1 Evidence as reasons for belief?

The expression ‘Q is evidence that P’ is used to refer to the evidential relation that relates one proposition (the evidence) to another proposition that is supported by the evidence. Every serious knowledge-seeking discipline employs it, and it is a relation that has also had a central place in epistemology. If reasons are to be ‘first’ in epistemology, something needs to be said about how reasons are related to that evidential relation and why the ideology of reasons is important for understanding the nature of this evidential relation.

Here is the principle Schroeder (2021: 16) provides in response:

Evidence as reasons [for belief] (ERB): Evidence matters in epistemology because if something is evidence that P, it is a reason to believe that P.

But here’s an alternative reasons-centric view of the evidence-that relation:

Evidence as Truth-Indicative Reasons (ETIR): Evidence matters in epistemology because: Q is evidence that P for some agent S iff: (i) Q is a reason that S has, and (ii) Q supports the truth of P.

The intended relations of ‘support’ are familiar deductive and ampliative inference relations among sets of propositions.

It’s notable that ETIR and ERB are logically compatible and logically independent. Their compatibility is of note when considering whether both are true. For were both true, it is arguable that ERB’s truth would in part be owed to ETIR’s truth. For if the aim of belief involves truth, then evidence of truth would seem to provide reasons for belief. Their independence is of note because there are objections to ERB that gain no traction at all against ETIR, allowing us to endorse ETIR without necessarily inheriting ERB’s problems. We’ll highlight a few challenges to ERB that ETIR seems to survive.

The first objection to ERB involves lotteries. Take a lottery case in which you have a cognitively mature ticketholder who recognizes that: they have significant evidence (L) that their ticket is a loser because they know that it’s highly objectively probable. But they also know that they cannot know (L). Now, if knowledge is the aim of belief, then this ticket holder can also know that they cannot achieve the aim of believing (L). In general, if you know that it is metaphysically impossible for you to achieve the aim of φ-ing, then you have no reason to φ. At the very least it seems plausible that: if you know that you cannot achieve the aim of believing that P by believing it on the basis of E, then E provides you with no reason to believe that P. This is inconsistent with ERB, but not ETIR. For the latter view is silent about how one ought to respond to their evidence.

Doubtless, a natural response is to deny the knowledge aim of belief or to deny that lottery propositions cannot be known. Here’s a way of driving this objection without reliance on those assumptions. Take a case with a self-defeating belief where the following conditions obtain:

  1. (i)

    P is true iff S does not believe that P.

  2. (ii)

    S does not believe that P.

  3. (iii)

    S knows that (i) and (ii) obtain.

It is easy enough to construct possible cases where these conditions are satisfied for some agent S, and many have done so.Footnote 1

If conditions (i)-(iii) obtain, then S has evidence which entails the truth of P. But again, it will be strange to think that S has any reason to believe that P because S knows that she cannot truly believe P. Provided truth is the aim of belief, we again have a case where an agent can know that she cannot fulfill the aim of belief by responding to E, and so E provides one with no reason to form a belief that P. This is inconsistent with ERB but consistent with ETIR since ETIR does not require that one’s evidence for P provide reason for believing P.

One last concern with ERB stems from cases of self-fulfilling belief. The relevant sorts of cases are ones where:

(i*) P is true if S will believe P at a future time t, and ¬P is true if one will believe ¬P at t.

(ii*) S does not know what she will believe at t.

(iii*) S’s total evidence does not presently support P over ¬P.

(iv*) S knows that (i*)-(iii*) obtain.

The possibility of such cases has been argued to involve cases where one has sufficient reason to believe P at t and sufficient reason to believe ¬P at t, despite presently lacking evidence that supports P over ¬P (cf. Silva, 2023a). Reasons-centric views of evidence should help us understand how such cases are possible. ETIR does this by identifying evidence with one’s reasons that stand in suitable deductive and ampliative support relations to propositions, but not by dictating what one’s doxastic response should be to one’s evidence. It is the job of a norm of belief to provide the rest (e.g., believe only truths, or believe only what you know). ERB does this by not saying that reasons to believe P are evidence that P, even though being evidence that P is a reason to believe P. But this is a surprising asymmetry. For one would expect that if ERB is true and, thus, that if Q is evidence that P for S, then Q is a reason for S to believe that P, then the converse of that claim should also be true: if Q is a reason for S to believe that P, then Q is evidence that P for S. Of course, cases of self-fulfillment challenge this converse claim since knowing (i*)-(iii*) is a reason to believe P (or ¬P), while not being evidence for either.

So, theoretically, there is a kind of harmony to the way in which ETIR manages lottery cases, cases of self-defeat, and cases of self-fulfillment: it is consistent with them all because it connects evidence to reasons that stand in support relations, without going on to say just how one ought to respond to that evidence. Again, that job is left for debates about the norm of belief. But ERB cannot manage lottery cases or cases of self-defeat. And ERB manages cases of self-fulfillment only by being silent about why reasons to believe P are not always evidence that P.

2 Evidence, factivity, and error

On Schroeder’s view, evidence need not be true, and it need not be factively accessible. This stands in contrast with other accounts of the nature of evidence and how we access it. To focus discussion, take the following:

Factive Content Views of Evidence (FCV). A proposition P is part of S’s evidence if and only if S stands in, or is in a position to stand in, some factive stative attitude that has P as its content.Footnote 2

Factive stative attitudes include amodal factives like knowing that P and being aware that P. But they also include modal factives like seeing that P, remembering that P, being informed that P, and P’s being self-evident to one. Some incarnations of FCV require that one actually stand in a factive relation to P (Williamson, 2000; Littlejohn, 2017) while others require only that one be in a position to do so (Lord, 2018; Silva 2023b).

Schroeder argues that FCV cannot account for the presence of rational belief in case-pairings involving beliefs formed in normal circumstances where one sees that P and minimally different cases where one unwittingly merely seems to see that P (Schroeder, 2021: 64–65). To illustrate his concern, take the following pair of cases:

Case 1

You inhabit a normal environment. You’re a security guard watching the entrance to your building. You watch your friend Job enter the building and believe (J) Job just entered the building because (Saw) you saw Job enter the building.

Case 2

You inhabit a normal environment. You’re a security guard watching the entrance to your building. Your personal history diverges in no way from your personal history in Case 1 except that you watch what seems to be your friend Job enter the building and believe (J) Job just entered the building, but not because: (Saw) you just saw Job enter the building. For it was Job’s twin brother that you saw. You did not know and had no reason to think that Job had any siblings, much less a twin.

In Case 1 you see Job and have (Saw) as evidence to justify your belief. In Case 2 you do not have (Saw) as evidence according to FCV because (Saw) is false. Even so, in both cases Schroeder would have it that your belief in (J) is propositionally and doxastically rational (= rational to believe and rationally believed), and this fact can be explained only if we allow the false proposition (Saw) to function as evidence you possess in both Case1 and Case 2.

Concessive responses to this kind of problem among advocates of FCV will be familiar. There are a range of positive epistemic properties that could be argued to be present in Case 2. Many have argued that in situations like Case 2 agents can be excusable (Littlejohn, 2017), or responsive to apparent facts (Sylvan, 2015), or exercise good belief forming dispositions (Lasonen-Aarnio, 2020; Williamson, forthcoming). Advocates of FCV can simply resist Schroeder’s argument by insisting that one-or-all of these properties which are present in Case 2 are being confused with responsiveness to facts which justify one’s beliefs.

However, advocates of FCV need not make any concession, and can affirm Schroeder’s judgment about Case 2. For even in Case 2 you are in a position to know that (SeemSaw) it seems to you that you just saw Job entering the building. FVC, and its anti-skeptical motivations, are consistent with the idea that facts about your internal mental life can provide you with sufficient reason to believe (J) in Case 2. How could this be? Reach back a bit. Decades ago, Alston (1989) argued for a certain reconciliation of internalism and externalism about justification. His basic idea was that internal states, like seeming states, can have the power to justify beliefs about the external world and this needn’t be explained along internalist lines, e.g., in terms of the phenomenology or introspective accessibility of such states. Rather, the justificatory power of internal states can lie with their reliability. That is, internal states justify beliefs because they stand in the right kind of reliability relations to the world. Moving away from the idea that mental states are justifiers to the idea that facts are, doesn’t undermine Alston’s basic idea. It is indeed a short step from arguing that mental states are justifiers for beliefs about the external world in virtue of their reliability to arguing that facts about one’s mental states are justifiers for beliefs about the external world provided there is a reliable connection between the two.

Can this Alstonian view give us doxastic rationality in addition to propositional rationality in Case 2? Yes. Notice that your belief in Case 2 would likely pass a counterfactual test that serves as a good indicator that your belief in (J) is based on (SeemSaw): you would not have believed (J) if (SeemSaw) were not true. Consider next, retrospective judgements about past doxastic mistakes. For example, suppose in Case 2 you were later told that it was Job’s twin, and not Job, that you saw enter the building. Were you then asked why you believed (J) when it was false you would likely say that: it seemed to you that you saw Job enter the building. If this retrospective judgment is true, it strongly suggests that your perceptual belief in Case 2 was in some way responsive to (SeemSaw), which is a fact that you knew or were at least in a position to know. If correct, then we have the makings for a non-concessive view on which there is no difference in propositional or doxastic rationality between your beliefs in Case 1 and Case 2.

It should be noted that not all instances of FCV can take advantage of the above Alstonian explanation of doxastic rationality in Case 2. Only those views on which merely being in a position to stand in a factive relation to P is enough for one to possess P as evidence will allow for it (Lord, 2018; Silva 2023b). For if actually standing in a factive relation to (SeemSaw) were required then every justified perceptual belief in a case like Case 2 would have to involve a backup knowledge-constituting belief about what you seem to perceive. But it’s implausible that for every perceptual belief we form we also form a second belief about how things perceptually seem (despite being in a position to do so). Further, making good on explaining how one’s belief that (J) could be suitably related to (SeemSaw) so as to ensure that it’s “properly based” on it will require some kind of non-causal theory of the basing relation to account for cases where (SeemSaw) is not believed and thus not part of the causal story for your belief in (J). Providing such a theory is beyond what can be done here.Footnote 3

3 Evidence, factivity, and subjective defeat

The following case will help us navigate Schroeder’s (2021: 106) next objection to FCV. It is an objection that centers around the phenomenon of subjective defeat:

Subjective Defeat. At t1 Job sees and remembers (B) that a barn is in the field, and then on that basis competently comes to believe (F) that a structure used for farming is in the field. As a result, Job comes to know (F). But later, at t2, Job comes to rationally believe (Pr¬B) that the structure in the field was very likely not a barn because he was in fake barn country. However, (Pr¬B) is false and Job really did see and remember that a barn was in the field as he was not in fake barn country.

Is it rational for Job to believe (F) on the basis of (B) after he comes to rationally believe (Pr¬B)? Schroeder rightly requires a negative answer: it is not rational for Job to believe (F) because coming to rationally believe (Pr¬B) ensures that it is irrational for Job to believe (B).

But, Schroeder says, this judgment cannot be explained by evidence that Job has if FCV is true. For the necessity direction (= right-to-left direction) of FCV implies that Job cannot have (Pr¬B) as evidence since it’s false, and the sufficiency direction (= left-to-right direction) of FCV implies that Job retains (B) as evidence since he really did see that a barn was in the field and he continues to remember that there was a barn in the field.Footnote 4 So it’s hard to see how FCV could give us the verdict that it is not rational for Job to continue hosting his belief in (F) when he comes to rationally believe (Pr¬B).

As noted above, concessive responses will be familiar. However, here they seem to involve allowing for some form of epistemic akrasia, where one has justification to believe: (B), and (B) is probably false. Many would wish to resist this.

Fortunately, non-concessive responses are available. Indeed, FCV is compatible with the judgment that it is not rational for Job to believe (F) or (B). For if FCV is true, then Job’s belief in (Pr¬B) is rational only if Job stands in, or could easily stand in, a factive relation to some fact that makes it rational. Advocates of FCV are free to argue that the facts that Job has that make his belief in (Pr¬B) rational also defeat his belief in (F) by defeating the rationality of believing (B). For example, suppose Job comes to rationally believe (Pr¬B) because Job came to know: (T) that a reliable person told him that he was in fake barn country. Advocates of FCV remain free to claim that Job’s knowledge of the fact (T) defeats the rationality of believing (B), and hence (F). This is consistent with holding that (B) remains a part of Job’s evidence, which an advocate of FCV must hold since Job really did see that (B) and correctly remembers that he saw that (B).

Why think (T) could defeat the normative significance of (B) in relation to believing (F)? After all, (B) is such good evidence for (F). Here there are two things to note. First, (B) can be part of one’s evidence even if one does not know that it is part of one’s evidence. Second, (B) can be part of one’s evidence even if one has strong reason to think that (B) is not part of one’s evidence. And if one has strong reason to think (B) is not part of one’s evidence, then there is something rationally defective and epistemically improper about continuing to believe (F) on the basis of (B)–even if (B) remains part of one’s evidence. So, arguably, the normative significance of (B) in relation to believing (F) gets high-jacked by your knowledge of (T) and its support of (Pr¬B).

What this non-concessive response demands is what was demonstrated in Sect. 1: one can have conclusive evidence for P without having sufficient reason to believe P. Cases of self-defeat are one sort of case where this possibility obtains. Cases involving defeating testimony are another–or so advocates of FCV who wish to retain Schroeder’s verdict about subjective defeat are hereby advised to think.

4 Knowledge from falsehood

According to Schroeder (2021: 228 − 31) knowledge is belief for sufficient objective and subjective reason. Objective reasons are facts about one’s factive attitudes, e.g., the fact that one sees that P, remembers that P, knows that P, and so forth. One’s objective reasons that favor believing P are sufficient only if they are weightier than all the objective reasons that exist. One believes for a reason only if one possesses that reason and one’s belief is based on that reason.

Limiting knowledge to objective reasons causes problems in connection with cases where agents form knowledge-constituting beliefs on the basis of falsehoods rather than facts, much less facts about one’s factive mental states. To see the problem, take a fairly standard case of knowledge from falsehood:

KFF. Ted knows that (100) he has 100 copies of his handout for the attendees of his talk. He wonders whether he brought enough for every attendee. He does a careful headcount and forms just one belief from his count: (53) there are exactly 53 attendees. From (53) and (100) he infers (Enough) that there are enough handouts for each attendee. But Ted’s headcount was wrong: there are only 52 attendees. One person, Bill, unobtrusively switched seats and got counted twice.

Intuitively, Ted knows (Enough). An adequate theory of knowledge should be able to explain this. At present, there are three kinds of explanations of knowledge-yielding falsehoods. Defeater-based accounts maintain that reasoning from a false premise yields knowledge if either the negation of the false premise is not a defeater for the true conclusion or the true conclusion is indefeasibly justified by a truth that is entailed by the false premise.Footnote 5Safety accounts have it that reasoning from a false premise generates knowledge if the inferential path from the false premise to the true conclusion is modally stable – if it could not have easily given rise to a false conclusion.Footnote 6 Finally, closeness-to-the-truth accounts hold that reasoning from a false premise yields knowledge if the false premise is semantically or epistemically close to the truth.Footnote 7

None of these proposals seem to be available for Schroeder. The trouble for Schroeder is that (53) is no fact, and since (53) is the only belief Ted forms upon counting attendees there is no other fact that Ted might rely on in forming his belief in (Enough), e.g., the fact that (≈ 53) there are approximately 53 attendees.

We suppose it’s possible for Schroeder to argue that there is a factive relation Ted stands in to (≈ 53) even if he does not consciously consider it or believe it. For suppose seeing that P is less demanding than knowing that P by not requiring belief or justification. Then, I might see that something is flying over a building even if I do not consciously consider that thought or come to believe it. So maybe Ted can see that he has enough handouts even if he doesn’t believe it. Supposing this is right, we still need to account for Ted’s knowledge in KFF. And Ted’s knowledge of (Enough) in KFF requires that his belief be based on objective reasons that are sufficient. But in the example above, Ted’s knowledge is not based on his seeing that he has enough handouts, but on his belief in (100) and (53). This strikes us as a problem.Footnote 8Footnote 9