FriSem

Erik Brockbank, Postdoc with Assistant Professor of Psychology, Tobias Gerstenberg, Stanford University

Title: Predicting and explaining behavior across time and contexts

Abstract: The ability to predict others’ actions is central to human collaboration and competition. In particular, people have a remarkable capacity to predict others across varying time scales and contexts. What sort of representations of others allow us to make sense of their behavior from moment to moment or over months and years, and how do we acquire them? Prior work has argued that the primary mechanism by which we understand others is through a theory of other minds in which behavior is attributable to an actor’s goals, beliefs, and desires. In my work, I explore how this model of others can be extended to accommodate predicting and explaining behavior over longer time horizons and in diverse contexts. In this talk, I’ll briefly review two lines of work exploring how we learn predictive behavior patterns over repeated adversarial interactions, and how we infer durable features of an agent’s physical world model in collaborative settings. Then, I’ll discuss ongoing work aimed at understanding how we combine transient situational factors and more stable, persistent features of an actor to explain their behavior. Taken together, this works points the way toward a precise account of how people learn and deploy representations of others that persist over time and in diverse environments.