Higher-Order Defeat and Withholding Judgment

  • Earl Conee (University of Rochester)


Defeat by higher-order evidence needs defending. Maria Lasonen-Aarnio argues powerfully for the conclusion that higher-order evidence does not have an unlimited capacity to defeat justification. While developing her main argument Lasonen-Aarnio poses two other problems for unlimited higher-order defeat. Some theories of higher-order defeat are not subject to the main argument. The other problems threaten those theories too. The problems will be developed with the aim of finding and evaluating optimal versions. The first offers a reductio argument. The argument will be criticized as employing an unjustifiable assumption. The second problem poses a dilemma. The dilemma will be avoided by supplementing plausible higher-order defeat theories with a view of what justifies withholding judgment.

How to Cite:

Conee, E., (2021) “Higher-Order Defeat and Withholding Judgment”, Ergo an Open Access Journal of Philosophy 7. doi:



Published on
22 Oct 2021
Peer Reviewed

1.1. Background

Maria Lasonen-Aarnio argues against unlimited higher-order defeat.1 Her main argument can be sketched as follows.2 Suppose for reductio that some undefeated higher-order negative evidence has a defeating effect on evidence that justifies attitudes, no matter how high in order the negative evidence might be. If so, then rules for justifying attitudes by evidence would have to have unlimited complexity. Such rules could not guide us toward forming justified attitudes. Yet such guidance is an essential function of epistemic rules. So undefeated negative higher-order evidence does not always defeat lower-order justification.3

This is a forceful line of reasoning against its target theories. Here is how Lasonen-Aarnio characterizes the theories.

In what follows, I will make an assumption about what the epistemic good-making properties of doxastic states are. It has its roots in what I will call a rule-driven picture of epistemic cognition. On such a picture, doxastic responses at least typically involve the application of epistemic rules, and whether or not a doxastic state is epistemically rational or justified depends on the goodness of the rules that were applied in forming or maintaining that state. (2014: 319)

Rule-driven theories as Lasonen-Aarnio describes them are vulnerable to her main objection. If justification derives from rule following, then the justifying rules have to be followable in order for justification to be available. Lasonen-Aarnio gives excellent reason to think that rules that take into account higher-order negative evidence of every height are too complex to be followable. As a result, it is highly doubtful that unrestricted higher-order defeat can be accommodated by followable rules.

Not all theories of justification are “rule-driven,” in the sense that they state rules aimed to give guidance. Some theories of justification are aimed at explaining its nature by specifying the conditions that constitute having a justified -doxastic attitude. In other words, these “constitutional accounts,” as we can call them, are aimed at telling us what makes attitudes justified. A constitutional account need not guide us to justification any more than a theory of what constitutes knowledge must guide us to knowledge. Thus, constitutional accounts are not threatened by Lasonen-Aarnio’s main line of argument against unlimited higher-order defeat.

Nevertheless Lasonen-Aarnio’s paper contains serious threats to constitutional accounts. The paper raises two concerns that challenge any theory that counts all higher-order negative evidence as potentially defeating justification.4 Lasonen-Aarnio does not fully develop these challenges. They are inessential to her main argument. But either could be a conclusive objection to the most plausible constitutional accounts that allow unlimited higher-order defeat. The challenges will be investigated here. The goal is to find their strongest versions and to assess them.

1.2. The Defended Theories of Higher-order Defeat

The focus here will be on a class of plausible constitutional accounts of justification that allow unlimited higher-order defeat. We can call the selected class Downward Defeat (DD) theories. It is hereby stipulated that DD theories are those that meet the following conditions. They are about the justification of the doxastic attitudes of belief, disbelief, and withholding judgment.5 Their distinctive features are manifested in cases in which S has evidence in support of p and S has either (1) reason to think that S lacks justification for believing p , or (2) reason to think that S is incompetent at rationally assessing S’s evidence regarding p. DD theories imply that having either (1) or (2) as evidence is having a potential defeater of S’s justification for believing pp 6 That is, having evidence of sort (1) or (2) renders belief in p less justified or unjustified, unless the justification-reducing effect is neutralized by other evidence that counters it. Additionally, DD theories imply that this defeat has a specific effect on justification. DD theories imply that when evidence that would justify belief in p is wholly defeated by (1) or (2) evidence, the justified attitude toward p is withholding judgment (unless the defeat is accompanied by new evidential support for p or its negation).

Here is an intuitive defense of DD theories. Epistemically justified belief is a sort of reasonable belief. It is the sort that is particularly concerned with the proposition’s truth. Belief by S in p is reasonable in this way only if its justification takes into account all of the information that S has about S’s access to p’s truth. The higher-order evidence cited in DD theories is about S’s access to p’s truth. The evidence indicates either (1) that S does not have justification for regarding p as true or (2) that S is likely to be wrong about what attitude toward p S’s evidence justifies. Again, we are assuming that S has evidence that would support p’s truth to S and thereby justify believing p in the absence of the higher-order negative evidence. When S also has sufficiently powerful negative higher-order evidence of type (1) or type (2), from S’s point of view S’s access to p’s truth is too unreliably related to the fact of the matter for believing p to be reasonable. Cases of type (1) occur when it is credibly denied to S that some evidence that prima facie supports p does give that support. With a sufficiently credible denial, the prima facie evidence for p does not justify believing p. This defeat is like other familiar cases of undermining. It is like the way in which evidence of misleading lighting conditions defeats otherwise supportive color experiences as justification for color judgments. Cases of type (2) occur when S has evidence for S’s being incompetent at assessing evidence that seems to S to support p. Sufficiently credible evidence of this incompetence renders it seriously doubtful to S that seeming support for p is genuine. It ceases to be reasonable for S to believe what the evidence seems to support. When S has good enough evidence of type (1) or type (2), S’s reasons to believe p no longer support p’s truth from S’s perspective. Evidence of type (1) or (2) gives S nothing that favors p’s falsehood. In light of all of the information that S has to go on, it is most reasonable for S to take a noncommittal attitude toward p’s truth. S is justified in withholding judgment on p.

These justification verdicts are what DD theories imply. The plausibility of this view makes such theories worth defending.

2.1. Preliminaries for the First Problem

Lasonen-Aarnio poses an argument against the possibility of an “Über-rule”. An Über-rule is “a function from epistemic circumstances to whatever the correct epistemic response is (or whatever the permitted doxastic responses are) in those circumstances” (2014: 330). There are two further features of Über-rules. First, an Über-rule is complete: “for any circumstances in which there is some epistemically rational doxastic state in the first place, the Über-rule codifies what that state (or range of states) is” (2014: 331). Second, “the following kind of situation can never arise: a subject does exactly as the rule recommends, but she has evidence that the resulting doxastic state is flawed” (2014: 331).

Lasonen-Aarnio’s argument against the possibility of an Über-rule is explicitly given against theories that are committed to this thesis:

Higher-order defeat Evidence that a cognitive process producing a -doxastic state S as output is flawed has defeating force with respect to S. (2014: 316)

Higher-order defeat says that the higher-order evidence “has defeating force” with respect to a doxastic state. Presumably this means that the evidence is a typical sort of defeater of justification for the state—the higher-order evidence reduces or nullifies the justification that the person has for being in the state, unless the higher-order evidence is itself defeated.

Not all DD theories are committed to Higher-order defeat. Unlike Higher--order defeat, the stipulated characterization of DD theories says nothing about the defeasibility of justification for withholding judgment. DD theories do share implications of Higher-order defeat concerning the justification of belief. DD theories imply that sufficiently strong higher-order evidence of the sorts that they cite is always a defeater of justification for belief, unless the potentially defeating higher-order evidence is itself defeated. This is implication concerning defeat for belief is enough for the argument to apply to DD theories.

2.2. The Argument

Here is Lasonen-Aarnio’s presentation of the argument against the existence of an Über-rule:

[A]ssume that you are staring at a chart representing the Über-rule: for each possible epistemic situation (or each relevant type of situation), the chart specifies what the recommendations made by the Über-rule in that situation are. (Let us set aside worries having to do with there being infinitely many such situations.) Now imagine that you hear an epistemology oracle tell you that the recommendations made by the Über- rule in the very situation you are in right now are incorrect. In so far as the rule is complete in the sense specified above, the chart must say something about your current situation. Imagine that, as the chart tells you, the rule recommends being in state S. But in so far as the oracle is to be trusted, doesn’t her testimony act as a higher-order defeater for any such recommendation? (2014: 331)

2.2.1. An Initial Paraphrase of the Argument and an Initial Evaluation

Here is an initial condensed reconstruction of the reasoning:

If any Higher-order defeat theory is true, then this is possible: A chart of the theory’s Über-rule tells you your justified attitude toward a proposition while a trustworthy oracle denies to you what the chart says. By Higher-order defeat theories, the denial defeats what the chart says. So by the theory your justified attitude is not what the Über-rule chart says. Yet the theory implies that the Über-rule is correct. So the theory is untrue.

We will soon seek a more fully articulated version of the argument. First we should note that Lasonen-Aarnio does not take the reasoning to be conclusive. She suggests that instead the Über-rule might be said to be undefined in the epistemic situation that the argument uses to pose the problem. (2014: 331)

DD theories would not be well defended by making that response. It alleges that the Über-rule is undefined in a situation. The situation is specified as a case in which an oracle denies a certain statement. The statement is specified as asserting something that the Über-rule implies to be justified under the circumstances. So the Über-rule cannot be undefined there because it implies an attitude to be justified.

Lasonen-Aarnio also suggests another response to the argument. She suggests that cases that are declared to be undefined by the Über-rule might be identified as follows: the ones in which the oracular testimony would make trouble for Higher-order defeat theories if the oracular testimony were a defeater.

This way of responding to the objection would be gravely problematic too. It denies defeat in some cases of testimony. Yet this would be testimony of the very sort that constitutes defeat that the theories recognize. The undefined status would be alleged just to avoid the objection. Selecting a theory’s implications in this way would be paradigmatically ad hoc.

In any event DD theories do not allow indeterminacy in some oracle cases. DD theories are never undefined when S starts out with a justified belief, S receives defeating higher-order evidence against the evidence favoring belief, and S receives nothing else that is relevant to the proposition’s truth. Those are clear cases of unmitigated DD defeat. About such cases DD theories allow no indeterminacy. The theories imply that the attitude of withholding judgment is justified. They need a better defense against the objection.

Before seeking that defense we should strengthen the argument against DD theories in a couple of ways.

2.2.2. Strengthening the Argument

Infinite charts can be avoided. The objection requires just the possibility of a statement of the particular implication of the Über-rule for the circumstances of the subject who receives the oracular testimony. Thus nothing infinite is required.

The argument can be further enhanced. For some situations and propositions, the oracular testimony that denies the justification of belief is a defeated defeater. The believed proposition is so clearly true that the oracle’s testimony is reasonably disregarded. Or at least, it is plausible that this sort of defeater defeat can happen.7 The challenge to DD theories does not depend on its being -impossible. To sidestep this issue, the reasoning can be limited to cases in which S’s other evidence is not so decisive. We can suppose that prior to S’s receipt of the oracle’s testimony, S has just enough support for p to justify belief. Highly trustworthy testimony seems capable of defeating such minimal justification for belief.

2.2.3. A Final Formulation of the Argument

Here is the argument against DD theories in a final version.

Two modest assumptions: First, if any DD theory is true, then the correct Über-rule exists. Second, if the Über-rule exists, then its implication for any given epistemic situation can be stated. Now we suppose that S has had evidence that was just enough to justify believing p.8 So S’s belief in p is justified until S’s situation changes. S’s situation changes as follows. S sees and understands a statement of what the Über-rule implies to be S’s justified attitude toward p in S’s situation. At the same time S hears from an epistemology oracle who S had excellent reason to trust. The oracle tells S that the statement is untrue.

The possibility of this situation is supposed to follow from uncontroversial facts and the assumed truth of some DD theory. This will be disputed below. First, here is the rest of the argument.

The argument’s aim is to derive that the implied situation is impossible and to infer that no correct Über-rule exists because its existence is the suspect assumption. The reasoning for the impossibility proceeds as follows. First suppose that the statement implied by the Über-rule says that S’s justified attitude is belief in p. The trustworthy oracular testimony denying the statement is defeating evidence against the modest evidence that S has for p. Since the statement says that belief is S’s justified attitude toward p, according to any DD theory belief is not justified for S. So by DD theories the statement is untrue. Yet the statement has been assumed to be an implication of the Über-rule. So the Über-rule is incorrect. Thus, if the statement says that belief is justified, then no DD theory is true because no correct Über-rule exists.

Now we suppose instead that the statement expressing what the Über-rule implies about S’s situation is that withholding judgment on p is S’s justified attitude. The oracle tells S that this statement is incorrect. This testimony denying that withholding is justified does not even appear to alter S’s justified attitude toward p. S remains justified in believing p. By assumption, S had enough evidence supporting p to justify belief until the testimony. An oracular denial that withholding is justified plainly does nothing to defeat that justification for believing. Yet the statement that we are now assuming to be the Über-rule’s implication asserts that the justified attitude is withholding. So the statement is untrue. Thus, if the implied statement says that withholding is justified, then the Über-rule is incorrect and no DD theory is true.

This exhausts the doxastic attitudes that DD theories are about. So there is a possible case in which the Über-rule for DD theories is incorrect about the justified attitude, no matter which one the Über-rule implies to be justified. One of our initial modest assumptions is that there is a correct Über-rule for any true DD theory. Therefore no DD theory is true.

2.3. A Criticism of the Argument

This argument unjustifiably assumes that it is possible for a proposition that is specified as an Über-rule implication also to be a proposition against which S receives the oracular testimony. The argument does not derive this possibility from DD theories together with some plain facts. If it did, then the source of the implied contradiction would be rightly assigned to the DD theories. There is no prospect of that derivation, however. DD theories imply that certain propositions attributing justification are defeated by certain testimony. The theories and the plain facts do not so much as suggest that the justification also survives the receipt of the testimony.

An analogy will highlight the problem for the argument. The objectionable assumption of a joint possibility is plain to see in the following relevantly similar argument.

Preliminaries: S might have considered any simple proposition—say, an attribution of a familiar color to a familiar object. There are always some simple propositions that S is not considering. Let’s call “NC” a theory that identifies, for some possible situations, some simple proposition that S can consider but is not considering. Clearly some such theories are true. It could not refute all NC theories to argue as follows:

First we assume that some proposition, PNC, is a proposition that NC implies that S is not considering in a possible situation PS. Next we note that any simple proposition can be expressed and S can see and understand an expression of it and thereby consider it. Now we assume that in PS an expression of PNC is something that S sees and understands. The expression is displayed and S is looking at it and comprehending it. Finally we infer that NC is incorrect, since by hypothesis NC implies that S is not considering PNC in PS, yet that is what S is doing in PS.

This anti-NC argument clearly fails. Its flaw is first to assume that an implication of NC for PS is that S is not considering PNC, and then just to assume that in the same PS situation things happen that make S consider PNC. The argument does not derive the possibility of S considering PNC in PS from what NC requires of its implications. The argument does not give any good reason to think that both the implication and the consideration can happen together. No such reason exists. Granted, any proposition that NC implies S not to be considering is one that S can consider. But no tenable basis exists for thinking that a true NC would imply the possibility of a case in which S does consider an NC implication.

Quite similarly, S can consider any proposition attributing to S a justified attitude. S can see a displayed expression of it and have it in mind. S can receive trustworthy testimony against any proposition that S can consider. But nothing indicates it to be possible that any proposition is at once both an Über-rule implication and the subject of the testimony. Clearly the Über-rule does not imply this possibility on its own. When the testimony to S denies that S is justified in believing p, for instance, the Über-rule counts the testimony as a defeater of S’s evidence for p. So the Über-rule does not imply that S is justified in believing p. Thus, the subject of the testimony, the proposition that S is justified in believing p, is not an Über-rule implication. The argument needs good reason to think that if there is an Über-rule, then it is possible that some proposition is at once both an Über-rule implication and the subject of the testimony. No such reason exists.

2.4. A Doubt about the Criticism of the Argument

Here is an objection to the criticism of the oracle argument.

An Über-rule implies, for any situation, a proposition asserting the justified attitude of S toward p in that situation. The reductio argument against an Über-rule can just specify that that proposition is the one that is stated by a displayed expression that S sees and understands. There is no need to appeal to implications of the Über-rule or otherwise to justify that possibility. The same goes for just assuming that S gets trustworthy testimony against whatever proposition S is considering. The possibility of getting trustworthy testimony against any considered proposition is clear. Whatever proposition is an Über-rule implication concerning p for S’s situation, S can get the testimony against that one.9

2.5. An Answer to the Doubt

Whatever is an Über-rule implication is something against which S can get the testimony, and possibly S does get testimony against whatever is an Über-rule implication about a possible case. It does not follow that is possible that S does get the testimony against what is an Über-rule implication. Again, the argument needs good reason to think that those things can happen together. No such reason is available. Compare again the counterpart reasoning about NC theories: Whatever proposition NC implies that S is not considering, S can consider it, and possibly S is considering whatever proposition NC implies not to be considered in some possible case. It does not follow that it is possible that a proposition that an NC theory implies that S is not considering is also a proposition that S is considering. Instead, for S to consider a proposition makes it not an implication of a correct NC theory. In the same way, when S considers a proposition attributing to S justified belief in p, the oracle’s testifying against it makes the considered proposition not an implication of an Über-rule. Thus, the oracle argument makes an unjustifiable assumption of a joint possibility.

The denial of this possibility does not imply any mysterious limit on what can happen. The higher-order defeat implications of Über-rules do not limit what propositions about justification can be considered while they are reliably denied. The implications just limit what propositions about justification can be true while they are reliably denied. Any Über-rule implication about any oracle case can be stated while the case occurs. The implication can be understood, believed with justification, and even known by anyone who does not have the oracular testimony or any other defeater for believing what the proposition says. DD theories even allow S to be considering an Über-rule implication while receiving the testimony, if S takes the oracle to be denying something else.

It remains highly plausible that certain evidence, such as the oracular denial to S that believing p is justified, does defeat at least modest justification for believing p. The justification is undercut. The testimony raises questions for S: “What am I missing about this evidence, or about p, or about justification? How does this evidence for p not justify belief?” S has available no support for answers that resolve S’s access to p’s truth. It is most reasonable for S to respond by withholding judgment, as the DD theories imply.10

3.1. Preliminaries for the Second Problem

The second problem for DD theories that is presented in Lasonen-Aarnio’s paper is stated briefly in a footnote:

Consider, for instance, evidence that whatever doxastic state S adopts, S is almost certain to commit some cognitive error. It seems that there simply cannot be any rational way of responding to such evidence, for the evidence has defeating force with respect to any attempt to take it into account. (2014: 331 n28)

This passage describes evidence that it suggests to be universally defeating. In order to make the possibility of such evidence maximally credible we can add a believable source for the evidence. To jeopardize DD theories more directly we can replace “cognitive error” with an error concerning the justified attitude. We can assume that the evidence is testimony from someone whom S knows to be trustworthy. We can assume that S is told that S has taken a drug with the effect that whatever attitude S adopts toward some particular proposition is almost certainly unjustified. Let’s call this evidence No Justified Attitude (NJA).

3.2. A First Argument

The cited passage says that Higher-order defeat theories apparently imply that there is no rational response to having the NJA evidence. Again, DD theories are not committed to the full thesis of Higher-order defeat, since they are not specified to imply anything about the defeat of justification for withholding. But it would refute DD theories for there to be no justified response to the NJA evidence when belief was justified. DD theories imply that there is always a justified response to defeated justification for believing, namely, withholding judgment.

3.3. A Criticism of the First Argument

A justified doxastic attitude remains available after receiving the NJA evidence. The previously justified attitude remains justified. The NJA evidence is too indiscriminate to defeat the justification for the previously justified attitude. If the NJA evidence were testimony against the justification of only some of the available doxastic attitudes, then the NJA evidence would leave it open that a mistake or a cognitive malfunction is making trouble. But evidence that uniformly opposes the justification of all attitudes, while giving no reason, is just mysterious. The NJA evidence has no intelligible bearing on either the truth of the proposition or what might be faulty about each doxastic response to it. When evidence is like that, the attitude that is justified given the rest of the evidence remains justified.

To test this out we can consider an example. Suppose that S has heard a weather forecaster predict rain in two days. S has found the forecaster to be pretty reliable. Suppose that S also has the NJA evidence concerning the proposition that it will rain in two days. In this situation S has good reason to think the following.

“The forecaster predicts rain in two days. The forecaster has been mostly right. I have good reason to believe the prediction and I do believe it. Now I am told by a trustworthy source that a drug I have taken makes me almost certain not to form a justified attitude. According to this evidence I am very likely to be making a mistake by believing that it will rain two days hence, though I have no clue of why that would be a mistake. By this same evidence, withholding judgment and disbelieving are as likely to be a mistake, if I opt for one of those attitudes. In those cases I see why. Those other attitudes are not responsive to the reliable forecast. The forecast supports the proposition that it will rain in two days. The testimony does not alter that. Belief remains justified.”

This thinking is quite reasonable. Relevantly similar thinking is available to -anyone with the NJA evidence. In the presence of NJA evidence the otherwise justified attitudes remain justified.11

3.4. A Revision of the NJA Evidence and a Second Argument

What has just been argued about the NJA evidence is not crucial to a defense of DD theories. Suppose that, contrary to what has just been argued, at least sometimes having the NJA evidence defeats the justification for believing a proposition. Whether or not the NJA evidence itself ever does that, the evidence can be enhanced to make it highly credible that it defeats the justification for belief. Suppose that, as in the previous example, S has the reliable forecaster evidence that it will rain in two days. As before, S is told that S is almost certain to fail to form a justified attitude toward the proposition that it will rain in two days, no matter which attitude S adopts. But now the testifier adds that S is especially likely to fail to have a justified attitude if S believes the proposition. We can call this the NJA* evidence.

When we replace the NJA evidence with the NJA* evidence, the forecast supporting the proposition that it will rain in two days seems to be neutralized by S’s higher-order evidence that belief in the proposition is especially likely to be unjustified. No reason is given for that, but it suggests that S’s inclination to believe is particularly rationally defective. This in turn suggests that another doxastic attitude is rationally better. Yet disbelief and withholding judgment are opposed by the testimony too. Additionally, they are out of keeping with S’s forecast evidence. In this version of the example DD theories seem not allow S to have any justified attitude toward the proposition.

3.5. A Criticism of the Second Argument

S does have available a justified attitude toward the proposition. Withholding judgment is justified. The reason is that quite generally, withholding judgment on a proposition is epistemically justified in the absence of undefeated evidence for or against its truth.12 The NJA* evidence denies that withholding judgment on the proposition is justified. But that denial is not a defeater of justification for withholding judgment. Justification for withholding is distinctively negative. Neither evidential support nor any other positive epistemic factor justifies it. What makes withholding judgment the justified attitude toward a considered proposition is the absence of good enough epistemic reason to believe it or to disbelieve it.13 In other words, withholding judgment on a proposition is rendered unjustified only by having justification for believing it or for disbelieving it.14 Having a balance of evidence that sufficiently supports the proposition or its negation would provide this justification. In the NJA* version of our example, S has no such balance of evidence.

S has the NJA* evidence for the proposition that withholding is not justified. That is support for this proposition about withholding. It does not support either that it will rain in two days or that it will not. Only such support is justification for belief or disbelief. Only such justification could prevent withholding from being justified.

3.6. A Third Argument

A new problem might seem to arise by strengthening S’s evidence against the justification of withholding judgment. Suppose that S receives tremendously trustworthy testimony that there is very nearly no chance that S is justified in either believing or withholding judgment on the proposition that it will rain in two days. The testimony asserts that S has taken a drug making S particularly prone to form those attitudes without justification, and that disbelief is unjustified too. We can call this evidence NJA**. S’s having the NJA** evidence gives S extremely good reason to think that withholding judgment on the rain proposition is unjustified. This fact might seem to be a defeater of any justification for withholding that S might have, thus making withholding judgment unjustified for S.

3.7. A Criticism of the Third Argument

The justification for withholding judgment is not subject to defeat.15 Defeaters defeat support for a proposition. Withholding judgment is not justified by support for any proposition. Evidence that withholding is unjustified does not justify not withholding. Not withholding requires believing or disbelieving the proposition. Justification for taking either attitude requires epistemic reason pro or con. S lacks that. S still has no undefeated evidence favoring either that it will rain in two days or that it will not. The forecast evidence is defeated by the NJA** evidence, with its trustworthy emphatic denial of the justification of belief. S still has no support for the proposition that it will not rain in two days (we can assume that rain in two days is as likely as not, given the rest of S’s rain evidence). The absence of undefeated support justifies withholding judgment.16 This bears out the implication of DD theories that defeated justification for belief with no new evidence about the proposition’s truth justifies withholding judgment.

4. Conclusion

Attractive DD theories avoid the two problems that we have investigated. The sort of defeat that DD theories imply remains quite credibly regarded as a factor in epistemic justification.


  1. Some philosophers pose problems for the distinction between first-order evidence and higher-order evidence; see for instance Dorst (2019). For present purposes S’s “higher-order evidence” concerning a proposition, p, at least includes evidence that bears on propositions asserting the justificatory status for S of p and evidence that bears on propositions asserting S’s competence at taking a justified attitude toward p.
  2. The argument is the principal theme of Lasonen-Aarnio (2014).
  3. This summary much simplifies the reasoning. For example, a hierarchy of less unwieldy epistemic rules might be thought to implement the higher-order defeat that the paper opposes. The main argument of the paper also criticizes the hierarchy idea. Soon it will be clear that the reason to set aside the main argument does not depend on the argument’s details.
  4. Numerous philosophers have advocated the view that this sort of evidence affects justification. Examples include Christensen (2010), Feldman (2005), Kelly (2010), and Sliwa and Horowitz (2015).
  5. Concerning the nature of disbelief, see Footnote 6 below; concerning the nature of withholding judgment, see Footnote 12 below.
  6. Assuming that to disbelieve a proposition is to believe its negation, disbelief does not require separate treatment.
  7. This sort of defeat of defeat is similar to David Christensen’s (2011: §3) view of what happens in the Careful Checking case that he discusses, though he is discussing credence and his defense of the preservation of a high credence in the face of a shocking disagreement is not phrased in terms of defeated defeaters. In any event, the present aim is to optimize the oracle argument. Nothing advocated here depends on the possibility of this sort of defeater defeat.
  8. We can continue to make the assumption of Footnote 6 that disbelief is belief in the negation, so that the reasoning equally covers the cases in which disbelief is initially justified.
  9. I thank a reviewer for raising an objection along these lines.
  10. Darren Bradley (2019: 15–16) briefly reconstructs and objects to Lasonen-Aarnio’s argument. Bradley’s treatment of the argument leaves room for the work done here. One of Bradley’s criticisms assumes that the Über-rule in question is correct, not just that it is correct if there is unlimited higher-order defeat. Another criticism assumes that the epistemology oracle tells only the truth. Those assumptions have not been made here. The argument as it is understood here does not need them. Bradley also discusses a case of trustworthy testimony that is not assumed to be true. The testimony denies to you what a correct Über-rule implies to have been your justified attitude toward a proposition. Bradley points out that the Über-rule can be correct about your pre-testimony justified attitude even if that attitude is no longer justified. This is possible because the testimony gives you new evidence. The Über-rule can also correctly state your justified attitude toward the proposition in the new situation. These points seem right about that case and they are helpful. As the argument is understood here, though, it describes another sort of case. It describes a case in which your receipt of the testimony is part of the situation about which the testimony asserts to you that a certain attitude is not justified. Lasonen-Aarnio’s presentation of the argument (quoted above) says that testimony is about “the very situation that you are right now in” when you receive it. The intention here is to assess an optimal version of an argument that includes the assumption that you have the testimony as evidence in the situation that the testimony addresses.
  11. Instead of the propositionally specific NJA evidence, trustworthy testimony might tell S that a drug has made S unlikely to form a justified attitude toward any proposition. This evidence too is too indiscriminate and mysterious to alter which attitudes are justified. Mostly the rest of S’s evidence continues to justify the same attitudes. S’s evidence supporting beliefs about what attitudes are justified might be defeated. Withholding judgment on those attitude propositions might be justified. Sections 3.43.6 below discuss what justifies withholding judgment.
  12. Withholding judgment as it is understood here requires nothing more than not judging a considered proposition. Jane Friedman (2013) has argued that for someone to refrain from believing and disbelieving a considered proposition is neither necessary nor sufficient for the person to suspend judgment on the proposition. Friedman advocates the view that suspending judgment is an independent doxastic attitude. Friedman’s view can be set aside here. If suspending judgment is an independent doxastic attitude, as Friedman holds, then the present topic is not suspending judgment. The topic is withholding judgment. It is not an independent attitude. Withholding judgment is not an attitude at all, if an attitude is understood to be taking some specific perspective on the truth of a proposition. In any case, the refraining from belief and disbelief that is taken here to be withholding judgment is a mental state that is subject to epistemic evaluation. What is under discussion here is the epistemic justification of this withholding state and its alternatives of belief and disbelief. Here is not the place to argue at length for the present views of what withholding judgment is and what justifies it. But it can be observed that one asset of the views is their help in defending plausible DD theories. (For further defense of the views see Feldman & Conee 2018.)
  13. For the present purpose of defending the existence of a justified attitude despite the NJA* evidence, it would be enough to hold that a lack of on balance evidence pro or con is sufficient to justify withholding judgment. The view being advocated, though, is that this lack is necessary as well as sufficient. This might seem wrong. It might seem that withholding judgment can also be justified by evidence for withholding. In our forecast example, for instance, S has modestly good evidence that it will rain in two days. Suppose that S receives reliable testimony that withholding judgment on the rain proposition is S’s justified attitude. If the testimony is credible enough, then it seems that S would be justified in withholding by the evidence that withholding is justified, despite having evidence for the rain proposition. Withholding would be justified, but it is not the testimonial evidence for withholding that justifies it. Rather, the testimony defeats the forecast evidence for the rain proposition. (Reasonable thinking that is available to S: “I must be missing something crucial about the forecast as a reason. I’d better back off from the belief.”) With this defeat, what justifies withholding here too is a lack of on balance support.
  14. A reviewer raises a worthwhile question about this claim. Here is preparation for the question: Suppose that an oracle says to S, “It is not so that withholding is rendered unjustified only by justification to believe or to disbelieve.” We can call the denied claim “W”. By DD theories, a sufficiently trustworthy oracular denial would be a defeater of any evidence S has for W. S not would be justified in believing W. Suppose that the oracle then asserts the following to S about some proposition, p, concerning which DD theories imply that withholding by S is justified: “Your withholding judgment on p is unjustified.” By DD theories in conjunction with the present view of the justification for withholding, S’s justified attitude toward p remains that of withholding judgment, since the oracle’s denial of justification for withholding on p gives S no new evidence concerning p’s truth. Yet W is unjustified for S, and so it seems that W could not be relied on to justify S’s continuing to withhold on p. With that preparation, here is the question: Is there any adequate defense of the DD implication that S remains justified in withholding judgment on p? Here is an answer in defense of DD theories: W is offered as a fact about what justifies withholding. W’s truth is not being held to depend on anyone’s attitude toward it. In the situation that raises the question, S would not be justified in believing W, given the oracle’s denial of W. S’s lacking justification to believe W, or even S’s having justification to disbelieve W, does not affect what actually justifies S’s withholding judgment. S is justified in withholding judgment on p by not having undefeated reason to believe p or to disbelieve p. S has the further oracular testimony as justification for denying that this withholding is justified. That testimony is justification for a falsehood about justification. It does not defeat S’s justification to withhold judgment on p. What justifies withholding is not subject to defeat. Defeat spoils justification only when it spoils support for a proposition. As W implies, it is not support for a proposition that justifies withholding, but rather a lack of support.
  15. This aspect of the advocated view of the justification for withholding judgment confirms Lasonen-Aarnio’s own tentative conclusion about higher-order defeat. She suggests that it might be best to deny that higher-order defeat is universal: “[I]n some cases a state can be perfectly epistemically rational even if one has what would seem like strong evidence for thinking that it is not” (2014: 342). The present view concurs. Evidence against justification for withholding judgment does not defeat the justification. The present view derives this exemption from the special negative character of the justification for withholding judgment. But the advocated DD theories still have it that the justification for belief allows unlimited higher-order defeat.
  16. A reviewer suggests commenting on the relation between this justification for withholding judgment and the opposition of advocates of higher-order defeat to epistemic akrasia. Here are a few words about that. Following Horowitz (2014: 718) I understand opposition to “epistemic akrasia” to be opposition to this thesis: it can be rational to believe something of the form “P, and my evidence does not support P”. So understood, the present view does not imply that there is any epistemic akrasia. That is, the combination of DD theories and the present negative view of what justifies withholding does not imply the possibility that a belief of that form is ever justified. Justification for believing that one’s evidence does not support P might always defeat one’s justification for P. The combination does imply the possibility that S is justified in withholding judgment on p while S is also justified in believing that S’s withholding judgment on p is not justified. These attitudes might seem inharmonious enough to count as some sort of epistemic akrasia. But in some possible cases they exhibit no rational shortcoming. In such cases the proposition about unjustified withholding on p is well supported to S and yet still S has no undefeated reason to believe or to disbelieve p. S could quite sensibly think: “According to my evidence, my withholding judgment on p is not justified. So I am justified in thinking that this is true. Still, it is strange. p has nothing going for it or against it as far as I can tell. Nothing justifies my believing p or disbelieving p. So I am not doing either one.” This combination of believing and withholding is entirely responsive to reason.


Bradley Darren (2019). Are There Indefeasible Epistemic Rules? Philosopher’s Imprint, 19(3), 1–19.

Christensen David (2010). Higher-Order Evidence. Philosophy and Phenomenological Research, 81(1), 85–215.

Christensen David (2011). Disagreement, Question-Begging, and Epistemic Self-Criticism. Philosopher’s Imprint, 11(6), 1–22.

Dorst Kevin (2019). Higher-Order Uncertainty. In Skipper Mattias and Steglich-Petersen Asbjorn, (Eds.) Higher-Order Evidence: New Essays (35–61). Oxford University Press.

Feldman Richard (2005). Respecting the Evidence. Philosophical Perspectives 19, 95–119.

Feldman Richard, and Conee Earl (2018). Between Belief and Disbelief. In McCain Kevin (Ed.) Believing in Accordance with the Evidence (71–89). Springer.

Friedman Jane (2013). Suspended Judgment. Philosophical Studies, 162(2), 165–181.

Horowitz Sophie (2014). Epistemic Akrasia. Noûs, 48(4), 718–744.

Kelly Thomas (2010). Peer Disagreement and Higher Order Evidence. In Feldman Richard and Warfield Ted (Eds.), Disagreement (121–174). Oxford University Press.

Lasonen-Aarnio Maria (2014). Higher-Order Evidence and the Limits of Defeat. Philosophy and Phenomenological Research, 88(2), 314–345.

Sliwa Pauline, and Horowitz Sophie (2015). Respecting All the Evidence. Philosophical Studies, 172(11), 2835–2858.