2. Disagreement and Three Responses

In this section, I will explain the problem and discuss some arguments for and against the three responses mentioned above. Let us start with the following well-known example from Christensen (Reference Christensen2007):

Setting up the case right involves adding that my friend and I know that we are epistemic peers, that is, we have the same evidence and are equally reliable. The obvious answer Christensen arrives at is:

Judgment-calibrationism delivers Christensen’s answer. Recall:

p is “my share is $43.” We can assume that the expected degree of reliability is 50% given that both agents are epistemic peers and setting aside the possibility that you are both wrong. The result is that the rational credence to assign to “my share is $43” is 50%.

Let us pause to note that the term “judgment” in Judgment-calibrationism is used in a technical way, as the proposition the agent regards as most likely to be correct on the basis of first-order evidence alone.Footnote ⁷ This use of “judgment” presupposes that we can neatly divide up first-order and higher-order evidence. This is contestable,Footnote ⁸ but for the purposes of our discussion, it is a harmless simplification. We can focus on cases like Restaurant, which can be divided into two stages—a first stage where the agent works through a specific set of evidence, for example, the contents of a bill, and a second stage in which she learns the judgment of a peer.

I take it that Judgment-calibrationism provides an intuitive answer. In light of the disagreement, it would be unacceptably dogmatic for either agent to maintain a strong belief in their original answer.

However, Judgment-calibrationism has two odd implications which Kelly (Reference Kelly, Feldman and Warfield2010) drew attention to and which continue to be a source of controversy.Footnote ⁹ The first oddity is that Judgment-calibrationism says that the agent who initially judged correctly and the agent who initially judged incorrectly should make equally extensive revisions to their beliefs (Kelly, Reference Kelly, Feldman and Warfield2010, p. 123). However, surely, says Kelly, there is an asymmetry. Should not the agent who initially judged correctly be required to make less extensive revisions than the agent who initially judged incorrectly?

The second oddity is that Judgment-calibrationism says that the agent who initially judged incorrectly becomes rational by following Judgment-calibrationism and assigning r to p. To see this, recall that the consequent of Judgment-calibrationism is “the rational credence for you to assign to p is r.” Judgment-calibrationism can be understood as a sufficiency condition for rationality, that is, the agent that assigns r to p becomes rational. However, surely it should not be that easy to become rational. After all, the agent who made the incorrect judgment was irrational. How could this irrationality drop out of the picture? Indeed, how could the evidence, which supports the belief that each owes $43, drop out of the picture?

These challenges might be answered by replacing Judgment-calibrationism with:

Evidence-calibrationism cannot be followed by the person who initially judges incorrectly, as they will be wrong about which proposition is best supported by their first-order evidence. So, Evidence-calibrationism will not deliver the odd verdicts that both agents should make equally extensive revision, nor that an agent that judges incorrectly can become rational by following Evidence-calibrationism.

However, the problem now is that following Evidence-calibrationism seems to require having access to which proposition is best supported by your first-order evidence, and if the agent has access to this, then they should believe that proposition (the one best supported by their first-order evidence) and ignore their expected degree of reliability.

Given these problems for Calibrationism, we might resort to:

This says that you should believe your share is $43 if $43 is the result of correctly doing the arithmetic, and you should believe your share is $45 if $45 is the result of correctly doing the arithmetic.

The problem here is that Right Reason seems to imply that you should ignore relevant evidence. It seems unacceptably dogmatic to receive evidence from a peer that you have made a mistake, and respond by refusing to shift your opinion one iota.

The right reason theorist will say that the person who got the correct answer is right not to shift opinion. They might emphasize that it is only the person who has done the arithmetic correctly who should stick with their opinion—the person who made a mistake should reconsider their opinion. However, this is puzzling, as neither agent knows whether they made a mistake. Their epistemic positions are symmetrical in this key sense. In my view, we can see that there is something right about all three responses once we embrace contextualism.

3. Contextualism

In this section, I will explain the core ideas of normative 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 in the context of discussing ordinary people, but false in the context of discussing basketball players.

A popular theory in linguistics is that normative terms like “ought” and “reason” are context-sensitive in a similar way. The traditional Kratzerian semantics posits two parameters which are determined by the conversational context (Kratzer, Reference Kratzer1981).Footnote ¹⁰

The first parameter is a modal base, which corresponds to a proposition, a set of information, or a set of live (partialFootnote ¹¹) possible worlds. “It must be that P’ means, roughly, that in all the live worlds, P.” The second parameter determines a ranking of the live worlds. “It ought to be that P” means roughly that in all the best live worlds, P. So “S ought to believe P” means, roughly, that S believes P in all the best live worlds. The modal base can be made explicit using “given,” for example, S ought to believe P given proposition B.

Candidate values for the second parameter include a moral ranking, which ranks worlds in order of how morally good they are, and an epistemic ranking, which ranks worlds in order of how epistemically good the beliefs/credences are. In this paper, we will focus exclusively on the epistemic ranking. The first parameter, the modal base, is the only moving part we need.

What are the candidates to be the modal base? The mechanics allow any proposition, but we will focus on four: all and only the true propositions; all and only the propositions believed; all and only the propositions supported by the first-order evidence; all and only the propositions supported by the total evidence. These correspond to the following “oughts”:Footnote ¹²

To make these explicit, instead of using “ought” (e.g., S ought to believe P), we can use one of these four (e.g., “S truth-relative-ought to believe P”). I will talk about these being used in conversation, but of course, they are all pronounced the same in English.

The familiar distinction between the objective and subjective “ought” can be plugged into this framework.Footnote ¹³ Suppose there will be dancing at a party, and Dancing Dave loves dancing while Talking Tim hates dancing. Dancing Dave correctly believes that there will be dancing at the party, while Talking Tim falsely believes that there will be no dancing, only talking. There are senses in which both ought to go to the party, but the “oughts” seem to be of different types. We can say that Dave objectively ought to go to the party because there will be dancing, but Tim only subjectively ought to go to the party because he only believes there will be dancing. As Schroeder (Reference Schroeder2004, p. 348) writes, “On a natural view, subjective reasons are simply things that you believe such that, if they are true, they are reasons for you to do something.” The objective ought will correspond to our truth-relative ought; the subjective ought identified by Schroeder will correspond to our belief-relative ought.

What determines the modal base? I suggest that part of what determines the modal base is the aim of the conversation. Several writers have distinguished two aims (among others) that people might have when using “ought” or other normative terms:Footnote ¹⁴

1. i) Expressing standards.

2. ii) Deliberating.

These most clearly come apart when the agent is missing important information. To take a well-known example, suppose there is petrol in Bernard’s glass but he believes, with good reason, that it contains gin.Footnote ¹⁵ What ought he to do? It is very plausible that there is no univocal answer—there is a sense in which he ought to drink and a sense in which he ought not to. These fit with the aims of expressing standards and deliberating which we can make explicit with “standards-ought” and “deliberative-ought,” that is, he standards-ought not to drink and deliberatively-ought to drink.Footnote ¹⁶ We get these results if the modal base in contexts where standards are expressed includes information the agent does not have (e.g., that the glass contains petrol), while the modal base in deliberative contexts includes only information the agent has.

We can make more fine-grained distinctions to recover the four senses of “ought” above. Starting with (i), we can distinguish two standards. One standard is to believe all and only truths.Footnote ¹⁷ One might say “one ought to believe all and only truths” and we can make sense of this “ought” as the truth-relative ought.

However, as we often do not have full information, in most contexts it is not appropriate to simply say that you ought to believe all and only truths. This leads to a second standard, which is to form the appropriate beliefs given the first-order evidence. We can think of this as the standard of not making any mistakes in reasoning and ignoring any higher-order evidence. (Higher-order evidence can only be misleading given that one has formed the appropriate beliefs given the first-order evidence.) Such contexts invoke the first-order-evidence-relative ought.

Moving to (ii), we can distinguish two types of deliberative context, which I will call a prospective-deliberative context and a retrospective-deliberative context.

In some deliberative contexts, the question is how the agent should proceed, holding fixed their current beliefs. The agent might have made mistakes in the past, but the question is how to proceed given what they currently believe. Call this a prospective-deliberative context. Such contexts invoke the belief-relative ought.

In other deliberative contexts, current beliefs are not held fixed. Agents have the time and cognitive capacity to reconsider their beliefs in the light of their total evidence. Call this a retrospective-deliberative context. Such contexts invoke the total-evidence-relative ought.

I suggest that these oughts allow us to identify what is true in Right Reason, Judgment-calibrationism and Evidence-calibrationism as follows:

1. a) Right reason is true for first-order-evidence-relative ought

2. b) Judgment-calibrationism is true for belief-relative ought

3. c) Evidence-calibrationism is true for total-evidence-relative ought

I will argue in the next three sections, respectively, that these principles explain the conflicting intuitions.

4. Right Reason is True for First-Order-Relative-Ought

Recall:

This says that the person who did the arithmetic correctly should ignore the evidence that their peer came to a different answer. Recall that the main objection was that this seems to ignore relevant evidence.

However, if the context invokes the standard of forming the appropriate beliefs given the first-order evidence, then we get (a):

There can be no debate over the truth of Right Reason First-Order Evidence. It is true by definition. I will argue that the arguments and intuitions of those who defend Right Reason are best understood as invoking the first-order-evidence-relative ought and Right Reasons First-Order Evidence. We will look at Field (2000), Titelbaum (Reference Titelbaum2015), and Weatherson (Reference Weatherson2019).Footnote ¹⁸

Hartry Field (2000) gave an early defense of what was later called Right Reason. His specific claim was that a priori justification is indefeasible by empirical evidence. In response to the argument that one could get evidence against propositions that have a priori justification, Field wrote:

It is not entirely clear what “ideal credibility” means, but one possibility is that it expresses a standard. In a context in which we are discussing standards, we might say that the agent ought to make no mistakes in their reasoning and ignore higher-order evidence, so first-order-evidence-relative-ought is relevant. Right Reason will therefore be true because it will be understood as Right Reason First-Order Evidence. So Field’s defense of Right Reason can be understood as a defense of Right Reason First-Order Evidence.

Relatedly, Titelbaum’s (Reference Titelbaum2015) defense of Right Reason relies on:

In response, Claire Field (Reference Field2019, p. 179) writes, “Assets requires some explanation. Titelbaum provides no such explanation when he introduces the claim, and as it stands, it is somewhat surprising. More recent advocates of a priori justification have typically thought of it as defeasible at best (see BonJour, 1998).” I agree with Field that Assets is not very plausible. The point I want to make here is that, to the extent that I can get myself to see how someone might accept Assets, it seems most plausible as expressing a standard. Thus, Right Reason is most plausible if understood as Right Reason First-Order Evidence.

Finally, Weatherson (Reference Weatherson2019) offers a book length defense of a Right Reason style theory. He distinguishes making normative claims about actions from giving normative advice. Put in our terms, ‘making normative claims’ seems to be relevant to the aim of expressing standards and ‘giving normative advice’ seems to be relevant to the aim of deliberating (or helping someone else deliebrate). In considering the view he opposes, according to which normative claims depend on the agent’s higher-order evidence, he writes:

It is not entirely clear what he means in saying that his theory is “least strongly opposed” to the view about advice. However, one plausible reading is that Weatherson is least confident that Right Reason is correct when it comes to advice. In addition, this would fit with my view that Right Reason is true in some contexts where agents express standards, but not true in contexts where agents deliberate.

So far I have argued that Right Reason can be motivated by thinking about the first-order-evidence-relative-ought. I add that it can also be motivated by thinking about the truth-relative-ought. For example, in Restaurant, the first-order evidence has guided the agent to the truth. So, in a context in which the relevant “ought” is what one truth-relative-ought to believe, Right Reason delivers the correct verdict. The agent truth-relative-ought to believe they owe $43; the agent also first-order-evidence-relative-ought to believe they owe $43. These “oughts” do not diverge in this example. This bolsters my argument that defenders of Right Reason can be understood to have in mind contexts where standards are being expressed—either the standard of forming the appropriate beliefs given the first-order evidence, or the standard of believing all and only truths.Footnote ¹⁹

5. Judgment-Calibrationism is True for Belief-Relative-Ought

In this section, I will argue that Judgment-calibrationism is true for contexts which invoke prospective deliberation, and that this explains calibrationist intuitions. Recall:

(b) says this is true for the belief-relative-ought:

Which set of beliefs are relevant? Contextualism offers a natural answer—the beliefs referred to in the antecedent.Footnote ²⁰ That is, we start with the belief/judgment that p and the expected reliability of r, then ask what the agent ought to believe given that starting point. In the restaurant case, the result is that the agent belief-relative-ought to assign 50% credence to their share being $43. Judgment-calibrationism Belief-relative sets aside whether the initial beliefs referred to in the antecedent are rational and simply tells us what to do given that we have them.

This fits with contexts that invoke prospective deliberation (which holds fixed current beliefs). We can create a context that invokes prospective deliberation by asking what you, the reader, would do in Restaurant. Imagine being in the situation described in Restaurant. What procedure would you follow? Notice you have not been told how you arrived at $43, so it is impossible to reconsider the way in which you arrived at that answer. I suggest that the most plausible procedure to follow is Judgment-calibrationism Belief-relative. Thus, we can explain intuitions in favor of Judgment-calibrationism Belief-relative as invoking contexts of prospective deliberation.Footnote ²¹

Of course, the proponent of Right Reason will reply that the agent does have access to which proposition is best supported by the evidence. They will claim that the first-order evidence is undefeated. My point is that this move is less plausible when we invoke a deliberative-ought than it is when we invoke a standards-ought. We can accommodate the intuitions of the Right Reason theorist without being committed to the view that Right Reason should be used in deliberation.Footnote ²²

Relatedly, the proponent of Right Reason might complain that the evidence that supports $43 should not have dropped out of the picture. Even in a deliberative context, why shouldn’t the agent use the total evidence, and why shouldn’t that total evidence support for the answer of $43? In response, the Calibrationist need not concede that the evidence that supports $43 has dropped out of the picture. Quite the contrary—the evidence for $43 provided the support for the agent’s initial judgment that they each owed $43. The evidence is still there, supporting this belief.

To be clear, it is not my aim to show that the Calibrationist is right about this. My aim is to show that the intuitions motivating Calibrationists (and Right Reason theorists) can be explained by a contextualist framework.

How does Judgment-calibrationism Belief-relative fare against Kelly’s objections? Kelly’s first objection was that Judgment-calibrationism misses the fact that an agent who irrationally judges that p should be required to make more extensive revisions to their opinions than an agent who rationally judges that p.

In response, Judgment-calibrationism Belief-relative does say that both agents are required to make equally extensive revisions to their opinions, but that is because we are starting with their judgments and asking how to proceed from there. Any mistake an irrational agent made in working out the bill has fed into their judgment. If we hold fixed those judgments and ask how to proceed from them, each should make equally extensive revisions.

Kelly’s second objection was that an agent who irrationally judges does not become rational merely by calibrating. Yet Calibrationism seems to imply that they do.

In response, Judgment-calibrationism Belief-relative does not make any claim about the overall rationality of the agent, so it does not provide a way of becoming rational. It says what the agent belief-relative-ought to believe given their prior beliefs, but does not offer a judgment on the rationality of those prior beliefs, so it does not offer a judgment on their overall rationality. The belief-relative-ought only offers a judgment on belief-relative rationality.Footnote ²³

Let us consider an objection to Judgment-calibrationism from David Christensen (Reference Christensen2016).Footnote ²⁴ If solid, it would be just as serious an objection to Judgment-calibrationism Belief-Relative. Christensen argues that Judgment-calibrationism is incomplete. The problem is that Judgment-calibrationism refers to an agent’s judgment, but there are cases where the agent has not formed a judgment. Christensen writes, “there are cases where agents have not formed any initial credence on the basis of a batch of first-order evidence before they get the higher-order evidence” (p. 406). And he points out that there are cases where the agent is told of a peer’s opinion before thinking through the first-order evidence. In such cases, the agent will never form a credence on the basis of just the first-order evidence. Christensen uses this to motivate the transition from Judgment-calibrationism to Evidence-calibrationism.

However, I don’t think this is a fatal objection, for an agent can work through what the first-order evidence seems to support, even if they believe they are unreliable. They can then take into account their believed unreliability to arrive at a considered view. Compare a case where you are first told that your color perception has been inverted. Then you appear to see a blue table. You can judge that your first-order evidence seems to support that the table is blue, then take into account your believed unreliability and move to the considered verdict that the table is orange. One can go through a similar process in cases like Restaurant. Recall, the term “judgment” in Calibrationism is used in a technical way, as the proposition the agent regards as most likely to be correct on the basis of first-order evidence alone. You might first get your peer’s opinion, then work out what the first-order evidence seems to show, then come to a considered judgment based on these two pieces of evidence in the manner suggested by Judgment-calibrationism.

6. Evidence-Calibrationism is True for Total-Evidence-Relative-Ought

Let us now turn to:

Evidence-calibrationism seems to have been motivated largely by the problems with Judgment-calibrationism (Christensen, Reference Christensen2016). In light of my defense of a version of Judgment-calibrationism, this motivation is less compelling. Still, what should we say about it?

Contextualists can allow that there is a reading that makes Evidence-calibrationism true. It is most plausible with the Total-evidence-relative-ought:

Is Evidence-calibrationism Total-evidence a principle of deliberation or a standard? Perhaps both.

Starting with deliberation, I suggested at the end of Section 3 that when we create a context that invokes retrospective deliberation (which does not hold current beliefs fixed), Evidence-calibrationism Total-evidence is intuitive. That is, when we do not hold the agent’s beliefs fixed and allow them to consider afresh what they should believe, Evidence-calibrationism Total-evidence seems like the correct rule for deliberation. Unlike Judgment-calibrationism Total-evidence, Evidence-calibrationism Total-evidence builds in the correctness condition that p is the proposition best supported by the first-order evidence, making it more demanding and harder to follow. Nevertheless, perhaps it could still be used in deliberation. If not, Evidence-calibrationism Total-evidence can be thought of as expressing a standard.

Evidence-calibrationism Total-evidence seems intuitive when the relevant standard is: believing what the total evidence supports, while allowing that you might have made a mistake. It tells the agent to come to the right judgment about what the first-order evidence supports, but still take into account that you might have made a mistake about what the first-order evidence supports.

Someone might object that Evidence-calibrationism Total-evidence seems imperfect both as a principle of deliberation and as a standard. I am happy to concede the point. Perhaps Evidence-calibrationism Total-evidence occupies an awkward halfway house between a principle for expressing standards and a principle for deliberation. Still, I see no reason to exclude this kind of “ought” from the language.

7. Contextualism and the Disappearance of Disagreement

A familiar objection is that contextualism wrongly predicts that there is no genuine disagreement between proponents of Right Reason, Judgment-calibrationism and Evidence-calibrationism. Footnote ²⁵ Once the parameters are made explicit, each participant in the debate should accept what the other is saying and disagreement disappears—and that looks like the wrong verdict.Footnote ²⁶

A first response is that it would not be so surprising to find no genuine disagreement. It is common for people to appear to disagree, only to find out that they have been using language slightly differently. Chalmers (Reference Chalmers2011) argues that this is a typical feature of philosophical debates. In addition, when it comes to normative terms like “ought,” we even have a widely accepted linguistic theory that tells us where the hidden parameters are and how they might lead to the appearance of disagreement.

Nevertheless, it would be unsatisfying to simply say that all participants in the debate are just confused by language and talking past each other. The challenge for the contextualist is to make sense of the intuition that there is genuine disagreement. Let me offer four compatible ways to retain disagreement.

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. Perhaps only one “ought” has normative authority. Indeed, Broome (Reference Broome2013, p. 24) talks about the “central ought” and Kiesewetter (Reference Kiesewetter2017, p. 9) talks about the “deliberative ought,” and these might be taken to be attempts to refer to one privileged “ought” with normative authority.Footnote ²⁷ Disagreement thus remains when speakers disagree about which “ought” has normative authority.

Someone might object that contextualism would be much less interesting if there were a single “ought” with normative authority. And indeed it would be. However, contextualism reduces the motivation for the thesis that there is a normatively authoritative “ought.” If there were a normatively authoritative “ought,” then we would face tricky metaphysical (what makes an “ought” privileged?) and epistemic questions (how do we know which is privileged?). Dispensing with a normatively authoritative “ought” allows us to avoid such questions and to explain apparent disagreements by appealing to different speakers using different parameters. What’s not to like?

Furthermore, 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. So, genuine disagreement is compatible with the absence of a privileged “ought.”

The second way to retain disagreement is to hold that sometimes disagreement involving contextualist terms should be understood as disagreement about which parameters to use. To motivate this view, note that contextualism says that the parameters are determined by the conversational context, and which parameters are operative might be in flux in a conversation. For example, consider the following conversation when all parties know that Michael Jordan is 1.98 m tall and Sun Mingming is 2.45 m tall:

1. A. Michael Jordan is tall

2. B. No he isn’t, Sun Mingming is tall

It is plausible that this conversation is best understood as an implicit negotiation about the standards for the word “tall.”

Lewis (1979; 1996) suggests that conversations should conform to the principle of accommodation, which 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.” Thus, B is attempting to shift the conversational score to one in which the relevant sense of “tall” is “tall-for-a-basketball player.” Applied to our example, defenders of Right Reason are arguing that the best norm of belief in some context is Right Reason and defenders of Judgment-calibrationism are arguing that the best norm of belief in some context is Judgment-calibrationism.

However, what if the hearer does not want to change the conversational score? Then, we get what Plunkett and Sundell (2013) call meta-linguistic negotiation, that is, an exchange in which speakers tacitly negotiate the proper deployment of a linguistic expression in a given context. There is no guarantee that there will be a successful conclusion to the metalinguistic negotiation, resulting in an unresolved disagreement.

This differs from the first way to retain disagreement because the speakers need not believe that any parameter is metaphysically privileged. They might still disagree about the aim of the conversation. That is, one person wants a conversation about the standards, and another wants a conversation about deliberation. This framework allows for genuine disagreement about which context to create.

This leads to the third way to retain disagreement—there might be disagreement about what to do. Björnsson and Finlay (2010) argue that the hidden goal of many conversations involving contextualist terms is to establish what to do. Thus, there need be no disagreement over which propositions are true, but still disagreement over what to do.Footnote ²⁸ (If determining the parameters is something we do, then the second way to retain disagreement could be considered a version of this third way.)

The fourth way to retain disagreement is to separate the metaphysics from the epistemology.Footnote ²⁹ One might hold that the full conversational context determines the values of the parameters (metaphysical), but the inference (epistemic) from the full conversational context to the values of the parameters is nonobvious. For example, the full conversational context might be determined by the intentions of the speakers, but speakers’ intentions are not always transparent, so hearers might infer false conclusions about the context. This leaves room for genuine disagreement about which beliefs one “ought” to have.

Therefore, the contextualist can make sense of the intuition that there is genuine disagreement between proponents of Right Reason, Judgment-calibrationism, and Evidence-calibrationism. However, there may well be less disagreement than originally appeared, and the remaining disagreement might not be what it initially seemed to be.