Photo by mikemacmarketing on commons.wikimedia.org
Meta announced: Artificial intelligence has begun to challenge humanity, this time in the classic game Diplomacy. So, what’s the end result?
Machine learning systems have been struggling with humanity for more than 10 years. Although the field is still in its infancy, the results of the tests are truly fascinating. For testing purposes, the two species, which shared their trump cards in the games, came face to face this time in the game Diplomacy.
First released in 1959, Diplomacy is a more refined version of the board game RISK, where two to seven players take on the role of the European power and their opponents try to win the game by conquering their territory. Unlike RISK, Diplomacy requires players to negotiate with each other—forming alliances or betraying each other—before moving their pieces at the same time in the next game phase. While the skills of reading and manipulating opponents, persuading players to form alliances and plan complex strategies, managing delicate partnerships, and knowing when to switch sides, are all a big part of the game, these are skills that machine learning systems have so far often lacked.
Last week, Meta AI researchers announced that they’d overcome these machine learning shortcomings with CICERO, the first AI to perform at a human level in Diplomacy. The team trained Cicero on 2.7 billion parameters for 50,000 rounds on webdiplomacy.net, an online version of the game, and placed 2nd (out of 19 participants), doubling the average score of their opponents in a 5-game league tournament.