LLM-Consensus: Multi-Agent Debate for Visual Misinformation Detection

2024-10-26

AI Summary

To address the challenge of cross-modal misinformation detection exacerbated by out-of-context (OOC) image-text misalignment, this paper proposes a domain-agnostic multi-agent debate framework. The method integrates multimodal large language models (MLLMs) with dynamic external knowledge retrieval, enabling specialized agents to collaboratively debate image-text consistency—achieving zero-shot, interpretable visual misinformation detection. Key contributions include: (i) the first debate-driven multi-agent reasoning mechanism, replacing single-point classification with knowledge-augmented consensus decision-making; and (ii) explicit modeling of cross-modal semantic alignment and OOC reasoning. Evaluated on mainstream benchmarks, the framework achieves state-of-the-art performance. Ablation studies confirm that external knowledge retrieval significantly improves accuracy. User studies further demonstrate its effectiveness in enhancing both expert and public discrimination capability and trustworthiness.

Technology Category

• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Computer Vision: Multi-modal Vision
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Responsible Web: Mitigating misinformation and disinformation

Abstract

One of the most challenging forms of misinformation involves the out-of-context (OOC) use of images paired with misleading text, creating false narratives. Existing AI-driven detection systems lack explainability and require expensive finetuning. We address these issues with LLM-Consensus, a multi-agent debate system for OOC misinformation detection. LLM-Consensus introduces a novel multi-agent debate framework where multimodal agents collaborate to assess contextual consistency and request external information to enhance cross-context reasoning and decision-making. Our framework enables explainable detection with state-of-the-art accuracy even without domain-specific fine-tuning. Extensive ablation studies confirm that external retrieval significantly improves detection accuracy, and user studies demonstrate that LLM-Consensus boosts performance for both experts and non-experts. These results position LLM-Consensus as a powerful tool for autonomous and citizen intelligence applications.

Problem

• Fake Information Detection
• Image-Text Integration
• AI Accuracy Improvement

Innovation

• LLM-Consensus
• Multi-perspective Analysis
• Automated Fact-checking