Prosociality by Coupling, Not Mere Observation: Homeostatic Sharing in an Inspectable Recurrent Artificial Life Agent

The paper presents a novel approach to prosocial behavior in artificial agents through homeostatic coupling, diverging from traditional methods that rely on social rewards or direct partner-state access. The authors introduce a scalar homeostat and a social coupling channel in the ReCoN-Ipsundrum architecture, demonstrating that coupled agents can effectively share resources in environments like the FoodShareToy and SocialCorridorWorld, while uncoupled agents do not exhibit helping behaviors. This research highlights the significance of self-regulation in fostering cooperative behavior, providing insights for practitioners designing AI systems that require adaptive social interactions.