Project summary

Pax Machina is an online publication for designing and debating the institutions a world with powerful AI will need. The format is concrete proposals and the responses that critique and extend them, by and for the best ~100 researchers in the world working on these questions across mechanism design, social choice theory, political philosophy, legal scholarship, and AI governance.

What are this project's goals? How will you achieve them?

The goal is to put concrete institutional proposals on the table, make visible the serious research already underway, and build an ongoing public debate where designs get critiqued and refined quickly. This work is currently scattered across separate journals and disciplines, so it never accumulates and stays illegible to decision-makers.

We achieve this by cohering the best people working on the problem, which we have largely already done, then giving them a format built for debate rather than isolated essays: proposals paired with commissioned responses, each piece carrying a short note from another researcher on why it matters.

We have already gathered the researchers for this project and gotten their buy in. For more, see this strategy doc.

How will this funding be used?

We are requesting $65,000 for the first year. The three co-founders edit the publication themselves, so this covers everything except their time.

Designer ($30,000): build the site and article template, plus ongoing visual work. External commissions ($30,000): pool for commissioning pieces and responses from outside the core network. Copy editing ($5,000): freelance per-piece copy editing (~10 pieces).

Who is on your team? What's your track record on similar projects?

The meaning alignment institute is Ryan, Joe and Oliver. Ryan Lowe led InstructGPT and co-led GPT-4 alignment at OpenAI. Joe Edelman co-founded the Center for Humane Technology. Oliver Klingefjord works on market mechanisms and institutional design, previously technical lead at the AI Objectives Institute. We've worked with OpenAI in the past, on democratic mechanisms.

Over the past year we built a 60+ researcher network across DeepMind, Oxford, MIT, and Stanford, and ran two workshops. The first produced a position paper with 30+ co-authors, presented at ICML. The second ran alongside IASEAI in Paris and was reported as the most high-impact part of that conference week.

What are the most likely causes and outcomes if this project fails?

The most likely failure is that the publication becomes a set of standalone essays rather than a debate: eliciting a proposal is easy, eliciting a sharp response on a deadline is harder. We mitigate this by paying for commissioned responses. The second risk is cadence, since editing falls to the co-founders and MAI's other work could crowd it out. In neither case is the money wasted: the worst realistic outcome is a body of concrete proposals that exists, is edited well, and reached fewer people than we hoped.

How much money have you raised in the last 12 months, and from where?

MAI's current funding is from ARIA, covering the core team's salaries.