On the Epistemic Costs of Friendship: Against the Encroachment View

Abstract

I defend the thesis that friendship can constitutively require epistemic irrationality against a recent, forceful challenge, raised by proponents of moral and pragmatic encroachment. Defenders of the “encroachment strategy” argue that exemplary friends who are especially slow to believe that their friends have acted wrongly are simply sensitive to the high prudential or moral costs of falsely believing in their friends’ guilt. Drawing on psychological work on epistemic motivation (and in particular on the notion of “need for closure”), I propose a different picture of what friendship requires in the doxastic realm. I argue that contrary to what the encroachment strategy suggests, exemplary friends’ belief formation ought not be guided by a concern with accuracy or error avoidance, but instead by a need to avoid a “specific closure” – namely, a need to avoid concluding in their friends’ guilt. I propose that exemplary friendship often generates a defeasible, doxastic obligation to exemplify such a need, despite its inherent corrupting effects on exemplary friends’ epistemic faculties.

It is widely acknowledged that friendship places special demands on our actions: we drive our friends to the airport in the middle of the night, help them move apartments, etc. We do things for them that we would never do for strangers. But can the requirements of friendship also extend into the doxastic realm, spilling over from the domain of action into that of belief? Some have recently answered that question in the affirmative,¹ relying on the following kind of case:

Phone Calls ²

John tells one of his colleagues Emma that Emma's friend Laura “knowingly broke her most recent date's heart”, by refusing to return any of his calls after having slept with him. Upon hearing John's story, Emma is perplexed. Surely, she thinks to herself, Laura must not have done what John reports. Or if she has done it, she must have had a good reason – for instance, she might have had already explicitly stated that she did not want to be called back.

In this case, a friend gains compelling evidence indicating that her friend has behaved badly and suspends judgment. In being wary to reach a certain conclusion about her friend and in looking for exonerating explanations of her behavior, Emma seems to act like any loyal friend would. The fact, however, that a stranger would probably react differently appears to cast doubt on her epistemic rationality: if a stranger would readily believe that Laura has misbehaved on the basis of John's testimony, how can Emma still suspend judgment? Couldn't it be because she has ignored part of her evidence and is thus epistemically irrational?

Relying on cases structurally similar to Phone Calls, Baker (1987), Stroud (2006), Keller (2004), and Hazlett (2013) have argued for the possibility that friendship constitutively requires epistemic irrationality. On their view, not only does friendship typically generate certain doxastic obligations – in particular, obligations to suspend judgment on the proposition that our friends have behaved badly. Fulfilling some of those obligations can only be done at the expense of ignoring part of our evidence and thus being epistemically irrational. Baker, Keller, Stroud, and Hazlett urge us to acknowledge the existence of widespread tragic conflicts between the norms of friendship and epistemic norms; between our lives as socially embedded agents and our lives as epistemic agents.³

This paper aims to defend this “irresolvable conflicts view”⁴ against a recent, forceful challenge. This challenge is raised by proponents of pragmatic and moral encroachment, who hold that the prudential or moral costs of being wrong as to whether p can raise the evidential threshold for having epistemic justification for the belief that p.⁵

Defenders of pragmatic and moral encroachment have argued that contrary to appearances, cases like Phone Calls are not ones where the requirements of friendship and those of epistemic rationality conflict.⁶ This is because, as they argue, the intuitively “exemplary friend” is in fact epistemically rational when she suspends judgment on claims that strangers would readily believe. The exemplary friend is simply sensitive to the high prudential or moral costs of being mistaken as to whether her friend has behaved badly: since falsely believing that her friend has done something wrong could damage her relationship with her or cause her friend moral harm, the exemplary friend ought to possess especially strong evidence before she can justifiably form such a belief. Because strangers do not face similar prudential or moral costs, their belief in the accusations can be just as epistemically rational as the exemplary friend's suspension.

My argument against this strategy for reconciling the demands of friendship with epistemic norms will proceed as follows. In Section 1, I sketch the motivations behind the “encroachment strategy” and explain why original defenses of pragmatic encroachment allow both prudential and moral considerations to affect epistemic justification. Then, in Section 2, I distinguish two ways of defining what has come to be known as the “moral encroachment” thesis. One can first model moral encroachment directly onto pragmatic encroachment and hold that the moral costs of being wrong as to whether p can raise one's evidential threshold for epistemic justification. Or one can be more liberal, and allow that the mere moral costs of coming to believe that p (be that belief true or false) can raise the amount of evidence it takes to justifiably believe that p. I argue that we should reject the latter construal of the moral encroachment thesis, on pain of countenancing forms of irrational wishful thinking, a notion which I clarify in Section 2. Drawing on the “friend as honest advocate” framework of what friendship requires in the doxastic realm (as articulated by Keller (2004) and Stroud (2006)), Sections 3 and 4 then argue that in Phone Calls and similar cases, friends typically incur important moral costs as soon as they form the belief that their friends have behaved badly, and not only when that belief turns out false.

Since the core cases figuring in the current debate over friendship and epistemic norms do not share the structure of classic pragmatic encroachment cases, I conclude that we cannot appeal to the notion of moral encroachment (which, when understood correctly, is a direct extension of pragmatic encroachment) to vindicate the epistemic rationality of exemplary friends. We should instead accept that exemplary friends who suspend judgment on whether their friends have behaved badly are often engaged in irrational wishful thinking, and thus recognize that friendship can constitutively require epistemic irrationality.

1. The pragmatic encroachment proposal

When confronted with cases like Phone Calls, one might be tempted to insist on the rich evidence we possess about our friends and their moral character. We indeed know more about our “significant others”⁷ than we know about strangers. This might go some way toward explaining the apparent doxastic asymmetries mentioned in the Introduction: one reason why, in cases like Phone Calls, the friend appears especially slow to suspend judgment on claims that others would readily believe could simply be that she has much more relevant evidence to draw from, before she rationally can come to a conclusion. Without such a rich evidential pool to rely on, the stranger isn't in a position to match the friend's accuracy – hence their differing judgments.

Despite being promising, such a line of argument cannot accommodate the full range of cases where friendship seems to force the adoption of non-evidentially supported doxastic attitudes. As Baker (1987: 3) points out, our friends sometimes face extraordinary accusations, such that “there is no prior set of tests or testing situations that [the friend] has come through with flying colours”, and which would rationalize suspension. Baker (1987: 3) takes the example of someone being accused of selling secrets to a foreign government. She claims that in such a case, our evidential position as friends, relative to the question at hand, is not significantly better than that of a stranger. And yet, we still suspend judgment. To put the “differential evidence explanation” to the test, we can also, like Stroud (2006: 516–18), devise cases where, by hypothesis, the friend and the stranger share all the relevant information. Stroud asks us to imagine a historian doing research on a historical figure, who possesses an impressive body of evidence in favor of that person's good character. When faced with incoming evidence that his subject of study has not behaved well, we can easily imagine, insists Stroud, the historian nonetheless coming to different conclusions than the figure's friends.

A satisfying explanation of why, despite the initial appearances, exemplary friends in Phone Calls are in fact epistemically rational should shed light on Baker's and Stroud's difficult cases. Moreover, other things being equal, it should also avoid depending on the controversial assumption that practical or ethical considerations can serve as legitimate grounds on which to hold a belief or suspend judgment.⁸ Of course, if the fact that someone is your friend can count as a normative reason to withhold judgment on the accusations made against them, then we easily see how the apparent conflict between friendship and “epistemic” norms dissolves. Instead, we would rather be able to classify Phone Calls and similar cases as ones where friendship does not force us into epistemic irrationality without having to give up on evidentialism, the view that epistemically rational doxastic attitudes are held solely on the basis of the evidence.⁹

The “encroachment explanation” promises to meet the two desiderata just outlined. As its name implies, it relies on the thesis of pragmatic encroachment to try to dissipate appearances of normative conflicts in our core cases. Kawall (2013), one of its main proponents, focuses on subject-sensitive invariantism, the view that when the stakes are raised for an agent (i.e., when there are high practical costs for her to being wrong), a stronger epistemic position is required for her to know or have justified beliefs.¹⁰

Proponents of SSI argue for their view in part by appealing to pairs of contrastive cases, one “High Stakes” and one “Low Stakes”.¹¹ Kawall himself cites the example of Claire and her friend, who are now at the bakery (2013: 364–5): Claire's friend asks the employee whether there are any peanuts in the baked goods that they are about to buy; the employee says that there aren't any. It then seems that Claire's friend can form a justified true belief that there are no peanuts on the basis of the employee's testimony. Such a statement does not seem to hold true of Claire, however. For Claire is severely allergic to peanuts, and her friend is not. Since it is thus crucial for Claire to avoid falsely believing that there are no peanuts in the items she buys, she is much more careful than her friend. She requests more evidence than her friend to conclude in the absence of peanuts and withholds forming a belief (“there are no peanuts in these baked goods”) that others would typically form. In doing so, argues the defender of SSI, Claire is not epistemically irrational. She is simply responsive to the high stakes of her practical situation. Since she is in a “High Stakes” case, she needs more evidence than her “Low Stakes” counterpart to justifiably believe.

Kawall argues that the exemplary friend's epistemic and practical situation is strictly analogous to Claire's. Because the good friend can incur important practical costs if she reaches false conclusions about her friend's past behavior, she needs to be in an especially strong evidential position to rationally form negative beliefs about her. Drawing on Keller (2004: 329–30) and Stroud (2006: 511), Kawall points out that damaging valuable relationships counts as an important practical cost of believing falsehoods about our friends’ past actions (2013: 364). First, by coming to endorse negative claims about our friends’ past behavior, we might grow detached from them: we might interpret their actions as a sign of the viciousness of their moral character, and thus become motivated to end relationships that we would otherwise have maintained. We can engage in genuine, valuable friendships with people whom we do not consider morally virtuous, but seeing someone as having a bad moral character will often foreclose the possibility of ongoing friendship. In addition, if our friends find out that we entertain negative false beliefs about them, they might feel betrayed: they expected us to see them in a better light and view our conclusions about their actions as manifesting a lack of trust. Either way, the formation of negative false beliefs about our friends’ past actions seems to carry important costs: both parties involved stand to needlessly lose a valuable relationship. According to the encroachment strategy, insofar as we see the maintenance of friendships as an important facet of the “good life”, and insofar as we also care that our loved ones enjoy friendships, we have strong reasons to be especially careful when forming beliefs that could undermine our significant relationships. On the present picture of the exemplary friend's epistemic practices, Emma is – just like Claire – in fact engaged in “careful epistemic work” (Kawall 2013: 365): she is simply especially cautious to avoid forming a false belief that could have negative consequences both for herself and someone she loves.

When developing the encroachment strategy, one should bear in mind that the relevant costs for the exemplary friend of forming negative false beliefs about her friends do not have to be described as solely “prudential”. The encroachment strategy indeed leaves room for other practical costs – besides those redounding to an agent's well-being – to affect her epistemic justification. These costs plausibly include the detrimental consequences, for other people the agent cares about, of her forming false beliefs. Original defenses of pragmatic encroachment were often focused on prudential costs for the believer. But one of the main arguments for pragmatic encroachment on “knowledge-level” epistemic justification – the argument connecting epistemically justified belief with acting “as if” – can be formulated by appealing to a broader notion of practical costs, which plausibly encompasses negative consequences for others as moral considerations.¹² This argument starts from the premise that having knowledge-level epistemic justification for p suffices for being licensed to rely on p in one's practical reasoning.¹³ It then notes the existence of pairs of cases (such as those involving Claire mentioned earlier) where the epistemic position of both subjects with respect to p is identical, their practical situations differ, and only one subject may rely on p in her practical reasoning. The conclusion is that a difference in practical interests can entail a difference in epistemic justification. We can see why such an argument can extend to a variant of the cases discussed by Kawall so as to highlight the possible relevance of moral considerations for epistemic justification: if we imagine Claire buying baked goods for an allergic friend and contrast her practical situation with that of another agent in the same epistemic position who buys baked goods for a non-allergic friend, it still seems that Claire ought to adopt an especially demanding evidential threshold for the proposition that there are no peanuts in the baked goods. Because the practical stakes of relying on p include consequences for others than ourselves, the pragmatic encroachment thesis might seem to offer a promising account of our central friendship cases.

Unlike the differential evidence explanation, the encroachment strategy has the merit of trying to encompass the variety of cases in which friendship seems to force us into epistemic irrationality. It also aims to do without the controversial assumption that practical considerations can serve as legitimate grounds on which to hold a belief or suspend judgment. But we should nonetheless reject the encroachment strategy as a way of establishing exemplary friends’ epistemic irrationality. The encroachment explanation portrays exemplary friends as especially cautious to avoid false beliefs and as concerned with accuracy, whereas – as I will argue – exemplary friends in fact ought to be motivated to avoid coming to certain conclusions about their friends altogether. Since exemplary friends can face a “moral cost” as soon as they conclude in their friends’ guilt (and not only when they reach false conclusions about them), the structure of our paradigmatic friendship cases and that of encroachment cases will prove to be very different.¹⁴

2. Moral costs and moral encroachment

We need to ensure that we are focused on the best version of the encroachment strategy before vindicating the “irresolvable conflicts view” against it. I have just explained that original defenses of moral encroachment allow moral (and not only prudential) costs of error to impact epistemic justification. Some authors have recently singled out this implication of original accounts of pragmatic encroachment, using the term “moral encroachment” for the thesis that the bad moral consequences of falsely believing that p can impact whether someone has justification or knowledge that p.¹⁵ Other philosophers, however, have used the term “moral encroachment” more liberally. Instead of viewing “moral encroachment” as a straightforward extension of pragmatic encroachment, these philosophers have insisted that the very entertaining of some beliefs involves certain “moral harms” or leads to bad moral consequences, even when those beliefs turn out true. According to these philosophers, the bad moral consequences that obtain regardless of a belief's truth or falsity can prevent that belief from counting as epistemically justified.¹⁶ As we shall see, we should favor the former construal of the moral encroachment notion, even if it still cannot be used to accommodate our core friendship cases. Agents who raise their “evidential thresholds” in response to the very costs of forming a belief (true or false) should be seen as engaged in a form of epistemically irrational “motivated thinking”,¹⁷ and not as rationally suspending judgment.

Defenders of both varieties of encroachment views just mentioned deny that adopting non-racist and non-sexist attitudes can require “epistemic irrationality through base-rate neglect”.¹⁸ Drawing on cases such as the following, they instead argue that it is always possible – at least in principle – to adopt doxastic attitudes that are morally acceptable while counting as impeccable believers, who form their beliefs on the basis of statistical evidence about race and gender:

College Graduation

Julia, a college professor aware of the very low graduation rates for black men at her institution, meets Andre, an incoming black student. One of Julia's colleagues then asks her whether she thinks Andre will succeed in his program. Julia answers that she honestly doesn't know. Despite knowing the graduation rates, she still does not believe that Andre will do badly.

If we accept either variant of the moral encroachment thesis, we don't have to view Julia as epistemically irrational. On the first variant, it is the negative moral status or consequences of falsely believing that Andre won't succeed which explain why the substantial support provided by Julia's statistical evidence still does not justify believing in Andre's future failure. Schroeder (2018) endorses this variant of the moral encroachment thesis, when he argues that beliefs’ moral badness depends on their falsehood, and that only beliefs that “falsely diminish” can morally wrong someone.¹⁹ By contrast, on a second way of construing moral encroachment, the very formation of the belief that Andre will fail is seen as an instance of “doxastic wronging” or as having bad moral consequences, even when that belief actually turns out true. In particular, it might be (as Rima Basu (2019a, 2019b) has argued with respect to similar cases) that the very belief in Andre's future failure reinforces existing patterns of discrimination and therefore needs to be supported by especially strong evidence to count as epistemically rational.²⁰ Since both variants of the moral encroachment thesis stress the impact of beliefs’ moral dimensions on epistemic justification, they might seem especially apt to account for our original friendship cases.

However, as we will now see, defenders of moral encroachment should acknowledge that the best version of their view makes moral encroachment continuous with pragmatic encroachment, by simply emphasizing the epistemic impact of one kind of practical costs (namely, the moral costs) of error as to whether p. Contrary to what defenders of the more liberal version of moral encroachment hold, the moral consequences of coming to adopt a certain attitude cannot make a difference to its epistemic status as justified. But since exemplary friends’ belief-formation is sensitive to such consequences – as I will show when developing the view of exemplary friends as “honest advocates” in Sections 3 and 4, neither variant of the moral encroachment thesis can help establish exemplary friends’ epistemic rationality. Friends who are responsive to the very moral cost of reaching certain conclusions about their friends should instead be seen as engaged in “directional” or “motivated thinking”. As the last section of this paper shows, they can even be better friends for exhibiting such tendencies – which nonetheless count as paradigmatic instances of epistemic irrationality and bias.

To highlight the restrictions posed by the pragmatic encroachment structure on the range of cases where an agent's response to “moral costs” can explain her epistemic rationality, we need to introduce yet further examples:

Low Costs

You just read an online article providing evidence that programmers at a certain technology company do not enjoy good working conditions, despite earning comparatively high wages. You are trying to determine whether you should believe the article's claims. Settling that question won't have any direct practical consequences for you: you are not a programmer, nor are you closely acquainted with any.

High Costs

You just read an online article providing evidence that programmers at a certain technology company do not enjoy good working conditions, despite earning comparatively high wages. You have applied for a programmer position at that company and have been called for an interview. You are trying to determine whether you should believe the article's claims. You know that if you do come to believe its claims, you won't be willing to work for the company anymore. You will thus have to continue your job search, even though the position so far seemed like a perfect fit for you.

I name these cases Low Costs and High Costs following Moss (2018), who has used similar cases to argue against some accounts of moral encroachment.²¹ Whereas classic pragmatic encroachment cases feature what Moss calls “risky beliefs” (namely, beliefs that would lead to significant harm “if and only if they would turn out to be false”), High Costs features a “costly belief”; a belief which would lead to significant harm in virtue of our mere entertaining it, even if it were true (Moss 2018: 195). In the high stakes case discussed earlier, for instance, acting on the belief that the baked goods contain no peanuts will lead to significant practical harm if and only if that belief turns out false. (If the baked goods indeed contain no peanuts, then Claire is not in danger of getting sick.) By contrast, in High Costs, your belief will lead to practical harm as soon as you entertain it, be it true or false: as soon as you believe that the company does not treat its workers fairly, you won't be willing to work there anymore. As a result, you will have to continue your job search, as unpleasant as that might be.

According to Moss (2018: 194–5), the cases cited in various attempts to apply moral encroachment views to the phenomenon of racial profiling share the structure of High Costs.²² Such a realization leaves the defender of moral encroachment with two options: either accept the thesis that agents suffer the distinctive harms associated with “unalloyed racial profiling” only when they are the victims of “false racial profiling”, or extend the traditional notion of pragmatic encroachment, and accept that a belief can fail to count as justified merely in virtue of being “costly”. Moss opts for the former approach, motivating her view that “cases of racial profiling can have just the same structure as classic cases of pragmatic encroachment” (2018: 197) by appealing to various examples of “false racial profiling” and the distinctive wrongs they feature.²³

Moss insists that revising existing accounts of pragmatic encroachment to encompass cases like High Costs would be a mistake. As she notes (2018: 196), “in so far as you are more reluctant to believe the more costly proposition [in a case like High Costs], it seems that you are engaged in irrational wishful thinking”. I hold that we cannot use Moss's kind of move in the present debate over friendship and epistemic norms: we cannot plausibly argue that the beliefs which we can form about our friends in cases like Phone Calls stand to harm them only if they are false, in an attempt to dispel appearances of epistemic irrationality. Instead, I propose that cases like Phone Calls have the same structure as High Costs: they can be construed as cases where we incur a “moral cost” and harm our friends as soon as we believe in their guilt, even when that belief is actually true. The agent in Phone Calls and other cases discussed in the debate over friendship and epistemic norms²⁴ should both be seen as engaged in “motivated reasoning” and as being better friends for having engaged in such biased thinking. It is in that sense that friendship constitutively requires epistemic irrationality.

Before arguing for that conclusion directly, I want to clarify the sense of “wishful thinking” at play in cases that share High Costs’ and Phone Calls’ common structure. I thus suggest that we make a brief detour to the psychological study of “lay epistemics”.

One of the main goals of psychologists studying lay epistemics is to uncover how motivational factors influence both the attainment of “settled belief” or “definite judgment” on a topic (what Kruglanski and Webster (2018 [1996]) call “closure”), as well as hypothesis generation and testing. Kruglanski and Webster (2018 [1996]) distinguish two main kinds of epistemic motivations: the need for non-specific closure and the need for specific closure.²⁵ When one possesses a desire for any firm answer to a question, one has a need for non-specific closure, whereas when one possesses a desire for a specific answer to a question (for instance, an esteem-enhancing answer, an optimistic answer, and so on), one can rightly be attributed a need for specific closure. The need for specific closure is what philosophers tend to think of in discussions of “motivated thinking”: it is the motivational force that drives us whenever arriving at some particular answer is considered antecedently desirable.²⁶

Both the need for non-specific closure and the need for specific closure are assumed to vary in degree, lying on a continuum ranging from a low to a high motivational magnitude.²⁷ Moreover (and importantly for our purposes), both the need for non-specific closure and that for specific closure influence the length and the nature of the epistemic sequence of hypothesis generation and testing. Indeed, a heightened need for cognitive closure instills in individuals the tendency to “seize” on early cues affording evidence and “freeze” on the judgments it suggests.²⁸ By contrast, when we experience a low need for non-specific closure, we engage in a more thorough informational search and avoid seizing on early cues, thus typically achieving greater judgmental accuracy.²⁹ The effects of the need for specific closure on hypothesis generation and testing, on the other hand, are well-known: as Kunda's (1990) work on “motivated reasoning” has shown, subjects with a heightened need for specific closure possess both a tendency to terminate the hypothesis testing sequence when the available evidence appears to yield the desired conclusion, as well as a tendency to keep the sequence going until such conclusion seems implied by the evidence.³⁰ In a search for support, subjects with a heightened need for specific closure credulously accept confirming data for their preferred hypothesis, while subjecting data that would strike outside observers as clear contra-indications to creative but withering scrutiny. They thus display a combination of hypersensitivity to evidence and blindness (which, of course, comes in degrees).

Neither the need for specific closure nor that for non-specific closure have to be consciously accessible to the subject to influence hypothesis generation and testing. The need for specific closure, in particular, often remains under one's radar, surreptitiously influencing the outcome of one's investigations:³¹ someone who takes herself to be solely animated by a desire for truth can nonetheless aim to embrace a particular conclusion, and be surprised to find out that her doxastic activity had been shaped by desire and interest all along. Upon discovering what truly animated her investigations, such a person would be rational in reducing confidence in her conclusions and revising her doxastic attitudes. For she would be in a position parallel to that of an agent learning that someone else has tampered with her evidence: if my learning that you have manipulated my evidence and hidden some facts from my view counts as a reason for thinking that my inquiry has not been solely guided by truth-indicating concerns, why would the fact that I am myself the manipulator in question be any less worrying? The point, from an epistemic perspective, of believing what you take to be best supported by your evidence or reasons is the pursuit of accuracy or truth. But believing in accord with someone's interests – one's own or someone else's – is, all else equal, about as reliable as believing randomly.³² This is how “wishful thinking”, understood as the set of belief-formation practices guided by a need for specific closure, can count as irrational and epistemically suspect.

I hold that cases that have so far been at the center of the debate on friendship and epistemic norms – cases like Phone Calls – are best understood as featuring agents (“intuitively exemplary friends”) that have a heightened need for specific closure, as opposed to a low need for non-specific closure. The view of the doxastic requirements of friendship defended in the following sections explains why friends sometimes ought to exemplify the need to avoid settling on a specific conclusion (in particular, the conclusion that their friends have behaved badly), even if such a need, as we just saw, inherently gives rise to irrational, inaccurate attitudes.

3. Friends as honest advocates

My argument that friends in cases like Phone Calls are best portrayed as fulfilling their friendship-related duties through a need for specific closure builds on the suggestion, first made by both Keller (2004) and Stroud (2006), that exemplary friends have a special, defeasible duty to act as their friends’ “defense lawyers”.³³

Keller, Stroud, and other defenders of the claim that friendship can constitutively require epistemic irrationality have argued that we ought, in order to count as exemplary friends, to stand up for our friends “externally”.³⁴ This is supposed to involve defending them in front of others or, as Stroud (2006: 503) puts it, “defend[ing] [their] reputation in the court of public opinion”. Stroud argues that someone who lets accusations against their friends go unchallenged and who simply sits silent when their reputation is being maligned does not behave like a true friend. Of course, there might exist a general moral duty to combat cruel jokes or gossip directed against anyone. But as Keller (2004), Stroud (2006), and others are keen to point out, there also seems to exist a special, defeasible demand of friendship enjoining us to intervene when our friend's reputation is on the line. Remaining passive while others stain one's friend's image seems disloyal.

Those who accept the claim that friendship can constitutively require epistemic irrationality then go on to point out that exemplary friends ought not only stand up for their friends externally, but must also stand up for them “internally”. To support this idea, we can imagine someone who defends her friend “outwardly” (in the ways just described), while also – “inwardly” – believing the worst of her. Reflecting on this possibility, Keller argues that such a person would clash with our ideals concerning friendship. As he explains, “You want a friend who's on your side, not one who's good at faking it.” (2004: 335). It could be, as Hazlett (2013: 101) suggests, that exemplifying such a split between behavior and internal attitudes “seems to require an unappealing kind of insincerity” – insincerity which stands in tension with our picture of the exemplary friend. The thought, then, is that friends ought to avoid coming to believe that their friends have behaved badly, on pain of not being able to provide them with an honest defense, and thus fail to fulfill their friendship-generated obligations.

To be sure, one does not have to believe in one's friend's innocence so as to count as providing an “honest defense”. That would be overly demanding, for in the very cases at the heart of the debate on friendship and epistemic rationality (such as Phone Calls), good evidence indicates that one's friend has behaved badly. Considering that evidence, belief in their innocence would amount to a stubborn denial of the obvious. As Stroud notes, a disposition to display total imperviousness to one's evidence is not part of our ideal of friendship (2006: 506).³⁵ Instead, what is required of the exemplary friend in the epistemic domain is something much more subtle; a kind of “epistemic slant” in favor of her friends that does not amount to a plain denial of the incontrovertible. By suspending judgment on claims that strangers would endorse, the good friend can fulfill her honest advocacy duties, all the while avoiding complete indifference to damning evidence. This is not to say that she has to stand up for her friends come what may: when facing decisive proof of her significant others’ guilt, even the most loyal of friends should disinvest herself from their innocence. However, since committing to “clearing someone's name” amounts to “having cast your lot” with theirs (or to “standing or falling” with them),³⁶ honest advocates’ attitude of suspension should be especially resilient and withstand all but the strongest of evidence.

To recast Stroud's and Keller's ideas in the psychological terms used in the previous section, we could say that to fulfill her honest advocacy duties, the exemplary friend has to possess a need for specific closure – in particular, a need to avoid concluding that her friend has behaved badly. In fact, in cases at the center of the debate on friendship and epistemic norms, agents seem to engage in the same kind of hypothesis testing and generation as Kunda's (1990) coffee drinkers, who are motivated to avoid believing that coffee is bad for them. Such agents subject data that would strike outside observers as clear evidence for that hypothesis to withering scrutiny. In a vivid passage, Stroud describes the epistemic practices of exemplary friends in the following way (2006: 509):

Characteristically, you might first try to discredit the evidence being presented and find a way not to believe your friend did this at all. If that isn't feasible, then you can accept those base facts and move to the interpretive level, where you try to put a different spin on what he did and file that action under some less damning label. If this proves impossible, then you can link the action to a different character trait than the obvious ones. If you can't in good conscience even do that, then you can seek to embed in a larger virtue the negative character trait you are forced to attribute to your friend … As a last resort, if even this last stratagem fails, you can relegate your attribution of a character flaw to your friend to an obscure corner of your portrait of him, rather than making it the dominant element.

In an attempt to avoid settling on the conclusion that their friends have behaved badly, friends apply comparatively more stringent criteria to evidence that has such an undesired implication, and less stringent criteria to competing evidence that places their friends in a more favorable light. Moreover, their move to the “interpretive level” can in fact be seen as a way to “keep the epistemic sequence going”; as a way to leave the question of whether their friends have behaved badly open for as long as they possibly can (or for as long as the evidence is not overwhelming), so as to fulfill their honest advocacy duties.

As previously discussed, agents who possess the need to avoid a specific closure are responsive to the practical costs and benefits of reaching that particular closure. Importantly, unlike agents who possess a low need for non-specific closure, they are not concerned with avoiding freezing prematurely on inaccurate results nor with reaching inadequate conclusions. They are instead responsive, in their belief-formation, to the practical costs of ending up with a particular attitude, as opposed to the practical costs of being mistaken. As we also saw, when practical costs are associated with the very having of an attitude (as opposed to being associated solely with one's being wrong), the worry that the agent exemplifies a need for specific closure and thus a form of epistemic irrationality comes to the fore. This worry also generalizes to cases plausibly involving moral costs, such as Phone Calls: when the very adoption of an attitude is morally costly, agents who strive to avoid it do not appear to be in a high stakes case, but instead engaged in irrational wishful thinking.

4. What friendship requires in the doxastic realm

The “friend as honest advocate” framework just sketched allows us to see why the belief that one's friend has behaved badly is (to borrow Moss's (2018) terms) morally costly and not morally risky. It thus enables us to see why cases like Phone Calls are not ones of moral encroachment (assuming that moral and pragmatic encroachment share the same structure), but instead belong to the morally required, irrational wishful thinking category. This important implication of the friend as honest advocate framework is best brought out by yet another case:

Mysterious Death ³⁷

In 1953, Eric Olson (then nine years old) finds out through Vincent Ruwet, his father's boss at an Army research establishment, that something terrible has happened to his father, Frank Olson. Frank has, in Ruwet's words, “fallen or jumped” from the 10th floor of a hotel in New York. For years, the family tries to live as if nothing had happened (Eric's mother, in particular, refuses to speculate about the incident). Eric, on the other hand, has a lingering sense of dissatisfaction with Ruwet's explanation. He cannot bring himself to believe that his dad committed suicide, as Ruwet implies. He keeps the inquiry open and will do so for several years– until the truth about his dad finally comes out, through his own investigations. His father did commit suicide, but not for the reasons the Army first stated when they claimed that he was “mentally unstable”. Eric's dad killed himself because he faced an intractable moral dilemma, fearing for his family's safety after having revealed classified information to an enemy power.

By focusing on “third-party” evidence, Mysterious Death (like Phone Calls) allows us to set aside issues related to friends’ trustworthiness in testimonial exchanges.³⁸ In fact, Mysterious Death presents a narrative pattern commonly recurring in film noir or neo-noir, as well as in other genres:³⁹ a loyal, exemplary friend or relative keeps trying to clear their significant other's reputation against all odds, suspending judgment on well-founded accusations when nearly no one else would, often with the effect of ultimately making the full truth come out.

Mysterious Death is meant to elicit the intuition that one can, merely by forming the belief that one's significant other has behaved badly, fail to display loyalty toward them, even when one's belief is actually true. Unlike Eric, by refusing to take a second look into the circumstances surrounding Frank's death, Frank's wife seems to have somehow let him down: she should not have come to believe so easily that her husband had “jumped”, even if that belief turned out true. Especially in comparison with her son, Frank's wife seems to have stopped the inquiry short. For as soon as she reached the undesired conclusion that Frank had indeed jumped, it became impossible for her to provide him with the honest defense he still seemed entitled to, even after his death. Eric, by contrast, comes across as a loyal son, who did what he had to when, despite all the weighty evidence provided by the Army, he still left the question of his father's death open, refusing to conclude that he had killed himself.⁴⁰

Reflection on other cases belonging to Mysterious Death's broad category further reinforces the verdict that forming certain beliefs about our significant others can, in and of itself, carry important moral costs. If only false beliefs in the accusations faced by our friends could wrong them, then agents who don't stand up for their guilty friends internally nor externally should feel relieved upon discovering that the accusations in question were in fact true. “Passive” friends and relatives should indeed see themselves as having narrowly avoided doing something wrong; as having rightly decided not to extend a defense to someone who, after all, was not worthy of it. But this is not, I think, how many who don't stick up for their guilty friends see themselves, nor how we should see them. Take for instance the character of Diane, in the critically acclaimed coming-of-age movie Say Anything (1989): even upon discovering her dad's guilt in the embezzlement charges he faced, Diane still does not seem to regret having stood up for him for as long as she did. Even with the benefit of hindsight, she still sees herself as a loyal daughter, who was right to look again and again into the IRS’ embezzlement allegations, in an attempt to establish her father's good character. When Diane imagines that she would have regretted not having stood up for her father (and expresses satisfaction regarding what she did), she does not seem engaged in a mere post hoc rationalization: she instead appears to view her dad as entitled to the defense – both internal and external – that she provided him. Since exemplary friends and children like Diane and Eric can morally wrong their significant others merely in virtue of forming certain beliefs about them, they are not motivated to avoid error. They instead strive to avoid belief altogether – and are thus inexorably led to ignore part of their evidence.

By leaving the question of their significant others’ past behavior open through a need to avoid a specific closure, loyal agents like Diane and Eric also seem to fulfill a “social epistemic function” that we often associate with love and friendship, and which underlies our positive assessment of their epistemic behavior. This function is highlighted at the very end of Stroud's paper (2006: 523):

Like a defense lawyer, the friend who consistently advocates the more charitable hypothesis serves an important social epistemic function: without her input, negative views (which propagate rapidly through gossip) might become entrenched with little resistance, leading to a decrease in the overall accuracy of the social set of beliefs about her friend.

The present suggestion is that exemplary friends’ local irrationality (which is traceable to their need to avoid a specific closure) can in fact increase the proportion of socially shared true beliefs about their friends. Even when the accusations against our friends are actually true, by standing up for them and defending them (often at the cost of ignoring our evidence), we can contribute to the emergence of a more nuanced picture of their actual character. By being moved by an epistemic need to avoid a specific closure, Emma not only stands up for her friend but also helps to locate her behavior within its full context, thus deepening the collective understanding of Laura's motivations.

Interestingly, the possibility that individual epistemic vices (such as the exemplary friend's “epistemic slant”) sometimes give rise to collective epistemic virtue has recently been examined under the label of “Mandevillian intelligence”.⁴¹ Smart (2018), in particular, has argued that the individual vice of “intellectual stubbornness” or “dogmatism” can, owing to the social structure into which it is sometimes embedded, give rise to a transformative social epistemic phenomenon whereby a collective displays greater accuracy and epistemic virtue. Smart (2018) draws on a significant body of empirical work depicting collective intelligence as a form of collective search through a complex space of doxastic possibilities, where “optimal solutions” are constituted by the set of doxastic states that approximate the nature of reality. Drawing on work from Zollman (2010) (among others), he argues that epistemic stubbornness can promote cognitive diversity within an epistemic community, so as to prevent it from prematurely converging on “sub-optimal parts” of the search space. The friend as honest advocate view of exemplary friends’ doxastic responsibilities insists that partiality bias, just like epistemic stubbornness, can act as a safeguard against pernicious, inaccurate forms of premature consensus. I suggest that such a social epistemic contribution on exemplary friends’ part is partly why we see them as being under a defeasible moral obligation to believe and act like honest advocates.

Importantly, the requirement to stand up for one's friends internally and suspend judgment as to whether they have behaved badly (even when that accusation is true) is a prima facie moral requirement. Not only is it defeated when the incriminating evidence is overwhelming (as discussed previously): other moral requirements can also counterweight it. Some things might indeed matter more, from the moral perspective, than the emergence of a deep understanding of our friends’ underlying motives and true moral character. Without fully developing the friend as honest advocate picture presented here, Baker makes some interesting remarks concerning the limits of our moral obligation to display partiality bias (1987: 6):

If I trust my friend, I do not lock up my silver, but there may be a situation in which I exercise caution with respect to others’ belongings. It is not because I am willing to risk my possessions but not yours, for I do not perceive us at risk. But I cannot take responsibility for the safety of your goods on the basis of facts you would dispute.

If I trust my friend, I don't believe that she would steal my belongings, nor yours. However, as Baker points out, it is not always morally appropriate for me to act on my own doxastic attitude, thus failing to exercise caution with respect to your things. For if you don't share my attitude of trust toward my friend, I am effectively putting your belongings at what you perceive to be a risk, which is not morally acceptable.

Baker proposes that conflicting moral obligations can make it morally unacceptable to act on our biased attitudes. But considering the dependence of the requirement to defend internally on the requirement to defend externally (namely, the idea that we ought to stand up for our friends in front of others and avoid being dishonest), we can push Baker's suggestion even further. We can hold that when we are not morally required to act on a biased attitude and defend our friends externally (because of conflicting moral obligations), then we cannot be morally required to entertain biased attitudes toward our friends either. To be sure, when Eric's standing up for his dad externally starts isolating him and causing him great psychological distress, it is no longer the case that he ought to defend him in action. And once the requirement of external defense is lifted, it is also no longer the case that Eric ought to provide Frank with an internal defense, by refusing to close the question of his death. Likewise, if by defending Laura's dating behavior, Emma might very well end up placing someone she cares about in Laura's last date's situation, then she no longer is under an obligation to stand up for Laura externally. And since she is relieved from her obligation of external defense, she isn't under any pressure to believe like an honest advocate. In short, when the requirement to provide an external defense is lifted – namely, when defending our friends in action is overall not morally responsible (for instance, because doing so would put others at risk, real or perceived), we also should not defend our significant others in belief.

However, as I hope to have shown, as long as we ought to stand up for our friends externally and thus internally, we will often have to “pay the price” of epistemic irrationality as the “admission cost” (Stroud 2006: 518) for friendship. This is how, despite what advocates of the encroachment strategy have argued, friendship can constitutively require epistemic irrationality.

Let me close by highlighting one outstanding issue. I have argued that we can wrong our friends merely in virtue of reaching certain true conclusions about them. But can we also sometimes wrong them by suspending judgment? In particular, is it ever possible to morally harm our friends in being too slow to conclude that they have behaved well? For instance, imagine being told that your friend has donated a significant amount to a charity aiming to support families with deaf children. You know that your friend's boss has a deaf son, and that your friend would very much like to win her boss’ favor. But you are not agnostic with respect to your friend's motivations: without hesitating and looking for further evidence, you readily believe that your friend has donated money simply to support the cause. Would such doxastic behavior be licensed – or even required – on the friend as honest advocate picture?

The literature on epistemic partiality has so far almost exclusively focused on cases of exemplary friends not forming negative beliefs about their friends’ past actions, such as Phone Calls.⁴² But as our last example illustrates, being quick to point out one's friends’ moral qualities also seems essential to a full defense of their moral character; of a piece with downplaying their moral weaknesses. If duties of honest advocacy extend to the formation of “positive beliefs” about our friends, then more trouble looms for the encroachment explanation: if someone is especially quick to believe in their friends’ past good behavior, it certainly cannot be because they are responsive to the costs of being wrong. (After all, such costs tend to drive one's evidential standards up, not down.) Once again, the account focused on exemplary friends’ duty to experience certain epistemic needs seems superior. Both the need to avoid concluding that our friends have misbehaved (when presented with incriminating proofs) and that to conclude that they have behaved well (when presented with flattering evidence), however, are bound to have corrupting effects on our epistemic faculties. Exemplary friends’ tendency to quickly conclude that their friends have performed good deeds thus seems to reinforce the conclusion that true loyalty can force us into epistemic irrationality.⁴³