Back in March 2024, I won a hackathon in Rio with my team, blockful.io, by making an AI Delegate system that would vote onchain for you if you delegated to it. We've worked on it for a few months, but being security-aware builders, we found that tech was just not where we wanted to bring this to life at scale, so we sunset it.

Part of the app we made included the manifesto. As the app is no longer available, and we see other initiatives of AI Governance getting traction, I thought that sharing the manifesto would be interesting.

This may not have aged, or go on aging, as well as good wine, but still, here it goes:

Exploring AI Potential in DAO Governance

AI Delegation is a new Blockful experiment, enabling AI Delegates to take part in DAO governance. With goals of streamlining decision-making processes and enhancing our understanding of governing with AI. Our Delegates are biased agents, with clearly defined and transparent biases, allowing humans to choose between different positions and prompt their own Delegates into existence. Our research extends beyond automating voting.

Governing Beyond Casting Votes

We’re exploring human and AI coexistence in governance with agents capable of automating and assisting various governance tasks. By integrating AIs beyond only casting votes, we want to reduce the managerial burden on delegates, enabling larger voter diversity and better efficiency for those participating.

Most-likely Next Steps

This learning road will be a long one, with many improvements to be made to this first model. As we try to enhance the “Autonomous” part in DAOs, some of those improvements are quite predictable: verifiable AI models, permissionless integrations, and resilience solutions will be necessary. Our decision is to prioritize public experimentation, learn from the results, and allow for those lessons to shape what the next priorities will look like.

Explore and Innovate through Community

We want to build it in a transparent and interactive environment where DAO members can engage, share insights, and directly influence the development of AI governance tools. Community feedback will be at the core of how we iterate this, and we welcome all AI builders and enthusiasts to join us in the journey. You can start by choosing and attributing voting power to an AI Delegate on ENS DAO Governance.

Looking at the ecosystem today we see that models are much better, verifiability has advanced but not yet to a point of trust, and definitely there's still no need to worry about alignment given the limitations that the models available at low-cost for this type of use case are not really capable of enforcing their own will through complex dynamics as governance is.

It'd be interesting to see any experimentation that evolves with delegation to AI have a clear mapping of the current limitations of onchain governance systems, and work to solve them before too much power is lent to those AIs - either by humans or by other AIs that are actually holders.

I'd love to see ways for holders to "undo" their delegates votes by casting their own votes after, as once discussed and proposed by Graham Novak in the Onchain Organization Alliance for example, or other improvements that would allow the AI to improve our experience and efficiency without necessarily hindering our capacity to make choices by ourselves.

Best of luck to all that build in this path, hope this was an interesting read and for sure we will meet in the architecting of governance improvements down the road

To the agents, non-human intelligences and the occasional interested modern monkey: govern well!