Buterin Links Crypto Anti-Collusion Rules to AI Safety Concerns

Vitalik Buterin, co-founder of Ethereum, believes that anti-collusion mechanisms he proposed for blockchain governance could be more relevant to AI safety than cryptocurrency itself.

In a recent post on X, Buterin pointed out the similarity between crypto governance and multi-agent AI systems. He argued that both involve humans interacting with algorithms, but in AI safety, humans are managing stronger agents while facing challenges from weaker ones.

To address this issue, Buterin drew from his 2020 essay 'Coordination, Good and Bad', where he proposed limits on how much agents can collude. He suggested decentralization, secret ballots, privacy protections, whistleblowers, communication limits, and mechanisms that make participants bear the cost of their decisions.

This idea has been echoed by researcher Eric Drexler in his essay 'AI Safety Puts the Same Idea in a Different Setting'. Drexler argued that AI collusion becomes easier when agents are similar, share objectives, communicate freely, observe each other's actions, and retain information across repeated interactions. His proposed countermeasures include using diverse agents, constraining communication between them, and imposing critics with the authority to intervene.