Speaker: Ignacio Ojea Quintana (MCMP, Ludwig-Maximilians-Universität München)
Date and Time: Thursday, February 12th 2026, 16:30-18:00
Venue: ILLC Seminar Room F1.15, Science Park 107 and online.
Title: Signaling Games with Distributed Rewards
Abstract. Lewis–Skyrms signaling games model the emergence of communication between a sender and a receiver through repeated interaction in a single cooperative game with aligned pay-offs. In this work, I generalize the framework to distributed systems, where agents adopt both sender and receiver roles within a network and engage in independent games with independent rewards. Despite the absence of explicitly cooperative incentives, agents can still learn to communicate meaningfully through reinforcement learning. Cognitively shallow agents can learn to have (simile) conversations even when they do not have a clear shared goal. I present simulation results for Roth–Erev learning, Q-learning, and Temporal-Difference learning. In this setting, Grice’s Cooperative Principle is not encoded in the payoff functions but in the network of signal channels. This points toward a broader lesson: that meaningful communication can be learned even under minimal and distributed assumptions.