Social character design for animated agents

Abstract

In order for a computer agent to be accepted and function properly within a community of people, an appropriate social interaction scheme has to be established. We conducted an experiment to investigate the effect of sharing opinions with interface agents on subsequent human behaviors. In the first stage of the experiment, either a positive sharing (Agreeable agent), a negative sharing (Disagreeable agent) or no sharing (Neutral agent) of opinions was reinforced between subjects and agents through simple human-agent interactions. The effect of the sharing of opinions was then tested, in the second stage of the experiment, in terms of the subjects' tendency to favor or disfavour agents' proposals. We found that the Agreeable agent was consistently in favor of the subject's decisions. We therefore examined how the subject indicated sympathetic responses to the Agreeable agent after he/she interacted with all of the agents. The results show that people tend to behave favorably toward agents that previously agreed with their decisions. This suggests that human-computer interaction has the same social dynamics as human-human interaction.