| Abstract |
A series of user studies resulted into an empirically validated
framework for the study of user-agent interaction. We formalized this
framework and made it the basic mechanism of how agents build up
affect for their human users, and extended this model with affective
decision making and the capability to simulate specific emotions. We
assured internal consistency through simulation experiments and
performed a successful pilot study. We compared the performance of an
agent equipped with our cognitive model to the performance of a human
that controlled the agent in a Wizard of Oz condition during a speed
dating experiment. Participants did not detect any differences between
the two conditions in the emotions the agent experienced and in the
way he perceived the participants. However, the cognitive structure
behind the agent’s perceptions was implicitly recognized by the
participants and differed significantly between a human and software
controlled dating partner. The structure recognized while a human
controlled the agent was more similar to the structure recognized in
humans than the structure recognized in our, apparently, still
imperfect model. This indicates that our model can well be used for
designing believable virtual humans or humanoid robots, but does not
completely reproduce human cognitive structure yet.
|