This paper is joint work of a philosopher and a well-known neuroscientist. You should have read the bulk of it by Tues., but our discussion will continue on to Thurs., when we will also discuss a short selection from a critical response. Section 4 of the paper is the longest and is also the place where Churchland and Sejnowski provide concrete examples of the sort of work they have in mind. I have included it in Tuesday’s assignment, making that much longer than Thursday’s will be. However, while it is important that you have some sense of the examples on Tues., we are unlikely to spend much time actually discussing them (most of the philosophy appears in other parts of the paper), so, if a more equal distribution of reading works better for you, you should read at least the introductory material in the section (i.e., through p. 256) and ideally also the discussion of NETtalk in §4.1, pp. 256-261.
Churchland and Sejnowski’s general aim is to suggest a model of cognition, “connectionism,” that is quite different from the sort of computational model described by Haugeland and Putnam. The latter approach is labeled in different ways: the term “symbolic computation” is sometimes used and that will fit the way Churchland and Sejnowski speak about it, and the critique of connectionism by Fodor and Pylyshyn that will discuss on the Thurs. will label the alternative they prefer the “Classical” view.
• Section 1 (pp. 247-249) is a discussion of general philosophical method that we probably won’t give much attention to.
• The key idea in section 2 (pp. 249-252) is “theory dualism.” The authors outline three motivations they see for this view and reply to two of them.
• Their response to the third motivation for “theory dualism” appears in section 3 (pp. 252-255). It is the key part of the discussion for our purposes. (You can regard the idea of “von Neumann architecture” as a rough equivalent of “Turing machine.” Von Neumann, a mathematician who made important contributions in wide variety of fields, was involved in the development, just after WWII, of actual computers that realized Turning’s theoretical idea of a “universal” Turing machine.)
• Section 4 is divided into four parts, an introductory section (pp. 255f) that is crucial for what follows (note especially fig. 17.2), discussions of two connectionist systems that Sejnowski was involved with (pp. 256-261 and 261-263) and suggestions of future prospects and problems (pp. 263f). Notice that the effort in this research is chiefly to provide a “proof of concept,” an indication that networks of this sort can perform certain functions, rather than to uncover the specific network structures that are exhibited in the actual neural systems.
• The final section (assigned for Thurs.) returns to the sort of issues discussed in sections 2 and 3. Most attention is given to the views of David Marr (1945-1980), a neuroscientist whose approach of vision was closer to classical symbolic computation than to connectionism.