Free-MAD: Consensus-Free Multi-Agent Debate

2025-09-13

AI Summary

Existing multi-agent debate (MAD) methods rely on iterative multi-round interaction followed by majority voting, suffering from three critical limitations: excessive token consumption, error propagation induced by large language models’ conformity bias, and decision randomness and unfairness inherent in voting-based aggregation. This paper proposes Free-MAD—the first consensus-free, single-round MAD framework. Its core innovations are: (1) eliminating iterative negotiation in favor of parallel, one-shot debate; (2) replacing majority voting with a holistic reasoning trajectory scoring mechanism that evaluates the quality of full inference paths; and (3) incorporating an anti-conformity strategy to suppress interference from erroneous agent responses. Evaluated across eight benchmarks, Free-MAD achieves significant gains in reasoning accuracy and robustness, reduces communication cost by approximately 50%, and demonstrates superior resilience under adversarial attacks.

Technology Category

• Multiagent Systems: Adversarial Agents
• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Multi-agent debate (MAD) is an emerging approach to improving the reasoning capabilities of large language models (LLMs). Existing MAD methods rely on multiple rounds of interaction among agents to reach consensus, and the final output is selected by majority voting in the last round. However, this consensus-based design faces several limitations. First, multiple rounds of communication increases token overhead and limits scalability. Second, due to the inherent conformity of LLMs, agents that initially produce correct responses may be influenced by incorrect ones during the debate process, causing error propagation. Third, majority voting introduces randomness and unfairness in the decision-making phase, and can degrade the reasoning performance. To address these issues, we propose extsc{Free-MAD}, a novel MAD framework that eliminates the need for consensus among agents. extsc{Free-MAD} introduces a novel score-based decision mechanism that evaluates the entire debate trajectory rather than relying on the last round only. This mechanism tracks how each agent's reasoning evolves, enabling more accurate and fair outcomes. In addition, extsc{Free-MAD} reconstructs the debate phase by introducing anti-conformity, a mechanism that enables agents to mitigate excessive influence from the majority. Experiments on eight benchmark datasets demonstrate that extsc{Free-MAD} significantly improves reasoning performance while requiring only a single-round debate and thus reducing token costs. We also show that compared to existing MAD approaches, extsc{Free-MAD} exhibits improved robustness in real-world attack scenarios.

Problem

• Reduces token overhead and improves scalability in multi-agent debates
• Mitigates error propagation from incorrect responses during debates
• Eliminates randomness and unfairness in consensus-based decision making

Innovation

• Score-based decision mechanism for debate trajectory
• Anti-conformity mechanism to reduce majority influence
• Single-round debate design reducing token costs