Exploring the mental space of autonomous intentional agents

The ability to infer the intentions of other agents on the basis of their motion is a critical psychological faculty. In the present study, we examine a key question underlying this process, namely: What are the psychologically natural categories of intentional agents and actions? To investigate this question empirically, we use displays containing a number of autonomous, independently programmed agents moving about a two-dimensional environment and interacting with one another. Each agent behaves according to its own simple program, controlled by a small number of parameters that define its "personality." We probe participants' impressions of the similarities among the behaviors of the various agents, and then use multidimensional scaling in an attempt to recover the subjective mental space of agent types. An important variable underlying this space turns out to be a parameter that determines how the agent reacts to a nearby agent at one critical distance. A follow-up experiment suggests that variation along this parameter ultimately contributes to modulating a more fundamental perceptual dimension that reflects how "hostile" or "friendly" the agents appear to be.