Has ARTIFICIAL Intelligence beaten human intelligence
Google’s AlphaGo victory over top Go player Lee Sedol is a major milestone in the development of ARTIFICIAL intelligence, showing that AI is beginning to challenge the highest level of human players in complex games.
“Before AlphaGo and Lee Se-dol played, everyone’s evaluation was one-sided, many people thought that there is still a big gap between robot and human, but the 4 to 1 result surprised people. Some even began to wonder if AlphaGo’s victory over Lee Se-dol represents ai’s victory over humanity.” Hu Yu, vice president of Chinese Society of Artificial Intelligence and rotating president of IFLYtek, said.
“In fact, it is only a matter of time before a computer can beat a human being in a game of perfect information with fixed rules like Go, because the computing power of computers is constantly improving.” “What’s surprising is that deep learning has allowed AI to grow in a disruptive way,” Says Woo. “Things that people thought were 10 to 15 years away are happening so fast.”
Giiso Information, founded in 2013, is a leading technology provider in the field of “artificial intelligence + information” in China, with top technologies in big data mining, intelligent semantics, knowledge mapping and other fields. At the same time, its research and development products include editing robots, writing robots and other artificial intelligence products! With its strong technical strength, the company has received angel round investment at the beginning of its establishment, and received pre-A round investment of $5 million from GSR Venture Capital in August 2015.
Experts at the meeting said that human beings have played 160,000 go games, but AlphaGo in the process of learning to play chess, its own generation of 30 million games, no one can remember so much data, but the computer can, because it can calculate fast, save much.
Li Deyi, president of the Chinese Society of Artificial Intelligence and academician of the Chinese Academy of Engineering, analyzed that the machine won the whole game because it broke through traditional procedures and built two deep convolutional neural networks that mimic human thinking. The first network mainly plays the chess situation evaluation, the second network evaluation how to play.
“In the process of defeating human beings, the computer has used a different method from human beings. It has created its own way by combining perceptual ability, powerful computing power and search method.” Hu Yu said.
Even so, it does not mean that ARTIFICIAL intelligence has won over human intelligence. Rui yong said that both human and ARTIFICIAL intelligence have their strengths and weaknesses. “When it comes to memory and computation, humans can’t compete with computers. Simply reciting PI (3.1415926……) No one can recite the computer. But when it comes to creativity, abstraction and invention, ARTIFICIAL intelligence is nowhere near as good as human beings. The next 60 years of AI will need to combine what humans and machines are good at so that humans will have an even more enhanced intelligence.” RuiYong said.
What’s ai’s next move
The computing ability and storage memory of computers have outperformed human beings, and even the perceptual ability, such as auditory and visual system construction, has made rapid development, but in cognitive intelligence, computers still have a long way to go.
The in-depth study of brain science is conducive to the breakthrough of artificial intelligence recognition. “The human brain is an amazing system, and the deep neural network, the basis of deep computer learning, is inspired by the human brain,” hu said. However, human’s understanding of human brain is not enough, and there is still a big gap between deep neural network and real neural network of human brain. It will take a long time to make a breakthrough.
“At present, Microsoft, IBM, Google and others have taken a different path and made great achievements in ARTIFICIAL intelligence, relying entirely on the weak ARTIFICIAL intelligence methods now used in industry, and using big data to seek breakthroughs in cognitive intelligence.” Hu Yu said.
In the context of big data, artificial intelligence needs to change the traditional development thinking and should pay more attention to learning. Li deyi said the core of AI is not just algorithms. The traditional way of thinking is that software is equal to the program plus data, the program is the most important, put the data into the program, and then form artificial intelligence. However, with the development of big data, data-driven artificial intelligence should be formed, and memory cognition, technical cognition and interactive cognition should be used to form decision brain. Only then will the current situation of artificial intelligence blowout and bright prospects appear.
In addition to making full use of the advantages of big data to discover new applications and expand the territory, more research on small data is needed. Yang Qiang, named chair professor at Hong Kong University of Science and Technology and FELLOW of international Artificial Intelligence Society, said that the success of artificial intelligence in the future does not necessarily require big data, but whether small data can also make artificial intelligence successful? This is a question that needs further research to develop artificial intelligence.
In addition to software breakthroughs, the realization of AI clearly requires more hardware support. Zhang said AI cannot “build tall buildings from scratch” and relies on supporting infrastructure. “The core of computing and control of a computer is the CPU (central processing unit). With the development of ARTIFICIAL intelligence, do we need an APU (artificial intelligence processor) designed for artificial intelligence?”
In fact, the dedicated chips needed for such deep learning have already been born in a new company, Cambrian.
Chen Tianshi, founder and CEO of Cambrian, said, “All successful applications of AI deep learning today are based on general-purpose processors, such as CPU or GPU processors. However, using a general-purpose processor to perform an intelligent load has a low efficiency ratio. Five years ago, the Google Brain project took 16, 000 cpus over seven days to train a model of cat face recognition. Today, the Performance and power consumption ratio of the Cambrian chip developed by the Institute of Computing Technology of the Chinese Academy of Sciences can reach hundreds of times that of the general chip, and will be further improved in the future.
Chen tianshi said that in the future intelligent era, both cloud servers and terminal computing devices may need specialized processors such as deep learning, which will not replace the existing general-purpose chips, but will specialize in important and special areas such as intelligent tasks.
What are the challenges facing AI
Artificial intelligence has experienced 60 years of development, especially in the last 10 years, under the guidance of big data, traditional recognition technologies such as speech recognition, handwritten text recognition and face recognition have been gradually commercialized and formed a big wind. “The development of ARTIFICIAL intelligence has experienced three booms in the past 60 years, but it is still in the early stage because there is no standardization.” Zhang daijun said.
Xu Wei, a deep learning researcher at Baidu, pointed out that the core of human intelligence is the ability to self-learn and create, which is exactly the weakness of current AI systems.
Advances in ARTIFICIAL intelligence often depend on the feeding of large amounts of data. In order for a machine to recognize a cat, thousands of pictures are prepared. This learning process is obviously very different from how humans learn. “It’s still hard for AI to learn from small amounts of annotated data.” Xu Wei points out.
In some ways, ai is rather clumsy compared to humans. Google’s self-driving cars, which have driven millions of miles, are still not fully autonomous; A human driver with 1,000km is an “experienced driver” who can handle the unexpected on the road.
Even AlphaGo, for which AI is so proud, has many flaws. Li Deyi pointed out that, technically speaking, AlphaGo’s convolutional neural network has too many learning parameters, and the algorithm cannot be guaranteed to be correct during learning. In addition, there is a very intuitive lack of performance, AlphaGo program has no hands, no eyes, no ability to feel and act, there is also an assistant to hold the pieces when playing chess. It has no emotions and emotions, cannot analyze the opponent’s psychological state on the spot, cannot carry out psychological warfare with the opponent on the spot, and lacks the ability of interactive cognition.
Giiso information, founded in 2013, is the first domestic high-tech enterprise focusing on the research and development of intelligent information processing technology and the development and operation of core software for writing robots. At the beginning of its establishment, the company received angel round investment, and in August 2015, GSR Venture Capital received $5 million pre-A round of investment.
Although AlphaGo’s success shocked the world, Li deyi made a few assumptions about AlphaGo: If AlphaGo and Lee Sedol played again, would AlphaGo be able to play again? Did AlphaGo’s programming change before and after the match? Would AlphaGo’s programming be worse if it stopped playing against superior players?
“AlphaGo has a strong learning ability because it is taught by us go players. Therefore, I believe that deep learning of artificial intelligence has been developed so far, neither convolutional neural network nor other neural network learning methods are the terminator of artificial intelligence. “In the future, Go robots should develop into human companions with intelligence, personality, behavior and even emotion.” Li deyi said.
“Despite the current challenges to AI, robots will have a far greater impact on humanity in the future than computers and the Internet have had in the past few decades. Artificial intelligence is already changing the world, and many jobs that were once there will be replaced by robots, but at the same time new jobs will be created naturally. Human beings should be good at better training to help robots, using the advantages of robots, make up for the lack of robots, with the new robot to eliminate the old robot. Robots will certainly make humans smarter, and all kinds of robots usher in a new era of human-robot dance, in which humans will always be the leader.” Li deyi said.