An artificial intelligence (AI) agent named CICERO has mastered the online board game of Diplomacy. This is according to a new study by the Meta Fundamental AI Research Diplomacy Team (FAIR) that will be published today (November 22) in the journal Science.
AI has already been successful at playing competitive games like chess and Go which can be learned using only self-play training. However, games like Diplomacy, which require natural language negotiation, cooperation, and competition between multiple players, have been challenging.
The new agent developed by FAIR is not only capable of imitating natural language, but more importantly, it also analyzes some of the goals, beliefs, and intentions of its human partners in the game. It uses that information to figure out a plan of action that accounts for aligned and competing interests, and to communicate that plan in natural language, the researchers say.
Playing anonymously against humans in 40 speed games in an online Diplomacy league, CICERO scored more than double the average score of human players and was in the top 10% of participants who played more than one game. CICERO “passed” as a human player against 82 unique players, and the researchers saw no in-game messages to suggest that the human players believed they were playing with an AI agent.
Reference: “Human-Level Play in the Game of Diplomacy by Combining Language Models with Strategic Reasoning” by Meta Fundamental AI Research Diplomacy Team (FAIR) , Anton Bakhtin, Noam Brown, Emily Dinan, Gabriele Farina, Colin Flaherty, Daniel Fried, Andrew Goff, Jonathan Gray, Hengyuan Hu, Athul Paul Jacob, Mojtaba Komeili, Karthik Konath, Minae Kwon, Adam Lerer, Mike Lewis, Alexander H. Miller, Sasha Mitts, Adithya Renduchintala, Stephen Roller, Dirk Rowe, Weiyan Shi, Joe Spisak, Alexander Wei, David Wu, Hugh Zhang and Markus Zijlstra, 22 November 2022, Science.