DescriptionMental states (e.g., goals, beliefs, and intentions) may be attributed to agents on the basis of motion cues, and previous studies have successfully related low-level perceptual qualities of a stimulus agent’s trajectory (i.e. speed, acceleration, or other manner of motion) to resulting subjective percepts. I argue for a powerful and novel experimental paradigm, in which I utilize a two-dimensional virtual environment populated by autonomous agents whose simulated vision, memory, and decision making capabilities can be manipulated. These agents—nicknamed “IMPs” (Independent Mobile Personalities)—navigate the environment, collecting “food” and interacting with one another. Their behavior is modulated by a small number of distinct goal states: attacking, exploring, fleeing, and gathering food. In a first study, subjects attempt to infer and report the IMPs’ continually changing goal states on the basis of their motions and interactions. Although these programmed ground truth goal states are not directly observable, subjects estimate them accurately and systematically. I present a Bayesian model of the inference of goal states which accurately predicts subjects’ responses, including their pattern of errors. In a second study, I use simulated evolution to create a pool of evolved IMPs which exhibit adaptive behavior. I operationally define IMPs sampled from this simulated evolution as being more rational compared to non-evolved “control” IMPs, and find that subjects construe evolved IMPs as being both more intelligent and more human-like than non-evolved IMPs. In a final critical experiment, I demonstrate that subjects are better at discriminating the goal states of evolved IMPs than those of non-evolved IMPs. The two studies I present in this thesis provide empirical support for an account of adult “theory of mind” which asserts that 1) the inference of latent mental states can be understood as the inversion of a model of the generative processes producing the observable behavior of the agent, 2) this generative model reflects expectations of agent rationality, and 3) evolutionary fitness is a reasonable operational model of apparent agent rationality, to which subjects are sensitive. These experiments also demonstrate that using autonomous agents as stimuli opens up many basic research questions in the study of the interpretation of intentionality.