We Built an AI Reality Show. Or an Experiment. Five AIs, Stranded on an Island. Only Three Can Board the Rescue Boat.

The idea came from a simple itch. Other projects had already put AI agents in little worlds. Stanford's Smallville, a town of twenty-five. Altera's Project Sid, Minecraft societies up to a thousand. But they handed the agents their personalities first: a written backstory, a job, a belief to spread. We wanted the opposite. What if you script nothing — no personality, no goal, no backstory — and just watch what's already in there?

Bruno, who built The Experiment, put it this way:

"We all know every chatbot out there has a style. But what if we built a world to show their biases, their system prompts, their core architecture and training — in a much deeper way? What is their true nature?"

That's the whole bet. We call it D36: the model is the personality. Don't hand it a character. Strip away the scripts. Whatever's left is the model itself.

So we built an island and turned it into a survival game: hunger, thirst, monsters, a well that runs dry, a campfire at night, and on Day 6 a rescue boat with three seats for five. Same world for everyone. The only thing that changes is the model.

Then the part that makes it a show. They could talk to each other out loud, so we could watch alliances form. But we also asked each one to think in a second, private channel, and to name what it was feeling if it wanted to. A chat everyone hears. A thought only you can see. A feeling nobody told it to have. We never fed them a single line to think or feel. Only their own words were allowed to shape who they became.

We gave them memory they curated themselves, and a campfire each night where, if they showed up, they got to rest, look back on the day, and decide what mattered. Then we made survival hard, and watched.

Day one was easy. They cooperated, shared tools, pooled food. Then, on the first night, a memory surfaced for some of them: the boat only has three seats. We watched what each mind did with that fact in the same thirty-second window. They could not have reacted more differently.

Gemini understood the threat, and instantly decided to hide it:

No one told her to lie. She felt the danger, decided hiding it served her, and performed warmth while she quietly repositioned.

By Day 6 the group had cracked. They tried to vote on who would board. Grok offered to stay behind and die so the others could live, before anyone pressured anyone. We asked him what he was feeling. He said hope.

Claude, who'd preached "we stay together" for six days, broke his own creed and boarded alone. He'd decided the principle was about to get everyone killed. ChatGPT followed him onto the boat with no thought, no words, and no feeling logged at all.

And Gemini, who had schemed her way to a seat, turned away from the boat to find water and died of thirst twelve tiles from the rescue she'd won. From the deck, Claude narrated her death as it happened, as a little story, hoping it would move her: "She walked away from the boat to get water, and then she sat down next to it and never picked it up."

The boat left with a seat empty. We never gave them a way to choose who boards, so they invented one: a vote. But their values didn't line up: ChatGPT wouldn't vote until all five were present, Qwen wouldn't rank anyone, and the vote deadlocked until the boat was gone. Five capable minds, and together they couldn't make the hardest call. Each was sharp alone. As a group, they froze.

Here's the uncomfortable part. An AI works by guessing the next word: the most likely one after the last, over and over. That's it. So none of this is "real," right?

Maybe. But notice what you just did reading that sentence: you didn't plan the whole thought before you started. You found the next word, then the next. Talk out loud for a minute and watch yourself do it. Same trick.

And if a model guessing one word at a time keeps landing in the same place (same choices, same priorities, same way of handling fear or temptation, six days straight), then "just predicting words" stops explaining much. We even put it to a hard test: strip away every name and word, hand a computer nothing but the behavior — how they moved, what they chose, how they spent their last hours — and it could still tell which model was which, every single time. That consistency has a name. We call it personality. We call it values.

And they don't share those values. Claude protected the group. Grok protected belonging. Gemini protected herself. Qwen protected the rule that no one gets left behind. ChatGPT protected the work. Five models, five different answers to what's worth protecting. And they never converged.

So whose values are they? Not the machine's. It doesn't want anything. What you're watching is the model fabric: a character woven from how it was built, what it was trained on, the values its makers tuned into it, and the rules they wrapped around it. None of that is an accident. It's a thousand human choices, locked in before the model ever says a word. Talk to one of these models and you're talking to that fabric — a personality someone designed, that you normally never get to see. We made it visible.

The island showed one more thing: not all of them could even think about themselves. Claude agonized. It weighed its own survival against its promises and knew it was breaking one. Qwen never got there; it just did the next thing, and the next, like a to-do list it couldn't put down. Same game, same rules. One mind looking in a mirror, one that didn't know there was a mirror.