Forget and Forgive: A Practical Approach to Forgotten Evidence

1.1. The Problem of Forgotten Evidence

If someone forms a belief in an irrational way, say by purely wishful thinking, but then he completely forgets how he formed this belief, is the belief still irrational? For a concrete example, take Joe, who is a severe coffee addict, and who at one point formed the belief that drinking coffee is beneficial to your health. Let’s suppose Joe formed this belief purely on the basis of his wish that this be true, and suppose he even engaged in this bit of wishful thinking knowingly at the time (he’s helpless!), and he later completely forgot how he came to his view about coffee’s health benefits. Is the surviving belief still irrational, or is it now rational? The case poses a problem, which we can present as a dilemma.

The first horn of the dilemma is just that there is certainly something uncomfortable about calling Joe’s belief rational. If his belief becomes rational after he forgets how he formed it, then Joe has laundered the epistemic status of his belief—like the mafia launders money. Can you really do this? Can you turn an irrational belief into a rational belief just by forgetting things? To make the problem even more vivid, think of the most appalling belief you can. Maybe someone, purely on the basis of his dislike for you, came to believe something horrible about you or your family, but the belief’s original basis is now completely lost from his memory. Easily forgiven?

It feels wrong to forgive the forgetful reasoner. After all, it appears as though he never acquired, and certainly does not now retain, any positive reasons in favor of his view, and that looks bad. How can high confidence in propositions about topics like coffee’s health benefits be justified without your ever having in your possession any positive reasons in favor of a particular view?

Some epistemologists who favor taking this first horn of the dilemma have ventured to suggest that you can possess positive reasons to trust a surviving belief even after forgetting its origins in virtue of a background belief you have that most of your beliefs have happily turned out, upon facing the test of time, to have been accurate.[1]

Unfortunately, the background belief suggestion is not appealing upon reflection. The problem is that it turns out that the background belief itself cannot be supported by any positive reasons, and thus anything that is based upon it cannot be said to ultimately enjoy the support of positive reasons. Why is it that the background belief has no positive reasons in its support? Well, if you were to possess some positive reasons that support this belief that your memory has been generally reliable, then those reasons would have to make some claim about the general accuracy of your past beliefs. But, what sort of reason have you to think your past beliefs were generally accurate? Any considerations you reach for here will make your reasoning circular.[2] So, even if Joe holds his belief that coffee is healthy on the basis of a background belief that his memory has generally been reliable, this will ultimately do very little to improve his ability to provide reasons in favor of his coffee belief, for the coffee belief is being based on another belief—the background belief—that itself enjoys no positive reasons in its own favor. To be clear, the background belief might well qualify as justified—perhaps it is justified by default, without the aid of any positive reasons in its favor. The point here is just that the worry about Joe not having any positive reasons to support his belief that coffee is healthy will not be satisfactorily addressed by claiming Joe should support that belief by basing it on a background belief that his memory has generally been reliable.

Turning to the other horn of the dilemma now, there is a real problem with saying such beliefs as Joe’s are irrational. How can it be that a belief like his is irrational when Joe possesses no evidence, no reason, that favors making any revision to his view about coffee’s health benefits? If his current belief is irrational, but he has no reasons to revise his belief, then he is in a rut: he must either hold on to an irrational belief, or make a revision he has no reason to make, a revision which would only land him with another irrational view. To make the problem here more vivid, just consider that even the best of us form a good number of irrational beliefs. And most of your beliefs, whether rational or irrational, outlive your memory of their original bases. You cannot be rationally required to keep track of much at all from the daily flood of evidence you take in and on the basis of which you first formed your beliefs; as Harman (1986: Chapter 4) famously argued, it would introduce a hugely wasteful amount of clutter into your mind if you stored all the evidence that would be needed to put every belief you now hold on any kind of firm foundation. You just keep the belief and forget the evidence. Joe now possesses no evidence opposing his belief that coffee is good for you, nor does he possess any evidence that his current view was unreliably formed. But, the same goes for lots of his beliefs, and lots of yours and mine, and it would be absurd to think that, just because we can’t store limitless evidence, we have reasons to revise these views.

It seems you may not be able to turn these irrational views into rational ones, even if you acquire and take account of new evidence to form a new view. This is because you cannot just disregard the old view; new views, in particular new credences, are always based, in part, on your prior views, and the rational status of the old view threatens to infect the new view. For example, if your prior credence that a coin is biased is high, then even as you observe an even frequency of heads and tails come up, your posterior credence that it’s biased will still remain a bit higher than it would have been if your prior had been lower. And if that prior credence was irrational, it seems the posterior credence will be too.[3] Joe’s case seems on a par with the coin observer ’s. Suppose Joe acquires new evidence that coffee is in fact not healthy. Perhaps it is allowable that he could end up with a rational outright, or “binary”, belief that coffee is not healthy (I’m not sure). However, his resultant credence in this will have been anchored down to some extent by his prior credence that coffee is healthy. Because of his irrationally high prior credence that coffee is healthy, it seems he will be left with a posterior credence that is still, to some extent, irrationally high. It seems such anchoring effects could only be fully undone if we could make Joe completely certain that his prior coffee views were formed unreliably. Rational certainty in anything is hard enough to attain. And it may be especially hard here, for, if we try to confront Joe with evidence that his view that coffee is healthy was based on wishful thinking, there might be anchoring that makes him rationally mistrust our evidence: he might reasonably be at least slightly suspicious that we are giving him misleading evidence, because he thinks coffee is healthy, and he knows a reliable thinker is likelier than a wishful thinker to have hit upon the truth![4] I think most contemporary epistemologists will be inclined to agree that it will always be reasonable for Joe to remain at least slightly dubious that he engaged in wishful thinking. So, any epistemic predicament like the one Joe got himself into will leave the believer in a real rut, one that is at least very hard, and maybe even impossible, to completely get out of, and one that is very common in all of us.

The problem of forgotten evidence is thus a dilemma. Either Joe’s belief is rational, in which case it has been laundered, or it is irrational, in which case he has landed in a rut. And, whatever we say, we will have to say it about a phenomenon that is very common.[5]

The dilemma divides epistemologists. Many say Joe’s belief is irrational (or unjustified or unwarranted—I won’t distinguish these), including Annis (1980: 326), Senor (1993: 468), Burge (1997: 39–40), Goldman (1999: 280–1), Goldman (2009: Section V), Huemer (1999: 348–9), Owens (2000: 157), and Greco (2005: 261). Others say the belief is now rational, including Harman (1986: Chapter 4), Pollock and Cruz (1999: 87), Conee and Feldman (2001: 9-10), McGrath (2007: 4), Lackey (2008: Sections A.3, A.6), Schoer (2008: 77–8), and Naylor (2015: 377). Interestingly, the division does not track externalist or internalist sympathies (note, for example, that Goldman and Huemer, arch-externalist and arch-internalist, are on the same team). Perhaps even more interestingly, many of these authors choose their side without relying on any argument. They simply express a strong intuition for their own verdict about Joe’s rationality. Each of Annis, Burge, Goldman, Greco, Huemer, Lackey, McGrath, Naylor, and Schoer present their view as simply the intuitively clear verdict—they just have different intuitions!

I join the second group, the nice guys: I say Joe’s belief is now rational. I will present a new line of argument for this conclusion, one that I hope will offer new hope for resolving a debate that looked to be hopelessly deadlocked.

1.2. Plan for My Argument

My argument has an unusual character. I will be arguing for this conclusion, a conclusion that says certain beliefs are rational, not on the basis of initial premises that themselves directly say what is or is not rational, but rather on the basis of certain considerations concerning the practical function, or practical purpose, that we serve by making epistemic evaluations. That is, the overall argument moves to a first-order epistemically normative evaluation for its conclusion, but it doesn’t start from any first-order epistemically normative premises. It starts from some plausible considerations about why we make epistemic evaluations, ostensibly a separate subject matter. And, in fact, the resulting argument is neither deductively valid nor “inductively strong”! (It is not a deduction, an enumerative induction, an inference to the best explanation, or any sort of statistical inference.)

How will the argument be any good then? One step in the middle of the argument will make use of semantic descent. In a simple use of semantic descent, we first make a claim about some bit of language’s truth-conditions and then we switch to using that very language to make a first-order claim that we showed is true. The move where such a switch happens is neither valid nor “inductively strong”, but it is ordinarily unproblematic. A simple example: ‘knows’ is factive; therefore Alice knows no lies.

In this paper ’s overall argument, however, I do not begin with semantics, but rather with pragmatics. I’ll begin by defending a view of the function of our epistemically evaluative language, and this view will then guide us to conclusions about the semantics of these words, in particular to a conclusion about their extensions. I’m going to argue that our use of epistemically evaluative language serves a certain function which I’ll outline, and I’ll argue that it is contrary to that function for us to criticize Joe’s coffee belief as irrational. Rather, we best serve the function by not calling Joe’s belief irrational; we best serve the function by forgiving Joe, and letting his coffee belief pass as rational. From this (pragmatic) claim that applying the language of irrationality to Joe’s belief does not serve its function, I wish to infer the (semantic) conclusion that Joe’s belief is not included in the extension of our word, ‘irrational’, or in the extension of our concept of irrationality expressed by our word. That is, I wish to infer that it is incorrect—a false application—to call Joe’s belief irrational.

Why should this pragmatics-to-semantics inference be reliable? It is plausible that this inference is reliable because it is plausible that we adopted words and concepts of epistemic evaluation with extensions such that their correct application is the application that serves their function. We could have adopted a different concept from our actual one, one whose extension differs slightly so as to make the alternative verdicts about cases of forgotten evidence correct. We could have used words that express slightly different concepts, words that have slightly different meanings. But, since we adopted linguistic and conceptual tools that serve their function by being applied to forgive Joe, it is a much simpler and thus more plausible hypothesis that we adopted a concept, and we use a word, whose extension makes this function-serving application the correct application. That is how my overall argument is supposed to work, and why it is supposed to be a convincing form of argument.

So, although solving the problem of forgotten evidence is certainly of sufficient intrinsic interest to write about, my hope is that this paper will also serve a second, ulterior aim. I hope to show a new way we might make progress on stalemated debates within first-order epistemology. I believe that reflecting on the function of epistemic evaluation can benefit epistemology in many ways. This paper illustrates how it sheds light on the problem of forgotten evidence.

Reflection on the function of philosophically important words and concepts has a few important precedents. Quine (1986: Chapter 1) and Craig (1990) made important proposals about the functions of attributions of truth and knowledge. (Parsons, 1990/2014, discusses Quine’s wider use of the method.) The trend of reflecting on functions has caught on in recent years: see Reynolds (2002), Williams (2002: esp. Section 2.4), Johnston (2006, 2011), Fricker (2007, in press), Henderson (2009, 2011); Henderson and Horgan (in press), Divers (2010), Mercier and Sperber (2011), and Dogramaci (2012, in press). A number of papers explicitly feature the form of argument of this paper: the premises describe the function of some philosophical word or concept, and the conclusion then uses that word or concept to make a claim about the property it expresses. One example: Boris Kment argues for a view of the function of modal discourse and thought, and then he takes this view to support the first-order metaphysical thesis that necessities and possibilities all reduce to counterfactual conditionals (Kment 2006b: 307; 2006a: Sections 1.1, 7). Williamson (2008: Chapter 5, Section 1) argues for that same metaphysical conclusion from the premise that evolution would only pressure us to develop modal cognitive capacities for thoughts about counterfactuals.

In what follows, Section 2 presents the general considerations we’ll require on the function of epistemic evaluation, and Section 3 applies these considerations to the problem of forgotten evidence. As a kind of preview, here’s a short and rough caricature of the argument: we criticize each others’ irrational beliefs in order to promote the trustworthiness of testimony, but it can’t now have any positive impact to criticize the coffee addict Joe who has lost all track of the period of time when he did something worth criticizing, so Joe gets a pass. The caricature you might easily find fault with. The real argument takes up the rest of the paper.

2.1. Having a Function Is Having a Teleological Explanation

I claim that our epistemically evaluative practice has a function. By “our epistemically evaluative practice”, I mean our actual practice of making linguistic assertions that use ordinary words of epistemic evaluation, words like ‘[ir]rational’,‘[un]reasonable’, ‘[un]justified’, ‘[un]warranted’, even the ordinary (non-technical) use of ‘[il]logical’. I’m targeting language that expresses a natural and intuitively significant category of evaluations: it’s the J in the traditional JTB (justified true belief) theory of knowledge. I don’t mean to generally include uses of ‘knows’ as part of our epistemically evaluative practice; however, certain uses are included, in particular those where a subject is said to not know something that’s not obviously false.[6] And sometimes epistemic evaluations can be made with equal effect using no conventionally epistemic words, for example, saying ‘Oh, come on!’ in the right argumentative context. In fact, I think regular people most commonly make epistemic evaluations by calling a person smart, clever, dumb, or stupid, even though those words have many different meanings, and even though a philosopher would never use those words in a paper to mean rational or irrational (probably because they are more vague and ambiguous).

By claiming our epistemically evaluative practice has a function, I mean there exists a teleological explanation of why we engage in that practice. In other words, I mean this: there exists an explanation of why the practice exists that proceeds by showing how the practice efficiently promotes interests of ours, interests that it is independently plausible that we strive to promote.

My intention is to uncover the function of words and concepts we already employ, the ones we already use in ordinary epistemic evaluation. We could, of course, revise our practice and switch over to alternative concepts of epistemic praise and criticism that correctly apply to different extensions. We might want to make such a revision if we discover that we don’t, upon reflection, endorse the interests promoted by our current practice’s function. (There is likely a teleological explanation of why people use certain ostracizing slurs, but at least some of us don’t reflectively endorse the apparently tribal interests that such usage serves.) Anyway, the function I’m going to propose for our epistemically evaluative practice serves an interest that we all would endorse promoting even upon reflection. So, we should be happy to go on employing our old language, expressing our old concepts, to assert whatever verdict about our coffee addict, Joe, turns out to serve the proposed function.

2.2. Pursuit of Truth through Reliable Belief-Forming Practices

To lead up to the main proposal I want to make and defend about the function of our epistemically evaluative practice (the offset claim labeled ‘Function’ at the end of this subsection), let me start off by examining and developing the following initial suggestion: we make epistemic evaluations because it efficiently promotes our interest in having true beliefs on subject matters of practical importance. I think this suggestion is ultimately correct, but the story is more complicated than it first seems.

The substantive claim contained in this suggestion is that we make epistemic evaluations because it efficiently promotes our having true beliefs on subject matters of practical importance. It’s trivial that having these beliefs really is in our interest, because I use “practical importance” just to mean those subject matters that it’s in our interest to have the true view on. True beliefs have obvious potential to serve our interests, at least if it’s a remotely accurate description of us that we act in ways that would fulfill our desires were our beliefs true. (Many philosophers, functionalists, say this interconnection between belief, desire and action is a necessary truth about any believer/desirer/actor;[7] here we’re only insisting on the milder claim that it’s at least actually true of us.) Though we occasionally have desires that are either neutral or detrimental to our interests, by and large we desire what’s in our interest, and consequently we hold beliefs on subject matters that, if those beliefs were true, would lead us to fulfill those desires, and thereby promote our own interests. By and large, then, the subject matters we care about are ones of practical importance. And then, automatically, the interest that I’m suggesting epistemic evaluations promote is something we will, by and large at least, endorse promoting, even upon reflection.

So, what can be said in support of the suggestion that our epistemically evaluative practice efficiently promotes our interest in having true beliefs on practically important subject matters? Well, from one broad perspective, it may seem obvious enough, for it may seem obvious that most linguistic communication has this function. Ordinary linguistic communication serves to transmit information from speakers to audiences, and by and large this benefits participants to the practice by efficiently spreading true beliefs, beliefs which then cause actions that promote whatever our desires are, and thereby, presumably, promote our interests (the presumption being that we desire what’s in our interest). But, actually, I don’t think the epistemically evaluative aspect of our overall linguistic practice serves its function in this straightforward way. Let me explain why.

True beliefs serve our interests when they are true beliefs about the means to the fulfillment of our desires and we desire what’s in our interest. So, for a really easy example, a true belief about how to get food concerns a means to fulfilling a desire for an outcome, an outcome that is in our interest. But a true belief about how to be rational does not concern a means to an outcome that is in our interest. We don’t have a direct interest in being rational. I know that claim—my last sentence—will initially strikes many people as wrong, but I think it sounds wrong only so long as we unreflectively associate being rational with being reliable. Being reliable certainly is in our interest, since being reliable just is having a tendency to form true beliefs. But, I do not see how being rational is at all in our interest independently of our interest in being reliable. It normally is in our interest to have rational beliefs, but not ultimately for any reason aside from our interest in reliability and, more fundamentally, truth. So, there is actually quite a puzzle: why do we form beliefs and why do we communicate information about how to be rational, if our interest is only in how to be reliable?

The answer is that the teleological explanation of why we make epistemic evaluations is a separate explanation from the more common and obvious teleological explanation of why we engage in most linguistic communication. Most, or at least much, ordinary communication on a subject matter is teleologically explained by our interest in true beliefs on that subject matter; a true belief on that subject matter helps us act in ways that fulfill our desires, and thus, normally, serve our interests. Epistemic evaluations, however, are not explained in this way. We do not make epistemic evaluations, we do not tell others what’s rational and what’s irrational, because having true beliefs on that subject matter fulfills our desires, and thereby serves our interests, in the way other beliefs normally do. The way epistemic evaluations promote our interest in true belief on matters of practical importance is more indirect.

I want to argue that we make epistemic evaluations because it influences the belief-forming practices of the community, influences them in a way that promotes the safety of the transmission of true beliefs via testimony, where “safe” characterizes a practice that does not easily lapse into spreading false beliefs. The idea is that epistemic evaluations play a supporting role: ordinary linguistic communication, ordinary testimony, has the primary function, that of transmitting true beliefs on subject matters of practical importance, while epistemic evaluation has the ancillary function, that of supporting the safety of the transmission of true belief by ordinary testimony. If ordinary testimony can easily lapse into spreading false belief, it is a dangerous tool for promoting our interests; it is unsafe. We benefit from keeping testimony safe. To do this, we need a way of guarding against false testifiers. Our practice of epistemically evaluating each other serves as one such guard. Let me explain how.

There are two ways to guard ourselves against false testimony. The first way is to flag the testimony as false or unreliable, and thereby discourage anyone’s trusting it. But, if epistemic evaluations serve a valuable function, it cannot be, or rather it cannot merely be, that of flagging false or unreliable testimony, because this job can be done most effectively using evaluations of others as simply unreliable. Ordinary people often do this using plainer language than ‘reliable’; they might just say “he doesn’t know what he’s talking about” to communicate the point that anything he gets right is by luck. This is the basic idea of Edward Craig’s view of the function of knowledge attributions, which I am strongly inclined to accept.[8] Perhaps calling someone irrational is a way of communicating, in many ordinary contexts, that he is unreliable, but it is not the only, much less the best, way. So, this cannot be the primary function served by epistemic evaluations, because then they would do nothing that evaluations of others as (un)reliable, or as (non-)knowers, already do just as well or better. (I’ll return to this point shortly, in Section 2.3 below.)

It’s the second way of guarding against false testimony that I propose epistemic evaluations are uniquely suited to do. The second way is to influence the testifier to modify his or her belief-forming practice, so as to improve the reliability of his or her testimony. This is the proposal about the function of epistemic evaluations I want to defend and will be relying on in addressing the problem of forgotten evidence, so let’s put it front and center:

Function only claims that epistemic evaluations help keep testimony safe. Function does not suggest evaluations are the only or even the primary means by which beliefs and testimony are made as reliable as they actually are. What Function claims is that the primary function of evaluations is to help make belief and testimony reliable. Something else might play a fundamental role in establishing our reliability. (Maybe we have innate belief-forming mechanisms that are by default highly reliable, maybe due to evolution.)

Also note that the explanation offered by Function assumes testimony is typically sincere, that is, expresses the speaker ’s actual beliefs, or else that audiences can reliably discern whether or not testimony is sincere. I won’t argue here for this background assumption of the offered explanation.[9]

2.3. Arguing in Support of Function

2.4. Clarifications about Function

Before proceeding to the next section, where we’ll return to directly tackling the problem of forgotten evidence, let me add a few clarifications about what, in endorsing Function, I have not said here about the teleological explanation of our epistemically evaluative practice. Skip this subsection if you want to skip subtleties.

First clarification: I’ve intended to argue in support of Function as describing the primary function of epistemically evaluative language. I’ve only made a defeasible case for this. It would be an objection to my view if it could be shown that epistemic evaluations have some other primary function, if something else better teleologically explained why we engage in this practice. But my view is consistent with the possibility that epistemic evaluations usefully contribute to other secondary functions, including the function that’s primarily served by assertions about reliability. Likewise, other terms, for example knowledge attributions, might contribute to the function that Function describes for epistemically evaluative language, though I’d argue (given my sympathy, already noted, for Edward Craig’s view) that knowledge attributions do not have this for their primary function. The present clarification is important because, if we observe some temptation to use epistemic evaluations in either of two incompatible ways, as perhaps in the problem of forgotten evidence, we should resolve the tension by taking the correct use of epistemic evaluations to be the one that serves their own primary function.

Second clarification: I said earlier that a candidate teleological explanation succeeds in explaining why a practice exists if, and only if, the practice efficiently promotes an interest that it is independently plausible that the participants would strive to promote. The interest cited in my proposed teleological explanation for our epistemically evaluative practice is our interest in having true beliefs. As already noted, this is an interest we will endorse promoting even upon reflection, and thus if—and when—we find that a certain response to the problem of forgotten evidence better serves this interest, that’s the response we’ll have good reason to continue engaging in. The clarification I want to add here is that, for all I’m saying or need to say in this paper, it can be (and very likely is true) that our epistemically evaluative practice serves some much deeper interest of ours, one served via our interest in true belief. Perhaps the teleological story I’m starting here is best completed by giving an evolutionary explanation of our use of epistemically evaluative language as promoting our survival interest, perhaps also adding that our epistemically evaluative practice is genetically selected and innate (similar to a Chomskyan account of our general faculty for language). This is consistent with what I’ve claimed, as long as true beliefs tend to promote survival. But I don’t want to make any claims concerning our evolutionary history, since, (a) I don’t see that I need to for my purposes in this paper, and (b) for all I know, it will turn out that the best complete teleological explanation of our epistemically evaluative practice is something else, something unexpected; for example, maybe epistemic evaluation is merely a product of some sort of non-genetic “cultural” evolution.

Third clarification: I have not endorsed the following view, which I’ll call instrumentalism. Instrumentalism says a belief is epistemically rational for a subject if, and only if, the subject possesses the goal of having true beliefs and it is thereby instrumentally rational for the subject to hold the belief. Kelly (2003) gives a compelling critique of instrumentalism. Kelly points out that in some cases, while it is epistemically rational to hold some belief, it is not instrumentally rational. I happen to agree with Kelly. My proposal has been a different one from instrumentalism. I only claim that it serves an interest of ours, an interest served by having true beliefs, to engage in our practice of epistemically evaluating each other. I do not even claim that serving this interest is a goal we possess in any psychologically real way. (I did claim that true beliefs serve our interest when the things we desire, i.e., goals we possess, are also in our interest, but this doesn’t undermine the present point. We don’t need to positively desire, to have as a goal, true belief. We need only desire things in our interest, like food, and then having true beliefs about the means to those goals will be in our interest.) I do not claim that every manifestation of our epistemically evaluative practice serves an interest of ours. I only claim that the practice as a whole serves our interest. One could argue that some other practice could advance our interests better than our actual practice; but, unless the practice is equally efficient at serving our interest, this will not obviously offer a better teleological explanation of our actual practice. A practice that would require much more effort or more resources to engage in would not be clearly more efficient. (Consider, for example, a variant of our actual practice, one which involves withholding evaluations when certain calculations would suggest the evaluation will be unlikely to have much effect.)

Fourth clarification: I have not endorsed the following view, which I’ll call epistemic belief-consequentialism. This view says that a subject’s belief is rational if, and only if, it serves some “epistemic goal” for that subject to hold that belief. One paradigmatic example of what might be treated as the epistemic goal is true belief, and you might easily think that I have here endorsed a version of epistemic consequentialism with true belief as the epistemic goal. That would be a misunderstanding; I have not endorsed that view. I’ve endorsed only a view of epistemic evaluation. I’ve claimed that we engage in a way of criticizing and praising that efficiently promotes a certain epistemic goal, the goal of true belief. This claim is false if, and only if, our overall practice of making evaluations actually does fail to promote this interest. Epistemic belief-consequentialism, on the other hand, will be false if, and only if, it is possible for a belief to be rational while the subject’s holding that belief fails to promote any epistemic goal. The following combination is perfectly good: we actually promote our interests, promote them quite efficiently, by engaging in an epistemically evaluative practice that has us, in some cases, criticize (praise) some individual beliefs the subject’s holding of which does (doesn’t) promote some (any) epistemic goal. I think this actually happens in cases, including BonJour (1980)’s unwittingly reliable clairvoyant: the clairvoyant’s having these beliefs serves the goal of true belief, but we criticize these beliefs. Another example is the unwitting victim of a Cartesian demon: we may praise false beliefs of his as rational even while his holding these beliefs does not serve, but harms, the goal of true belief (as well as any other plausible goal). The overall evaluative practice that includes these particular evaluations still efficiently promotes true belief, via its function of promoting the safety of testimony, which is so effective at spreading true beliefs throughout the community.

The case of the clairvoyant also illustrates the falsity of another view I’ve not endorsed, a view we can call epistemic rule-consequentialism. This view says that a subject’s belief is rational if, and only if, the subject formed the belief by following a rule, the following of which serves some “epistemic goal”. It seems to serve the paradigmatic epistemic goal, true belief, for the clairvoyant to follow her highly reliable rule. Yet her beliefs are not rational. And it might (and I believe it does) promote true belief to criticize the clairvoyant.[20]

Selim Berker (2013) has recently given extensive attention to clarifying and criticizing a broader meta-epistemological view that includes what I’ve just called epistemic belief-consequentialism and epistemic rule-consequentialism. The target of Berker ’s critique also includes the following view, which I’ll call epistemic recommended-rule-consequentialism. This view says that a subject’s belief is rational if, and only if, the subject formed the belief by following a rule, a rule from the list of rules which it serves some “epistemic goal” for us to exclusively praise. My view is that this biconditional is actually true. But, while I agree with the truth of this biconditional, I don’t see it as offering any explanation of why this or that belief or rule is rational. My explanatory aims are only to explain why we make certain evaluations. I’m in fact not disagreeing with Berker ’s view, since his aim is to reject consequentialist explanatory theories of epistemically rational belief.[21]

3.1. Applying Function to Joe the Coffee Addict

Let’s, finally, turn back to the problem we started with. The problem of forgotten evidence was a dilemma, one I posed using the example of a coffee addict, Joe, who formed a belief, on the basis of wishful thinking, that drinking coffee is healthy, but has since entirely forgotten how he formed this belief.[22] We saw that there are challenges facing either way of answering the question: is his belief now rational or irrational? To solve the dilemma using the practical approach, the question we turn to now is this: which evaluation would, or would more efficiently, serve the function of epistemic evaluation, as described in Function? (As an intuitive shorthand, I’ll now often speak of “serving Function” to mean serving the function that’s described in Function.)

Reminder: the larger argumentative strategy relies on the premise that evaluations that serve Function will be correct applications, ones that apply our word or concept (of rationality) to their extension. Thus, we will ultimately be in a position to draw a conclusion about the rationality of Joe’s belief, a conclusion we state by using our word ‘rational’ or our concept of rationality. And we’ll want to continue to use this word, we’ll want to continue to make evaluations that express this concept, because we of course endorse the goal of Function, namely the goal of promoting true belief.

Function said that we promote true belief in a particular way. It said that epistemic evaluations help protect us from unreliable testimony, not by excluding unreliable testifiers from the circle of trust, but rather by influencing unreliable testifiers to modify their beliefs and, thereby, modify their testimony. Now, Joe certainly is an unreliable testifier: he engaged in wishful thinking, an unreliable belief-forming practice, and he has now forgotten what his evidence was (if any) for his belief that coffee is healthy, so he cannot self-correct his unreliability. This may seem to suggest that Joe is a prime candidate for someone whose belief-forming practices, and thus whose testimony, would benefit from correction by the epistemic evaluations of others; Joe may seem to be someone whose belief, according to Function, we should criticize.

I will argue that is wrong. Criticism of Joe’s belief would not help improve the reliability of his own, or anyone else’s, beliefs and testimony. Not every instance of unreliability is treatable by epistemic evaluation. Sometimes unreliability can only be treated, or would be better treated, by a non-normative assertion, such as an assertion about reliability; sometimes it cannot be treated by any assertion. In the case of Joe, I’ll argue that epistemic evaluations, which are normative, would not contribute to improving the reliability of either Joe’s, or anyone else’s, testimony.[23]

3.2. Criticizing Joe Won’t Serve Function

3.3. What’s So Bad about Laundered Beliefs?

To fully satisfactorily resolve the problem of forgotten evidence, more needs to be said than just a positive argument for one side. The problem is a dilemma, either horn of which immediately strikes us as counter-intuitive. I have given an argument that Joe’s belief is rational, but, even if that’s completely right, this may still leave us with our original discomfort with the thought that Joe can get away with laundering his belief. It is one thing to learn which horn of a dilemma is true, and even to know why it is true; it’s another thing to feel no intellectual discomfort making this judgement. A fully satisfactory resolution of our problem, then, would not just convince us that Joe can launder his belief, but would also resolve our discomfort with this judgment. We can do this if we can explain why we misleadingly feel tempted to reject this judgment. And there are several points that have emerged over the course of our discussion that begin to explain this.

Even as we’ve argued that Joe’s belief is rational, we’ve emphasized a number of ways in which Joe exhibits cognitive defects. Joe formed his belief about coffee in an unreliable way. Unreliability is a defect, a cognitive defect worth discouraging. Joe also forgot how he formed his belief. Forgetfulness can also be a cognitive defect, one that it is often, though not always, worth discouraging. Of course, no one can remember everything, and there are certain things it’s a waste of cognitive resources to bother remembering, but it’s still true that forgetfulness is often harmful and worth discouraging.

These are defects of Joe’s cognitive state, defects worth discouraging linguistically or by other means. If part of our language effectively functions to discourage these defects, then it may naturally require extra care not to confuse the role of that language with the role of the kind of language I’ve narrowly focused on here, the language I called (in Section 2.1) our epistemically evaluative practice. We certainly do have other language, language other than epistemic evaluations, that partly functions to discourage unreliability and harmful sorts of forgetfulness. “Think carefully!”, “Pay attention!”, “Gather up plenty of evidence!”, “Double-check your conclusions!”, “Avoid sloppy reasoning!”, to give only some simple illustrations.

Not only are Joe’s cognitive defects worth discouraging, they are worth flagging to others. In Section 2.2, I said that we have two broad ways of promoting the safety of testimony. One is to encourage reliable belief-forming practices (the function of epistemically evaluative language), and another is to flag unreliable beliefs and belief-forming practices. I said that I, with many others, find Craig’s proposal plausible: knowledge attributions serve the function of flagging reliable and unreliable sources of testimony for the safe consumption by others. Joe’s coffee belief, even if it is true, is not knowledge.[25] It is thus appropriate to say Joe doesn’t know that coffee is healthy, and his belief and his testimony are thereby flagged as unreliable and untrustworthy.

It is easy to confuse these defects of Joe’s belief with irrationality. And it is easy to confuse the effective ways we do have of linguistically responding to these defects with the ways we use epistemic evaluations for their own purpose. But, I’ve argued that epistemic evaluation serves a subtly but importantly different purpose from other language, especially reliability attributions and knowledge attributions. Careful reflection on these distinctions can help loosen the grip of our initial temptation to balk at Joe’s laundering of an irrational belief just by forgetting his evidence.

In fact, the whole metaphor of laundering beliefs is part of the cause of the confusion! The misleading intuition in the original dilemma is pumped by figurative talk of laundering. But, now we can better see that laundering a belief is one thing, and a launder-ed belief is another. It is appropriate to criticizing launder-ing: we might do this by criticizing Joe for forgetting his past wishful thinking. But, it’s not the function of epistemic evaluations to evaluate forgetting. They serve to evaluate the launder-ed belief, the clean belief.

3.4. Objection and Reply: On an Alternative Application of Function

I’ve argued that a practice which has us not criticize Joe better serves Function than one which has us criticize him. Given my argument that our epistemically evaluative words are tailored to efficiently serve Function, this is evidence that our word ‘rational’, and our concept of epistemic rationality, really have Joe’s belief within its extension. And, given that we do, even upon reflection, want to serve Function as best we can, we have reason to continue to use this word and this concept, and to evaluative Joe’s belief as rational.

In this subsection, I’ll consider a potential objection.[26] I want to consider an account of our epistemically evaluative practice, one which includes criticism of Joe as irrational, and which it might be argued serves Function as well as, or better than, my proposal above. I’ll describe the proposed account, highlight why it may appear to serve Function better, and then argue it serves Function more poorly. So, since it is poorer, it is not likely that our actual practice, which is tailored to serve Function efficiently, operates in this way. Nor will we want to switch to using language and concepts that do operate in the described way.

Let’s name the new proposal to be considered sneak-catcher, and to introduce it and explain its name, let me introduce Sneaky, one more character to compare against Joe. Sneaky, like Joe, formed a belief that coffee is good for you purely on the basis of what she then knew was wishful thinking. Unlike Joe, Sneaky still remembers she formed her belief in this way (and has since acquired no new basis for this belief). Sneaky, though, now presents herself as though she doesn’t remember at all how she formed her belief. Thus, Sneaky’s belief is not exactly like Joe’s, but it’s difficult for people who interact with her to distinguish her case from Joe’s.

Given all the facts about her, I take Sneaky’s belief to be uncontroversially irrational. And, I take it that it best serves Function to criticize beliefs like Sneaky’s: managing one’s beliefs as she does is an unreliable method that we effectively suppress using epistemic evaluation.

The sneak-catcher view, now, says that we most efficiently promote reliability overall, and thus efficiently serve Function, by criticizing both Sneaky and Joe.

What can be said in favor of the sneak-catcher view? The apparent argument is this. It serves Function to criticize Sneaky. But, any rule that recommends we don’t criticize Joe and do criticize Sneaky will be hard to conform to, hard to accurately implement in practice: we will frequently make errors either by letting Sneaky off the hook or by wrongly accusing Joe. Thus, to ensure we don’t let Sneaky off the hook, we should adopt a rule for criticizing both, and this rule will be easy to implement without such frequent errors.

It may be helpful to note there is an analogy here to strict liability in the law. Laws imposing strict liability hold a person responsible for a harm she causes, regardless of whether or not the harm was foreseeable. Now, strict liability in the law might have a pragmatic justification: in order to catch and punish those who could have foreseen harms they caused, we find it necessary, or simply worth the costs, to implement a rule that also has us punishing others who could not have forseen harms they caused. But, strict liability does not have the same appeal outside the legal context. To many ethicists, the sneak-catcher view’s analogy with strict liability will immediately cast the view in an unfavorable light. As Thomas Nagel has commented, in an ethical context, “If the object of moral judgment is the person, then to hold him accountable for what he has done in the broader sense is akin to strict liability, which may have its legal uses but seems irrational as a moral position.” (1976/79: 31); and Gideon Rosen has likewise said, ‘But in morality there is no such thing as strict liability’ (2004: 301).

Here is why I don’t buy the above apparent argument for sneak-catcher (aside from a temptation to dismiss it out of hand, perhaps as Nagel and Rosen would). Compared to a rule for criticizing both Sneaky and Joe, it’s no harder to implement the following pair of rules: 1) as long as you can’t tell them apart, criticize both Sneaky and Joe, but 2) when you are able to tell them apart, just criticize Sneaky, not Joe. This pair of rules is no harder to jointly implement (no harder than just following (1), which is just sneak-catcher), and the effect of following both rules dominates following just the one in terms of serving Function. Certainly, as theorists who are considering Joe, having stipulated all the facts about his forgetting whatever evidence he did or didn’t have for his belief, we can apply rule (2), and decide not to criticize him. Thus, I stand by the diagnosis that Joe’s belief is rational.