The Affective Reasoner (AR) is a broad platform for research on various aspects of computing emotions. The work is constrained to a descriptive model (based originally on the work of Ortony, et al. [Ortony, Clore, & Collins1988]) wherein a broad comprehensive model of human emotion is used as a basis for describing, and manipulating, the social-emotional fabric of interaction between (1) agents and their perceived world, (2) between agents and other agents, and (3) between agents and humans. A key element of the ``emotionally intelligent'' processing that agents perform, is that they each have idiosyncratic, dispositional, ways they construe the world around them, and manifest responses to internal states that arise. It is from this processing that their relatively rich personalities arise. A second constraint is that agents do not experience emotions themselves (no body processes are represented), and emotions are not used functionally in any sophisticated ways; agents thus may react to situations that arise in a manner consistent with the motivations their descriptions are intended to capture, but do not often act because of those motivations. In short, agents appraise, in real time, the world that unfolds around them (including their own actions), and express their emotional reactions to this world, but their emotional reactions only minimally participate in the causal structure of the unfolding events.
Despite the above constraints, AR agents have broad capabilities, some of which address the three areas of research mentioned by Danny Hillis in his talk. As a vehicle for presenting background on this work, we will discuss three ways the AR platform has been used: as a general test system for a real-time computable model of emotion, as supporting theoretically rich, emotionally expressive, virtual actors, and as effecting a computable model of story-telling that uses a sophisticated representation of emotion interaction, and personality, to build a robust, dynamic, model of stories.