Strategic Control of Experience-Weighted Attraction Model in Human-AI Interactions

This paper investigates a novel control framework for strategic Artificial Intelligence (AI) in human interactions. We leverage the Experience-Weighted Attraction (EWA) model, a widely used method for capturing human learning dynamics. EWA incorporates experiences to influence future choices through “attraction values” assigned to different actions. By treating these attraction values as the system state, we formulate the interaction between the AI and human as a stochastic control problem. This approach allows the AI to strategically influence the human’s behavior by manipulating the environment or offering incentives that alter the attraction landscape, even under conditions of partial knowledge about the human agent’s learning process. Our framework contributes to the field of human-AI interaction by providing a novel control method driven by the dynamics of human learning through EWA.