Paper Overview
Field: Machine Learning Authors: Michał Wawer, Jarosław A. Chudziak Published: 2025-06-01 arXiv: 2506.00633
Summary
Multi-agent systems are commonly designed to reduce disagreement through voting, consensus protocols, debate, or fault-tolerant aggregation. This paper argues that this objective is insufficient for value-laden tasks, where disagreement may reflect genuine normative uncertainty rather than agent error.
Building on prior work on reasoning-trace disagreement in human-AI collaborative moderation, the authors propose a knowledge-representation layer in which reasoning traces and agent decisions are abstracted into symbolic disagreement states. Given agents producing explicit reasoning traces and binary decisions, four states are distinguished according to reasoning similarity and conclusion agreement:
- Convergent agreement — similar reasoning, same conclusion
- Divergent agreement — different reasoning, same conclusion
- Convergent disagreement — similar reasoning, different conclusions
- Divergent disagreement — different reasoning, different conclusions
- Paper: https://arxiv.org/abs/2506.00633
These states support defeasible strategic routing rules. The framework is instantiated in content moderation, and the authors argue that disagreement-aware routing provides a bridge between sub-symbolic LLM reasoning and symbolic knowledge representation for multi-agent strategic reasoning.