A team of developers from DeepMind (a division of Google) taught the AI game in the Quake III Arena in capture mode. With the help of reinforcement training (kind of machine learning), artificial intelligence has the ability to adapt to ever-changing maps. The For The Win algorithm from DeepMind does not have a sample on which it could learn, but learns everything by trial and error, receiving a response from the environment.
The DeepMind laboratory organized a tournament for which attracted 40 real players. They were mixed with AI agents in a random order. As a result, the characters managed by the algorithm got more wins than real people. And they, in turn, noted that the algorithm was more cooperative than other real players.