After chess and go, the machine beats the human in yet another board game. Meta Platforms AI beat human players in Diplomacy, a game that requires not only strategic planning but also verbal negotiations with opponents.
The system, named Cicero, is presented in the leading journal Science (https://www.science.org/doi/10.1126/science.ade9097) by researchers from the Meta Fundamental AI Research (FAIR) department and leading American universities such as MIT, Harvard and Stanford.
Unlike chess or other strategy games, Diplomacy requires negotiation, cooperation, multi-level competition with more than one player at the same time, and bluffing, which is considered more difficult for machines.
Artificial intelligence is based on two subsystems, one for natural language dialogue and one for strategic reasoning. Thanks to the new learning algorithms it is based on, it can analyze the goals, beliefs and intentions of the seven teammates based on their movements on the game map.
It then uses this evidence to draw up a strategic plan of action that takes into account the diverse and often conflicting interests to propose coordination based on mutual benefit.
Playing anonymously against humans in 40 online matches between August and October, Cicero averaged more than twice the score (25.8%) as humans (12.4%).
The amazing thing was that none of his 82 opponents realized that he was playing against a computer and not a human.
