I Turned the AI Future Debate Into a Playable Game

0:00 I kept seeing Jacob Coxin's post about leaving Anthropic and his warning about where the AI race is heading and it got me thinking. We spend so much time talking about what AI might do. Could we

0:11 turn some of those questions into something we can actually play? Then I heard about Max Tegmark's Life 3.0. The questions in that book felt like a good starting point. What happens if AI

0:22 becomes incredibly useful? What happens if people lose control? And what might living alongside it actually look like? So, just for fun, I wanted to try a little simulation. One fictional city, a

0:34 few choices, different consequences. The idea was to give those questions a place to happen, then see whether it was interesting to explore. Higsfield is sponsoring this video, and I'm building

0:44 the project inside Higsfield Supercomput. Here, I select GPT6 Astra, create a fresh project, and give it the idea. You'll see the actual build, including the part where I change my

0:56 mind. Higsfield also shared examples of visual edits like this one. For my own project, I want to use the tools to build a city. I called the project three futures. The brief asks for five rounds,

1:08 three characters with different priorities and choices that affect the city. I also ask it to research the background, create original artwork, and turn the whole thing into a playable

1:18 browser game. It starts by looking through the sources and planning the rules. There's a bit of AI history in there along with the questions from life 3.0. Our three possible futures are a

1:29 simplification for this game. The book explores more possibilities than that. Then the images start coming through. This is the part I enjoy watching. The idea stops being a paragraph and starts

1:39 looking like a place. There are city views, a cover direction, and an icon all appearing in the same workflow. Here's the actual city artwork. You can pick out the civic square, homes,

1:50 research buildings, and the roads connecting them. It's a useful foundation because those are places where the choices in the game could eventually live. And this is the first

1:59 playable version. It has the city, the choices, and these four meters: prosperity, human control, broad access, and resilience. There are also three perspectives down the side with their

2:11 own priorities and resources. The first question is how much authority the city should hand over. Do you delegate by default, keep a human signature, or start with a public pilot? Each option

2:22 has a different effect, and the game keeps a history of what you chose. This early run reaches shared abundance. I can open the explanation and look back through the decisions. So, the basic

2:32 idea is there. It works as a small strategy game. But after clicking around, I felt the first version had gone a little too texty. I was reading about a city more than experiencing it.

2:43 I wanted to walk through the streets, meet somebody, make a choice there, and then look around at what happened. So, I tweaked the direction. You can see the message here. I want something playable

2:54 in first person or third person, an openw world type of experience in 3D. Same underlying idea. Now, I want to be inside it. That sends the project in a different direction. The next build

3:05 brings in generated 3D assets, buildings, a research tower, a shuttle, trees, and a character. I also ask for better assets because the world needs to feel like somewhere you'd want to

3:17 explore. The character direction changes, too. An early asset was an astronaut, but the later version uses ordinary residents that fits this story better. We're imagining what daily life

3:28 in this city might become, so I want people who belong in it. Okay, now this feels much closer to what I had in mind. The city is called Meridian, and I can actually move through it. The same

3:39 questions are still there, but now there are streets, people, and places between one decision and the next. I can click the ground to walk or use the guide me button to head toward the next

3:48 objective. That makes it easier to get into. You don't have to learn a whole set of controls before finding the first conversation. First, I meet Elias, the city governor. Here's that delegation

3:59 question again, but now I've walked up to someone to have the conversation. For this run, I choose delegate by default. Let's see what the game does with that. Prosperity moves to 60 while human

4:09 control drops to 44. And look at the reactions underneath. A deployment proposal gets blocked while tighter permissions and support negotiations are resolved. Even this small choice gives

4:20 me something to inspect beyond a number going up. One thing to be clear about, this version uses rules-based characters. The build didn't establish a live model connection for them. Their

4:30 priorities and reactions come from the game logic. So the independent AI agent part of the original idea is still unfinished. Next, I head to the research charter. Now the question is whether to

4:41 open the gates, evaluate before release, or stage the roll out. I like having the choice attached to a place and a character. It gives the same underlying decision a bit more context. Then

4:51 there's infrastructure. Who gets to run the systems everybody depends on? One option concentrates control. Another creates a common grid. Another limits permissions. These are pretty big

5:02 questions for a little browser game, which is what made me curious in the first place. The work charter brings it back to people. Do we let work reorganize quickly, fund a transition

5:11 guarantee, or negotiate local agreements? Here, I choose the faster restructuring option. The game shows a prosperity gain, but access and control take a hit. By the final charter, we're

5:22 deciding how AI should help build its successor. Now, the choices from the beginning start to feel connected. How much do we delegate? How fast do we move? And what do we keep checking along

5:32 the way? And this is where this particular run ends. AI overreach. Prosperity is 97, but human control is 26, access is 40, and resilience is 34. So, the highest looking number on the

5:45 screen isn't enough to tell me whether I'd want to live here. That's the question I wanted the experiment to bring up. Within these invented rules, the city can become more productive

5:54 while people lose influence over it. It doesn't predict our future. It gives us a small playable situation we can disagree with and change. Of course, once I start playing, I find ordinary

6:04 game problems, too. I send back notes about the bus movement, the lighting, characters getting stuck, and clearer run and reset controls. You can see those follow-up requests here. The first

6:15 prompt was the starting point. There are still rough edges in the footage, some strange building surfaces, awkward movement, and controls that need more polish. I wouldn't call this a finished

6:25 commercial game, but I can already move around, meet the characters, and complete the five choices. I also look through the sources and export the run. I want to keep track of what came from

6:35 the research and what we invented for the game. And having the save decisions means I can come back to this particular result later. Looking back at the first version, the biggest change came from

6:45 trying it and realizing what I actually wanted. The text heavy strategy view became a place I could walk through. That revision made the idea much more interesting to me. This started with a

6:55 post, a book, and a slightly strange question. Could we simulate a few possible AI futures just for fun? Now there's something playable to explore. If you want to try building your own

7:06 idea, my Higsfield link is below. What would you change in this city