Trust of Virtual Agent in Multi Actor Interactions

Abstract

Trust is crucial when integrating virtual agents in human teams. Our study investigated the combined use of psychological and physiological measures in predicting human trust of agents undertaking social tasks. The psychological measures comprised trust scores on ability, benevolence and integrity. The physiological measures included facial expressions, voice, heart rate and gestural postures. Subjects interacted with two avatars. A neurofuzzy algorithm extracted rules from the psychophysiological data to predict trust. Results revealed that trust can be predicted with 88% accuracy.