Affect control theory (ACT), a social psychological theory of the dynamics of social interaction, may be a viable candidate for modeling human affective responses to systems. ACT proposes that actors assign meaning to the entities with which they interact in three dimensions: Evaluation, potency, and activity. These meanings generate expectations for interactions with the entities (i.e., how they and the entity should respond to one another). Violations of expectations generate affective responses leading actors to adjust their behavior toward the entities and/or redefine the meaning of the entities, which, in turn, leads to new expectations and repertoires of action and reactions. An initial study demonstrating how ACT might be extended from its current domain of human-human interaction to human-computer interaction is presented. ACT’s mathematical models provide the machinery for using such an extension as the foundation for a socially intelligent architecture.