Given a formally well-defined task, a rational actor model defines a rationally justified, optimal response. Rational actor models are desirable goals in the behavioral, cognitive, and social sciences, but in this chapter we use the distinction between small and large worlds to question the cachet associated with the terms “rational” and “optimal.” Ideally suited to the analysis of small world problems, both concepts can be counterproductive in the analysis of large world problems. In small worlds, the relevant problem characteristics are certain and uncontroversial in their formalization. For example, a tin can manufacturer seeking to minimize the tin used to package 12 ounces of soup might use solid geometry to determine an optimal can design. In this small world the manufacturer is safe in claiming that no other can design uses less tin. Large worlds are characterized by inherent uncertainty and ignorance, properties which undermine the validity and existence of optimal responses. An aircraft manufacturer designing a flight control system, for instance, faces a large world problem due to the complexity and uncertainty of the operating conditions. Rational actor models may rest on rigorous formal foundations, but they can also signify the questionable use of small world methods to understand large world problems. With similar concerns, Savage introduced the distinction between small and grand worlds when assessing the limits of Bayesian decision theory (Savage 1954). Savage’s decision theory for small worlds – those where an agent has access to a decision matrix defining states of the world, consequences, and actions – shows that the agent will maximize subjective expected utility, providing their preferences satisfy Savage’s axioms. Savage saw the problem of casting large world problems, those involving uncertainty and ignorance, in these terms as “utterly ridiculous” (p. 16). Consequently, we will use the term “Savage’s problem” to refer to obstacles and potential dangers in using analytic methods geared for small worlds to theorize, and make statements about, large worlds. Specifically, we consider Savage’s problem in the context of inductive inference, where a decision maker is required to generalize from observations and infer statistical properties of the environment. For example, an organism foraging for food may infer regularities in the distribution of food items from observations of previous food items. What distinguishes small from large worlds in inductive inference? Moreover, how significant is Savage’s problem to the study of inductive inference, where rational actors and optimal responses play a particularly influential role?
|Name||Evolution and Rationality: Decisions, Co-Operation and Strategic Behaviour|