In order to create artistically effective, interactive characters (or believable agents, I have taken lessons from other arts as a starting point. Most traditional AI research has focused on either creating competent/intelligent agents or agents that are cognitively plausible models of humans. The goal of creating good interactive characters is different from either of these goals and requires a different approach (though I have used ideas from these other research areas).
My goal to aid in the creation of emotional agents draws heavily on other emotion research. Just to drop a few names: Ortony, Clore, and Collins; Sloman (Aaron heads the Cognition and Affect project at the University of Birmingham); Frijda; Oatley. My work is somewhat different, though, in that I am willing to pass up cognitive plausibility to achieve characters that simply appear emotional. I have designed an architecture, called Em, that allows artists to create a wide variety of emotional characters. Em is both more powerful and easier to use because it has been tightly integrated into a broad agent architecture.
To help artists create believable social agents, I have designed a two-part methodology that help artists create personality-rich social behaviors for believable agents. The first part of the methodology suggests a number of important elements of personality that should be incorporated into social behaviors and some heuristics for doing so. The second part of the methodology suggests that artists create behaviors that rely on minimal amounts of modeling of other agents in the environment. Using this methodology I have been able to create a number of social behaviors that are robust and personality-rich.
For more on my efforts to create social and emotional agents, check out my thesis, "Believable Social and Emotional Agents" in postscript and gzipped postscript.
I also spend some time hanging out with the members of the Prodigy project. Prodigy is a planning system that acts as a testbed for research in machine learning. My main research effort involving Prodigy was a joint effort with Jim Blythe to combine the behavior-based Hap architecture used in Oz agents with the deliberative Prodigy system. Our results are recorded for all to see in a very nice tech report.
For (vaguely research-related) fun, I have also dabbled a bit with using genetic algorithms to interactively create art and I maintain the interactive fiction page.