Multi-Agent Systems¶
Multi-agent systems (MAS) in the context of information integrity research involve coordinated autonomous agents—often powered by large language models (LLMs)—working together to generate, evaluate, refine, or detect misinformation and propaganda.
Scope¶
In fake-news and misinformation research, multi-agent approaches have emerged as a tool for: - Automated content generation: Multiple agents with specialized roles (generator, evaluator, refiner) coordinate to produce narratives (counter-narratives, fact-checks). - Simulated evaluation: Agent personas impersonating target audiences provide feedback on narrative effectiveness without exposing real users. - Iterative refinement: Agents iteratively improve content quality by exchanging feedback loops, reducing human labeling burden. - Distributed analysis: Agents handle different subtasks (stance detection, veracity assessment, frame analysis) and aggregate findings.
Key papers¶
- A Multi-Stage Agentic Framework for Effective Counter-Narrative Generation and Refinement — three-agent framework (Generator, Evaluator, Mediator) coordinating to refine counter-narratives via simulated pro-Russian user feedback.
- Agentic Mixed-Source Multi-Modal Misinformation Detection with Adaptive Test-Time — multi-modal misinformation detection using agentic test-time scaling.
Related topics¶
- Generative AI (underlying technology)
- Computational Propaganda (application domain)
- Misinformation and fake news detection (complementary task)