New Releases by Andy Clark

Clark charts a fundamental shift from a static, inner-code-oriented conception of the subject matter of cognitive science to a more dynamic, developmentally rich, process-oriented view.

Abstract: "The paper considers the problems involved in getting neural networks to learn about highly structured task domains. A central problem concerns the tendency of networks to learn only a set of shallow (non-generalizable) representations for the task i.e. to ''miss'' the deep organizing features of the domain. Various solutions are examined including task specific network configuration and incremental learning. The latter strategy is the more attractive since it holds out the promise of a task-independent solution to the problem. Once we see exactly how the solution works, however, it becomes clear that it is limited to a special class of cases in which (1) statistically driven undersampling is (luckily) equivalent to task decomposition, and (2) the dangers of unlearning are somehow being minimized. The technique is suggestive nonetheless, for a variety of developmental factors may yield the functional equivalent of both statistical and ''informed'' undersampling in early learning."

Connectionism is currently one of the most flourishing and interdisciplinary areas of cognitive science. Drawing on research in neural computation and networks it has found applications in areas such as psychology and animal intelligence. By using types of network which attempt to mirror our own cognitive architecture, connectionism is making breakthroughs in the understanding of the human mind a real possibility.

We show that a popular class of connectionist models (which we label ''first order connectionism'') looks unlikely to provide the kind of resources required by the hypothesis. We examine some alternative hybrid models that seem more promising. Finally, we raise a more purely philosophical issue concerning the conditions under which a being can count as a genuine believer or cognizer."

Microcognition provides a clear, readable guide to parallel distributed processing from a cognitive philosopher''s point of view.

Abstract: "NETtalk is a connectionist system able to negotiate a certain problem domain. It may even be said, somewhat loosely, to have knowledge of that domain. Likewise, various lower animals know their way around certain problem domains. But, in a sense to be explained, such systems lack structured knowledge of the domain. They lack explicit abstractions which represent constraints on properties common to a whole class of inputs all of which they can nonetheless deal with appropriately. NETtalk''s knowledge of the entity ''vowel'' is a good example. The paper draws together and criticises a number of perspectives on this basic idea.

Abstract: "Dennett''s treatment of belief and Smolsnsky''s [i.e. Smolensky''s] description of cognition are united by a common lacuna. Each seeks to accomodate troublesome facts by augmenting a basic cognitive system with a language manipulating device. In Dennett, this surfaces as the distinction between belief and opinion. In Smolensky, it surfaces as the Conscious Rule Interpreter. But neither author comments on the architectural conditions necessary for the development of such a device. In the present paper I first exhibit this lacuna, and try to make it pressing. I then sketch a possible solution based on some of the computational architecture which it seems to require. The upshot of this is a two factor theory of belief in which broadly architectural constraints must be met before a system is even a candidate for having beliefs. But the fuzziness and indeterminacy of the latter no longer commit us (as they commit Dennett) to treating the distinction between believers and non-believers in an instrumental or quasi-instrumental fashion."