Reply to Holmen on Williamson

In his review of Jennifer Hornsby’s “Belief and Reasons for Acting”, Holmen briefly considers Williamson’s example of the burglar who persists in searching for a diamond in a house at great risk of being caught. Holmen takes the burglar example to establish the following claim:

[K]nowledge sometimes must figure in the best explanation for why some agent F-d. According to him [Williamson], attributions of knowledge may be a better predictor for determining someone’s actions by lending more probability to a certain way of conduct.

Roughly, the idea is supposed to be that the burglar’s persistence can only be satisfactorily explained by attributing him with knowledge. Consequently, it is alleged that knowledge attribution allows us to better explain or predict what the burglar would do.

In my previous post, I expressed reservations about whether Williamson’s burglar example establishes the conclusion Holmen claims it does. I consider the burglar’s psychophysical doppelgänger, who displays the same level of psychological conviction and engages in the same exact behaviour, but who nevertheless lacks knowledge because his belief has been rendered false by a highly improbably quantum occurrence. I maintain that so long as we keep his level of psychological conviction the same as that of the original burglar, we are equally able to explain (i.e., make sense of) and predict his behaviour. The upshot is that knowledge attribution is not necessary for a satisfactory explanation of the burglar’s conduct. What knowledge attribution adds (if anything at all) is not explanatory or predictive power, but rather a justificatory dimension—it helps us to understand why the burglar who has knowledge was justified in his conduct.

In his follow-up post, "Knowledge in Explanation: A reply to Avery Archer", Holmen attempts to sidestep my criticism by arguing that the fact that we can explain an agent’s conduct even when she lacks knowledge does not undermine the claim that knowledge is sometimes necessary for explaining or predicting an agent’s conduct. His point seems to be that, at best, my argument shows that something other than knowledge can explain or cause the burglar’s conduct. Even so, I have not shown that knowledge is not the explanation or cause of the agent’s behaviour in the particular example Williamson posits. Holmen summarises what is problematic about my argument as follows:

Avery's case is more like saying that "Shakespeare did not write Macbeth since there is a doppelgänger case where Macbeth gets written but by Marlowe." And even granted the presence of such a case Shakespeare surely wrote Macbeth and is the cause of Macbeth's existence. Likewise, there could be dozens of cases where the burglar acts in the same way but for other reasons than that he knew P. They say nothing whatsoever about whether Kp is the cause in our case for his behaviour.

However, it seems to me that Holmen mischaracterises the dialectical context in which my objection to the burglar example is articulated. He depicts my argument as attempting to establish a positive conclusion when the conclusion it seeks to establish is purely negative. This point may be illustrated by considering the following single-premise argument:

(1) The biological complexity we observe can only be satisfactorily explained by the existence of a Divine designer

(2) Therefore, there is a Divine designer

We may effectively refute (1)-(2) if it could be shown that:

(1*) The biological complexity we observe can be satisfactorily explained by natural selection.

If true, (1*) impugns (1)-(2) by revealing (1) to be false. However, (1*) only establishes the negative conclusion that (1)-(2) fails to establish its conclusion. It is not (nor should it be taken to be) a positive argument showing that (2) is false. For example, the following is clearly a bad argument:

(1*) The biological complexity we observe can be satisfactorily explained by natural selection.

(2*) Therefore, there is no Divine designer.

The lesson is that the same consideration can be effective in a negative argument (i.e., one intended to show that a specific argument fails to establish its conclusion) but ineffective in a positive argument against that same conclusion. It is therefore important to get clear on the aim of an argument (whether it is negative or positive) before one can evaluate its cogency. Since my objection only targets the efficacy of Williamson’s burglar argument it is clearly a negative argument.

Now Holmen’s point seems to be that my claim that beliefs are equally able to explain or predict an agent’s conduct does not entail that knowledge is not a cause. This is of course true if one assumes that I am offering a positive argument in which the conclusion “knowledge is not a cause” is supposed to directly follow from the claim that “things besides knowledge may also be causes”. But to construe my argument along these lines would be to overlook the dialectical fact that I am simply challenging one of the premises in Williamson’s argument (i.e., a purely negative endeavor). To see that this is so, we may summarise Williamson’s burglar example in terms of the following single-premise argument:

(3) The burglar’s conduct can only be satisfactorily explained by attributing him with knowledge.

(4) Therefore, knowledge is sometimes necessary for a satisfactory explanation of an agent’s conduct.

What Williamson’s burglar example alleges to show is that there are certain kinds of conduct (such as the burglar's persistence in the face of countervailing evidence) that simply cannot be satisfactorily explained without attributing knowledge to the relevant agent. (In this regard, his argument is analogous to the creationist who insists that a Divine designer is the only satisfactory explanation of the observed biological complexity.) Construed along these lines, all one needs to do in order to impugn Williamson’s argument is to show that the very same conduct can be satisfactorily explained without attributing knowledge. Since the agent’s conduct is one of the things that is held fixed between the burglar and his doppelgänger, then our ability to explain the doppelgänger’s conduct without attributing knowledge to him shows that knowledge attribution is not necessary to satisfactorily explain or predict the original burglar’s conduct. Thus, we arrive at my counterclaim:

(3*) The burglar’s conduct can be satisfactorily explained by attributing to him a belief with the appropriate level of psychological conviction.

If true, (3*) shows that (3)-(4) fails to establish its conclusion by revealing (3) to be false.

Humanity's Primate Heritage

Dilworth's Functional Consonance

In his paper, “Perception, Introspection and Functional Consonance”, Dilworth attempts to provide a broadly functionalist account of first-person authority with respect to how things perceptually seem to an agent. The account is functionalist in the sense that it individuates cognitive states in terms of their role in the aetiology of a disposition to engage in a certain kind of classification behaviour. For example, if I perceive that X is red or I believe X is red, then my perceiving or believing are to be unpacked in terms of my disposition to sort X as red in the relevant classification task (i.e., to engage in red-classification behaviour). When my perceptual experience and belief match in terms of classification behaviour, they are said to be “consonant” with each other. Moreover, Dilworth holds that in any given situation, S, it is logically impossible for an agent to be disposed to behave in two mutually exclusive ways in response to S. It follows that if an agent has both a perceptual experience and belief in response to a given situation, the experience and belief must be consonant with each other. This leads Dilworth to postulate what he refers to as the “perceptual consonance principle”:

(PCP): If one is having at time t a perceptual experience concerning the F-ness of X, and if at that same time t one is also having a perceptually derived belief concerning the F-ness of X, then the experience and the belief must be consonant with each other.

Unfortunately, it is possible to construct a counterexample to (PCP). Consider a case in which an agent is given the task of classifying objects as straight or bent. Moreover, let us suppose that the agent is unfamiliar with the phenomenon of refraction. Looking at a partially submerged stick the agent has the visual experience of the stick being bent. Given Dilworth’s dispositional model, this entails that the agent is disposed to classify the stick as bent. Suppose further that the agent slides her fingers down the length of the stick, giving rise to the tactile experience of the stick being straight. By Dilworth’s lights, this entails that the agent is disposed to classify the stick as straight. Since it is possible for the agent to have the visual experience that the partially submerged stick as bent (at time t) and have the tactile experience of it being straight (at t), it follows that an agent may be simultaneously disposed to engage in conflicting classification behaviours. Furthermore, we can imagine that the agent forms the perceptually derived belief that the stick is straight (based on her tactile experience). In such a case, the agent would have a perceptual experience concerning the shape of the stick (a la sight) and a perceptually derived belief concerning the shape of stick (a la touch), without the experience and belief being consonant with each other.

UPSHOT: What must be added to (PCP) is a restriction to a single sensory modality, and not simply a restriction to a given time or situation. However, as far as the paper under consideration is concerned, Dilworth fails to provide us with the theoretical tools we need for such a restriction.

Dilworth appears to anticipate this objection in his discussion of the Muller-Lyer illusion; a case in which a pair of equal vertical lines (A and B below) appear to have different lengths.

According to Dilworth, one can be said to perceive the lines as being of different lengths because one is temporarily disposed to classify them as such. However, this perceptual disposition is quickly displaced by one’s prior belief that the lines are actually the same length. Moreover, Dilworth points out that this offers no counterexample to his claim that there cannot be conflicting perceptual dispositions since the “relevant knowledge that the lines are actually the same length is clearly not a belief that is perceptually acquired...” He summarises the point as follows:

[W]hat we have in such cases is not a conflict of perceptual dispositions—which has already been shown to be, strictly speaking, impossible—but instead a refusal by a perceiver to believe what she would otherwise believe on ‘the evidence of her own eyes’, so that the temporary or initial consonant belief that the lines are of different lengths is rejected in favour of a longer-term belief based on prior knowledge.[1] (Italics mine)

CAVEAT: Insofar as Dilworth’s functionalist analysis commits him to the thesis that being disposed to engage in classification behaviour is a necessary condition for perception, his claim that one is only “temporarily” disposed to engage in the relevant classification behaviour already concedes too much. If one perceives that X is F only if one is disposed to engage in F-classification behaviour with respect to X, then the point at which one ceases to be so disposed one also ceases to have the perceptual experience in question. However, in the case of a Muller-Lyer illusion, the agent continues to have the perceptual experience of the lines being different lengths even after she is no longer disposed to classify them as such.

Putting aside the above caveat, Dilworth’s treatment of the Muller-Lyer illusion case fails to generalise to that of the partially submerged stick. In the latter the conflict is not between a perceptual experience and a prior belief (as is plausibly the case in the Muller-Lyer illusion), but rather between two conflicting perceptual experiences. In fact, we can imagine a modified partially submerged stick example in which the agent fails to form a belief altogether. We may suppose that, confronted with conflicting perceptual experiences, the agent simply decides to remain agnostic as to whether the stick is straight or bent. (Recall, we have assumed that the agent is unfamiliar with, and therefore has no prior beliefs about, the phenomenon of refraction.) However, even if the agent were to decide to withhold belief altogether, she may still continue to enjoy both sets of conflicting perceptual experiences. Thus, unlike the Muller-Lyer case, the partially submerged stick example presupposes no competing beliefs or knowledge.

It is important to get clear on why this poses a problem for Dilworth’s account. The problem is not that the partially submerged stick case threatens to undermine Dilworth’s claim that any given perceptual experience may be accompanied by a corresponding consonant belief. Since Dilworth wishes to defend the claim that a belief about how things seem to one must be consonant with the corresponding perceptual experience (if one were to form such a belief at all), the mere fact that such a consonant belief is absent (as we may suppose it is in the partially submerged stick case) is no counterexample to the functional consonance thesis. (On at least this much, Dilworth and I agree.) Moreover, I hold that if the agent were to form beliefs about both her visual and tactile experiences, then each respective belief would be consonant with the corresponding perceptual experience. This is something I believe Dilworth would also wish to say; though I maintain that he is not entitled to do so until he has filled in the gap described by Upshot. However, this is not the worry I presently wish to press.

Instead, the present worry has to do with Dilworth’s claim that in any given situation, S, it is logically impossible for an agent to be disposed to engage in mutually exclusive classification behaviours in response to S. Insofar as he is also committed to the claim that an agent perceives that X is F only if that agent is disposed to engage in F-classification behaviour with respect to X, then the partially submerged stick example shows that an agent may be disposed to classify the same object as both bent and straight. Dilworth must therefore either give up the claim that it is logically impossible to have conflicting perceptual dispositions or the claim that being disposed towards classification behaviour (of the relevant kind) is a necessary condition for perception.

Another tempting (but ultimately unsuccessful) response available to Dilworth is to argue that the partially submerged stick example is an instance of a “mixed case”; one “in which several properties are simultaneously perceived, each of which would require its own separate experience versus belief consonance analysis.” The example that Dilworth provides is that of perceiving an object as being a certain colour (with one sensory modality) and perceiving it as having a certain shape (with another sensory modality). Regarding such mixed cases Dilworth maintains there cannot be any conflict:

Now arguably mixed cases of simultaneous perception of different properties would always be cases involving properties falling under different determinable, since one could not simultaneously perceive an object both to have determinate colour F1 and also to have a distinct determinate colour F2.

The above observations may be perfectly correct as far as they go, but they fail to address the issue at hand. Recall, the objection with which this post is concerned is not directed against the notion of experience/belief consonance per se (to which I am sympathetic), but against the conjunction of the claims that (a) it is logically impossible to have two conflicting dispositions and (b) that being disposed to engage in classification behaviour is a necessary condition for perceiving that X is F. Dilworth fails to consider cases in which different sensory modalities present an agent with conflicting information with respect to the same property. The partially submerged stick example is one such case; an instance in which both the visual and tactile sensory modalities provide information about an object’s shape. Thus, unlike the mixed cases that he considers, the partially submerged stick example appears to be a genuine instance in which Dilworth is forced to either (a) acknowledge that an agent may be simultaneously disposed to engage in conflicting classification behaviours or (b) give up on the idea that being disposed to engage in the classification behaviour is a necessary condition for perception.

Littlejohn and Comesana on Epistemic Justification

Reply to Holmen on Hornsby

I wish to respond to Heine Holmen’s very insightful analysis of Jennifer Hornsby’s paper, “Knowledge, Belief and Reasons for Acting”. Since Holmen already does a fine job adumbrating Hornsby’s position (see here), I will skip the summary and jump right into the analysis.

Let me begin by laying my philosophical cards on the table. I am very sympathetic with Hornsby’s claim that knowledge is a reason for acting. However, I believe Hornsby’s account suffers because of her failure to distinguish between (what may be referred to as) explanatory and justificatory reasons. Explanatory reasons (or what I sometimes prefer to call “motive-giving explanations”) have to do with our attempt to make sense of or explain the purposive activity of intentional agents (rational and non-rational alike). By contrast, justificatory reasons have to do with our attempt to ascertain whether or not the actions of an intentional agent are rationally recommended (and is therefore only applicable to agents with rational capacities). Both explanatory and justificatory reasons are normative (since they both allow for the possibility of error); but while the former answers to a why question (vis-a-vis the actions of any intentional agent) the latter answers to a should question (vis-a-vis the actions of a rational agent).

Note: I define a rational agent as an intentional agent with linguistic abilities.

According to Hornsby, “acting for reasons does not come into play unless the agent has knowledge” (I am here quoting Holmen’s summary of Hornsby’s position). Here, Hornsby clearly has justificatory reasons in mind, for we often are able to make perfect sense of the actions of agents who act from mere belief (including false ones) rather than knowledge. For example, suppose I’m playing catch with my dog in the park. I pretend to throw the ball, and the dog runs over to the place where the ball would have landed (had I thrown it) and proceeds to investigate the location(sniffing around without success). I can explain or make sense of the dog’s behaviour by observing that the dog believes that I threw the ball. The fact that I did not throw the ball (and that the dog’s belief is therefore false) does nothing to undermine my ability to explain the dog’s actions. Thus, as far as explanatory reasons are concerned, I am perfectly capable of making sense of the dog’s actions by simply referring to the dog’s beliefs. No recourse to talk of knowledge is required (in fact, since we are talking about a non-linguistic animal, some would argue that talk of knowledge is not even appropriate). Finally, it is worth emphasising that since there are correctness conditions attaching to the dog’s behaviour (i.e., the dog get’s things wrong when it believes I threw the ball), such explanations remain normative in nature. (I am here assuming that the possibility of getting things wrong is necessary and sufficient for normativity.) Thus, we arrive at my earlier claims that explanatory reasons do not require knowledge, are normative, and can be applied to non-linguistic animals.

If I am right, then Hornsby’s requirement that acting for reasons implicates knowledge is limited to justificatory reasons. Consequently, she errs by assuming that what may be a perfectly good point (i.e., that having a justificatory reason for acting requires knowledge) applies to explanatory reasons as well. Holmen observes Hornsby’s indebtedness to Williamson on precisely this score:

Hornsby seems to start out by picking up a clue from Williamson’s suggestion (2000, p. 62) that knowledge sometimes must figure in the best explanation for why some agent F-d. According to him, attributions of knowledge may be a better predictor for determining someone’s actions by lending more probability to a certain way of conduct. The intuitive example is the rational burglar who risks a lot by searching the whole building for a valuable diamond.

Holmen goes on to summarise what he takes to be the upshot of Williamson’s burglar example:

The only way to understand why a burglar would take such a risk is, according to Williamson, by attributing her with the knowledge that the diamond is in the building. Otherwise it would be hard to explain why she apparently disregards evidence to the contrary—i.e. as the time goes and her search is not successful—on pain of diminishing her rationality (like declaring her to be just plain stubborn, insensitive to evidence etc.).

I am sceptical about the above argument. It seems to me that what is really doing the work in the burglar example is the certainty or conviction with which the relevant beliefs are held, rather than the fact that the beliefs are true (or, a fortiori, that the agent has knowledge). First, consider the burglar’s psycho-physical doppelganger who is just as convinced that there is a diamond in the building (based on the same body of evidence), but does not have knowledge because (unknown to him) a ridiculously improbably quantum occurrence has transformed the diamond into a block of coal. We would not expect the behaviour of the doppelganger to diverge from that of his twin just because his justified belief turns out (quite improbably) to be false. In short, if we stipulate that the level of certainty is identical in both cases, this is sufficient to explain or make sense of why each respective burglar persists in his search for the diamond. It seems to me that the fact that in one case the justified belief happens to be false (thus baring the agent of knowledge) in no way reduces our ability to make sense of or explain the agent’s behaviour.

It may be objected that while we can still make sense of or explain the doppelganger’s actions in light of his firm conviction, we would also be inclined to regard him as rationally irresponsible for having such a firm conviction, given that he lacks knowledge. But even if we grant that this is so (and I have my misgivings) the question is no longer one of making sense or explaining why the burglar did what he did. Instead, it has become one of his rational standing. In other words, this is to switch our focus from one of motivation (why did he do what he did) to one of justification (is he justified in doing what he did). Since belief-desire explanations (as I understand them) are only concerned with the former, the burglar argument is simply changing the subject of debate. A similar point can be made with regards to Hornsby’s knowledge requirement. Hornsby endorses the following claim:

(E) Where ‘x Φ-d because p’ gives a reason-explanation: x Φ-d because p iff x Φ-d because x knew that p.

On the view I have been urging, (E) is true so long as we understand a “reason-explanation” as a justificatory rather than explanatory explanation. Since the latter (but not the former) is a necessary condition for intentional explanation, the truth of (E) is irrelevant to the type of intentional explanation that the belief-desire theorist is concerned with. Holmen, however, advances a slightly different criticism of (E)—namely, that it entails a version of the KK principle:

I think I see a problem with (E): it seems to be a version of the KK principle and thus leads one to the absurd consequence that follows when one applies an S4 model for the accessibility relation that operates on the epistemic operator.

However, it is not clear that (E) entails any such control principle. For example, consider a rudimentary reliabilist account of knowledge; one according to which the naive chicken-sexer counts as knowing the sex of the chicks she reliably sorts. Let us suppose further that the naive chicken-sexer, being naive, does not know that she knows the sex of the chicks. In short, her knowledge is strictly of the first order. According to (E), if the naive chicken-sexer sorts some male chick M as male because she knows that M is male, then she sorts M as male for a reason. This remains true even though she does not know that she knows the sex of the chick. Thus, (E) does not entail the KK principle.

In summary, while I share both Holmen’s sympathy for and critical attitude towards Hornsby’s knowledge requirement for reasonable action, I do so for very different reasons. I believe Hornsby’s analysis goes wrong due to a conflation of explanatory and justificatory reasons. While Hornsby’s claims may be be true of justificatory reasons, they are not true of the type of explanatory reasons that constitute the bread and butter of the belief-desire theorist. Moreover, I have argued that Holmen’s primary objection to Hornsby’s theory—i.e., that it commits her to an implausible KK principle—misses its mark.