Researchers at DeepMind, Google’s artificial intelligence (AI) subsidiary, have made their AI agents able to play and win a multiplayer online game , something impossible to date , and which involves cooperating and competing with other players
According to a study published today in the journal Science , AI has demonstrated an “amazing ability” to develop and use strategies learned independently, without human intervention.
This method had demonstrated its success in first-person games -with a single player-, and two -like chess or Go-, but until now it had not broken the barrier of the multiplayer stage, where elements such as the game in team and the competition against more than one opponent.
The video game with which this feat has been demonstrated is “Quake III Arena Capture the Flag “, a first-person shooter.
The study shows that the Google AI can play at the level of a player of flesh and blood, and beat both other artificial intelligence and human players.
The artificial intelligences involved in the study were not trained or received information about the game or the players, but learned from their own experience playing, the score obtained and what they “saw” in the pixels on the screen.
The generated artificial intelligences learned to play from thousands of games that were developed between them and in randomly generated scenarios.