DeepMind of Google teaches AI to work in teams by playing Quake III Arena

Google DeepMind recently published the results of its research and experiment in which many AIs are taught how to play chess in Quake III Arena, a first-person multiplayer shooter game.The AI ​​taught in this way now plays better than humans, whether it's playing in groups with people or with other AI.

The AI ​​named For the Win (FTW) played nearly 450,000 copies of the Quake III Arena to understand how to effectively cooperate with colleagues, whether it be a person or an AI.

"We teach each AI individually how to play with other members of the group, be it human or artificial intelligence," the company said in its blog post.From a multi-stakeholder perspective, the robbery game requires players to cooperate with teammates and compete with opponents while dealing with any type of play. "

Game is the way researchers are using to teach AI how to cooperate

DeepMind is known as the intelligence that created AlphaGo, the AI ​​system defeated the world's best Go player in 5/2017.AlphaGo Zero, AlphaGo's successor, then trained himself by playing with himself.

In the round with 40 human players, the only groups that consisted of AI fought against a human-only team and had a 95% chance of winning compared to groups with both humans and AI.Interestingly, AI is more cooperative than human.

DeepMind's research is the latest effort of researchers who use the game to train AI tactics, memory ability . which are not usually on computers.Last month, OpenAI said it also used Dota 2 to teach AI.

What comes from a multi-object environment can be used to teach AI when interacting with people, teaching AI how to work together.

See more:

1. Google AI can create 3D images from "flat" 2D images.
2. Try out with 3 difficult Go games like AlphaGo
3. What is "artificial intelligence" AlphaGo is what makes people admire?