Abstract: 

The classical approach to AI was to design systems that were rational at run-time: they had explicit representations of beliefs, goals, and plans and ran inference algorithms, online, to select actions. The rational approach was criticized (by the behaviorists) and modified (by the probabilists) but persisted in some form. More recently, relatively unstructured data-driven end-to-end approaches have demonstrated great success in a wide variety of domains, and began to seem like a plausible route to general-purpose AI agents.

However, most recently, we have begun to see the limits of pure behavior learning and many practitioners are re-integration forms of search and explicit reasoning into their approaches.

I will revisit the rational-agent approach to the design of intelligent robots, from the perspectives of engineering effort, computational efficiency, cognitive modeling and understandability. I will present some current research focused on understanding the roles of learning in runtime-rational agents with the ultimate aim of constructing general-purpose human-level intelligent robots.