An Accuracy-Based Argument for Conciliation. When we try to figure out what to believe, having evidence is good and having more evidence is

Transcription

1 1 1 An Accuracy-Based Argument for Conciliation 1. Introduction When we try to figure out what to believe, having evidence is good and having more evidence is even better. We like the following sequence of events: I get some new evidence, and then, on that basis, I update my beliefs. Call this The Process. The Process: I hold some beliefs P1, P2, etc. on the basis of some evidence E. Upon acquiring some additional evidence, E*, I attempt to rationally update my beliefs; I thereby arrive at some new P1*, P2*, etc. I take it to be a basic datum in epistemology that The Process generally improves our beliefs. That s why we go to such great trouble looking for more evidence, and why we are especially concerned to do so before making a decision of great practical importance. So, I say that The Process generally improves our beliefs. But what is it to improve our beliefs? One natural standard is matching; beliefs improve when they match the world better. When we evaluate full or all-out beliefs this can be cashed out in terms of truth and falsity; we say our full beliefs match the world when they re true. But, by contrast, when we focus on partial beliefs, truth can no longer play that role directly on the usual understanding, a degree of confidence, or credence, cannot strictly speaking be true or false. So instead we speak of the analogue, graded notion of accuracy ; we say a partial belief 2 matches the world better when it s more accurate. The platitude that The Process generally improves our beliefs is true on either framework; getting and responding to evidence tends to give us true full beliefs and accurate partial ones. But since 1 This is a version of my Anticipating Failure and Avoiding It (2018a) that has been streamlined for use as a writing sample. Throughout I will indicate where interested readers may consider consulting the longer version, which is available at my website: 2 To the extent that the differences among them are relevant, I discuss the particular ways in which accuracy can be quantitatively measured in section 8.

2 2 this paper is interested particularly in rational degrees of belief, I will focus on the second version. So: The Process is generally good, where by that I mean that going through The Process generally increases the accuracy of one s beliefs. But even if The Process is generally good, it need not be good in every instance. Sometimes it fails, where by that I just mean the converse of what I meant by success namely, that it sometimes issues in beliefs less accurate than those with which it began. Here are two (non-exhaustive) ways for it to fail. First: sometimes I receive misleading evidence, and so even though I rationally update my beliefs in light of that evidence it points me astray. Second: sometimes I receive non-misleading evidence, but yet because I manifest some irrationality I update my beliefs in ways contrary to the truth that evidence in fact suggests. Even though the evidence points in the right direction, I manage to go astray anyway. I am interested in cases where The Process not only fails, but where we can see that failure coming in advance. What should we do in such cases? What should I do when I can anticipate, of some particular piece of evidence I am going to receive, that in responding to it The Process is likely to fail rather than succeed? The terms failure and success here themselves suggest a natural answer: if I really anticipate that The Process is likely to fail, I ought not engage in it. So I ought not take account of this evidence; instead, I ought to ignore it. This thought will be especially compelling if we think not only that The Process does usually yield more accurate beliefs, but also that this fact is actually what gives The Process its value. We might say: getting and responding to evidence is just a means to accurate beliefs. On this understanding, it would involve a confused fetishization of the means to try to take account of evidence even in the deviant case where it is counterproductive to the end of getting and retaining accurate beliefs. 3 My goal here is not to argue for such an accuracy-oriented understanding. Rather, I am going to 3 The promise of accuracy-oriented epistemology has been something of a hot topic; it is tackled under such headings as epistemic consequentialism, epistemic teleology, and epistemic decision theory. Recent literature has taken up the following concern: initially promising results such as those of Joyce (1998) and Greaves & Wallace (2006) have relied on a false assumption of independence between

3 3 skip straight to applying it. I want to show how prioritizing accuracy leads to further interesting conclusions; the ones I shall be particularly interested in concern the current debate on peer disagreement. Specifically, I intend to show how a focus on accuracy can be used to militate in favor of a conciliatory view of disagreement. At a big picture level, conciliatory views are those that hold we should take the set of expert judges who have considered a topic as equivalent indicators of the truth of the matter and, importantly, when we are experts we should include ourselves as just another member in that set, no more significant than any other; in practice, this may often result in something close to splitting the difference with those who disagree with us. I spell out the details of my particular development of the view in section 2. Non-conciliatory views push back against the conciliatory picture by positing some evidence, unrecognized by the conciliationist, that gets lost in the assimilation of judges to indicators, and of oneself to others. This unrecognized evidence, when properly attended to, rationalizes the non-conciliatory degrees of confidence they take to be appropriate or, so they argue. In this paper, though, I show that the evidence they invoke uniformly possesses an unfortunate feature: trying to take account of it tends to decrease rather than enhance our accuracy. But if this is so, then it lacks the essential feature which we have taken to imbue evidence with its value. We are better off ignoring it. The upshot, then, is that there s an argument for conciliation that s surprisingly robust in the face 4 of non-conciliatory claims about what the evidence supports. We can allow that the conciliationist is mind and world; when that assumption is removed, it is no longer clear that accuracy-oriented epistemology gives intuitively acceptable answers. There is no consensus on the lesson we ought draw from this (for instance, whether we ought to thereby abandon accuracy-oriented epistemology). For an informal entry point into the current debate, see Berker (2013); for more technical treatments see Greaves (2013), Carr (ms.), and Konek & Levinstein (2018). I believe the arguments of this paper to be compatible with many different resolutions to these foundational disputes, though I will not legislate this here; for more explanation, see section 15 of Steel (2018a). 4 Some related literature: Van Wietmarschen (2013) argues in a very different way for a conclusion with a similar ring to it, namely, that conciliation is a bad theory of what the evidence supports but that it may nonetheless be a good theory of well-grounded belief. I am obviously broadly sympathetic, but I take it that (on ordinary assumptions about basing) Van Wietmarchen s account will struggle to explain cases where higher-order evidence rationally requires one to become more confident, as, for instance, in the

4 4 indeed ignoring all sorts of genuine evidential features of the situation, just as their critics allege: we are still left with an argument that ignoring those features is exactly what they ought to be doing. Or, so I aim to convince you. 2. What is Peer Disagreement? Think of something interesting you have an opinion on: which candidate is going to win the coming election, if there are any gods, whether a diet high in cholesterol causes heart disease, and so on. If you indeed picked something interesting, then almost certainly there are very smart people who vehemently disagree with you. Noticing this, and being correspondingly disturbed, is not new. Nearly two millennia ago, considering the existence of entrenched disagreement was one of Sextus Empiricus routes to skepticism. And it is easy to impress on just about anyone the force of the question: how can you really be so confident, when after all it might be you who is the one who s gotten it wrong This is a useful backdrop when motivating an attention to disagreement. I will not, however, work on such a broad scale. Rather, I restrict my attention to a special class of case, a class of case I will refer to as peer disagreement. Namely, I consider how one ought respond to a disagreement 1) with precisely one other person 2) over a single yes-or-no question 3) where one takes that disagreeing other to be an epistemic peer with respect to the question under dispute. These restrictions are in some sense artificial; we should expect our ultimate theory to also tell us how to handle multi-party disagreements over multiple propositions, we should expect it to go beyond yes-or-no questions (e.g., not just has the less-often discussed case of peer agreement. But I take disagreement and agreement to be two sides of the same coin, and we should be partial to a unified account of each. See Schoenfield (2012) for a lucid argument for separating out claims about what the evidence supports from claims about what we should believe in a different sort of case, namely, those where permissivism about rational belief is tempting. And finally, see Worsnip s (2018) for an extended argument that rationality and responsivity to evidence can come apart, particularly in cases of misleading higher-order evidence, though he there resists drawing conclusions about any all things considered -style ought.

5 5 gas run out? but also how long until the gas runs out? ), we should expect it to cover not just disagreement with peers, but also disagreement with experts and neophytes, and so on. Nonetheless, I introduce these restrictions for the standard reason that simplicity brings tractability. Furthermore, although I will not address them here, I do think there are natural paths for generalizing the verdicts I reach; as such, I take it that beginning with the special case will ultimately be fruitful in addressing the big picture concerns mentioned above. But regardless, before we can worry about any such generalizations we need to get the case itself right and given the many competing treatments it has received, that s already a big enough task. I take it that the first two conditions I gave above (that disagreement be with one other person, and that it be over a yes-or-no question) are both clear. The third, however, needs some spelling out: what is it to consider a person to be your epistemic peer with respect to a question? As I conceive of it, taking someone to be your peer with respect to a question consists in holding a special belief relating you, them, and a proposition. Namely, for a person A, a person B, and a proposition P: A considers B their peer with respect to P = def A believes that if B disagrees with them over P, then it s 50/50 which of them is right. Or, in the language of credences: A considers B their peer with respect to P = def A s 5 confidence in P, conditional on A and B disagreeing over P, is ½. That s a mouthful, but here s a straightforward example: Two Bright Students: Misty and Ash are good friends. They are also both clearly the best students in their class. Their averages have hovered around the same stellar number throughout the course so far, and they have traded the top class rank back and forth. In the past, when they have disagreed over the answer to a problem, they have each been right an equal portion of the time. Their final exam, rapidly approaching, consists in a 5 That is to say, when Cr A (P B disagrees with me over P) = ½. Now, it s worth noting that there s something curious about this formulation. Namely, it mixes attitude types it describes a situation in terms of full beliefs, disagreeing over P, but then specifies a particular credence conditional on that situation. This may seem spooky. I take it, however, that the example I m about to give is one where this sort of mixed attitude is intuitively fitting, and, indeed, there isn t any ready-to-hand homogenous replacement that would capture the same content. This general understanding of what it is to be a peer in terms of confidences conditional on disagreement descends from Elga (2007).

6 6 single true or false question. This question will be neither so easy that they can just see the answer, nor so difficult that they are just hopeless at figuring it out: it lies in the broad range of questions that they have been tackling with equal success over the course of the semester. I claim that in the above case, it is rationally required for Misty to think that if, on the final exam, Ash disagrees with her over whether the answer to the final question is true, then it s 50 / 50 which of them is right and, by extension, it s 50 / 50 whether the proposition in question is true. So the story above is one on which Misty is rationally required to have the belief I ve identified with taking Ash to be her peer with respect to the question on the final. Suppose that Misty, being epistemically excellent, does in fact have this belief. Now continue the story as follows: A Frustrating Final: Misty and Ash take the final exam. As they walk out, Ash turns to Misty and says I got T. Misty replies: oh, I got F. With that unfortunate development, we have a case of the sort of I set out to investigate. That is to say, we have a case of peer disagreement. And so we are in a position to ask: in such a case, what should Misty think? On first blush, the answer may look trivial: she should think it s 50 / 50! If Misty had the conditional credence described, doesn t it just follow that she should update according to it? It does not. Misty s earlier belief was conditional on her and Ash disagreeing over the final question; call this fact, the fact that they got contrary answers, the mere fact of disagreement. Even supposing that conditionalization is the correct general way to update one s beliefs, it s still the case that one ought conditionalize on the strongest evidential proposition one learned. As such, a commitment to conditionalization would only require Misty to think it was 50 / 50 if the mere fact of disagreement exhausted the epistemically relevant things she learned. And this needn t be true.

7 7 To wit, there are lots of ways to fill in the story such that it is immediately obvious both that Misty learns something over and above the mere fact of disagreement and that what she learns is epistemically relevant to the confidence she ought to have that she (rather than Ash) has gotten the right of things. So, for instance, she may notice that Ash is visibly drunk when he s taking the test. Or perhaps when she sees the test she sees that it is identical to an online worksheet she had just been studying from; she easily recalls the answer. In either case, it would be rational for Misty to massively discount Ash s disagreement. On the other side of things, she may see that the final exam question concerns material from a day she missed and never studied for. Then, by contrast, when she discovers Ash s disagreement she should become almost certain that he was right and she was wrong. The interesting question, though, is not over such cases. Rather, it arises when we instead imagine that the test proceeds quite unremarkably. What happens in a case of unadorned disagreement, or disagreement per se? In those cases, does Misty learns anything that would justify deviating from the 50 / 50 confidence she would adopt were she to respond just to the mere fact of disagreement? I say: no. So, in the end, I endorse the same result that you might have initially thought was trivial: upon discovering the disagreement of her peer, Misty should think it s 50 / 50 which of them was 6 right. Call this thesis conciliationism. 6 Conciliatory views are popular in the literature, and leading defenders include e.g. Christensen (2007), (2009), (2011), and Elga (2007) and as revised in (2010a); the discussion, pro and con, is now voluminous. As such, the precise meaning people attribute to conciliation has also naturally varied as it s been deployed in different hands not all uses of the word follow my characterization in the main text. For the sake of clarification, I will now note two particular features of the approach I take here. The first feature is that I identify conciliation with a particular way of responding to peers. But there are people who can be conciliationists in that sense while still retaining what are intuitively egoistic and/or nonconformist overall views of disagreement as, for instance, if they thought it a priori almost impossible to get into the situation of regarding someone else as your peer. See for instance Schafer (2015), and, subject to complications, Wedgwood (2010). The second feature is that none of my general statements of the conditions under which one should conciliate include the requirement that one s peerhood belief be rational. This is not an accident: I don t think that it matters, in the final balance, whether Misty is rational in taking Ash to be her peer. But nothing in the argument turns on this; readers who, unlike me, are only willing to endorse conciliation for rational peerhood beliefs may mentally insert the qualifier as desired throughout (eliding this issue is why I chose an example where the peerhood belief inessentially, in my view does happen to be rational).

8 8 3. Bootstrapping and Independence Conciliationists hold that nothing Misty learns over the course of a disagreement per se ought to change her confidence. But the import of this claim is not immediately obvious, insofar as it is not immediately obvious what she learns in disagreement per se, and so not immediately obvious what it is that the conciliationist is insisting is irrelevant. Return to the case. What happens over the course of Misty s taking the final and comparing answers with Ash? Well, she sees the question itself. She reasons through it. She arrives at her answer. Beforehand, she was agnostic about who would be right if they disagreed on the final question, whatever that was. But now she sees that Ash is disagreeing with her not just on a question, but on this question, and that he is doing so by getting that answer. And all of this, unlike seeing that Ash is drunk or etc., is something that she learns whenever she has a disagreement, just in virtue of it being a disagreement; it is, in other words, something she learns in a case of disagreement per se. How does this change things? Suppose Misty reasons as follows: Disputed Reasoning : I got T on the final question. And now that I have seen the content of the final question, I can also see that T is in fact the answer. Ash got F. So Ash must have reasoned incorrectly. Since I can now see that he reasoned incorrectly, our situation is no longer symmetric. I no longer take him to be my peer when it comes to this specific question, and if I need take any account of his opinion at all it need not be as much as I antecedently would have thought. In reasoning in this way, Misty leverages the very facts under dispute in the disagreement while considering how she ought to respond to it. If she can do this, then it will never be the case that thinking it was 50/50 in advance of the disagreement requires thinking it continues to be 50/50 afterward. Even in cases of plain or unadorned disagreement, one will always discover something new, something

9 9 capable of disturbing one s original assessment: namely, the question and the answer one got. It is this form of reasoning that the conciliationist insists is illegitimate. It is no great mystery why one might want to block that Disputed Reasoning. It has an air of dogmatic circularity. Imagine if Misty actually gave such an argument to Ash. Presumably, he would take it to be rather insulting of course if the answer is T then I ve made a mistake but the whole point is that I think it s F! And there seems to be some force to his complaint. So, seeing things through Ash s eyes, conciliationists have proposed the following principle: Independence : In evaluating the epistemic credentials of another person s belief about P, in order to determine how (if at all) to modify one s own belief about P, one should do so 7 in a way that is independent of the reasoning behind one s own initial belief about P. If Independence is true, then reasoning as in Disputed Reasoning is illegitimate. And if Disputed Reasoning is blocked, then it is plausible that there is nothing relevant that Misty learns over the course of the disagreement per se. If she learns nothing relevant, then all she has to respond to is the mere fact of disagreement. So both ways, then, conciliationism seems to stand or fall with Independence. The question of whether conciliatory or non-conciliatory views are correct thus reduces to a question about whether Independence is correct and Disputed Reasoning should thereby be rejected, or whether Disputed Reasoning is legitimate and Independence should thereby be rejected. What we d like, to motivate conciliation, is some motivation for adopting the first package rather than the second. We already have something pushing us in this direction, in the form of Ash s complaint that Disputed Reasoning is obnoxiously dogmatic. Can we develop this thought any further? Here s one tack 7 This principle is proposed in e.g. Christensen (2007, p.16-17) and Christensen (2011, p.1-2). Independence is not intended to block off making reference to the existence or even the properties of one s reasoning--again, provided that those properties are picked out in a way that does not presuppose the correctness of that reasoning (e.g., this reasoning had many steps, but each seemed clear not this reasoning was valid ). How to precisely characterize what is kosher and what isn t with Independence is a fraught question, as it so often is with philosophically important principles, but it is nonetheless usually clear enough how to apply it to a given case.

10 10 we might take: we might say that Disputed Reasoning involves Misty taking her own answer more seriously than Ash s merely on the basis that it is her own. But this is absurd. And, indeed, this seems like a natural way to unpack Ash s ire. Adam Elga seminally gave an argument for conciliation very like this. His argument sought to demonstrate the absurdity of what we might call the extra weight view the view that one may permissibly be extra confident in one s answer merely on the basis that it is one s own. He argued that if giving one s own answer such extra weight is permissible, then Misty can bootstrap her way into undeserved confidence in her own abilities; since such bootstrapping is epistemically abhorrent extra weight must not be permissible. Paraphrased, the argument ran: Suppose it really were permissible to give one s own view extra weight. If that were so, then after disagreeing with a peer, one could be rationally confident that one was right; but if that were so, then one could also become rationally confident in the propositions one s having been right entails, namely that one has a better track-record than one s friend, and hence that one is more reliable after all and all this would be possible in advance of actually receiving any independent confirmation of one s rightness. If this were really rational, that must be because disagreement with a peer is evidence that one is the better judge. But that s absurd. The mere fact of disagreement, absent any independent confirmation that one is right, is no evidence at all that one is the better 8 judge. Hence, the extra weight view is false. As stated, this line of objection targets views which would allot oneself extra weight, but it also applies just as much to views which would allot one s interlocutor extra weight. There the absurd conclusion is just the reverse of the one derived before, namely, that one can take the mere fact of disagreement to be evidence that one s interlocutor is more reliable. The point here is just that the existence of disagreement on its own is no evidence at all either way. Therefore, Elga s bootstrapping objection aims to rule out any deviation from the conciliatory view whatsoever. But it has become clear, though, since this argument s debut, that although it is an excellent 8 Elga (2007, p.12-15).

11 11 argument against the extra weight view it cannot do the work of establishing conciliation. The problem is that it assumes that the mere fact of disagreement is the only epistemically relevant thing learned. Then, on that basis, it is able to construct a dilemma where the only choices are conciliation or extra weight. If that dilemma were genuine, then conciliation would indeed follow. But the claim that the mere fact of disagreement is the only thing you learn and so that any extra confidence gained must be sourced in brute extra weight, rather than new evidence is itself just another way of phrasing the thing we really want to demonstrate, namely a conciliation-friendly Independence-style principle. Indeed, non-conciliatory theorists have responded by allowing that bootstrapping considerations are fatal to the extra weight view, but then going on to carefully distinguish their own views from it. They have done so by clarifying their account of the relevant evidence. In so doing, they show that at no point do they appeal to the mere fact of disagreement, and so at no point are they guilty of taking to be evidence something that obviously isn t. Instead, they characterize some further, richer evidence they claim to be present in disagreement cases that looks like it does have the right sort of rational force. The upshot of this dialectic, then, may look grim for the conciliationist. Non-conciliationists have begun to flesh out their conceptions of the relevant evidence in ways that absolve them of any appeal to extra weight. These conceptions, when admitted, put serious pressure on Independence; they describe prima facie good evidence that Independence would require one to ignore. Worse yet, the best-known defense of Independence as expressed in the bootstrapping argument turns out to presuppose rather than motivate it. So the overall appearance may be of a route; conciliation needs Independence, and Independence is in serious trouble. I will argue that things are not so gloomy as all that. I agree with its non-conciliatory critics that the original bootstrapping argument cannot do the work required of it. I hold, nonetheless, that worries about extra weight can be rehabilitated by way of a closely related argument. This new presentation reformulates Elga s objection out of the language of evidence and into the language of accuracy. So

12 12 reformulated, the new objection can catch non-conciliatory views in its net. For although they have managed to give a description of the evidence that differentiates their justificatory structure from the dubious justificatory structure of the extra weight view, non-conciliatory views nonetheless bear an important relationship to it when we focus just on their outputs. I call this relationship functional entailment : temporarily suppressing some complications, my claim will be that one who tries to follow a non-conciliatory view will in fact reach all the same results as the ones outputted by the extra weight view. But the outputs of the extra weight view are bad. They re losers from the perspective of accuracy, and thus so too are the identical outputs of non-conciliatory views. Independence, then, is rehabilitated by attending to considerations of accuracy. Before I give this argument in greater depth, I need to introduce the non-conciliatory views I aim 9 to criticize. I will canvass two major alternatives. First I ll describe Enoch s steadfast view. Then I ll 10 describe Kelly s total evidence view. These views capture the two most obvious ways of breaking the apparent symmetry in peer disagreement: either by reference to who s who, or by reference to who s right. So, first let s get onto the table their conceptions of the relevant evidence that arises over the course of disagreement; once that s done, I can go on to show how, as so specified, 1) trying to respond to such evidence functionally entails extra weight and 2) that extra weight sabotages accuracy. 4. The Steadfast View 9 Enoch (2010). 10 Kelly (2010) and (2013). One significant view that I do not explicitly discuss is Lackey s justificationist view. I take the justificationist view to be, in the respects relevant to the argument of this paper, isomorphic to the total evidence view; my same criticisms will apply mutatis mutandis. See Lackey (2010a), (2010b). For similar reasons, I also eschew discussion of a third family of view, the right reason view of e.g. Kelly (2005), Weatherson (2013) and (ms.), and Titelbaum (2015); again, the short reason is that my criticism of total evidence will apply mutatis mutandis to it. I elaborate on this in section 7 and passim in Steel (2018a); I also give a separate, paper-length treatment to right reason in my (2018b).

13 13 First, what is Enoch s position? He argues as follows. The only way in which it is possible to form judgments about who is a peer and who is not is to look to someone s track record: one must ask how reliable has she been on this subject in the past. Furthermore, the only way to do that is to compare her history of judgments to one s own and see how well they match we have nowhere else to start, when judging the reliability of our peers, than from our own views. But if we methodologically take ourselves to be right in our evaluation of putative peer s past reliability, then it would be arbitrary not to do the same with our present conflict. Thus, Enoch advises, we ought respond to peer disagreement by, among other things, demoting our putative peer to some degree; her new track record, this judgment included, is worse than their old one. So Independence is false, and Disputed Reasoning is legitimate. But doesn t this involve treating the mere fact of disagreement as evidence that one is a better judge that one s peer? It does not. Enoch is careful in differentiating his view from the extra weight view: he insists that the grounds on which we demote our peer is not the mere fact of disagreement, but rather, it is the deterioration of their track record. And the deterioration of their track record is good 11 evidence that they are not good judges. To illustrate: suppose I take you to be my peer on matters p-related. I judge p. I then discover that you have judged not-p. On Enoch s view it is appropriate for me to demote you, but not on the grounds that I judged p and that you judged not-p, but rather, on the grounds that p and that you judged not-p. Of course, what makes it the case that it s appropriate for me to demote you on the grounds that p is that I have judged p, but that does not make judging p itself my grounds. Our beliefs are, and necessarily must be, transparent insofar as we act from their contents directly, rather than from hedging propositions like I judge p. Since I am acting directly from the content of my judgment, rather than from the fact that I so judged, my judgment is based in your deteriorating track record itself rather than merely the fact that I so 11 Enoch (2011, p ).

14 14 judged. There is some relevant evidence here, the evidence of track records, and that distinguishes the steadfast view from the extra weight view. 5. The Total Evidence View In cases of peer disagreement, Independence screens off the particular belief in question thus prohibiting p, therefore you re wrong about ~p but that s not all it screens off. It also screens off the reasoning used to arrive at p. There are obvious reasons for this. Imagine p was screened off but the reasoning supporting it was not. Then one could simply re-conclude p from that reasoning, and then proceed as in Disputed Reasoning. But Independence is designed to block Disputed Reasoning. So 12 Independence must screen off that supporting reasoning as well. This, however, opens up Independence to the charge that it throws out evidence. After all, if I was correct in my reasoning from my evidence to p, then that evidence really does support p. I should not ignore it when coming to my final opinion. Conciliation falsely treats the case as if the only evidence available in the wake of a disagreement with a peer were the fact that we had differing beliefs. But there is more evidence than that there is the evidence on which we based those differing judgments in the first place. So, for instance, Kelly imagines a case of peer disagreement where we both initially form our views on the basis of some evidence E. After we consult with each other, and discover our differing conclusions, our new total evidence includes both the original evidence E and the fact that we reached contrary responses; call this new pool of evidence E*. He then puts the point as follows: Notice that, on the Equal Weight View, the bearing of E on H turns out to be completely irrelevant to the bearing of E* on H. In effect, what it is reasonable for you and I to believe about H... supervenes on on how you and I 12 Christensen makes the point in (2011, p.18) when introducing a problem about how to define the scope of Independence.

15 15 respond to E... E gets completely swamped by purely psychological facts about what you and I believe But why should the normative significance of E completely vanish in this way? The original evidence, E, is still available to us, and so it should not vanish. But what of that original motivation for Independence ruling out Disputed Reasoning? Would Kelly assent to p; you think ~p; therefore you are a worse judge of p than I? The answer, it turns out, is it depends. If, after adding the fact that you think ~p to my total evidence, it still supports p on the 14 balance, then I am allowed to so proceed. If it does not, I am not. The philosophical assumption that there is any general, informative epistemic rule to decide the question is false. Deciding what the total evidence supports in any particular case is hard, and certainly more difficult than simply applying a 15 formal rule but such are the burdens of judgment. Kelly s view, then, is also distinct from the extra weight view. Whereas the extra weight view tells you to assign extra weight to your own answer as such, Kelly s view tells you to attend to the total weight of the evidence; if the evidence really does favor the answer you initially came to, then you should respond to it by retaining a higher credence in your own answer. But if it favors something else, you ought to believe that other thing. What is important here is just the disposition of the evidence, and nothing in that description makes reference to your view as such. 6. Non-Conciliatory Views Functionally Entail Extra Weight So: all of these non-conciliatory views posit something epistemically relevant which comes into Misty s possession over the course of the disagreement. Call this epistemically relevant stuff, whatever it is, E. The non-conciliatory views say: E is the sort of thing that rationalizes deviating from the 50/50 13 Kelly (2010, p. 124). 14 Kelly (2013, p ). 15 Kelly (2013, p. 52).

16 16 verdict with which Misty began. So that is the non-conciliatory story about why both Independence and the conciliatory views that rely on it are false. Independence would exclude, and conciliatory views ignore, this E. But forget for a moment what E rationalizes. That is to say: forget what Misty should do with it. It s worth asking, instead, what Misty will do with it. If Misty picks a non-conciliatory view and tries to follow it, what will happen? We have to exercise some care with the register in which we ask this question. After all, what someone will do is typically an empirical question, and as such subject to all sorts of contingencies. So, for instance, perhaps Misty happens to be emotionally attached to the conciliatory view, whereas she strongly dislikes the steadfast and total evidence views; if she tried to follow either of the latter then she would be so distressed that her poor mood would compromise her reasoning. That is the sort of thing we can t rule out without knowing Misty. When I ask what Misty will do when she picks a non-conciliatory view and tries to follow it, those are not the sorts of influences I am interested in tracing out. Rather, the question I m getting at isn t really empirical at all; it s about what happens when a person tries to follow e.g. the total evidence view specifically in virtue of their trying to follow an epistemic view with that particular content. This is an a priori question about our understanding of what it is to try to follow a norm, not a question about anyone s empirical-psychological quirks. This formulation may sound threateningly opaque, but here it s rather straightforward; I take it that it will be eminently clear what I mean as we see how things play out for each view. To prefigure, though, the answer will be: on this, the non-conciliatory views we have considered all agree. If Misty tries to follow them, and in so doing to take proper account of their E, she will in fact become more confident that she, rather than Ash, has had the better of the disagreement. To repeat our earlier terminology: epistemic view A functionally entails epistemic view B iff

17 17 someone who attempts to act in conformance with A will, in fact, wind up arriving at all the same attitudes as someone who was actually acting in conformance with B. What non-conciliatory theorists concede and it is hard to see how they could not is that their views functionally entail the extra weight view. People who try to follow non-conciliatory views do not justify their answers in terms of extra weight; rather, they justify them in terms of E. Nonetheless, you couldn t tell the difference from looking at the answers they get. This is easiest to see with respect to the steadfast view. On the steadfast view, both Misty and Ash receive an E such that each of them may become rationally confident in their own correctness on its basis. But each becomes more confident in their own correctness is just the same result that would be arrived at by parties following the extra weight view. As such, the steadfast view functionally entails the extra weight view. Enoch himself explicitly acknowledges this functional entailment. But he is untroubled. He says: one can foresee, on his view, that one will arrive at all the same verdicts as the person who favored their own view with extra weight. Still, the person following his view does not act under the intention of giving herself extra weight as such, and this difference between foresight and intention is epistemically 16 relevant. So he acknowledges the functional entailment while disputing its significance. That s fine; the point for now is just that there is such a functional entailment. The total evidence view is not symmetrical in the way that the steadfast view is. On the total evidence view, Misty and Ash receive a single E, and that E is such that it is tilted toward the person who was in fact better responding to the evidence when they formed their initial judgment. So both of them ought to arrive at the same judgment, namely one tilted toward the person who was in fact better responding to the evidence when they formed their initial judgment. Does this lack of justificatory symmetry functionally distinguish it from the (symmetric) extra weight view? 16 Enoch (2011, p ).

18 18 It does not. Since the answer to the question of which party E tilts toward is itself dependent on who was right in their initial judgment, in attempting to ascertain and thereby properly respond to the tilt of E both parties will have no alternative to re-deploying those same differing initial judgments. And this is something that Kelly also explicitly acknowledges. As he puts it: there is no warning bell that goes off when you are mis-evaluating the evidence, and which lets you know that you re the one in the dispute who ought to lay down your arms; furthermore, the relevant facts may not seem to be facts at all 17 from your perspective. So in trying to take account of the tilt of E, it s inevitable that one will de facto take it to tilt toward oneself. The total evidence view, just as much as the steadfast view, also 18 functionally entails the extra weight view. So, these non-conciliatory views functionally entail the extra weight view. When we forget about what people should do and focus just on what they will do, we see: people with non-conciliatory views will respond to the evidence E that arises in disagreement by acquiring just the same attitudes that one would acquire by putting some extra weight on one s own view, taken as such. Given that some there is some, how much extra weight is functionally entailed by these views? Misty starts out, before the final, taking things to be 50 / 50. How far away from that will she get, on the 19 basis of E? Let s name that distance, whatever it is, N. Misty gets up to 50 + N / 50 - N. 17 Kelly (2010, p.165, ). 18 Indeed, this is why Setiya (2012) suggests, in friendly development of the total evidence view, that although the stronger status of justification should be reserved only for the correct party, the status of blamelessness should still be allowed to the incorrect. 19 I make only two assumptions about N, both of which I take to be harmless. The first thing I assume is that N is sometimes non-trivial. Non-conciliatory views would not present a very interesting or exciting alternative if they posited some E such that it allowed Misty to be as much as an extra thousandth of a percent more confident in her correctness. Rather, the difference ought to be sometimes substantial. The second thing I assume is that, even if the value of N is not precisely knowable in advance of any particular disagreement, nonetheless a reflective and rational subject could at least estimate it. On some views rational subjects will have perfectly sharp credences in even the most outlandish of propositions: see, for instance Elga (2010b). That would be congenial to my argument, but I need not assume it. All I need to assume is that: however well behaved things need to be to yield an estimate that is itself well behaved enough to figure in practical reasoning, N is at least that well behaved. This is satisfiable both if N is poorly behaved but good behavior is not required, or if good behavior is required but N turns out to be well behaved.

19 19 7. Conciliation Maximizes Expected Accuracy: The Informal Case Having argued that non-conciliatory views functionally entail extra weight, I now turn to drawing out the consequences of that fact. I am specifically interested in what happens when someone who holds a non-conciliatory view considers the fact that their view functionally entails extra weight; my goal is to argue that such a person would no longer be interested in responding to the sort of evidence their view posits. Rather, they would see responding to such evidence as a bad deal. To illustrate, let s return to Misty s case. Suppose she takes up the position we re interested in: she both holds a non-conciliatory view and realizes that her view functionally entails extra weight. Now suppose she looks forward to the possibility of disagreeing with Ash on the final. How does she expect things to go? Misty now believes that if they disagree it s 50 / 50 whether she or Ash will be right (because she takes him to be a peer). But she also believes that if they disagree, then in acquiring and responding to E she will then become convinced that it s 50 + N% / 50 - N% favoring her (because she recognizes that her view functionally entails extra weight). So she believes she will gain a certain degree of confidence no matter what. Our question now is: should she foresee that gain in confidence as making her more or less accurate? I answer: less. So, insofar as she is interested in gaining and retaining accurate beliefs, she should prefer not to undergo that gain in confidence. And so she should prefer to ignore E 20 rather than take account of it. The remainder of this section marshalls some informal considerations in defense of the claim that Misty ought to not want to undergo the predictable gain in confidence that responding to E 20 My focus on Misty s present evaluation of her future performance may invite the suspicion that my argument implicitly relies on or otherwise reduces to a reflection principle of the sort seminally discussed in Van Fraassen (1995). See section 15 of Steel (2018a) for my explanation of why this is not the right way to gloss my argument.

20 20 occasions. My argument takes the following shape: for each of the non-conciliatory views we ve been considering (steadfast and total evidence) I will first explain how E -type evidence, as they conceive it, predictably prejudices one toward one s own correctness. I then use that explanation to construct a parallel case containing relevantly similar evidence. I argue it is intuitively clear in those parallel cases that, when it comes to getting and retaining accurate beliefs, we should not want to respond to that analogous evidence. But since that evidence is relevantly similar to E -type evidence, this then furnishes us with an argument we should not want to respond to E -type evidence either. So, let s begin with the steadfast view. Recall that it is is symmetric; on the steadfast view, both parties to a disagreement may, on the basis of the E they acquire, become rationally more confident in their correctness. This because the E each receives supports the claim that the other s track record as a judge has gotten worse, and so, on the basis of this evidence, each is rationally entitled to downgrade the other. Thereby, they may each rationally retain more confidence in their own answer and less in the other s. But, of course, note that it will not be true for each of them that their opponent s track record has gotten worse; that will be true only for the person who got it right. Thus, the person who got it wrong will be downgrading their opponent on the basis of misleading evidence. The summary, then, is that on the steadfast view, each disagreement situation is one where two new E s are generated, with each of them symmetrically supporting opposite conclusions about who was right. Because these E s are by nature generated in opposing pairs, for each non-misleading E there is a paired misleading E. When Misty thinks ahead to undergoing disagreement with Ash, does she think she is any more likely to get the non-misleading E as opposed to the misleading E? No. The person getting a non-misleading E is the correct respondent, and Misty currently thinks she is no more likely than Ash to be the correct respondent. That s just the content of her peerhood belief, as we earlier defined it. All together, then, we get the result that a steadfast and reflective Misty anticipates disagreement with Ash proceeding as follows: she will get some evidence E, and, no matter what, this evidence will support her

21 21 coming to be more rationally confident in her own correctness--but, at the same time, it will be perfectly random whether this evidence will be misleading or not. This, I take it, is a disturbing state of affairs. More specifically, it is a state of affairs in which Misty should anticipate that getting and responding to E will make her beliefs worse, not better. To bring this out, consider the following thought experiment. Suppose we are all ruled by an absurdist demon. One day you run into her, and while chatting she lets it slip that she finds the idea of people coming to understand the world by way of reading newspapers to offend her sensibilities. And so every time some piece of non-misleading evidence is reported in a newspaper, she then exercises her powers to ensure that some other misleading report is also generated. The nature of this second report is that it always contains misleading evidence of the exact same strength as in the original report, and this misleading evidence always instead indicates the opposite conclusion. Some newspaper reports evidence for P, which is true; she ensures that some other newspaper reports equivalently strong evidence for ~P, which, of course, is false. She does the reverse as well: every time some newspaper reports evidence for ~P, which is misleading, she ensures that some other newspaper reports equivalently strong evidence for P, which isn t. The result, overall, is a perfectly even balance. Newspapers are composed, 50-50, of misleading and non-misleading evidence. I submit that if one were to learn this was happening, then the first thing to think would be that reading newspaper reports and believing them was a bad way to come to have accurate beliefs about the world. By opening a newspaper, one would be no more likely to get non-misleading evidence than misleading evidence. As such, there would be no expectation that getting and responding to newspaper evidence would lead one systematically toward truth. Rather, it would have a brute scrambling effect on one s beliefs. In terms of its truth-conduciveness, reading and responding to the compromised newspapers would be no better (or worse) than being hit on the head with a special stick which simply induced, via trauma, random additions and subtractions of confidence in one s beliefs. The effects, in