Dim(N) Week 5 - Joshua Tenenbaum (youtube lecture) (08/01/24) What Kind of Computation Is Cognition?
[Abstract] Recent successes in artificial intelligence have been largely driven by neural networks and other sophisticated machine learning tools for pattern recognition and function approximation. But human intelligence is much more than finding patterns or approximating functions. And no machine system yet built has anything like the flexible, general-purpose common-sense grasp of the world that every human being does. This talk will address prospects for capturing human common sense in computational terms. I will briefly illustrate recent work modeling intuitive physics and intuitive psychology—that is, computer-implemented formal models of people’s intuitive mental models of physical objects, intentional agents, and their causal interactions; how these systems guide our perception, prediction, and planning; and the learning mechanisms that extend our thinking into new domains. I will introduce in non-technical terms the computational ideas underlying these models, such as probabilistic programming, inference over video-game-style simulations, and probabilistic program synthesis. And I will close with some speculative thoughts for discussion about the metaphysics of computation: In what sense can we say that these models capture the real workings of the human mind, or brain?
Comments