AI has gained a lot of learning opportunities by playing 450 thousand games with human players and their own clones. AI does not give a description of how to play the game, but only a strategy to win the game, which usually means that a version of the AI is playing the same game for the same clone.
DeepMind introduces the diversity of game styles by training 30 AI teams, thus providing extra depth for this formula. Training AI in this way requires nearly five hundred thousand games, each lasting five minutes. As in the past, how this concept of simple technology represents the complex behavior of robots is impressive.
DeepMind's AI not only learned the basic rules to capture the flag, but also learned to protect its own flag, and followed the teammates. In order to make AI more difficult to challenge, each game was carried out on a new, program generated map to let AI learn to deal with various challenges.