The Adaptive Control of Action and Thought: Boundedly Rational Behavior Under Cognitive Constraint

This talk will describe and illustrate a new approach to cognitive theory and modeling based on boundedly rational models of control. The approach assumes that individuals adapt to a utility function given constraints on their cognitive architecture. It narrows the space of predicted behaviors through analysis of the payoff achieved by alternative strategies, rather than through data fitting (it thus has strong ties to reinforcement learning). It thereby addresses a major problem in applying general programmable cognitive theories (e.g., Epic, Soar, ACT-R) to specific domains: How do we determine what strategy such architectures should be programmed with to accomplish a given task? The approach will be illustrated with modeling and empirical results from several domains, including an analysis of elementary dual-tasking that yields the first quantitative and predictive account of individually varying dual-task costs shaped by system noise.