Leslie Kaelbling: Reinforcement Learning, Planning, and Robotics | Lex Fridman Podcast #15

Robot Challenges

Creating intelligent robots involves navigating complex technical and conceptual challenges. Leslie Kaelbling discusses the intricacies of building robots with human-level intelligence, emphasizing the need for systems that can observe and evaluate their own processes. She notes, "You could have robots that are zombies, right, that behave like humans but are not conscious, and I, at this moment, would be happy enough with that" 1. This highlights the current focus on functionality over consciousness in robotics. Additionally, Kaelbling argues against viewing human life solely as a planning problem, suggesting that a multifaceted approach is necessary for both human and robotic reasoning 2.

   

Machine Consciousness

The debate over machine consciousness raises both philosophical and practical questions. Leslie Kaelbling explores the concept of self-awareness in machines, questioning whether it is essential for achieving human-level intelligence. She suggests that self-awareness might simply involve a system's ability to monitor its own functions, rather than possessing consciousness as humans understand it 1. Kaelbling also emphasizes the importance of determining what aspects of intelligence should be built into machines versus what should be learned, acknowledging the gaps in current understanding 3.