Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate
Стимулирование дивергентного мышления у больших языковых моделей с помощью дебатов между несколькими агентами
2024-01-01
SCID: 54.1/kxth2ejn
Discuss with AI
Degeneration-of-Thoughtcounterintuitive arithmetic reasoningdivergent thinkinglarge language modelsmulti-agent debate
Figures from the paper
Abstract (AI)
Modern large language models (LLMs) like ChatGPT have shown remarkable performance on general language tasks but still struggle on complex reasoning tasks, which drives the research on cognitive behaviors of LLMs to explore human-like problem-solving strategies.Along this direction, one representative strategy is self-reflection, which asks an LLM to refine the solution with the feedback generated by itself iteratively.However, our study shows that such reflection-style methods suffer from the Degeneration-of-Thought (DoT) problem: once the LLM has established confidence in its solutions, it is unable to generate novel thoughts later through reflection even if its initial stance is incorrect.To address the DoT problem, we propose a Multi-Agent Debate (MAD) framework, in which multiple agents express their arguments in the state of "tit for tat" and a judge manages the debate process to obtain a final solution.Clearly, our MAD framework encourages divergent thinking in LLMs which would be helpful for tasks that require deep levels of contemplation.Experiment results on two challenging datasets, commonsense machine translation and counterintuitive arithmetic reasoning, demonstrate the effectiveness of our MAD framework.Extensive analyses suggest that the adaptive break of debate and the modest level of "tit for tat" state are required for MAD to obtain good performance.Moreover, we find that LLMs might not be a fair judge if different LLMs are used for agents.Code is available at https://github. com/Skytliang/Multi-Agents-Debate.
Key Findings
1
Effective debate requires adaptive termination and a modest level of reciprocal “tit for tat” interaction between agents.
2
Experiments on commonsense machine translation and counterintuitive arithmetic reasoning demonstrate that Multi-Agent Debate improves performance on challenging reasoning tasks.
3
Language models may act as unfair judges when the debating agents use different underlying LLMs.
4
Self-reflection methods exhibit a Degeneration-of-Thought problem: confident but incorrect initial solutions prevent later generation of novel thoughts.
5
The proposed Multi-Agent Debate framework uses multiple arguing agents and a judge to promote divergent thinking and derive final solutions.
Research Object
large language models (LLMs) engaged in multi-agent debate
Research Subject
the encouragement of divergent thinking and its effects on complex reasoning performance, including the Degeneration-of-Thought problem and debate-process factors
Publication Details
Publication Date
2024-01-01
Journal
Publisher
ISSN
Cited by
1097
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
Cited by9
A Survey of Large Language Models2026
Empowering biomedical discovery with AI agents2024
Exploring Large Language Model‐Based Intelligent Agents: Definitions, Methods, and Prospects2026
A review of large language models and autonomous agents in chemistry2024
AI Agents Under Threat: A Survey of Key Security Challenges and Future Pathways2025
Evaluating clinical AI summaries with large language models as judges2025
Towards trustworthy agentic AI: a comprehensive survey of safety, robustness, privacy, and system security2026
The Agentic Digital Twin-Led Assessment System (ATLAS): A new architecture for psychological assessment in the era of agentic AI2026
Heterogeneous Swarms: Jointly Optimizing Model Roles and Weights for Multi-LLM Systems2025