Researchers argue AI agents need social theory as structural priors
A new framework formalizes how multi-agent AI systems must be modeled with social science laws.
A new position paper from Lynnette Hui Xian Ng, Iain J. Cruickshank, Adrian Xuan Wei Lim, and Kathleen M. Carley argues that as agentic AI systems are deployed into social environments—such as social media platforms, multi-agent LLM pipelines, or autonomous robotics fleets—their emergent behaviors can no longer be understood from individual agents alone. The paper, titled 'Social Theory Should Be a Structural Prior for Agentic AI: A Formal Framework for Multi-Agent Social Systems,' proposes that AI agents must be modeled with social theory as a structural prior to capture the dynamics of multi-agent interactions over time.
The authors formalize a Multi-Agent Social Systems (MASS) framework, representing it as a class of dynamical systems of information generation, local influence, and interaction structure. They anchor the framework in four structural priors: strategic heterogeneity (agents have diverse goals), networked-constrained dependence (interactions follow network structures), co-evolution (agents and environment change together), and distributional instability (system outcomes are sensitive to initial conditions). Through formal propositions, they demonstrate the importance of each prior and articulate a research agenda for modeling, evaluating, and governing such systems—pushing AI development closer to real-world social complexity.
- MASS framework formalizes four social theory priors: strategic heterogeneity, networked-constrained dependence, co-evolution, distributional instability
- Paper argues single-agent models fail for multi-agent environments like social media, LLM pipelines, and robot fleets
- Proposes a dynamical system view where agent interactions generate emergent, system-level outcomes over time
Why It Matters
Brings decades of social science into AI to make multi-agent systems more predictable and governable.