Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics

Leslie Kaelbling is an MIT roboticist known for reinforcement learning, planning, and robot navigation, a winner of the IJCAI Computers and Thought Award and former editor-in-chief of JMLR. This opening frames the conversation.

Her excitement for AI began in high school reading Gödel, Escher, Bach, which exposed her to how complex behavior can be built from simple primitives and combination. Robots entered later: her first job at SRI's AI lab put her on a Shakey-successor robot, and getting that machine to do things is what drew her to robotics.

She did her Stanford bachelor's in philosophy because there was no CS undergraduate degree there at the time, and Stanford's symbolic-systems track — logic, model theory, and formal semantics of natural language — turned out to be ideal preparation for AI.

symbolic systems, which is logic, model, theory, formal semantics of natural language. And so that's actually a perfect preparation for work in AI and computer science.

Back then too few people pursued AI for it to be a named path; her graduating class of philosophy majors split roughly between computer science and law, with only one or two staying in philosophy.

On whether AI researchers should philosophize, she keeps to the formal corner closest to her work — belief, knowledge, denotation — and declares herself a materialist who sees no reason a robot couldn't be behaviorally indistinguishable from a human, while the zombie question of internal distinguishability is one she neither knows nor much cares about.