Belief Engine: Configurable and Inspectable Stance Dynamics in Multi-Agent LLM Deliberation
Joshua C. Yang, Maurice Flechtner, Damian Dailisan, Michiel A. Bakker
Why It Matters
What makes this one worth your time
Understanding and controlling stance dynamics in AI agents can enhance their reliability and transparency in applications like negotiation and conflict resolution.
Belief Engine offers a transparent method for tracking stance changes in LLM-based multi-agent deliberations.
Summary
The paper introduces the Belief Engine, a belief-update layer for LLM-based agents that simulates deliberative interactions by treating beliefs as evidential states and exposing them as scalar stances. It uses a log-odds rule to update stances based on evidence uptake and prior anchoring, aiming to provide a transparent infrastructure for studying evidence-grounded deliberation.
Key contributions
- Introduction of the Belief Engine for auditable belief updates in LLM-based agents.
- Demonstration of stance dynamics control through parameter sweeps across multiple LLMs.
- Application to a human deliberation dataset to validate the system's ability to reconstruct participant stances.
Notable insights
- The use of a log-odds rule to update beliefs based on evidence uptake and prior anchoring is a clever approach to model stance dynamics.
- The system's ability to reconstruct participant stances from a human deliberation dataset highlights its potential for analyzing real-world deliberations.
Possible limitations
- Not stated in the abstract
Abstract
arXiv:2605.15343v1 Announce Type: new Abstract: LLM-based agents are increasingly used to simulate deliberative interactions such as negotiation, conflict resolution, and multi-turn opinion exchange. Yet generated transcripts often do not reveal why an agent's stance changes: movement may reflect evidence uptake, anchoring, role drift, echoing, or changed prompt and retrieval context. We introduce the Belief Engine (BE), an auditable belief-update layer that treats "belief" as an evidential state over a proposition and exposes it as scalar stance. BE extracts arguments into structured memory and updates stance with a log-odds rule controlled by evidence uptake u and prior anchoring a. Across multiple base LLMs, parameter sweeps show that these controls reliably shape stance dynamics while preserving an evidence-level update trail. On DEBATE, a human deliberation dataset with pre/post opinions, BE best reconstructs participants whose final stance follows extracted evidence; stable and evidence-opposed cases instead point to anchoring or factors outside the extracted evidence stream. BE provides configurable infrastructure for studying evidence-grounded deliberation, where openness, commitment, convergence, and disagreement can be tied to explicit update assumptions rather than hidden prompt effects.