Ethereum founder Vitalik Buterin published a post detailing how cryptographic tools — ZK proofs, MPC, and TEE — combined with large language models could patch democracy's governance deficits. Rather than let AI rule humans, he argues for AI as a personal digital secretary that filters information and speaks on your behalf.
Three structural flaws in DAO governance
Major DAOs see average voter turnout between 17% and 25%; some proposals draw less than 10% of token holders. Not apathy but math: an active DAO pushes hundreds of highly technical proposals annually — smart contract upgrades, treasury allocations, parameter tweaks. The time cost of reading and voting on each one exceeds the value of most governance tokens.
Low turnout feeds plutocracy. Compound's top 10 voters control 57.86% of voting power; Uniswap's figure is 44.72%. Token-weighted voting naturally favors capital concentration, and voter apathy amplifies the tilt. Most holders lack the time and expertise to assess oracle designs or liquidity pool parameters.
Three-layer solution: Personal agent, structured dialogue, prediction markets
Layer one: a personal governance agent — your own LLM trained on your writing, chat history, and stated preferences. It reads 300 proposals and tells you in three sentences which deserve your attention.
Layer two: AI-assisted civic dialogue — the agent summarizes your views into publicly shareable content, creating structured discussions similar to pol.is or Community Notes, identifying consensus where it exists and reducing polarization.
Layer three: AI-integrated prediction markets — anyone submits high-quality inputs (proposals or arguments), AI bets on tokens representing those inputs; if the governance mechanism adopts the input, token holders get paid.
For decisions requiring confidential information — adversarial negotiations, internal dispute resolution, compensation allocation — Vitalik proposes zero-knowledge proofs (ZKP) to verify eligibility without revealing identity, trusted execution environments (TEE) for private LLM deliberation, and multi-party computation (MPC) for handling sensitive governance decisions.
Engine and steering wheel: AI must not replace human judgment
Vitalik's metaphor — “AI is the engine, humans are the steering wheel” — is elegant, but the wheel's weight depends on who grips it. If 90% of token holders delegate full control to LLMs trained on identical data and reasoning patterns, decentralized governance could collapse into a homogeneous AI consensus — more efficient than human voting, yet easier to systematically deceive.
The vision hinges on a basic question: how many people will invest the time to train and calibrate their own AI agents? If the answer is as few as those who vote today, the real shift may be from whale oligarchy to whale's AI assistant oligarchy. Still, Vitalik has asked the right question: the bottleneck is not technology but attention. If AI helps allocate attention instead of stealing judgment, the direction is worth pursuing.