The building of computer emotion models has long been viewed with a certain amount of skepticism: after all, even deciding upon what comprises emotion in humans is rife with controversy. Whispering a phrase like basic emotions in the wrong company is liable to cause verbal fireworks, if not actual fisticuffs. However, it is clear that computer agents are well capable of expressing a rich variety of emotions [Elliott1997b, Koda1997], and that these can be effective in conveying much about the intended personality of the agent. Few would argue that some working model of emotion systems will greatly benefit both the believability, and intuitiveness, of automated characters.
The first problem in sorting out the field is that work labeled as centering on affect, emotion, or personality tends to be grouped together out of hand regardless of the work's focus and goals, although these vary widely. In some cases the researchers are attempting to faithfully model emotion subsystems (if not architecture), both human and otherwise; in other cases they are attempting to support a believable social structure based on a descriptive model of eliciting scenarios, and response behaviors. Rosalind Picard at the MIT media lab has a good starting list of references on affective computing (starting with her new book, ``Affective Computing''), and many of these have relevance to the intelligent, interactive, agent paradigm (http://vismod.www.media.mit.edu/vismod/demos/affect/AC_bibliography.html). Here we consider only a few of the many labs which are primarily dedicated to building computer models of emotion and personality appropriate for use in computer agents. Additionally, two groups are certainly also working in this area but are presented elsewhere in the article. They are the Affective Computing group at the MIT media lab (http:/vismod.www.media. mit.edu/vismod/demos/affect/), and the OZ project (http://128.2.242.152/afs/ cs.cmu.edu/project/oz/web/) -- and its offshoot Zoesis.