New LLM framework models social capital with AI agents
Researchers build LLM agents that simulate Putnam's Social Capital Theory in action
Researchers from the University of Science and Technology of China and collaborators have developed SocaSim, a groundbreaking LLM-based multi-agent simulation framework designed to model and apply Putnam's Social Capital Theory in controlled, replicable environments. Published on arXiv, the work addresses a longstanding gap in AI-driven social simulations by integrating social network evolution, trust dynamics, and norm propagation into a single, theory-aligned environment. Unlike traditional behavior-driven simulations, SocaSim enables researchers to trace micro-level causal pathways through round-by-round simulations and counterfactual interventions, offering process-level interpretability.
The framework was validated by reproducing Putnam’s macro-level social capital patterns and demonstrating strong alignment with human group behaviors. As a practical application, the team used SocaSim to analyze adaptation challenges in smart elderly care, showcasing its utility in real-world social systems. With 23 pages, 13 figures, and 11 tables, the paper establishes a new research paradigm that leverages LLM agents to bridge social science theory with computational modeling.
- SocaSim is the first LLM-based multi-agent framework to model Putnam's Social Capital Theory with integrated social network evolution, trust dynamics, and norm propagation
- The framework reproduces macro-level social patterns and achieves strong human-agent alignment at the group level
- Researchers applied SocaSim to analyze adaptation challenges in smart elderly care, demonstrating real-world utility
Why It Matters
This work enables AI-driven social science research with unprecedented fidelity, allowing professionals to model and optimize collective action and community resilience in real-world applications.