Artificial intelligence has long since conquered the realm of classic board games like chess, checkers, and go. But there’s always been a segment of conversational games that seemed out of reach for AI. At least, until now. Meta AI has just revealed Cicero, an AI that can beat humans at the game of Diplomacy. For those who haven’t played, Diplomacy is a board game that was first invented in 1959. The game is set in Europe during World War 1, with the players each controlling a major European power. The objective is to capture and control strategic cities and provinces, which allow that player to make more military units to eventually take over all of Europe. What sets Diplomacy apart from other games is that it’s a conversational game at its core. Each round has a negotiation phase, where players converse with everyone else to try and figure out what to do during the next battle. Those conversations will inevitably result in alliances with weaker players teaming up against a stronger rival, or they’ll result in spectacular betrayals that could see players eliminated from the game. There is some strategy involved, but your ability to negotiate is what determines your success in Diplomacy. Cicero combines the strategic thinking made possible by the AI that conquered games like chess and language-processing AI like BlenderBot and LaMDA. Meta AI then programmed Cicero with a 2.7 billion parameter language model and trained it over 40,000 rounds of webDiplomacy.net, a free-to-play web version of Diplomacy. Cicero’s language was often getting human players to prefer working with it “over other human participants.” But don’t think that Cicero is just a smooth-talking robot. According to three-time Diplomacy world champ Andrew Goff, Cicero is “ruthless in executing its strategy.” (https://www.thegamer.com/meta-cicero-ai-diplomacy/) The result was an AI that came in second place out of 19 participants in a five-game league tournament with double the average score of its opponents.