Achievement

The ability of humans to infer intentionality of living, moving entities was studied by an interdisciplinary team of trainees and affiliates, led by Pantelis. They built a virtual environment of moving autonomous “agents” (small shapes) that were programmed to perform tasks, such as gathering food, or attacking or fleeing each other. The model driving the agents allowed the agents to evolve their actions on the basis of prior success. Human subjects viewing these simulations were able to accurately infer the agents' goals and intentions. The performance of the humans was well accounted for by a Bayesian model in which intentions of agents were inferred from the observed action at time t, as well as the probability of the agent being in a given state given the previous states at time t-1. This work is important both for modeling the human perception of intentions, as well as the design of effective virtual environments. (Pantelis, Feldman et al., under review).