UC San Diego

Human interaction relies on the ability to form accurate internal models of other people. What is the structure of our mental representations of others? Existing theories in psychology broadly fall into two classes: those which view people as constructing rich generative models of those around us, and those which argue for more simplified predictive representations based on past behavior. In this dissertation, I explore the conditions under which people employ different representations of others and the constraints they face in each case. My work probes dyadic behavior across repeated interactions, thereby exposing the precise structure of the representations that people form in diverse settings. In Chapter 1, I begin by investigating how people develop predictive models of others based purely on simple, sequential patterns in their previous actions. I present evidence that in mixed strategy equilibrium (MSE) games, people acquire an adaptive model of their opponent over many interactions and argue that behavior in such games offers a novel perspective on people’s opponent modeling. In Chapter 2, I present two studies characterizing the basis of people’s opponent modeling in MSE games and exploring the scope of this ability. Results suggest that people show substantial limitations in their capacity to develop predictive models of others using patterns in their behavior alone. In light of these findings, Chapter 3 explores the process by which people develop more abstract and sophisticated representations of others in domains where they have rich mental models of their own. Specifically, this work focuses on how people incorporate the competence of another agent into collaborative interactions in a physical task. I first show that people infer latent and dynamic properties of others’ behavior in this setting; in a second study, I show that such inferences extend to features of their collaborator’s internal model of the task. Broadly, this work suggests that our representations of others can take on surprisingly diverse forms but their complexity is heavily context-dependent. I conclude with a discussion of future directions aimed at understanding the structure of people’s representations of others and how they calibrate these representations to the context at hand.