Meta AI introduces CICERO, an AI agent capable of negotiating and cooperating with humans
Meta recently introduced CICERO, an AI that achieves human-level performance in the online version of Diplomacy, a strategy game in which it communicated and negotiated in natural language with human players without them knowing that they were dealing with a bot. Cicero ranked in the top 10% of players who played more than one game and even won an eight-game tournament involving 21 players.
AIs like DeepMind’s AlphaZero, based on reinforcement learning, have demonstrated skills matching or surpassing those of humans for games of chess, shogi or go. Cicero integrates a natural language processing model with planning and reinforcement learning algorithms and is the first AI system to achieve human-level performance in the strategy game Diplomacy.
In this game, seven great powers in 1901: France, England, Germany, Austria-Hungary, Russia, Turkey, Italy, each seek to dominate Europe. Seven players, each controlling a country, compete to possess at least half of the supply centers on the game board. To succeed in this challenge, the players must cooperate, negotiate, forge alliances. Each round is preceded by two-by-two negotiations, the players form alliances there, league against their adversaries. When playing and moving their pieces, it is up to them whether or not they keep their commitments.
At each stage of the game, Cicero models the likely moves of other players based on the current state of the board and previous conversations.
For an AI to play this game, it is not enough for it to use natural language, it must communicate persuasively. According to an article in Science, Ciceron was trained on a dataset of 125,261 games played online on webDiplomacy.net. Of these, 40,408 games contained dialogue, with a total of 12,901,662 messages exchanged between players.
Cicero was entered, without revealing that it was a bot, in 40 games taking place between August 19 and October 13, 2022: he doubled the average score of players. During these 72 hours of play involving the sending of 5,277 messages, the other players did not suspect that he was not human. According to Meta AI, they even prioritized trading with CICERO over other human players.
While he performed well, Cicero still has room for improvement, he sent messages containing errors or contradicting his own plans. His dialogue only focused on player actions on the current turn, he didn’t model how his dialogue might affect his relationship with other players longer term in the game.
Meta AI states in its blog:
“CICERO marks the start of a new era for AI that can collaborate with people in-game using strategic reasoning and natural language processing, and learnings from technology like this could one day lead to smart assistants that can collaborate with people. »
#Meta #introduces #CICERO #agent #capable #negotiating #cooperating #humans