Building Theories of the World: Human and Machine Learning Perspectives

Human knowledge of the world is often expressed in the form of intuitive theories: systems of abstract concepts that organize, predict and explain our observations of the world. How are these powerful knowledge structures represented and acquired? I will describe computational frameworks for modeling people's intuitive theories and theory-building processes, and some ways of testing these models experimentally with human learners. Our models of human learning and inference build on core approaches in Bayesian artificial intelligence, statistical relational learning and inductive logic programming, but also suggest new ways to extend these machine learning and reasoning approaches to more human-like capacities.