Physical Problem-Solving in Minds and Machines

Explore the fascinating intersection of AI, psychology, and neuroscience in this 46-minute talk by Kelsey Allen from DeepMind. Delve into the structured nature of the world and how cognitive and machine models can achieve remarkable efficiency and generalization by respecting these structures. Examine the factorization of objects, relations, and physics to support flexible physical problem-solving in both minds and machines. Discover how these elements can explain complex cognitive phenomena, such as effortless learning of new tool use in humans, and complex behaviors in machines like highly realistic simulation and tool innovation. Learn how leveraging problem structure, combined with general-purpose methods for statistical learning, can lead to the development of more robust and data-efficient machine agents while also shedding light on how natural intelligence learns so much from so little.