The hottest 101deepmind AI in StarCraft

  • Detail

10:1! Deepmind AI wins the man-machine battle of StarCraft II against professional players

[techweb] on January 25, according to foreign media reports, after defeating Ke Jie, lishishi and other go masters in the man-machine go war, deepmind, an AI company under Google, turned the man-machine war to computer games. In the man-machine war of starcraft II, Google deepmind's artificial intelligence program defeated human professionals 10:1

the artificial intelligence program developed by Google deepmind for man-machine combat is called alpha. It challenges two professional players, TLO and mana, who start the beam motor of StarCraft II. The game was held in December last year. Blizzard, the developer of StarCraft II, and Google deepmind have announced the video of the game at that time

in the battle between alphastar, TLO and mana last year, a slightly outdated version of StarCraft II was used. This version is designed to facilitate the relevant research of artificial intelligence programs

tlo and mana played five games with alphasar respectively, but TLO and mana did not occupy an advantage in the game. The final result was alphasar's complete victory, and the showdown result was set at the awkward 10:0

in the published competition video, Google deepmind launched a new version of alpha star to compete with mana. The newly launched alpha star limited the vision. Mana was able to discover some of the shortcomings of alpha star and finally overcome it. It saved a little face for human professional players with a one game victory. Alpha star also suffered its first defeat against professional players

in the video of the battle with alphastar, TLO once said that he was confident of defeating alphastar, but he never thought that the latter would win all five sets, and each time he adopted a completely different strategy

After TLO, mana changed the hydraulic oil and tried his best in the competition, but he did not get rid of the same fate as TLO. He also failed in the five games with alphastar

foreign media reported that AI programs have more advantages in computer games. Professional players, including TLO and mana, have theoretical restrictions on the number of mouse clicks per minute when playing computer games, while AI programs have no such restrictions

secondly, foreign media reported that the reaction time of AI programs is about 350 milliseconds. Alpha star is slower than most professional players in terms of reaction speed. However, after a lot of training, AI programs become smarter and more efficient decisions make up for the lack of reaction time, which is also a major advantage of AI programs in games

thirdly, although the ancient saying has the legend of seeing and hearing from all directions, and there are also legends of experts with broad vision in history, this is not very applicable to professional players in the game, but more applicable to artificial intelligence programs. Relevant technologies enable them to better control all areas of the game, while human players need to spend time focusing on different areas

alpha has better control over all areas of the game, which was also confirmed in its last game with mana. In that game, Google deepmind restricted the perspective of the newly launched alpha, so it no longer has obvious advantages in paying attention to all areas of the game. Mana will have time to find out the shortcomings of alpha and finally overcome it

alpha star's skills in StarCraft II come from the deep training conducted by deepmind, the developer. Deepvoh's plastic containers can replace glass and metal containers. Mind replays a large number of videos of human players during games, and also trains neural networks based on relevant data. In a week, alpha star played StarCraft II, which is equivalent to 200 years

alpha's current game level and super fast learning ability are not good news for professional players in StarCraft, but not all bad news. Professional players can also learn some useful strategies. After all, alpha has accumulated much more game experience through training than professional players

Copyright © 2011 JIN SHI