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. More recently, relatively unstructured, data-driven end-to-end approaches have achieved great success across a wide range of domains and have begun to seem like a plausible path to general-purpose intelligent robots. However, we are now seeing the limits of pure behavior learning, and many practitioners are reintegrating forms of search and explicit reasoning into their approaches. Leslie Kaelbling will revisit the rational-agent approach to designing intelligent robots from the perspectives of engineering effort, computational efficiency, cognitive modeling, and interpretability. She will present current research aimed at understanding the role of learning in runtime-rational agents, with the ultimate goal of constructing general-purpose, human-level intelligent robots.