Abstract

This thesis explores how the mind uses the motion of animate objects to make inferences about these objects' underlying mental states, intentions, goals, or dispositions. We present dynamic scenes to subjects in which autonomously programmed triangular "agents"' interact with each other and--in two of the experiments--an additional agent that is controlled by the subject. We strive for the autonomous agents to be simple in their underlying programming but to also engage in a rich array of lifelike behaviors. Subjects watch short simulations populated with these agents and then are asked questions designed to probe their perceptions of the similarities among the agents' behaviors. We use the responses to derive a multidimensional scaling solution for the agents in our set. The aim is to relate this MDS solution to the underlying programming on one hand, and to also discover interesting structure in subjects' perception of the "agent space." Clusters in this space, for example, could provide insight into the perceptual biases subjects bring to the experiment for which types of agents are a priori more likely to be observed. The most robust result from the experiments is that subjects, in making inferences about an agent's intentions, pay special attention to how the agent reacts when another agent is at a "critical distance" of about 10-17.5 agent lengths away.