Ethereum co-founder Vitalik Buterin pointed out that the limits of human attention span are the core problem plaguing decentralized autonomous organizations (DAOs) and democratic governance systems.
summary
- Buterin says the limits of human attention span are a core flaw in DAO governance.
- Personal AI agents can vote using your preferences and context.
- Proposal markets and MPCs have the potential to improve privacy and decision-making.
Writing in X, Buterin argued that participants are faced with thousands of decisions across multiple disciplines that they do not have enough time or skills to properly evaluate.
The usual solution of delegation is that a small group controls decision-making and creates disempowerment, with supporters having no influence after clicking the delegate button.
Buterin proposed a large-scale language model of the individual as a solution to the attention problem and shared four approaches. Privacy-preserving multiparty computation for private governance agents, public conversation agents, proposal markets, and sensitive decisions.
You may also like: Zcash price falls as Ethereum plans stealth addresses and ZK privacy features
Individual LLMs can vote based on their preferences
Personal governance agents perform all necessary voting based on preferences inferred from personal posts, conversation history, and direct statements.
If an agent faces uncertainty about voting preferences and believes the issue is important, it should ask the user questions directly while providing all relevant context.
“AI becomes the government” is dystopian. When the AI is weak, it leads to stagnation, and once the AI is strong, it maximizes destruction. But when used well, AI can be empowering and push the frontiers of democratic/decentralized forms of governance.
Core issues of democracy/…
— vitalik.eth (@VitalikButerin) February 21, 2026
A public conversation agent aggregates information from many participants before giving each individual or their LLM a chance to respond.
The system summarizes individual views, transforms them into a shareable format without exposing personal information, and identifies commonalities between inputs similar to the LLM-powered Polis system.
Buterin pointed out that good decision-making does not result from “a linear process of taking people’s opinions based solely on their own information and averaging them (even quadratic).” “The process must first aggregate collective information and then enable an informed response.
You may also like: Tether to downsize offshore yuan stablecoin CNH₮ amid weak demand
High-quality proposals may surface in the proposal market.
Governance mechanisms that emphasize high-quality inputs could implement prediction markets where everyone submits suggestions while AI agents bet on tokens. Once the mechanism accepts the input, the token holder will be paid.
This approach applies to suggestions, discussions, or any conversation unit that the system passes to participants. Market structures create financial incentives to surface valuable contributions.
Buterin argued that decentralized governance won’t work if sensitive information is needed to make important decisions. Organizations typically handle adversarial disputes, internal disputes, and compensation decisions by appointing individuals with significant authority.
Multiparty computations using a trusted execution environment can incorporate input from many people without compromising privacy.
“When you send your personal LLM to a black box, the LLM looks at your personal information, makes a decision based on it, and outputs only that decision,” Buterin explained.
Privacy protection is important as participants submit a large amount of input, including more personal information. Anonymity requires zero-knowledge proofs, which Buterin said should be built into all governance tools.
