Bayesian belief layer enables controlled opinion shifts in AI agents
Bayesian Belief Layer for Controllable Opinion Dynamics in LLM Agents
Multiagent SystemsArtificial Intelligence
Summary
It's hard to know how AI agents change their opinions when they talk, because those changes happen inside their 'black box' brains. The authors introduce a simple system that keeps track of what each AI agent truly believes, separately from how it talks. This system uses probabilities that update logically every time the agent hears something new, and a stubbornness setting that controls how easily opinions change. This helps simulations show different group behaviors like agreement or persistent disagreement and lets people see hidden biases in the AI agents’ thinking.
What this means in practice
- •For simulation developers: Create social simulations where agent beliefs update transparently and controllably through Bayesian steps.
- •For ai system auditors: Audit large language model agents for hidden biases by explicitly modeling and inspecting their belief states.
Authors
Hafsa Akbar, Daniel Platnick, Marjan Alirezaie, Hossein Rahnama
Abstract
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.