United States time in January 27th, Facebook and Google between the two play a no smoke of war. In less than a few hours of time, the two sides have announced their own artificial intelligence technology (AI) in the development of the software on the level and level. Such a tight bite, I am afraid it is not a coincidence. Let's see who's more artificial intelligence......
Face book voice
On Wednesday morning, Facebook's founder, Zuckerberg, knocked on his face book:
The story behind
Facebook last year November is on the International Conference on learning representations (ICLR): the paper "better computer go player with neural network and long term prediction, the author Yuandong Tian and Yan Zhu two Chinese. This paper introduces the artificial intelligence system of the go in the black forest. Almost at the same time Zuckerberg released the news, Facebook artificial intelligence research and Development Center released the update version of the paper, the center is responsible for the LeCun Yann also made the relevant introduction on Facebook.
The project independent leader Tian Yuandong said:
(Facebook Go program has reached the level of five amateur level)
This time, they added a Monte Carlo tree search to the new version of the system in the dark forest 2, with the previous convolutionneural networkAfter the combination, the performance has been further enhanced. Of war human players, dark forest 2 can reach amateur three levels; by 7.5 million times after the operation, you can remained stable in amateur five horizontal section in kgs server.
On the same day, Google said in an official blog:
On both sides of the opening are introduced from the beginning of this ancient Chinese thinking, Google even mentioned the Kong Zi and the four art. But behind the words, Google technology seems to be better.
The story behind
Google go paper can be described as early as come.
Also in January 27th, the DeepMind team published "the game of Mastering Go with deep neural networks and tree search Nature" as the cover papers on the line.
DeepMind is a London based artificial intelligence team, in 2014 the company was Google to 400 million pounds of the price of the acquisition. Hassabis Demis, the team's founder, said in an official blog post on Wednesday that the chess program AlphaGo, which was developed by them, is a combination of advanced tree search and deep neural networks. Through the neural network 3000 million step training (all chess were derived from human professional players of the game). AlphaGo predict human opponents next move right rate has reached 57. The previous record was 44.
(5-0 AlphaGo beat Fan Hui)
When it comes to the game record, Google obviously should be more proud of some. AlphaGo in October 5, 2015 to five straight beat European chess champion fan hui. The match is in full compliance with the rules of the game, and there is no set of rules for the game. In March this year, AlphaGo will be against the world's top players
(Korean players Li Shi)
Experts how to see?
Fan Hui, 2013-2015 three European Chess Championship
In China, go is not only a game, it is a mirror of life. We say that if your chess has a problem, it is very likely to be your mental problems
Really sorry to lose. And AlphaGo against the war before, I think I can win. After losing the first game, I changed the tactics, increased the attack, but still lost. The problem is that human beings sometimes make fatal mistakes, because we are human beings. Sometimes we will be tired, sometimes we are eager to win, we always have this kind of pressure. Computer programs will not be so, it is very strong, very stable, just like a wall. That's a big difference to me. I know AlphaGo is a computer, but if no one told me, I might think it is a little strange but very powerful opponent, is a real person.
Of course, losing the game made me unhappy, but as a professional, we lost a lot of games. So, the failure of the study, may be harvested. In a word, it is a good thing in the long run.
Toby Manning, the British Chess Association treasurer, referee fan Hui of man-machine war
Go players are aware that artificial intelligence has not yet solved the problem of go. So I think people will want to see the computer to reach the level of human professionals, but I think it may be 10 years time. In this game, I was optimistic about the fan hui.
It's amazing that in the course of the game it's hard to tell who is the human and who is the computer. In many chess game software, you can see a lot of reasonable and computer chess Road, will instantly collapse. But this time it's different, it's really hard to tell who is who.
A major feature of AlphaGo is different with human control time, fan Hui chess for longer than computer. The number of AlphaGo doesn't look like human being aggressive. It will quietly layout, rather than launch attacks or try to eat a piece of.
I think the main response to the game will be, just like IBM's deep blue chess master status, people want to be able to control the software, and in their own game to make up for their deficiencies.
When I know the computer to challenge the occupation top players Li Shi, really surprised. At the time, I thought that the challenger was absolutely no concept to the top players. But in fact, maybe I'm the one who doesn't know how strong this computer is. Now I'm very excited about this game.
Who will win? I don't know. Li Shi himself thinks that the computer is as strong as him. May be I heard too many AlphaGo news, I am now surprised by its powerful. At the same time I also know the level of Li Shi very much, so I think they both open five five.
I think the game is a very competitive, I do not think if the computer AI defeated the human will cause what harm.I think people will accept that they are going beyond the computer technology.
Schaeffer Jonathan, a computer scientist at University of Alberta, Chinook Designer
I think this is not the game version of the deep blue moment. The real achievement is that this program can be used as a player in a top competition. Deep blue was often overcome all the masters in 1989, eight years after the last summit is. But in terms of the current situation, I think the gap between the AlphaGo and the top human players has been greatly reduced. May only need to make a little effort and improvement, and then improve the ability to calculate a little less than one or two years, it will be able to defeat the human.
On March of this year, the game is not against the AlphaGo team, I am optimistic about the human.We can think of AlphaGo as a juvenile genius. It suddenly learned a good hand of chess, great progress, but still limited experience.From chess and checkers, experience is very important.
Hassabis Demis, founder of DeepMind
AlphaGo is advancing by leaps and bounds, and even beyond the best of human beings. Can see it in the rules of go to create new things, feel very magical. We have a very close relationship with the system we create, especially the way it is created.It will study, we will teach it, its style is like a person.You don't know what is going on in the end, because it can learn by itself.
Facebook Yuandong artificial intelligence research center, said the words, the two companies do have some kind of Tian
1 doi:10.1038/nature.2016.19255 Nature
3 Facebook, Googleblog, Wired
The academic circle of knowledge society, initiated by Tsinghua University professor of overseas scholars launched the public interest academic exchange platform, designed to share academic information, integration of academic resources, strengthen academic exchanges, promote academic progress.