33

Reasoning and Reflection: A Reply to Kornblith 1. Introduction Hilary Kornblith’s book is motivated by the conviction that philosophers have tended to overvalue and overemphasize reflection in their accounts of central philosophical phenomena. He seeks to pinpoint this tendency and to correct it. Kornblith’s claim is not without precedent. It is an oft-repeated theme of 20th-century philosophy that philosophers have tended to give ‘overly intellectualized’ accounts of important phenomena. One thinks here of Wittgenstein, Ryle and many others. One version of this charge is that philosophers have tended to appeal to higher-order thoughts when first-order thoughts about the world are all that’s needed. A more specific version of this charge is that philosophers have tended to appeal to second-order thoughts with normative , or quasi-normative, contents when all that’s needed are first-order thoughts with factual contents. It is this second version of the charge that Kornblith is particularly interested in pressing. Although he doesn’t spell it out, the connection between this project and Kornblith’s previous work on naturalistic conceptions of epistemology should be fairly obvious. Very roughly, if you want humans to look a lot closer to the lower animals, then you’d better think that most central human abilities can be explained without appeal to reflection and without appeal to normative thought. What’s good and important about Kornblith’s book is that he gives this charge a sustained and illuminating treatment. He looks in detail at accounts of knowledge, reasoning, epistemic agency, free will and normativity; he identifies sympathetically some of the temptations to think that we must resort to second-order resources to explain these phenomena; and he attempts to show that the appeal never works and is, in any case, not needed, since first-order accounts manage very well. 2. Causality + ? = reasoning I will look at the case of reasoning in some detail. I pick this case partly because it interests me, but partly also because it provides Kornblith with his strongest hand: reflection-based accounts of free will, or normativity, or justification, may appeal to varying degrees; but reflection-based accounts of reasoning look downright treacherous. As we will see, though, even in this most unpromising of cases, naturalist accounts that give no role at all to reflection seem utterly implausible. 2 What is it for you to reason from p to q ? All parties are agreed that for such reasoning to take place your acceptance of p must cause your acceptance of q. But, clearly, not every such causal transition would be reasoning. 3 Kornblith agrees in setting up the issue in this way: If I am to reason from one belief to another – if the first belief is to be part of my reason for believing the second – then the first belief must have played some causal role in producing (or sustaining) the second belief. But surely this is not enough. There are, after all, lots of ways in which one belief might play a role in producing or sustaining a second belief which would not constitute either reasoning or believing for a reason. … So what else is required if one’s belief is to be one’s reason, or part of one’s reason, for holding a second? (42) In response (as we will further see below), some philosophers have suggested that it is enough to add that the transition from the first belief to the second be sensitive to the judgements’ contents. It must be in virtue of the belief’s contents that the first causes the second. But that is clearly not enough. In a habitually depressive person, the judgement ‘I am having so much fun’ may routinely cause and explain his judging ‘But there is so much suffering in the world’; this transition would be content-sensitive, but would not be reasoning. It would be content-sensitive associative thinking. At this point, you might be attracted to the suggestion that the depressive’s thinking cannot count as reasoning because his first judgement – that he is having so much fun – doesn’t support his second – that there is so much suffering in the world. Indeed, the first judgement might be thought to (somewhat) undermine the second. But that can’t be a good suggestion – it conflates reasoning per se with good reasoning. Although this conflation is quite common in the literature, it must be avoided: our theory of reasoning must allow for bad reasoning. And one way that reasoning can be bad is for its premise to fail to support its conclusion. What, then, should we say about the difference between a merely causal transition and reasoning? If we think about our depressive’s associative thinking, the answer seems to leap out at us. The reason that his thinking doesn’t count as reasoning is not (as we’ve said) because his first judgement doesn’t support his second, but, rather, because he doesn’t take it that his first judgement supports his second. The first judgement simply suggests the second one to him; he doesn’t draw the conclusion because he takes it to be supported by the first. This way of thinking about reasoning echoes a remark of Frege’s ( 1979 ): To make a judgment because we are cognisant of other truths as providing a justification for it is known as inferring. (3) Frege’s characterization needs some tweaking. We needn’t always infer from truths, if we are to count as inferring. We may reason from falsehoods, or even from imperatives, which are incapable of truth or falsehood. Furthermore, ‘cognisant’ has a success grammar built into it, and so Frege’s characterization implies that one cannot reason badly. But, as I’ve been emphasizing, it is possible to reason badly. So, I would rather put it like this: But this is clearly a view in the same spirit as Frege’s. And it is one that gives a crucial role to some sort of reflective element – the taking. • (Taking) A transition from some acceptances to a conclusion counts as inference only if the thinker takes her conclusion to be supported by the accepted truth of those other acceptances, and draws her conclusion because of that fact. Now, there are, of course, many different ways of elaborating such a view. As formulated so far, it is just a schema for a view, one which signals that something has to play the role of ‘taking’ but without specifying exactly what it is that plays that role, nor how it plays it. In some of my other work, I have looked at different versions of such a view in some detail ( Boghossian 2014 ). I will return to this issue further below. 3. The adequacy of first-order views of reasoning Before doing so, however, I want to look at Kornblith’s arguments for the claim that no account of reasoning requiring a reflective element could be correct. Kornblith has both a positive and a negative argument for this claim. His positive argument proceeds via the claim that: His negative argument, on the other hand, depends on claiming that: • (A) We know of clear-cut cases of reasoning that incorporate no reflective element, hence no such condition could be necessary. • (B) No view of reasoning incorporating such a second-order element can be made to work. I will look at each argument in turn. The supposedly clear-cut cases of non-reflective reasoning include both non-human and human instances. The non-human instances are of special interest since, if they were convincing, they would clinch the case, its being widely accepted (it’s not entirely clear why) that non-human animals are incapable of reflective thought. 4. The case of the clever plover What are these allegedly clear cases of non-human reasoning? Kornblith offers us a contrast between the tongue-flicking behaviour of a frog and the broken-wing display of a piping plover. The first is said to illustrate mere stimulus-bound behaviour; but the latter is said to furnish a clear case of reasoning: The frog’s response to flies is … [stimulus bound]. If one rolls a piece of birdshot (BB) in front of the frog, it will suck it up as if it were a fly. Well, we all make mistakes. But the frog will do it a second time if a second BB is rolled in front of it; and a third; … The frog will not learn from its experience. (51) By contrast, the piping plover seems to show a greater capacity for adapting productively to its experience. When a nesting piping plover detects a potential predator approaching its nest, it feigns a broken wing, luring the predator away from the nest; once the predator is far enough away, the bird flies off. Kornblith writes: One may reasonably ask whether this behavior is simply stimulus-bound. After all, there are behaviors like this in the animal world which are simply wired in, like the frog’s tongue-flicking, and show no sign of sensitivity to other information which the animal may, in some sense, possess. But the piping plover’s broken wing display is not like this. Thus, for example, if a human being walks on a path which would come close to the nest, the plover will approach the individual and perform the broken wing display. But if that person continues on her path without disturbing the nest, and follows the same path on a daily basis, then the plover will stop performing the display. When a different individual approaches, however, on the very same path, the plover goes back to the broken wing performance. This is not some stimulus-bound performance. In order to explain it, we need to see the plover as picking up information about its environment and integrating it with stored information about the past. (51–52) There is no disputing that there is an important difference between the behaviour of the frog and that ofthe plover. The frog is programmed to perform a particular action (flick its tongue) whenever it registers a certain bit of information about its environment – the presence of a fly-like object. The plover is capable not only of registering information about its environment, but of having those informational states interact with one another, thus allowing it to have more sensitive interactions with its environment. Kornblith, however, goes on to interpret this distinction between the frog and the plover as a distinction between behaviour that is merely ‘wired in’, versus behaviour that manifests a capacity to reason. He says: When the plover discovers, after repeated interactions with this particular individual, that no threat is present, she ceases to respond by engaging in the display. This requires the integration of new information with old, and this is why we need to posit a representational system to account for such sophisticated behavior. But, as I see it, there is now every reason to regard these informational interactions as cases of reasoning : they are, after all, transitions involving the interaction among representational states on the basis of their content. (55, emphasis added) This, of course, is the point at which, if we are to get clear on the philosophical issue about reasoning, we need to go slowly. Why is there ‘every reason’ to regard the interactions between the plover’s informational states as reasoning? Kornblith could just be thinking, I suppose, that it is intuitively clear that the plover reasons. But that really doesn’t seem right. Intuitively, one recoils not only from the claim that the plover reasons , but even from what looks to be the more modest claim that it has beliefs. Rather, the plover looks to have only informational states and algorithms governing the transitions between them. And prima facie, anyway, the difference between the frog and the plover is not a difference between behaviour that’s ‘wired in’ as opposed to behaviour that’s the result of reasoning; rather it appears to be the difference between one sort of behaviour that’s wired in and a slightly more complex form of behaviour that’s wired in. Whereas the frog’s behaviour looks to be governed by a rule with a single type of input and a single type of output, If C, stick out your tongue and suck it up! the plover’s broken wing behaviour is governed by a more complex rule: If C and not D, display broken wing behaviour; otherwise, carry on! Or something like that. I don’t see that a slight increase in the complexity of a hard-wired algorithm governing a creature’s behaviour can, all by itself, make for the difference between reasoning and mere mechanism, no matter how much the more complex algorithms might contribute to a ‘more sensitive’ interaction with the environment. In fact, Kornblith doesn’t seem to be primarily relying on intuition at all here. Instead, he seems to think that his case is secured by an argument , one of whose premises is that reasoning is nothing other than transitions among informational states on the basis of their content: But, as I see it, there is now every reason to regard these informational interactions as cases of reasoning: they are, after all, transitions involving the interaction among representational states on the basis of their content. (55, emphasis added) This is clearly not a sound argument. An alarm system has the capacity to make transitions between informational states on the basis of their content, without its being true that it is capable of reasoning. Kornblith could try insisting that we restrict our attention to processes that occur in living organisms; but that still wouldn’t help. The process of breathing , for example, would count as reasoning on that criterion. Breathing is controlled automatically and subconsciously by the respiratory centre at the base of the brain. Sensory organs in the brain and in the aorta and carotid arteries monitor the blood and sense its oxygen and carbon dioxide levels. When these sensors detect an increased concentration of carbon dioxide in the blood, the brain stimulates the lungs to breathe more deeply and frequently. These transitions ‘involve interaction among representational states on the basis of their content’, but do not count as reasoning. Kornblith could also try insisting that we should restrict our attention to transitions between informational states that are beliefs , thereby ruling out both the alarm systems example and the breathing example. This would be a problematic move in at least two respects. First, we don’t want to rule out practical reasoning or reasoning with imperatives or suppositions. But, secondly, and more importantly, why should we be so confident that the plover has beliefs? Kornblith is just as confident that the plover has beliefs as that it reasons. He says: I want to agree … that having beliefs requires a sensitivity to reasons, and I believe that the behavior of the plovers … shows that they are sensitive to reason. When the plover encounters a reason to stop engaging in the broken wing display, it stops; … As I see it, meta-cognitive evaluation is only one way to show a sensitivity to reasons. (56)