Reverse-engineering the core of common sense in the human mind and brain

Pour rencontrer Josh Tenenbaum, merci de contacter Coralie Chevallier : coralie.chevallier@ens.fr.

Abstract : Recent successes in artificial intelligence have been largely driven by methods for sophisticated pattern recognition, including deep neural networks and other machine learning methods.  But human intelligence is more than just pattern recognition.  And no machine system yet built has anything like the flexible, general-purpose commonsense grasp of the world that we can see in even a one-year-old infant.  At the heart of human common sense is our ability to model the physical and social environment around us: to explain and understand what we see, to imagine things we could see but haven’t yet, to solve problems and plan actions to make these things real, and to build new models as we learn more about the world.  I will talk about prospects for reverse-engineering these capacities in human adults and infants, primarily at the cognitive level but also recently extending to the neural level.  I will describe at a high level some of the core technical concepts, especially probabilistic programs and program induction, which together with tools from deep learning and modern video game engines drive both our models of human commonsense and enable us to make AI systems smarter in more human-like ways.