2. Truth and rationality

Consider two hypotheses about what one ought to believe:

Suppose there is petrol in Bernard’s glass, but he has good reason to believe it contains gin.Footnote ⁵ Rationality says Bernard ought to believe his glass contains gin, and Truth says Bernard ought to believe his glass contains petrol. Thus, Rationality conflicts with Truth.Footnote ⁶ But there are strong arguments for both Truth and Rationality.

Inspiration for Truth can be traced to at least William James:

The list of proponents runs from Chisholm (Reference Chisholm1977) to Joyce (Reference Joyce1998), Goldman (Reference Goldman1999, Reference Goldman, Fairweather and Zagzebski2001, Reference Goldman2015), Wedgwood (Reference Wedgwood2002), and McHugh (Reference McHugh2012).Footnote ⁷

One powerful argument for Truth is that Rationality breaks the close connection we want between norms of belief and truth. As Gibbons puts it:

Whereas Truth says that in believing that it is raining when it is not, you are not believing as you ought to.

On the other hand, a powerful argument for Rationality is that Truth breaks the close connection we want between norms of belief and guidance: Footnote ⁹

We cannot always be guided by Truth, but, the thought goes, we can always be guided by Rationality. If we add that genuine norms must always be able to provide guidance, it follows that Truth is not a genuine norm.

Most epistemologists in this debate defend one of Truth and Rationality and reject the other.Footnote ¹⁰ An alternative approach, recently defended by Hughes (Reference Hughes2019, Reference Hughes2021a, 2021b, forthcomin), is to defend both of Truth and Rationality. He argues that each places an incompatible demand on us, with the result that we face epistemic dilemmas. I also defend both Truth and Rationality, but I don’t think we face epistemic dilemmas. Instead, each of Truth and Rationality is true in specific contexts; there are no contexts where both are true and so no epistemic dilemmas. I return to Hughes in Section 6.

3. Contextualism

It is a familiar thought that whether someone is correctly described as tall depends on the details of the conversation. For example, Michael Jordan, at 1.98 m, is tall for an ordinary person, but not tall for a basketball player. So the truth of ‘Michael Jordan is tall’ depends on the conversational context. It is true given a context in which ordinary people are being discussed but false given a context in which basketball players are being discussed. So the word ‘tall’ has hidden parameters that are filled by the conversational context. As the meaning of ‘tall’ depends on the conversational context, the word is context-sensitive.

The standard view in linguistics is that normative terms like ‘ought’ are context-sensitive in a similar way. They have hidden parameters that are fixed by the conversational context, resulting in different semantic values in different contexts. The standard view is that there are at least two parameters needed to fix the content of ‘ought’ – a set of live possibilities and a standard.Footnote ¹¹

The first parameter, the live possibilities (called the modal base), are the possibilities treated as live in the conversation. For example, suppose we are discussing whether you ought to take an umbrella when you leave the house. If rain is a live possibility, then you ought to; if rain is not a live possibility, then it is not the case that you ought to. If I tell you that the forecast is for fine weather, this can be taken as eliminating the possibility in which it rains, with the result that it is not the case that you ought to take an umbrella.

The second parameter is a standard or goal that determines an ordering of the live possible worlds. The standard proposal is that ‘S ought to A’ is true iff S A’s in every live world at the top of the ranking. Assuming the goal is to stay dry, then, if there is a live possible world in which it rains, the highest ranked worlds will be those in which you take an umbrella. It follows that you ought to take an umbrella. But things change if the goal changes. Perhaps you love getting wet, so the goal is to get wet. Now the highest ranked worlds will include those in which you get wet, and it will not follow that you ought to take an umbrella.

The standard need not be one that the subject cares about. If I say ‘you ought to start with the cutlery on the outer edge’, the standard might be the rules of etiquette. The more explicit sentence is ‘by standards of etiquette, you ought to start with the cutlery on the outer edge’. This sentence can remain true even if you don’t care about etiquette. This allows us to say to the psychopath ‘you shouldn’t kill people’; the full sentence is ‘by standards of morality, you shouldn’t kill people’, which is true even if the psychopath doesn’t care about morality. More importantly for us, it allows us to say ‘you should believe things that are true’ even to someone who doesn’t care about the truth.

For our purposes, we can distinguish two goals, which might be relevant to a conversation about what someone ought to do or believe. The goal might be to:

1. a) set a standard or

2. b) state what is rational.

A similar distinction can be found in many places: Bales (Reference Bales1971) distinguishes decision procedures from right-making characteristics; Schroeder (Reference Schroeder2011, p. 1–2) distinguishes deliberative from evaluative oughts; McHugh (Reference McHugh2012, p. 9–10) distinguishes prescriptive from evaluative norms; Schoenfield (Reference Schoenfield2018, p. 690) distinguishes plans to make from procedures to conform to; and Steinberger (Reference Steinberger2019, p. 7) distinguishes directives that guide from evaluations which set standards. The distinction is needed in cases where agents aren’t sure what is best. For example, consequentialists can say that agents ought – in the standard-setting sense – to perform the action that brings about the best consequences, and they ought – in the rationality sense – to perform the action they think will bring about the best consequences.

Returning to norms of belief, the contextualist view is that there are two goals, corresponding to two meanings of ‘ought’ and two norms. In the standard-setting sense, agents ought to believe what is true, and in the rationality sense, agents ought to believe what is rational. Which sense is relevant depends on the context:

It will be useful to make explicit the sense of ‘ought’ being used with ‘veritistically-ought’ and ‘rationally-ought’:

And we can say that the context in which ‘veritistically-ought’ is relevant is a veritistic context, and a context in which rationally-ought is relevant is a rational context.

The question of precisely what fixes the values of these two parameters is a difficult open question. The contextualist says that it is determined by features of the conversation. But what features? Plausible candidates include the interests, goals, evidence, and beliefs of the participants in the conversation. Two plausible principles for turning these ideas into something more concrete are the principles of charity and the principle of accommodation (Lewis Reference Lewis1979, Reference Lewis1996). The principle of charity holds that we should interpret people so what they say makes a sensible contribution to the conversation. The principle of accommodation says that when a speaker makes an utterance involving a context-sensitive term, the parameters shift to make the sentence true. In Lewis’s terms, the utterance changes the ’conversational score’. In some cases, the participants in the conversation might disagree about which parameters to use. In such a case, they might engage in a process of metalinguistic negotiation in an attempt to resolve the disagreement. I’ll return to metalinguistic negotiation in Section 7.

Technically, from now on, I should semantically ascend and only mention ‘ought’ rather than using it. I will balance this with readability. I think the sloppiness used and apologized for by David Lewis (Reference Lewis1996, p. 566–7) helps beginners even if it irritates experts.

I note in passing that contextualism explains the debate about whether ‘ought’ is subjective or objective (see fn. 3). The contextualist can (i) associate the objective ought with the goal of believing the truth and a modal base that includes all the facts and (ii) associate the subjective ought with the goal of believing what is rational and a modal base consisting of the agent’s evidence or beliefs. The subjective and objective ‘oughts’ are two locations in the wider space opened up by contextualism.

Now that we have this machinery on the table, I will argue that the problems regarding the norms of belief can be solved.

4. Truth and rationality

Let’s work through a couple of examples. First, recall the case of Bernard, who reasonably believes he has gin in his glass, when in fact it is petrol. Truth and Rationality give different verdicts about what Bernard ought to believe. Rationality says he ought to believe there is gin in his glass; Truth says he ought to believe there is petrol in his glass.

The contextualist says either can be true depending on the context. We can make this explicit with ‘rationally-ought’ and ‘veritistically-ought’:

To put the point in terms of the contextualist semantics, the ranking of the worlds changes in different contexts (see Figure 1). In a veritistic context, the worlds are ranked by the ratio of true to false beliefs, so the highest ranking world is the one in which Bernard believes there is petrol in his glass (left); in a rational context, the worlds are ranked by how well beliefs fit the evidence (or whatever determines rationality), so the highest ranking world is the one in which Bernard believes there is gin in his glass (right).

Figure 1. Two rankings for belief.

For another example, consider a question on a student S’s exam: ‘what is the capital of Russia?’ As the answer is Moscow, there is a sense in which S ought to believe it is Moscow. And this remains true even if they have no idea what the correct answer is. Suppose S was sick for a month and missed all the work on Russia. Consider the following dialogue:

The contextualist can explain that the headteacher is invoking a context in which the student ought to believe what is trueFootnote ¹² :

The teacher shifts the context to one in which the highest ranked worlds are those where the student has rational beliefs:

Someone might object that there is no change in context and that the dialogue shows that the Headteacher was simply mistaken about what the student ought to believe. But the conversation might continue in ways that further strengthen the contextualist reading. For example, the headteacher might say:

The contextualist can understand this as a case in which the headteacher has not accommodated the context introduced by the teacher and in which Truth remains relevant.Footnote ¹⁴

5. The truth connection objection and guidance objection redux

Let’s return to the arguments of Section 2. The objection to Rationality was that it breaks the close connection we want between norms of belief and truth. One quick contextualist response is to point out that the connection to truth is still there – in some contexts, Truth is relevant, and obviously, Truth provides a connection to truth. But the complaint might continue that this doesn’t address the objection to Rationality. After all, given that beliefs should be closely connected to the truth, we need an explanation of how Rationality could be a norm of belief at all. As Gibbons put it, Rationality seems to make truth merely an optional extra for a belief.

However, the discussion above suggests a natural response – Truth sets the standard, but following Rationality is the best way to try to meet this standard. As we don’t have direct access to what is true, we cannot always follow Truth, but the best attempt for satisfying Truth is to follow Rationality. Thus, the objection that Rationality is not sufficiently connected to the truth can be answered by pointing out that Rationality is the best method we mere (non-omniscient) mortals have for meeting the standard set by Truth.Footnote ¹⁵

The objection to Truth was that genuine norms must always be able to provide guidance, but Truth cannot always provide guidance, so Truth is not a genuine norm. In response, if Truth sets standards, it is no objection to point out that it cannot always provide guidance. The point is analogous to that made by Bales (Reference Bales1971). It is no problem for consequentialism if agents do not always know which action will bring about the best consequences; consequentialism sets the standard for a right act. Similarly, it is not a problem for Truth if agents do not always know which propositions are true; Truth sets the standard for correct beliefs.

It is worth mentioning a further advantage of the thesis that Truth sets the standard rather than states what is rational. It has proven quite difficult to work out how strong Truth should be. The following have all been suggestedFootnote ¹⁶ :

What’s wrong with the straightforward bi-conditional? It seems to demand too much of us. There are truths that are inaccessible, or too complicated, or impossible for us to believe. However, if Truth is understood as setting the standard, these objections have no force.Footnote ¹⁷ It is not a problem if the standard demands too much of us in some cases. The standard doesn’t tell us what to do or give advice, it simply sets the standard for when a belief is correct. So if Truth sets the standard, we can defend Bi-conditional.

This completes my main argument. It remains to discuss two further objections – that even rationality cannot always guide and that it is implausible that disagreement disappears.

6. Can rationality always guide?

Hughes (Reference Hughes2021a) claims that Rationality cannot always provide guidance.Footnote ¹⁸ This seems to threaten my claim that the Guidance Objection to Rationality is dealt with by contextualism. I will offer a response that demonstrates the resources available to the contextualist.

Hughes argues that Rationality, like Truth, often fails to provide guidance.Footnote ¹⁹ He draws on the cognitive biases psychologists have discovered. To pick just one, everyone seems to have confirmation bias; that is, evidence that undermines our beliefs tends to be assigned too little weight, with the result that we do not change our minds as often as we should. Consider an agent who is unaware that they have this bias. They will systematically assign too little weight to evidence that undermines their beliefs, and they will have no way of knowing that they are doing this. So they are unable to rationally follow the evidence. This suggests that Rationality fails to give us the advertised benefit of a close connection between norms and guidance.

I suggest that we distinguish bounded rationality from ideal rationality.Footnote ²⁰ Ideal rationality does not take into account any cognitive limitations; bounded rationality does. This allows us to distinguish two versions of Contextualist Rationality:

Where an agent has confirmation bias, they will often be unable to do what is ideally rational.Footnote ²¹ But there will still be better or worse ways to respond to the evidence. An agent who shifts their beliefs in the right direction in the light of new evidence will be doing better than an agent who ignores all new evidence. In moving in the right direction they are satisfying the demands of bounded rationality. Where the conversational context takes into account the limitations of the agent, the relevant norm in the conversation is Contextualist Bounded Rationality.

I suggest that Contextualist Bounded Rationality can always provide guidance, allowing us to make peace with the Guidance Objection.Footnote ²² Of course, what exactly is boundedly rational is highly sensitive to the knowledge and abilities of the agent in their context, as well as the allowances that the speakers make. Leaving space for these complexities is one of the virtues of contextualism.

Someone might object that this distinction is ad hoc. But as well as being intuitive and defended by many others, it fits neatly with contextualist semantics. Recall that the context determines a set of possible worlds. These can be thought of as the relevant worlds. In an ideal context, a wide range of worlds are relevant i.e. worlds in which the agent has a wide range of beliefs, including the ideally rational beliefs. We can model a non-ideal (bounded rationality) context using a narrower range of relevant worlds. Making allowances for the limitations of the agent amounts to restricting the live possible worlds.

A different objection is that some biases are not rational in any sense of ‘rational’ and so not even boundedly rational. I agree. I do not say that all biases are boundedly rational; I say that some are boundedly rational. I am committed only to the view that there is always some norm of rationality that can guide an agent (even if they have cognitive limitations).

A counterexample to my thesis that some norm of rationality (e.g. Contextualist Bounded Rationality) can always provide guidance would be an agent who cannot be guided by any norm that deserves to be called a norm of rationality. And I think it is difficult to tell a coherent story of what that would look like. For example, consider an agent so benighted that, necessarily, they can only respond irrationally (not just non-ideally) to some piece of evidence; for example, they respond to seeing a white swan by increasing their credence that there are black swans. Note: It is not just that they do respond irrationally – they are only able to respond irrationality. I think we would be reluctant to ascribe beliefs to this agent at all. For example, the standard functionalist theory of beliefs (Lewis Reference Lewis1972) would say that beliefs are caused by relevant evidence; a mental state that is not sensitive to evidence fails to be a belief. An alternative theory is that beliefs are whatever would rationalize behaviour given experiences, in which case it might be metaphysically impossible for an agent who sees a white swan to infer that there are black swans (Dennett Reference Dennett1971, Reference Dennett1987).

7. Disagreement regained

Someone might object that it is implausible that there is no genuine disagreement in the debate about the norms of belief.Footnote ²³ The challenge for the contextualist is to make sense of the intuition that there is genuine disagreement. Let’s consider three ways in which a genuine disagreement might be retained.

First, even if statements about what one ought to believe are context-sensitive in English, there might be a metaphysically privileged parameter that picks out a normatively privileged property. Worsnip writes:

Worsnip is working with a primitive concept of ‘normative authority’. He is allowing that there might be lots of ‘oughts’, just as contextualism predicts, but that not all of them have normative authority. Indeed, perhaps only one ‘ought’ has normative authority.Footnote ²⁴ Disagreement thus remains when speakers disagree about which ‘ought’ is privileged. And if Truth and Rationality are both normatively authoritative, then we have Hughes’ view that there are epistemic dilemmas.

Still, I take it that contextualism reduces the motivation for the thesis that there is a normatively authoritative ‘ought’. Rather than facing tricky metaphysical (what makes an ‘ought’ privileged?) and epistemic (how do we know which is privileged?) questions, we can explain apparent disagreements by appealing to different speakers using different parameters.

And notice that there can be genuine disagreement even if there is no privileged ‘ought’. All that’s needed for genuine disagreement is differing beliefs about which ‘ought’ is privileged (even if none of them actually are privileged). Compare: If you think Santa Claus is married and I think he is a bachelor, we disagree, even though we are both wrong.

The second way to retain disagreement is to hold that the full conversational context determines (metaphysical) the values of the parameters, but the inference (epistemic) from the full conversational context to the values of the parameters is non-obvious. Just because the facts are determined doesn’t mean we will all figure out what they are. This leaves room for genuine disagreement about which context one is in and so genuine disagreement about which beliefs one ought to have.

The third way to retain disagreement is to hold that some instances of disagreement involve metalinguistic negotiation, that is, an exchange in which speakers tacitly negotiate the proper deployment of some linguistic expression in a context.Footnote ²⁵ For example, defenders of Truth argue that the best norm of belief in some context is Truth while defenders of Rationality argue that the best norm of belief in some context is Rationality. This differs from the first way to retain disagreement because speakers need not believe that any parameter is metaphysically privileged to disagree about the aim of the conversation e.g. one person wants a conversation about the standards and another wants a conversation about rationality. Contextualism allows for genuine disagreement about which context to create. So this is best understood as a practical disagreement about what to do; that is, should we create a veritistic context or a rational context?

So the contextualist can make sense of the intuition that there is genuine disagreement between proponents of Truth and proponents of Rationality. But there may well be less disagreement than originally appeared, and the remaining disagreement might not be what it initially seemed to be i.e. a disagreement about the correct unique universal norm.