Believable Emotional Agents
Scott Neal Reilly, Ph.D.
Zoesis Studios, Inc.
March 18, 2005
11 a.m. - 12 noon
Fuller Labs 320
Abastract
Characters matter. We use characters to teach, to sell, to entertain, to enlighten. There is little doubt that interactive characters will continue to play the same important and omnipresent roles that their non-interactive counterparts have for millennia; it is just a matter of time. Unfortunately, artificial intelligence, for all its advances in creating interactive agents, has taught us very little about how to make interactive characters--how to make characters that seem emotional, quirky, endearing, alive. If, however, we take these traits (and others that artists tell us are important) as important goals, we can make progress. We can even use some existing AI and cognitive modeling to help us get there if we use it carefully.
In this talk, I will focus on one part of this larger goal, the problem of creating interactive characters that seem to have emotions. I will present an implemented and tested architecture for the generation, manipulation, and expression of emotions that is flexible, powerful, and authorable enough that it has been used for the creation of dozens of characters over the past decade. I will compare this approach to cognitive-appraisal approaches which are quite common in computational emotion modeling, I will describe some important lessons learned along the way, and I will describe various forms of evaluations used over the years to gain confidence that this is a powerful approach.
Host
Michael A. Gennert
Refreshments will be served.
Last modified: September 25, 2006 14:00:11