2016 is about to pass. In the middle of the year, I believe that the most vocal words in the ICT industry are AI (Artificial Intelligence). No matter the enterprise or the media, all the words and ICTs must be labeled with AI. Otherwise, It is suspected of OUT. But what about the actual situation?
The so-called traceability. Here we may wish to go back to this year's more important or cause of media or industry AI attention or speculation of several nodes or landmark events, and the interpretation and extension of these landmark events will undoubtedly let us see the essence of AI.
First, Google's AI company DeepMind developed the Alpha Go neural network to win the absolute advantage with the world's Go champion Li Shishi's "human-machine war", which led to the industry's attention to AI, after which AI began to appear frequently in technology media and enterprises. In the report;
Second, Google's main AI driverless car appeared in the road side of the first accident and Tesla's Autopilot (automatic driving system) frequent deaths, although it is negative news, but still stimulated the industry's attention to AI And reflected in the hot and automatic driving of driverless cars;
Third, Amazon's Echo speakers with Alex voice recognition technology, the so-called best-selling and Mary Meeker, who is known as the "Internet Queen", released the 2016 Network Trend Report, which is a good prediction for Echo and AI.
Fourth, the US presidential election called MogIA's artificial intelligence system successfully predicted that Trump will become the president of the United States;
Fifth, the graphics chip company or its own advertised as artificial intelligence company's Nvidia share price surge.
First of all, let's see what Alpha Go relies on to win Li Shishi in Go. In fact, for the game between computer and human beings in chess, as early as 1997, the computer defeated the first-ranked player in the world. Gary Kasparov lost to IBM's computer program "Deep Blue" with 2.5:3.5 1 wins, 2 losses and 3 draws. At that time, the global media and high-tech circles exclaimed that artificial intelligence entered a new era.
Prior to 1988, the "Deep Blue" previous generation "Thinking" was the first computer to win a chess master. In 1996, "Deep Blue" became the first computer to win the world championship in chess. It should be noted. Dark blue weighs 1270 kilograms and has 32 brains (microprocessors). It can calculate 200 million steps per second and enters more than two million games with more than one hundred years of excellent players.
In contrast, AlphaGo initially tried to match the professional chess player's past game by imitating the human player. The database contains about 30 million moves, and the computing power is 30,000 times that of the original "dark blue". The biggest difference we see here compared to "dark blue" is the advantage of AlphaGo in terms of data and computing power.
Some people may say that AlphaGo wins in its huge and complex neural network, but according to the papers of TIan yuandong and AlphaGo, if you don't do any search (actually test the computing power), just based on the "game feel" (in fact, it is estimated Value function), CNN (neural network) is best to reach the level of KGS 3d, which is the level of amateur 1 segment.
The MCTS algorithm can defeat human masters on the 9x9 board without Value Network. In fact, it proves that AlphaGo's strength in the endgame plays an important role in search (calculation), that is, the computing power of the endgame crushes humans. But the well-known fact is that the computer's computing power is far stronger than humans have long been common sense.
In response, the executive vice president of Microsoft Research Asia, Yong Yong, told the media when evaluating AlphaGo:
"All of today's artificial intelligence comes from the big data of human past. No ability in any field comes from self-awareness. Whether it is chess or go, computers learn from the past chess of human beings. Other fields are similar. When computers do image recognition, they also learn a lot of pictures from the big data that humans already have.
In the face of problems that humans have never taught, computers will be ignorant. If you let AlphaGo go to checkers, it will be completely stupid. Even saying that the chess board of Go is slightly modified, from the grid of 19&TImes;19 to the grid of 21&TImes;21, AlphaGo can't stand, but humans have no problem. â€
The Oxford English Dictionary defines intelligence as "the ability to acquire and apply knowledge." In the words of Tom Davenport, a researcher at the Massachusetts Institute of Technology's (MIT) Digital Economy Initiative and AI opinion leader, "Deep learning is not a profound study."
Another expert, Allen InsTItute of AI, has similar opinions: “AI is just a simple mathematical implementation of large-scale execution.†Simply put, the current AI is only a powerful calculation method, and it has not been reached. The human brain is a smart way.
Michelle Zhou, an expert who spent 15 years working with IBM Research and the IBM Watson team, is an expert in the field, which divides AI into three phases.
The first stage is to identify intelligence. Algorithms running on more powerful computers can recognize patterns and get topics from a large amount of text, and even get the meaning of the entire article from several sentences. The second stage is cognitive intelligence, machines. It has transcended pattern recognition and began to make inferences from the data; the implementation of the third stage will wait until we can create virtual humans who think and act like humans.
And now we are only in the first stage, "identifying intelligence", that is to say, a large part of what people say "artificial intelligence" is actually data analysis, or the original routine or "old bottled new wine".
Coincidentally, if the above AlphaGo finally relies on powerful computing power to reflect the so-called AI advantage, then we will say that Google and Tesla's automatic and driverless cars are biased in simple data analysis. .
The most typical performance is the Google driverless car that has been touted before. This year, a traffic accident occurred at a speed of less than 2 miles per hour, and it was Google by responsibility.
If we compare the choices and results of the accident at the time of Google’s driverless car with the choices and results of each move in this man-machine war, for the AI ​​(such as AlphaGo), the former does not know how many times easier. (The biggest advantage of Google’s driverless system over humans is to anticipate each other’s behavior and respond.)
It is a pity that the Google driverless car in this accident shows that the intelligent system failed to fully judge the behavior of human beings, and also made the most disappointing and probably the most sinister choice for human drivers. Eventually led to the accident.
As for Tesla, after repeated accidents this year, it upgraded to the Autopilot 2.0 system and released a demonstration video of the second autopilot technology.
Disposable Vape,High 2.0 Disposable Vape,Ecig Products,Flavors Disposable
Maskking(Shenzhen) Technology CO., LTD , https://www.szdisposablepods.com