The realism of behavioral theory-based vs. non-theory-based AI agents during a simulated infant formula shortage

Introduction: AI-driven digital twins and autonomous AI agents are increasingly used to simulate human behavior during crises. Incorporating behavioral science frameworks may improve agent realism, but this practice is still in its infancy. This research evaluates the realism of behavioral theory-based agents in a controlled experimental design. Methods: Using a simulated infant formula shortage in South Dallas County, we compare two conditions: one with theory-based agents, and another without. Participants (human raters) assessed the perceived realism of agent decisions across both conditions. Results: Results showed significantly higher realism ratings for the theory-based agents, supporting our hypothesis. Discussion: This study constitutes an early effort to assess behavioral theory in simulation frameworks and establish a repeatable method for assessing behavioral fidelity. It provides policymakers and researchers with a theory-informed approach for enhancing AI agent realism, with the goal of increasing trust in digital twin models used for decision support in high-stakes environments.