Tomer Ullman (MIT Tenenbaum Lab)

Title:  “The Structure and Construction of Intuitive Theories: Reasoning about Objects, Agents, and Beyond”

Abstract:  From an early age, human beings use intuitive theories to make sense of the world around them. Akin to scientific theories, intuitive theories posit hidden variables that can predict and explain observed data. While this view has been both useful and influential in cognitive science and development, fundamental questions regarding the structure and acquisition of intuitive theories remain open.  In this talk I will argue that probabilistic generative programs are best suited for capturing the structure of these representations, and show evidence for this argument from my own work. I will also discuss my work on algorithmic-level approaches to hierarchical Bayesian reasoning over intuitive theories, which connect to the real-world psychology of learning new programs. I will consider in particular two key intuitive theories: intuitive physics, and intuitive psychology.