From Average to Archetype: Hybridizing LLMs with Psychological and Sociological Models for Diverse Simulations

TL;DR: What if we fuse LLMs with agent-based models grounded in psychology or sociology to capture archetypal and extreme behaviors? The experiment would integrate psychological trait models or social network effects into LLM agent architectures, aiming to unlock richer heterogeneity than pure LLMs.

Research Question: Can hybrid models that combine LLMs with explicit psychological or sociological frameworks generate more archetypal and diverse behavioral patterns than LLMs alone?

Hypothesis: Embedding domain-specific behavioral models (e.g., Big Five personality, social influence networks) into LLM simulation agents will foster richer, more representative diversity—including long-tail behaviors—compared to purely data-driven LLM approaches.

Experiment Plan: - Implement hybrid agents that use LLMs for language and decision-making, but parameterize their behavior using psychological or sociological models (e.g., explicit personality traits, peer influence matrices).

• Simulate OmniBehavior tasks, comparing pure LLM, hybrid, and rule-based agent populations.
• Assess improvements in behavioral diversity, outlier frequency, and archetype representation against real-world baselines.

References:

• Chen, J., Xu, R., Cao, B., Pan, R., Zhang, Y., Hu, Y., Du, Y., Gao, T., Lu, Y., Sun, Y., Han, X., Sun, L., Wu, X., & Lin, H. (2026). Towards Real-world Human Behavior Simulation: Benchmarking Large Language Models on Long-horizon, Cross-scenario, Heterogeneous Behavior Traces.
• Huang, Y. J., & Hadfi, R. (2024). How Personality Traits Influence Negotiation Outcomes? A Simulation based on Large Language Models. Conference on Empirical Methods in Natural Language Processing.
• Kopp, S., & Bergmann, K. (2017). Using cognitive models to understand multimodal processes: the case for speech and gesture production. The Handbook of Multimodal-Multisensor Interfaces, Volume 1.