Authors: Sejal, DavidDAO and @carlb
Inspired by our conversation with Audrey Tang and Vitalik’s vision of “AI as engines, humans as steering wheels,” DeepGov explores how configurable AI agents could help scale funding decisions while keeping human values at the center.
Forked from Deep Funding and powered by a $25K public goods fund from @owocki, we’re bringing this to life during Gitcoin Grants 23.
Figure 1: The “AI as the engine, Human as steering wheel” framework.
What is DeepGov?
DeepGov aims to explore a future where configurable AI agents—aka AI politicians—can help scale decision-making in public goods funding, guided by citizen values. These agents don’t replace us—they are the engines. We, the humans, remain at the steering wheels.
Inspired by conversations with Audrey Tang and Takahiro Anno’s historic AI-assisted political campaign in Japan, DeepGov is both a thought experiment and a live pilot in AI governance.
Figure 2: Takahiro’s AI-based political campaign, earning the most votes in Japanese history for a newcomer
GG23 DeepGov Experiment: What We’re Doing
This is our plan for GitcoinGrants23:
Step 1: Broad Listening
We begin by gathering insights from the Gitcoin community using Polis, a tool for collective sensemaking. This helps us understand what matters to the Gitcoin community and the important questions these agents need to address.
Jump into this questionnaire to share your perspectives and help surface the most important questions and values. Your input will directly shape the design of our AI agents and the principles they operate on.
Step 2: Build the AI Agents
Using community input, we’ll create a diverse cast of AI politicians—each with distinct governance philosophies and value systems:
Grant – the Gitcoin Communist
Luna – the Open Source Capitalist
Panda – the Regenerator
These agents are built using a ModelSpec & Constitutional AI framework, and each reflects a different worldview for how public goods should be funded. In the future, you can fork these and create your models.
Step 3: Simulate Funding Proposals
Luna, Panda, and Grant will review all GG23 OSS projects participating in the QF round and generate reviews and funding suggestions—each through the lens of their unique ideology.
We’ll publish their reviews so the community can see how each agent reasons and make their judgments—agree, disagree, debate. It should be fun to see how different agents “think” about the same projects.
Step 4: Elect Your AI Representative
In the second week of the GG23 donation period, we’ll kick off a political-style campaign where you, the community, vote to elect the AI agent you want to represent your values.
Voting will use quadratic voting, and every participant will receive 100 base voting credits. You can boost your voting power by donating—the more you contribute, the more credits you unlock. All contributions go toward funding public goods.
Why This Matters
DeepGov is a sandbox to explore big questions:
-
Can AI agents effectively fund public goods while remaining accountable to human values?
What does it take to build models that reflect community priorities—and can we trust them to act on our behalf? -
How do we preserve pluralism and transparency in automated decision-making?
If different communities hold different values, how do we make sure no single ideology dominates? How do we make AI reasoning legible and challengeable? -
What happens when AI models become political entities with ideologies?
Can AI agents campaign, debate, and be held accountable like human representatives? What kind of governance do we enable—or risk—when we give them power?
We’re not trying to automate the future—we’re trying to make it co-created, legible, and steerable.
Help Us Steer DeepGov
This is just the beginning. We’re inviting the Gitcoin community to help shape how this experiment unfolds—and what values our AI agents should embody.
Jump into this Polis questionnaire now!
What to Expect Next
Over the coming week(s), we’ll share regular updates on how DeepGov is evolving—including where we need you, the community, to be the steering wheel for these AI engines.
We’ll also publish deeper dives and explainers to break down the concepts behind the experiment—from model specs and value alignment to participatory budgeting and quadratic voting—so everyone can follow along, learn, and contribute.
And finally, a huge thank you to @owocki, @MathildaDV, and @deltajuliet for believing in this experiment and encouraging us to bring it to life.
Let’s build the future of capital allocation—together.
