Ethereum Co-Founder Proposes AI Agents for DAO Governance
Ethereum co-founder Vitalik Buterin is advocating for the use of AI agents to represent stakeholder interests in decentralized autonomous organization (DAO) voting. The proposal aims to address challenges in decentralized governance by creating a more scalable and potentially more rational decision-making process. This experiment could significantly alter onchain governance and market microstructure by introducing programmable, agent-driven capital allocation.
- The proposal directly targets low voter turnout in DAOs, which typically ranges from 15% to 25%, a condition that concentrates power and creates vulnerabilities to governance attacks. - To maintain voter privacy and prevent coercion, the framework suggests implementing zero-knowledge proofs (ZKPs), multi-party computation (MPC), and trusted execution environments (TEEs), allowing AI agents to process private user data without exposing it on a public blockchain. - A key technical challenge is the computational cost of running large language models on-chain; a proposed solution is Zero-Knowledge Machine Learning (ZKML), where model execution happens off-chain and a cryptographic proof of the result is verified on-chain. - To combat the risk of AI-generated spam proposals, Buterin suggests integrating prediction markets where agents can bet on the probability of a proposal's acceptance, creating a financial incentive to submit high-quality ideas. - The concept extends beyond voting to include AI agents that summarize complex technical debates and a five-step plan from Ethereum Foundation's Tomasz Stańczak for LLM-driven creation, review, and moderation of Ethereum Improvement Proposals (EIPs). - This idea is not unique to Ethereum; Lane Rettig, a researcher at the Near Foundation, is developing a similar concept of "AI-powered digital twins" to vote on behalf of DAO members to address low participation. - A case study simulating an AI-assisted governance framework on KlimaDAO, a carbon-focused DAO, demonstrated a 97% alignment with historical human voting decisions and projected a potential 40% increase in voter participation.
