Toward Unifying Cognitive Architecture and Neural Task Set Theories
- Bryan Stearns, Department of Computer Science and Engineering, University of Michigan, Ann Arbor, Michigan, United States
- John Laird, Department of Computer Science and Engineering, University of Michigan, Ann Arbor, Michigan, United States
AbstractPRIMs theory describes a computational foundation for understanding task-general human learning and transfer using rule-based cognitive architectures. Integration with ACT-R has yielded Actransfer, which replicates human learning and transfer across many tasks. However, this model requires task-specific latency scaling parameters, implying that there is missing computation in the theory. Neuroscience literature has separately defined the "task set" as the neural encoding that configures stimulus-response rule behavior in working memory. Switching between different task sets is often used to explain human latency costs. This paper introduces an alternate instantiation of PRIMs theory that enacts task set processing to account for the missing computation via a novel memory structure called a procedure context. Human tasks of varying complexity are modeled across two experiments. Procedure contexts model human latencies and interference effects in all tasks by integrating latency, decision making, task representation, and learning as aspects of a single unified process.