The Metacognitive Loop and Reasoning about Anomalies
Matthew D. Schmill, Michael L. Anderson, Scott Fults, Darsana P. Josyula, Tim Oates, Don Perlis, Hamid Haidarian Shahri, Shomir Wilson, Dean Wright · The MIT Press eBooks · 2011
This chapter describes an architecture for generalized metacognition—called the metacognitive loop (MCL)—aimed at making artificial intelligence (AI) systems more robust. The key to this enhancement is to characterize a system by its expectations each time it engages in activity, to watch for violations of system expectations, and to attempt to reason in an application-general way about the violation to arrive at a diagnosis and plan for recovery.