established principles, and solid evidence. The only requirement at the outset is that we have some inclination to accept them. Because we do, they are initially tenable. Clearly that inclination does not endow them with a high level of tenability. To be initially tenable is not to be presumptively acceptable, or prima facie acceptable, or ceteris paribus acceptable. It is merely to have some very slight measure of epistemic standing. If a consideration is initially tenable, we need a reason to give it up. But since initial tenability is weak and precarious, even a weak reason may suffice to abandon it. So construing a consideration as initially tenable is not saying much in its favor. Reasons to give up initially tenable commitments are legion.
That being so, why should we think that the mere fact that we have some inclination to accept a consideration gives it any level of epistemic standing? This much can be said in its favor. Since beliefs form the basis for action, the success of our actions affords some evidence that the beliefs that underwrite them are true enough. And the persistence of the inclination, however slight that inclination, is evidence that the consideration has not unduly impeded our actions. Granted, brute luck may explain why certain untoward initially tenable commitments have not yet caused our efforts to go awry. Or the untoward commitments may bear on factors that are too distant from day-to-day experience for the success or failure of our actions to make much difference to their continued acceptance. Or we may have made compensating errors. Nevertheless, that they have not caused enough trouble to induce us to abandon them is at least a minuscule mark in their favor. Moreover, we learn from experience. When we learn that premonitions tend not to be borne out, we cease to credit them. Although we may continue to experience feelings of foreboding, once we lose any inclination to credit their deliverances, they cease to be initially tenable. The fact that we have not yet had reason to give a commitment up is then a small reason to accept it.
Our initially tenable commitments pertaining to a topic may at the outset constitute a motley collection; they include our current convictions about the topic, the methods for investigating it, the standards to which a cogent account of that topic should be held, and the cognitive ends such an account should serve. Taken together they are apt to leave relevant questions unanswered and relevant problems unsolved. The elements of the collection may be in tension with one another. Our convictions may fail to satisfy the standards we consider them subject to. We may have incompatible or noncotenable convictions and/or mutually unsatisfiable cognitive desiderata. Considerations that seem independently plausible may seem jointly implausible. To achieve an acceptable account, we correct, reject, refine, and extend the commitments we started with. Some, such as ‘Fish have gills’, may come to be seen as crude approximations. Others, such as ‘Objects falling near the Earth fall at a rate of 32 ft/sec2’, turn out to be acceptable only within certain limits—for example, where air resistance is negligible. Yet others turn out, like Frege’s Basic Law V, to be hopeless, regardless of their antecedent plausibility.
Nor is our task just to corral and tame our initially tenable commitments. For even duly pruned and connected, a constellation of initially tenable commitments is apt to be gappy. Achieving reflective equilibrium often requires integrating into our emerging account commitments that we have no independent inclination to believe. The scientific community was originally committed to the existence of positrons not because it had any direct evidence of them, but because it was strongly committed to symmetry principles and to the existence of electrons. If electrons exist and symmetry holds, then there exist positively charged counterparts to electrons—that is, positrons. The relation between the embedding account and the positron hypothesis is one of mutual support. The hypothesis gains tenability from its place in the account; the account increases in tenability as well. Given the other commitments, the asymmetry that would result from denying the electron a positively charged counterpart would be an irregularity that required explanation that physics was in no position to provide. By rejecting, correcting, revising, and augmenting, we bring commitments into accord. Because the elements of the resulting account are reasonable in light of one another, they are in equilibrium; because the account as a whole is as reasonable as any available alternative in light of the relevant antecedent commitments, its equilibrium is reflective. It is an account that the community of inquiry can, on reflection, endorse.
Initially tenable commitments perform two functions. Being, at the outset, our current best take on how matters stand and on how to figure out, measure, and assess how they stand, they provide the raw material out of which to construct tenable accounts. They capture what we start out thinking we understand about the domain. In constructing a better account, we leverage that understanding. Their mutual support makes for equilibrium. By playing a second role, initially tenable commitments ensure that the equilibrium is reflective. They serve as touchstones against which to assess revisions.
Their capacity to play the second role may seem doubtful. I conceded that the initial commitments are to one degree or another inadequate. To use them as a touchstone for judging revisions may therefore appear misguided. How can an account be vindicated by being shown to be reasonable in light of considerations that are acknowledged to be unsatisfactory? Because our initial commitments constitute our previous best guesses about the topic, we use those commitments as a threshold for assessment. The account we arrive at should be recognizable as an improvement upon them. We engage in inquiry because we suspect that they are inadequate and in need of correction, fortification, elaboration, and/or extension. We revise our previous commitments when we arrive at something we consider better. The requirement that an acceptable account answer to our initial commitments is not a requirement that it incorporate or validate them. But where it does not, it should, at least for the most part, show why they seemed as reasonable as they did when they did.
Epistemic advancement, then, consists in improving on the commitments we currently hold, where improvement itself must be measured by current standards. We are in no position to adopt a God’s eye view and say that in absolute terms one account is better than another. But we do not need a God’s eye view to characterize cognitive changes as improvements. We can recognize, for example, that Newton’s account of planetary motion is an improvement over Kepler’s because Kepler’s account accommodates only the relation between the Earth and the sun, while Newton’s also accommodates the gravitational attraction of one planet on another. We appreciate that having acknowledged the significance of the sun’s gravitation, we have no grounds for ignoring other sources of gravity.
The method of reflective equilibrium is dialectical and its results are provisional. We judge our commitments and potential revisions of our commitments against one another to see which combination seems best on balance. But we also judge our criteria for being best on balance by seeing if they yield verdicts that we can on reflection endorse. Reflective equilibrium can be upset by new findings. However well it fits with our other ornithological commitments, the conviction that birds cannot hover is discredited by the discovery that hummingbirds hover. (We might accommodate the discovery by abandoning the conviction that birds cannot hover or by abandoning the conviction that hummingbirds are birds. But something’s got to give.) No account or component of an account is irrevocable, but an account in reflective equilibrium is acceptable until problems emerge or improvements are envisioned. The justification for this verdict lies in the fact that the commitments that comprise such an account are collectively at least as reasonable as any available alternatives in the epistemic circumstances.
The requirement that each element be more reasonable in light of the others than it was alone may seem too demanding. After all, one might think, when an observer sees a camel before her very eyes, in good light, in the middle distance, and uncamouflaged, her belief that she is looking at a camel needs no further support. I agree that her perceptual deliverance alone yields a presumption that she sees a camel. That she is actually looking at a camel is initially tenable. But that deliverance does not stand alone. It gains or loses backing from collateral commitments. If the perceiver harbors an account (even a Moorean commonsensical account) that includes the beliefs that camels are visible and that organisms like her are capable of seeing them, her perceptual belief is more reasonable than it would be if she had no views about the visibility of camels or if she thought that they were, like proverbial pink elephants, figments of the imagination that appear only to the inebriated. Her belief is strengthened if backed by a plausible reason for why there might be a camel in front of her. Is she at the zoo, at a circus, wandering the Sahara? If not—if, for example, her experience is as of a camel on Main Street—she might need to discredit the hypothesis that her apparent camel sighting is a hallucination or a dream. If she has no reason to think that she is taking mind-altering drugs or is under posthypnotic suggestion, the absence of such competing potential explanations contributes to her warrant. But if she has no idea whether camels are visible, whether she has been taking mind-altering drugs, whether she has recently been hypnotized, or whether human beings can see, her belief that she sees a camel is epistemically precarious. Much of the support we depend on in cases where things seem obvious is so deeply tacit and uncontroversial that we may be unaware of it (Adler, 2002). But that tacit background contributes to the tenability of even the most seemingly obvious deliverances.
Some elements of a tenable account have no independent presumption in their favor. They gain all their support from their place in the account they figure in. Had physics abandoned or attenuated its commitment to electrons or to symmetry, early twentieth-century physicists would have had no reason to believe that positrons exist. Similarly, Robinson Crusoe, seeing what he takes to be human footprints in the sand, concludes that he is not alone on the island, basing his conclusion on the conviction that the presence of what look to be footprints is a strong indication of the recent presence of human feet. If he discovered that the configuration that he took to be human footprints could as easily be produced by local sand crabs, his belief would lose its support.
Mutual Support
A critical question is how reflective equilibrium provides justification. Why should we think that mutual accord among our commitments, backed by a tie to our previous commitments affords any indication of how things are? Perhaps this will become clearer if we consider a case.
Yesterday Meg’s Latin book was stolen from her school locker. Three students may have witnessed the theft. None of them is very reliable. Anne is given to proving theorems in her head and tends to be oblivious to her surroundings when preoccupied with a tricky proof. To compensate for her habitual distractedness, she draws plausible inferences about mundane events, and often fails to register whether an opinion is due to observation or to such an inference. Being rather vain, Ben frequently does not wear his glasses. Like Anne, he draws plausible inferences about events around him, and tends not to remember having done so. Chauncy is simply a liar. Presumably he knows when he is speaking sincerely, but given the fluency and frequency of his lies, virtually nothing he says is trustworthy. Not surprisingly, the social circles of the three students do not intersect; none would deign to speak to the others. When questioned about the theft, Anne and Ben report what they think they saw, but confess that they are not sure what they actually witnessed and what they inferred. Chauncy insists that his report is accurate, but in view of his record, his claim is suspect.
Individually, none of the reports would count for much. Had only one of the putative witnesses been present, the most we could reasonably conclude would be that the thief might fit the description. But all three reports agree, and agree in alleging that the thief had an unusual appearance: he had spiked green hair. This makes a difference. Even though individually each report is dubious, and the probability of a green-haired textbook thief is low, the fact that the three reports provide the same antecedently improbable description inclines us to believe it. Their accord evidently enhances the epistemic standing of the individual reports (C. I. Lewis, 1946, 346). We have more reason to believe each of them in light of the others than we have to believe them separately. The question is why? How can multiple statements, none of which is tenable on its own, coalesce to yield a tenable conclusion? How can their relation to other less than tenable claims enhance their tenability?
Given the unreliability of the witnesses, we would not be surprised if they were wrong about the thief. But we would not expect them to all be wrong in the same way. Their agreeing needs an explanation. Were they coconspirators, the explanation would be straightforward: They plotted to tell the same tale. But not being on speaking terms with one another, they are unlikely to be in cahoots. If the descriptions they provided fit a relevant stereotype, a penchant for plausibility might explain their accord. But someone with green spiked hair is far from anyone’s stereotype of a textbook thief. So despite Anne’s and Ben’s propensity to draw inferences based on plausibility, their descriptions of the thief seem not to result from such an inference. Evidently the best explanation of the agreement is that the reports are accurate (Lipton, 2004).
It is not just our ability to exclude obvious alternatives that leads us to credit the allegation. A variety of collateral considerations support it. Some bear directly, albeit weakly, on the content of the claim. The principal’s secretary dimly recalls seeing an odd-looking stranger lurking in the hallway, but cannot remember what made the stranger look odd. The custodian thinks he recently saw a container of hair dye in the trash, but cannot remember which day. Although the tentativeness of these reports makes them less than wholly creditable, they are suggestive enough to buttress the eyewitness testimony. Other collateral considerations concern the witnesses and their circumstances. Book thefts are observable events, so there is nothing inherently dubious about a claim to have seen someone steal a book. The light and sight lines were such that the witnesses could have seen what they report from where they said they were standing. The witnesses, teenagers themselves, are adept at recognizing furtive adolescent behavior. None was subject to psychological experiments with implanted memories. None was on drugs. And so on. Separately, these factors count for little. Either their credibility is low, their bearing slight, or their evidential support minimal. But they coalesce to make a solid case. This suggests that the epistemic tenability of the several reports and the conclusion they sanction derives from their mutual supportiveness.
Although our focus is on the status of the allegation, it is really the account as a whole that is or is not acceptable. Many of the justificatory relations are reciprocal. The allegation is acceptable only if (at least much of) the rest of the constellation of supporting considerations is. Since the eyewitnesses are individually unreliable, and the contentions of the collateral witnesses or the bearing of those contentions on the theft is tenuous, the acceptability of the testimony likewise depends on the acceptability of the allegation. The epistemic status of the allegation is inseparable from the status of the rest of the story. Some of the background information may be separately secured, but to a considerable extent, the various components of the story stand or fall together. Epistemological holism contends that the primary unit of acceptability is an account comprised of mutually supportive elements. The acceptability of individual elements largely derives from the acceptability of the account they belong to or follow from. This suggests that epistemic justification is at base a property of a suitably comprehensive, coherent account, when the best explanation of its coherence is that the account is at least roughly correct. The epistemic justification of individual claims derives from their figuring in such an account.
Coherence evidently plays a major role. Although there is no universally accepted criterion of coherence, at least this is required: the components of a coherent account must be mutually consistent, cotenable, and supportive. Logical consistency is not enough. Two claims can be consistent with one another because they are mutually irrelevant. ‘Napoleon died in exile’ and ‘Penguins cannot fly’ are consistent with one another, but neither lends support to the other. There is no more reason to accept their conjunction than there is to accept the separate conjuncts. Since both cotenability and supportiveness are matters of degree, coherence is too. So if it can be shown that epistemic justification requires coherence, there remains the question of how coherent an account must be in order for it to be epistemically justified. Before facing that worry, though, other challenges need to be met. At least two worries immediately arise. The first is that coherence is too demanding an epistemic requirement. The second is that it is not demanding enough.
Even where we take ourselves to be on solid ground, contravening considerations are not uncommon. Mrs. Abercrombie, the aging geometry teacher, says that during the relevant period she saw a young man sporting a green hat. A green hat is not green hair, so strictly speaking her report is in tension with the reports of the other putative witnesses. Mr. Polk, the hall monitor, insists that no one was in the corridor at the time of the alleged theft. Mr. Miller, the classics teacher, disputes the allegation on the strong sociological grounds that students do not steal Latin books; they shun them. These reports are clearly relevant to and at odds with the emerging account of the theft. If we incorporate them into it, we render it incoherent. But they are initially tenable, so we have no legitimate reason to disregard them. The problem is this: The discussion so far suggests that the credibility of the various claims comprising an account depends on how well they hang together. If so, the failure of other, equally relevant information to cohere with them threatens to discredit the account.
Although true, this is not so daunting as it appears. The immediate threat of incoherence comes from assuming that we must take seemingly contravening considerations at face value, incorporating them into the account of the theft as they stand. We need do no such thing. Rather, we assess contravening considerations just as we do the rest of our evidence. Recall that we did not take the eyewitness reports at face value. We initially deemed them suspect because our background information indicated that the informants are unreliable. The credibility of the reports increased because of their agreement with one another and the support provided by collateral information. That agreement gave us reason to think that the general unreliability of the witnesses did not affect the standing of these particular reports. Contravening considerations are subject to similar assessments. Mrs. Abercrombie, being dreadfully near-sighted and woefully out of date, cannot even imagine, much less see, that a green thatch on someone’s head is his hair. That being so, her characterization of the suspect as wearing a green hat seems close enough to count as supporting rather than undermining the original allegation. It is, then, true enough. Although Mr. Polk flatly denies what the others say, there are reasons to doubt his veracity. The three putative eyewitnesses purportedly saw each other in the corridor during the interval when Mr. Polk denies that anyone was there, so his contention is dubious on independent grounds. Since he occasionally goes AWOL to smoke a cigarette, it is not implausible that he was absent when the theft occurred. Mr. Miller’s argument cannot be so easily discredited. The prior probability of a Latin book theft is slight. But the book is gone. Meg put it in her locker when she arrived at school. It was not there when she returned. None of her classmates admits to having borrowed it. Even if Latin books are not normally attractive targets for teenage thieves, the book’s having been stolen may explain its absence better than any available alternative would. Just as other considerations compensate for the improbability of a green-haired thief, other considerations compensate for the improbability of a Latin book’s being stolen. In determining the acceptability of a claim, we assess the considerations that afford evidence pertaining to its tenability. This is not always a simple yes/no matter. We may find that although an evidence statement is unacceptable or unsupportive as it stands, with suitable modifications, it would be. And we may find that the modifications themselves are acceptable. Coherence remains crucial. Sometimes it is achieved directly, sometimes by discrediting or disarming threats.
Although we are initially inclined to take putatively eyewitness testimony at face value, we reject Mr. Polk’s claims when we find that they conflict with the testimony of the other witnesses, and we reconstrue Mrs. Abercrombie’s claim so that it supports rather than conflicts with the contention that the thief had green hair. Since none of the school’s students, faculty, or staff has green hair, we entertain and finally accept the hypothesis that the thief was an outsider, despite our initial conviction that it was an inside job. If our methods prove wanting, or our standards inadequate, they too are subject to modification or rejection. If we seek to understand the theft rather than merely identify the thief, we may appeal to psychological and sociological insights about motivation and the conditions that give rise to it.
The coherence that affords epistemic justification is not just coherence among object-level considerations. We have higher-order commitments about what sorts of object-level considerations are trustworthy, what level of credibility they enjoy, how they ought to mesh, and what to do when commitments clash. These higher-order commitments supply reasons to revise or reject some contentions rather than others when conflicts occur. The coherence that constitutes epistemic justification is something we achieve, not something that simply falls out of the relations in which our object-level deliverances happen to stand to one another.
Although epistemology generally focuses on the beliefs of a single individual, I began with a public case because the otherwise unlikely agreement of independent witnesses clearly shows how the best explanation of the coherence of a given body of claims enhances their acceptability. The case of a single individual can be trickier. People sometimes confabulate. This is a matter not of lying or intentionally seeking to mislead, but of self-deception. A confabulator composes a coherent narrative by unconsciously ignoring, bracketing, or downplaying factors that detract from the story he seeks to construct. It might seem that such self-deception undermines my case for epistemic holism, since an account that achieves coherence by bracketing, ignoring, or misinterpreting relevant factors has no claim to be epistemically acceptable.
But coherence through confabulation is not so easy to achieve as philosophers imagine. Adler (2002, 67–72) argues that self-deception typically undermines coherence. Confabulators achieve coherence among their first-order beliefs by sacrificing coherence between first- and second-order beliefs. A doting father deceives himself into thinking that his daughter’s dismal grades are due to her dimwitted teachers’ failure to recognize her quirky brilliance. To sustain this belief he needs to overlook factors that he normally takes to bear on the assessment of student performance—her dreadful study habits, her incomplete homework, her ignorance of the most basic facts about the subjects she is studying. He may ‘explain away’ these factors to his own satisfaction, thereby sustaining his self-deceptive belief. Nevertheless, he weakens the overall coherence of the relevant neighborhood of his belief system by carving out exceptions for a special case. His thinking violates his otherwise accepted methodological commitments about how to assess such matters. He would not, for example, invoke them to account for his paperboy’s poor grades. This is not to say that the self-deceptive agent can recognize his error. He has, after all, managed to deceive himself. But because self-deception weakens coherence, it does not automatically discredit the contention that coherence plays a major role in justification.
Even if mundane self-deception is not a serious worry, delusions might pose a stronger challenge. Suppose a delusional mental patient believes he is Napoleon in exile. He interprets all of his experiences in the locked ward in terms of his delusion. He takes the nurses, doctors, and attendants to be lackeys, courtiers, and palace guards; visitors to be loyal subjects; those who refuse to do his bidding to be traitors; those who tell him he is ill or mistaken in his beliefs to be part of the plot to prevent him from retaking the throne. Unlike the self-deceptive father, he does not carve out special exceptions. He explains all of his experiences in terms of the delusion. Whenever a tension occurs, he rejects or radically reinterprets the deliverance that conflicts with his self-identification with Napoleon. His beliefs may well be more coherent than the self-deceptive father’s. But notice how deeply deluded he would have to be to achieve anything like the level of coherence that normal epistemic agents effortlessly achieve. He would have to believe that he and everyone around him was speaking French rather than English. He would have to believe that oil lamps or candles rather than electric lights were illuminating the area. He would have to believe that the vehicles he sees are horse-drawn carriages rather than cars and trucks. He would have to construe computers, televisions, and elevators as early nineteenth-century artifacts. He would probably have to blind himself to a vast number of familiar items—Velcro fasteners, ballpoint pens, cell phones, and so forth. I do not deny that this is possible. But it is important to see how thoroughgoing a delusion must be to retain coherence. It is doubtful that he could achieve anything like the level of coherence that ordinary clusters of beliefs typically exhibit.
Obviously, when epistemic agents are self-deceptive or deluded, the coherence of their beliefs is not explained by their being true or true enough. If it is hard to tell whether they are deceiving themselves, it is hard to tell whether coherence confers epistemic standing on their beliefs. But to understand how, why, and when coherence engenders credibility, it is best to put this complication aside. Then we see that the story I have told about the book theft could be told of a single epistemic agent as well. If the agent’s account reaches an acceptable threshold of coherence and the best explanation of that coherence is that it is at least true enough, then so long as she has no overriding reason to think otherwise, she is justified. Anne is aware of what she thinks she saw, and what she thinks the other witnesses report. She is privy to the relevant background information about apparent sight lines and the like. Since her various relevant cognitive commitments mesh and the best explanation of their meshing is that they are at least roughly right, according to my epistemological holism, she is justified in accepting them.
The second worry is that coherence is not demanding enough, for it can readily be achieved through epistemically illicit means. A good nineteenth-century novel is highly coherent, but not credible on that account. Even though Middlemarch is far more coherent than our regrettably fragmentary and disjointed views about the book theft, the best explanation of its coherence lies in the novelist’s craft, not in the truth or approximate truth of the story. The novel’s coherence affords virtually no reason to think it is anywhere near literally true. This is surely right. But rather than taking this objection to completely discredit the contention that coherence conduces to epistemic acceptability, I suggest that it indicates something different: coherence conduces to epistemic acceptability only when the best explanation of the coherence of a constellation of claims is that they are (at least roughly) correct. That is, the claims that purport to be true are at least roughly true and those that purport to be true enough are true enough.
One might argue that even the best nineteenth-century novel does not pose as great a threat as we sometimes suppose. No matter how deeply immersed I am in the story, a single glance up from the page suffices to convince me that I am not in a drawing room in nineteenth-century England. The content of the story, though internally coherent, manifestly fails to mesh with the rest of my experience. This is so, but the question is what to make of it. On the one hand, too restricted a cluster of mutually supportive claims seems inadequate to engender tenability. We cannot render the story tenable simply by ignoring everything else we believe. On the other hand, insisting that all our commitments need to cohere seems unduly demanding. If acceptability requires coherence with everything we accept, or with everything we accept for cognitive purposes (Lehrer, 1986), it is but a short step to skepticism. One wayward belief, however remote from current concerns, could discredit an entire constellation of beliefs. Epistemological theories that ground justification in coherence then face a problem of scope.
Worries about scope, however, seem not to do justice to the difficulty that confronts us here. Faced with a clash between the deliverances of the novel and those of my glance, it is obvious which I should accept.1 There is no temptation to resolve the tension by dismissing perceptual deliverances or taking them to be fiction. They seem to possess an epistemic privilege that prevents considerations of coherence from overriding them. The capacity of perceptual deliverances to override the claims of a tightly knit novel may seem conclusively to demonstrate that epistemological justification cannot consist in coherence.
Perception’s Putative Privilege
Although I do not contend that justification consists in coherence alone, the matter deserves further consideration. Until the source of perception’s epistemic privilege is clear, it is premature to rule coherence out. Foundationalists argue that knowledge requires that there be some independently credible beliefs. Many foundationalists hold that perceptual deliverances are among the independently credible beliefs because perceptual deliverances derive at least some of their warrant from the circumstances in which they occur, not their relation to other deliverances. Exactly how credible they are is a matter of dispute (BonJour, 1985, 26–30). But they must, foundationalists contend, have some measure of credibility that does not derive from their accord with other convictions. Reliabilists argue that a deliverance is epistemically acceptable if produced by a reliable mechanism. Since some perceptual mechanisms are reliable, some perceptual deliverances are acceptable. Since the reliability of perceptual mechanisms is independent of the relations of their deliverances to other deliverances, perceptual deliverances are independently credible.
There are at least two separate insights here. The reliabilist argument targets the need for a link to the world. The reason for crediting the deliverances of a casual glance while dismissing those of the novel is, they maintain, that perception provides the link. The way the world is constrains our perceptual deliverances more immediately and directly than it does our other beliefs. Insofar as the contents of knowledge claims concern the way the world is, it makes sense that the constraints the world supplies should override other considerations. The foundationalist position underscores the idea that some deliverances—in particular, those of perception—seem to display some measure of credibility independently of their connections to other beliefs.
What the objections show is that if perception is to provide the sort of check on theorizing that we think it should, egalitarianism with respect to object-level deliverances will not do. An egalitarian account would hold that each deliverance has an equal claim on our epistemic allegiance. On an epistemic principle akin to ‘One Person, One Vote’, it would maintain that there is no basis for privileging some commitments over others. If a perceptual deliverance fails to cohere with an otherwise coherent account, the perceptual deliverance ought to be rejected, since the claims of the many outweigh the claims of the one. But no matter how comprehensive and integrated an empirical account is, no matter how many other beliefs the account manages to incorporate, observations should have the capacity to discredit it. They have that capacity only if the epistemic claims of perceptual deliverances at least sometimes outweigh those of an antecedently accepted account. Still, it does not follow that perceptual deliverances must be immune to revision or rejection on the basis of considerations of coherence. Nor does it follow that the epistemic privilege granted to perceptual deliverances is independent of considerations of coherence.
If we think about our situation when we glance away from the novel, we recognize that we draw on more than the sentences comprising the novel and our current perceptual deliverances. We tacitly rely on a fairly extensive and epistemologically informed understanding of novels and of perception. We know enough about underlying mechanisms and track records to have reason to credit some perceptual deliverances. We know enough about literature to realize that novels typically are not literally true. That constitutes sufficient reason for even casual perceptual deliverances to override the claims of the novel.
Juxtaposing the novel with perception might seem to make the problem too easy, though. Regardless of what we think about perception, if we recognize that a novel is a work of fiction, we have reason to discount any direct claims it may seem to make on our epistemic allegiance.2 The serious challenge comes from a coherent factual account that conflicts with perceptual deliverances. If holism holds that such an account always overrides perceptual deliverances, it seems plainly unacceptable. However tightly woven an empirical account may be, we would be epistemically irresponsible to ignore recalcitrant evidence. Foundationalists take this latter point to be decisive: if observation can show an account to be unjustified, then coherence cannot be the locus of justification.
This would be so if observation worked in isolation. For then, owing to its epistemic privilege, a single perceptual deliverance would have the capacity to discredit an entire system of thought. But that is a myth. Only observations we have reason to trust have the power to unseat theories. It is not an observation in isolation, but an observation backed by reasons, that actually discredits the account.
The holist response to the challenge presented by observation is this: a priori, perceptual deliverances have no special weight. They are just initially tenable commitments jockeying for inclusion in epistemically acceptable accounts. But as we attend to the fates of our various initially tenable commitments, we learn that the incorporation of some, but not others, yields accounts that are borne out by further experience and that retain their place in a coherent system over time. This gives us grounds for discrimination. We realize that the deliverances we take to be perceptual are more likely to be confirmed than spontaneous deliverances that just leap to mind. So we assign greater weight to perceptual deliverances than to hunches and premonitions. Eventually, hunches and premonitions cease to qualify as initially tenable. Moreover, we learn that not all perceptual deliverances are on a par. Those that are credible tend to come in mutually reinforcing streams, so isolated perceptual deliverances count for relatively little. We begin to draw distinctions among perceptual deliverances. We discover that peripheral vision is less trustworthy than foveal vision. So we have reason to discount what we see out of the corner of the eye. This is not to say that we dismiss the deliverances of peripheral vision out of hand, only that we demand more in the way of corroboration. A passing glance that reveals what looks to be a goldfinch at the bird feeder requires additional support if there is good reason to think that goldfinches have flown south for the winter. Some of us discover that we are color-blind or tone deaf or myopic. That is, we learn that our perceptions as of colors, pitches, or the dimensions of distant objects are not to be trusted. And so on. We come to assign different weights to perceptual deliverances depending on how well they accord with other things we take ourselves to have reason to credit—other appearances of the same object, the reports of other observers, the implications of our best accounts of the perceptible properties of items of the kind in question, and so forth.
The issue is not simply how well a given content meshes with other things we believe, but how well a given content from a given source in given circumstances does. The weight we attach to perceptual deliverances derives from our understanding of the world and our access to it. Perhaps initially this is just a matter of track records. Some perceptual deliverances seem to integrate better into acceptable accounts than spontaneous thoughts that just leap to mind. Later, as we develop physiological and psychological accounts that explain our perceptual mechanisms, we gain additional reasons to credit some perceptual deliverances. The epistemic privilege that some perceptual deliverances enjoy then derives from an understanding of ourselves as perceiving organisms. That is, the reason for assigning those deliverances significant epistemic weight derives from the coherent account of perception that backs the assignment. Contrary to the foundationalist’s contention, the justification for privileging perception derives from the relation of perceptual judgments to the rest of our account of ourselves as cognitive agents interacting with a mind-independent world.
The reliabilist account seems to fare slightly better. What justifies assigning visual inputs significant epistemic weight seems to be that vision is a reliable perceptual mechanism. What justifies dismissing forebodings is that premonition is not. This is not quite right, though. It is not the brute reliability or unreliability of a source that supplies the justification, but a recognition of that reliability or unreliability. Reliabilists disagree. They hold that the sheer fact that a process is reliable renders its deliverances at least prima facie justified (see Goldman, 1986). They would say that if my forebodings were reliable, they would be justified. Were I to believe them, their deliverances would be knowledge. Maybe so. But the knowledge would avail me little. If I had no reason to trust my forebodings, they would not be acceptable. It would be irresponsible for me to use them as a basis for inference or action when my ends are cognitive. From my perspective, they would appear no different from lucky guesses. Even if my forebodings are accurate and their source is reliable, so long as we have no reason to trust them, they rightly bear little epistemic weight.
This argument explains both why some perceptual deliverances have the capacity to unsettle an account, and why those deliverances are not intrinsically privileged. They owe their epistemic status to their place in our evolving understanding of the world and our modes of access to it. This has two welcome consequences. The first is that the privilege they enjoy is revocable. When Bill learns that he is color-blind, he needs to revise his views about which of his visual deliverances are acceptable. The second is that nonperceptual deliverances can in principle be equally weighty. This is an advantage in accounting for the epistemic status of scientific evidence and of testimony.
A look at modern science shows that it is not just (and perhaps not even mainly) bare perceptual deliverances that have the capacity to discredit a scientific account. The outputs of measuring devices do too. In an effort to retain a tie to classical empiricism, some philosophers of science argue that measuring devices are simply extensions of our senses. Just as eyeglasses enable the myopic to see what otherwise they could not, telescopes and microscopes enable everyone to see what otherwise we could not. So if seeing something in suitable circumstances has sufficient weight to undermine a coherent cluster of commitments, seeing something through a telescope or microscope should be able to do so too.
This idea is not unreasonable so long as we restrict ourselves to devices like optical telescopes and microscopes. But it stretches the bounds of plausibility to contend that radio telescopes, electron microscopes, MRIs, and the like are also mere extenders of the sense of sight. It seems preferable to forgo the strained analogy and simply characterize such devices as detectors. Then an understanding of what they detect, how they detect, and why they should be trusted supplies reason to accord their outputs considerable epistemic weight. Even without the strained analogy, the argument for crediting outputs of scientific instruments thus parallels the argument for crediting perceptual deliverances. For although they are not perceptual mechanisms or even quasi-perceptual mechanisms, the devices are among our modes of access to the world.
Testimony poses a similar problem. We acquire many of our beliefs from the testimony of others, and we consider those beliefs justified. Some philosophers say that the justification for accepting testimony is a priori, albeit defeasible (e.g., Burge, 1993). Ceteris paribus, we are justified in accepting what people tell us. Others say it is inductive. We should believe only those who have shown themselves to be relevantly reliable in the past (E. Fricker, 1994). The former threaten to endorse gullibility, the latter to unduly limit acceptability. Something more sensitive is wanted. Evidently the question is not whether testimony per se is or is not prima facie acceptable. Some testimony is frankly incredible; some requires a good deal of corroboration; some is straightforwardly acceptable. The acceptability of a bit of testimony depends on how well its content coheres with other relevant deliverances, how well the belief that the testifier is competent with respect to her allegation coheres, and how well the belief that she is sincere coheres (Hume, 1977). Because of its mesh with our background beliefs, straightforwardly acceptable testimony scores high on all these measures. Just as different perceptual deliverances are accorded different weights, so are different bits of testimony. Testimony with sufficiently strong backing can discredit a hitherto tenable cluster of beliefs. Even though the deliverances of perception, testimony, and instrumental readings have no special standing a priori, in light of our evolving theories of the world and our modes of access to it, some of them turn out to bear considerable epistemic weight. This satisfies the demand that acceptable beliefs be appropriately constrained by the way the world is. It also reveals that holism has the resources to recognize that initially tenable commitments can differ in weight, some being more credible than others. The claims of the few can in suitable circumstances outweigh the claims of the many.
Adjudication
In adjudicating among competing claims, we prioritize. We do not consider all initial commitments on a par. When they clash, we preserve some at the expense of others, and consider our decisions reasonable. Systematizing is informed by higher-order commitments that assign weights to first-order commitments. Although in principle any commitment can be revised or rejected in the interests of systematicity, commitments differ in cognitive inertia. Some are readily given up; others are abandoned with great reluctance. We readily relinquish the initially tenable commitment that there is a cowbird at the bird feeder in Massachusetts in January. Since it is at best only weakly interwoven with our other commitments, little is lost by abandoning it.
But, as Quine recognizes, even central, tightly woven contentions may justifiably be surrendered. ‘Inanimate material objects are identical when and only when their parts are identical’ is a highly plausible principle that many metaphysicians consider necessary. If so, it cannot be false. Being a universal claim, it should hold for particles of a viscous fluid. However, if it holds, ‘F = ma’ does not. The reason is this: The molecules in a viscous fluid move at different rates. In prototypical applications of ‘F = ma’, forces act on objects like billiard balls that have some sort of material integrity or sharp boundary. But in viscous fluids, the ‘forces’ on the ‘object’ are the effects on momentum of molecules moving in and out of that ‘object’. So preserving ‘F = ma’ requires continually redefining what constitutes a particle, letting different molecules comprise it at different times. Although physicists concede that ‘F = ma’ does not hold at the quantum level or at speeds and distances where relativistic factors play a major role, it is an extremely valuable law for characterizing the behavior of slow-moving middle-sized items in this region of the cosmos. That includes the viscous fluids flowing around here. Fluid mechanics thus characterizes its particles so as to comport with the law. Rather than insisting that all component molecules of a particle be the same from one instant to the next, they let the individual molecules come and go, but keep the average enclosed mass constant (Wilson, 2006, 158–159). ‘F = ma’ evidently is so central a law of physics that scientists are willing to make drastic revisions in the criteria for the identity of a particle over time in order to preserve it. In this case, the tension is acute. One way or another, a radical revision in antecedently plausible principles is needed. Either scientists must revise a very reasonable metaphysical commitment about the identity of an object over time, or they must revise a fundamental law of physics.
The issue of scope remains. The totality of a person’s epistemic commitments is unlikely to be coherent. Not only are there outliers and inconsistencies among commitments, but an epistemic agent also harbors constellations of commitments that are relatively isolated from one another. Meg’s cluster of beliefs about the pituitary gland, the evidence that bears on the acceptability of these beliefs, the trustworthiness of testimony on the subject, and the proper methods for assessing such matters has few and loose connections to her cluster of views about parliamentary procedure, the evidence that bears on these views, the trustworthiness of testimony about the subject, and the proper methods for assessing them. It seems that she could easily be badly wrong about the former without her error having any significant effect on the tenability of her views about the latter. Outliers and inconsistencies among beliefs are in principle relatively unproblematic. According to holists, outliers lack justification. Because they lack suitable connections to other epistemic commitments, although they are initially tenable, the epistemic agent has little reason to credit them. Inconsistencies among commitments conclusively demonstrate that some of them are false. But it is not obvious that mutual indifference of clusters of commitments is objectionable. It would be unreasonable to consider Meg epistemically defective because of the lack of close ties between the two clusters. On the other hand, if the clusters of commitments are too small and too numerous, complacency over their mutual indifference seems problematic. We do not want to license ignoring inconvenient tensions among commitments by consigning them to mutually irrelevant clusters.
The problem neither has nor needs an a priori resolution. Our evolving theories of the world and our access to it provide us with an appreciation of the relations in which our various clusters of commitments should stand to one another and the requirements they should satisfy. Such a laissez-faire attitude might seem to allow for the acceptability of crazy constellations of views. If we leave it to our evolving worldviews to decide what range of considerations acceptable accounts must answer to, we may be forced to endorse isolated islands of claptrap. The worry is more apparent than real. We have metalevel accounts that enable us to assess the reasons, methods, standards, and evidence that our various object-level accounts appeal to. Some requirements, such as logical consistency, apply globally. Regardless of the distance between them, unless they can be conjoined without contradiction, Meg’s political and endocrinological commitments are not all acceptable. Other requirements, like the need to respect judicial precedents or to accord with biochemical findings, are more limited in range. But even these do not enable us to completely isolate clusters of commitments. Even if Meg’s views about endocrinology and politics have few points of contact, her views about endocrinology and hematology have many.
Consistency requirements do more than rule out express contradictions. The general requirement that like cases be treated alike demands that if a consideration has weight in one area but not in another, there must be an acceptable reason for the difference. To be tenable, a system of mutually reinforcing claims must either answer to the logical and evidential standards to which other theories are subject or be backed by a tenable account of why those standards do not apply. Some theories have such backing. There are, for example, cogent reasons why mathematics is not subject to empirical testing. So infinitary mathematics is not threatened by the absence of empirical evidence for its findings. In epistemically objectionable cases, no such reasons are available. The claims of ‘creation science’, although purportedly mutually reinforcing, are epistemically unacceptable because they are indefeasible. According to their adherents, nothing could count as evidence against them, for whatever we find is the way God created things. To say that something cannot be ruled out a priori is not to say that it cannot be definitely and decisively ruled out.
A familiar objection to coherentism is that coherence is so easily achieved that it affords no reason to believe that the contentions in a coherent account are true, hence no justification for them. This assumes that any coherent collection of claims is as good as any other and that whenever incoherence threatens, we can restore coherence by throwing out whatever poses the threat. Whatever its merits as an objection to pure coherentism, it does not apply to positions that ground acceptability in reflective equilibrium. First, not all initially tenable commitments are object-level statements pertaining to the topic at hand. Among those we seek to accommodate are ostensible epistemic values, such as simplicity and relevance; ostensible epistemic standards, such as standards of statistical significance and of evidential support; ostensible epistemic criteria, such as those specifying what counts as validity and as argument; and ostensibly acceptable methods, such as controlled experiments and random sampling. It is by no means easy to bring all of these into accord. Second, the objection overlooks the special role that initially tenable commitments play. Because they are considerations that present themselves as creditable, they serve as touchstones for theorizing. They supply epistemic ballast, preventing us from restoring coherence by simply rejecting considerations cavalierly.
Not all initially tenable commitments, of course, present themselves as candidates for belief. Putatively felicitous falsehoods present themselves as being presumptively true enough. Putatively acceptable values, methods, standards, and criteria are elements of the constellations within which doxastic commitments function. So the slight presumption in favor of doxastic commitments extends to axiological and methodological commitments and felicitous falsehoods as well. They, too, are initially tenable.
We devise a flexible network of cognitive commitments that, through continual readjustments, achieves an understanding of the topic that is on balance reasonable. None of the commitments is absolutely unrevisable. Different potential revisions have different costs and benefits. To decide among potential revisions requires asking what sort of understanding we seek, what resources we have to draw on, and what limitations we currently face. We have multiple cognitive desiderata—simplicity, fecundity, elegance, predictive power, and so on. Insofar as is feasible, revisions in our initially tenable commitments should yield an account that satisfies them. Moreover, with the advancement of inquiry, methods, standards, values, and goals evolve. There emerge new understandings of how to find or figure things out, what sorts of methods and results stand the test of time, what sorts of representations are apt, what desiderata are worth pursuing and when and how they might conflict. When the account that emerges is grounded in considerations that are collectively as good as any available alternatives, it is an account we can on reflection accept.
Some might dispute this. If our initial commitments were quite wide of the mark, the account that results might still be far from the truth. This is so. Nevertheless, such an account is an epistemic improvement over our starting point and is as good as anything we have reason to accept in those dismal epistemic circumstances. It affords a relatively stable platform for research that may yield yet further improvements.
To see this, consider a skeptical scenario. Imagine that unbeknownst to themselves, a group of scientists live in a Berkeleyan universe. What they take to be material objects are really immaterial ideas in the mind of God. Initially, they have fairly crude, unsystematic opinions about how things behave and about how to find out how they behave. Over time, they correct, extend, and systematize their approach to the point where they develop the scientific method. They subject their findings to peer review. They insist on controlled, repeatable experiments whose results are statistically significant. The geniuses among them come up with laws that reveal an astonishing order and regularity among the phenomena they observe. The order those laws describe obtains. The ‘orbits’ of the ‘planets’, the ‘tides’, and ‘falling bodies’ fall under the same equations. Their account is wrong, however, in that it takes the substrate for that order to be material rather than mental.
There is no denying that this is a big mistake. But, I would urge, we should also acknowledge their significant epistemic achievement. Otherwise we would have to construe the success of their science as a fluke. They can, like all scientists, make further progress by extending, correcting, and deepening their account, improving their methods, fine-tuning their criteria, and so on. Maybe someday they will be in a position to recognize that matter, as they conceive it, does not exist; maybe not. But it seems churlish to discount their achievement because it yields insight only into the structure of reality, not into its metaphysical ground. Rather, I contend, we should recognize that because their opinions, methods, criteria, and standards are in reflective equilibrium, they are reasonable in the epistemic circumstances. The system they have constructed is as good epistemologically as could have been achieved given their unfortunate starting points.
Some might concede that the method of reflective equilibrium is the way we go about developing our accounts. It is, after all, a schematic idealization of the self-correcting scientific method. But, they would insist, it is only a method. What makes an account a good account is not that it was generated by the method, but that it satisfies some independently specified standard. Typically the standard is truth. And what makes the method a good method is that it tends to produce theories that satisfy the external standard. The justification for the method of reflective equilibrium, they would maintain, is that it is truth conducive.3
There are at least two replies. One is that we have no assurance that the accounts we accept satisfy the standard. If the standard is really independent of the method, we might legitimately wonder whether the accounts we credit are true. Going back and garnering more evidence would plunge us back into the method of reflective equilibrium, whose bona fides are in question. This is a version of the problem of the criterion (Chisholm, 1973). Moreover, if the standard is truth, the account devised in the Berkeleyan world is devoid of epistemic merit. Perhaps we can concede that the scientists did the best they could, but in the circumstances, there was no way to come up with an epistemically estimable account. From a purely epistemological point of view, they are no better than their compatriots who cavalierly jump to conclusions without even trying to amass or assess evidence.
This is not just a point about some fictional scientists in a made-up world. We have no conclusive evidence that we are not in their situation. That being so, the acceptability of accounts that satisfy the standards we actually use is doubtful. Unless we can explain why being in reflective equilibrium is truth conducive, we seem hostages to epistemic fortune. From our perspective, whether we live in a Berkeleyan world is a matter of epistemic luck. If we can explain it, we can treat being true or truth conducive as an objective that we attempt to accommodate in devising theories in reflective equilibrium. And, if we like, we can assign truth or truth-conduciveness a high priority so that other desiderata would be given up to achieve it. I will argue below that science tends to be much more flexible in its attitude toward truth, but in principle we could take a hard line.
It might seem that the requirement on epistemic acceptability it is too lenient. ‘As good as any available alternative’ is not ‘better than any available alternative’, much less ‘better than any alternative, available or not’. Why settle for so little? The method I sketched plainly leaves room for ties. When commitments clash, different trade-offs may be equally good on balance. One might sacrifice a measure of sensitivity for an increase in scope; the other might sacrifice scope for sensitivity. It seems dogmatic to insist that one must be better than the other when, by our standards, both do an equally good job and none does a better job at realizing our cognitive desiderata.
Some philosophers think that such ties just reveal our epistemic inadequacy. There is one way the world is, so there is one and only one correct account of the world. If at a certain point in our investigation, we have two or more equally good accounts (and none better), that just shows that our methods, standards, or criteria are too weak. We need to introduce tie breakers (D. Lewis, 1999a). But even if there is just one way the world is,4 the contention that there is exactly one best account of the world is unwarranted. What makes for a good account depends not just on the subject matter of the account but also on our goals and resources in theorizing. Even if the world’s contribution is fixed, different accommodations to our inevitably limited resources may be equally reasonable. No argument is given that would privilege a preference for a more sensitive account over one with greater scope, or conversely. Sensitivity and scope are both desiderata. We can explain why each of them is desirable. But the conviction that there is a unique optimal balance between them is unwarranted. Requiring that we refine our standards until we achieve uniqueness is idle. Without an epistemically accessible indication of what justifies one refinement over another, any decision would be arbitrary.
To insist that an account is unacceptable unless it is better than any alternative, whether available or not, is perilously close to an argument for skepticism. To suspend judgment until we have reason to believe that an account we have devised is better than any imaginable or unimaginable alternative is to suspend judgment permanently; to accept an account on the assumption that it is better than any imaginable or unimaginable alternative is intellectually arrogant. It intimates that the account satisfies a standard it cannot plausibly be known or reasonably believed to satisfy. To insist that uniqueness is required is to consign our accounts to inadequacy, whatever merits they display. Or it is to contend disingenuously that the accounts we consider acceptable satisfy standards we have no evidence that they satisfy, indeed standards we cannot even articulate. The best policy, then, is to take the standard of acceptability to be ‘as good as any available alternative’ and to forgo a demand for uniqueness.
Earlier I said that truth has a more peripheral role in understanding than epistemologists standardly think. We are now in a position to see what that role is. A tenable account affords an understanding of its topic. An account is tenable only if it is in reflective equilibrium. This may require that claims in the account that purport to be true actually be true, or anyway not off by much. Although science endorses felicitously false models, it does not consider false observation reports to be felicitous. Fidelity to the evidence seems mandatory. Observation reports are not, of course, the only assertions in scientific accounts. Assertions—about laws, boundary conditions, methods, and so forth—are scattered throughout. It might seem, then, that there is a streamlined answer to our question: tenability requires that the assertions in an account be true and the professions be felicitously false. For assertions purport to be true, and professions purport to be felicitous. Only then, we might think, would an account embody an understanding of the phenomena it concerns.
This does not quite work, for understanding grows over time. One of the ways it grows is by overthrowing previously accepted accounts. Sellars (1968) maintains that when one theory supplants another, it should have the resources to explain why its predecessor did as well as it did. Often the reason the predecessor was so successful is that some of its central assertions, which were put forth as true, turned out to be felicitously false. Kepler’s laws, presented as true, neglected the gravitational force on the Earth of everything except the sun. As Newton recognized, Kepler’s laws are false. For the gravitational forces exerted by the other planets affect the Earth’s orbit. But because the forces exerted by the planets are relatively insignificant compared with that of the sun, Kepler’s laws are not off by much. So Newton’s theory satisfies Sellars’s demand by reinterpreting statements that purport to be true as felicitously false. That a sentence is asserted shows only that it purports to be true. It does not show that its truth is what it contributes to the tenability of account it belongs to. Truth evidently is epistemically important only to the extent that it contributes to tenability.
My position does not privilege any sorts of beliefs or representations a priori. Nothing favors the verbal, the literal, the descriptive, or the true. Acceptability derives from reflective equilibrium. So whether a representation (true or false, verbal or pictorial, literal or figurative, denotative or exemplificational) is acceptable turns on whether it is an element of an account in reflective equilibrium. What commitments are worthy of acceptance is something we learn by developing increasingly comprehensive, coherent accounts of the world and our access to it. The method enables us to start from whatever initially tenable commitments we happen to have. But because it insists that we subject them to rigorous assessment, such a starting point is not question begging. The standards of assessment are themselves the fruits of epistemic activity and can change in response to feedback (Goodman, 1984, 69). Hence, nothing is in principle immune to revision. An account that we can on reflection accept today may be one that we cannot on reflection accept tomorrow. But so long as an account is in reflective equilibrium, it and its components are justified. What results is neither certainty nor skepticism but a fallible, provisional, reasonable epistemic stance. We judge our commitments and potential revisions of our commitments against one another to see which combination seems best on balance. But we also judge our criteria for being best on balance by seeing if they yield verdicts that we can on reflection endorse. No account or component of an account is irrevocable, but an account in reflective equilibrium is acceptable until problems emerge or improvements are envisioned. The justification for this verdict lies in the fact that the commitments in reflective equilibrium are as reasonable as any alternative in the epistemic circumstances.
Notes
1. One might wonder whether the contents of a novel qualify as deliverances. Arguably, they do not present themselves as candidates for epistemic acceptance. If this is right, the objection does not get off the ground. In chapter 11 I discuss the epistemic functions of fiction. Here I am willing to treat the contents of a novel as deliverances, and let the nineteenth-century novel serve as a challenge to theories that take coherence to be conducive of tenability.
2. I say ‘direct claims’ because I believe that novels play a significant role in the advancement of understanding. Just how they do so will emerge in chapter 11. See Elgin (1996), 183–200.
3. I thank Fred Feldman for this objection.
4. Which I think is probably an unintelligible claim.