As project leader for the Adaptive Intelligent Systems group at Stanford University's Knowledge Systems Laboratory (KSL), Barbara Hayes-Roth has long been one of the main proponents of character-rich agents. Hayes-Roth was an area chair for the inaugural Autonomous Agents (97) conference in Marina del Rey, and has been tireless in her efforts to legitimize interactive agents research. Most closely related to this article, is the Hayes-Roth group's work on the Virtual Theater project, and on a commercial offshoot, Extempo Systems. The latest Virtual Theater communications can be found at http://ksl-web.stanford.edu/projects/CAIT/index.html. The latest Extempo developments can be viewed at http://www.extempo.com/.
One of the research lines for this group is in developing user-guided characters, and fully autonomous characters, for use in shared virtual environments. These environments such as KSL's CyberCafe, and Extempo's award-winning Spence's Bar, allow users to meet and engage in directed social interaction. The user-controlled avatars, and autonomous characters, build interactive stories together in an online ``chat room''. (see figure 2.)
At KSL, Daniel Rosseau, and Hayes-Roth, have been designing schemes for a high-level descriptive representation of personality appropriate for such interactive characters [Rousseau & Hayes-Roth1997b]. The goal of the research is to allow authors to build distinct and recognizable personality types suitable for use in the interactive virtual environments. The models are based partly on trait theory (wherein a multidimensional space of characteristics like sociability and extraversion is seen as determining how one will act in society), and social learning theories (wherein behavior is partly determined by context characteristics and the individual's past experiences in similar situations).
The personality profiling model allows for the specification of traits, such as self-confidence, activity, and friendliness, which can be varied along a numeric continuum. These, in part, determine how an agent reacts to situations in the virtual environment. The traits, in turn, can depend upon values of agent states, such as happiness-sadness (self-oriented affective states), gratitude-anger (other-oriented affective states - e.g., grateful to someone), and liking and hatred (attraction-oriented affective states). Characteristics like these are used to create the dispositional, and dynamically variable, personalities of agents used in the interactive environment.
Using such controls over agent behavior, one is able to define personalities that reflect the intended high-level characteristics of labeled lay-personality types (and c.f [Elliott1993]). For example, one might create agents with general types of nasty, friendly, shy, lazy, choleric, and selective (friendly with some, nasty with others).
Early exploratory studies using this approach have shown promise in that users are able to recognize the intended personality characterizations, and that they respond to them in socially coherent ways. [Rousseau & Hayes-Roth1997a]
The efforts at the Virtual Theater project, and Extempo Systems share many common elements with the groups that follow in the next section, in that they support the creation of virtual human-computer communities through the use of agents.