14th Annual Lecture
March 23, 2012
Wu & Chen Auditorium
101 Levine Hall
Modeling common-sense reasoning with probabilistic programs
Artificial intelligence (AI) has made great strides over its 60 year history, building computer systems with abilities to perceive, reason, learn and communicate that come increasingly close to human capacities. Yet there is still a huge gap. Even the best current AI systems make mistakes in reasoning that no normal human child would ever make, because they seem to lack a basic common-sense understanding of the world: an understanding of how physical objects move and interact with each other, how and why people act as they do, and how people interact with objects, their environment and other people to achieve their goals. I will talk about recent efforts to capture these core aspects of human common sense in computational models that can be compared with the judgments of both adults and young children in precise quantitative experiments, and used for building more human-like AI systems.
These models of intuitive physics and intuitive psychology take the form of “probabilistic programs”: probabilistic generative models defined not over graphs, as in many current AI and machine learning systems, but over programs whose execution traces describe the causal processes giving rise to the behavior of physical objects and intentional agents. Perceiving, reasoning, predicting, and learning in these common-sense physical and psychological domains can then all be characterized as approximate forms of Bayesian inference over probabilistic programs.