Purchase this article with an account.
Peter Pantelis, Steven Cholewiak, Paul Ringstad, Kevin Sanik, Ari Weinstein, Chia-Chien Wu, Jacob Feldman; Perception of intentions and mental states in autonomous virtual agents. Journal of Vision 2011;11(11):733. doi: https://doi.org/10.1167/11.11.733.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Comprehension of goal-directed, intentional motion is an important but understudied visual function. To study it, we created a two-dimensional virtual environment populated by independently-programmed autonomous virtual agents. These agents (depicted as oriented triangles) navigate the environment, playing a game with a simple goal (collecting “food” and bringing it back to a cache location). The agents' behavior is controlled by a small number of distinct states or subgoals–including exploring, gathering food, attacking, and fleeing–which can be thought of as “mental” states. Our subjects watched short vignettes of a small number of agents interacting. We studied their ability to detect and classify agents' mental states on the basis of their motions and interactions. In one version of our experiment, the four mental states were explicitly explained, and subjects were asked to continually classify one target agent with respect to these states, via keypresses. At each point in time, we were able to compare subjects' responses to the “ground truth” (the actual state of the target agent at that time). Although the internal state of the target agent is inherently hidden and can only be inferred, subjects reliably classified it: “ground truth” accuracy was 52%–more than twice chance performance. Interestingly, the percentage of time when subjects' responses were in agreement with one another (63%) was higher than accuracy with respect to “ground truth.” In a second experiment, we allowed subjects to invent their own behavioral categories based on sample vignettes, without being told the nature (or even the number) of distinct states in the agents' actual programming. Even under this condition, the number of perceived state transitions was strongly correlated with the number of actual transitions made by a target agent. Our methods facilitate a rigorous and more comprehensive study of the “psychophysics of intention.” For details and demos, see ruccs.rutgers.edu/∼jacob/demos/imps.html
This PDF is available to Subscribers Only