Proceedings of the Annual Meeting of the Cognitive Science Society

Intelligent behaviour requires monitoring and adjusting ongoing cognitive processes to the changing demands and constraints of internal and external environments. This is studied in terms of meta-cognition, cognitive control and meta-reasoning, and, recently, as the expected value of control and the value of computation. However, integrated consideration of core problems such as state representation, generalization, and prospective versus retrospective calculations of the long-run value of meta-computational policies is lacking. We offer a recursive Markov decision process framework called recursive-MDP in which meta-cognition is a form of meta-level perception, and, just as for control itself, meta-control algorithms can be fractionated into costly but adaptive model-based, or fast model-free strategies. Understanding how the brain implements self-adjustment has significance to both psychiatry and artificial intelligence.