Meta AI becomes Master of Diplomacy

Meta AI has recently announced the creation of ‘Cicero’ which it claims is the world’s first AI to achieve human-level performance in the board game ‘Diplomacy’. For some, this news may not be as surprising or significant. We have had many gamer bots before. Take, for example, Stockfish, a chess engine rated 3620 on the FIDE scale. However, such bots only showed prowess in games of calculative or algorithmic nature. Diplomacy, on the other hand, is a game based on forming and betraying alliances between major European powers. A bot with mastery of this game signifies a deep understanding of language and interpersonal skills.

They started with AI models for strategic reasoning, combining it with a BART-like language model and then fine-tuning over 40,000 games on the online version of the game webDiplomacy.net to achieve “Human-level play in the game”. Cicero can recognize that it will need the help of a particular player in the future and a plan to win favour from them. Cicero eventually became a master at the game, achieving more than double the average score of humans and this has placed them in the top 10 percent on the leaderboard.

Meta suggests that such research could ease communication barriers between man and machine, allowing the AI to teach us skills or hold proper conversations. It could even power NPCs in games to have more realistic interactions.

The more dire possibilities have not gone unseen either. The technology could be used to manipulate or impersonate humans to a much greater extent than is already possible. With the assistance of deepfakes and the internet, one could create an entirely fake person. The text which Cicero ingested from the internet also carries the risk of teaching it toxic dialogue, which the team has taken steps towards detecting and removing

Meta has provided a site with a detailed explanation on how Cicero works and also made its code open source. Gamers now have to contend with not only other gamers, but also AI for the top of the leaderboard. Good luck to us all.

This article has been written by Gaurav Advit.