Treating connectionism properly: Reflections on Smolensky

Summary

In this essay, I undertake to examine the principal theses of Paul Smolensky's 1988Behavioral and Brain Sciences target article, “On the proper treatment of connectionism,” from the point of view of the methodology and epistemology of science, that is, the philosophical theory of theories in general. After exploring the instrumentalist and realist views of the relationships between micro- and macrotheories on their ”home ground” in the natural sciences, the procedures by which a phenomenally described cognitive task is “prepared” for symbolic or subsymbolic modeling, and the contrast between the deliberate conscious reasoning processes of a novice and “intuitive” behavior of an expert in solving a given family of cognitive problems, I argue that although Smolensky is right about what it would take for connectionist subsymbolic models to relate to symbolic models as micro- to macrotheories, he is wrong in concluding that they do. On the contrary, it would be something of a miracle if the idealized nomological structure of the behavior of stochastic patterns of activity over large numbers of subsymbolic units in a connectionist machine corresponded even approximately to the nomological structure of the “conceptual level” behavior of a Von Neumann computer running off a program whose syntax had been explicitly designed to structurally operationalize a determinate fragment of intentional semantics — unless, of course, the connectionist machine had been deliberately constructed to implement the symbol processor in the first place.

I conclude that the proper treatment of connectionism is to be found among the “blandly ecumenical” proposals for irenic cooperation and division of labor that Smolensky considers and rejects.