Deepmind KI learns and plays Quake 3



[ad_1]

Cooperation has brought us far ahead over the past thousand years. We learned that in a group, you can better survive and protect yourself than yourself. And the principle of cooperation accompanies us in our daily lives – we must constantly gather together for collective intelligence. Teaching this artificial intelligence behavior is not so easy.

Google's sister company, Deepmind, has been dealing with artificial intelligence and potential challenges for years. They often use computer games and so here too. The developers put an AI before the classic game Quake III Arena and she taught the game. In a short time, the program has surpassed all human players in terms of skills and has become virtually invincible.

Quake III Arena laid the foundation for many first-person video games and created a long-time electronic sports scene. The rules are simple, but the dynamics are a bit more complex. Two teams are fighting on a card in order to capture the flag of the opposing team while protecting theirs. There are some tactical tricks and the team with the most flags after five minutes wins the game.

So it's about cooperation, competition and adapting to different play styles The AI ​​takes the game as real humans, so it comes with the same game screen and as the card changes every time, the program must constantly reevaluate it. He is forced to devise new strategies and recognize the right plan of action.

The AI ​​was formed by many small artificial intelligences, who participated and learned in the game. The whole concept of training recalls the Pac-Man history of the subsidiary of Microsoft Maluuba. The procedure was similar. Over time, the "memory" and visual attention of artificial intelligence are trained, making them invincible in the end. Not to mention the quick response time.

More importantly, AI actually learns behaviors similar to those of humans, such as teammate tracking and the presence of opponents. All this is made possible through cooperation and only one, that an AI understands this principle and knows how to apply it, is already quite impressive.

via: deepmind

[ad_2]
Source link