This project aims to develop a computational model of human Theory of Mind (ToM): our ability to reason about the intentions and beliefs of other agents through observing their actions. ToM is a key component of social intelligence, allowing us to work with and learn from others. We want to model ToM as inference over a partiallyobservable Markov decision process (POMDP), in which the observer tries to guess the POMDP parameters that best explain the agent’s actions. In particular, we want to apply POMDPs to understanding the development of ToM in infants, under the hypothesis that infants use some weaker version of POMDP inference en route to developing full ToM capabilities.

I’ve worked on diverse projects in my MIT career, including stops at edX, where I crowdsourced teacher-student interactions, and Google, where I visualized data for Google Maps developers; the whole time holding a long-term interest in AI that began when I read Gödel, Escher, Bach in high school. I got involved in this UROP through Professor Tenenbaum’s course in computational cognitive science (6.804).