Tom Kelly, "The epistemic significance of disgreement," OSE 1 (2005)

Kelly defines peers as being your equals in respect of general intellectual virtues (as does Gutting 1982), but also in respect of their "exposure to evidence and arguments which bear on the question at issue." In section 2.3, he also stresses that the agents have scrutinized the evidence equally carefully. Elga 2007 defines peerhood differently, as meaning that "conditionally on the two of you disagreeing about the claim, the two of you are equally likely to be mistaken."

Like Feldman (esp. 2006), Kelly in this paper only considers cases where the one peer believes p and the other disbelieves (instead of cases where the other peer merely withholds).

At the end of section 3, Kelly sketches an argument for the Fully-Conciliate/Equal Weight View: When I know that you are better-informed than me, I should defer to your judgment. When I acquire the evidence you had but I lacked, shouldn't I then still give your judgment about what our (now shared) evidence supports "considerable weight"? Indeed, aren't our situations symmetrical, so that your judgment deserves equal weight? What reason would I have to privilege my own judgment?

Kelly responds to this argument in section 4. His response initially seems to be one available to both of the peers:

Suppose that, as it turns out, you and I disagree. From my perspective, of course, this means that you have misjudged the probative force of the evidence. The question then is this: why shouldn't I take this difference between us as a relevant difference, one which effectively breaks the otherwise perfect symmetry?... From my vantage point --- as one of the parties within the dispute, as opposed to some on-looking third party --- it is just this undeniably relevant difference [namely, how well each has evaluated the evidence] that divides us on this particular occasion.

But a bit later it emerges that he's thinking that the peer who orginally assessed the evidence correctly is in a stronger position. So the view he advocates is a form of what Elga calls the "Right Reasons" View: it's only the subject who originally assessed the evidence correctly who is entitled to retain their judgment in the face of disclosed peer disagreement.

Earlier in the paper Kelly announced his conclusion as:

Once I have thoroughly scrutinized the available evidence and arguments that bear on some question, the mere fact that an epistemic peer strongly disagrees with me about how that question should be answered does not itself tend to undermine the rationality of my continuing to believe as I do. Even if I confidently retain my original view in the face of such disagreement, my doing so need not constitute a failure of rationality. Indeed, confidently retaining my original belief might very well be the uniquely reasonable response in such circumstances.

The underlined qualifications are contingent upon me being the one who initially assessed the evidence correctly. The phrase "does not itself tend to undermine" could be interpreted in a stronger or a weaker sense. In the stronger sense, it means that your FO justification is not diminished at all. In the weaker sense, it means that even if your FO justification is diminished somewhat, it needn't always be to a degree that mandates agnosticism. This paper primarily defends the weaker thesis, but also expresses sympathy and some motivation for the stronger thesis. In later papers, Kelly abandons the stronger thesis, and only defends the weaker thesis.

Section 5 defends the idea that what is fundamentally providing a threat in cases of disagreement is not the existence of actual disagreers, but rather how good the reasons of actual (or possible) disagreers are (or would be). Kelly posits that the non-existence of actual skeptics is not usually taken in itself to be an objection to skepticism. (I don't think this is as clear as he does.) Kelly endorses this. He says:

The case for skepticism stands or falls with the probative force of skeptical arguments and does not depend on contingent and empirical facts concerning the actual existence or nonexistence of skeptics.

Section 6 asks whether the responses of actual peers might be higher-order evidence, which should have some defeating first-order effect, even in cases where it's false/misleading about the higher-order question. Kelly articulates an argument for not treating the HOE as having a FO effect: I shouldn't treat my peer's judgment as a new piece of evidence about the FO issue, because my peer won't (reasonably) count her own judgment that way. At one point in this argument, Kelly observes that the peer won't count her judgment as a reason to increase her confidence about the FO question; which is certainly true but not something Kelly's opponent needs to commit to. So I think this is a red herring. It's also true, and perhaps all that Kelly needs here, that the peer won't be entitled to count her own judgment as additional evidence relevant to the FO question (regardless of whether it'd be evidence that entitles her to increase her confidence). It seems that if I or the peer were so to count her judgment, when they have access to the evidence on which the judgment is based, we'd be guilty of a kind of "double-counting." Shouldn't we instead let the peer's actual evidence "screen off" the judgment she bases on that evidence?

Kelly doesn't wholeheartedly endorse this argument; at least, he doesn't count it as decisive. (And as I said, in his later papers, he abandons it.) Instead, he argues that even if the HOE provided by the peer's judgment has FO bearing, "it doesn't follow that skepticism or agnosticism is the reasonable response." His reason is that, after disclosure of the disagreement, we still have our original evidence, and it won't always be swamped by the HOE. Kelly says:

Our original evidence E does not simply vanish or become irrelevant once we learn what the other person believes on the basis of that evidence: rather, it continues to play a role as an important subset of the new total evidence E'.

In paradigm cases, the HOE may be equivocal, because my own judgment and my peer's "cancel each other out"; but see the discussion of section 4 in the next paper.

See also Weatherson, which is very close in spirit to this paper of Kelly's. Weatherson also argues that the FOE shouldn't be screened off by the HO evidence or judgment (see also Lasonen-Aarnio 2013, end of §3), and like Kelly in this paper, Weatherson resists the idea that the HO judgment would have any undermining effect on the FO belief. Weatherson's reasons for thinking the latter are: (i) without the idea that the HO judgment screens off the FOE, he doubts that FOE that entails P can be undermined; and (ii) he rejects the view I called "credulism" earlier in the term, and so in particular, thinks that immediately justified inferences can't be undermined by any evidence. I reject both (i) and (ii).

Section 7 discusses epistemic roles that Kelly thinks actual disagreement can play:

Tom Kelly, "Peer disagreement and higher-order evidence," Kelly completed this paper in 2007, and it's sometimes cited with that year; but it appeared in Feldman and Warfield (2010), and in Goldman and Whitcomb (2011)

In this paper, Kelly works with more fine-grained disagreements, including subjects who have different degrees of confidence in P. He also considers cases where the number of peers on each side is not balanced.

In this paper, Kelly interprets "splitting the difference" between different views as a matter of averaging the subjects' credences; in his 2013 he acknowledges that this can't be a generally suitable way to understand that informal notion. In class, I gestured at a couple of reasons people have articulated for thinking so. I also proposed a loose heuristic to use for notions like "splitting the difference" or "fully conciliating." According to my proposal, A and B "ought to fully conciliate" when they ought to arrive at a single shared doxastic response, and which response that is isn't sensitive to which of A and B originally assessed the FOE correctly. (One would also like to have a heuristic for when A and B in fact "fully conciliate," regardless of whether they ought to. But I don't know how to translate the second clause of my proposal into a form suitable for explicating this notion.)

Section 1 describes some alternatives to the Fully-Conciliate/Equal Weight View (EWV).

One dimension of variation is how far the subjects ought to conciliate: not at all? or only partially? A second dimension of variation is whether the view is more symmetrical and subjectivist, or more non-symmetric/objectivist/Right-Reasonsy.

The view Elga calls the "Extra Weight" View (I won't ever abbreviate this, EWV will instead always be used as above) is symmetric, and doesn't require fully conciliating: each peer is entitled to privilege her own judgment to some degree. Kelly's "Total Evidence" View (not so helpfully named; it roughly coincides with what Elga calls the "Right Reasons" View) is more objectivist: at most one of the subjects is being reasonable, the one whose original opinion better tracked the FOE.

Kelly also articulates a symmetrical version of the No Weight View (NWV). He expresses some sympathy for it in section 2. If I correctly accept Permissivism (as do Rosen, Schoenfield, most Bayesians), and my peer's judgment is within the constraints of what (also) counts as reasonable, I can reasonably remain steadfast.

Kelly says that defenders of the EWV will/should reject Permissivism, and he will go along with that for the same of argument. With Permissivism off the table, Kelly no longer finds the symmetric competitors to EWV plausible. He doesn't want to go for any Extra Weight View.

Section 3 argues against the EWV.

Case 4 (which Christensen calls Right/Wrong): You rightly have low confidence in H on FOE E; I wrongly have high. Before disclosure, it's uncontroversial that Right should retain her confidence and Wrong should reduce his. EWV says that asymmetry "completely washes out" once the disagreement is disclosed. How peers should respond supervenes on the purely psychological evidence about how they each assessed the FOE; the actual force of the FOE is extraneous.

As in 2005, Kelly instead argues that the objective upshot of the FOE should matter. In this paper, he is prepared to allow that Right should be somewhat less confident after disclosure (see section 4, discussed below); but he denies that Right and Wrong are always required to make "equally extensive revisions" in their opinions.

Some other cases, discussed at various places in the paper:

Four additional arguments against the EWV:

  1. (section 3.1): in intrapersonal conflict (two of my beliefs are inconsistent), it can sometimes be reasonable to drop one of the beliefs and retain the other, which was in fact better supported by your evidence. This makes it more plausible that the interpersonal case can have a similar structure.

  2. (section 3.2, from Bronfman): Case 5 (which Christensen calls Wrong/Wronger). EWV seems to say these two unreasonable judges can bootstrap into a reasonable intermediate view just by comparing notes. That seems incorrect.

    Kelly considers the possibility that an EWV will only say peers are "rationally required" to split the difference in their opinions, where this is merely necessary and not sufficient for being reasonable. (For Worsnip, it's neither necessary nor sufficient.) Kelly's objection to that view is that in the Right/Wrong case, it would say that the subjects differ in their reasonableness even after fully conciliating, despite responding to the same total evidence with a single shared state of confidence. (Christensen construes their evidence as de se, and so denies that their evidence is the same.)

    Fn 17 discusses my suggestion that EWV might only be addressing the question of how Right should revise, while being silent about Wrong/Wronger.

  3. (section 3.3): the EWV seems to legitimate "bootstrapping in the single person case." After you've formed a judgment, you shouldn't be able to screen off and ignore the actual FOE. (Compare the "double-counting" argument in Kelly 2005.)

    The EWV doesn't allow mere facts about who's made a performance error to make a difference to whose view deserves more weight. Only independently available evidence about subjects' having made a performance error is admissibly relevant. Kelly argues that this makes it:

    arbitrary and unmotivated to continue to maintain that actual performance makes a significant difference in single person cases... Rather, on the suggested picture, if I am generally competent in the way that I respond to evidence (and I know that I am), then this should be enough to guarantee that I am reasonable in responding to my evidence in whatever way that I do. But this contradicts our initial assumption, viz. that one way of ending up with an unreasonable belief is to respond incorrectly to one’s evidence, despite possessing the ability to respond to that evidence correctly.
  4. (section 3.4): Litmus paper. You should regard someone's judgments as evidence relevant to P to the extent that you take those judgments to be a reliable indication of P. This is no different from the way in which you treat litmus paper to have evidential force. Why then in paradigm cases of peer disagreement should the HO/psychological evidence about you and your peer's judgments overwhelm the FO evidence "given that the way in which the two kinds of evidence qualify as [bearing on the FO question] is exactly the same?"

Section 4 presents Kelly's positive view, the "Total Evidence" View (a version of what Elga calls a "Right Reasons" View). The subject who initially assessed the FO evidence correctly is not required to split the difference upon learning that her peer disagrees. This is as in the 2005 paper, though Kelly is now prepared to think the subject ought to be somewhat less confident. (And correlatively, he's now prepared to think that actual disagreements have a kind of significance that merely possibly disagreements lack. It's important though that peers' judgments are relevant only to the extent that they are independent --- of each other, or of their teachers, whose views you have already taken into account.)

If Uniqueness, we have to reject the Symmetrical NWV. Should we accept the Asymmetrical NWV, which says that at least the subject who initially responded correctly would reasonably give no weight to their peer's judgment?

No, Kelly now thinks typically you should conciliate to some degree. He gives a case of a mathematician in fact proving a Conjecture, but the entire mathematical community (wrongly) rejects his proof as unsound. Eventually, Kelly thinks, the subject ought to become agnostic, and plausibly this means when confronting even a single peer, there is some defeating effect. (Kelly's example of weather forecasting could also be used to make this point, without the distraction of having the FOE entail the conclusion it supports.)

Kelly cautions that we shouldn't assume that a single pair of competing judgments straightforwardly "cancel each other out": they make a greater proportion of our total evidence now support agnosticism. When the number of peers is large enough, as in the mathematician example, the HOE will swamp the FOE, and agnosticism will be conclusively called for.

Kelly's view also allows cases where it's the FOE that overwhelms the HOE, as in Christensen's and Elga's cases where one's apparent peer arrives at an absurd result.

The EWV (and Symmetric NWV) says that what it's reasonable to believe in cases of peer disagremeent supervenes on the distribution of peer opinion: the psychological evidence about the peers' judgments trumps/screens off the FOE. (As mentioned before, Weatherson and Lasonen-Aarnio 2013 §3 frame the debate similarly.) The Asymmetric NWV says that it supervenes on the FOE: that screens off the psychological/HOE. Kelly rejects both of these supervenience claims.

An interesting case:

If I hold some belief on the basis of fallacious reasoning, then it will typically not be reasonable for me to hold that belief. However, in the unlikely but possible situation in which a large number of generally reliable peers mistakenly arrives at the same conclusion by independently committing the same fallacy, it will typically be reasonable for them to believe that conclusion upon comparing notes, even if there is no legitimate first-order reasoning by which they could have arrived at the conclusion.

Section 5.2: inanimate measuring devices like thermometers

Of course the fact that a thermometer is mine is in itself no reason to trust it better, and so too with my initial assessment of some evidence. Kelly is not defending an Extra Weight View.

Objection: the mere facts about which thermometer is functioning properly don't (on most views) by themselves have evidential relevance. So why should the fact that one subject has in fact evaluated the FOE correctly?

Kelly's reply: cases of peer disagreement are more like you've opened up the two thermometers and seen how their verdicts are produced, so "you have witnessed the malfunction occur."

Worry: isn't this scenario one that you "can't discriminate" from the "symmetrical" case where it's you who have malfunctioned, and your peer who gave the initially reasonable verdict? You should only favor an initial verdict that you have independent reason to think is more likely to be correct. (Compare Christensen's and Elga's Independence Principles.)

Kelly imagines two replies to this worry: (1) one might argue that it falsely assumes that justification for P requires justification for believing one has justification for P. He doesn't develop that reply, but rather: (2) the subjects aren't in symmetrical situations wrt their HOE, because the subject who has assessed the FOE correctly will usually have more reason to think they have done so:

[I]n order to preserve the higher level normative symmetry which lends The Equal Weight View its plausibility, it seems that the proponent of The Equal Weight View will have to maintain the following: when we met, the justification which I possessed for [incorrectly] thinking that my original response to the evidence is reasonable was just as strong as the justification which you possessed for [correctly] thinking that your original response to the evidence is reasonable. For if you were better justified in thinking that your response was reasonable than I was in thinking that my response was reasonable, then this would break the putative higher level symmetry and provide a basis for favoring your original belief over mine.

Kelly finds dubious this claim that the incorrect HO belief will always be at least as justified as the correct one.

Section 5.3: Kelly acknowledges both "downward epistemic push" --- the HOE can have FO bearing --- and "upward epistemic push": when E is genuinely good evidence for H, you will also have good evidence that this is so. (Kelly says, more specifically, that "this very fact" that E is good evidence for H contributes to your justification for believing it; but this seems controversial and more than is needed.)

An argument for Kelly's claim that your justification for a HO belief needn't always come from some independent evidence, beyond the FOE itself:

Let E represent all of the evidence that you currently possess. Surely you can recognize that E supports the belief that the sun is larger than the moon. But this recognition is not based on some independent evidence that you possess (i.e., evidence not included in E), since ex hypothesi, E exhausts what evidence you have.

Lasonen-Aarnio 2013 also argues for the existence of "upward epistemic pushes" --- she calls positive such pushes "evidential attenuation" (because they weaken the force of disagreement-based defeaters) and she also discusses negative such pushes, which she calls "evidential amplification." (Her paper is very good, recommended reading.)

Section 5.4: Elga argues that competitors to the EWV would legitimate an intuitively objectionable kind of bootstrapping, where you could become confident that you have a better track record than your apparent peer, just by virtue of your competing judgments (especially over a long run).

Kelly admits that Elga's Extra Weight View (which he rejects) is vulnerable to this complaint, but his own view denies that you should be confident that your friend is wrong every time you disagree. You should only have that attitude in the cases where you were in fact the one who initially assessed the evidence correctly. There is no "general answer" to the question of how confident you should be that it's your friend who made the mistake, that "floats entirely free" of the FOE and whose verdict it in fact supported. (Lasonen-Aarnio 2013 makes similar claims about "blanket views of disagreement"; see also Kelly 2013's claim that "dogmatism is not a formal vice.")

Kelly's Cases 8 and 9 (the Philosophy Seminar) are ones where you do consistently assess the FOE correctly, and here Kelly argues it is more reasonable to demote your apparent peer. Elga has difficulty accommodating that intuition.

Kelly's footnote 39 argues that Elga's overall position seems to commit him to an objectionable non-commutativity.

Section 6: Kelly's view might be resisted on account of its being "externalist" in some sense:

[T]here is no magic red light that illuminates when one responds to the evidence correctly, no warning bell which sounds when one does not. Indeed, as a phenomenological matter, there might be no introspectible difference between how things seem when one is responding correctly and how things seem when one is not.

But Kelly observes that he hasn't committed to "externalism" in other, more familiar senses. What it's reasonable for you to believe doesn't depend on anything more than what total evidence you possess/are aware of.

Kelly's view might also be resisted on account of a desire to identify "good evidence" with "potentially persuasive evidence." Kelly argues (as have both Williamson and I) against conflating these notions.

Tom Kelly, "Disagreement and the burdens of judgment," in Christensen and Lackey (2013)

This paper repeats and further develops a number of points already mentioned above.