A Method to the Machine: An Architecture for Argument-Driven, Dynamic Character Performance

The evolution of architectures for virtual agents reveals a history of principled trade-offs. While dominant paradigms have successfully optimized for scalability and robustness, other valuable agent characteristics—such as the legibility of intent and the capacity for expressive performance—were often de-prioritized. This paper investigates an alternative set of architectural choices aimed at reclaiming these qualities. We introduce a novel agent architecture inspired by Stanislavskian acting, which operationalizes performance techniques by having agents reason from a core "supertask" and "ask questions" about their circumstances. The computational framework is composed of three layers: a strategic layer for character-centric decision-making, a defeasible logic programming (DELP) reasoning layer for argumentation, and a tactical layer using A Behavior Language (ABL). This paper provides a detailed architectural analysis of the system, demonstrated through a case study in a Unity-based game scenario. We conclude by discussing the system's limitations and engineering challenges, outlining a path for future work.