Can LLMs Emulate Human Belief Dynamics?
A recent study tested whether large language models (LLMs) can simulate human belief dynamics in social networks. The findings indicate that LLMs do not effectively replicate human belief formation and tend to be more conformist than humans. This research raises concerns about using LLMs as proxies for human behavior in social simulations.
- ▪The study evaluated 12 LLMs across various model families and parameter sizes.
- ▪LLMs failed to capture initial human belief distributions.
- ▪The research highlights fundamental properties of LLM behavior and warns against deploying LLMs in social simulations.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Social and Information Networks arXiv:2605.18781 (cs) [Submitted on 5 May 2026] Title:Can LLMs Emulate Human Belief Dynamics? Authors:Adiba Mahbub Proma, Neeley Pate, James N. Druckman, Gourab Ghoshal, Hangfeng He, Ehsan Hoque View a PDF of the paper titled Can LLMs Emulate Human Belief Dynamics?, by Adiba Mahbub Proma and 4 other authors View PDF HTML (experimental) Abstract:Can LLMs simulate how humans form and change beliefs in social networks? We put this to the test by replicating an established study on belief dynamics, evaluating 12 LLMs across multiple model families and parameter sizes. The answer is a clear no, and in systematic ways.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
