Alison Gopnik (UC Berkeley)

Babies and Bayes Nets: A computational and cognitive account of theory formation

Alison Gopnik, Dept. of Psychology, University of California at Berkeley, Visiting Moore Fellow

How do children manage to learn as much about the world as quickly as they do? For the past thirty years developmental psychologists have explored "the theory theory" - the idea that children's cognitive development is like the development of theories in science. Recently, we have been using the formalism of directed causal graphical models or "Bayes Nets" to make the theory theory more precise and testable. The Bayes net formalism represents causal relations as directed graphs that are systematically related to patterns of probability and to outcomes of interventions. They allow both causal reasoning and learning - in particular, given a particular causal graph only certain patterns of conditional probabilities and outcomes of interventions will follow. I will report a series of results with 2-4 year old children showing that children may implicitly use similar representations and learning mechanisms. These representations and learning mechanisms allow them to construct intuitive theories of the physical and biological world, and especially, allow them to construct a theory of mind.