Invitation

Join an AI war game.

A structured role-play of AI takeoff, based on the AI Futures Project's tabletop exercise. The goal is to better understand the dynamics of the coming years by playing them out as realistically as we can.

Each player takes a role, such as the US government, a leading AI lab, its safety team, China, or the press, and acts as that actor realistically would. A facilitator plays the rest of the world and rolls dice when an outcome is genuinely uncertain. We start in January 2027 with AI progress on its current trend, play in three-month turns until superintelligence arrives or time runs out, then go over what happened.

Please read AI 2027 and AI 2040 beforehand.

When
Date TBDStarts 19:00, about 3.5 hours
Where
Eitan's placeJaffa. Address in the group chat.
Players
6 to 10possibly in pairs, see below

Roles

The roles below are a first draft. The final list depends on how many people come and on more thinking about the scenario. We may also play in teams of two per role, which is still to be decided. Each role gets public goals, private goals and one secret.

US President

Wants the US ahead, a second term, and no disaster on their watch.

Leading lab CEO

About six months ahead of everyone else, under pressure from investors and Washington.

Head of Alignment

Knows the most about whether the model is safe. Controls almost nothing.

China

Behind on chips. Can steal, centralize compute or negotiate.

Chinese lab lead

Answers to Beijing, with ambitions of their own.

Rival US lab

In second place. Can lobby, hire away researchers or cut corners.

Congress and the press

Reacts to leaks and public opinion. Can hold hearings and pass laws.

Taiwan and TSMC

Almost every advanced chip is made here.

How a turn works

Each turn is three months of game time and takes about 20 minutes.

  1. Briefing

    News from the last quarter, and sometimes random events we didn't plan for.

  2. Talk

    Negotiate and make deals, openly or in private.

  3. Act

    Write your moves on a card and hand them in secretly.

  4. Roll

    The facilitator states the odds out loud and rolls the dice. The world updates.

At the end, all secrets and hidden moves are revealed, including what the AI actually wanted, and we discuss what happened.

Who wins?

There is no single winner. Each player is scored on whether their actor achieved its own goals by the end of the game. There is also a shared outcome: how the world ends up once superintelligence arrives. It's possible for a player to achieve their goals while the world as a whole does badly, and the reverse.

RSVP

Roles will be assigned and sent out in advance, so everyone has time to read up and think about their role. Please let me know early if you're coming.

If you know someone who would be relevant to invite, talk to me about it first. The aim is for this to be as serious as possible, so please think carefully before suggesting anyone: people who will play their role realistically and engage seriously with the subject.

Reply on WhatsApp