How Common is Peer Disagreement? On Self-Trust and Rational Symmetry

Philosophy and Phenomenological Researchdoi: 10.1111/phpr.12169© 2015 Philosophy and Phenomenological Research, LLC

How Common is Peer Disagreement?On Self-Trust and Rational Symmetry

KARL SCHAFER

University of Pittsburgh

In this paper I offer an argument for a view about the epistemology of peer disagreement,which I call the “Rational Symmetry View”. I argue that this view follows from a natural(if controversial) conception of the epistemology of testimony, together with a basic enti-tlement to trust our own faculties for belief formation. I then discuss some objections tothis view, focusing on its relationship to other well-known views in the literature. Theupshot of this discussion is that, if the Rational Symmetry View is correct, much of theaction in the epistemology of disagreement relates—not to how one should treat thoseone regards as an “epistemic peer” in the sense popular in that literature—but rather towho one should treat as such.

Consider the following two views about the epistemology of peer disagree-ment:

Equal Weight View: When two individuals, who rationally regard eachother as “epistemic peers”, and who are equally aware of each other’s evi-dence, have a disagreement about some issue, they should both give equalweight to both of their opinions about that issue.1

Rational Asymmetry View: When two individuals, who rationally regardeach other as “epistemic peers”, and who are equally aware of eachother’s evidence, have a disagreement about some issue, those individu-als that have responded correctly to their initial evidence should be

1 See, e.g., Adam Elga. “Reflection and Disagreement.” Nous 41 (3): 478–502 (2007) andRichard Feldman. “Epistemological Puzzles About Disagreement.” in Stephen Hethering-ton (ed.), Epistemology Futures. (Oxford: Oxford University Press, 2006).

I use “who are equally aware of each other’s evidence” as opposed to “who share allof the relevant evidence” to allow for views that characterize an individual’s evidence interms that make it misleading to speak of disagreeing parties as ever literally sharing allthe evidence.

HOW COMMON IS PEER DISAGREEMENT? 1

Philosophy andPhenomenological Research

more confident in the wake of the disagreement than those who havenot.2

Each of these views has important proponents in the contemporary literatureon these issues. But, while it has received a good deal of attention, the samecannot be said of the following:3

Rational Symmetry View: When two individuals, who rationally regardeach other as “epistemic peers”, and who are equally aware of each other’sevidence, have a disagreement about some issue, both should give moreweight to their own opinion than they do to their apparent peer’s.

As stated each of these views is quite vague. Indeed, as I will discuss inmore detail below, depending on how they are made precise, they are notnecessarily incompatible with one another. For it may be the case that oneof them is true, given one interpretation of the crucial idea of an “epistemicpeer”, while another is true, given another interpretation of it. Thus, as willbecome clear, although these views may seem to be competitors with oneanother, ultimately they may be best understood as addressing related, butdistinct questions in the epistemology of disagreement.

Nonetheless, for the moment, I want to leave open how “epistemicpeer” is best read. But it is important to stress at the outset that all ofthese principles are meant to apply only to cases in which neither partyhas any special reason (all things considered) to treat themselves as moreor less reliable than the other party. For if either party did have specialevidence of this sort, they would obviously not be in a position to ratio-nally regard the other party as an epistemic peer in any interesting senseof this term.

As we proceed the variant of “epistemic peerhood” that is of interest tome will become clearer.4 In this discussion, my initial aim will be to showhow a natural, although not at all uncontroversial, story about the epistemo-logical significance of testimony generates a version of the Rational Sym-metry View. This, I hope, will provide us with some reason to take this

2 See e.g. Thomas Kelly. “Peer Disagreement and Higher Order Evidence.” in Alvin I.Goldman & Dennis Whitcomb (eds.), Social Epistemology: Essential Readings. (Oxford:Oxford University Press, 2010).

3 But see Ralph Wedgwood. “The Moral Evil Demons.” in Richard Feldman & Ted War-field (eds.), Disagreement. (Oxford: Oxford University Press, 2010) and David Enoch.“Not Just a Truthometer: Taking Oneself Seriously (but not Too Seriously) in Cases ofPeer Disagreement.” Mind (2010) 119 (476): 953–997 for two important exceptions tothis generalization, although neither endorse the precise view defended below.

4 In particular, I discuss different ways of thinking of epistemic peerhood at some length inthe Objections and Replies at the end of this essay.

2 KARL SCHAFER

view more seriously than it often is—even if it is far from a knockdownargument in its favor.

What story about testimony do I have in mind? The picture of interest tome can be found in many places, but an elegant statement of it is the fol-lowing, taken from R.M. Hare’s The Language of Morals:

A scientist would not have become a scientist unless he had convincedhimself that the observations of other scientists were in general reliable.He did this by making some observations of his own. When we learnt ele-mentary chemistry at school, we had some theoretical periods and somepractical. In the theoretical periods we studied books; in the practical peri-ods we made experiments, and found, if we were lucky, that the results tal-lied with what the books said. This showed us that what the books saidwas not all nonsense; so that even if, by reason of disturbing factorsignored by us, our experiments came out wrong, we were inclined to trustthe books and acknowledge that we had made a mistake. We were con-firmed in this assumption by the fact that we often discovered later whatthe mistake had been. If our observations, however carefully we did them,were always at variance with the textbooks, we should not be tempted tomake science our profession. Thus the confidence of the scientist in otherpeople’s observations is ultimately based, among other things, on his ownobservations and his own judgements about what is reliable. He has in theend to rely on himself.5

For our purposes here, two elements of Hare’s story are particularly impor-tant:

Self-Trust: Every normal individual is entitled, from the beginning ofinquiry, to rely on her own basic belief-forming faculties, at least to somedegree.6

Testimonial Entitlement: Our entitlement or obligation to rely on thebeliefs and belief-forming faculties of others is derivative from the evi-dence generated by compliance with this first entitlement.

Both of these ideas are very controversial. But nonetheless, they are quitenatural views to take on these issues, and thus it is not surprising that theyhave found many proponents.

5 Oxford: Oxford University Press, 1952.6 Self-Trust as stated applies to all our “basic” faculties equally. But it leaves open which

faculties count as “basic” in this sense—which will obviously have to be settled by fur-ther epistemological theorizing. Moreover, it may be restricted in various ways to accom-modate a variety of epistemological views. Also note that, of course, as we gain evidenceabout the relative reliability of our various faculties, the degree to which we are entitledto trust them will vary from case to case.


In particular, some version of Self-Trust sits at the core of many anti-skepti-cal views in epistemology. To be sure, this does not mean that every anti-skep-tical epistemologist must accept a principle of this sort. For example, onemight reject it, while avoiding skepticism, by modifying Self-Trust so that weare only entitled to rely on a belief-forming faculty when it possesses some fur-ther epistemic virtue such as reliability or safety. But while it is plausible thatthese sorts of characteristics matter with respect to a faculty’s ability to serveas a potential source of knowledge, it is also quite natural to think that there isa basic sense in which we are rationally entitled to rely on all our basic belief-forming faculties, whatever they are, provided that we do not have any positiveevidence that they are not reliable. Thus, a principle like Self-Trust, interpretedin this general fashion, has seemed attractive to many philosophers.7

Testimonial Entitlement, on the other hand, is a good deal more contro-versial. But while this principle may call to mind “reductive” or “inferen-tial” views of testimony, it is weaker than what such views normallyrequire. All that Testimonial Entitlement requires is that our entitlement togenerally rely on others in forming beliefs is somehow dependent upon theevidence provided by our own faculties.8 But it leaves entirely open howthe one might depend on the other. Thus, while Testimonial Entitlementmay call to mind views on which our entitlement to trust the testimony ofothers is based on inference from past a posteriori evidence of their reliabil-ity, it does not imply that any such view is true. Rather, it is entirely com-patible with views on which we have an a priori entitlement to trust suchtestimony—so long as that entitlement is dependent in some way upon thea priori evidence provided by our own faculties. Thus, while TestimonialEntitlement represents a “reductive” view of testimony in some sense, thisis true only if we use “reductive” in a quite broad sense.

Of course, even this weaker claim is very controversial. For, many phi-losophers, myself included, are inclined to think that we have an entitlementto rely on the testimony of others that is not dependent in this fashion onthe evidence provided by our own faculties. By the end of the paper, I’llweaken Testimonial Entitlement to accommodate this thought. In particular,

7 For two recent examples, see Richard Foley. Intellectual Trust in Oneself and Others.(Cambridge University Press, 2001) and Michael Huemer. “Compassionate Phenomeno-logical Conservatism.” Philosophy and Phenomenological Research (2007) 74:1. Andcompare Keith Lehrer’s Self-Trust: A Study of Reason, Knowledge, and Autonomy.(Oxford: Oxford University Press, 1997), which builds upon Reid’s classic discussion ofrelated ideas.

For some very general arguments in favor of the necessity of a certain sort of “self-trust”, see Adam Elga and Andy Egan’s “I Can’t Believe I’m Stupid.” PhilosophicalPerspectives (2005) 19(1): 77–93.

8 Views of this sort are, of course, often associated with Hume’s discussion in “On Mira-cles”. For a more recent discussion of similar ideas, see Jennifer Lackey’s Learning fromWords (Oxford: Oxford University Press, 2008).

4 KARL SCHAFER

we’ll see that the argument below only rests on the claim that the basiclevel of confidence we are entitled to have in others is not as high as thebasic level of confidence we are entitled to have in ourselves. And this issomething many philosophers who rejected Testimonial Entitlement wouldnonetheless accept.

That having been said, for the sake of simplicity, I’m going to begin byusing Testimonial Entitlement as a working hypothesis in order to arguethat a version of the Rational Symmetry View follows from Self-Trust,when it is paired with Testimonial Entitlement. Then I’ll explore whetherthe argument I’ve given will go through using the weaker claim just noted.

***

In some sense, this should not be terribly surprising. After all, given theseassumptions, our rational level of confidence in the reliability of others willbe derivative from our rational level of confidence in our own reliability.Thus, our level of confidence in our own faculties’ reliability will, all otherthings being equal, constrain our level of confidence in the reliability of thefaculties of others in a manner that generates a certain sort of universalfirst-person bias in favor of our own reliability; and so a view of disagree-ment that may be regarded as a version of the Rational Symmetry View.

In order to explore this line of thought more carefully, let me begin bymaking some terminological remarks. Throughout the following, I will usephrases like “treating X as reliable”, “treating X as more reliable than Y”,“regarding X as reliable”, and so forth. By these phrases, I mean somethingvery simple, but there are ambiguities in this neighborhood that should beavoided. In the sense of interest to me here, when someone treats X as n%reliable (about some question) this simply means that the probability theyassign to P conditional on the proposition that X endorses P is .n.9 Of course,insofar as one is rational, this conditional probability will generally be alignedwith the probability one assigns to propositions about X’s reliability. But thisalignment may be more or less complex—since one may come to treat X as n% reliable in this sense either by (say) being certain that X is n% reliable orby giving equal credence to the proposition that X is less than n% reliable andthe proposition that X is more than n% reliable, and so on.10

9 Note that on this way of speaking, treating someone as 0% reliable involves treating himas perfectly anti-reliable.

10 On the other hand, both of these states of mind should be sharply distinguished fromone’s credence in the proposition that X rationally ought to treat herself as n% reliable.For given the discussion below, there is no easy inference from the proposition that Xrationally should treat herself as n% reliable to the proposition that anyone else shouldtreat X as so reliable or (indeed) to the proposition that X is in fact so reliable. (Formore discussion of this issue, see objection 3 below.)


Thus although such phrases may call to mind the idea of treating oneselfas a mere reliable indicator of the truth (a mere “truthometer”) that DavidEnoch has recently criticized, they needn’t have the implications Enochfinds problematic. Indeed, the conclusions of this essay are very much inkeeping with Enoch’s own about these issues.11

With this in mind, let’s move on to consider the implications of Self-Trust plus Testimonial Entitlement. To do so, let’s consider a very simpleexample that brings out the relevant features of such cases. Consider twoindividuals, Carla and Boris, who regularly discuss some area of knowl-edge, and that area only, with each other via email. Sometimes they agreewith one another. But sometimes they do not. Moreover, for the sake ofsimplicity, suppose that neither Carla nor Boris has any special evidence,independent of their disagreement, that either of them is unreliable aboutthe area of knowledge in question. So, for example, suppose that whenBoris reviews his past opinions about the area of knowledge in question, hediscovers nothing that would give him cause to lower his own evaluation ofhis reliability. Thus, independently of their disagreement, both Carla andBoris will (in this area at least) retain the level of confidence in their ownreliability that Self-Trust entitles them to.

Now, let’s consider how Boris and Carla should react to their disagree-ments. By Testimonial Entitlement, whatever epistemic significance this dis-agreement has for them must be based in the evidence that their ownfaculties have provided each of them about this or related matters. Giventhis, it is important to stress that, given these assumptions, this is not a casein which Boris should begin inquiry by treating Carla as equally reliable ashe is. Rather, it is a case in which Boris begins by having no view aboutCarla’s relative reliability and then attempts (in accordance with TestimonialEntitlement) to use the verdicts of his own faculties to arrive at a viewabout Carla’s reliability.12

11 In particular, while I assume that in forming beliefs one will implicitly treat one’s facul-ties as being reliable to such and such a degree, I don’t think Enoch would take issuewith this. What is important here is that in doing so, I am not treating myself as a mere“truthometer”. Thus, to speak in Enoch’s terms for a moment, it is perfectly possible todevelop the Rational Symmetry View so that it involves forming beliefs in a manner thatone knows will involve giving oneself extra weight—as opposed to forming beliefs inthe manner one does because one gives oneself extra weight. For the sake of simplicityI leave such complications to the side for the most part.

12 This points to an important potential ambiguity in phrases like “Boris starts out with theassumption that there is no reason to treat Carla as either less or more reliable than heis.” One way to understand this phrase is as implying that Boris should begin by treatingCarla as likely to be just as reliable as he is. But, as noted above, this misdescribes thecase of interest here. On the other hand, this phrase might be read as implying that (priorto consulting his own faculties) Boris has no reason for having any view at all aboutCarla’s reliability. It is this reading of this phrase that accurately describes the caseabove.

6 KARL SCHAFER

Given this, to consider this issue, we need to consider how Boris andCarla’s faculties might provide them with evidence that is relevant in thisway. One way in which this might occur is obvious. For, insofar as Boristrusts his own faculties, when something seems true to him in the area inquestion, he will take that to be evidence of its truth. And so, when Carlaagrees with Boris, he will take this to be evidence of her reliability. Thus,insofar as Carla agrees with Boris, all other things being equal, he will takeher to be a reliable source of information in this area.

Now, suppose for the moment that this is the only ground that Boris hasfor treating Carla as reliable—and more generally for thinking about therelationship between their levels of reliability.13 Of course, under normalconditions, Boris’s faculties will provide him with many other forms of evi-dence about Carla’s reliability. For example, he will normally have all sortsof evidence about the reliability of human cognitive faculties in general. Ina moment, we’ll consider the importance of these further forms of evidence.But let’s suppose for the moment that no such evidence is available toBoris. The point I want to stress here is that, if this is the case, then givenTestimonial Entitlement, Boris will never have cause to treat Carla as morereliable than himself. After all, if their agreement and disagreement is theonly way in which Self-Trust gives Boris a basis for trusting Carla, and theabove assumptions are in place, then Boris’s confidence in Carla’s reliabilityshould be a simple function of his confidence in his own reliability and thedegree to which they agree (or have agreed). And this will mean that Boriswill count Carla’s agreement with him as having positive epistemic signifi-cance only to the degree that the verdicts of her faculties agree with verdictsof his. Thus, this will mean that Boris’s confidence in his own reliabilitywill constrain his confidence in Carla’s reliability so that the former isgreater than or equal to the latter.

***

To see why this is the case, consider an even simpler case with the samebasic structure. Suppose I have a list of 100 yes/no questions and answersto each. And suppose that I know that each of these answers is 95% likelyto be correct. Now suppose I am given a second list with the same 100questions and answers to each, where these answers agree with the answersin the original list in the first 95 instances and disagree with them in theremaining 5. And suppose I know nothing at all about the second list. In

13 Note that this means that we are not making any assumptions here about whether Carlaand Boris’s views are probabilistically independent of one another. Indeed, as we willsee in a moment, the reasoning described below is best understood as simultaneouslyproviding Boris with a rational view of Carla’s reliability and a rational view of the levelof independence between their views.


particular, I have no evidence of its reliability or the degree to which itsanswers are independent of the answers in the first list. (Of course, this isan assumption that would never be realized in the real world, but we canstill consider what would be true if it were.)

Given this, and the assumptions above, what should I think about thenumber of correct answers on the second list? If we make one further sim-plifying assumption about this case, we can calculate the expected numberof correct answers on the second list as follows. For suppose that the ques-tions on each list are independent of the other questions on that list—so thatone list getting a question right or wrong has no effect on the likelihoodthat that same list will get any other question right or wrong.14 Then theexpected number of correct answers on the second list will equal the sumof the probabilities that each particular answer on the second list is correct.But in this case, my only source of evidence about whether each questionon the second list is correct is provided by whether it agrees with the corre-sponding answer on the first. Thus, in cases in which these answers agree, Ishould conclude that the answer on the second list has a 95% chance ofbeing correct. While in cases in which they disagree, I should conclude thatthe answer on the second list has only a 5% chance of being correct. Thus,given all this, I can calculate the number of expected correct answers on thesecond list as follows: (95)(.95)+(5)(.05) = 90.5.15

In this way, I will also be led to the conclusion that the second list is slightlyless reliable than the first. I suspect that many readers will be suspicious of thisreasoning.16 For example, one might note that if the errors on one list areprobabilistically independent of the errors on the other list, the long-term rateof agreement between the lists should equal the frequency that both entries arecorrect plus the frequency that both entries are wrong. Thus, the long-termfrequency of agreement between the lists = (the reliability of the first list)(thereliability of the second list) + (1—the reliability of the first list)(1—thereliability of the second list). And so, given this assumption, if we assume thatthe two lists’ level of agreement is indicative of their general reliability, wecan calculate the reliability of the second list via the following equation:

.95 = .95(the reliability of the second list) + .05(1—the reliabilityof the second list)

14 Thus, for example, none of the answers on the list are inferred from other answers on it.This simplifying assumption is innocuous, since, while it simplifies the relevant calcula-tions, it does not change the basic result we are interested in here.

15 In effect, as will become clearer, this reasoning involves considering a number of differ-ent hypotheses about what explains the agreement or disagreement between the two lists.

16 Thanks to David J. Barnett and Wlodek Rabinowicz for pushing me to say more aboutthese and related issues.

8 KARL SCHAFER

Which gives us the (surprising) result that reliability of the second listequals 1, and not .905.

If this is right, then we should expect that the second list contains nothingbut correct answers. But surely, given the assumptions we are working with,this is not what we should conclude! So what has gone wrong in this secondcalculation? The problem here is that, by making the assumption of probabi-listic independence between the two lists, we are ignoring some of the evi-dence available to us in this case. In particular, we are ignoring the manner inwhich the high degree of agreement between the two lists provides one with astrong source of evidence in favor of the proposition that the errors in the twolists are not probabilistically independent from one another. In other words, inthinking about this case, it is extremely important to recognize that the degreeto which the lists agree or disagree (together with my knowledge of the firstlist’s reliability) is evidence concerning two questions: (i) how reliable the sec-ond list is, and (ii) how independent the two lists are from one another. Forthis reason, it would be a mistake to think about such cases by first assumingthat the errors in one list are probabilistically independent of the errors in theother list and then proceeding from that starting point.

Upon reflection, it is very natural to think of the agreement betweenthese two lists as providing one with evidence of both these sorts.17 Afterall, when two lists generally agree with one another, this fact should greatlyincrease the probability we assign to possibilities in which their results areexplained in part by common causal factors. And so it should greatlydecrease the probability that we assign to possibilities in which their errorsare probabilistically independent of one another. Thus, on an intuitive level,the agreement between the two lists is evidence in favor of both the propo-sition that the errors of the two lists are not probabilistically independentfrom one another and the proposition that the two lists have similar levelsof reliability—albeit with the second list being slightly less reliable than thefirst (for the reasons laid out above).

17 If this is unclear, consider a case in which the two lists agree with one another 100%over the long term, but the first is known to be 95% reliable. In this case, the only con-clusion to reach is that the errors of the second list are not probabilistically independentof the errors of the second list. For if we assume that the errors of the two lists are inde-pendent, we would be left with the equation:

1 = .95(the reliability of the second list) + .05(1—the reliability of the second list)

Which would generate the absurd result that the reliability of the second list is greaterthan 1. But what is true of this extreme case will also be true in less extreme cases likethe one just described, provided they are not very abnormal. Thus, as is intuitive, thelevel of agreement between the lists (and the reliability of the first) will normally be asource of evidence both for the reliability of the second and for its level of independencefrom the first.


Of course, in reasoning in this fashion, I am using the first list as aguide to what the most likely explanation of the disagreements betweenthe two lists is. But this is exactly as it should be, given the assumptionsin place. And if I reason in this way, then given our simplifyingassumptions, I will be led to the conclusion that the probability that thesecond list is correct about a generic (yes/no) question is equal to .95 *(percentage of known cases in that area in which the lists agree) + .05 *(percentage of known cases in that area in which they disagree), whilealso being lead to the conclusion that very often, when the first list iswrong, the second list is wrong as well.

***

Hopefully the analogy between the case of the two lists and Boris andCarla is clear enough. In particular, in the (admittedly unrealistically)simple example described above, Boris’s only guide to Carla’s reliabilitylies in the level of agreement between them (plus his confidence in hisreliability). Thus, if we again make the simplifying assumption that thequestions at issue between Boris and Carla are independent of oneanother—so that getting one question right or wrong has no effect onthe likelihood that one will get any other question right or wrong, Borisshould calculate Carla’s expected level of reliability in the same fashionoutlined above.18 Thus, if Boris believes that there is .95 chance that heis right about any question in the area, then he should think that theprobability that Carla is correct about a generic (yes/no) question in thatarea is equal to .95 * (percentage of known cases in that area in whichCarla agrees with Boris) + .05 * (percentage of known cases in that areain which Carla does not agrees with Boris).19 Once again, in doing so,Boris will be reaching conclusions about both Carla’s reliability and thelevel of interdependence between their respective verdicts. Again, this isjust as it should be.

18 Again, this assumption is innocuous here because it does not affect the basic point beingmade about Boris’s views about the general relationship between the two lists—although it does, of course, simplify the calculations we are doing.

19 Again, it is important to stress that this is only true because of the extreme simplicity ofthe case under consideration. And, of course, even in this simple case, things are proba-bly more complex than this. For instance, it is plausible that Boris’ probability assign-ment should be less precise than this, to account for Boris’ own uncertainty about hisestimates of reliability. For the sake of simplicity, I ignore this complication here. But itis important for some purposes—e.g. insofar as we want to model, say, the effect theamount of Boris’s past experience should have on his level of confidence in his esti-mates of reliability. The important point is that moving to (say) interval-based credencesto account for this does nothing to affect the main point here, though—for it will remaintrue that Boris’s estimate of his own reliability involves an interval that is centered on ahigher value than his estimate of Carla’s reliability does.

10 KARL SCHAFER

***

Of course, this only follows from Self-Trust and Testimonial Entitlementbecause of a variety of simplifying assumptions we have made about thiscase. For one, we have assumed that the questions at issue between Borisand Carla are independent of one another. But while dropping this assump-tion will complicate the calculations here, it will not change the outcome ofthese calculations in a manner that matters. More relevant is the assumptionthat Boris’s faculties provide him with a basis for regarding Carla as reliableonly insofar as they agree with one another about the issues they are dis-cussing. For of course this assumption grossly simplifies the nature of suchcases under real world conditions. For example, under normal circum-stances, Boris’s faculties will have given him a great deal of backgroundevidence about the general reliability of individuals other than himself.Thus, in any normal case, even given Testimonial Entitlement, Boris willhave a wide variety of Self-Trust-based-grounds for treating Carla as reli-able that extend well beyond the degree to which they agree or disagree intheir conversation.

This certainly complicates such cases, but it does not remove the basicfirst-person bias that the combination of Self-Trust and Testimonial Enti-tlement induces. Rather, as we increase the set of evidence that is rele-vant here, the degree to which this bias will be present may be reduced,but it will not disappear completely. For example, under normal circum-stances, it is true that Boris will have a great deal of evidence in supportof the claim that other normal human beings are roughly as reliable ashe is. But what will the nature of this evidence be? Well, given theabove assumptions, it will be entirely based on the evidence that Boris’sown faculties provide him about the world. So on the most basic level,the structure of this case is fundamentally the same as the structure ofthe simple case we have just been discussing. After all, Boris will onlyhave reason to treat other human beings in general as reliable insofar ashis own faculties provide him with evidence of their reliability. And hisfaculties (either a posteriori or a priori) will provide him with such evi-dence only insofar as they either (i) indicate that other individuals gener-ally agree with their own verdicts or (ii) indicate that Boris’s ownfaculties are less than reliable by, say, delivering conflicting results. Thus,if (as we are assuming) Boris does not have any positive reason to treathimself as unreliable, the degree to which Boris regards others as reliablewill once again be a function of the degree to which these individuals(collectively) agree with the verdicts of his faculties (in general). And hislevel of confidence in the general reliability of others will once again bea (much more complicated, of course) function of the degree of generalagreement between them and Boris. So, for example, if Boris’s faculties


do not provide him with any reason to lower the initial level of self-con-fidence supported by Self-Trust, then, he ought to think that the probabil-ity that a generic human being other than himself is correct about ageneric (yes/no) question is a function of his confidence in his own reli-ability and the degree to which other individuals generally agree with theverdicts of his own faculties in the same general fashion as is true in thesimple cases above.20

Moreover, this continues to be true if we move from a focus on questionsof reliability to other epistemically relevant judgments about others. Forexample, it might be thought that Boris’s general understanding of what aproperly functioning mind or brain is like might provide him with some rea-son to treat Carla as more reliable than he is. But, once again, given theabove assumptions, how should Boris decide what counts as properfunctioning with respect to some domain of facts? Boris might, of course,use his own answers within that domain to determine this—but then afirst-person bias will still be present. Or Boris might appeal to some generalprinciples about what the proper functioning of a mind or brain consistsin—but here too, sooner or later we’ll reach a stage where the relevant biasreappears—since (given our operative assumptions) there will still be asense in Boris’ conception of what a probably functioning human mind islike is rooted in the verdicts of his own faculties. Thus, while these sorts ofconsiderations can push the source of the bias in question back, it remainsthe case that Boris’ picture of reality will be systematically biased in favorof the reliability of his own faculties as compared to the reliability of others.

To be clear, this does not mean that Boris might not come to view cer-tain other individuals as more reliable than himself about certain matters.For of course this is possible. Thus, I do not mean to claim here that Boriscould not come to regard others as epistemically superior to him.21 But suchcases would be cases in which Boris has special reason to treat others asmore reliable than he is. That is, they are cases that involve a clear episte-mic superiority of one party over the other. Given this, the basic pointremains, namely that the basic structure of Boris’s views continues in suchcases to retain a systematic bias in favor of his own faculties—one which

20 Of course, as above, Boris should plausibly have some level of uncertainty about bothhis estimate of his own reliability and his estimates of the reliability of others. And inso-far as he is uncertain about this matter, he should plausibly give equal weight to the pos-sibility that he is underestimating and the possibility that he is overestimating someone’sreliability. But since this applies equally to all of these estimates here, it does nothing toaffect the point I am making above. Rather, even if we model these effects via thinkingof Boris’s credences in terms of intervals centered on the values discussed above, it willremain the case that the interval which is relevant to his estimate of his own reliabilityis centered on a higher value than the interval which is relevant to his estimate of thereliability of others.

21 For further discussion of how this is possible on such views, see below.

12 KARL SCHAFER

makes it the case that he will continue to view himself as more likely to bereliable than others when there is no special reason to treat them as morereliable than himself.22

In this way, expanding our understanding of the Self-Trust-based-evi-dence that is relevant to Boris’s confidence in others does not affect thebasic claim that Boris is rationally entitled to a certain sort of systematicbias in favor of the reliability of his own faculties over the faculties of otherindividuals. But this only follows from our treatment of these cases becausewe have assumed that Boris’s own faculties do not provide him with anyreason to lower his initial level of confidence in his own reliability. So whatif we remove this simplifying assumption?

Well, suppose that when Boris reflects on his past track record, he dis-covers that only 75% of the judgments he makes appear to be correct byhis present lights. And suppose he discovers that this ratio of agreementbetween his current views and his past and future views holds in generalacross time. Given this, Boris should plausibly reduce his evaluation of hisown reliability. But this alone need not change his belief that he is morereliable than other individuals. For suppose that the same is true of otherindividuals’ self-evaluations as well—so that their views are no more stablethan Boris’s over time. Then, given the above assumptions, there will con-tinue to be no reason for Boris to take other people to be more reliable thanhim—and his evaluation of a generic individual’s reliability (about genericyes/no questions) will continue to be a simple function of his confidence inhis own reliability and the degree to which they generally agree.

But now suppose that we introduce an asymmetry into Carla and Boris’sself-evaluation so that Boris is less reliable by his own lights than Carla isby her own lights. For instance, let us suppose that Carla’s views about thearea in question are more stable over time. On its own, this will still giveBoris no reason to take Carla to be more reliable—for stability can be evi-dence of consistent error just as much as it can be evidence of consistentaccuracy. But now suppose that Boris also notices that his views about anissue show a tendency over time to converge with Carla’s (more stable)view about it—so that over time his view about an issue tends to becomemore stable by tending to come into agreement with Carla’s view.23 Thismight well give Boris reason to treat Carla as more reliable than him aboutthe matter in question. After all, given this, at any given time, Carla’s viewswill be a better predictor of Boris’s eventual views than Boris’s own views

22 To be clear, I don’t mean to take a stand on how large such a bias would be under nor-mal conditions. In particular, it seems reasonable that it will decrease as we gain moreevidence from our faculties.

23 Of course, in order for this to be meaningful, the convergence must be due to epistemicfactors.


at that time. Thus, if Boris wants to try to approximate his eventual, morestable views, he should do so via following Carla’s lead, at least to a largedegree. And, of course, the same will be true of Boris’s evaluation of otherindividuals in general, if the same assumptions are true of them.24

Thus, even given Self-Trust and Testimonial Entitlement, it is perfectlypossible for Boris to come to regard other individuals, even in general, asmore reliable than him. For example, as we have just seen, he will do soinsofar as he has evidence that his views will (or would, under ideal cir-cumstances) converge over time to their views in this way. But, once again,if he has evidence of this sort, he will also have evidence that the individu-als in question are not epistemic peers, but rather epistemic superiors.

In the absence of such evidence, on the other hand, Boris will continue(quite rightly, given the views under discussion) to regard himself as morereliable than others.25 And of course others will do the same. Moreover,these cases are cases in which Boris has no evidence that is independent ofhis disagreements with others that supports the proposition that he is moreor less reliable than they are. Thus, the class of cases at issue here are oneof the sorts of cases that the peer disagreement literature means toaddress.26

In particular, given all, it follows from Self-Trust and Testimonial Enti-tlement that:

First-Personal Bias: Provided one has no special reason to treat oneself asless (or more) reliable than other people, one should give somewhat moreweight to one’s own opinions than one does to the opinions of others.

From which it follows that a version of the Rational Symmetry View istrue—namely:

Rational Symmetry View*: Two individuals, who are equally aware ofeach other’s evidence, and neither of whom have any special reason to

24 Note that this assumes that Self-Trust applies, not just the immediate verdicts of my fac-ulties, but also their verdicts in the future. Whether this assumption is defensible is a fur-ther question. For some skepticism about principles of diachronic rationality, see BrianHedden’s “Time-Slice Rationality.” forthcoming in Mind.

25 Of course, there will be a point along this spectrum of evidence at which Boris rationallyought to regard the other party as equally reliable as himself. In this case, he will regardthis person as an “epistemic peer” in the very demanding sense associated with, say,Elga’s work on these issues. For more on this, see below.

26 Depending on how we define an “epistemic peer”, we might not regard these as cases of“full-fledged peer disagreement”. (This is true, for example, with respect to Elga’s defini-tion of an epistemic peer—as I discuss below.) If so, the principle above is best regardedas a principle about when one should treat someone else as an epistemic peer asopposed to a principle about how one should respond to peer disagreement. For morediscussion of this issue, see Objection 1 below.

14 KARL SCHAFER

treat themselves as less (or more) reliable than the other, have a disagree-ment about some issue, both should give more weight to their own opinionthan they do to the other’s.

But perhaps all that this shows is that Testimonial Entitlement is an overlysimplistic picture of the epistemic significance of testimony? For example,just as we have a basic entitlement to trust our own faculties, perhaps wealso have a basic entitlement to trust the views of others to some degree? Ithink some version of this idea is very plausible.27 But consider the follow-ing question: Is the basic level of confidence that we are entitled to have inothers as high as the basic level of confidence we are entitled to have inourselves? If the answer to this question is yes, then the argument I havebeen sketching collapses. But if the answer is no, then while the force ofthis argument is reduced, the basic point remains unaltered. For so long asthere is an asymmetry between the level of trust we are entitled to have inourselves at the beginning of inquiry and the level of trust we are entitledto have in others under such conditions, the basic line of thought sketchedabove will continue to apply.28

Thus, in order to resist this argument for the Rational Symmetry View,one must accept a very strong sort of symmetry between the epistemic sig-nificance of one’s own faculties and the epistemic significance of the facul-ties of others at the beginning of inquiry. Whether or not such a view iscorrect, it involves a very strong and very controversial claim about our epi-stemic relation to other rational subjects. And this gives us good reason totake views like Rational Symmetry View* seriously—much more seriouslythan they generally are.

Objections and Replies

1. Isn’t the Rational Symmetry View* just a variant of what Adam Elgacalls the “Extra Weight View”? If so, isn’t it subject to Elga’s objections to

this view?

It is tempting to associate these views with one another, but in fact theyaddress related, but distinct, topics. Elga’s Extra Weight View claims that“one should give one’s own assessment more weight than the assessmentsof those one counts as epistemic peers”—where you count someone as your

27 Thus, I mean for many anti-reductive views about testimony to fall within the scope ofthis argument—including, for example, the view that we ought to trust others (to somedegree) because it would be irrational to endorse a principle like Self-Trust and give nocredence to the testimony of others.

28 To be clear, I don’t take it to be obvious that any such asymmetry exists. Indeed, I amvery much of two minds about this issue. For some discussion of these issues, see EarlConee’s “Peerage.” Episteme (2009) 6 (3): 313–323.


epistemic peer just in case you count them “as being as good as you atevaluating such claims”.29 Thus, the Extra Weight View (and Elga’s com-peting Equal Weight View) address the question of how you should respondto a disagreement with someone whom one antecedently regards as equallyreliable about the issue in question.30

The Rational Symmetry View* does not directly address this question.Rather, it addresses the question of how one should regard someone whenone has no special reason to treat that person as less (or more) reliable thanoneself and when one knows that they are equally aware all of the relevantevidence. The Rational Symmetry View* (perhaps surprisingly) claims that insuch situations one should treat oneself as somewhat more reliable than theother party. But in saying this, it says nothing about how one should respondto cases in which one does regard the other party as equally reliable.

Given this, the Rational Symmetry View* simply addresses a differenttopic than Elga’s Equal or Extra Weight Views. Using Elga’s (admittedlynarrow) definition of an “epistemic peer”, we can describe this difference asfollows: The Rational Symmetry View* is a view about when we shouldcount someone as an “epistemic peer” in Elga’s sense, while the EqualWeight View is a view about how to respond to situations in which wehave done so. Thus, while the spirit of these two views may appear verydifferent, they are entirely compatible with one another as stated.

That having been said, if the Rational Symmetry View* is true, cases ofgenuine (and rational) “peer disagreement” in Elga’s demanding sense arelikely to be less common than we might imagine. For if the Rational Sym-metry View* is correct, we are entitled all other things being equal to havea systematic bias in favor of our own reliability. And if such an entitlementexists, cases of rational “peer disagreement” in Elga’s sense of the term willonly arise when this default bias is overridden by sufficiently strong evi-dence of equal reliability. Thus, if the Rational Symmetry View* is correct,much of the real action in the epistemology of disagreement lies, not inhow we ought to respond to an epistemic peer, but rather in who we shouldtreat as such. In this way, one might argue that the Rational Symmetry

29“Reflection and Disagreement”, 7 and 11. Elga defines the Equal Weight View as fol-lows: “For according to the equal weight view, your probability that you are right shouldequal your prior probability that you would be right, conditional on what you later learnabout the circumstances of the disagreement.” (19)

30 Compare David Christensen: “Much of the recent discussion has centered on the specialcase in which one forms some opinion on P, then discovers that another person hasformed an opposite opinion, where one has good reason to believe that the other personis one’s (at least approximate) equal in terms of exposure to the evidence, intelligence,freedom from bias, etc. (Such a person is often referred to as one’s ‘epistemic peer’.)”“Disagreement as Evidence: The Epistemology of Controversy.” Philosophy Compass(2009) 4/5: 756–767.

16 KARL SCHAFER

View* addresses a question that is prior to the questions that have domi-nated much of the recent discussion of these issues.

2. Don’t we normally have background evidence that makes it irrational totreat oneself as more likely than others to be right about a given issue?

Does the Rational Symmetry View* conflict with this intuition?

In short, the answer to this question is that the Rational Symmetry View*allows for the existence of many cases of this sort, although it does ofcourse predict that they will be less common than they would be if manyother views were correct. But to consider this question in more detail, let’sfocus on an example of this sort of phenomenon made famous by DavidChristensen:

Mental Math: You and your friend have been going out to dinner togetherregularly for many years. You always tip 20% and split the check (witheach person’s share rounded up to the nearest dollar), and you each do therequisite calculation in your head upon receiving the check. Most of thetime you have agreed, but in the instances when you have not, you havetaken out a calculator to check; over the years, you and your friend havebeen right in these situations equally often. Tonight, you figure out thatyour shares are $43, and become quite confident of this. But then yourfriend announces that she is quite confident that your shares are $45. Nei-ther of you has had more wine or coffee, and you do not feel (nor doesyour friend appear) especially tired or especially perky. How confidentshould you now be that your shares are $43? Many people agree that inthis sort of case, strong conciliation is called for: you should become muchless confident in $43—indeed, you should be about as confident in $45 asin $43.31

Does the Rational Symmetry View* conflict with this reaction? To somedegree, this will depend on the strength of the relevant track record evi-dence in this case. But there is no reason for a supporter of the RationalSymmetry View* to deny that there is some level of such evidence thatwould override one’s default bias in favor of one’s own reliability so as todemand that one regard the other party as (say) an epistemic peer in Elga’ssense. And once this level had been reached, there is no reason for thedefender of the Rational Symmetry View* to disagree with Christensenabout the proper response to this case or other cases like it.

In effect, in doing so, she would simply be agreeing with Christensen’ssuggestion that, “One might hold that conciliation is required only to theextent that the dispute-neutral assessment provides one with strong positive

31“Disagreement as Evidence: The Epistemology of Controversy,” 757. Compare Christen-sen’s “Epistemology of Disagreement: The Good News.” Philosophical Review (2007)116: 187–217.


reason for thinking the other person likely to have gotten things right”.32

Once again, the Rational Symmetry View* does not conflict with such casesbecause it does not address cases in which one has strong positive reasonsto regard the other party as equally reliable as oneself.

3. Perhaps this response is appropriate in some cases, but consider thefollowing sort of case:

Suppose that Boris’s cognitive faculty F tells Boris that P, whereas Carla’scognitive faculty F tells Carla that not P. And suppose that Boris knowsthis. Moreover, suppose that Boris regards his own faculty F as being 70%reliable about questions like whether P. Then given the argument above,all things being equal, Boris should regard Carla’s faculty F as being lessthan 70% reliable about whether P. And so he should give his opinionabout P more weight than he gives Carla’s. But now suppose that Borishas additional information, in the form of several other sources that assessthe reliability of Carla’s cognitive faculty F. These sources are independentof one another, and they all agree that Carla’s cognitive faculty F deservesto be treated by Carla as 95% reliable about whether P. As above, supposethat Boris is entitled to treat each of these sources of evidence as some-what less than 70% reliable about this question—e.g. suppose he is entitledto treat each of them as 65% reliable. Even so, they seem to give Borisstrong reason to accept that it is in fact rational for Carla to regard her fac-ulty F as very reliable. But if so, doesn’t this also give Boris compellinggrounds to treat Carla’s faculty F as more reliable than 70%?

This case may be further broken down into two subcases. The crucial questionis whether the further sources of evidence that Boris has access to provide himwith evidence about both (i) the degree to which Carla rationally ought toregard herself as reliable and (ii) the degree to which she is in fact reliable—orwhether they simply provide him with evidence about the first of these ques-tions without providing him with evidence about the second. In the most nor-mal cases, these sources will provide Boris with both of these sorts ofevidence. But (at least given Rational Symmetry View*) we cannot simplyassume that this is the case—since (given this principle) there is no easy infer-ence from the claim that Carla should treat herself as reliable to such-and-sucha degree to the claim that Boris should treat her as reliable to the same degree.

In any case, if the sources of evidence in question do provide Boris withboth of these sorts of evidence (either directly or indirectly) then there is noconflict between Rational Symmetry View* and the intuitive verdict that Borisshould react to his total evidence by coming to regard Carla’s faculty F asmore reliable than his own. For in this case, Boris does have compelling

32“Disagreement as Evidence: The Epistemology of Controversy,” 758.

18 KARL SCHAFER

special reasons to treat Carla as more reliable and so Rational SymmetryView* does not apply. On the other hand, if these sources of evidence onlyspeak to what Carla would be rational in believing about her own facultieswithout in any way speaking to the question of whether she is in fact so reli-able, then there is no reason to think that they give Boris any reason to shifthis evaluation of her reliability—since, by stipulation, the evidence in ques-tion simply does not speak to this question. Thus, I don’t think there is any-thing in such cases that conflicts with Rational Symmetry View* as stated.

4. How does the Rational Symmetry View* interact with the “independenceprinciples” that have been the focus of much of the recent debate about

these issues?

“Conciliatory” views about disagreement like Elga’s Equal Weight Vieware often motivated via an appeal to principles like the following:

Independence: In evaluating the epistemic credentials of another person’sbelief about P, to determine how (if at all) to modify one’s own beliefabout P, one should do so in a way that is independent of the reasoningbehind one’s own initial belief about P.33

I am somewhat skeptical of such principles when stated in full generality. Butonce again, there is nothing in the Rational Symmetry View*, as stated, thatconflicts with them. For the Rational Symmetry View* can be understood to beconcerned about what one should think about one’s relative reliability prior toand independently of arriving at any particular view about the issue in dispute.That is, it can be made more precise in something like the following manner:

Rational Symmetry View**: When two individuals, who are equallyaware of each other’s evidence, and neither of whom have any special rea-son to treat themselves as less (or more) reliable than the other, have a dis-agreement about some issue, then prior to and independently of arriving atany particular view about the issue in dispute, both should treat themselvesas more likely to be reliable about that issue than the other party is. Thus,once they have discovered their disagreement, both should give moreweight to their own opinion than they do to the other’s.

At least when interpreted in this manner, the bias towards the self that theRational Symmetry View* endorses is entirely compatible with the claim thatone should respond to the discovery of disagreement in the manner Indepen-dence describes. Thus, the Rational Symmetry View* is in fact compatiblewith Independence, and we need not choose between these two principles.

33“Disagreement as Evidence: The Epistemology of Controversy,” 760. Compare Christen-sen’s “Epistemology of Disagreement: The Good News” and Elga’s “Reflection and Dis-agreement”.


5. Doesn’t the Rational Symmetry View* license unacceptable forms ofbootstrapping?

The answer to this question will, of course, depend a great deal on whatsorts of bootstrapping one judges to be “unacceptable”. But I think there isa good case to be made that the answer to this question is no. One way tosee this is to note that the Rational Symmetry View* is wholly compatiblewith a “rationalist” understanding of Self-Trust on which our entitlement totrust our faculties is the product of an a priori entitlement to believe thatthese faculties are reliable.34 If we adopt such a reading of Self-Trust andtake our prior confidence in the reliability of some faculty or process of rea-soning to constrain the level of credence that can be the rational output ofthis faculty or process, it will be free of the forms of bootstrapping thathave seemed most obviously problematic to philosophers working on theseissues. In particular, if this is the case, whatever evidence my faculties pro-vide me about my own relative reliability will simply be the product of mya priori entitlement to believe that I am somewhat more reliable thanaverage. Of course, the very existence of such an entitlement will seemproblematic to many, but the problems with it are not primarily related tothe manner in which it licenses unacceptable forms of bootstrapping.35

Once again, it is important to stress the differences between this viewand Elga’s Extra Weight View. Views of the latter sort may face seriousbootstrapping concerns, but the same concerns about bootstrapping do notrise with respect to the Rational Symmetry View*.36

34 For the relevant sort of “rationalism” see, e.g., Roger White “Problems for Dogmatism.”Philosophical Studies (2005) 131(3): 525–557.

35 Of course, these issues are complex, and there is no space to enter into these debates indetail here. For some of the relevant complexities see (e.g.) Jonathan Weisberg’s “Boot-strapping in General.” Philosophy and Phenomenological Research (2010), 81(3): 525–48.

36 In particular, there is no need for the defender of the Rational Symmetry View* to dis-agree with any of what Elga says here (although he may disagree with Elga about howcommon such situations are):

You and a friend are to judge the same contest, a race between Horse A and HorseB. Initially, you think that your friend is as good as you at judging such races. Inother words, you think that in case of disagreement about the race, the two of youare equally likely to be mistaken. The race is run, and the two of you form indepen-dent judgments. As it happens, you become confident that Horse A won, and yourfriend becomes equally confident that Horse B won.When you learn of your friend’s opposing judgment, you should think that the twoof you are equally likely to be correct. For suppose not—suppose it were reasonablefor you to be, say, 70% confident that you are correct. Then you would have gottensome evidence that you are a better judge than your friend, since you would havegotten some evidence that you judged this race correctly, while she misjudged it.But that is absurd. It is absurd that in this situation you get any evidence that youare a better judge. (“Reflection and Disagreement”, 13)

20 KARL SCHAFER

6. If the Rational Symmetry View* doesn’t actually conflict with views likethe Equal Weight View, why should we consider it relevant tocontemporary debates about the epistemology of disagreement?

The temptation to dismiss the Rational Symmetry View* on these groundsis common, but mistaken. For we should not assume that these debates havefocused on the only issues surrounding the epistemology of disagreementthat have genuine philosophical interest. Indeed, if the Rational SymmetryView* is correct, these debates have been missing where much of the realphilosophical action lies. For if something like Rational Symmetry View* istrue, then we can often rationally treat those around us as less reliable thanwe are, even though we can give no special reason for doing so. And, ifthis is true, then genuine cases of “epistemic peerhood” in the demandingsense that authors like Elga have focused on will generally be less commonthan we might have assumed. In which case, the question of how we shouldreact to genuine “peer disagreement” in this demanding sense will often bebeside the point with respect to the question of how we ought to respond tothe real world disagreements that provide the literature on the epistemologyof disagreement with much of its interest.

7. Doesn’t the argument above about Boris rely on the fact that Boris hasresponded correctly to his evidence? Would the same argument go through

if he had not responded to his evidence correctly? If not, doesn’t thatundermine the Rational Symmetry View*?

The argument above does assume that Boris is responding correctly to hisevidence. But it is also wholly compatible with the claim that the same istrue of Carla as well. Indeed, the point of the argument is precisely that it ispossible for both parties to a disagreement to do just this and yet continueto disagree.37 So there is no reason to think that this will undermine thisargument.

Conclusion

If the arguments above are correct, then a version of the Rational SymmetryView follows from the conjunction of Self-Trust and Testimonial Entitle-ment. Indeed, if these arguments are correct, the Rational Symmetry Viewfollows from something considerably weaker than this. As we have seen,this is best regarded, not as a claim about how we should respond to “peerdisagreement” in the narrow sense common to much of the contemporaryliterature. Rather, it is best regarded as a claim about who one should treat

37 Note that this does not commit me to “permissivism” about rationality—it simply com-mits me to the claim that the rational response to some set of information can vary fromagent to agent. But agent-relativity is not the same as permissivism—for relative to anyparticular agent, the requirements of rationality may be wholly non-permissive.


as a “peer” in this demanding sense of the term. Thus, if these argumentsare sound, “peer disagreement” in the sense that has dominated much of therecent discussion of these issues is much rarer than it might appear. If so,much of the action in the epistemology of disagreement relates, not to howone should treat those one regards as “epistemic peers” in the sense popularin the literature, but rather to whom one should treat as such.38

38 For very helpful comments on this paper I would like to thank Yuval Avnur, David Bar-nett, Sinan Dogramaci, Matt Kotzen, Steve Makin, Berislav Marusic, John Morrison,Wlodek Rabinowicz, Kieran Setiya, Robert Steel, Jack Woods, and several anonymousreferees.

22 KARL SCHAFER

How Common is Peer Disagreement? On Self-Trust and Rational Symmetry

Documents