Emergent Relational Order in LLM Agent Societies: From Collective Affect to Authority Stratification

This paper investigates whether the Differential Order Pattern, a sociological theory describing rural society as a series of ego-centered, relationally graded circles of obligation, is a culturally specific phenomenon or a universal emergent outcome of general social mechanisms. To test this, the authors developed CAREB-MAS, a multi-agent simulation framework that avoids hard-coding cultural rules. Instead, agents are equipped with a cognitive architecture grounded in Affect Control Theory, Social Identity Theory, and Durkheimian collective affect. The agents interact in a simulated environment where they must manage production and resource allocation through deliberation and kinship-based social ties.

Across long-horizon simulations, the agents spontaneously reproduced five core phenomena associated with the Differential Order Pattern:

• Stable Labor Specialization: Agents developed consistent roles without centralized coordination.
• Guanxi-based Economic Ethics: Ethical obligations within close relationships sustained cooperation even when short-term efficiency was low.
• Relational Decay of Cooperation: Cooperative intensity and efficiency systematically attenuated as social distance increased.
• Emergent Relational Authority: Stable hierarchies formed through long-term interaction and social recognition rather than formal power.
• Clan-based Stratification: Economic returns and authority stratified based on relational proximity, creating a clear center-periphery structure.

The study further found that these patterns are structure-sensitive: symmetric production capacities lead to kin-centered integration (mechanical solidarity), while complementary capacities drive functional interdependence (organic solidarity).

By successfully generating complex social structures from basic cognitive and affective mechanisms, this research provides a powerful computational testbed for social science. It suggests that many observed social structures, often attributed to unique cultural traditions, may actually be predictable outcomes of how humans (or agents) process emotion, ethics, and identity in the face of varying economic constraints. This approach bridges the gap between qualitative sociological theory and quantitative agent-based modeling.