Emergent Communication in Multi-Agent Systems

Two independent agents — a Sender that sees a target object and a Receiver that only hears a symbol — are trained end-to-end with nothing but a shared reward for guessing correctly. This simulator renders the resulting Lewis signaling game in 3D: five distinct concept objects on a shelf, a sender orb that lights up with a chosen symbol, a message that crosses a channel with no built-in meaning, and a receiver orb that points at its best guess. Live readouts track the rolling success rate, the number of training episodes, and the protocol's entropy — watch it fall from near-random noise to a crisp, consistent code as the two agents converge on a shared language purely through interaction, the same underlying mechanism that lets modern LLM-based multi-agent frameworks like AutoGen and CrewAI coordinate through learned or negotiated message conventions rather than a protocol anyone hand-wrote.