Ethereum co-founder Vitalik Buterin has proposed a significant overhaul of how decentralized autonomous organizations (DAOs) function, advocating for the use of personal artificial intelligence (AI) agents to automate voting and address issues of low participation and centralized power. The proposal, detailed on social media platform X, comes roughly a month after Buterin publicly criticized the current state of many DAOs, suggesting they were drifting towards traditional, centralized control structures.
Buterin’s plan centers on shifting away from the current model where individuals delegate their voting power to large token holders. Instead, users would deploy their own AI models – large language models (LLMs) – trained on their individual values and past communications. These AI “stewards” would then vote on the numerous decisions facing DAOs, effectively scaling governance participation.
“There are many thousands of decisions to produce, 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,” Buterin wrote, as reported by CoinDesk. “So what can we do? We use personal LLMs to solve the attention problem.”
A core component of Buterin’s proposal focuses on ensuring privacy, and security. The system would utilize zero-knowledge proofs (ZKPs) to allow users to verify their eligibility to vote without revealing their wallet addresses or voting choices, according to CoinDesk. This aims to prevent coercion, bribery, and the practice of “whale watching,” where smaller token holders simply mimic the decisions of larger ones.
To further safeguard sensitive data, Buterin suggests employing secure computing environments like multi-party computation (MPC) and trusted execution environments (TEEs). These technologies would allow AI agents to process confidential information – such as job applications or legal disputes – without exposing it on the public blockchain, as outlined in the CoinDesk report.
Addressing the growing problem of spam and low-quality proposals, exacerbated by the increasing accessibility of generative AI, Buterin proposes the implementation of prediction markets. These markets would incentivize AI agents to identify and support valuable proposals, while penalizing the submission of irrelevant or malicious content. Agents could wager on the likelihood of a proposal’s acceptance, with successful predictions earning payouts.
Buterin’s call for a redesign of DAOs extends beyond simply improving voting mechanisms. He has previously argued that DAOs should be tailored to solve specific infrastructure problems, including the need for more reliable oracles, onchain dispute resolution systems, and long-term project stewardship, as noted by Cointelegraph. He frames this thinking through a “convex vs concave” governance lens, suggesting that DAOs should prioritize robustness and aggregation of input for “concave” problems where compromise is beneficial.
The Block reported that Buterin urged the crypto community to develop more advanced and effective DAOs, stating, “We need more DAOs — but different and better DAOs.” This sentiment echoes his earlier criticism that current DAOs often function as little more than “a treasury controlled by token holder voting,” a model he deems inefficient and vulnerable to manipulation.
BitgetApp highlighted Buterin’s identification of human attention limits as a fundamental obstacle to effective DAO and democratic governance systems. The proposed AI stewards are presented as a potential solution to this challenge, automating routine governance tasks and flagging only critical issues for human review.