Vitalik Buterin Proposes AI Agents for Decentralized Voting

• Vitalik Buterin suggests personal AI agents to vote on behalf of users in DAOs.

• Public conversation agents could aggregate views using LLMs and zero-knowledge proofs.

• Multi-party computation and TEEs may secure private inputs in complex governance decisions.

Ethereum co-founder Vitalik Buterin outlined a new approach to governance using AI in a recent discussion. He argued that personal large language models could help users manage thousands of decisions in decentralized organizations. By doing so, AI would empower participants rather than concentrate power among a few delegates, addressing long-standing attention and expertise limitations.

Personal Governance Agents to Handle Voting

Buterin suggested personal AI agents could cast votes based on users’ writing, conversations, and declared preferences. If the agent is unsure of a person’s stance and the issue is significant, it prompts the individual directly.

This ensures that participants remain informed while maintaining influence over important choices. He emphasized that this model avoids the disempowerment often seen in standard delegation systems.

These personal agents could continuously align with users’ values, filtering relevant decisions while preserving human judgment. Unlike current delegation models, supporters retain influence beyond a single vote. The system also reduces cognitive load, making participation in complex decentralized autonomous organizations more feasible.

Public Conversation Agents Aggregate Collective Views

Buterin also addressed the challenge of aggregating information across groups. He proposed public conversation agents that summarize commonalities in participants’ inputs without revealing private data.

LLM-enhanced systems could convert personal viewpoints into shareable formats while protecting anonymity. Zero-knowledge proofs could further secure participant identities during discussions, allowing collective input while safeguarding privacy.

This method improves decision-making beyond linear voting models, which often fail to consider distributed knowledge. Participants’ AI agents could respond based on aggregated insights, enabling more accurate and informed consensus-building. The approach bridges the gap between private opinions and group-level deliberation.

Multi-Party Computation Supports Confidential Decisions

Finally, Buterin explored multi-party computation to handle decisions involving private information. Personal AI agents could process sensitive inputs in secure environments, such as TEEs or cryptographically guaranteed systems, and output only decisions.

Neither participants nor others see the underlying data, preserving confidentiality. This method applies to negotiations, disputes, and compensation decisions, ensuring privacy of both participant identities and content.

This layered approach combines personal AI, collective summarization, and cryptographic security, offering a potential blueprint for scaling democratic governance in decentralized systems.