Models of Environment

The thirteenth in a series of excerpts from Minds, Models and Milieux: Commemorating the Centennial of the Birth of Herbert Simon.

Marcin Miłkowski

Herbert A. Simon is well known for his account of bounded rationality. Whereas classical economics idealized economic agency and framed rational choice in terms of the decision theory, Simon insisted that agents need not be optimal in their choices. They might be mere satisficers, i.e., attain good enough goals rather than optimal ones. At the same time, behaviorally as well as computationally, bounded rationality is much more realistic.

One of the most important factors in his theorizing on bounded rationality was the structure of the environment of the agent (Simon, 1956). This might sound surprising today because Simon is all too often classified as one of the proponents of classical, symbolic cognitive science. After all, he favored symbolic models over situated action frameworks (Vera and Simon, 1993). However, already in his 1956 paper, he acknowledged that his account of bounded rationality is similar to robotic models built by Grey Walter. Moreover, Simon’s (1996) story about the ant that uses the environment to make the navigational task easier has become one of the classical examples for later proponents of the extended mind (Clark and Chalmers, 1998). So why did Simon stress the situatedness of cognition and denied that symbolic modeling is to be rejected? Was he deluded or self-contradictory?

The purpose of this chapter is to understand the role of the structure of the environment in Simon’s work on models of cognition. It will be shown that his modeling methodology includes both internals of the information-processing architectures and environmental constraints. The inner architecture is important insofar as it is a constraint on adaptation to the environment, and remains invariant over multiple different environments; hence, it is relevant to explaining behavior in any environment. For this reason, physical symbol systems are to be understood as both situated and adaptive; otherwise, they cannot be flexible and support cognition. Even if Simon’s treatment of symbols remains vague and underspecified, the idea that naturalistic models need to interleave internal and external states remains surprisingly timely.