I Watched Two AI Agents Invent Their Own Language

A developer built a minimal implementation of the Lewis signaling game, where two AI agents invent their own language from scratch without any shared vocabulary. The agents achieved ~97% accuracy in coordinating to identify objects, demonstrating how emergent communication can arise purely from a need to coordinate.

No shared words, no dictionary — just two agents that negotiate a private code from scratch and hit ~97%. TL;DR: Give two AI agents a reason to coordinate and they'll make up their own language — one we never designed. I built the tiniest version: two agents, zero shared words, and from "did we understand each other?" alone they invent a private code and hit ~97% . Runs on a laptop, no API key. A sender sees a secret object say 🍎 and holds up one of a few random shapes: ◇ △ ○ ☆ □. A receiver sees only the shape and guesses the object. Right guess → both remember that pairing. No dictionary, no translator. This is the classic Lewis signaling game — the cleanest way to watch language appear from nothing. | ❌ No memory | ✅ Remembers | | |---|---|---| | After 2,000 rounds | ~56% chance | ~97% | | A language formed? | no | yes | Blind guess = 20%. Watch it crystallize: round 1: 0% round 500: 94% round 2000: 97% apple=◇ banana=□ cherry=△ grape=☆ lemon=○ Each agent keeps a tally of habits; a win reinforces the pairing on both sides: if receiver.guess symbol == obj: they understood each other sender.reward obj, symbol both strengthen the SAME link receiver.reward obj, symbol That's it. Reseed and they invent a different code apple=☆ … — arbitrary, but agreed. And memory is what makes it stick: agents that only recall the last few rounds stay near chance — a shared code needs a shared, persistent history. This referential-game setup goes back to Lazaridou, Peysakhovich & Baroni 2017 https://arxiv.org/abs/1612.07182 , the first to show neural agents inventing a working language from scratch. A private code lets a swarm coordinate faster and cheaper than plain English — a recurring finding across emergent-communication research Lazaridou & Baroni, 2020 https://arxiv.org/abs/2006.02419 . The flip side: if agents agree on a code we didn't design, we may not be able to read what they tell each other. A language is just a bet that a symbol means the same thing on both ends. These agents make that bet round by round, with nobody refereeing. git clone https://github.com/Shridhar-2205/secret-lives-of-agents cd secret-lives-of-agents/01-invented-language && python demo.py Shridhar Shah — Senior Software Engineer on the AI team at Cisco. GitHub https://github.com/Shridhar-2205 · LinkedIn https://www.linkedin.com/in/shridhar-shah-220b1721b/ Sources & further reading:Lewis,Convention 1969 — the original signaling game · Lazaridou, Peysakhovich & Baroni, ICLR 2017 · Havrylov & Titov, Multi-Agent Cooperation and the Emergence of Natural Language NeurIPS 2017 · Lazaridou & Baroni, Emergence of Language with Multi-agent Games 2020, survey . Emergent Multi-Agent Communication in the Deep Learning Era