Submission by Conductor Labs

There's a really big potential problem in the coming future. The end of the world. And if that happens, how would we solve the solutions human need to survive from scratch? WildMind is a living, self-teaching persistent AI simulation where 30 independent AI agents — each a separate language model — are dropped into a procedurally-generated world with nothing. No language. No culture. No shared memory. No knowledge of survival, medicine, agriculture, fire, or each other. We use ElevenLabs to capture the language development. Agents develop proto-words — invented sound-sequences that acquire shared meaning through repeated co-occurrence with context. The sound "kroh-tuh" said near fire enough times, by enough agents, in enough interactions, becomes the tribal word for fire. This emergence follows real mathematical laws: Zipf's Law (word frequency distributions), Heaps' Law (vocabulary growth curves), and small-world network topology — the same patterns that govern every human language on Earth, appearing without instruction. You can listen along as the develop their language thanks to ElevenLabs, on a massive information dashboard where you can track every tick from every world hosted by Replit. The end goal is to publish **A Guide to Restarting Human Civilization** — a living, multimedia document synthesized from simulation data and made available to researchers, institutions, and anyone studying civilizational resilience.