AINeutralarXiv – CS AI · 7h ago6/10
🧠
Hidden Coalitions in Multi-Agent AI: A Spectral Diagnostic from Internal Representations
Researchers introduce a spectral diagnostic method to detect hidden coalitions in multi-agent AI systems by analyzing mutual information patterns in internal neural representations rather than observable behavior. The technique successfully identifies hierarchical and dynamic coalition structures in reinforcement learning and language models, providing a scalable tool for monitoring emergent organization in distributed AI systems.