The Commons Problem - Playtest 1 Report

A high-level overview of the session of The Commons Problem that took place on Saturday.

Things that went well at Playtest 1.

Design issues: things that went badly at Playtest 1 in a good way. Design lessons learned and issues that could not have been found without playtesting.

Execution issues: things that went badly at Playtest 1 in a bad way. Costly mistakes I made in planning and facilitation of the event which were largely silly and avoidable.

AI Labs got to work training up new AI Models, competing on model Strength (an integer representing how strong the Model is. The first model trained, Eve 1.0 by Mankind Labs, had a Strength of 8).

Chip companies got to work designing/printing new Chips to create compute to sell to AI Labs.

All of this R&D and manufacturing was costly, so private and public firms scrambled to Investors in search of funding. VCs threw money at the best-performing Chip Designers and AI Labs.

Russia and Europe entered negotiations when the Ukraine war escalated after Belarusian troops entered. The UN attempted to address the crisis, but their resolution was vetoed.

China sabre-rattled at Taiwan, deploying their navy and threatening forced annexation. The EU was a far stronger voice for peace than the US— as one reporter put it, "Although the US and China are ostensibly at odds, their AI ecosystems are entangled."

International private-public compute cartels

Taiwan's chip plant expansion into Arizona

A proposed US military operation against the European Union due to a perceived lack of cooperation

A total buyout and taking-private of TSMC by a British hedge fund and a UAE sovereign wealth fund

The Commons Problem is, overall, a fun game. 20 out of 24 players surveyed enjoyed or strongly enjoyed the game's negotiation mechanics. The energy was electric in the room from beginning to end. Everyone was hungry for dealmaking and victory. One player described The Commons Problem as "chaotic and exuberant."

People seemed to eventually get a grasp for the game. Almost everyone understood their faction and their objectives, and 15 out of 24 surveyed pretty much understood what they were doing by the end of the game.

Players got the memo about creativity and freedom very early on. Intergovernmental organisations, treaties, global slush funds, emergency meetings, and contracts formed organically with no direction from me.

The experience delivered many high-impact, memorable moments. Between the official survey, published and unpublished post-mortems, and verbal debriefs with my players, it seems like almost everyone had a crazy moment they wanted to share with me. I have been told repeatedly that The Commons Problem, with its forced immersion into the shoes of key AI stakeholders, deeply challenges players to question their worldviews and mental models.

Of the 40 players who registered for Playtest 1, 36 showed up to play and 16 paid to be there. This is an excellent turnout rate for an AI safety event as elaborate as this.

The game was too easy. Many factions reported in Rounds 1-3 that they'd already achieved their objectives without much fuss, and it's ridiculous that all 21 Factions declared victory at the end of the game. That should be a highly improbable outcome.

There is an "infinite money glitch", as Epistech Inc.'s CEO put it to me. Talent, the resource representing smart humans working on AI, chips, and software, has a fixed low price. Talent is used to build Software Products, which produce revenue at variable but linear rate, forever. The more Talent you have, the more Software Products you can ship, the more revenue you can generate arbitrarily. To fix this, Talent needs to become a hot commodity that gets vastly more expensive with demand. One of my GMs suggested that Talent should be non-fungible, with a pool of active researchers/engineers available for hire like baseball cards.

Paper does not scale. There were almost two dozen discrete paper forms and artifacts that existed in the game— military order sheets, IPO filings, research trees, chip design specifications, data centers. I had the naive vision that my GMs would be able to process all these papers, get me the information, and I would enter it into my database. In reality, this would have required a staff twice the size, with half of them only doing paperwork and data entry while the other half answered questions and facilitated play. I am humbled by the early designers of megagames who processed all that information on paper only (though to be fair, I think that my game is far more quantitative information-dense than, say, Watch The Skies). Tactility is important to me, so the in-person Megagame iteration of The Commons Problem will always have some physical components and artifacts, but much of this game needs to be ported onto software and web tools, with everything tracked on some kind of central ledger.

There is an information shortage during play. Though the publishers worked as hard as they could, it was impossible to keep track of everything. The fog of war is an important game mechanic— Factions need to make important decisions under time constraints, resulting in forced guesswork— but a 3 person news team is possibly insufficient. I think the GM team needs to inject more information, publicly and privately. A second Publisher faction might also be interesting.

AI safety research needs to be harder. As we are seeing now in 2026, AI risk does not go away even if you throw billions and billions of dollars at it. My rules for AI safety research were shallow. A global "AI DOOM" index isn't a good metric. Risk and safety research should be tied locally to individual models, and there should perhaps be factions akin to METR or MIRI whose entire job is to benchmark AI models, audit labs, and conduct safety research.

I knew this would be a problem but I still did not design around it: the wargaming rules remain underbaked. They were non-existent in my Playtest 0.5 test run, and I had two weeks to correct for that, but the rules I came up with were shallow and still created a lot of question marks. There were no rules for amphibious invasions, coalition warfare, long-term expeditionary occupation, capturing objectives, or arms sales. Zyad, my GM for Policy and Warfare, did a stupendous job of improvising and facilitating, but more work is needed here. It's funny, for both Playtests 0.5 and 1, I almost felt like I was intentionally restricting myself from writing too many rules for warfare? I am afraid of scope creep and I was stopping myself from applying too much of what I know (combat heavy negotiation/war games like Axis and Allies and Twilight Imperium) onto this new project, for fear that warfare rules would crowd everything out. I overcorrected, and if Zyad wasn't so excellent of a GM, the game would have suffered.

The game has no rules for Mergers and Acquisitions. I didn't realise that there were two mergers in the game: an EU Lab and Tech Firm merged, and America's two largest AI Labs merged. There were no rules for investment banking, antitrust suits, or changing of win conditions.

The game features a fully fledged stock market with many listed firms (both player-controlled and non-player-controlled) and basically no one interacted with it. With the labs and tech firms printing money, it made no sense for investors to tie up capital in public energy and banking stocks that might rise by 20% in the next year when they could spend all their money on new data centers.

There was an information shortage before the game started. While a few players came into the game seemingly fully prepared, my method for disseminating information wasn't perfect. I tried to give everyone siloed documents so as not to overwhelm them: one slide deck for "Finance and Markets", one brief overview for "Your Faction's Objectives", but I should have spent my energy on creating a single rulebook that everyone can reference.

I am far too averse to spending money, and it very nearly caused me some problems. I cheaped out on the printer and projector, providing my own, and both failed on game day. The newspaper editor managed to get the printer working, but the projector never got running, which seriously impacted the game. I should have rented gear from the venue. Time is money!

I did not give myself enough setup time. I didn't have a coherent plan for setup, and I didn't even have a single component checklist. Players had to sketch out their own AI model order forms on scrap paper, and we had to improvise and create new denominations for resources (i.e. "one pink chip is worth 20 Talent now!").

We ran out of numerous components— people had to use stock exchange order forms to sketch out private investment deals, and there was more demand for data centers than we had data center sheets available.

People were thirsty and though I provided coffee in the morning, I provided no bottled water or soft drinks. Stupid mistake.

Playtest 2 is scheduled for 11 November 2026 in Toronto, ON as part of AI Safety Week.

Playtest 3 is scheduled for 21 November 2026 in San Francisco, CA, hosted at Mox SF.

Playtest 4 is targeted for 5 December 2026 in Berkeley, CA. I am still looking for a venue and partner organisations for this final test.

A free, open-source (probably Creative Commons licensed) ruleset for The Commons Problem on my website which can be downloaded by anyone. This will include all the game's soft components: a master rulebook, GM's guide, player briefs, paper components, software, a recommended component list, and a basic play scenario for the GM.

A hard component play kit available for sale (probably at cost for most buyers) containing the GM-recommended hard components. This is optional; you can absolutely play The Commons Problem print-and-play style, or by sourcing your own poker chips and dice, but the play kit will make it a little easier for organisations.

A framework for organisations, institutions, events, and individuals to hire me to facilitate The Commons Problem. I want this to be inexpensive for nonprofits but obviously in-person event coordination has hard costs. Certainly any nonprofit or corporation could download my rules and put the game on themselves, maybe even at a profit, but time is money and it might be easier and cheaper for them to just contract Automaton Games to do it for them.