LessWrong Event Link (LW RSVPs appreciated but please also register on the main website)
EXEC EXEC SUMMARY: Come to Toronto on October 17th and playtest my AI themed wargame. You can LARP as an AI company or a military commander. It’ll be about 5 or 6 hours with breaks. Coffee and lunch included. thecommonsproblem.com
Executive Summary
I am developing an AI Governance Megagame called The Commons Problem, inspired by AI wargames like D. Scott Phoenix’s “The Endgame” and Shahar Avin’s “Intelligence Rising”. In The Commons Problem, 30-60 players gather in a physical space and break into teams representing key entities in the global AI space.
Over a few hours of play, the players and the Game Masters simulate the future of AI governance. Some play sessions might see a cyber security meltdown, while others might feature a novel AI-designed supervirus or a Third World War. The purpose of the game is to learn about and observe the emergent properties of the complex, incentive-driven AI landscape of today. When faced with crisis, will the world come together to stop catastrophe? Or will each faction’s incentives pull them into self-interested behaviour, leading to a tragedy of the commons?
Tickets are on sale now for Playtest 1 on October 17th in Toronto. It’s going to be a lot of chaos, a lot of fun, or both. Go to thecommonsproblem.com to register. More playtests will be taking place in California and Ontario throughout Q4 2026/Q1 2027.
I am generously supported by BlueDot Impact and the Toronto Event Generator. The Commons Problem is currently a not-for-profit enterprise and I am interested in funding for further work. If that interests you or someone you know, please write me at hello@thecommonsproblem.com.
What is The Commons Problem?
In April 2026 while I was on a writer’s residency/vision quest in at Inkhaven 2 in Berkeley, I played a rather interesting “game” called The Endgame hosted by D. Scott Phoenix in which ~40 players broke into teams like “xAI”, “China” and “The Public” to roleplay the near future and simulate AI outcomes. I wrote a whole post about this in April, but in short my response to Phoenix’s “game” was “this would be a cool design project”.
I say “game” in quotation marks because, with no disrespect to Phoenix, The Endgame didn’t exactly play like a real game. A defining criterion of a “game” in my opinion is the presence of meaningful choices, and The Endgame had none of those. There were no resources, no action economy, no real constraints whatsoever. My team, xAI, could just as easily declare “we release a new model” or “we build the new YottaFabricator 69 in Texas, making the US totally fab-independent for frontier chips”. I regard this as a failure of the exercise. Without constraints, it was more of an improv exercise than a real game.
The Commons Problem (TCP) is a real game. A Megagame, to use the hobbyist term. I’ve described it as a hybrid of Model UN + Professional Wargame + LARP + Interactive Workshop. I wanted to take Phoenix’s basic model— sticking 30-60 people in a room and sorting them into teams to roleplay AI governance— but inject rules, mechanics, design, and structure into the mix so that the experience and outcome is more useful as a learning experience and creates more interesting after-action reports, which might be of some value on their own. More on that later.
A sample play area layout for Beta v0.5. (Not final)
The players gather into a large space with tables for each faction to store their faction sheet and other team components. The space needs a centralised stage area for the Game Masters (GMs) to stand and address the crowd, as well as plenty of space to walk around and negotiate with each other. Each team represents a key player in AI, and every member of the team has a specific role to play. One player might be “COO of Huaxin Inc” and another might be “Minister of Defence for the Russian Federation”.
TCP is built around Resources, of which there are several types. Resources are the primary first-order constraint of what you can and cannot do. Compute is needed to train new AI Models and operate existing Models. Influence is needed by Great Powers to create Policy. Fabrication is needed by Chip Firms to manufacture Chips (which can be turned into Compute with the construction of Data Centers). Resources are used to take Actions, which are all resolved simultaneously by the GMs at the end of each Round.
Faction sheets modelling resources and information about 2 of the game’s many Factions. (Not final)
The core of the game’s economy revolves around AI Models, which the AI Labs can both deploy commercially to earn revenue, while also licensing the Models out to other factions. Paying Mankind Labs to licence “Shakespeare 6.6 Mega” or whatever might be useful to Chip Firms trying to AI-assist their designs, Investors trying to perfectly time transactions and get better yields, or Great Powers trying to airstrike targets more effectively.
Ultimately the game is trying to simulate a marketplace, a global economy, and a geopolitical balance of power. There is a live equities market, updating macroeconomic variables, and 4 nuclear-armed Great Powers. But all the while, as the GMs are refereeing the game, they are also keeping an eye on the game state and, based on that game’s specific secret Scenario, they are injecting crisis. If a frontier AI Model passes a certain threshold of power, the GMs might announce at the end of the round that the model was used to hack a country’s infrastructure. If an Investor makes too many transactions with frontier Models, the GMs might announce a catastrophic market collapse due to trade agents gaming the system. Et cetera.
The Commons Problem is “semi-cooperative” to use the hobbyist strategy gaming term. Every Faction has their own set win conditions— Grapefruit Capital might wish to simply 3x their portfolio by the end of the game, the United States might wish to hold certain economic and technological benchmarks— but any number of Factions can win in theory. The catch is that per the game’s Scenario (which only the GMs know at the beginning), there are certain conditions under which a “catastrophic outcome” occurs and everyone loses.[1]
Why am I doing this?
As Tom Scott said of his early internet escapades, “because the alternative was not doing it.”
I think my personal motivation is a combination of a few things:
I’d been looking for a complex long-term project for some time now
I’m particularly interested in the design of games and event management
My increased x-risk anxiety made me want to pivot into AI somehow
There seems to be oodles of American philanthropic money flying around to support AI governance/fieldbuilding/communications projects
Most of the AI-knowledgeable people with whom I’ve discussed TCP seem to think it’s a very cool, high upside project with a good chance of scaling and receiving future funding
I don’t imagine I will get rich off of The Commons Problem or build any kind of career around it. But I’d nonetheless been picking away at TCP in my afternoons and weekends from April-August, from Berkeley to Toronto to Honolulu. I developed a little Alpha and did two five-person playtests over Discord. And then one afternoon on a train ride to Stouffville, ON, I decided to throw my name in for a BlueDot Impact Rapid Grant. 36 hours later, BlueDot Impact awarded me a Rapid Grant to develop a 30-60 person game and conduct two playtests in Toronto.
Received help from AIGS Canada to prototype the Beta of the game in a secret Playtest 0.5, also here in Toronto
Been invited to run the game at a large AI safety conference in San Francisco, CA
It’s harder for me to answer why I am doing this on, like, a spiritual level, dude. I mean certainly I have a theory of impact for The Commons Problem: there are probably a lot of wargame-enjoying, systems-oriented people out there who maybe know a bit about AI safety, and would think more deeply about the issue (as I did) when they’re exposed in a structured way to the problem of conflicting incentives in the AI governance ecosystem. But what is the trajectory of this project? I’m still not sure. I’m hoping this will become clearer as I get more feedback during playtesting.
What next?
My initial plan called for 2 playtests here in Toronto. I am still in early talks for some of them but it’s possible that I will be able to stretch this to 6 playtests in Ontario and the Bay Area by the new year if things fall into place, only slightly over budget. By Christmas I have a good idea of what the future of TCP is.
My problem is that I am currently bootstrapping and my runway is limited. BlueDot Impact’s Rapid Grants program does not include a personal work stipend. In three weeks when Playtest 0.5 and Playtest 1 are concluded (and hopefully successful) I will be applying for a larger grant to continue R&D + subsidised facilitation of the game, including a lean stipend so I can eat for awhile as I work. It feels weird asking large organisations for money, but my current model of testing is so expensive that if I were running Playtest 1 independently the ticket price would be 3x or 4x what it currently is. I am theoretically also open to equity investment rather than a free grant, but I do not know whether this game will ever make a profit and honestly the AI safety grantmaking space is so liquid that a grant seems far easier to acquire and fulfill.
I think the model of me independently organising play sessions of TCP isn’t sustainable long term. It’s just too expensive. I initially proposed three simultaneous pathways for scaling the project to BlueDot Impact:
Make the rules open source so that anyone in the world can download the game and print-and-play it with their group or organisation.
Sell component kits for people who want to play with some better quality, “creator-recommended” components rather than cheap B&W printouts.
Organisations, institutions, and events who are interested in the game but don’t want to set it up themselves could hire my facilitator services to run the game in-person for them, similar to Intelligence Rising and such.
That third pathway is the most speculative, and I’m still not sure who exactly would be interested in this. Universities? Tech conferences? Corporations looking for retreat activities? Nonprofits?
If I were given a grant I would use it to run TCP as much as I can, train and assist other people to become TCP facilitators so they can run the game independently, and develop other versions of the game beyond the 5 hour 30-60 person wargaming extravaganza. Taking some inspiration from the AI 2027 tabletop exercise, I’ve considered:
An asynchronous play-by-mail version run over Discord or something
A smaller, quicker, 10-15 player version more akin to other wargames
A packaged version that you could play like a board game
An online version that works sort of like a video game (increasingly not-that-expensive in this post Fable/Astra world)
A much larger 100-200 person version that would require an entire event/conference to be built around it (most ambitious, most expensive, least probable, but this is the IRL version that makes the most sense to organise independently I think)
Worst case scenario if this project doesn’t pan out, I’ll go back to freelance marketing or join a commune or something. But the sheer amount of positive feedback toward the project that I’ve gotten so far makes me think total failure is unlikely.
For now
I’m going to get back to work. Of you, I ask only this:
If this seems at all interesting to you and you are near Toronto on October 17th, register for The Commons Problem Playtest 1 at thecommonsproblem.com
If this does not interest you, believe me, I get it. But if you know someone who might find this cool, please direct them to this post or the website.
If you happen to know anybody who might want to write a grant for this project, or have the game facilitated for their organisation in the next six months, have them write me at hello [at] thecommonsproblem [dot] com
If you’d like, follow the game on Twitter or the company on LinkedIn. We also have a mailing list on our site.
Phoenix’s The Endgame had a player faction called “The AI” rather than leaving it to the GM’s control. This is a cool idea, but I do not know how to write rules and win conditions for “The AI” factions. No one knows what a theoretical AGI, once it’s achieved, might want. Each of the game’s eventual Scenarios will model a different kind of x-risk contingency— cyber risk, biosecurity, authoritarian disempowerment, AI-assisted terrorism, etc.
LessWrong Event Link (LW RSVPs appreciated but please also register on the main website)
EXEC EXEC SUMMARY: Come to Toronto on October 17th and playtest my AI themed wargame. You can LARP as an AI company or a military commander. It’ll be about 5 or 6 hours with breaks. Coffee and lunch included. thecommonsproblem.com
Executive Summary
I am developing an AI Governance Megagame called The Commons Problem, inspired by AI wargames like D. Scott Phoenix’s “The Endgame” and Shahar Avin’s “Intelligence Rising”. In The Commons Problem, 30-60 players gather in a physical space and break into teams representing key entities in the global AI space.
Over a few hours of play, the players and the Game Masters simulate the future of AI governance. Some play sessions might see a cyber security meltdown, while others might feature a novel AI-designed supervirus or a Third World War. The purpose of the game is to learn about and observe the emergent properties of the complex, incentive-driven AI landscape of today. When faced with crisis, will the world come together to stop catastrophe? Or will each faction’s incentives pull them into self-interested behaviour, leading to a tragedy of the commons?
Tickets are on sale now for Playtest 1 on October 17th in Toronto. It’s going to be a lot of chaos, a lot of fun, or both. Go to thecommonsproblem.com to register. More playtests will be taking place in California and Ontario throughout Q4 2026/Q1 2027.
I am generously supported by BlueDot Impact and the Toronto Event Generator. The Commons Problem is currently a not-for-profit enterprise and I am interested in funding for further work. If that interests you or someone you know, please write me at hello@thecommonsproblem.com.
What is The Commons Problem?
In April 2026 while I was on a writer’s residency/vision quest in at Inkhaven 2 in Berkeley, I played a rather interesting “game” called The Endgame hosted by D. Scott Phoenix in which ~40 players broke into teams like “xAI”, “China” and “The Public” to roleplay the near future and simulate AI outcomes. I wrote a whole post about this in April, but in short my response to Phoenix’s “game” was “this would be a cool design project”.
I say “game” in quotation marks because, with no disrespect to Phoenix, The Endgame didn’t exactly play like a real game. A defining criterion of a “game” in my opinion is the presence of meaningful choices, and The Endgame had none of those. There were no resources, no action economy, no real constraints whatsoever. My team, xAI, could just as easily declare “we release a new model” or “we build the new YottaFabricator 69 in Texas, making the US totally fab-independent for frontier chips”. I regard this as a failure of the exercise. Without constraints, it was more of an improv exercise than a real game.
The Commons Problem (TCP) is a real game. A Megagame, to use the hobbyist term. I’ve described it as a hybrid of Model UN + Professional Wargame + LARP + Interactive Workshop. I wanted to take Phoenix’s basic model— sticking 30-60 people in a room and sorting them into teams to roleplay AI governance— but inject rules, mechanics, design, and structure into the mix so that the experience and outcome is more useful as a learning experience and creates more interesting after-action reports, which might be of some value on their own. More on that later.
A sample play area layout for Beta v0.5. (Not final)
The players gather into a large space with tables for each faction to store their faction sheet and other team components. The space needs a centralised stage area for the Game Masters (GMs) to stand and address the crowd, as well as plenty of space to walk around and negotiate with each other. Each team represents a key player in AI, and every member of the team has a specific role to play. One player might be “COO of Huaxin Inc” and another might be “Minister of Defence for the Russian Federation”.
TCP is built around Resources, of which there are several types. Resources are the primary first-order constraint of what you can and cannot do. Compute is needed to train new AI Models and operate existing Models. Influence is needed by Great Powers to create Policy. Fabrication is needed by Chip Firms to manufacture Chips (which can be turned into Compute with the construction of Data Centers). Resources are used to take Actions, which are all resolved simultaneously by the GMs at the end of each Round.
Faction sheets modelling resources and information about 2 of the game’s many Factions. (Not final)
The core of the game’s economy revolves around AI Models, which the AI Labs can both deploy commercially to earn revenue, while also licensing the Models out to other factions. Paying Mankind Labs to licence “Shakespeare 6.6 Mega” or whatever might be useful to Chip Firms trying to AI-assist their designs, Investors trying to perfectly time transactions and get better yields, or Great Powers trying to airstrike targets more effectively.
Ultimately the game is trying to simulate a marketplace, a global economy, and a geopolitical balance of power. There is a live equities market, updating macroeconomic variables, and 4 nuclear-armed Great Powers. But all the while, as the GMs are refereeing the game, they are also keeping an eye on the game state and, based on that game’s specific secret Scenario, they are injecting crisis. If a frontier AI Model passes a certain threshold of power, the GMs might announce at the end of the round that the model was used to hack a country’s infrastructure. If an Investor makes too many transactions with frontier Models, the GMs might announce a catastrophic market collapse due to trade agents gaming the system. Et cetera.
The Commons Problem is “semi-cooperative” to use the hobbyist strategy gaming term. Every Faction has their own set win conditions— Grapefruit Capital might wish to simply 3x their portfolio by the end of the game, the United States might wish to hold certain economic and technological benchmarks— but any number of Factions can win in theory. The catch is that per the game’s Scenario (which only the GMs know at the beginning), there are certain conditions under which a “catastrophic outcome” occurs and everyone loses.[1]
Why am I doing this?
As Tom Scott said of his early internet escapades, “because the alternative was not doing it.”
I think my personal motivation is a combination of a few things:
I don’t imagine I will get rich off of The Commons Problem or build any kind of career around it. But I’d nonetheless been picking away at TCP in my afternoons and weekends from April-August, from Berkeley to Toronto to Honolulu. I developed a little Alpha and did two five-person playtests over Discord. And then one afternoon on a train ride to Stouffville, ON, I decided to throw my name in for a BlueDot Impact Rapid Grant. 36 hours later, BlueDot Impact awarded me a Rapid Grant to develop a 30-60 person game and conduct two playtests in Toronto.
Since then, I have:
It’s harder for me to answer why I am doing this on, like, a spiritual level, dude. I mean certainly I have a theory of impact for The Commons Problem: there are probably a lot of wargame-enjoying, systems-oriented people out there who maybe know a bit about AI safety, and would think more deeply about the issue (as I did) when they’re exposed in a structured way to the problem of conflicting incentives in the AI governance ecosystem. But what is the trajectory of this project? I’m still not sure. I’m hoping this will become clearer as I get more feedback during playtesting.
What next?
My initial plan called for 2 playtests here in Toronto. I am still in early talks for some of them but it’s possible that I will be able to stretch this to 6 playtests in Ontario and the Bay Area by the new year if things fall into place, only slightly over budget. By Christmas I have a good idea of what the future of TCP is.
You should come to Playtest 1, by the way.
My problem is that I am currently bootstrapping and my runway is limited. BlueDot Impact’s Rapid Grants program does not include a personal work stipend. In three weeks when Playtest 0.5 and Playtest 1 are concluded (and hopefully successful) I will be applying for a larger grant to continue R&D + subsidised facilitation of the game, including a lean stipend so I can eat for awhile as I work. It feels weird asking large organisations for money, but my current model of testing is so expensive that if I were running Playtest 1 independently the ticket price would be 3x or 4x what it currently is. I am theoretically also open to equity investment rather than a free grant, but I do not know whether this game will ever make a profit and honestly the AI safety grantmaking space is so liquid that a grant seems far easier to acquire and fulfill.
I think the model of me independently organising play sessions of TCP isn’t sustainable long term. It’s just too expensive. I initially proposed three simultaneous pathways for scaling the project to BlueDot Impact:
That third pathway is the most speculative, and I’m still not sure who exactly would be interested in this. Universities? Tech conferences? Corporations looking for retreat activities? Nonprofits?
If I were given a grant I would use it to run TCP as much as I can, train and assist other people to become TCP facilitators so they can run the game independently, and develop other versions of the game beyond the 5 hour 30-60 person wargaming extravaganza. Taking some inspiration from the AI 2027 tabletop exercise, I’ve considered:
Worst case scenario if this project doesn’t pan out, I’ll go back to freelance marketing or join a commune or something. But the sheer amount of positive feedback toward the project that I’ve gotten so far makes me think total failure is unlikely.
For now
I’m going to get back to work. Of you, I ask only this:
Phoenix’s The Endgame had a player faction called “The AI” rather than leaving it to the GM’s control. This is a cool idea, but I do not know how to write rules and win conditions for “The AI” factions. No one knows what a theoretical AGI, once it’s achieved, might want. Each of the game’s eventual Scenarios will model a different kind of x-risk contingency— cyber risk, biosecurity, authoritarian disempowerment, AI-assisted terrorism, etc.