Research
Cognitive grounding improves LLM simulation of collective human behavior
Nature · By Hongan Zhu ·
The Brief
- A study in Nature Computational Science reports that guiding AI agents with descriptions of how people make decisions improves how well they reproduce collective human behavior.
- The study uses a supply-chain game as its test setting.
- The paper was published online on 16 November 2026 with DOI 10.1038/s43588-026-01080-z.
Why it matters
Researchers who use LLM agents to simulate human groups may get more realistic collective behavior if agents are grounded in cognitive descriptions of decision-making rather than left unguided.
Who’s affected
- computational social scientists
- researchers using LLM agents for simulation
- supply-chain and behavioural economics researchers
- #llm
- #agent-based-modeling
- #collective-behavior
- #cognitive-science
- #simulation
Nature Computational Science, Published online: 16 November 2026; doi:10.1038/s43588-026-01080-z The study shows that guiding AI agents with descriptions of how people make decisions in a supply-chain game helps agents reproduce collective human behavior.