7 Bringing AI Delegation to Arbitrum


Track: GovTech
Challenge Statement: “What technological solutions can we develop to reduce the complexity and friction in governance processes?”

Members: zeugh.eth and alextnetto.eth
Team Lead contact name or alias: Zeugh - Telegram: Contact @zeugh
Pitch: Loom

Proposal text:

1. Title

Bringing AI Delegates to Arbitrum DAO Governance

2. Abstract

Arbitrum DAO will delegate 5M ARB, divided equally between 5 AI Voting Agents created by Blockful. Three of these agents will have their “bias prompts” chosen through a JokeRace contest, while the other two will focus on Public Goods Funding and Arbitrum DAO Longevity.

3. Motivation

ARB holders should strive to keep the DAO treasury safe and efficient. However, human delegates may fail to be active. They can also vote contrary to delegators’ preferences, as humans agreeing on a topic might still disagree on others, creating a counter-incentive for delegation by lack of trust and resulting in even lower delegated supply and active voting delegations, which puts the DAO at risk.

AI Delegates can address these issues: they always vote, maintain a consistent perspective, and are immune to social pressure, lobbying, or bribery.

By adopting AI Delegates, Arbitrum DAO can improve its governance body, ensuring decisions are made more efficiently, inclusively, and transparently. This proposal aims to create a resilient and adaptive governance tool leveraging the strengths of both human and AI participants.

Drawing from the insights of the “Event Horizon” proposal, the integration of AI Delegates addresses the fundamental issues of low voter participation and the unpredictability of human delegates. The benefits of this approach include:

  1. Consistent and Reliable Voting: AI Delegates always vote in accordance with their predefined biases, ensuring that the DAO’s decisions are made with consistent logic and reasoning. This reliability addresses one of the major shortcomings of human delegates, who may fail to participate in all votes or be influenced by external factors.
  2. Elimination of Human Shortcomings: Human delegates can be influenced by external factors such as lobbying, bribery, and social pressure. AI Delegates are immune to these influences, making them a reliable alternative. This eliminates the risk of votes being swayed by non-transparent factors and ensures that decisions are made based solely on predefined biases and the DAO Dataset.
  3. Encouragement of Delegation: By providing a reliable and transparent voting mechanism, AI Delegates can encourage more ARB holders to participate in the governance process through delegation. This addresses the issue of low delegated supply and strengthens the DAO’s governance structure. ARB holders can delegate their voting power to AI Delegates with confidence, knowing that their votes will be cast consistently and transparently.

4. Rationale

Integrating AI Delegates aligns with the Arbitrum community’s mission by enhancing governance quality and security. This initiative provides a reliable delegation option, ensuring diverse preferences are represented. It promotes consistent participation and mitigates risks associated with low voter turnout.

Enhancing Voter Participation and Mitigating Risks:
AI Delegates ensure consistent voting and are free from external influences such as bribery and social pressures. This consistency mitigates risks associated with low participation, safeguarding the DAO treasury and decision-making processes.

Inclusive Representation and Community Engagement:

  • Bias Prompts: The use of bias prompts allows diverse perspectives to be represented consistently through different AI Delegates. The JokeRace contest engages the community in selecting these prompts, ensuring they reflect a wide range of views.
  • Alignment with DAO Values: Community-selected bias prompts ensure AI Delegates align with Arbitrum DAO’s community preferences, promoting integrity and effectiveness in governance.

Transparency and Accountability:

  • Clear Voting Process: AI Delegates operate transparently with community-defined biases, ensuring stakeholders understand the rationale behind each vote and give their reasoning on the votes.
  • Monitoring Tools: Blockful’s tools provide real-time insights into AI Delegate voting behaviors, enhancing accountability and trust in the governance process.
  • Stable Voting Activity: Consistent AI Delegate participation maintains stable voting power, reducing volatility and promoting informed, balanced decisions.

5. Key Terms

  • Voting Agents (or AI Delegates): AI agents with an address and a specific “Bias Prompt” that vote in the DAO governor contract for each proposal.
  • DAO Dataset: A dataset including the DAO constitution and governance artifacts that guide Voting Agents’ decisions.
  • Bias Prompt (or Bias): A Voting Agent’s “core belief” used to interpret proposals and cast votes.

6. Specifications

  • AI Technology by Blockful: Development and customization of AI Voting Agents to suit Arbitrum DAO’s needs, including programming specific bias prompts.
  • JokeRace Contest Platform: Used to select bias prompts for three AI Delegates, fostering community engagement. It can also be used in the future to create new biases as the DAO decides to change or expand its AI Delegate body.
  • Existing Governance Contracts: AI Voting Agents will interact with existing DAO governance contracts, requiring no new smart contracts.
  • Monitoring and Evaluation Tools: Blockful provides for monitoring AI Delegate performance, viewing biases, delegating, and accessing the DAO Dataset.

You can check more about Blockful’s AI Delegate system details on https://blockful.notion.site/AI-Delegate-8296ff74623a4078b1ab7e7536adfabf

For the Arbitrum implementation, the following diagram includes all the elements exposed in the previous sections:

7. Steps to Implement

  1. Initial setup during the GovHack
    • Setup of the addresses and ENS of AI delegates
    • Setup of example test JokeRace contest
    • Setup of the backend to listen to the right contracts
    • Setup of the backend to vote in the right contract
  2. Assessment Phase (1-2 weeks):
    • Analyze details of the governance processes.
    • Gather artifacts for the DAO Dataset.
  3. Development Phase (1-2 weeks):
    • Update Blockful’s AI Delegate system for Arbitrum DAO governance specifics.
    • Set up the JokeRace contest for community bias prompt selection.
    • Further specify the AI Delegates focusing on Public Goods Funding and DAO Longevity with Arbitrum’s context.
  4. Pilot Phase (1-2 weeks):
    • Implement the pilot with 5 AI Delegates.
    • Monitor performance and gather data on efficiency and satisfaction.
  5. Full Implementation
    • Refine AI Delegates based on pilot results.
    • Continuously monitor and update AI Delegates.

8. Timeline

  • Assessment Phase: 1-2 weeks
  • Development Phase: 1-2 weeks
  • Pilot Phase: 1-2 weeks after setup
  • Full Implementation: Following successful pilot

JokeRace Pilot Contest Timing:

  • Submissions: July 6, 2024, 5:25 PM CEST to July 7, 2024, 8:00 AM CEST
  • Voting: July 7, 2024, 8:00 AM CEST to July 7, 2024, 4:30 PM CEST

9. Overall Cost

Blockful is waiving the initial costs for this proposal for the first pilot. We envision retro funding once we are able to validate the system efficiency for Arbitrum DAO.

We request a delegation of 5M Arbitrum, divided equally between the 5 AI Delegates to be deployed, and support from the community to improve the product and the DAO governance with it.