Learning Expertise with Bounded Rationality and Self-Awareness
Susan L. Epstein, Smiljana Petrovic · The MIT Press eBooks · 2011
This chapter describes how an architecture enhanced by metareasoning assesses expertise, manages large bodies of heuristics, and learns to think more effectively. FORR (FOr the Right Reasons) is a learning and problem-solving architecture that models the development of expertise with bounded rationality and self-awareness. Metareasoning is essential in FORR’s ability to learn to solve problems within a given class. As it learns to manage heuristics, FORR uses metareasoning to decide when to abandon an unpromising learning attempt, when to stop learning, how to select heuristics during learning, and how to prioritize heuristics. FORR also reasons about its performance on previous problems, its previous decisions, and the relative discriminatory power of its heuristics. Thinking about thinking makes FORR-based applications more incisive and more successful.