From AlphaGO to the current Master, artificial intelligence AI is not necessarily invincible

March 9th, 2016 may become a day in the history of mankind - Li Shishi, the Google AlphaGo battle that lasted for five days, started a game in which the world’s most popular go-player and artificial intelligence were supervised by all humans. Commonly onlookers.

Before the game, many people in the world of chess and science made predictions about the game. Google is confident in the robots it develops, claiming that according to computer calculations, Li Shishi’s probability of defeating the “Alpha Dog” is zero. The majority of the chess world is on the side of Li Shishi. Chinese chess Saint Nie Weiping thinks that Li Shishi’s defeat is impossible.

In this first game, Li Shishi lost to the "Alpha Dog" in the case of the previous advantage, and declared defeat in the final game. Before the game, Li Shishi had confidently stated that he could win 5 sets. So the flag should not stand up. Where is AlphaGO sacred? So strong enough to let the Go generation arrogant Li Shishi defeated the nine paragraphs?

A brief introduction, what is Google AlphaGo? AlphaGo, an artificial intelligence, and IBM have defeated the deep blue supercomputer of chess master Kasparov in the last century, as well as the contemporary Apple Siri and Google Now. To interpret AlphaGo, first we need to understand what is behind AlphaGo. In fact, behind Google AlphaGo is a neural network system developed by DeepMind, a British artificial intelligence company acquired by Google in 2014. This system is different from Deep Blue, not a supercomputer, but a neural network system with many data centers connected as nodes and multiple supercomputers in each node. Just like the human brain, it is made up of 5-10 billion neurons, which is why this machine learning architecture is called neural network.

AlphaGo is an example developed on this neural network system specifically for Go. However, although the name has already introduced its features, the neural network system behind AlphaGo is suitable for any intellectual sports project.

In fact, the "Master" player is an upgraded version of AlphaGo.

On January 5th, the AI ​​developed by Google DeepMind returned again after defeating the Korean World Championship on the Go board. From December 29, 2016 to January 4, 2017, the upgraded version of Google AlphaGo was named after the Master. In the fast chess game of Chengweiqi and Wild Fox Go, the player with the highest level of humanity achieved a 60:0 overwhelming record, once again making people shocked the strength of Go AI. Earlier on the 5th, Google DeepMind founder Demis Hassabis posted a message on Twitter confirming that the upgraded version of AlphaGo was online with a real person.

Hasabis wrote: "In the past few days, we have conducted some informal Go games online, and the game is in the form of fast chess... Our purpose is just to see if the system is as good as expected." Also said: "Thanks to the players who played against us in Yicheng Go and Wild Fox Go. Our accounts are Magister(P) and Master(P). We would also like to thank those who watched the game."

Many people have been guessing who this Go master is. Everyone is not sure. Some people suspect that it is a computer. Chinese chess player Coulee offered a reward of 100,000 yuan to find a player who could defeat the "Master".

The news released by DeepMind shows that the informal test may have ended, and later this year, AlphaGo will participate in some official competitions.

All along, we are working hard to improve AlphaGo. In the past few days, we have conducted some informal online competitions. The game was played in the form of fast chess. The latest development of the prototype system was participated in the competition. Check the system to see if it is as good as expected. Thanks to the players who played against us in Yicheng Go and Wild Fox Go. Our accounts are Magister(P) and Master(P), and we would like to thank those who watched the game.

The new version of AlphaGo made some creative and very beautiful moves in the game. We learned a lot from the Go community, and we were very excited and satisfied.

AI talks about everyone: AI has no rivals in the world of chess, but it does not prove that it is the same in other fields.

After playing against AlphaGo, the player Gu Li posted: "Human and AI work together and will soon reveal the deep secrets of Go." Now our informal test is over. This year we are ready to cooperate with Go organization and experts. Hold a formal and complete competition to further explore the secrets of Go, and let humans and AI inspire each other and make progress together. Soon we will announce more news.

Master has not lost in the current game. Does it mean that there is no more opponent in the world of chess in the world of artificial intelligence? The fourth paradigm CEO Dai Wenyuan believes that the basics are a foregone conclusion. Dai Wenyuan said that the ability of machine learning is too strong. It is a tireless child who is diligent and eager to learn. No one can maintain such strength, so the gap between humans and machines will become larger and larger. Now that I want to win the machine, it is possible to have a black swan event like a blind spot in the top master.

For the question of whether the Master is AlphaGo, Dai Wenyuan said that this is no longer important. Because there is no secret in AlphaGo's own technology, as long as there is enough computing resources, and then follow the AlphaGo way to train a game of Go, you can achieve such a high level. Dai Wenyuan said that the technology does not need to have a particularly big breakthrough. It only needs to train for a period of time in the original way, and it will be stronger than the original AlphaGo. However, although it is technically possible for other companies to implement the Master, it should not be a small company in terms of cost, because the calculation cost is still quite high.

For Master's move to kill the Quartet in the world of Go, go out and ask CEO Li Zhifei, I don't know if Master is AlphoGo, but I believe that all ArTIficial Games will eventually win over humans. AlphoGo is a good demo of AI's ability, but if AlphoGo wins humans, it means that AI IQ is more nonsense than human beings. For example, AlphaGo does not involve the composition of many people's IQ such as natural language understanding and dialogue.

Is it easier to train an AlphaGo in other fields? Dai Wenyuan believes that this is not the case. "The scene of playing chess has a natural advantage. That is, the machine only needs to play chess with itself every day. It does not require a lot of input from the outside world. Similar scenes include the game field and the machine. You only need to keep practicing with yourself and you can practice it. There are also real-life scenarios such as driving and flying. Although it involves cost issues, you only need to accumulate data to complete the training.” Dai Wenyuan said that in other fields, for example, medical treatment. In the field, it is difficult to achieve the goal of training AlphaGo, because it is not that the machine wants to learn the case, it must have enough suitable cases to appear. Such a scenario requires sufficient input and cooperation from the outside.

Adhseive To LSE Plastics Membrane Switch

This is some kind of Membrane Switch with plasitc case, soushine can provide good quality membrane switch adhesive to suitable case, then achieve the waterproof effect, also reduce the assembly operation.

Adheres To Lse Plastics Membrane Switch,Exibdl Membrane Switch,Alarm Membrane Switch,Methane Oxygen Detection Membrane Switch

Dongguan Nanhuang Industry Co., Ltd , https://www.soushine-nanhuang.com