This week at The Junkyard, we’re hosting a symposium on Peter Langland-Hassan’s recent book Explaining Imagination (Oxford University Press, 2020). The book is available as an Open Access download. See here for an introduction from Peter. Commentaries and replies will follow Tuesday through Thursday.

* * *

Commentary from Margot Strohminger

We often evaluate conditionals, or statements of the form, ‘If p, q’, in the imagination. Roughly, we perform a ‘thought experiment’ by imagining a scenario in which the antecedent p holds and then considering whether the consequent q holds in it, too. When someone ends up believing the conditional on the basis of such a thought experiment, one imagines both the antecedent and the consequent of the conditional. The attitudes towards the antecedent and the consequent (not to mention the attitudes towards many other propositions used to ‘fill in’ the details of the imagined scenario) constitute sui generis imaginings.

Or so anyways it is widely assumed in discussions of the imagination and related theories of folk psychology and pretense. (The term ‘imagination’ is not always used for this sui generis attitude. Nevertheless, the consensus posits a propositional attitude that resembles but is nevertheless distinct from belief, and which maps more or less well on to our ordinary talk of ‘imagination’.) Peter Langland-Hassan challenges this assumption in Chapters 5 and 6 of his provocative and engaging new monograph, Explaining Imagination. The challenge is part of a broader challenge to the idea that there are sui generis imaginings at all. Langland-Hassan maintains that they can be ‘reduced’ to other familiar folk-psychological states and processes in the following sense. Any imagining that p is constituted by some other folk-psychological state or process, which may or may not be an attitude towards the proposition that p. In some cases, the state constituting an imagining is a belief, but in others it is a judgment, a desire, an intention, or a decision (15).

The evaluation of conditionals represents an important battleground between Langland-Hassan’s ‘reductionism’ and ‘primitivist’ orthodoxy. This is because it is such a central platitude about the imagination that it features in the evaluation of conditionals. In fact, it is sometimes suggested that the original function of the imagination is that it provided a way of evaluating conditionals when other means were not available or were simply too risky to implement.[1]

For Langland-Hassan to carry off his project, he needs to convince us that in every case where we (correctly) attribute an imagining that p in the evaluation of conditionals, this imagining should not be thought of as sui generis; instead, it is some other attitude. Looking at an example might help. We can look in more detail at an example that Langland-Hassan considers (Ch. 6, Sect. 7). The example is a shepherd using his imagination to come to believe the conditional, ‘If the sheep have broken out of the pen and disappeared, they have gone down to the river.’[2]

I’ll start by describing how the shepherd performs this feat, by my (primitivist) light’s.[3] First, the shepherd imagines that the sheep have broken out of the pen and disappeared and then considers what else has happened (‘supposing’ is the term often used for this imagining, which initiates the construction of an imagined scenario). In response the shepherd imagines various further propositions. The process by which one imagines these further propositions deploys the same inferential mechanisms used in belief updating. The imaginings have contents that match beliefs the shepherd would either form or retain were he to come to believe the antecedent. For example, if the shepherd would still believe that the sheep like to frolic by the river even after coming to believe that the sheep have broken out of the pen, then this claim is something that the shepherd might imagine here. Likewise if the shepherd would come to believe that the sheep have gone down to the river, then this claim is also something the shepherd might imagine. Supposing he does the latter, the shepherd can then come to believe the conditional, ‘If the sheep have broken out of the pen and disappeared, they have gone down to the river.’ In coming to believe this conditional, the shepherd is relying on a heuristic, which allows the subject to form the belief in the conditional, ‘If p, q’, when S imagines that q on the initial supposition that p. The heuristic can be thought of as an instance of a more general heuristic, the “Suppositional Rule”, proposed by Williamson (2020):

Take an attitude unconditionally to ‘If A, C’ just in case you take it conditionally to C on the supposition A. (2020, 19)

Heuristics like these harken back to Ramsey (1929, 155) and subsequent appeals to the ‘Ramsey test’ in the literature on conditionals.

Langland-Hassan argues that the shepherd’s imaginative feat is better described without positing sui generis imaginings (134ff.). What the shepherd is doing, Langland-Hassan proposes, is simply the following (135). First, “[he] registers the question, ‘Where are the sheep likely to be if they have broken out of their pen and disappeared?’ and begins relevant processing.” Second, he accesses “[b]eliefs about the sheep’s preferences and tendencies”. Third, and he finally, he relies on these beliefs in an inductive inference to reach the conditional of interest. According to Langland-Hassan, this reductionist description of the example is “clearly preferable” to primitivist descriptions like the one I rehearsed in the last paragraph (137). Here my main goal will be to instill some doubts in the reader that it is preferable, let alone clearly preferable. But first I will suggest that even if Langland-Hassan’s description of the shepherd example is accurate, it is not enough for his purposes.

To see why it is not enough, recall how Langland-Hassan characterizes his reductionism in Chapter 1: whenever someone S imagines that p, that imagining is constituted by some other folk-psychological attitude—be it a belief, a judgment, a desire, an intention or a decision.[4] In the present example the shepherd imagines the propositions that the sheep have broken out of their pen and disappeared and that they have gone down to the river (among other propositions). What attitudes might the reductionist reach for? The shepherd recollects various beliefs he has about the sheep and their surroundings, and then comes to judge the conditional. Is the idea supposed to be that the imagining that the sheep have broken out of their pen and disappeared is constituted by his consideration of the question, ‘‘Where are the sheep likely to be if they have broken out of their pen and disappeared?’ (or else the various beliefs he relies on in the inference?); and likewise, that the imagining that the sheep have gone down to the river is constituted by the judgment that the sheep have gone down to the river if they have broken out of the pen? These don’t strike me as plausible candidates, in part because the alleged underlying attitudes’ contents are so different from the imaginings. However, I cannot think of any more plausible candidates on the reductionist’s behalf. Nor does Langland-Hassan address the issue as far as I can see.

Setting aside whether Langland-Hassan’s description of the case is genuinely reductionist or not, let’s now consider whether it should be preferred to a primitivist description. According to Langland-Hassan, his description is simpler (136–7). In one respect, it is clearly simpler: it makes use of one fewer attitude. However, there is a respect in which it is more complex that I think Langland-Hassan skates over. The primitivist story recycles machinery already in place for the purposes of belief updating. We can think of it as relying essentially on three pieces of machinery. First, there is a gadget that determines which beliefs are ‘compatible’ with the supposition of the antecedent. Second, there is a gadget that determines which imaginings can be formed given earlier imaginings. Finally, there is the Ramsey-esque heuristic. The first two of these are already needed in order to accomplish belief updating. All that is new is the heuristic. In the primitivist story, it is natural to think that the way the shepherd would reach the belief that the sheep have gone down to the river were he to come to believe that they escaped from their pen by an inductive inference that draws on other beliefs he has. The shepherd is performing the same inference modulo the fact that it is performed ‘offline’ in the imagination. Langland-Hassan’s story also uses an inductive inference. However, the inference is not just like the one the shepherd uses in this case of belief updating. In comparing the two descriptions, Langland-Hassan depicts both as deploying essentially the same kind of inference because they are inductive (137). But this misses the fact that the kind of inference used in his story relies on a new rule, one that is not already used in the shepherd’s belief updating. There is thus a respect in which the reductionist story is more complex than Langland-Hassan lets on.

In any case explanatory adequacy strikes me as more important at this stage of investigation than simplicity. At this stage we do not even know if the reductionist can explain the full range of cases about conditionals and related claims. Exploring this would of course take us outside the bounds of the case we have been exploring, and elsewhere in Chapter 6 Langland-Hassan starts to take up this task—for example, by considering whether the reductionist can explain how thought experiments might produce knowledge. Still, without straying too far, we might consider how the shepherd might come to believe the negation of some conditional, or merely accept the conditional as highly probable or, say, 0.6 likely without outright believing it. Williamson’s “Suppositional Rule” is powerful enough to explain this achievement. While Langland-Hassan does not address this issue, I take it he will end up positing new inference rules. The once simple-looking reductionism thereby grows increasingly baroque and mysterious, resting on a black box of ever more inductive inference rules, the nature of which is to be left “a matter for empirical investigation” (137).

[1] See especially Nichols and Stich 2003, 28–9.

[2] Langland-Hassan chooses this example because it appears already in Williamson 2016, 118.

[3] Cf. Nichols and Stich 2003, 30–2 and (especially because it treats this very example) Williamson 2016, 118–9.

[4] Langland-Hassan writes, “I will argue instead that, for any situation where we attribute an imagining to someone, we could alternatively, and equally plausibly, attribute either a belief, desire, or intention—or one of their occurrent counterparts (viz., a judgment, desire, or decision)” (15). Elsewhere he includes memories among the relevant attitudes (11), but this doesn’t fundamentally alter my observation in the main text.

REFERENCES

Kind, Amy, and Kung, Peter, eds. Knowledge Through Imagination. Oxford: Oxford University Press.

Langland-Hassan, Peter. 2020. Explaining Imagination. Oxford: Oxford University Press.

Nichols, Shaun, and Stich, Stephen P. 2003. Mindreading: An Integrated Account of Pretence, Self-Awareness and Understanding Other Minds. Oxford: Oxford University Press.

Ramsey, F. P. 1929/1990. General propositions and causality. Reprinted in D. H. Mellor, ed., Philosophical Papers, 145-63. Cambridge: Cambridge University Press.

Williamson, Timothy. 2016. “Knowing by imagining.” In Kind and Kung 2016, 113-23.

—2020. Suppose and Tell: The Semantics and Heuristics of Conditionals. Oxford: Oxford University Press.

Response from Peter Langland-Hassan

Questions concerning the relationship of imagination to conditionals are some of the most difficult and foundational in the literature on imagination.  I’m grateful to Margot Strohminger for articulating the tensions she sees in the revisionary approach I recommend.

Margot’s first concern is that the contents of the beliefs I hold to constitute imagining that p, in cases of conditional reasoning, do not themselves have the content p.  On my view, a shepherd’s imagining that the sheep have broken out of their pen does not amount to his tokening a mental representation with the content: the sheep have broken out of their pen.  Instead, I identify his imagining with the entire process by which relevant information about the sheep’s proclivities is retrieved and used to drive an inference that takes the form of a believed conditional:  if the sheep have broken out of the pen, then they have gone down to the river.   According to me, this process only involves inferences among one’s beliefs, none of which have “the sheep have broken out of the pen” as their content.   Margot finds it problematic that “the alleged underlying attitudes’ contents are so different from the imaginings” used to describe the process on a superficial level.   

However, the maxim, “don’t assume content-mirroring” is a central pillar of my approach, argued for in Chapter 1.  In order for a token mental state of φ-ing that p to consist in one’s being in some more basic token state (or collection of states), I argue, that more basic state needn’t also have the content p.  In favor of this maxim I give the simple example of suspecting that p.  Plausibly, to suspect that p is nothing other than to believe that it is somewhat likely that p.  (Though see Nick Wiltsher’s commentary.)  The content of the belief is different from that of the suspicion it serves to reduce.  I make a similar argument about regretting that p.  It is plausible that regretting that p is not a primitive attitude but rather reduces to a combination of more basic attitudes and emotional dispositions, none of which need have the precise content p.  So, in response to Margot’s question about which of the states described is to be identified with the shepherd’s imagining that the sheep have broken out of their pen, I would say “all of them.”  To imagine that the sheep have broken out of the pen (or anything else) is something we do, and we need not do it in exactly the same way each time. (That is, incidentally, the second maxim of Chapter 1: “don’t assume homogeneity.”) We should focus instead on what goes on in specific instances where we are inclined to ascribe an imagining and assess what reductions are available in those cases. 

Margot’s second worry is that my reductive proposal, while coherent, is not in fact simpler than the primitivist view I seek to overturn.  I hold that the inference rule whereby the shepherd moves from beliefs about the past tendencies of the sheep to the conditional conclusion that, if the sheep have broken out of their pen, then they have gone down to the river, to be inductive in nature.  The primitivist also holds that there is an inductive inference at work; however, they hold that it is the very same inference rule that is followed “online” in cases where a person in fact comes to believe that (e.g.) the sheep have broken out of the pen.  The only difference is that the rule is followed “offline” among one’s suppositions, on their view.  This is a rule I am also committed to our being able to follow, albeit only as applied to our beliefs.  It seems that I am committed to our being able to follow two rules and the primitivist only one.  Of course, the primitivist has the additional cost of appealing to the offline heuristic Margot mentions—specifically, of the ability to move from whatever occurs “offline” to a judgment in favor of a specific conditional.  They are also committed to the sui generis attitude itself, of course, and whatever else comes along with it.  (In the book, I argue that the hidden costs are considerable (pp. 133-137)). 

Who wins the simplicity standoff?  While the matter deserves closer attention than I can give it here, it seems to me that where Margot sees two rules there may be only one.  The ability to follow a rule of the (rough) form “if most Xs have been Ys in the past, then believe that, probably, the next X will be a Y,”—which Margot allows we have—seems one and the same with the ability to follow the rule “if most Xs have been Ys in the past, then believe that, probably, if Z is a X then Z is a Y.”  It is certainly odd, on its face, to suggest that following one rule requires sui generis imaginative states while following the other does not.  Both are rooted in the same principle:  believe that unobserved Xs are relevantly like observed Xs.  We would need good reason to think humans follow some finer-grained version of that rule to the exclusion of another.

Margot ends with a worry that “we do not even know if the reductionist can explain the full range of cases about conditionals and related claims,” including how we come to believe the negation of a conditional.  By contrast, she thinks Williamson’s Suppositional Rule “is powerful enough to explain this.”  In response, it is not clear to me what special puzzles are raised by the cases Margot mentions.  As she notes, I consider in the book a number of examples that seem far tricker, including philosophical thought experiments, the method conditional proof, and reductio ad absurdum.  More importantly, Williamson’s Suppositional Rule itself does not explain the mechanics of how we arrive at any inferences.  It simply advises us to take an attitude unconditionally to ‘If A, C’ just in case we take it conditionally to C on the supposition A.  That sounds like good advice, but how do we go about following it in all the cases that we do?  Williamson says that we do offline what we would do online if we believed the antecedent—without saying much about the online inferences themselves.  I offer a reductive explanation of what’s involved in following the Suppositional Rule.  In carefully considering what happens in ordinary “online” inferences, we see that we don’t need the “offline” part at all.