1. WILLIAMSON'S ARGUMENT FOR AMBIGUOUS RATIONALITY

Williamson (Reference Williamson2017) argues that debates about epistemic rationality are plagued by an ambiguity that has further deleterious effects on epistemology. I will summarize Williamson's argument for the ambiguity thesis first and then turn to the alleged harmful consequences.

Williamson's starting point is that the question “What is rational to believe for me in my current circumstances?”, even if restricted exclusively to epistemic rationality, can be understood in two ways. We can understand it as asking which contents of potential beliefs are adequately supported by one's current evidence. Williamson suggests that the original question is answered in accordance with the following schema:

But we use the term ‘rational’ not only to assess beliefs, but also to assess epistemic agents. Therefore, the question “What is rational to believe for me in my current circumstances?” can be understood as asking what a rational agent would believe in my current circumstances. This approach might be less familiar, so it is helpful to consider an analogous case. We can attempt to answer the question “What is the righteous thing to do?” by considering what a righteous person would do. We consider what we take to be a prototype of a righteous person (whoever you deem suitable to name as a paragon of virtue), and ask what that person would do in our circumstances, in order to determine what is righteous to do. Likewise, we can ask what a rational person in our circumstances would believe in order to answer the question what is rational to believe. Williamson suggests that the original question is answered in accordance with this schema:

Both schemas are propositional, not doxastic. Accordingly, they do not hold that a belief is doxastically rational if the condition on the right-hand side is met. Suppose your belief that p is the result of a blow to the head. Even when one meets the condition on the right-hand side, this belief is not doxastically rational, although it is propositionally rational.

Williamson contends that in the epistemology literature, there is a tendency to conflate these two schemas in assessments of statements of ‘it is rational to believe that p’. Sometimes, such statements are assessed along the lines of the Content-oriented schema, at other times, along the Disposition-oriented schema. But the legitimacy of this practice depends on the equivalence of both schemas, such that it is rational _cont to believe p iff rational _disp to believe p. In effect, it is tacitly assumed that the Equivalence Schema is true:

If (ES) is false, then the notion of rational belief is ambiguous between rational _cont and rational _disp, as it would be possible to satisfy one but not the other. Thus the question of whether the notion of rational belief is ambiguous crucially depends on (ES).

Williamson argues against (ES) by using a number of thought experiments that involve an imaginary device that he calls a ‘brain scrambler’. The device is imagined to emit waves which alter the content of unconscious short-term memory, causing people to make fallacious inferences from their evidence. For example, a person affected by the brain scrambler might confidently declare that 6 + 6 = 13, without the person noticing that she is under the influence of the brain scrambler (p. 268).

Against the right-to-left direction of (ES), Williamson gives the following case:

The problem is that no evidence supports a contradiction, so Innocent's evidence does not support believing this contradiction. While Innocent does have the disposition to confirm her beliefs to what her evidence supports, the influence of the brain scrambler leads her to have a belief that is not supported by her evidence. Williamson points out that the brain scrambler does not change Innocent's general disposition. It merely temporarily interferes with her disposition, but does not destroy it, just like a bubble-wrapped vase remains fragile, even when the bubble-wrap interferes with its fragility-disposition. Thus, the right-to-left direction of (ES) is falsified. Innocent has the relevant disposition and believes p, but her evidence does not support p.

Williamson offers a similar counterexample against the left-to-right direction of (ES). Only this time, “[t]he brain-scrambler may cause Innocent to refuse to believe a tautology, even though her evidence entails it (because everything does)” (p. 269). But while Innocent has the disposition to conform her beliefs to what her evidence supports, she does not believe the tautology, as the brain scrambler prevents her from forming the belief. Thus, the left-to-right direction of (ES) is falsified. Innocent's evidence supports the tautology, but even though she has the relevant disposition, she does not believe it.

Williamson moves on to further counterexamples that involve contingently true propositions, but I omit these cases because they are structurally identical with the ones just introduced. If both of the above cases are correct, then either direction of (ES) fails and thus we should reject (ES). And if we must reject (ES), then the notion of what is rational to believe is ambiguous between rational _cont and rational _disp.

Overlooking this ambiguity has distorted the debate about evidence in epistemology, Williamson holds. Many argue in favor of a phenomenal conception of evidence, according to which one's evidence consists of one's non-factive mental states. A popular argument for this view compares subjects in a skeptical scenario with those in a non-skeptical scenario. While both subjects differ in what they know, their beliefs are said to be equally rational. Consequently, their evidence must be identical, which suggests the phenomenal conception of evidence.

But if rationality is ambiguous, this type of argument can be evaded. A brain-in-a-vat (BIV) is said to have rational beliefs, as it has good cognitive habits, but just happens to be in a bad situation. Williamson writes that this amounts to an ascription of rationality _disp (p. 271). But since (ES) fails, we need not hold that the BIV is rational _cont, and thus that it has beliefs that are supported by the evidence. Thus we need no longer hold that subjects in skeptical scenarios and in contrasting non-skeptical scenarios have the same evidence. This makes room for externalist conceptions of evidence, such as Williamson's thesis that one's evidence is what one knows (“E = K”).

2. RESISTING WILLIAMSON'S ARGUMENT

I believe that we should resist Williamson's argument for the ambiguity of rationality. While I see some reason to doubt his counterexamples to (ES)Footnote ², my belief is based on a more substantial worry. We should reject (DS) because it is unfaithful to the key thought behind the idea of capturing what is rational to believe in terms of what a rational agent would believe. Without (DS), Williamson's argument against the equivalence between the content-oriented and the disposition-oriented notion cannot get started and the replacement of (DS) I will offer does not succumb to Williamson's counterexamples.

Williamson cannot just help himself to (DS) gratuitously.Footnote ³ The aim of (DS) is to capture a legitimate notion of what is propositionally rational to believe. We should only accept (DS) if it gets paradigmatic cases of rational and irrational belief right. I will demonstrate that (DS) does not seem to be particularly successful at this. If (DS) has this flaw, then we should resist any argument based on (DS), including Williamson's argument to the conclusion that rationality is ambiguous.

Hopefully, we can all agree that there are paradigmatically irrational beliefs, beliefs that are irrational qua content and structure and which we can pre-theoretically classify as irrational. Here is an incomplete list of what I have in mind. Belief in a contradiction, such as that it is raining and that it is not raining; self-defeating beliefs, such as believing that one has no beliefs; conceptually confused beliefs, such as believing that there is a round triangle; delusional beliefs, such as believing that one is God.

We should judge a theory of rationality at least partly by how well it stands up to those judgments. (CS) can capture those judgments. As Williamson himself holds, no evidence supports a contradiction, so according to (CS) beliefs in contradictions count as irrational. Likewise, it seems that one can simply have no evidence in favor of having no beliefs, since believing that one has such evidence is already self-defeating. Nobody competent with the concepts ‘round’ and ‘triangle’ can have evidence for believing that there is a round triangle. And it is quite hard to imagine anybody having good evidence for believing that they are God.

In comparison, (DS) allows that instances of said beliefs can be rational. For any of the mentioned beliefs, we can construct a brain scrambler case. In fact, Williamson's argument against the right-to-left direction of (ES) has Innocent believing an outright contradiction. His case not only demonstrates that (ES) fails, it also entails that according to (DS), Innocent's belief in the contradiction counts as rational. In the same circumstances, that is, being manipulated with the brain scrambler and with the same evidence as Innocent, anybody disposed to conform their beliefs to what their evidence supports would believe the contradiction. Thus, according to (DS), this belief is rational. For all the other examples of paradigmatic irrational beliefs mentioned above, we can give parallel cases in which the brain scrambler causes Innocent to have such beliefs. Since all these beliefs satisfy the demands of (DS), they count as rational. This seems to be a questionable result.

(DS) should provide an answer to the question “What is rational to believe for me in my current circumstances?” that captures pre-theoretical judgments. But the cases from the previous paragraph demonstrate that (DS) glaringly fails to meet this aim. Therefore, we should reject (DS) as a theory of what is rational to believe.

If we reject (DS), then the argument for the ambiguity of rationality fails. The argument was based on the claim that (ES) is false, but (ES) is partly based on (DS). Without (DS), the whole chain of argument cannot get started.

My rejection of (DS) does not amount to a rejection of the idea to capture what is rational to believe in terms of what a rational agent would believe. I merely believe that (DS) inadequately captures this idea. As Williamson himself puts it, the key idea is that we “judge the rationality of an act by whether a rational agent would have done it” (p. 264). We approach the question whether it is rational to believe p by answering the question whether a rational agent would believe p. But the question in this form is imprecise and needs sharpening to develop an account of rationality in terms of rational agents.

Consider the following example, which, for intuitiveness, is based on a moral case. Suppose I face the choice whether to Φ, where Φ-ing has potentially immoral consequences. I attempt to answer the question “What is the moral thing to do?”, by considering what the archetype of a moral person would do. For vividness, suppose I ask myself: what would Jesus do? By asking myself what Jesus would do in my situation, I am not interested in knowing what Jesus would do if he was manipulated by a brain scrambler, even if in my own situation, I am manipulated with a brain scrambler. That would be absurd. What I do want to know is what Jesus would do in my situation if he were to fully display his property of being a moral exemplar. What I am interested in is not what somebody with the relevant disposition would do, but rather what somebody manifesting that disposition would do. The answer to the former question includes scenarios in which that disposition might not be triggered (e.g. due to manipulation with the brain scrambler). The latter abstracts from such situations and is sensitive only to scenarios in which the disposition is manifested.

The same holds for rational belief. When we ask ourselves what a rational agent given our evidence would believe, we are not asking ourselves what a rational agent would believe when she does, perhaps understandably and through no fault of her own, not manifest her disposition to believe what her evidence supports. What we do ask is what a rational agent would believe if she were to manifest her disposition to conform her beliefs to the evidence. Thus, to do justice to the idea to capture what is rational to believe in terms of what a rational agent would believe, we should replace (DS) with what I call the Disposition-manifestation schema.

(DMS) can handle ordinary cases, but also more extravagant cases like the brain scrambler. Let us return to Innocent and assume that her circumstances are normal, no manipulation is going on. Given her mathematical competence, she has excellent evidence for believing that 7 is a prime number. Innocent is also rational _disp to believe that 7 is a prime number. In her circumstances and with her evidence, someone who manifests their disposition to conform their beliefs to what their evidence supports would do so by believing that 7 is a prime number.

Now let us assume that Innocent's circumstances involve the manipulative influence of the brain scrambler that leads her to believing that 7 is not a prime number. Yet, Innocent is not rational _disp to believe that 7 is not a prime number. Innocent's evidence remains as good as before, and even in her circumstances, which include manipulation, it is not the case that someone who manifests their disposition to conform their beliefs to what the evidence supports would manifest it by believing that 7 is not a prime number. Thus on (DMS), beliefs induced by the brain scrambler that go against what the evidence supports do not count as rational _disp.

A bit of elaboration on ‘would manifest that disposition by …’ is in order. A vase is fragile, which means that it has the disposition to shatter when struck. Suppose that a vase that is wrapped in 1,000 layers of bubble wrap is struck – nothing happens to the vase, it does not shatter. Yet, we can ask what a manifestation of the disposition in those circumstances, being wrapped in 1,000 layers of bubble wrap, would be. The answer in this case is simple. The manifestation would be that the vase would shatter. More generally, for any circumstances, we can still ask what a manifestation of a disposition in those circumstances would amount to, even if in those circumstances, the disposition is not manifested. We often have a pretty firm grip on what the answer to such questions would be.

Things are only slightly more complicated in the case of rational belief. One's disposition to conform one's belief to the evidence can be manifested in different ways: believing that p if the evidence strongly supports p or withholding belief if the evidence does not favor p over its negation. But qua stipulation, the evidence in Innocent's case supports that 7 is a prime number, which means that someone with Innocent's evidence could manifest her disposition to conform her beliefs to the evidence only by believing that 7 is a prime number (and by refraining from believing that 7 is not a prime number).Footnote ⁴ Thus, even in her circumstances, under the influence of the brain scrambler, it is rational _disp to believe that 7 is a prime number and not rational _disp to believe that 7 is not a prime number. This is entirely compatible with the fact that in the given case, Innocent, due to the brain scrambler, could never have a doxastically rational belief that 7 is a prime number, since we are only concerned with propositionally rational belief.

(DMS) seems better at capturing the key idea of accounting for rational belief in terms of rational agents that was suggested at the outset. This gives us at least prima facie reason to adopt (DMS) and those who are drawn to the idea behind rational _disp to prefer it over (DS). But my main concern here is not whether (DMS) is true. My main concern is the question whether there is an ambiguity between rational _disp and rational _cont. Once we drop (DS) in favor of (DMS), we must slightly change our approach to answering this question. Since (ES) depends on (DS), we must drop (ES) in favor of a new principle that adopts the right-hand side of (DMS):

Now we must simply check whether (ES*) is also subject to Williamson's counterexamples to decide whether Williamson's main thesis of an ambiguity between rational _disp and rational _cont can be maintained.

Perhaps it is already clear that (ES*) is immune to the counterexamples that Williamson leveled against (ES). In any case, we can easily establish it explicitly. In his counterexample to the left-to-right direction of (ES) mentioned in the previous section, Innocent's evidence supports the tautology, but even somebody with the relevant disposition would not believe the tautology under the influence of the brain scrambler. Yet, it remains true that – even under the influence of the brain scrambler – the manifestation of the disposition would consist in believing the tautology. Thus, on the scenario envisioned, the left-hand-side and the right-hand-side of (ES*) do not come apart.

The same holds for the counterexample for the right-to-left direction. Innocent's evidence does not support the contradiction. Therefore, it remains true that for somebody with the relevant disposition, her evidence and in her circumstances, if they were to manifest their disposition, they would not manifest it by believing the contradiction. According to (ES*), this means that the evidence does not support believing that p, which is in accord with the stipulation of the case. Therefore, the counterexample does not accomplish to bring the left-hand-side and the right-hand-side of (ES*) apart.

But if (ES*) stands, then Williamson's argument for the ambiguity of rationality breaks down. Once we replace (DS) with (DMS) and consequently (ES) with (ES*), the cases Williamson uses do not pass muster as counterexamples to (ES*). If the cases fail to bring both sides of (ES*) apart, they fail to demonstrate that rational _disp and rational _cont come apart. Thus we no longer have reason to believe that rationality is ambiguous between rational _disp and rational _cont.

3. THE CASE FOR UNAMBIGUOUS RATIONALITY

So far, we have only seen that Williamson's argument against (ES) cannot be extended against (ES*). But we can do better and provide an argument in favor of (ES*). If we have a positive argument for (ES*), then we have a positive argument for the thesis that rational _disp and rational _cont do not come apart and that the notion of rational belief is unambiguous between these two readings.

(ES*) is true because denying each direction of the biconditional is contradictory. This can simply be demonstrated by breaking (ES*) down into two conditionals that each captures one direction of the biconditional.

Let us begin with (ES*LR). To falsify (ES*LR), there would have to be a case in which one's evidence supports that p, but in which it is not the case that someone in the same circumstances, with the same evidence, and the disposition to conform their beliefs to what their evidence supports would manifest that disposition by believing that p. But such a case is simply contradictory. If the evidence supports p, then the manifestation of the relevant disposition consists in believing that p. It would be contradictory to maintain that even though one's evidence supports that p, it is not the case that somebody with the same evidence and the relevant disposition does not manifest that disposition by believing that p. As things are set up, there is simply no room for a case in which the antecedent is true, but not the consequent. Thus (ES*LR) is true.

The same kind of argument can be given for (ES*RL). To falsify (ES*RL), there would have to be a case with the following four features. First, there is someone in the same circumstances as me. Second, this someone has the same evidence as me. Third, this someone has the disposition to conform beliefs to the evidence and a manifestation of this disposition in these circumstances would consist in believing that p. And fourth, it would be that my evidence does not support that p. But this fourth feature turns this into a contradictory case, as the second feature stipulates that the evidence is the same and the evidence must support p, since the antecedent of (ES*RL) holds that the manifestation of the relevant disposition consists in believing that p. Thus (ES*RL) is true.

Since both (ES*LR) and (ES*RL) are true, (ES*) is true. And since (ES*) is true, rational _disp and rational _cont do not come apart, which means that the notion of rational belief is unambiguous between these two readings.

4. OBJECTIONS AND UPSHOTS

Before I turn to some of the upshots for other debates in epistemology, I will consider an objection to (ES)* and (DMS). The simplicity of the argument for (ES)* presented in the last section could suggest that (ES)* is trivially true. If overlooked, even trivial truths can be worth pointing out. However, the objection continues, the triviality of (ES)* is due to (DMS). The problem with (DMS) is that it is uninteresting and useless in comparison to its predecessor (DS).Footnote ⁵

Behind this charge of uselessness lies the following worry. Williamson claims that we sometimes have better access to what a moral or rational person would do than to the demands of the abstract standards of morality or rationality (p. 264). Hence, when we ask the question “What is rational to believe?”, it can be useful to ask what the rational person would believe. This accessibility advantage makes (DS) interesting and useful and what earns it its keep even if it were equivalent to (CS). The worry is that it is not obvious that the same can be said for (DMS). Why bother with the question whether a person would manifest a disposition to conform their beliefs to the evidence by believing instead of focusing on the question what the evidence supports directly? While (DMS) helps to make (ES)* true, it does so at the cost of robbing the notion rational _disp of significance.

While this objection needs to be addressed, it does not jeopardize what I intended to argue for in this paper. My aim was to argue that even if we adopt both notions, rational _disp and rational _cont, once we have a proper understanding of the former, Williamson's claim of the ambiguity of rationality cannot be maintained. Keeping this aim in mind, there are (at least) two ways to respond to the objection that I want to consider.

First, with my aim in mind, one could simply accept the objection because the aim was not to defend a version of rational _disp that has the alleged accessibility advantage. Suppose it turns out that no account of rational _disp that does not succumb to the problems for (DS) pointed out in section 2 has the alleged accessibility advantage. This is bad news for those who endorse a version of rational _disp because of its alleged advantages. It is not bad news for the defender of the unambiguousness of rationality. In fact, if it was true that no plausible account of rational _disp adds anything over and above (CS), then an equivalence of both rational _disp and rational _cont seems guaranteed, and hence rationality is not an ambiguous notion.

Second, there is a less conceding reply, which I prefer, and which comes in two steps. Before I spell this reply out, here are a few words on why I opted for (DMS). (DMS) is as closely modeled on Williamson's (DS) as possible. My concern with (DS) is that it overlooks that we are not interested in what somebody with a disposition would do under any circumstances, but rather only under circumstances in which the disposition is manifested. In order to fix this while staying as close as possible to (DS), as to not beg the question against Williamson or other critics, I went for (DMS).Footnote ⁶

With this in mind, let's turn to the first step of the reply, which is a reminder of the key idea behind rational _disp. As pointed out in section 2, when we ask ‘What would Jesus do?’, we are not asking what would Jesus do under just any circumstances, but rather just those in which Jesus manifests his dispositions that make him a moral exemplar. Thus, it is the manifestation question that we are after anyway. In light of this, the above skepticism about (DMS) seems to be a general skepticism about the usefulness of the key idea behind rational _disp. Since both (DS) and (DMS) are simply different attempts of making good on the key idea, we should be equally suspicious of (DS) as of (DMS).

However, this is the second step, what can be said in favor of the usefulness of (DS) can be said of (DMS) as well. Before introducing (DS), Williamson provides a brief argument in favor of the usefulness of the notion of rational _disp. He holds that asking what the rational person would believe brings in the cognitive power of the imagination in a way that it does not get involved if we ask the more abstract question what the evidence supports. Since our imaginative grip on what the rational person would believe may be firmer than our grip on the abstract question of what the evidence supports (p. 264), rational _disp can be a useful notion. If we accept this as a motivation for (DS), we should accept it to motivate (DMS), too. That is because, as pointed out before, we are interested in what somebody would believe if they manifest the disposition, not in what somebody with the disposition would believe when they do not manifest it. Hence, if the imagination is of any help, it must be of help in answering the question what somebody would believe if they manifested the relevant disposition. Consider an additional example with another role model. When I ask “What would Sherlock Holmes believe if he had my evidence?”, I wonder what Holmes would believe if he manifested his disposition to conform his beliefs to the evidence. And it seems possible that the imagination could be of help in answering my question. It might give me an idea how Holmes would go about assessing the evidence and what it supports. Consequently, it seems that the powers of imagination can provide the very same accessibility advantage for (DMS) that they confer to (DS) according to Williamson. Hence, the worry behind the objection can be assuaged.

Of course, one can be skeptical about whether the imagination provides said accessibility advantage. This is not the place to settle this question. But even if it were answered negatively, this does not undermine my aim. One could then go back to the first reply. All this would show is that the notion of rational _disp does not have the alleged accessibility advantage. It does not show that there is an ambiguity in the underlying notion of rationality.

Let us now turn to one upshot. Williamson argued that the assumption of (ES) has distorted the debate about evidence in epistemology. Now that I have replaced (ES) with (ES*), it is natural to ask about potential consequences for the evidence debate. I think that there are none. At best, we can show how to make progress in this debate.

(ES*) is compatible with various accounts of the nature of evidence. It is compatible with Williamson's view that E = K. But it is also compatible with the phenomenal conception of evidence, which denies E = K and holds that one's evidence consists of one's non-factive mental states. If one assumes that E = K, then the BIV's impression that Jones’ fingerprints are on the gun simply does not amount to evidence. And if that impression is all the BIV has supporting its belief that Jones killed Smith, then this belief is not supported by the evidence. But then somebody with the same evidence, or better lack thereof, if she were to manifest her disposition to conform her beliefs to the evidence, would therefore not believe that Jones killed Smith. All that follows from that is that it is not rational for the BIV to believe Smith killed Jones. If one assumes that the phenomenal conception is true and that the BIV's impression is evidence for believing that Jones killed Smith, then somebody with the same evidence and manifesting her disposition to believe in accordance with the evidence would also believe that Jones killed Smith. Since (ES*) is compatible with both of these contrary views of evidence, it cannot be used to decide the debate for either side.

But if rationality is not ambiguous and (CS) is a legitimate schema, then this gives us an idea on how to proceed in the debate. (CS) ties evidence to what is rational to believe. But then the battleground about the nature of evidence extends to what is rational to believe. To defend E = K, one ought to defend the resulting theory of rational belief that (CS) commits one to. Williamson may do so by arguing that following good cognitive instincts, as the BIV does, is not sufficient for rational belief. Likewise, opponents to E = K should defend that following good cognitive instincts suffices for rational belief.

If there is no ambiguity of rationality, then there is room for a genuine debate here. I think that this should be a welcome consequence for both sides to the debate. It favors neither side over the other, but it means that there is at bottom a single question both sides disagree on and that the current debate was not an unfortunate instance of talking past each other.