PRIMs theory describes a computational foundation for un-derstanding task-general human learning and transfer usingrule-based cognitive architectures. Integration with ACT-Rhas yielded Actransfer, a model that replicates human learn-ing and transfer across many tasks. However, this model re-quires task-specific latency scaling parameters from ACT-Rto model different tasks, implying that there is missing com-putation in the theory. Neuroscience literature has separatelydefined the “task set” as the neural encoding that configuresstimulus-response rule behavior in working memory. The pro-cess of switching between different task sets is often used toexplain human latency costs. This paper introduces an alter-nate instantiation of PRIMs theory that enacts task set process-ing to account for the missing computation via a novel memorystructure called a procedure context. Human tasks of varyingcomplexity are modeled across two experiments. Procedurecontexts model human latencies and interference effects in alltasks by integrating latency, decision making, task representa-tion, and learning as aspects of a single unified process. Thisapproach offers promise for future modeling within cognitivescience by uniting theories from neuroscience and cognitivearchitectures.
