Projectibility and reference

of the original article: The computational view of mind rests on certain intuitions regarding the fundamental similarity between computation and cognition. We examine some of these intuitions and suggest that they derive from the fact that computers and human organisms are both physical systems whose behavior is correctly described as being governed by rules acting on symbolic representations. Some of the implications of this view are discussed. It is suggested that a fundamental hypothesis of this approach (the "proprietary vocabulary hypothesis") is that there is a natural domain of human functioning (roughly what we intuitively associate with perceiving, reasoning, and acting) that can be addressed exclusively in terms of a formal symbolic or algorithmic vocabulary or level of analysis. Much of the paper elaborates various conditions that need to be met if a literal view of mental activity as computation is to serve as the basis for explanatory theories. The coherence of such a view depends on there being a principled distinction between functions whose explanation requires that we posit internal representations and those that we can appropriately describe as merely instantiating causal physical or biological laws. In this paper the distinction is empirically grounded in a methodological criterion called the "cognitive impenetrability condition." Functions are said to be cognitively impenetrable if they cannot be 302 THE BEHAVIORAL AND BRAIN SCIENCES (5) 2 Continuing Commentary influenced by such purely cognitive factors as goals, beliefs, inferences, tacit knowledge, and so on. Such a criterion makes it possible to empirically separate the fixed capacities of mind (called its "functional architecture") from the particular representations and algorithms used on specific occasions. In order for computational theories to avoid being ad hoc, they must deal effectively with the "degrees of freedom" problem by constraining the extent to which they can be arbitrarily adjusted post hoc to fit some particular set of observations. This in turn requires that the fixed architectural function and the algorithms be independently validated. It is argued that the architectural assumptions implicit in many contemporary models run afoul of the cognitive impenetrability condition, since the required fixed functions are demonstrably sensitive to tacit knowledge and goals. The paper concludes with some tactical suggestions for the development of computational cognitive theories. Is the "cognitive penetrability" criterion invalidated by contemporary physics? Peter N. Kugler", M. T. Tutvey, and Robert Shaw" 'Department of Psychology, University of Connecticut, Storrs, Conn. 06268; Haskins Laboratories, New Haven, Conn. 06510 and "Department of Psychology, University of Connecticut, Storrs, Conn. 06268 Pylyshyn (1980) advocates and extends a popular view that the model source for the explanatory concepts of cognitive science is the science of formal symbol-manipulating machines. The argument is that the proper vocabulary for constructing adequate explanatory theories of the knowings of animals and humans is the representational-computational vocabulary of computational science and artificial intelligence. The representational-computational perspective on knowings is far from recent; it has appeared in various forms for over two millennia, being intimately linked with philosophical attitudes variously termed "representational realism," "indirect realism," "idealism," and "phenomenalism." By and large, these attitudes follow from a distinction between the "physical" object of reference and the "phenomenal," or intentional, object that is said to be directly experienced and to which behavior is referred. It has been commonplace over the ages to question the coordination of the two kinds of objects, and it has seemed a simple enough matter to identify slippage between them. In consequence, it has frequently been concluded that the reference object might just as well be excluded from explanatory accounts; there are doubts that it can be known, and even doubts that it actually exists. The representationalcomputational vocabulary and its allied philosophical postures question or deny that the world is knowable. Animals and humans can only know the phenomena (sense data, representations, etc.) that their brains or minds supply (see Fodor 1980). In sum, philosophy and science have been unable to characterize the animal-environment relation in a way that allows that what animals know is real, existing independently of their knowing it. This state of affairs is curiously tolerated despite its obvious contradiction of the scientific enterprise (see Shaw & Turvey 1980 on Fodor 1980). Among the many assumptions and intellectual commitments that prohibit a realist posture (see Shaw & Turvey 1981; Shaw, Turvey & Mace, in press) is the assumption that contemporary physical theory is complete. The complete theory's failure to accommodate regularities in biology or psychology gives license to propose new, often special in the sense of extraphysical principles. Pylyshyn proposes "cognitive penetrability" as a methodological criterion that is sufficient (but not necessary) to distinguish those phenomena whose explanation requires the privileged vocabulary of representation and computation from those phenomena that can be appropriately described by physical law. Our reading of what is necessary for the "cognitive penetrability" criterion is a good deal more general than Pylyshyn's, but we believe it to be accurate. The necessary condition is that the behavior of the system in question be nondetenninate, that is, not dominated by boundary and initial conditions. As we describe below, this necessary condition is met by a broad class of physical systems termed "dissipative structures," systems that are indeed "mere" instantiations of physical law and, therefore, by the criterion, systems that do not entail the representationalcomputational vocabulary. It seems to us that the criterion is diluted, if not invalidated, by recent extensions of physical theory. Because of this fact, we question its completeness and its propriety for natural phenomena. Before turning to a description of dissipative structures, let us remark on an aspect of Pylyshyn's argument that we find especially puzzling the conjunction of Pylyshyn's pursuit of nondeterminacy as the necessary condition for genuine cognitive processes and his advocacy of formal symbol-manipulating machines as the model source for explaining such processes. Pylyshyn wishes to earmark for cognitive science behavior that is not determinately bound to environmental events; such behavior, it is argued, can be accounted for exclusively by the representational-computational vocabulary. However, no suggestion is given of how the various algorithms and representatons are to be nondeterminately selected. Computational devices are all determinate machines in which the output is completely specified by the initial conditions (input) and boundary conditions (algorithms and representations). Oddly, by selecting the formal symbol-manipulating machine as his model source, Pylyshyn, like other proponents of his view, fails to offer any nontrivial distinction between the popular model of cognition and any prototypic behaviorist model, that is, between cognitive science and behaviorism. Dissipative structures as consequences of conditions on natural law. An analogue to Pylyshyn's "penetrability" condition can be shown to exist in physical systems governed by natural law when such systems are construed as dissipative structures. Although this idea requires careful and complete development, a sketch of the argument can be given as follows: Classical reversible equilibrium thermodynamics describes the thermodynamic behavior of a system only when the system is in or near a state (condition) of equilibrium. In addition, the system may exchange neither matter nor energy with its surrounds. Systems meeting these conditions are referred to as isolated closed systems. The behavior of these systems is characterized by a tendency to run down to a maximum state of disorder, zero information, and loss of the ability to do work (Bridgeman 1941). This behavioral state is entropic equilibrium, and once a system is in this state nothing new can emerge as long as the conditions of the system remain isolated and closed. Under these conditions, the thermodynamic analysis is complete. The reversible quality of these systems is evident in the fact that if a perturbation occurs to the system under these conditions, the system responds by going through a succession of states, all of which are at entropic equilibrium. In short, the entire event occurs in a state space in which all points in the space are homogeneous with respect to entropic equilibrium. The concept of reversibility is reflected by the fact that there are no preferred points in the entropic state space: States may reverse themselves and still maintain the condition of entropic equilibrium. Under these conditions the system's behavior is completely determinate and specified by initial and boundary conditions. Such conditions do not allow for the possibility of autonomy or self-organization. While some real events (such as very slow processes in the macroworld) are THE BEHAVIORAL AND BRAIN SCIENCES (5) 2 303 Continuing Commentary rather well described by the conditions surrounding classical reversible equilibrium thermodynamics, most interesting events regarding biological and psychological systems are not. Our suspicion is that Pylyshyn's concept of "natural laws" is based on the above conditions, namely those of an isolated, closed (thermodynamic) system. We would suggest, however, that a model for a biological or cognitive system is poorly represented by the conditions of isolated, closed systems. A more appropriate model might be found in the less familiar conditions of open physic