Ethereum cofounder Vitalik Buterin has suggested using personal AI agents to automate voting in decentralized autonomous organizations, aiming to boost participation and protect privacy. The proposal, shared on social media platform X, addresses issues like low voter turnout and power concentration among large token holders. It incorporates cryptographic tools to safeguard sensitive data and prevent coercion.
Vitalik Buterin, cofounder of Ethereum, outlined a plan to overhaul governance in decentralized autonomous organizations (DAOs) by deploying AI 'stewards.' Published on social media platform X about one month after Buterin critiqued DAOs for declining participation and centralized power, the idea shifts away from users delegating votes to major token holders.
Under the proposal, individuals would use AI models trained on their past messages and values to handle thousands of DAO decisions across various expertise areas. Buterin noted the challenge: “There are many thousands of decisions to make, involving many domains of expertise, and most people don't have the time or skill to be experts in even one, let alone all of them.” He added, “So what can we do? We use personal LLMs to solve the attention problem.”
To ensure privacy, the AI agents would operate in secure environments like multi-party computation (MPC) or trusted execution environments (TEEs), processing private data without public exposure on the blockchain. Anonymity would be maintained through zero-knowledge proofs (ZKPs), allowing users to verify eligibility without revealing wallet addresses or votes. This setup aims to counter coercion, bribery, and the practice of smaller voters copying large holders' decisions, known as whale watching.
The AI stewards would manage routine governance tasks and alert humans only to critical matters. To combat low-quality or spammy proposals, especially amid rising generative AI use, Buterin recommended prediction markets where agents bet on proposal acceptance. Successful predictions would yield rewards, promoting useful input while discouraging irrelevant ones.
Additionally, the system would enable privacy tools for evaluating sensitive information, such as job applications or legal disputes, without blockchain leaks.
