Affective Agent Architecture: A Preliminary Research Report

Abstract

Our work with affective agents is motivated by a desire to develop a biologicallyinspired model of the human system that includes sensation, feeling, and affect integrated with the logical and cognitive models traditionally used in AI. Our work can be situated alongside Breazeal’s robot Kismet (1998), Elliott’s Affective Reasoner (1992), as well as Baillie’s architecture (2002) for intuitive reasoning in affective agents; we too seek to model the motivations and reasoning of agents with an affective foundation. Our approach differs in how we model affect and the transition of biological impulses from sensation through affect into action. We base our work on the Affect Theory of Silvan Tomkins; Thomkins’ (2008) theory comprises both a biologically hardwired notion of affect, similar to Ekman’s basic emotions and Izard’s DET, as well as a more culturally and biographically situated theory of scripts for describing how affect moves from feeling into action (Nathanson, 1996). As a proof of concept, we have consciously chosen to work with a simplified implementation of Tomkins’ idea of affect, and explore how both internal (drives, organ sensations) and external stimuli move through the agents via the affect system and transition into action.