4.1. Case 1

On the first possibility S’s higher-order evidence supports being undecided about what her first-order evidence, E, supports. In light of this higher-order evidence, S rationally suspends judgment about whether E supports p. Is it possible for S in this situation to rationally believe p? I’ll argue that it’s not.

To better see the point consider the following case, adapted from a case presented by ^{Horowitz (2014)}:

SLEEPY DETECTIVE*: Sam is a police detective, working to identify a jewel thief. He knows he has good evidence -out of the many suspects, it will strongly support one of them. Late one night, after hours spent assessing all the evidence, he finally comes to the conclusion that the thief was Lucy. Sam is quite confident that his evidence points to Lucy’s guilt, and he is quite confident that Lucy committed the crime. In fact, he has accommodated his evidence correctly, and his beliefs are rational. He calls his partner, Alex. “I’ve gone through all the evidence”, Sam says, “and it all points to one person! I’ve found the thief!” But Alex is unimpressed. She replies: “I can tell you’ve been up all night working on this. But five times out of the last ten, your late-night reasoning has been quite sloppy. You’re always very confident that you’ve found the culprit, but you’re only half of the time right about what the evidence supports. So there is only a 50% chance your evidence supports Lucy in this case.” Though Sam hadn’t attended to his track record before, he rationally trusts Alex and believes that she is correct -i.e. that he is right only 50% of the times about what the evidence supports on occasions similar to this one.

Alex’s testimony provides Sam evidence that indicates that it is equally likely that his evidence supports Lucy’s guilt and that it supports her innocence (let’s assume that since there is a presumption of innocence, if the evidence does not support Lucy’s guilt it supports her innocence). In virtue of this, Sam rationally suspends judgment about whether his evidence supports Lucy’s guilt. Now, suppose that despite this Sam continues believing that Lucy is guilty. Is it doxastically rational for Sam to maintain his belief?

I’ll argue that the answer is “no”. The problem is that in this situation Sam’s belief that Lucy is the thief is improperly based, on account of being maintained arbitrarily. This is so despite the fact that his evidence supports Lucy’s guilt and that he based his belief on that evidence. Schematically, the argument goes like this: as I’ll argue, Sam’s case involves some sort of arbitrariness. But believing arbitrarily is a certain kind of bad epistemic performance. Moreover, it is a kind of bad epistemic performance that might be present even though the subject has based her belief on good reasons. Now, by stipulation, Sam has based his belief on good reasons, so Sam’s case is a case in which he bases his belief on good reasons but still performs poorly in forming beliefs. But this is just how we characterized improper basing in the foregoing paragraphs. Hence, Sam’s case involves a kind of improper basing. Finally, since improper basing prevents beliefs from being doxastically rational, Sam’s belief is not doxastically rational.

I motivated the above mentioned characterization of improper basing in section 2, and I do not think it is controversial that forming beliefs arbitrarily falls into the category of “bad epistemic performance”, so ultimately the argument hinges on the premise that Sam is being in some sense arbitrary. Now, there is one sense of “arbitrariness” according to which believing arbitrarily consists just in believing without reasons. Admittedly, in this sense Sam’s belief is not arbitrary; by stipulation he has good reasons to believe that Lucy is the thief. This sense of arbitrariness is what is at stake when assessing a belief in terms of propositional rationality. But as we said earlier doxastic rationality also involves an assessment of the way in which the subject performs in forming beliefs; assessing a belief for doxastic rationality thus requires considering a procedural dimension, and when assessing a belief from this point of view, it becomes relevant how things look from the subject’s perspective.

With this in mind, we can say that what’s arbitrary in Sam’s epistemic performance is that he treats his evidence as supporting Lucy’s guilt even though he consciously evaluated that he cannot determine whether such evidence is in fact indicative of Lucy’s guilt, and moreover, even though he is committed to the idea that it is equally likely that his evidence supports Lucy’s innocence. In other words, from Sam’s perspective believing that Lucy is the thief and believing that she is not the thief are epistemically on a par. The arbitrariness in Sam’s epistemic performance stems from the fact that although he considers both alternatives to be epistemically on a par, he proceeds as if they were not, i.e. as if one of them were better supported by the evidence than the other. From the procedural point of view that is relevant to doxastic rationality this is reproachable: in general, it is a bad epistemic policy to treat what seem to be equally good alternatives as if they were not.

Now, remember the lesson from Turri’s work discussion: the way in which a subject performs can prevent her from being doxastically rational, even if she has based her belief on good reasons. Applying that lesson to this case: Sam proceeds arbitrarily by not treating in the same way what seem to him to be equally good alternatives. Hence, although he bases his belief that Lucy is the thief on good evidence, that belief is not doxastically rational, on account of being improperly based.

A final comment: I grant that Sleepy Detective* is not exactly analogous to the Mr. F.A. Lacy case. In the latter, improper basing stems from the fact the subject makes use of an invalid inference rule when deducing the conclusion from the premises; in the former, the inappropriateness comes from the fact that the subject treats what looks to him as equally well supported beliefs as if they were not equally well supported. I don’t think this is a problem though, since improper basing as a general phenomenon need not involve the use of an invalid inference rule: we can legitimately say that cases of moderate epistemic akrasia are cases of improper basing as long as they involve a kind of questionable epistemic performance on the part of the subject, and I think that being arbitrary in the sense previously discussed fits this description.

4.2. Case 2

On the second view earlier considered (Feldman’s view), a subject, S, rationally suspends judgment about whether her evidence supports p only if S possesses higher-order evidence that supports that her first-order evidence supports neither p nor ~p. To illustrate the point consider the following case (adapted from ^{Coates 2012}):

HOLMES AND WATSON: Watson is a promising apprentice of Holmes, a master sleuth. As part of his training, Watson will often accompany Holmes to crime scenes and other locations, size up the evidence as best he can, and tell Holmes what conclusion he has drawn and how he has drawn it. Now suppose that Holmes brings Watson to a crime scene, assume that the evidence indicates that the butler is guilty, and that Watson uses good reasoning to arrive at that conclusion. In short, Watson rationally believes that the butler did it. Later, Holmes, after having assessed the evidence himself, tells Watson that in his opinion the evidence supports neither the butler’s guilt nor his innocence. Authoritative though he is, Holmes is not infallible, and this is one of the rare occasions in which he is wrong. However, given that Holmes is a master sleuth Watson is rational in accepting his conclusion: Holmes’s testimony on these matters is very authoritative.

Holmes’s testimony provides Watson higher-order evidence that the first-order evidence supports neither the butler’s guilt nor his innocence. Suppose then that in light of this evidence Watson rationally suspends judgment about the matter, in line with Feldman’s view. Similar considerations concerning arbitrariness apply to this case. If Watson suspends judgment after hearing Holmes’s testimony then he comes to think of his evidence as supporting neither the butler’s guilt nor his innocence. That is, Watson considers both alternatives to be on a par. But then, on what grounds could he prefer one over the other? In other words: from Watson’s perspective his evidence is as indicative of the butler’s guilt as it is of his innocence, so believing that he is guilty (i.e. treating the evidence as if it were indicative of the butler’s guilt) is arbitrary. Now, from the point of view of propositional rationality this is irrelevant: all that matters is that Watson has good reasons to maintain his belief. From the point of view of doxastic rationality, however, the way in which the agent bases her beliefs matters: in this case, the agent judges that there is no consideration licensing the conclusion that the butler did it (and to eliminate the other alternative), and still he maintains that belief. Since by believing arbitrarily Watson performs poorly, his belief is not properly based. In consequence, even if he bases his belief on good reasons he does not rationally believe that the butler is guilty.

Here’s another way of putting the point. If Watson, after hearing Holmes’s testimony, is committedly undecided about whether his evidence supports the butler’s guilt or his innocence, and he continues nonetheless to believe the former, it seems that he is just guessing, and hoping to be lucky (even if, unknown to him, the evidence does happen to support the butler’s guilt). To be clear, I am not saying that this is a case of epistemic luck; after all, by stipulation the evidence in fact supports the butler’s guilt, so following the evidence will not lead Watson to the truth just by accident. However, it looks like a case of luck from Watson’s perspective, in the sense that he maintains his belief despite consciously judging that in the face of his evidence the odds of holding a true belief by believing that the butler is guilty are no better than the odds of holding a true belief by believing that the butler is innocent. In other words, he retains his belief despite judging that the odds of getting to the truth in the face of his evidence are no better than chance. In general, this is a bad epistemic policy.

Now, on closer inspection intentional reliance on luck seems to be just another way of forming beliefs arbitrarily: the subject seems not to care about basing her beliefs on good enough reasons. This is bad enough to preclude Watson from properly basing his belief. In consequence, his belief that the butler is guilty is not doxastically rational. Thus, even if we admit that the overall evidence Watson possesses makes it rational for him to believe that the butler is guilty and he makes use of that evidence, we must conclude that Watson does not rationally believe p, on account of believing arbitrarily.¹⁹

As a final remark, let me address the following concern. As was mentioned in section 2 (see also fn. 8) I adopted an attitudinal account of suspension of judgment. I characterized this attitude as involving some sort of commitment to being neutral or undecided regarding some matter. One may worry that while this characterization of suspension of judgment is not mandatory (viz. there is room for an attitudinal view of suspension of judgment that does not ascribe any commitment to the subject) it is necessary for my argument to work: if higher-order suspension of judgment does not involve any commitment to being undecided, one could think, first-order belief will not look arbitrary to the subject. Moreover, not only is the notion of commitment avoidable in this context; it is also suspicious: after all the whole point of suspending judgment seems to be to avoid commitment.

First, one cannot just assume that the whole point of suspension of judgment is to avoid commitment. In fact, several philosophers working on the topic characterize suspension of judgment as involving a kind of commitment: Friedman argues for an attitudinal account of suspension of judgment, which she characterizes as an attitude of “committed indecision” (2013, fn. 5), while Sturgeon (2010) describes it as a kind of “committed neutrality”. Also, in ^{Crawford’s view (2004)} suspension of judgment is understood as a kind of higher-order belief, so it could also be plausibly described in terms of commitments. That being said, I concede that it might be possible to formulate an attitudinal account of suspension of judgment which does not characterize it in terms of commitments. But, to be sure, the onus of proof is on my opponents to show how such an account would go and that such a view is in fact tenable; and even if it is, it is not entirely clear how it would affect my arguments.

5.1. Internalism/Externalism

One possible worry regarding the view proposed earlier is that it presupposes an internalist view about rationality or justification.²⁰ If this were so, the scope of the argument would be very limited: it would leave those who advocate an externalist (or otherwise non-internalist) theory unmoved. However, I don’t think the view I defended presupposes internalism. In this section I will explain why.

Internalist theories come in different flavors, but there is a core idea that underlies most of them, viz. in order for a belief to be justified or rational the subject must be aware (or potentially aware) of all (or at least some of) the reasons or justifiers for that belief. Bergmann states this condition as follows:²¹

The Awareness Requirement: S’s belief B is justified only if (i) there is something, X, that contributes to the justification of B -e.g. evidence for B or a truth-indicator for B or the satisfaction of some necessary condition of B’s justification- and (ii) S is aware (or potentially aware) of X. (^{Bergmann 2006, p. 9})

As Bergmann notes, one of the most compelling motivations for this kind of internalist requirement comes from BonJour’s famous case of Norman, the Clairvoyant (^{BonJour 1985, p. 41}). Very roughly, according to BonJour whenever Norman is not aware of whatever makes his belief justified his belief looks accidental or arbitrary from his perspective, thus he is not justified. Of course, it might be that from an external point of view the belief would not look accidental or arbitrary, but according to BonJour it is his own perspective that matters. Bergmann summarizes this motivation as follows:

The Subject’s Perspective Objection: If the subject holding a belief isn’t aware of what that belief has going for it, then she isn’t aware of how its status is any different from a stray hunch or an arbitrary conviction. From that we may conclude that from her perspective it is an accident that her belief is true. And that implies that it isn’t a justified belief. (^{Bergmann 2006, p. 12})

Admittedly, the arguments I developed in the previous sections exhibit some resemblance with BonJour’s remarks. However, I’ll argue that the view I defended is consistent with the rejection of both the awareness requirement and the subject’s perspective objection. Hence, it does not presuppose internalism.

First, I’ll argue that my view is consistent with the rejection of the awareness requirement. Let’s call rationality-contributors whatever facts are relevant for the rationality of a belief. These facts might include the actual basis for a belief as well as facts about the obtaining of a support relation between that basis and the belief. Internalism about rationality is the idea that it is a necessary condition for a belief to be rational that the subject is aware (or potentially aware) of (at least some of) her rationality-contributors for that belief. Externalism, by contrast, contends that at least sometimes a belief can be rational even if the cognizer is not (or is incapable of being) aware of (at least some of) the rationality-contributors for her belief. It is worth noting that externalism is not the claim that agents are never aware of the rationality-contributors for their beliefs: it is compatible with externalism that agents are sometimes aware of the rationalitycontributors for their beliefs.

Now, in the previous sections I argued that if you rationally suspend judgment about the epistemic status of your first-order belief then such belief is not doxastically rational. It is plausible to assume that if you rationally suspend judgment about the epistemic status of your first-order belief then you must be aware of (at least some of) the rationality-contributors for that belief. However, what follows from this is that your actual awareness of the rationality-contributors for a belief B might affect B’s doxastic rationality. What doesn’t follow is that your lack of awareness of the rationality-contributors for B turns B irrational (either doxastically or propositionally). It is one thing to say that when you are in fact aware of the rationality-contributors for B this might affect B’s rationality (according to my view, only insofar as you rationally suspend judgment regarding the obtaining of a support relation between the basis for B and B), and another thing to say that it is a necessary condition for B to be rational that you are aware of the rationality-contributors for B. I maintained the former, but this is consistent with rejecting the latter.

In order to illustrate the point consider the following example. Very roughly, reliabilism is the view that a belief is rational only if it is the output of a reliable belief-forming process, i.e. a process that is truth-conducive. Whether a belief is the output of a truth-conducive belief-forming process is not something to which agents typically have reflective access. However, reliabilism is compatible with adding some further requirement according to which in order for a belief to be rational one must not have a reliable belief that such a belief is not reliable.²² In the same way, one could add a negative requirement according to which in order for a belief to be rational one must not reliably suspend judgment about whether such a belief is reliable.²³ Since reliably believing or suspending judgment about whether one’s belief was reliably formed plausibly involves some kind of awareness of the rationality-contributors for that belief, these requisites entail that in some circumstances awareness of one’s rationality-contributors might have an impact on rationality. However, reliabilism is still an externalist view. The negative requirement that one is not aware of one’s belief being unreliable, or that one doesn’t suspend judgment about the matter, does not entail the positive requirement that one is aware of one’s belief being reliable.

Something analogous might be said about the subject’s perspective objection. In arguing for my view, I claimed that being aware of the rationality-contributors for one’s belief and rationally suspending judgment about whether they make the belief rational will make such belief look accidental or arbitrary from one’s perspective, thus making it doxastically irrational. Admittedly, the appeal to the perspective of the subject gives to the argument an internalist appearance. However, from what I have claimed in the previous section it doesn’t follow that when you are not aware of the rationality-contributors for your belief and you have not reflected on the epistemic status of your first-order belief such belief will look accidental or arbitrary from your perspective, thus being irrational.

To better see the point, consider one frequently cited motivation for externalism. We are willing to grant justification to some beings, like small children, that do not have the cognitive resources to be reflectively aware of the rationality-contributors for their beliefs or to reflect on the epistemic status thereof. However, arguably a small child’s belief that there is an object before him will not look accidental or arbitrary to him. So it doesn’t seem to be true in general that the lack of awareness of one’s reasons makes one’s belief look accidental or arbitrary from one’s own perspective. However, cases of moderate epistemic akrasia are very much unlike the one just mentioned. In cases of moderate epistemic akrasia the subject is in fact reflectively aware of (at least some of) the rationality-contributors for her belief and she has adopted a particular attitude concerning their epistemic status, i.e. suspension of judgment. In these cases in which there is in fact reflective awareness it is much less obvious, and I think counterintuitive, to say that the subject’s belief will not look accidental or arbitrary to her. If a subject is reflectively aware that she is not able to determine whether her evidence supports her belief then it seems that forming or maintaining such belief will look arbitrary to her, in the sense that she is aware that she is treating her reasons as being indicative of the truth of her belief even though she believes that she is not able to determine whether they are. But the most important point is this: it is consistent to maintain that being aware of one’s reasons and suspending judgment about the epistemic status of one’s belief could make one’s belief look accidental or arbitrary from one’s perspective while rejecting that the same is so in cases in which one is not aware of his own reasons or has not reflected upon the epistemic status of his own belief. So my view is consistent with rejecting the subject’s perspective objection: I do not need to maintain that if a subject is not aware of the rationality-contributors for her belief then the belief will look accidental or arbitrary to her; I need only maintain that in cases in which the subject is reflectively aware that she is not in a position to determine whether her evidence supports her belief, her belief will look accidental or arbitrary to her.

So my view presupposes neither the awareness requirement nor the subject’s perspective objection. Since these theses characterize internalism, we can conclude that my view does not presuppose internalism.

There is a related worry worth considering. Even granting that my view does not presuppose internalism, one might argue that it does presupposes mentalism (^{Conee and Feldman 2004}) -assuming that mentalism is not a form of internalism (^{Bergmann 2006, pp. 49-57}). Let’s follow Bergmann in characterizing mentalism as follows:

Mentalism: A belief’s justification is a function solely of (i) which mental states the subject is in and (ii) which mental states of the subject the belief is based on. (I.e. if two possible subjects are exactly alike mentally and in terms of which of their mental states their beliefs are based on, then they are exactly alike justificationally.) (2006, p. 45)

Again, if this were correct my arguments would be very limited in their scope: anyone who believes that mentalism is false would remain unconvinced. Since mentalism is a controversial view, this is a legitimate worry. I will argue, however, that my view does not presuppose mentalism. First, it is important to note that rejecting mentalism is consistent with maintaining that sometimes a belief’s rationality depends on some of the mental states the subject is in. The idea that not only mental states can be rationality-contributors does not entail that no mental state can be a rationality-contributor. Now, in the previous sections I argued that in the cases in which you suspend judgment about the epistemic status of your first-order belief, that belief is not doxastically rational. This presupposes that some mental states, e.g. some of the attitudes you hold towards the rationality-contributors for your belief, are relevant to the rationality of that belief. But this in no way entails that only mental states are relevant to determine the rationality of a belief. It is consistent to maintain that more than just mental states can play the role of rationality-contributors for a belief while contending that sometimes some mental states can affect one’s belief rationality. If what I said is on the right track, my view does not presuppose mentalism neither.

To end this section, allow me a final remark in order to dissipate the worries that my argument is not sufficiently general. Let’s call X whatever property confers epistemic rationality on a belief. A moderately akratic agent is someone who believes p and suspends judgment about whether her belief that p has X. Of course, a rational moderately akratic agent is someone whose attitude of suspension of judgment is rational, i.e. someone whose attitude of suspension of judgment has property X. Now, put in these general terms, my argument in section 4 was that someone who rationally suspends judgment concerning whether her belief that p possesses property X and still maintains her belief that p is being doxastically irrational, on account of being arbitrary. The arbitrariness stems from the fact that the subject believes that p despite being committedly undecided concerning whether p has X. In other words, the agent is aware that rationality dictates being undecided concerning whether her belief possesses rationality-conferring properties, but still she goes on believing as if her belief possessed such properties. Admittedly, in the cases at hand her belief has, by stipulation, rationality-conferring properties, but the point is that she cannot rationally take this for granted while being rationally undecided regarding whether that is actually the case.

It should be clear then that the argument can be formulated in general, my point being just that whatever you think of the merits of this argument, its success does not depend on any particular view on epistemic rationality.

5.2. Perspective

There is a third worry that needs to be addressed. Even granting that my view does not presuppose the subject’s perspective objection, one might wonder why, in the cases at hand, one should grant such importance to the subject’s perspective. Why should the believer’s perspective have this kind of authority?

One way of putting the worry is the following. One might argue that all that matters to doxastic rationality is being sensitive to the evidence that makes one’s belief propositionally rational. In discussing cases of radical epistemic akrasia, Coates presents the case of Huck and Jim:²⁴

Huck accepts that slaves are property and believes he should turn Jim in to the authorities. But when the opportunity arises, he finds he cannot bring himself to do so and continues to help Jim escape. Like Bob, he judges his action to be morally wrong, even though it is in fact morally right. (^{Coates 2012, p. 116})

It might be tempting to argue that Huck’s actions are subjectively blameworthy, since he believes that acting as he does is morally wrong. However, Coates argues that Huck’s action is morally praiseworthy because the properties that motivate him to act as he does are the same that make his action morally right. In other words, his action responds to the right sort of moral properties even if his second-order moral belief does not. That is all we need in order to say, in Coates’ words, that “he genuinely cares about morality”. Analogously, one could say that in cases of moderate epistemic akrasia all that matters to the subject being doxastically rational is that the properties that motivate her in forming or sustaining her belief are the same that make the belief propositionally rational; if her belief responds to the right sort of epistemic properties then her belief is doxastically rational. If this is on the right track the perspective of the subject seems irrelevant: as long as she is sensitive to the properties that make her belief propositionally rational we could say that she really “cares” about forming rational beliefs.

The problem with this view, I think, is that it cannot make sense of cases of improper basing. To be sure, the reasons that motivate Mr. F.A. Lacy to form the belief that the Spurs will win are the same that make that belief propositionally rational for him. Still, his beliefs do not seem to be doxastically rational at all. As Turri notes, there is something wrong with the way in which those subjects perform in making use of their reasons. The whole point here is that being doxastically rational is not just about being motivated by the reasons that make a belief propositionally rational, it is also necessary that the subject’s use of those reasons is competent.

Once it is acknowledged that proper basing is important, it becomes clearer that there is a place for arguing that the subject’s perspective might also be important in the cases at hand. One might argue that a change in one’s perspective might affect whether one’s epistemic performance is appropriate, thereby affecting doxastic rationality. To better see the point, consider the following. Arguably, forming true beliefs and avoiding forming false ones are two important epistemic goals. Following the evidence is the right thing to do because by doing it we (are supposed to) attain those goals. Now, in cases of moderate epistemic akrasia the subject is committedly undecided about whether her evidence is indicative of the truth of her belief. It’s not just that she is not aware of her evidence being indicative of the truth of her belief: she is rationally committed to the idea that she is not able to determine whether her evidence is in fact indicative of the truth of her belief -or in other words: she is rationally committed with the idea that the evidence equally supports the belief and its rejection. Still, she treats her evidence as if it were indicative of the truth of her belief. Treating evidence that one rationally believes is not indicative of the truth of a belief as being indicative of the truth of a belief shows some kind of epistemic incompetence, in the following sense: behaving like that can only mean either that i) the subject does not care about forming true beliefs and avoiding forming false ones, or ii) she does care about forming true beliefs and avoiding false ones, but she is trying to attain those goals by ignoring the evidence, just by guessing and hoping to get lucky or iii) she is letting non-epistemic reasons play a role in belief-formation. In any of these cases, it is clear that the subject “does not care” about epistemic rationality.

5.3. Forming vs. Sustaining a Belief

There is a final worry we need to discuss.²⁵ Consider the following possible case:

SLEEPY DETECTIVE**: Sam is a police detective, working to identify a jewel thief. He knows he has good evidence -out of the many suspects, it will strongly support one of them. One day, very late at night, he is getting ready to go through the evidence. Before assessing the evidence, however, he calls his partner Alex to discuss the case. In that conversation Alex says to him: “Try not to go through the evidence now so late at night. Five times out of the last ten, your late-night reasoning has been quite sloppy. You’re always very confident that you’ve found the culprit, but you were only half of the time right about what the evidence supported. So if you assess the evidence now there would be only a 50% chance your evidence supports your belief.” Though Sam hadn’t attended to his track record before, he rationally trusts Alex and believes that she is right, that is, that he is right only 50% of the times about what the evidence supports on occasions of late-night reasoning. However, he does not follow her advice and assesses the evidence anyway. Now, since Sam trusts Alex, before going through the evidence he rationally suspends judgment about whether his evidence will support his belief. Later, after hours spent assessing all the evidence, he finally comes to the conclusion that Lucy is the thief. Sam is quite confident that his evidence points to Lucy’s guilt, and he is quite confident that Lucy committed the crime. In fact, he has accommodated his evidence correctly, and his belief is rational.

This case is similar to Sleepy Detective* but it has one important difference, viz. Sam assesses his first-order evidence and forms his belief after he rationally suspends judgment about whether his evidence will support his future belief. The worry is that there may be important differences between Sleepy Detective** and Sleepy Detective*. One might feel that even if it is not doxastically rational to form a belief when one has rationally suspended judgment about whether one’s belief will be supported by the evidence, it is doxastically rational to maintain a belief one already has in the face of rational suspension of judgment about whether one’s evidence supports the belief.

The question then arises: What differentiates the case of forming a belief from the case of maintaining a belief? One possible answer appeals to epistemic conservatism -see ^{Harman 1986}, ^{Foley 1983}, ^{Lycan 1988}, ^{Sklar 1975}-. Roughly, epistemic conservatism’s main thesis contends that the mere fact of holding a belief confers some degree of justification on that belief. As ^{Vahid (2004}) has noted, however, epistemic conservatism is not a unified view but a collection of related theses that have been defended by different philosophers. He distinguishes three different forms of epistemic conservatism:

From these options, differential conservatism (DC hereinafter) might provide a reason for sustaining that Sam can rationally maintain his belief that Lucy is guilty in a case like Sleepy Detective* but he cannot rationally form the belief that Lucy is guilty in Sleepy Detective**.²⁶ DC contends that an agent need not reject a hypothesis, once the hypothesis is believed, just because she comes to know of an evidentially equivalent alternative to it. In other words, according to DC “if an agent believes p on the basis of evidence E, he is justified in believing that p despite coming to recognize that an incompatible hypothesis, q, is equally supported by E” (^{Vahid 2004, p. 103}). Now, the arguments provided in the previous paragraphs maintain that holding a belief in the face of moderate epistemic akrasia would be arbitrary. Roughly, this is because in those cases, from the subject’s perspective both believing p and believing ~p look to be equally supported by the evidence. But here a defender of epistemic conservatism might object that the arguments only work in cases of belief formation. If the subject already holds the belief when she rationally suspends judgment in the second level she should not abandon it, since the situation looks from the subject’s perspective like one in which DC applies: one should not abandon a belief just because one acknowledges that there is an incompatible belief that enjoys an equally strong support from the evidence.

This objection depends on whether DC is correct. But there are reasons to reject DC. ²⁷ First, one might press the charge of arbitrariness against DC itself. If the subject thinks that both p and ~p are equally likely in light of her evidence, why would the mere fact that she happened to form the belief that p first make a difference? The belief that ~p could have had the same epistemic status had the agent formed that belief first. But plausibly the facts that determine that a subject forms one belief before another are not epistemically relevant. So it just seems arbitrary to confer a privileged epistemic status to a belief just because it was produced at an earlier time.

However, I find Vahid’s argument the most compelling. The problem is that DC contravenes the thesis of epistemic supervenience according to which the justification of a belief supervenes on certain non-epistemic properties of that belief (e.g., being part of a coherent belief system, being reliably produced, being adequately grounded, etc.). A consequence of that thesis is that if two beliefs share the same subvening property and one is justified, so is the other. Now, consider the following:

Suppose now two scientists S ₁ and S ₂ who, faced with the task of explaining the same data, come up with incompatible but evidentially equivalent hypotheses H ₁ and H ₂ respectively. Suppose further that the hypotheses in question become known to both some time later during a meeting where S ₂, for some reason of his own, gives up H ₂ and follows S ₁ in believing H ₁. Now, according to DC, S ₂ should have stuck with H ₂. His belief that H ₁ is, thus, unjustified while S ₁ is justified in believing H ₁ and rejecting H ₂. So, assuming that DC is normatively correct, we are faced with a situation in which while believing H ₁ is rational (justified) for S₁, another token of the same belief fails to be rational (justified) for S ₂ despite the two tokens sharing the same subvening justification-conferring property, and this contravenes the thesis of epistemic supervenience. (^{Vahid 2004, p. 105})

As Vahid also notes, we could understand the objection from arbitrariness not in terms of the time at which the belief was formed but in terms of the supervenience thesis: a subject that behaves in accordance with DC is prepared to ascribe the property of justification or rationality to one belief while refraining from treating another belief as justified or rational despite the fact that both beliefs seem to him to share the same subvening justification-conferring property, and this “introduces an element of arbitrariness into our canons of rationality” (^{Vahid 2004, p. 106}).

So appealing to epistemic conservatism in order to defend that there are cases of rational moderate epistemic akrasia, viz. those involving the sustaining of a belief, faces its own problems. There might be other strategies available in order to defend this view, but I think in this case the onus of proof is on my potential opponents.