Disagreement

Interview by Richard Marshall.

[Photo by Allegra Boverman]

'We found that as you vary the cultural distance between the speakers, people tended to agree less with the statement that at least one of the two speakers’ claims must be false. But we found no similar effect in judgments about whether the two speakers disagreed by making those claims – people continued to think that the two speakers disagreed in every case.'

'Consider the following situation. Amy is playing cards with Beth. I’ve seen both hands, and I saw that Amy’s hand is weaker than Beth’s. Amy folds her hand. I now say, “If Amy had called, she would have lost.” That seems right. But now suppose you were also on the scene, and knew that Amy was cheating and knew Beth’s hand. You might now say the following, “Amy knew Beth’s hand, so she would have called only if she (Amy) had the winning hand. Therefore, if Amy had called, she would have won.” 

My counterfactual seems true when I said it, but so does yours when you say it; yet, prima facie, mine and yours are contraries, so they can’t both be true. A natural conclusion to draw is that counterfactuals admit of two different interpretations.'

Justin Khoo's research is in philosophy of language and related issues in metaphysics. Here he discusses moral disagreement and contextualism, why disagreeing with someone doesn't have to mean you think the other person wrong, what that tells us about disagreement, modal disagreement, the distinction between indicatives and subjunctives, backtracking counterfactuals, Ramsey's thesis and its counterfactuals, and then he explains his ideas about how code-words like 'Inner'city' work to disguise toxic thoughts.

3:AM:What made you become a philosopher?

Justin Khoo:Philosophy was an accident. I started college as a music major, but quickly realized I didn’t love music enough to make a career of it (nor did I have the chops). After music school, I went out to California and, on a whim, signed up for a philosophy course at the Santa Rosa Junior College (taught by Michael Aparicio). Actually, I had signed up initially for two classes at the same time – Michael’s “Introduction to Philosophy,” and a course on rhetoric. After the first meeting, I asked Michael which class I should take. He told me, “Rhetoric will teach you how to win a debate, but in philosophy we are concerned not with winning debates, but with the truth.” I later wrote my writing sample for graduate school on truth (well, deflationism about the truth predicate).

I finished my undergraduate degree at UC-Davis, where I took classes with a stellar group of philosophers. One of my favorite classes was a metaphysics class taught by the late Josh Parsons. Josh taught in a way that empowered students to find their own solutions to philosophical puzzles, and it was during his class that I first thought I might want to continue studying philosophy in graduate school. Josh was also unbelievably cool and talented and fun to talk philosophy with. One of my favorite memories from that time was chatting with him about movies at the cafe Delta of Venus and learning about his favorite scene from “Lost Highway,” one that involves a character who at one time is located in two different locations. Josh will be sorely missed – my heart goes out to his friends and family.

3:AM:One of the things you’ve investigated is the problem of moral disagreement and epistemic contextualism. So first, could you set out what the issue is that arises when contextualists deny that a disagreement may not be over a proposition literally asserted? Can you give us a concrete example of the kind of thing we’re talking about here.

JK:First, we should get the kind of view and the kind of challenge to it on the table. A contextualist theory of expression-type X holds that the content of X may vary depending on the context in which it is used. So, a moral contextualist might hold that ‘morally wrong’ expresses a different content in different contexts – perhaps, ‘morally wrong’ means the same as ‘wrong as judged by the standards of my community’.

One challenge to this kind of view, which goes back at least to G.E. Moore, is this: if I say ‘killing is wrong’ and someone else says, ‘No, killing is not wrong’, it seems that we thereby disagree about whether killing is wrong. But if the contextualist view above is correct, then I would have said that killing is wrong as judged by the standards of my community and you would have said that killing is not wrong as judged by the standards of your community. But clearly, we could both be right about what we (each) said, and so, it would seem, we cannot thereby disagree. Thus, if (this version of) moral contextualism is right, then we would not disagree when we clearly do, so (this version of) moral contextualism must be wrong.

3:AM:And in your paper with Josh Knobeyou argue that although two people may disagree about a moral position they are not necessarily saying that the other view is wrong. How does this work?

JK:An assumption in the above argument against contextualism is something like the following: the best explanation of why someone uttering a moral sentence S and someone else uttering its negation ~S disagree is that one of their claims must be false. Given this assumption, it may seem that what we want is a moral semantic theory that is invariantist -- one according to which moral sentences like "killing is wrong" and "killing is not wrong" always express incompatible propositions.

What Josh Knobe and I found was that there are cases in which someone utters S and someone else utters ~S (where S is a moral sentence) and they thereby disagree, but in which it is also not the case that at least one of their claims must be false. We did this by asking people either (i) whether it was appropriate for one of the speakers to reject the other’s claim by using “No” (this was our way of measuring their judgment about whether the two speakers disagreed) or (ii) whether at least one of the two speakers’ claims must be false.

In our experiment, we looked at cases varying the cultural distance between the two speakers. One speaker was an American college student and the other was either (a) also an American college student, (b) a person from an isolated warrior culture, or (c) an alien, called a Pentar, whose species cares only about maximizing the number equilateral pentagons in the universe.

We found that as you vary the cultural distance between the speakers, people tended to agree less with the statement that at least one of the two speakers’ claims must be false. But we found no similar effect in judgments about whether the two speakers disagreed by making those claims – people continued to think that the two speakers disagreed in every case. Thus, when you get to the human-alien case, ordinary speakers were overall strongly inclined to think that the two speakers disagreed and weakly inclined overall to think that it’s not the case that at least one of their claims is false.

We take this pattern of judgments to be evidence that the ordinary notion of disagreement allows for two people to disagree by making non-incompatible moral claims (i.e., two claims such that it is not the case that one must be false). If that is right, then, not only is the disagreement argument against contextualism undermined, but the correct theory may in fact be a contextualist theory, since that kind of theory does allow for the possibility of disagreements that do not involve making incompatible claims.

3:AM:So how should we understand the notion of disagreement in these cases and what model of moral semantics does it suggest?

JK:We propose a version of moral contextualism on which moral sentences are evaluated relative to systems of norms; basically, “X is morally wrong” means something like, norm-system N forbids X (where N is picked out by the context of utterance). However, we depart from a simplistic contextualist analysis in which there is a unique norm-system in every context. Rather, in many contexts, we think there is no single system of norms that is uniquely identified at that context. Instead, this contextual parameter might be something the conversational participants might intend to negotiate in the course of their conversation.

When no unique system of norms is settled on in the context, there will be contextual indeterminacy as to whether a particular moral claim is true or false. One way this could happen is if conversational participants differ in normative outlook.

We then think of assertion as a proposal both to update our shared information (as in the view of Stalnaker and others), but also as a proposal to resolve contextual indeterminacy. When the college student says, “X is morally wrong,” he proposes to resolve the indeterminacy in favor of a system of norms that forbids X. When the Pentar says, “X is not morally wrong,” she advocates for a system of norms that does not forbid X. Since no system of norms can both forbid and not forbid X, the two context change proposals are incompatible, and this is the source of disagreement between the two speakers.

But notice that in this kind of case, since the two speakers differ in normative outlook, there remains contextual indeterminacy in the system of norms, and so both of their claims are neither true nor false. So, our theory predicts that it is not the case that at least one of their claims must be false.

3:AM:How does this connect with your arguments regarding modal disagreements? (Perhaps you could sketch for the uninitiated what a modal disagreement is!)

JK:One reaction to the above discussion is that we are making a lot out of an unusual case involving aliens with psychologies radically different from our own. Why draw such surprising conclusions from such a marginal case? One reason is that the same kind of pattern of non-incompatible disagreement arises in other domains as well.

Epistemic modals are words like “might,” “must,” and “probably” that (prima facie) express claims about information or evidence. The same conflict between contextualism and invariantism that we saw in the moral domain arises for epistemic modals as well. Contextualists say that “it might be raining” means something like: rain is compatible with evidence E (picked out by the context of utterance). Here, as in the moral domain, disagreement arguments have been used to put pressure on modal contextualism: it seems that people with very different evidence disagree when one says “it might be raining” and the other says, “No, it is not raining.”

In my paper on modal disagreement, I found a pattern of ordinary judgments very similar to the one Josh and I found regarding moral disagreements. In this experiment, participants are told that Fat Tony has faked his death, but then learn that an investigator who does not know this (and instead is only aware of the evidence Fat Tony planted suggesting his death) says, “Fat Tony might be dead.” I asked participants either (i) whether they would respond to the investigator by saying, “No, Fat Tony is alive” or (ii) whether they agreed or disagreed with the statement that the investigator’s claim is false.

The resulting pattern of responses matches the pattern in the moral case: while participants were strongly inclined to disagree with the investigator by rejecting his claim, they did not think that the investigator’s claim was false.

I think that the overall pattern in the two kinds of cases suggests that disagreeing with a claim is one thing and thinking that the claim is false is another – the two often go together, but they need not!

3:AM:You’ve written about the distinction between indicatives and subjunctives – and this looks dry and difficult to see why it’s philosophically important. So can you whip us into a frenzy of excitement about this issue – what is at stake here?

JK:Take the contrast between the following:

1. If Oswald didn’t shoot Kennedy, someone else did.
2. If Oswald hadn’t shot Kennedy, someone else would have.

Here are two pretty mundane observations about these sentences. The first is that (1) and (2) are morphologically different. (2) has the past perfect ‘had’ and the temporal auxiliary ‘would have’, whereas (1) lacks both. The second observation is that (1) and (2) mean very different things. (1) seems to be a reasonable thing to think just because you know someone shot Kennedy, but (2) seems to be something only a conspiracy theorist would think. Philosophers call conditionals like (1) indicative conditionals and those like (2) subjunctive (or counterfactual) conditionals.

Now, it turns out that these observations are related in a deep way: a wide range of historically unrelated languages use past tense morphology to encode the distinctive meaning expressed by subjunctive conditionals (for more on this, see my colleague Sabine Iatridou’s work). The question I am interested in is: why is this? In particular, could it be something about the way we as humans think about and represent counterfactual possibilities?

There are two kinds of answers to this question in the literature. The first is that past tense in some environments has a modal meaning – so, basically, instead of temporal distancing, it means modal distancing (roughly, “go to a possible world that’s further away than ones you think might be actual”). The second kind of answer is that this past tense on conditionals just has its ordinary temporal meaning (roughly, “go to a time before now”). On the latter view, (2) is what you get when you evaluate the conditional (3) at a past time:

3. If Oswald doesn’t shoot Kennedy, someone else will.

3:AM: So what’s your theory and why is it better than serious alternatives?

JK:I have defended a version of the second kind of view – that past marking on subjunctive conditionals is genuine past tense (not modal past). One reason I like this view is that it allows us to tell a cool story about the cross-linguistic data.

The basic idea is that the way we think about counterfactual possibilities constrains the linguistic devices we use to talk about them: counterfactual possibilities are historically structured in a way that other, informational possibilities, are not. What I mean by this is that the counterfactual possibilities at a time t are those which share a history up until t but diverge thereafter, whereas informational possibilities are not constrained in this way. So, all of the present counterfactual possibilities share our actual past, but it need not be that all of the present informational possibilities share our actual past.

Now, suppose you want to express that there is a counterfactually necessary connection between a certain coin being flipped and it landing on heads, even though the coin wasn’t flipped and also didn’t land heads. Since the possible coin flip we want to talk about is in the past, there are no present counterfactual possibilities in which the coin flip happened (and so no present counterfactual possibilities in which the coin landed heads, or tails). But there are past counterfactual possibilities in which the coin was flipped. It is for this reason that we use past tense on conditionals like (4) – the past tense is being used to talk about these past counterfactual possibilities:

4. If the coin had been tossed, it would have landed heads.

Another way to put it is this: the past tense on (4) allows you to talk about a past “branch” point at which the coin counterfactually could have been tossed.

Therefore, if counterfactual possibilities are historically structured in this way and the temporal past view is right, we expect to find a cross-linguistically robust pattern of using past tense (or some linguistic device for picking out a past time) to talk about counterfactual possibilities.

3:AM:What’s the philosophical challenge about ‘backtracking counterfactuals’? How do you propose to meet the challenge?

JK:Consider the following situation. Amy is playing cards with Beth. I’ve seen both hands, and I saw that Amy’s hand is weaker than Beth’s. Amy folds her hand. I now say, “If Amy had called, she would have lost.” That seems right. But now suppose you were also on the scene, and knew that Amy was cheating and knew Beth’s hand. You might now say the following, “Amy knew Beth’s hand, so she would have called only if she (Amy) had the winning hand. Therefore, if Amy had called, she would have won.”

My counterfactual seems true when I said it, but so does yours when you say it; yet, prima facie, mine and yours are contraries, so they can’t both be true. A natural conclusion to draw is that counterfactuals admit of two different interpretations. Philosophers call the interpretation on which mine is true a non-backtracking interpretation, while the interpretation on which yours is true a backtracking interpretation, because in yours you backtrack further (undo more of the past, so to speak) in supposing the antecedent had happened.

Backtracking counterfactuals are sometimes dismissed as an oddity, but I think they raise an important challenge: why do backtracking interpretations require more work to bring out than non-backtracking interpretations, and why does uttering a related backwards counterfactual (“she would have called only if she had the winning hand” above) help to clear the way for a backtracking interpretation?

I appeal to the temporal past theory to account for these observations. On my theory, the choice of past time is what determines whether an interpretation of a counterfactual is backtracking or not – as you find more and more distant past branch times, the more backtracking is involved in the interpretation of the counterfactual. Non-backtracking interpretations are generally default because they are more likely true in a null context. The reason for this comes from a fact about historically structured counterfactual possibilities: as you go from past to future, the fewer (in a certain sense) counterfactual possibilities there are. Thus, less distant past times result in a more restricted domain of counterfactual possibilities, and hence fewer possibilities of falsifying the counterfactual. Since we generally want to interpret people charitably, we default to a “less temporally distant” past interpretation of counterfactuals, and this accounts for the default non-backtracking interpretation.

But one way to talk about a further past branch time is to make that time salient in the context. And one way to do this is to talk about a backwards counterfactual evaluated at that time. That is why uttering a relevant backwards counterfactual is a helpful way to set up a backtracking interpretation of a subsequent counterfactual.

3:AM:What is Ramsey’s thesis regarding indicatives and conditionals – and what is the problem with rare counterexamples?

JK:Ramsey’s thesis, so-called because something like it goes back at least to Frank Ramsey, is that the probability of an indicative conditional if A, then C is equal to the probability of its consequent C conditional on its antecedent A. This is a very plausible claim. Suppose I have just rolled a fair die D and then I ask you, “How likely is it that if D landed on a prime, it landed on an odd number?” The most plausible answer to this is two-thirds, and this seems so because the probability that the die landed on an odd conditional on it landing on a prime is two-thirds (that is: two out of three possible prime outcomes are odd: 3 and 5, vs. 2, 3, and 5).

However, there are purported counterexamples to Ramsey’s Thesis in the literature (due to Vann McGee and Stefan Kaufmann). Here is one that I owe to my colleague Steve Yablo. Suppose a coin, C, has just been tossed. You know C was weighted towards tails, so you think it likely landed on tails. You then hear that Jones caught a glimpse of the way C landed. Jones goes to tell you what she saw, but you can’t quite make out what she says. Nonetheless, you’re pretty sure Jones said “heads.” On this basis, it seems plausible to conclude that it is very likely that:

5. If Jones is correct, C landed heads.

Nonetheless, it still seems plausible to think that the following is unlikely:

6. If Jones is correct, she said “heads.”

And that is because you still think C landed on tails (since it was weighted towards tails). So, it is more likely that Jones said “tails” if she was correct.

Here’s the rub: conditional on Jones being correct, “C landed heads” and “Jones said ‘heads’” are equivalent! So, the conditional probability of C landing heads given that Jones is correct is equal to the conditional probability that Jones said “heads” given that Jones is correct. Yet, there seems to be an intuitive difference in the probabilities of (5) and (6). Thus, (5) and (6) jointly comprise a counterexample to Ramsey’s Thesis – for at least one of them (maybe both), its probability is not equal to its corresponding conditional probability.

3:AM:How do you explain these puzzling failures?

JK:We have a prima facie extremely appealing thesis with counterexamples, so what should we do? We could try to explain away the counterexamples, or we could accept them. The latter approach would then require us to say why the thesis felt so intuitively plausible in the first place. In my paper, I offer a theory that tells us why such counterexamples arise and why they are rare.

My theory is basically a generalization of one proposed by Robert Stalnakerin his paper “Indicative Conditionals.” On that kind of theory, an indicative conditional if A, then C is true at world w just if the closest A-world to w is a C-world. Stalnaker proposes that indicative conditionals are constrained by the context set – so that the closest A-world to any in the context set must also be within the context set (which means that it must also make true everything that is currently presupposed in the conversation).

On my theory, conditionals are evaluated relative to a question under discussion in the context. The question under discussion further constrains the closeness relation, so that the closest A-world to any in the context set must be an A-world that answers the question under discussion the same way.

Go back to the coin flip case. Start with (5): “If Jones is correct, C landed heads.” Plausibly, there are at least two questions under discussion in this context: whether C landed heads or tails (H/T) and whether Jones said ‘heads’ or ‘tails’ (SH/ST). (Probably also the question of whether Jones is correct is under discussion, but I’ll set that aside here – see my paper for the full theory.) Now, we need to figure out which of these questions the conditional (5) is evaluated relative to. It turns out that if (5) evaluated relative to the question whether C landed heads or tails, then it will end up equivalent to “C landed heads.” In the paper, I argue that there is independent reason to think that this kind of interpretation should almost never happen.

Thus, we evaluate (5) relative to the question whether Jones said ‘heads’ or ‘tails’ (SH/ST). By the question-constraint, for any world in the context set where Jones said ‘heads’, the closest world to it in which Jones is correct must also be one in which Jones said ‘heads’. But any world where Jones is correct and said ‘heads’ is one in which the coin landed heads. Therefore, we have shown that (5), interpreted relative to SH/ST, is true at every world in the context set in which Jones said ‘heads’. Thus, since we think it’s likely Jones said ‘heads’, it follows that it is likely that (5) is true.

But now consider (6): “If Jones is correct, she said ‘heads’.” For the same reason as above, evaluating (6) relative to the question whether Jones said ‘heads’ or ‘tails’ will make it equivalent to “Jones said ‘heads’” and so we have reason to rule out this interpretation. Thus, we evaluate (6) relative to the question whether C landed heads or tails (H/T). Again, by the question-constraint, for any world in the context set where the coin landed tails, the closest world to it in which Jones is correct must also be one in which the coin landed tails. But any world where Jones is correct and the coin landed tails is one in which Jones said ‘tails’ (not ‘heads’!). Therefore, we have shown that (6), interpreted relative to H/T, is false at every world in the context set where the coin landed tails. So, since we think it’s likely that C landed tails, it follows that it is likely that (6) is false, or alternatively that it is unlikely that (6) is true.

So, that’s how I account for the failures of Ramsey’s thesis. How, then, do I account for why it felt intuitively plausible in the first place? My answer to this is that when there is no admissible question under discussion to interpret the conditional relative to, the conditional is default interpreted relative to the universal question that has as its one answer the entire context set. I show that my semantics entails Ramsey’s Thesis on this interpretation of if A, then C.

3:AM:You’ve argued that political code words like ‘inner-city’ don’t encode hidden meanings but nevertheless you do agree that they bring about some surprising effects. So what are these effects and how do you explain them?

JK:Compare the following sentences:

7. Food stamp programs help many inner-city families.
8. Food stamp programs help many African American families.

The political scientist Ismail White found experimental evidence that, for participants reading (7), stronger prejudice towards African Americans predicts greater opposition to food stamp programs. However, this effect is neutralized for participants reading (8). This puzzling phenomenon is explained by the theory put forward by Tali Mendelberg, which is that, since there are social costs to being perceived as violating norms of racial equality, individuals inclined to act in racist ways will look for ways to ambiguate that behavior, allowing them the veneer of plausible deniability against charges of racism. Thus, opposing food stamp programs after reading (8) is too overtly racist – to do so would risk being seen as opposing a policy because it helps African American families. But opposing food stamp programs after reading (7) allows for the possibility that the grounds for one’s opposition are non-racial: for instance, maybe one thinks urban schools already receive sufficient government assistance.

This hypothesis raises a question about language, which is: what do code words like “inner city” mean? In particular, how does their meaning enable them to play this role in creating plausible deniability?

In my paper, I argue for a very simple theory that can explain this phenomenon. The theory is that “inner-city” does not have any racial meaning, but it carries racial connotations because there are widespread stereotypical beliefs about the racial makeup of American urban cores. Thus, when people learn that food stamp programs help inner city families, they come to learn that such programs help African American families by inferring this from their (pre-existing) beliefs about the racial makeup of the inner city. Yet, since none of this is part of the meaning of “inner-city,” using the word affords plausible deniability, since one can always retreat to its thin lexical meaning.

This brings the issue of code words in line with a longstanding issue in philosophy of language, regarding the analytic/synthetic distinction. In effect, I propose that sentences about the racial makeup of the inner city are synthetic, while others such as Jason Stanley have proposed that such sentences are analytic (in a certain sense). There is a lot more to go into here, but maybe I should just leave it there…

3:AM:And finally for the readers here at 3:AM, are there five books you can recommend to take us further into your philosophical world?

For more on disagreement, contextualism, and a relativist alternative, I recommend John Macfarlane’s Assessment Sensitivity.

The book that got me into conditionals was Jonathan Bennett’s A Philosophical Guide to Conditionals.

I think that Angelika Kratzer’s work in semantics is incredibly deep and insightful. I recommend her book, Modals and Conditionals.

The book that had the greatest impact overall on me and my philosophical thinking, and one that I come back to more than any other, is Robert Stalnaker’s Context and Content.

For more on code words, I recommend Tali Mendelberg’s The Race Card.

ABOUT THE INTERVIEWER
Richard Marshallis still biding his time.

Buy his book hereto keep him biding!