Emergent Negotiation Arena

Emergent Negotiation Arena is an interactive multi-agent AI experiment that explores how communication can emerge from necessity. Three autonomous agents enter a resource-scarce 10×10 grid world with no shared language or predefined vocabulary. To survive, they must negotiate trades. Instead of communicating with ordinary words, the agents invent and exchange symbols. As the simulation progresses, the system tracks how often those symbols are reused, in which trade contexts they appear, and whether multiple agents begin assigning them the same meaning. When a symbol consistently stabilizes across agents for the same context, it becomes an emergent shared “word.” The project supports multiple execution modes. Fireworks enables live LLM-powered agents, while heuristic mode provides a rule-based live simulation and replay mode lets judges inspect a recorded run without consuming API credits. A real-time Gradio dashboard visualizes the world, agent activity, trade outcomes, vocabulary formation, emergent words, and benchmark data. The project is designed as both an AI research experiment and an accessible interactive demo. It demonstrates how coordination, negotiation, and primitive language can emerge without giving agents a shared vocabulary in advance. The application includes a tested multi-agent simulation engine, parallel agent execution, live AI integration, replayable experiment artifacts, and 84/84 passing tests.