SOAR as a world view, not a theory

Newell's SOAR proposal is just that, soaring. It attempts to gather almost all of cognitive psychology under a single theoretical tent. Newel1 distinguishes several levels or, as he sometimes calls them, bands of human behavior and suggests that qualitatively different laws apply to each. In Newell's terms, there are seals between each level. In particular, he distinguishes between biological, cognitive, and rational bands, with their own explanatory laws. (He also discusses a social band, but we shall not.) The thrust of his argument is that the SOAR program is an appropriate vehicle for building theories of thought at the cognitive and rational levels. This argument implicitly assumes that complex programs such as SOAR are appropriate ways of expressing theories of cognition. We raise two auestions. Do these three bands indeed "carve nature at its joints," so that self-contained theories are possible within each? And is SOAR, or anything like it, an acceptable form for a theory? Newell's biological band is defined, conceptually, by such physical mechanisms as neural transmission. He argues that basic events in this band span from 10 to 100 msec, and the relevant laws are physical and causal. The cognitive band spans from 0.1 to 10 sec and encompasses the phenomenon of information-processing psychology. In particular, symbols are manipulated with little regard to their semantics. An example is the operation of fetching an item from long-term to working memory. Finally, the knowledge level deals with the semantics of symbol manipulation. The congressman who says "When in doubt I always vote against Congressman Y" is operating at this level. Causal relations here reflect the semantics of the world. The notion of a clear distinction between the biological and cognitive levels is directly challenged by the success of connectionist models designed to derive information-processing functions, such as the Hick-Hyman law (Keele 1986, pp. 30-35) from models of biological organization. To the extent that connectionism is successful, it directly challenges Newell's position. We are equally skeptical about a seal between the upper-level cognitive and the representational bands. According to Newell, at the upper level of cognition actions take upwards of 10 sec to execute, which is well beyond the time it takes to comprehend a moderately complex sentence and incorporate its meaning into one's representation ofthe text. Clearly, the semantics ofwhat is being read affects both the strategy used and the ability to incorporate new information into a text. The seal leaks downward from knowledge to cognition. Similarly, the knowledge level is influenced by working memory limitations, especially in situations in which knowledge must be accessed quickly. This is an example of how actions in Newell's knowledge band may be sharply constrained by causal relations at the informationprocessing level. This is not to deny the existence of purely