Architectural Roles of Affect and How to Evaluate Them in Artificial Agents

Is it possible to design affective artificial agents and if so, why should we care? This paper addresses both questions by laying out a program for systematically defining and evaluating possible functional roles of affective states in architectures for virtual and robotic artificial agents. It provides functional and architectural characterizations for simple and complex affective states, discusses possible interactions between affective and nonaffective processes, and proposes an experimental evaluation framework that allows for the rigorous quantification of the utility of architectural components (for affective and nonaffective agents alike). In doing so, it also provides a brief overview of past findings about the utility of affect mechanisms for artificial agents that were obtained following the proposed methodology.