Belief Engine: Configurable and Inspectable Stance Dynamics in Multi-Agent LLM Deliberation

2026-05-14

AI Summary

Current large language model–driven multi-agent negotiation systems lack interpretability regarding the reasons behind stance shifts, such as evidence uptake, anchoring effects, or prompt variations. This work proposes the Belief Engine (BE), which, for the first time, models stances as evidence states over propositions and explicitly represents them in scalar form. BE enables auditable and configurable dynamic belief updating by extracting arguments from structured memory and integrating a log-odds rule with parameterized evidence adoption rates and prior anchoring strengths. Experiments on the DEBATE dataset demonstrate that the approach accurately replicates human-like, evidence-driven stance evolution and reveals stable or counter-evidential behaviors arising from strong anchoring or external influences, substantially enhancing the interpretability and mechanistic analysis of negotiation processes.

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.

Problem

• stance dynamics
• multi-agent deliberation
• belief update
• evidence uptake
• LLM-based agents

Innovation

• Belief Engine
• stance dynamics
• evidence-grounded deliberation
• auditable belief update
• multi-agent LLM