Modeling Affect-Based Relationships and Social Allostasis as Mechanisms for Adaptation in Social Groups

Abstract

Artificial social agents are becoming increasingly present in the real world. For these agents, their success is tied to their ability to remain truly autonomous—to appropriately adapt their decisions and behaviours across changing physical and social contexts—and still remains a key area of research in the field of embodied autonomous agents. Numerous approaches towards models of adaptation have been proposed, including biologically-inspired models based on the regulation of an agent’s internal environment. A more recent concept from biological systems, called “(social) allostasis”, proposes a mechanism for the anticipatory adaptation of the internal environment through (affect-based) interactions with the (social) environment. Despite the long-term adaptive benefits proposed by social allostasis, current approaches towards models of adaptation for artificial social agents are yet to consider its principles in their approaches. In this manuscript, we address this shortcoming and investigate how mechanisms of social allostasis through affect-based relationships can be used for long-term adaptation of an embodied agent model. Using a biologically-inspired artificial life approach, we systematically investigate numerous hypothesised, hormonal mechanisms that underpin social allostasis and the survival-related benefits associated with affect-based social relationships. We conduct these investigations using a small society of simulated agents across numerous dynamic physical and social contexts. Our results find that adaptation of an embodied agent model through social allostasis mechanisms can provide survival-related advantages across several dynamic conditions in some contexts through physiological and behavioural adaptation. Throughout the manuscript, we also find numerous social interactions and dynamics that mirror biological systems. Our work addresses limitations in current approaches to adaptive models for embodied artificial agents, and presents a novel framework to guide future work in this field. The work also contributes a biologically-inspired artificial life model as a scientific tool to address some methodological challenges in natural systems. . .