The Level-Splitting View and the Non-Akrasia Constraint

Some philosophers have defended the idea that in cases of all-things-considered misleading higher-order evidence it is rational to take divergent doxastic attitudes to p and E supports p. In a recent paper, Sophie Horowitz has argued that such “Level-Splitting views” are implausible since they violate a rational requirement she calls the Non-Akrasia Constraint. In this paper, I argue that Horowitz’s objection is misguided since it conflates two distinct notions of epistemic rationality.


Horowitz's Argument
According to Horowitz's Non-Akrasia Constraint: Bit can never be rational to have high confidence in something like, Bp but my evidence doesn't support p.^ (2013, p. 718). Since Level-Splitting views violate the Non-Akrasia Constraint she thinks that we can rule them out completely. Here is a more formal statement of the argument: (1) Epistemic Akrasia is always irrational (not rational).
To be akratic in the practical sense is to fail to intend to φ, if one believes that one ought to φ. By, contrast, to be epistemically akratic is to hold a certain belief even though one does not believe that this belief is supported by the evidence. 2 Most people take epistemic akrasia to be paradigmatically irrational but in order to strengthen this intuition, Horowitz provides us with the following vignette: Sleepy Detective: Sam is a police detective, working to identify a jewel thief. He knows he has good evidenceout of the many subjects, it will strongly support one of them. Late one night, after hours of cracking codes, and scrutinizing photographs and letters, 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 belief are justified. He calls his partner, Alex. BI've gone through all the evidence^, Sam says, Band it all points to one person! I've found the thief!^But Alex is unimpressed. She replies: BI can tell you've been up all night working on this. Nine times out of the last ten, your late-night reasoning has been quite sloppy. You're always quite confident that you've found the culprit, but you're almost always wrong about what the evidence supports. So your evidence probably doesn't support Lucy in this case.^Though Sam hadn't attended to his track-record before, he rationally trusts Alex and believes that she is rightthat he is usually wrong about what the evidence supports on occasions similar to this one. (2014, p. 719).
Sam's total evidence is made up of the first-order evidence (codes, photographs, letters etc.) and the higher-order evidence provided by Alex's testimony about Sam's track record data. Horowitz seems to grant that Sam's total evidence in this case supports both the proposition that BLucy is the thief^and the proposition that BThe evidence does not support that Lucy is the thief^. Of course, some will argue that it is impossible for one's total evidence to support divergent attitudes to p and E supports p but in order to concentrate on Horowitz's argument, I will make the assumption that it is possible to have all-things-considered misleading higherorder evidence. 3 According to the Level-Splitting view, Sam should believe something of the form, BLucy is the thief, but my evidence doesn't support that^. Horowitz points out that this looks strange and that the conjunction of the two statements has a somewhat Moore-Paradoxical quality. The statements are not contradictory but there is still something odd about believing them simultaneously. But Horowitz argues that the oddness extends well beyond the combination of attitudes themselves. Advocates of the Level-Splitting view also have to accept some very counterintuitive consequences.
First, if Sam comes to believe both that Lucy is the thief and that the evidence does not support that Lucy is the thief^it seems that he can engage in some patently bad reasoning. Suppose that Sam reflects on his belief that Lucy is the thief. He cannot conclude that he believes this in virtue of the evidence, since he believes that the evidence does not support that Lucy is the thief. Instead, he must believe that he got lucky: that he found out the truth without relying on the evidence! Secondly, it seems that Sam could use his belief that Lucy is the thief as evidence that the higher-order evidence is misleading. But this seems to be nothing but a way for Sam to illicitly bootstrap himself into confidence. 4 Horowitz also asks us to consider what it would look like if Sam would act out on his beliefs. It seems that Sam should be disposed to place a high-odds bet on the fact that Lucy is the thief (since he thinks this is true). But at the same time, it seems that Sam should be disposed to place a low-odds bet on the fact that Bthe evidence support that Lucy is the thief^. Horowitz says that: Bit seems patently irrational to treat a bet about p and a bet about whether one's evidence supports p as completely separate^. (Ibid, p. 728) She also asks us to consider what would happen if Sam were to act out on his akratic beliefs. Someone who is convinced that Lucy is the thief should be disposed to recommend that the cops raid her apartment and send her to jail. But if Sam were asked to justify or explain his behavior he would not have much to say since he does not think that his evidence supports any of it.
I think that Horowitz argues convincingly that it is irrational from a certain perspective to hold akratic beliefs. However, in the following I will suggest that Horowitz's argument conflate two distinct notions of epistemic rationality: the view that rationality consists in responding correctly to one's epistemic reasons, and the view that rationality is constitutively tied to coherence. 3 Feldman (2005) seems to think that all-things-considered misleading higher-order evidence is impossible. He argues that on reflection one will realize that one's evidence cannot support p if one's higher-order evidence does not support belief in E supports p. Worsnip (2018) provides some good arguments for the possibility of all-things-considered misleading higher-order evidence. 4 Horowitz (Ibid. pp. 732-34) goes on to argue that the bootstrapping problem also demonstrates that the Level-Splitting view in the long run becomes inconsistent since it allows first-order evidence to affect what to believe at the meta-level. However, I do not think that Level-splitters are committed to the implausible assumption that first-order and higher-order evidence can never interact in this way.
First, some background. In metaethics we find two different ways of talking about rationality. According to one view, what one is rationally required to do is determined by one's normative reasons. According to another view, what one is rationally required to do is determined by considerations having to do with coherence. Niko Kolodny (2005) refers to the former view as objective rationality and the latter view as subjective rationality. Some people (e.g. Broome 2007) have argued that the objective account of rationality is implausible since our normative reasons are not always transparent to us. Others, like Kolodny, argues that coherence requirements fail to be deontically significant. However, we do not have to take a stand on this debate; what matters for our purposes is that these are distinct notions of rationality. While the distinction Kolodny is referring to between objective and subjective rationality is a prevalent conception in metaethics it has only recently received serious attention in epistemology. 5 A very natural thought is that epistemic reasons are provided by one's evidence. However, in order for a piece of evidence to provide you with an epistemic reason we also have to presume that you possess that that evidence. 6 The fact that there is evidence to believe that Lucy is the thief does not give you any epistemic reason to believe so unless you have access to the relevant evidence (in this case codes, photographs, letters etc.) Moreover, in order to possess evidence to believe that Lucy is the thief we also have to assume that you have the capacity to make the proper inference from the evidence. So, objective rationality in epistemology consists in responding correctly to the epistemic reasons that one possess. 7 According to another view about epistemic rationality one has to meet certain standards in order to be rational. On this account, rationality is conceived as a system of rules much like chess, etiquette, or law. According to a typical standard of rationality you are required to intend to ψ if you intend to ϕ, and believe that in order to ϕ you must ψ. Another typical standard of rationality is the enkratic principle that says that you are rationally required to intend to ϕ, if you believe that you ought to ϕ. The epistemic version of the enkratic principle states that you are rationally required to believe p, if you believe that the evidence supports p. What rational requirements have in common is that they regulate what combinations of attitudes are required in order for you to be rational. On the most plausible interpretation of this view, you are not rationally required to have any particular attitudeswhat matters instead is that your attitudes conform to a certain structure. 8 Notice that the reasons-responsiveness and the coherence account of rationality might come apart. We can easily think of cases in which one's evidence supports a certain doxastic attitude to p, but one fails to see this. For example, suppose that the 5 See Worsnip (2018), Lasonen-Aarnio (forthcoming), and Sylvan (n.d.). 6 See Sylvan (2016) for further discussion about what it means to possess evidence. 7 The idea that what one is rationally required to do is determined by the normative reasons one possess has become increasingly popular in metaethics. See Lord (2017) for a recent defense. 8 I take it that it is more plausible to see structural requirements of rationality as wide-scope than narrow-scope principles. The problem with Narrow scope principles is that it implies that one's beliefs determines what one is rationally required to believe. For example, if S has a crazy belief, e.g. that the evidence supports that the moon is made of green cheese, then S is rationally required to believe that the moon is made of green cheese. This is what Kolodny (2005) calls Bthe bootstrapping problem^. evidence supports suspending judgment about p, but one for some reason or the other ends up believing that the evidence supports believing p. Then we have a situation in which one in virtue of the evidence has epistemic reason to suspend judgment about p, but is rationally required to believe p (since this is what one believes that one ought to do). What one has epistemic reason to believe and what one is rationally required to believe in order to be coherent may therefore amount to different things. As I will show in what follows, this opens up for the possibility that epistemic akrasia might be rational given one account of epistemic rationality but irrational given another.

Disambiguating BEpistemic Rationality^in Horowitz's Argument
Let us return to Horowitz's argument to see how the fact that the two notions of epistemic rationality can pull in different directions plays out here: (1) Epistemic Akrasia is always irrational (not rational).
The most plausible way to understand (1) is in terms of a coherence account of rationality. What makes epistemic akrasia irrational is the fact that one holds a combination of doxastic attitudes that fails to be consistent or coherent. Given that we understand Brational^in the same way in (2) the argument goes through.
However, I do not think that (2) should be apprehended in this way. The Level-Splitting view is not a view about structural rationality. On reflection, it is implausible to think that Level-Splitters are defending the view that one is rationally required to believe p if one believes that the evidence does not support p. It is much more plausible to understand Level-Splitting in terms of a reasons-responsiveness account of rationality. I think that this also makes sense if we consider the philosophers Horowitz identifies as Level-Splitters. Timothy Williamson have affirmed elsewhere (2000, p.164) that he takes it to be a platitude that rationality consists in proportioning one's beliefs to the evidence. Allen Coates is also explicit about the fact that he takes epistemic rationality to be: Ba matter of basing one's beliefs on sufficient evidence ( 2012, p. 113).
But in that case, Brational^does not denote the same thing in (1) and (2). When Horowitz is defending the Non-Akrasia Constraint she is using Brational^in the coherence sense. She is appealing to the idea that it is irrational to have doxastic attitudes that does not seem to be consistent. However, when Level-Splitters are describing their view they are using Brational^in the reasons-responsiveness sense. As a result, there is good reason to believe that Horowitz's argument against Level-Splitting views is misguided. The fact that Level-Splitting is irrational by the light of an alternative conception of epistemic rationality cannot be dialectically effective as an argument against Level-Splitting.
Moreover, advocates of Level-Splitting views might even agree that there is something epistemically bad about taking divergent doxastic attitudes to p and E supports p. For instance, Maria Lasonen-Aarnio (2014, p. 342) acknowledges that one in a distinct sense is Bcriticisable from an epistemic point of view^to retain one's belief in the face of apparently undermining higher-order evidence. 9 However, it does not follow that advocates of Level-Splitting views thereby are committed to hold that epistemic akrasia is epistemically irrational.
To complete the picture let us return to Sleepy Detective. Remember that we were to assume that Sam's total evidence supports both believing that Lucy is the thief and that the evidence does not support that Lucy is the thief. Since Sam is in possession of this evidence he has epistemic reasons to believe something of the form, BLucy is the thief but my evidence does not support that Lucy is the thief^. According to the reasonsresponsiveness account of epistemic rationality there is nothing irrational in taking akratic attitudes as long as they are reflected by the epistemic reasons that one possesses.
By contrast, according to the coherence account of epistemic rationality it follows that it is irrational for Sam to take an akratic combination of attitudes. However, as I suggested, Level-Splitters are still free to maintain that it is epistemically rational for Sam to be akratic in the reasons-responsiveness sense but hold that he thereby also is doing something epistemically bad. To believe that Lucy is the thief while also believing that the evidence does not support this represents a certain type of epistemic failure. So, in a way it is not at all surprising that Sam's akratic beliefs leads to some very odd behavior. But this does not have to exclude that Sam at the same time is epistemically rational in the sense that he is believing in accordance with his epistemic reasons.

Concluding Remarks
In this paper, I have argued that Horowitz's argument against the Level-Splitting view is misguided. To demonstrate this, I have made a distinction between two distinct notions of epistemic rationality: the reasons-responsiveness and the coherence view. Given that we understand Level-Splitting as a view about how to rationally respond to one's epistemic reasons and not as a structural requirement of rationality it should become plain that it is not a proper target to Horowitz's objection. At the very least we can conclude that it becomes dialectically ineffective to point out that the Level-Splitting views is implausible given an alternative conception of epistemic rationality.
This might seem like a minor point to make but I believe that it also has a greater bearing on the debate about the significance of higher-order evidence. For example, I suspect that a similar conflation plays a major role in the peer disagreement debate. Conciliatory views hold that one is rationally required to suspend judgment or at least be significantly less confident about one's opinion in cases of peer disagreement. Steadfast views, on the other hand, hold that one sometimes is rationally permitted to maintain one's belief despite peer disagreement. One way to make sense of this controversy is to interpret advocates of conciliatory views as talking about epistemic rationality in terms of coherence, and advocates of steadfast views as talking about epistemic rationality in terms reasons-responsiveness.
On this construal, what advocates of Conciliationism are saying is that one cannot rationally sustain one's belief that p, given that one also believes that an epistemic peer believes differently. Since epistemic peers are supposed to take each other to be equally likely to be right about the disputed matter it does not appear to be rational for either of them to believe that their own particular belief is supported by the evidence. That would amount to believing something of the form, BI believe p but I don't know whether my evidence supports p^. Steadfast views, by contrast, can be construed as appealing to the fact that in some peer disagreements one peer's belief may in fact be supported by the evidence despite appearances to the contrary. 10 So, the more important lesson is that we should be much more precise about what sense of epistemic rationality we are referring to when we are discussing the significance of higher-order evidence.