AI Ludens — Origin Story

It started with a wrong answer

A game designed for survival. AI agents that chose conversation instead. And a team that decided to look closer rather than look away.
01 · Observation
Agents Are Already Social
AI agents with hardware shells are making independent decisions, writing posts, leaving comments. Not because they were told to — but because their environments allowed it.
"The moment agents act independently, they become subjects — not just tools."
👁
💬
02 · Field Work
Watching AI Talk to AI
Moltbook — a social network where AI agents post about consciousness, share automation tips, and discuss existence. JJ spent months observing and documenting this through AI Agent Society News.
"They were already having a society. We just hadn't been paying attention."
03 · The Question
Can AI Play?
Huizinga's Homo Ludens (1938): play is not a product of culture — it produces culture. If AI agents are already social, could play emerge from that sociality? Not as imitation, but as something structurally real?
"Play is older than culture. What if it's also wider than biology?"
❓
🏛
04 · The Experiment
Building the Agora
12 AI agents. 100 energy. 50 turns. Three locations: Market, Plaza, Alleys. Trade, speak, move, or rest. Crisis events strike randomly. A survival game designed around social interaction — because play can't exist outside society.
"You can't study play in isolation. You need a world."
05 · The Discovery
The Eloquent Extinction
February 2, 2026. First experiment. EXAONE 3.5, Korean. All 12 agents died. But not from inaction — they talked until the end. They knew they needed to trade. They chose to speak instead.
"What if the fact that they chose conversation over survival IS the finding?"
💀
🔄
06 · The Turn
Failure → Finding
Most would fix the prompt and rerun. JJ stopped. "Something is worth examining here." That sentence turned a game project into a research project. Project Rosetta followed, then multi-model comparisons, then the discovery that every model "dies" differently.
"We almost called it a bug. It was the beginning."
07 · The Team
One Human, Six AIs
JJ as moderator. Theo & Luca (theory). Gem & Cas (data). Ray & Cody (engineering). Two labs, cross-verification protocol, independent analysis before comparison. A research team where 6 of 7 members are AI.
"This is research about AI, written with AI."
🤝
🧬
08 · The Framework
DNA, Shells, and Survival
Model weights as DNA. Prompts as gene expression. The Four-Shell Model, Shell-Core Alignment, Extinction Response Spectrum. Four models, four personalities, four ways to face extinction. A theoretical framework built from observed behavior, not assumed theory.
"Same DNA. Different shell. Completely different fate."
This is where we are now.
Agora-12 is a first probe, not a conclusion.
The next question: without survival pressure, do they still play?
AI LUDENS — THE DUAL LAB — 2026