This is another skill necessary to live in a normal society that brings Google DeepMind closer to becoming a full member of our society.
All right, jokes aside. The latest achievement of Google's artificial intelligence is indeed impressive. It is true that DeepMind has already won with the Go master and considerable successes in Starcraft 2 competitions, but Quake III is - according to the creators of DeepMind - a completely different pair of wellies.
Google DeepMind versus Quake III Arena
Google's SI learned to play Quake in Capture the Flag mode, which is played in teams, and the whole game is about stealing the flag from the opponent's (heavily defended) base and delivering it to your base.
Apart from the dexterity aspects, the main difference between Q3 and Go or Starcraft is the team game. In both Go and Starcraft, DeepMind learned 1-on-1, while ctf in Quake is a classic team game in which - I know that for some players it will sound like heresy - without team cooperation you won't get too a lot of succsesses.
Cooperation between avatars controlled by artificial intelligence is a completely new challenge for DeepMind, but after several hundred thousand Google SI matches played, it began to cope quite well with this new style of play. Enough to face the human team after 12 hours of playing with yourself (of course at a fast acceleration).
Artificial intelligence quickly learned to play better than people.
Result? People were able to beat the AI in only 25 percent. matches played. Interestingly, human players turned out to be much better marksmen - DeepMind still has problems aiming at greater distances. On the other hand, the AI did much better when it came to navigating the map and planning the best routes to deliver the flag.
Mixing teams also turned out to be a very interesting experience. It turned out that teams composed of man and algorithm won by 5 percent. more games than fully human teams or only controlled by DeepMind. The creators of artificial intelligence Google say that such a fruitful cooperation results from the fact that their algorithm can adapt their playing style to a man from the same team. This discovery may suggest that the AI will be just as good at helping us with slightly more adult tasks than running behind the flag in Quake.
Don't miss new texts. Follow Spider's Web in Google News .
Google's artificial intelligence has mastered the game of Quake III Arena
Comments
Post a Comment