Bayesian Belief Layer for Controllable Opinion Dynamics in LLM Agents
Bayesian Chronicle Agents add an explicit belief layer so LLM opinion dynamics can be controlled by stubbornness.
Bayesian Chronicle Agents add a minimal belief layer that separates what an LLM agent believes from how it speaks. Each stance is a probability updated by one Bayesian step per heard utterance, and a prior-strength parameter kappa controls stubbornness in the style of Friedkin-Johnsen opinion dynamics. Sweeping kappa produces consensus, persistent disagreement, and committed-minority influence; persistent disagreement matches the Friedkin-Johnsen closed-form fixed points at R-squared 0.93 to 0.99. Prescribed kappa remains recoverable after the language round-trip, with perfect rank-order recovery across four models, and the layer exposes per-model stance biases.
- Each stance is a probability updated by one Bayesian step per utterance heard.
- A single prior-strength parameter kappa sets stubbornness, following Friedkin-Johnsen dynamics.
- Sweeping kappa yields consensus, persistent disagreement, or committed-minority influence.
- Persistent disagreement matches Friedkin-Johnsen fixed points at R-squared 0.93 to 0.99.
- Prescribed kappa remains perfectly rank-ordered after the language round-trip on four models.
Full article150 words · extracted from arxiv.org · click to collapse
LLM agents in social simulation revise their opinions implicitly, in context: how open an agent is to persuasion can neither be specified nor verified, and collective outcomes inherit the model's training prior. We introduce Bayesian Chronicle Agents (BCA), a minimal belief layer separating \emph{what} an agent believes from \emph{how} it speaks. Each stance is a probability, updated by one Bayesian step per utterance heard. A single prior-strength parameter $κ$ encodes stubbornness, modeled after its role in Friedkin--Johnsen (FJ) opinion dynamics. We then sweep this parameter to yield three canonical regimes of opinion dynamics on demand (consensus, persistent disagreement, committed-minority influence), with persistent disagreement matching the FJ closed-form fixed points at $R^2\!=\!0.93$--$0.99$. We further show that prescribed $κ$ remains recoverable after the language round-trip, with perfect rank-order recovery across all four models. Explicit belief also makes simulation auditable: the layer surfaces systematic per-model stance biases that end-to-end simulation would silently absorb.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.21997