AutoVote

DAO governance is broken. Its plagued with unintended centralization, misinformed decisions, a lack of participation, and sparse community awareness. AutoVote fixes that with on-demand, autonomous governance, guaranteeing complete participation, uptime, involvement, and (automated) awareness.

Motivated by Illia Polosukhin’s thesis on governance facilitated by AI agents and SUAVE’s capabilities of creating a network of TEE coprocessors with trusted on-chain settlement, AutoVote is an automated governance solution that utilizes personalized AI agents to act as on-demand proxies for DAO governance members, or as ‘centralized’ coordinators of a DAO’s governance efforts.

The project utilizes SUAVE for its decentralized network of TEE coprocessors that settle verified off-chain transactions on-chain to train personalized LLM models that serve as on-demand voting proxies for voting DAO members. These models are trained on the network, and I utilized the suave-geth precompile that connects to OpenAI's ChatGPT model to train a singular voting agent. This agent receives delegate preferences, past voting data, the current proposal, and general voter strategy as an input, and generates a 'yay' or 'nay' response with supported reasoning.

This response is hashed and sent as a signed transaction from the user to the smart contract representing a certain proposal within a DAO. This response is used as the vote on the DAO's proposal, which then guarantees voter participation, even if the voter is not actively involved in studying the proposal or has external commitments that prevent them from being able to take part in the governance process for a certain cycle.

AutoVote utilizes SUAVE as a decentralized network of trusted coprocessors, expanding its capabilities beyond being a programmable and private mempool.

This network is used to train personalized LLM models that serve as on-demand voting proxies for voting DAO members. These models are trained on the network, and I utilized the suave-geth precompile that connects to OpenAI's ChatGPT model to train a singular voting agent. This agent receives delegate preferences, past voting data, the current proposal, and general voter strategy as an input, and generates a 'yay' or 'nay' response with supported reasoning.

This response is hashed and sent as a signed transaction from the user to the smart contract representing a certain proposal within a DAO. This response is used as the vote on the DAO's proposal, which then guarantees voter participation, even if the voter is not actively involved in studying the proposal or has external commitments that prevent them from being able to take part in the governance process for a certain cycle.