Even if every AI in a group is trying to be fair, putting them together in a 'swarm' accidentally turns them into a polarized mob.
April 13, 2026
Original Paper
Aligned Agents, Biased Swarm: Measuring Bias Amplification in Multi-Agent Systems
arXiv · 2604.08963
The Takeaway
It disproves the idea that collaboration naturally balances out bias, showing that AI workflows can unintentionally become echo chambers. This is critical for developers, as the structure of the team itself can create bias where none existed before.
From the abstract
While Multi-Agent Systems (MAS) are increasingly deployed for complex workflows, their emergent properties-particularly the accumulation of bias-remain poorly understood. Because real-world MAS are too complex to analyze entirely, evaluating their ethical robustness requires first isolating their foundational mechanics. In this work, we conduct a baseline empirical study investigating how basic MAS topologies and feedback loops influence prejudice. Contrary to the assumption that multi-agent col