What Master brings to the world: Is it "out of control" or evolution?

[51CTO.com original article] Master shocked the Go world with its 60-game winning streak in early 2017. ***DeepMind officially confirmed that the Go program "Master", which swept the Go masters from China, Japan and South Korea in 60 games on the Go website, was also from DeepMind. This is an important progress made by DeepMind in artificial intelligence in the past six months based on AlphaGo.

DeepMind's official summary reads, "The most exciting thing is the creativity shown in the game of AlphaGo. Sometimes, its moves even challenge the ancient wisdom of Go. Go, one of the most far-sighted games of all time, AlphaGO can identify and share insights."

How to view Master or how to view the amazing changes in artificial intelligence, I think this is what we all pay attention to and think about. In fact, with the invention of computers, people have been discussing what kind of artificial intelligence this will lead to? One foreseeable possibility is that functional artificial intelligence can be produced, which is the result widely achieved by supervised deep learning today. Another view is that artificial intelligence can imitate human thinking and emotional activities, which is the future that unsupervised deep learning will create.

[[180790]]

Below we will analyze the thoughts and significance behind this great victory of artificial intelligence from three major aspects.

1. Technological progress is the key

Artificial intelligence actually has a history of more than 60 years. The development of its technology has experienced ups and downs, which is quite dramatic and representative.

The Dartmouth Conference was first initiated by John McCarthy and others on August 31, 1955, with the aim of gathering like-minded people to discuss "artificial intelligence" (this definition was proposed at that time). The conference lasted for a month and was basically a large-scale brainstorming, which gave birth to the AI ​​revolution that is now known to everyone. At that time, many innovative inventions appeared in the field of algorithms, including a prototype of reinforcement learning (i.e. Bellman's formula), which is the core idea of ​​Google's AlphaGo algorithm today.

However, for a long time afterwards, artificial intelligence did not become popular because people found that artificial intelligence could only do very simple, very specific and narrow tasks, and could not cope with anything slightly beyond the scope. There are two limitations: on the one hand, the mathematical models and mathematical methods based on artificial intelligence were found to have certain defects; on the other hand, many computational complexities increased exponentially, making them impossible computational tasks. These problems caused artificial intelligence to encounter bottlenecks in its early development, which directly led to the reduction or cancellation of many artificial intelligence projects.

The concept of decision support system was proposed in the 1970s, and then decision support system has been greatly developed. In the late 1980s and early 1990s, decision support system began to combine with expert system to form intelligent decision support system. Intelligent decision support system has achieved the organic combination of qualitative analysis and quantitative analysis, which has greatly developed the ability and scope of machine problem solving. At this time, there were also major inventions in artificial intelligence mathematical models, including the famous multi-layer neural network and BP back propagation algorithm, and there was also a highly intelligent machine Deep Blue that could play chess with humans. Then in 1997, Deep Blue defeated chess master Kasparov, which shocked the world. At that time, Deep Blue was a behemoth, weighing 1.4 tons, with 32 nodes, each node had 8 processors specially designed for chess games, and the average computing speed was 2 million steps per second. But Deep Blue at that time could not actually think, and its endless computing power made up for its shortcomings to a large extent.

With the rapid development of computers, computing power is becoming more and more powerful. When more powerful computing power is transferred to artificial intelligence research, the research effect of artificial intelligence is significantly improved. Due to this series of breakthroughs, artificial intelligence has produced a new boom. In terms of more general functionality, machines can also reach or exceed human standards in math competitions and image recognition competitions.

We can see that in the early research of artificial intelligence, because early artificial intelligence research was more defined as mathematics and algorithm research. In the 21st century, with the rise of cloud computing and big data, distributed large-scale computing gradually became popular. Hadoop is an excellent representative of it. It can use different machines to execute subtasks in parallel to complete large-scale computing tasks. At the end of 2013, Carnegie Mellon University in the United States made an open source release of the distributed machine learning system and named it Petuum. This Petuum is a deep optimization of distributed computing systems such as Hadoop and Spark from the perspective of software optimization. On the other hand, there are other scientific research institutions trying to completely solve the bottleneck of the von Neumann architecture from the perspective of hardware, which is the neuron chip and quantum computing in the longer term. Then in 2016, we saw that GOOGLE's AlphaGo defeated Lee Sedol with the help of deep neural networks and super-powerful massive distributed computing capabilities.

Looking back at the 60 years of development of artificial intelligence, we see that technology is the key. From the initial reinforcement learning, the machine continuously interacts with the external environment, "learns" through continuous trial and error and cumulative rewards, to machine learning and AlphaGo's exploration of reinforcement learning algorithms, the door to "infinity" has been opened. It is through the research and hard work of countless people that the best combination of theory and practice is found between ups and downs, disappointment and hope, and the balance between technology ecology and artificial intelligence is achieved.

2. The known and unknown world

First of all, we know that practice is the basis and source of knowledge. Knowledge is neither "innate knowledge" nor subjective, nor directly derived from the object, but obtained through the practical activities of the subject actively transforming the object. As Engels said: "The most essential and immediate basis of human thinking is the changes in nature caused by man, not nature itself." Human thinking depends on our cognition of the objective world and how much we know about ourselves.

We divide our understanding of ourselves into two parts, one is "we know" and the other is "we don't know". For example, when playing Chinese chess, we can explain it clearly, we know why it is so, and we know why it is so. This can be solved with traditional artificial intelligence models.

Of course, there are still many questions that we cannot explain clearly. A typical example is Go. We can explain some aspects of Go and chess clearly. For example, why the knight moves in this way, why the pawn moves forward, and why a white piece falls in this place in most cases, even the master cannot explain it clearly. He will say that this is my feeling of the game.

How do you get a sense of chess? This is AlphaGo's major contribution. It treats the sense of chess as pattern recognition. When you see the layout, you should know how to place the pieces. The layout is a pattern. We can say that a Go master is good at Go not because he is too smart, but because he has a strong ability to recognize patterns. If the pieces change a little, he will know how the situation will change in the future, and then my strategy will also change accordingly. This is done by using traditional models + deep learning. AlphaGo does this.

In the past, we said that machines have no emotions and consciousness, but now this can also be simulated using deep learning methods, at least on the surface. We have found that artificial intelligence can do so much, and there are fewer and fewer things that it cannot do. There are some things that we humans cannot do, such as no emotional fluctuations, full of energy, and never tired. These artificial intelligence can easily do all of this.

In a world unknown to us humans, when we don't know what to do, what will humans do? That is to try and explore. Through continuous attempts and explorations, we slowly understand and gradually clarify what we don't know. We turn what we don't know into what we know, and then pass it on. However, due to the various materials and research conditions at the time, our understanding of the world is always limited. Our understanding of the world is like a "circle". The larger the radius of cognition, the larger the circumference, and the larger the unknown "external" world, so we are more confused. The reason why there are many things we don't understand is that we always see only part of the world, and we see the world within the "circle" of our cognition.

With the development of science and technology, our pace of understanding the world will accelerate. With the help of machines and artificial intelligence, the world will become clearer and clearer in front of us, and the unknown things will gradually decrease. So, this is why we say that artificial intelligence has such great hopes.

3. The relationship between humans and artificial intelligence

The emergence of artificial intelligence will not reduce the level of human chess players. On the contrary, it can help us analyze and improve ourselves. After Master defeated Ke Jie, chess master Nie Weiping said, "Master has changed our traditional concept of thickness and thinness and overturned the stereotypes of many years. Go is far from being as simple as we imagined. There is still a huge space waiting for us humans to explore. Whether it is AlphaGo or Master, they are all sent by the "Go God" to guide humans."

The development of artificial intelligence has greatly promoted the progress of society and deepened people's research on epistemological issues. Compared with computers, in general, the human brain has the ability to process fuzzy information and is good at judging and processing fuzzy phenomena. However, computers have poor recognition ability for fuzzy phenomena. In order to improve the ability of computers to recognize fuzzy phenomena, it is necessary to design the fuzzy language commonly used by people into instructions and programs that can be accepted by machines, so that machines can make corresponding judgments as concisely and flexibly as human brains, thereby improving the efficiency of automatic recognition and control of fuzzy phenomena. In the end, artificial intelligence will become smarter and the improvement points it learns in any situation will be enhanced.

Artificial intelligence has broken the inertia of human thinking to some extent. Some of the special tricks used by MASTER are very creative and worth learning and thinking about.

The Future of Artificial Intelligence

We see that although artificial intelligence has surpassed humans in some aspects, it is also advancing by standing on the shoulders of humans.

When intelligent machines can acquire intelligence through autonomous learning and exploration of the world, the changes that may occur in the future will be "infinite", and the future of artificial intelligence is destined to be exciting. Of course, this road is not smooth and may be tortuous.

Facing the wave of artificial intelligence, don’t be afraid or hesitate, just rush forward and ride it, that’s it!

[51CTO original article, please indicate the original author and source as 51CTO.com when reprinting on partner sites]