Below, let's look at the different elements of the information technology industry, in the chain from science to application, in which position. In order to not reveal unnecessary details and reveal the essence, we divide this chain into Five links: scientific principles, basic common technologies, specific application technologies, basic system principles/technologies and specific application systems, see Figure 1.
Figure 1 From scientific principles to application systems, image source: author for the picture, the same below
The scientific principle is a summary of the basic laws of motion, and technology is the application of the law. Therefore, the introduction of new scientific principles often has a profound and extensive impact on society. It is precisely because the meaning of scientific principles is so great, so ' Science' is also often stolen. Many technical outputs have also been brought to the 'scientific' hat. In the computer field, Turing machine and computational complexity theory are basically in the category of scientific principles. It is precisely because of this, computers Was named as 'scientific'.
Fundamentally, the artificial intelligence boom cooled in the 1990s because people have worked hard in the field of artificial intelligence for decades, and have not been able to understand the nature of intelligent processes in the general sense, and thus have not been able to achieve scientific significance. The principle breakthrough, in theory, abstracts the basic intelligent operation like digital basic calculation to support more advanced and complex intelligent processes. Therefore, the output of artificial intelligence field, although rich and influential, has never been achieved. The height of scientific principles.
In an industry, there are some basic common technologies, sometimes called core technologies, which support the entire industry. In the information technology industry, operating systems, databases, integrated circuits, etc. belong to this level of technology. These technologies The progress of the whole industry is also global. It is the progress of integrated circuit technology that has led to a historic turning point in the entire information technology industry around 2010 (see "Turning - Looking at IT Peaks" Chapter 2 Section III) The field of artificial intelligence has not only the results of scientific principles, but also has not been able to produce basic common technologies supporting an industry, no matter what name we give to those technologies/methods.
On the basis of basic common technology, there are specific application techniques for different problems to solve different types of problems. At this level, we have encountered the trace of artificial intelligence. For example, we are in the book "Turning - Looking at IT Peak" In the second section of Chapter 8: 'When people realize that we are not capable of effectively solving various 'intelligent problems' with some basic basic logic rules or mechanisms, the research of artificial intelligence has entered various kinds. Among the specific problems. Different kinds of problems have been developed, and a lot of solutions have been developed. Great progress has also been made.. .. Because of this, artificial intelligence is now more and more regarded as specific. The application tool method is integrated into different types of applications, appearing under its own specific technical name, and playing its own role in obscurity. The name of a classic textbook about artificial intelligence in these years is 'artificial intelligence--a kind of The modern method ', its subtitle 'A Modern Approach' refers to the attempt to adopt the concept of 'Agent' to artificial intelligence in different fields.
In fact, the use of the concept of agent to integrate technical methods related to artificial intelligence is also a way to show no help. It shows a helpless reality in this field: only practical specific technical methods, lack of scientific principles or basic common technologies. Support, there is no effective theory at the basic system level. The 'deep learning' that has been popular these years is also the technology at this level.
The concept of 'deep learning' includes deep belief networks, convolutional neural networks, cyclical and recursive networks, and many different specific network models and corresponding algorithms to solve different types of problems. They are actually based on computers' Violence' computing power, using large-scale, nonlinear artificial neural networks with tunable parameters of up to tens of millions or more, using specific 'learning/training' algorithms, adjusting these parameters through statistical processing of large numbers of samples, achieving non- Linear fitting (transformation) to achieve the functions of extracting and subsequently classifying the input data features.
It's a concrete way of solving a particular type of problem, not a general learning ability like humans, although the name does trigger the imagination of many people who don't understand the technology. In fact, information technology Most of the technologies in the field belong to this level, including technologies related to big data, and they all belong to technologies that assist in intelligent nature. Therefore, the boundaries between big data, artificial intelligence and other technologies are increasingly blurred.
These specific practical techniques, including 'deep learning' (artificial neural networks), are often experimental techniques. Before applying to a new specific problem, we are not sure whether it can effectively solve the problem, or can solve the problem. To what extent.
Because of this, take deep learning as an example, in Deep Learning ('Mei' Ian Goodfellow, MIT Press, 2016), which is considered to be the foundation of the 'deep learning' field. In the textbook, in order to explain the experimental characteristics of deep learning, the author specifically sets Chapter 11 to discuss this issue. Its topic is taken as 'practical methodology'.
At the beginning of this chapter, the author wrote a passage: 'To successfully use deep learning techniques, it is not enough to know which algorithms exist and why they are effective. A good machine learning practitioner also needs to know how to Select a suitable algorithm for the specific application and how to monitor it, and improve the machine learning system based on experimental feedback. In the daily development of the machine learning system, the practitioner needs to decide whether to collect more data, increase or decrease the model capacity, add or delete Regularization, improved model optimization, improved model inference or adjustment of the software implementation of the model. Trying these operations requires a lot of time, so it is especially important to determine the correct approach without blind guessing. Deep learning The experimental features of this specific technology.
This state of artificial intelligence is somewhat like the traditional field. Before the advent of modern science, people have been able to design and manufacture many different types of sophisticated tools to solve various specific problems. Exquisite, it may not be able to produce a deeper principle, the result of universality. In the history of China, countless skilled craftsmen have not been able to let China catch up with the trend of modern science and technology, which illustrates this problem.
The output of the above layers cannot be directly served to people. So on top of them, there are principles and techniques for turning technology into a practical product/system. Here we have only two layers for the sake of simplicity. The fact that technologies such as artificial intelligence belong to specific application technologies rather than system-level technologies, so in fact they cannot be the basis for constructing actual application systems/products, but must be attached to system-level principles and related technologies to function. At the beginning, there were efforts to construct systems based mainly on artificial intelligence technology, such as the fifth-generation computer in Japan. In the future, this kind of effort will not completely disappear. However, from the objective nature of artificial intelligence technology, it is applied as a specific level. Technology to use, is a reasonable choice.
What is decisive for the underlying technology to play its own value is the basic system principle and related technologies. For example, the von Neumann architecture belongs to this category. It has become a core achievement in the computer field because it allows us to take advantage of this architecture. The related specific technology design and manufacture of the computer system products that can be actually used, and the computer system products enable the relevant specific technologies to play their due value in various fields.
The basic system principle and related technologies of this layer are relatively independent from the specific application fields, so the impact is also global. They include not only the principle and phase relationship technology of independent basic systems, but also a large number of independent systems. The interactive links constitute the principle and related technology of the macroscopic basic large system. Internet, cloud computing belongs to this category. In the field of networked information technology, the principles and technologies that constitute the macroscopic system play an increasingly important role. Of course, the innovation of independent basic systems is the premise of large-scale interconnected systems, and its role is more fundamental.
The importance of this layer of basic system principles and related technical innovations far exceeds those specific application technologies, at least comparable to basic common technologies, and some even close to scientific principles. So the von Neumann architecture is only available in the computer field. Such a lofty position. The powerful function of the human brain is not only reflected in the specific intellectual ability, but also in its system level. This system level advantage, in the first section of this chapter, we use examples to point out that it It is not just a highly distributed network formed by the distributed connection between a large number of neurons. In the next section, we will analyze another important aspect of the human brain at the system level, which is not fully valued.
In addition to the Internet, cloud computing has a recently-recognized 'blockchain'. The blockchain technology that followed Bitcoin in 2009 is the best way to explain the breakthrough at the system level. How can the lower layer be specific? Application technology has the greatest value.
In November 2008, an anonymous person who called himself 'Zhong Ben Cong' published a short but influential article on the Internet: 'Bitcoin: A Peer-to-Peer Electronic Cash System' (Bitcoin: One Equivalent The cash system on the network.) On January 3, 2009, Nakamoto founded his first block (Creation Block) in the Bitcoin system, and the Bitcoin system supported by the blockchain began to operate. Chain technology officially debut. The schematic of the technology is shown in Figure 2.
A closer look at the specific techniques in the blockchain reveals that the techniques used to solve specific problems in blockchains, such as asymmetric encryption, tamper-proof, peer-to-peer networks, etc., are all existing technologies, none of which It was invented or improved by Nakamoto. Using only these off-the-shelf technologies, Nakamoto has created a distributed operation on the open Internet. Without supervision, everyone can participate in accounting, and the accounts are transparent and transparent. Trustworthy, reliable, secure, accurate cash system.
If there is no basic feasibility yet, but countless people are still eager to create an intelligent system that is as intelligent as people, even surpassing people, a cash system can be said to be basically anyone before it is created. I have never thought of achieving this goal, nor will I believe that someone can create it. This is an information technology miracle that is built on the specific technology of existing information and beyond everyone's imagination.
This miracle relies not on new breakthroughs in technology to solve specific problems, but in an innovative systemic principle and design. Nakamoto takes full advantage of the behavioral characteristics of people in a distributed open network environment, and skillfully creates a Open block-based distributed system on the Internet. The so-called 'consensus algorithm' in Bitcoin to ensure accurate accounting records is part of the system-level operating mechanism of this distributed system. Or the consensus algorithm is adopted. The design of the system-level operating mechanism of a distributed system is realized. The reason why blockchain technology is difficult to be accurately grasped and understood is that many of its essential characteristics are determined by system-level mechanisms rather than relying on single-point technology. achieve.
Regardless of the application of the blockchain technology in the future, its emergence has brought us a very profound inspiration.
First of all, it allows us to see in the information technology industry a living, using mature and specific technology to create system-level innovation, beyond everyone's imagination, beyond the specific technology can achieve, miraculous subversive Effect. The importance of the system was highly valued in the 1980s. At that time, there were famous three theories in the scientific and technological world, namely, information theory, cybernetics and system theory. The result of deepening the understanding of the development of modern science and technology. The most important historical contribution made by Qian Xuesen is to deeply implant the viewpoints and methods of system theory into many fields of China, especially in the aerospace field.
In the traditional industry, we can also see a large number of systematic innovations surpassing specific technologies. Let's take a brief look at the example of the aviation industry. The power of the early aircraft was the piston engine. Later in the late 1930s. Germany took the lead in sending jets to the sky. Then the turbojet became the protagonist of aerodynamics. The two engines are scientifically identical, using fuel combustion to cause gas expansion and thermal energy into mechanical energy. But the system-level principles of the two are completely different. Turbojet engines are a fundamental system-level disruptive innovation in aerospace systems. The advances it brings are not the ability of piston engines to improve technology or materials at any particular level. Comparable. It has brought about tremendous changes in the aviation industry.
In today's military field, the decision of the battlefield advantage is no longer just the advancement of the specific weapon in the traditional sense, but the super system of sea, land and air integration. This is a qualitative leap in contemporary warfare.
Secondly, in the face of the dazzling results brought by the blockchain, we must not only ask: After the information technology industry has opened its golden age, will there be more, beyond our imagination, based on system level Innovations are constantly emerging, which have a subversive impact on all areas of human society.
Compared to the miracle created by blockchain, we focus our imagination on the improvement of tool intelligence. Although it makes sense, the horizon may be too narrow. At present, the miracle of blockchain and the more complicated intelligent technology are not too much. The direct connection is not a repetition or imitation of human intelligent activities, but it has an amazing subversive impact. Of course, if the blockchain is widely used in the future, it is an inevitable trend that various specific intelligent technologies are integrated into it.
In today's information technology industry, compared with big data and artificial intelligence, because its professional blockchain is a relatively lonely technology, it is it that opens a unique observational information technology industry. The windows of future development provide a unique perspective. The potentials revealed by it may have more essential and important significance for us to effectively grasp the future development of the information technology industry and the entire human civilization.
The current information technology application system has increasingly become the design of software, and as we pointed out in the first section of Chapter 2 of "Turning to the Peak of IT", 'Compared with the design and development of physical products, software development It is almost free to play in the virtual space of an infinite resource. So software design development is even considered to be a purely spiritual art creation process. 'It is this freedom of free creation that has a blockchain Miracles will also spawn more disruptive system-level innovations of varying system sizes beyond our most romantic imagination today.
The functions that many systems will achieve will far exceed our expectations for artificial intelligence as a whole. They will not have the subjective status of people, but they will greatly surpass the human beings in intelligence as an individual. The height and breadth of the body - just like the mechanical tools of auxiliary physical energy achieve a comprehensive transcendence of human beings in all aspects of physical fitness. The 'intelligence' of these systems will be significantly different in nature from human intelligence, just as The blockchain system is shown to us. These systems are being gradually created and will form an unprecedentedly complex virtual world.
While paying full attention to the progress and impact of specific application technologies such as artificial intelligence, we should expand our imagination and pay more attention to the innovation of basic system level principles and related technologies. The future of information technology industry development may be more This level of subversive innovation is determined. And specific technologies such as artificial intelligence will find better and bigger stage in these innovations.
The intelligent system with the main virtual image as the core in the book “Turning – Looking at IT Peak” is the basic system principle and technology level in the information technology industry. It is an important innovation with systemic intelligence meaning, representing information. An extremely important direction for the application of the class is a great advancement in human intelligence pursuit, and a qualitative leap in human-assisted intelligence tools.
About the author: Dr. Xie Wei, a leading talent in the capital science and technology department, Ph.D., Department of Electronic Engineering, Tsinghua University. Author of "My Workplace Ten Years", "Growth - From Campus to Workplace", "Turning - Looking at IT Summit".