Abstract When Robert Axelrod organized his 1979 tournament of repeated Prisoner’s Dilemma strategies, few anticipated that the winning approach would be so simple or “nice”: Tit-for-Tat, a strategy built on cooperation rather than aggression, outperformed far more sophisticated rivals, a result that reshaped thinking across economics, political science, and evolutionary biology. By creating a common arena in which diverse strategies could compete against each other, Axelrod’s competition also showed that bottom-up discovery through structured comparison could surface findings that deductive theory alone may not anticipate. Four decades later, artificial intelligence (AI) has transformed what this method can do by expanding the scale, complexity, and realism with which negotiation strategies can be systematically compared. The Program on Negotiation (PON) AI Negotiation Summit showcased three such competitions—the Massachusetts Institute of Technology (MIT) AI Negotiation Competition, the Melting Pot and Concordia Contests, and the Automated Negotiating Agent Competition (ANAC)—each designed around distinct assumptions about environment structure, agent architecture, and evaluation criteria. Taken together, they demonstrate that negotiation competitions are powerful tools for generating discovery at scale, benchmarking strategies across multiple criteria, and advancing the science of negotiation.
Curhan et al. (2026) studied this question.