The New Science of the Mind Read online

  The distinction between making information available to a subject and making it available to subsequent processing operations is an important one and, indeed, will play a crucial role in the arguments to be developed in the second half of this book. The distinction corresponds to that between what are known as personal and subpersonal cognitive processes. Processes that make information available to a subject are personal-level processes, and this is true even if they also make information available to subsequent processing operations.' However, processes that make information available only to subsequent processing operations are subpersonal cognitive processes. This distinction corresponds to two different ways in which a cognitive process might belong to a representational subject, and will be discussed in more detail later on.

  Condition (3)

  It is currently fashionable in some quarters to suppose that cognitive science can do away with the idea of representation. I think this fashion is seriously confused, but perhaps that is to be expected when one is dealing with a multiply ambiguous concept such as representation. However, recall that this criterion is presented only as a sufficient condition for a process to count as cognitive, and not a necessary condition. So, if you believe that cognition does not require representation, that is fine with me-at least for the purposes of the arguments to be developed in this chapter. There are, however, important strategic or dialectical reasons for me to formulate the criterion of cognition in terms of the idea of representation.

  First of all, as we shall see, certain objections to the thesis of the extended mind-versions of the differences argument-are based on the assumption that cognition involves representation (indeed, representation of a quite specific sort). To deny this at the outset would, therefore, invite charges of question-begging. Second, the invocation of mental representations has, at least until recently, been a staple of cognitive-scientific theorizing. The denial that representations are required for cognition is still controversial, and in developing my argument for the theses of the extended and embodied mind, I want to rely on as few controversial assumptions as possible. I am going to use the criterion of cognition to argue for the theses of extended and embodied cognition, and so to defend the new science in general. So, in developing the criterion of cognition, I want to give the objector everything he or she could reasonably want. I am going to focus on the most respectable, conservative, even reactionary versions of cognitive-scientific practice it is possible to find, and I am going to defend the criterion of cognition on the basis of this sort of practice. And then I am going to argue that the theses of the extended and embodied mind still follow. In other words, the criterion of cognition, I shall argue, has conservative origins-origins that even the most dyed-in-the-wool defender of tradition would have to accept-but radical consequences.

  Now that I have explained why the criterion of cognition invokes the idea of representation, some further clarifications are required concerning what this invocation involves. The third condition takes no stand on whether representation is, ultimately, a naturalistic phenomenon. Many suppose that it is possible to supply broadly naturalistic conditions of representation, and that a state will count as representational if, and perhaps only if, it satisfies these conditions. This naturalistic assumption is neither mandated by, nor incompatible with, condition (3).

  There are three further distinctions we need to observe in the formulation of condition (3). The first is the distinction between representational and semantically evaluable; the second is that between derived and nonderived representation; and the third is that between personal and subpersonal representations.

  (a) Representational versus semantically evaluable As I shall use the expression, a state is semantically evaluable if and only if it has truth conditions: that is, it is the sort of thing that can be true or false. The notion of a representational state is not coextensive, still less synonymous, with the notion of a semantically evaluable state. It would be implausible to suppose that all representational states are semantically evaluable: such states must possess adequacy conditions, but they need not possess truth conditions. Mental models, or cognitive maps, possess adequacy conditions but not truth conditions (McGinn 1989a,b). That is, mental models or cognitive maps can be accurate or inaccurate, but they cannot be true or false. They cannot be true or false because the concept of truth is essentially bound up-for familiar Davidsonian reasons-with the logical connectives (not, and, or, and variants thereof); and mental models do not relate to these connectives in the same way as sentences. Thus, for example, the negation of a sentence is another specific sentence; but to the extent that it makes sense to speak of the negation of a map or model, this can mean nothing more than a distinct, but nonspecific, map or model. Similarly, the disjunction of two sentences is another sentence; but the disjunction of two maps or models is not another map or model. Once we understand the connection between truth and the logical connectives, I take it that this point is incontrovertible, though often oddly neglected. However, nothing much will turn on it in the arguments to follow. But it is perhaps worth emphasizing that the commitment to representational states in condition (3) does not involve commitment to the idea that these representations are sentence-like, and so does not entail commitment to a language of thought. (b) Derived and nonderived content The type of representational state invoked in (3) is, in a sense to be clarified shortly, a nonderived one. That is, it is a representational state that possesses nonderived content. Derived content is content, possessed by a given state, that derives from other representational states of a cognizing subject or from the social conventions that constitute that agent's linguistic milieu. Nonderived content is content that does not so derive. A form of content's being nonderived is not equivalent to its being irreducible or sui generis: nonderived content can, for example, "derive" from, and be explained in terms of, the history or information-carrying profile of the state that has it. It is what content is derived from that is crucial. Nonderived content is content that is not derived from other content; it is not content that is irreducible or sui generis.

  The existence of nonderived content is controversial. However, there are at least three reasons for understanding the representational state invoked in condition (3) as being a state that possesses nonderived content.

  First, the claim that nonderived content is central to cognition has been used to attack the thesis of the extended mind. For example, Adams and Aizawa (2001) attack Clark and Chalmers's claim that the sentences in Otto's notebook constitute a subset of his beliefs precisely on the grounds that these sentences possess only derived content. These sentiments are echoed by Fodor (2009). We shall return to this attack later, where I shall argue that it is unsustainable. However, at this stage, I might be regarded as unacceptably slanting discussion in favor of the extended mind were I to allow that the representational states invoked in condition (3) need possess only derived content. Recall that the strategy to be prosecuted in this chapter is to give the objector to the amalgamated mind everything he or she could reasonably request, to develop a criterion of cognition on this basis, and then argue that the theses of the embodied and extended mind still follow. Insisting that the representational states invoked in condition (3) are nonderived is required to give the objector everything he or she could reasonably want.

  Second, the strategy I am going to employ in defending the criterion of cognition turns on an examination of cognitive-scientific practice. I shall argue that the criterion can be extracted, in a relatively straightforward way, from this practice. As we shall see, however, the postulation of nonderived representational states (states that possess nonderived content) is a staple of such practice.

  Third, there are certain reasonable expectations concerning what a criterion of the cognitive will allow us to do. The goal is, of course, to understand what cognition is.3 And the strategy is to do this by way of the provision of a criterion that allows us to demarcate, with a reliable though not necessarily infallible level of precision,
those processes that count as cognitive from those that do not. Derived content, however, is content that derives from the operation of cognitive processes. So to allow that the representational state specified in (3) need be one that possesses only derived content would, it seems, undermine the primary rationale for the criterion.

  One final, but important, point of clarification regarding the notion of the "nonderived" is required, and this can best be made with respect to Marr's model of vision we examined earlier. Marr's account begins with the retinal image. Visual processing transforms this successively into the raw primal sketch, full primal sketch, and 21/2D representation. Does the content of the full primal sketch derive from that of the raw primal sketch? In one sense, it obviously does. The full primal sketch has the content it has only because the raw primal sketch has the content it has, and this content was transformed by various rules of visual processing. However, this is not the sense of derived presupposed in the condition of nonderived content. If this counted as derived content, then there would be almost no nonderived content in cognitive theorizing. At most, the beginning of processing streams would consist in structures with nonderived content. All structures downstream would possess only derived content.

  With this in mind, recall the dialectical situation. I am agreeing to the condition of nonderived content because such content has been used as an objection to the extended mind, and I don't want to be accused of stacking the deck at the outset. However, if we adopt a conception of nonderived content such that it can be possessed only by structures at the very beginning of processing streams, then almost nothing turns out to have nonderived content. But then it becomes unclear, to say the least, how nonderived content can be used as an objection to the extended mind. In other words, those who wish to object to the extended mind on the grounds of nonderived content (e.g., Adams and Aizawa, Fodor) are committed to denying that the content of the full primal sketch is derived simply because it comes from that of the raw primal sketch. And, as I have been at pains to emphasize, I am simply giving the objectors to the extended mind what they ask for. Thus, the relation between the content of two subpersonal states within the same processing stream, one of which causally succeeds the other, cannot be regarded as a relation of derivation in the required sense. If we are to make a case that the content of one of these states is derived, we must look to factors outside the processing stream.'

  (c) Personal and subpersonal The distinction between derived and nonderived representations is not equivalent to the distinction between personal and subpersonal states. Personal and subpersonal states can, in principle, have both derived and nonderived content (although the conditions under which the latter would possess derived content would be extremely unusual). More important, for the purposes of this book, is the claim that both personal and subpersonal states can possess nonderived content; they can both be representational in a nonderived sense. This claim is disputed by some, but any disagreement with the position advanced here is likely to be a matter of stipulation rather than substance. Thus, I shall assume that subpersonal states can be representational; but I shall not assume that they are representational in the same way that personal states are. I shall assume, equivalently, that subpersonal states can possess a content of sorts; but I shall not assume that it is the same sort of content as that possessed by personal states. The dispute concerning whether subpersonal states are representational or possess content almost invariably turns on the issue of whether the (allegedly) representational properties, hence content, possessed by subpersonal states are the same sort of thing as the representational properties or content possessed by personal states. And the denial of the status of representation or content to the former derives from the claim that the two are insufficiently similar. The content of my claim that subpersonal states can be representational is given by the claim that they satisfy broadly naturalistic criteria of representation of a familiar sort. This is compatible with being noncommittal on the issue of whether representation is, in general, a naturalistic phenomenon. If personal-level representation turns out to not be naturalistic, then it remains true that subpersonal representation almost certainly is. The primary reasons for denying naturalistic status to personal-level representation turns on the relation between such representation and consciousness-the failure to naturalistically explain the latter being thought to legislate against the possibility of a naturalistic account of the former (e.g., McGinn 1991). However, since subpersonal states are not conscious, these sorts of considerations have no echo at the level of subpersonal states.

  The relevant naturalistic criteria of representation applied to subpersonal representations will almost certainly look something like this: Subpersonal representational states must (i) carry information about the world, (ii) have the function of carrying such information, or the function of allowing the cognizing organism to achieve a given task in virtue of carrying this information, (iii) be capable of misrepresenting the world, (iv) are decouplable from the world, (v) have an appropriate structure that permits combination with other states, and (vi) play an appropriate role in guiding the organism's behavior. One need not, of course, accept all of these conditions; and each condition is susceptible to a variety of interpretations. They are provided merely as the sorts of conditions a naturalistic account of representation will supply. I take no stand on the precise character of these naturalistic conditions for one very simple reason. The strategy I am going to follow is to defend the criterion of the cognitive presented above by way, in the first instance, of a careful examination of cognitive-scientific practice. And the postulation of subpersonal representational states is a staple of traditional cognitive-scientific practice.

  Condition (4)

  As we shall see, perhaps the most difficult implication of taking the thesis of the amalgamated mind seriously lies in understanding the sense in which cognitive processes belong to cognizing subjects or organisms. I am going to argue that cognitive processes can belong to a subject in two different ways, one pertaining to the personal level, the other to the subpersonal level. The personal level, I shall argue, is primary. Cognitive processes occurring at the subpersonal level belong to the subject to the extent that they make an appropriate contribution to cognitive states and processes occurring at the personal level. This idea is straightforward, but surprisingly difficult to render precise. I shall attempt to do so in terms of the concept of integration: subpersonal cognitive processes belong to the subject to the extent that they are appropriately integrated into the personal-level states and processes possessed by a representational subject. The key, then, is to provide an account of how a personal-level cognitive process can belong to a representational subject. This is the task of the second half of the book.

  I am going to motivate and defend conditions (1)-(3) separately from condition (4). Conditions (1)-(3), I shall argue, can be extracted from examination of various standard models of cognitive-scientific practice. Condition (4) is also implicit in this practice, but the bulk of its defense will be based on more general grounds of plausibility. Defense of (4) will be postponed until the next chapter.

  4 Defending the Criterion: Cognitive-Scientific Practice

  I shall defend conditions (1)-(3) of the criterion by showing that they can be extracted, in a relatively straightforward manner, from examination of cognitive-scientific practice. The guiding thought is that if we want to identify a mark of the cognitive, then we had better pay close attention to the sorts of processes that cognitive scientists regard as cognitive, and then try to identify the general features of processes of these kinds. However, to avoid the charge that the criterion is motivated by amalgamated mind aforethought, the cognitive-scientific practice in question must be internalist cognitive science-and the more typical or paradigmatic the form of internalism the better. In this respect, they simply do not come any more typical and paradigmatic than David Marr's (1982) theory of vision. Although many of the details of this theory are now starting to look decided
ly quaint, the general approach instigated by Marr has, as I argued earlier, both dominated and shaped internalist-inspired theorizing in cognitive science.

  To recap: for Marr (1982), visual perception begins with the formation of an informationally impoverished retinal image. The function of properly perceptual processing (which is cognitive with a small "c" but not with a large "C") is to transform this retinal image into, successively, the raw primal sketch, the full primal sketch, and then the 21/2D sketch-the culmination of properly perceptual processing. At each stage in the operation, one information-bearing structure is transformed into another. The retinal image, reputedly, contains very little information, but it does contain some. The retinal image is made up of a distribution of light-intensity values across the retina. Since the distribution of intensity values is nomically dependent on the way in which light is reflected by the physical structures that the organism is viewing, the image carries some information about these structures. The first stage in visual processing consists in transforming the retinal image into the raw primal sketch. In the raw primal sketch, information about the edges and textures of objects has been added. Application of various grouping principles (e.g., proximity, similarity, common fate, good continuation, closure, and so on) to the raw primal sketch results in the identification of larger structures, boundaries, and regions. This more refined representation is the full primal sketch. And so on.