On Modularity

If Lashley's (1929) idea of mass action were valid, then neuropsychology would be of little relevance for understanding normal function. Any form of neurological damage would deplete by a greater or lesser degree the available amount of some general resource, say the mythical g. Knowing which tasks a patient could or could not perform would enable us to partition tasks on a difficulty scale. It would tell us little, if anything, about how the system operated.

If one considers the design principles that might underlie cognitive systems, a rough contrast can be drawn between systems based on equipotentiality (e.g., those that follow principles such as mass action) and more modular ones. At a metatheo-retical level, some form of the ‘modularity’ thesis is probably the most widely accepted position in the philosophy of psychology today (e.g. Chomsky, 1980; Morton, 1981; Marr, 1982; Fodor, 1983). Over the past 30 years, the arguments for the position have become increasingly compelling and diverse; they are of a number of different types: computational, linguistic, physiological, and psychological.

The most basic argument is the computational one, which we owe to Simon (1969) and Marr (1976). In Marr's words.