Beyond Arrow's Impossibility: Fairness Emerges from Multi-Agent LLM Collaboration

arXiv·low signal

Paper reframes AI fairness from a single-model optimization problem to an emergent property of multi-agent interaction, drawing on social choice theory (Arrow's impossibility theorem) to show that what no single model can guarantee, collaborating agents can produce through exchange. As agentic AI systems increasingly involve multiple LLMs coordinating, this provides theoretical grounding for why multi-agent architectures may naturally produce fairer outcomes than monolithic models. Early-stage but directionally important for multi-agent system designers.