At the end of the year, if you want to summarize the major events of the technology industry in 2023, artificial intelligence can almost be regarded as one of the main lines throughout the year.
From the end of last year, the GPT-3 general language model appeared "wisdom emerged", to the beginning of this year, ChatGPT was out of the circle in a high-profile manner;From the "Thousand Model War" in full swing to "all businesses are worth doing all over again with artificial intelligence";From Nvidia's market value soaring, to OpenAI's personnel change drama ......
However, for most of the past few decades, there have been no IPO gongs and flowers on this AI track, and there are only long thorny roads.
This is especially true for Chinese AI people. Due to missing the first wave of artificial intelligence in the world in the 50s of the last century, there are no authoritative scholars such as Minsky, McCarthy, Sima He, Feigenbaum, and Hinton in the Chinese artificial intelligence community, but since the 70s of the last century, we have caught up, with relatively limited resource investment, not only achieved world-class results in the research of symbolist artificial intelligence, but also caught up with the boom of neural network research, and gradually caught up with the second position in the world today.
In this process, a major event is the national "863" plan, in the 90s of the last century, when the global artificial intelligence fell into a trough, the "863-306" theme through continuous investment, in high-performance computers, intelligent interfaces, intelligent applications and other aspects of a number of major scientific research achievements, for China to cultivate a large number of computer talents into the international high-tech frontier, but also become the source of a large number of high-tech companies such as Shuguang, Hanwang, Zhongke Xingtu, iFLYTEK, Cambrian, etc.
The recently published "A Brief History of Chinese Artificial Intelligence (Volume 1)" was written from 1979 to 1993, through a large number of interviews with the parties, with the artificial intelligence and pattern recognition committee of the China Computer Federation, and some personnel of the Chinese Artificial Intelligence Society as the main line, systematically sorted out the early source flow of Chinese artificial intelligence development, and provided a rare map for friends who want to understand and devote themselves to the field of artificial intelligence today.
The development of artificial intelligence is a spiraling process. Taking history as a mirror can help us understand how to grasp the rules in scientific research work, not follow the trend, and better meet these challenges. Gao Wen, director of the Pengcheng Laboratory and academician of the Chinese Academy of Engineering, wrote in the preface to the book.
This article is the author of "A Brief History of Chinese Artificial Intelligence (Volume 1)" by the scientific and industrial forces.
1. Interview record of Lin Jun, founder of Leifeng.com.
Science and Engineering Power: Tell us about the background and purpose of writing this book from your perspective
Lin Jun:Teacher Gao Wen also mentioned in the preface that it was New Year's Day in 2018, I was on vacation with my daughter on white beach in San Diego, USA, and suddenly received *** from the teacher He said that there was a ppt about "The History of Chinese Artificial Intelligence" and wanted to ask my opinion.
I listened and listened, and found that the style of painting was wrong, and it turned out that he was trying to fool me into writing. As soon as my head was hot, I agreed.
A few weeks later, I returned to China and had an interview with Mr. Gao Wen in the office. Mr. Gao gave an idea to start our interview with the Artificial Intelligence and Pattern Recognition Committee of the China Computer Federation and some personnel of the Chinese Artificial Intelligence Society.
The China Computer Federation (CCF) is a first-level society, and its Artificial Intelligence and Pattern Recognition Committee is a second-level society equivalent to other societies, which has great influence. The Chinese Society for Artificial Intelligence (AAI) was not a pure artificial intelligence society at first, and it also included many scholars in the fields of social science and philosophy, which to a certain extent led to a certain degree of incompatibility between them and computer science-based societies. Over time, the Chinese Society for Artificial Intelligence has also gradually shifted to a direction that focuses more on the fields of natural sciences and engineering.
Taking the personnel and development of these two universities as clues, to a certain extent, it can reflect the development process of Chinese's artificial intelligence field, of course, it does not fully cover all scholars and institutions that have made important contributions to the field of artificial intelligence.
Then I went to Northeast China and visited Wang Kaizhu, an important teacher in the field of natural language in China and the old head of Harbin Institute of Technology, and Liu Dayou, a student of Wang Xianghao, at Jilin University. When I came back, I began to conceive and prepare for writing.
But the publication of this book has taken a long time and is a bit tortuous, first of all, because I was concentrating on writing "Boiling New 10 Years", and in 2020, there was the impact of the epidemic, and the topic of artificial intelligence was not a hot topic in the ** circle at that time, and the publishing house was not very enthusiastic, so the publication progress was delayed. But the interview kept advancing.
By the end of 2022, ChatGPT became popular again, and we accelerated the progress and published the first volume in 2023. At present, more than half of the second volume has been written, and 30% of the third volume has been written, and we want to finish the second volume before the Spring Festival, and we will be able to finish the third volume by July next year.
Scientific and industrial forces: Since the end of last year, has the progress of the general model affected your writing of the third volume?
Lin Jun:The logical design of our three books is that the first volume writes about the academic origins, clarifying where the source of Chinese artificial intelligence is, and who the grandmaster of the founding school is;The second volume is to spread out the leaves, more to write the story of educators, this stage is at the trough of global artificial intelligence, the grandmasters began to pass on**, most of these students are now the heads of major universities, the directors of major scientific research institutes, and the leaders of various systems;In the third volume, we believe that we should write about the past and the future, and talk about the progress of artificial intelligence industrialization, because after the rise of the Internet in 2008, there began to be algorithmic recommendation engines, machine vision, autonomous driving and other industrial applications.
One of the reasons why our writing came to a halt in 2021 was that we found it a bit difficult to finish this series.
At that time, although various application scenarios such as voice, semantics, machines, security, and autonomous driving had been derived, it was only everyone's investment in some cutting-edge applications in the process of AI industrialization, and there was also a certain overestimation and bubble in the capital market.
Originally, we didn't know how to end, we couldn't write an industrial history, and finally wrote about the industry's chicken feathers, and in the second half of 2022, GPT appeared, and we could end up with the new industrialization brought by the GPT model. At the same time, artificial intelligence has developed rapidly in the past two years, and we will write another chapter for the future.
Science and Engineering Power: What types of readers did you envision when you wrote?
Lin Jun:Again, we want to attract researchers and practitioners of artificial intelligence first.
First of all, the core readers should be those who do research, especially some young friends who are new to the industry, and should understand the context of academics and disciplines. Secondly, we want to attract AI practitioners and help them understand the development of scholarship, and our target is basically these two types of readers.
Science and industry force: Academician Gao Wen wrote in the preface that the development of artificial intelligence in the world and China for decades is a spiral forward process, and now that artificial intelligence is on fire, many researchers and practitioners do not know the history, there are many plausible, what do you think of this problem?
Lin Jun:In the past few decades, AI has not been a dominant science.
I studied artificial intelligence, majoring in image processing and pattern recognition, now called computer vision, in fact, it was difficult for us to find a job when we graduated, one direction was to go to the public security bureau, because there are criminals to arrest to take pictures and identify;The other direction is to go to space systems, do data transmission and remote sensing, these directions, there are few suitable job opportunities.
So in our opinion, artificial intelligence has been at a low point in the industry for a long time. We've been through the lows and highs completely, so we're not so pessimistic at the lows, and we're not so optimistic at the highs.
Now the hype and advocacy of artificial intelligence, some of them come from science fiction enthusiasts, they rely more on imagination, but we are from computer backgrounds, and we still have to talk more about whether these things are feasible, theoretically and logically, and whether they can be successful. At present, some ** hot spots have nothing to do with the computer itself, but the large model is the result of the improvement of computing power, which is feasible, although there are still some problems with the explanation.
The science fiction work "Three-Body Problem" We believe that the current artificial intelligence climax brought about by the improvement of computing power is another wave of spiraling, which will bring a number of new applications to the market, attract new investment in research and development in the market, and develop new computing power. Of course, there will be some bubbles in this process, and the continuous iteration of computing power itself is also a cyclical process, showing the characteristics of wave progress.
There are a lot of young people who are jumping in on the bubble process right now, and that can be problematic. We believe that the possibilities offered by new technologies should never be overestimated or underestimated.
Science and Industry Power: Chapters 8 and 15 of the book focus on the story of the "863-306 Project", Academician Gao Wen led the research of intelligent interfaces at that time, Academician Li Guojie led the "Dawn" series of computers, in the trough of global artificial intelligence, why can the research and industrialization of Chinese artificial intelligence accelerate to catch up?
Lin Jun:Admittedly, in the early days of the 863-306 project, we did want to "copy homework" from Japan.
In the mid-90s of last century, Japan vigorously promoted the fifth generation of computers, not only to accelerate the catch-up in artificial intelligence, but also to make strong efforts in memory chips. As a result, Intel gave up the production of memory chips and outsourced this business to TSMC foundry, which led to the rebirth of its own chip business.
Japan has been in the limelight for a while, and in China there is a debate over whether to refer to the United States or Japan. At that time, many people in China were inclined to the Japanese line, and two academicians, Gao Wen and Li Guojie, as young people at that time, played a role in determining the line.
This is the content of the eighth chapter of the book, at the time, Gao Wen was studying in Japan, Xi artificial intelligence, and he gave the opinion that there was a problem with Japan's artificial intelligence route. Li Guojie was studying at Purdue University in the United States at that time, and he was very familiar with Wang Kai, Huang Yingjie and others, provided a lot of information about the United States, and also put forward his own opinions.
Both of them have adopted a realistic approach, which has led to China's choice of high-performance computing (HPC) route, that is, the American route, rather than Japan's fifth-generation aircraft route, which is more in line with China's national conditions and will also bring more help to China's industry and application level, including the support of the 863-306 project to iFLYTEK later, which is a later story.
When it comes to computers and artificial intelligence, there are a lot of people who are immersed in their own business, but it turns out that with limited resource constraints, philosophical thinking and strategic direction are important.
Science and Industry Power: Before the planning of this book in 2018, alphago in 2016 also set off a wave of artificial intelligence boom, what do you think is the situation at that time compared with the present, what are the similarities and differences?You are also a practitioner, how do you see the current investment environment in artificial intelligence?
Lin Jun:In 2016, Leifeng cooperated with the China Computer Federation to create the "Global Conference on Artificial Intelligence and Robotics" (GAIR), which may be my personal metaphysical fascination with the year of "6", you see, the beginning of 2006 is the "iPhone moment";In 1996 Microsoft released Windows 951986 was the release of the first 386 computer, and it was ...... important moment for the PC
Therefore, we believe that the 2016 alphago incident is a breakthrough in the public domain of artificial intelligence, and it is likely to mark an important moment for artificial intelligence, and it has a small range of fans and investment pioneers, and some speculators have entered this field.
In 2022, ChatGPT appeared, when the industry has been developing for more than six years, the first batch of speculators have been squeezed out, and new investment and research institutions will join in.
Once it becomes an industry, it is very unlikely that it will go down to a new trough, and I personally believe that the AI industry will reach a high point around 2026 and then enter a sideways trend.
Science and Industry Force: At present, there are some large domestic enterprises and national teams in the benchmarking GPT, but there are also some ** that the domestic level is still far behind, how do you think it is objectively evaluated?
Lin Jun:By the end of the year, the country will be able to reach GPT-3The level of 5 will probably be able to catch up to the level of GPT-4 by the middle of next year, which is a relatively objective and unified perception at present.
Science and industry force: If you catch up like this, is the gap narrowing or expanding?
Lin Jun:In shrinking.
We had an interesting observation that we once met in Silicon Valley with Wang Meihong, the youngest Chinese E9 talent at Facebook at the time, who later served as the head of overseas technology at Kuaishou.
Wang Meihong, who was the first person in the world to build a recommendation engine, shared his observations with us.
In 2012, he worked on Facebook as a recommendation engine, and in 2014, he returned to China once, when everyone in China had only heard of these engines, and no one was doing them;When he returned to China in 2016, he found that the recommendation engine made in China already looked good;By 2018, he found that domestic recommendation engines had done better than foreign ones.
From this small example, it can be seen that as long as there is a driving force, in terms of application innovation, Chinese are leading, but in terms of innovation of underlying technical capabilities, it is indeed not as strong as the Americans.
Americans can make Android, Chinese can make WeChat, probably this logic.
TV series "Silicon Valley".Scientific and industrial strength: foreign artificial intelligence has first-mover advantages and ecological advantages. AlphaGo was out of the circle before, and now ChatGPT is out of the circle, plus NVIDIA's GPU chips, CUDA tool chain, TensorFlow and PyTorch's mainstream AI framework, a large number of developers, a strong English corpus, and a mature capital and financing environment, together constitute a strong AI ecosystem with first-mover advantages. Faced with such a situation, how should Chinese companies respond?
Lin Jun:It depends on how we define "Chinese enterprises". If the so-called "Chinese enterprises" are enterprises that were founded in China, use only original Chinese technology, and do business in China, then such a discussion may not be necessary. But if you're talking about Chinese-led companies with global presence, there's a lot of room for change and discussion.
Let me give you a simple example, after the introduction of the "Data Security Law", ByteDance is promoting the "first case" of cross-border data flow. The background of this matter is that the amount of domestic ** content is much larger than that of the United States, and the ecological richness is much stronger than that of the United States, and the amount of data on Douyin is 10 times that of tik tok.
In fact, the top technology companies in the United States are the same, and computing power providers like Nvidia also have a strong urge to sell products in the Chinese market. It's an interesting topic, and the giants of the United States are also facing a conflict between their national position and their own values, because they want to do global business, especially to support globalization.
Global giants such as Coca-Cola, Hewlett-Packard, IBM, Microsoft, Google, Facebook, Nvidia, etc., have transcended the conflict between countries, and their business is all over the world, with only 20%-30% of the business in the United States, and China's huge market is here, and they can't give up if they want to.
I am optimistic on this issue, as long as we are willing to adhere to openness and communication, with the diligence and wisdom of the Chinese, we will be able to use the world's resources for our own use.
Science and industry power: large models need so much computing power, it has obvious infrastructure characteristics, the marginal cost is decreasing, and there is a strong scale effect. The current situation of foreign development is that by multinational giants or giant alliances to provide such a public infrastructure for global use, as China, does it need a new type of *** system to compete with it?
Lin Jun:Supercomputers and computing power networks are undoubtedly the key infrastructure of the country, and their importance can even be compared to the core**.
It is true that China does not have the ability to take the lead in building this "nuclear **", but it is possible that we will follow the nuclear ** when others build it, and as long as there is communication and information transmission, it is impossible to stop the proliferation of technology.
Prof. Gao Wen led the Pengcheng Laboratory, which has been built around the mission of providing computing infrastructure for Huawei from the very beginning. It is a laboratory that combines scientific research objectives and industrial transformation, which is a new mechanism, traditional universities may focus more on scientific research goals, while Pengcheng laboratory is also industry-oriented, has its own engineering management capabilities, and concentrates the best talents in the country and the world, which can be said to be a representative of the "new first-class system".
Academician Gao Wen, director of Pengcheng LaboratoryScience and engineering strength: Chinese artificial intelligence development has the shortcomings of high-end talents, and at the same time has a large number of outstanding young people, how to view this situation?What advice would you give to young people who are interested in pursuing a career in the field of artificial intelligence?
Lin Jun:First, in terms of artificial intelligence, China has the world's best batch training system for higher education talentsSecond, there is a shortage of masters in China, which are two facts.
By the way, the American side, on the contrary to us, has a system of master growth, but not in terms of mass training of talents.
This has a historical origin, the world's artificial intelligence before this wave of momentum, has two ups and downs, China's artificial intelligence development started at the beginning of the second wave of artificial intelligence, so it is born a cycle later than the world level, which is why we now lack masters.
However, China has a unique talent training mechanism at the application level, and now it also has a rich application scenario that is unique in the world. Around 2016, the Machine Vision (CV) Four Tigers appeared in China, why did they first appear in China and not in the United States?In this matter, the labor cost of the United States is very high, and since 1979, China has cultivated a group of talents with a solid foundation and are good at solving scene problems, and the landing scenarios are also richer.
In the future, computing power will increasingly become a core resource in the computer field, and algorithms will become a new driving force in business. Therefore, if Chinese artificial intelligence practitioners want to make a difference, my suggestion is to choose the direction of computing power or algorithms, go overseas when they are young, see the world, and return to China for development after achieving certain achievements overseas, which is still a fast and efficient path.
Science and Engineering Power: Are there any regrets in the writing process of this book?
Lin Jun:We want to say something quite sentimental. In the course of interviewing and editing and publishing this book, some of the older gentlemen left us. In the history of Chinese science and technology, through our small efforts, we have left some precious historical materials of pioneers.
P.S. The (old) gentlemen who died during the compilation of "A Brief History".
Dong ZhendongHe is a well-known Chinese information processing expert and inventor of CNKI. He passed away in February 2019.
Chapter 15, "306 Getting Better", mentions the relevant contributions to machine translation.
thomas huangChinese computer vision masters, peach and plum all over the world. He passed away in April 2020.
There is a special section in Chapter 6, "Computer Vision".
Liu XuhuaHe is a pioneer in the field of Chinese artificial intelligence research and a professor of Jilin University. In 1994, he suffered a cerebral hemorrhage and went into a vegetative state. He passed away in May 2020.
The first chapter, "Symbolism and Mathematicians," mentions Liu Xuhua's promotion of early AI as Wang Xianghao's main assistant.
Bian ZhaoqiPioneer of pattern recognition, author of the textbook "Pattern Recognition". He passed away in March 2021.
Chapter 3 talks about pattern recognition and Chapter 7 talks about Tsinghua AI research.
Xu KongshiComputer software scientist, founding director of the Institute of Software, Chinese Academy of Sciences. He passed away in September 2021.
Chapter 1, "The Dartmouth Conference in China," mentions Xu Kongshi as a panelist at the 1979 CCF Conference at Jilin UniversityChapter 5, "The Spring of Computational Theory and the Rise of Computational Linguistics," mentions Xu Kongshi's help to Ma Xiwen.
Yu ShiwenHe is a pioneer in computational linguistics and the former director of the Institute of Computational Linguistics of Peking University. He passed away in November 2021.
Chapter 5, "The Spring of Computational Theory and the Rise of Computational Linguistics", mentions that Ma Xiwen presided over the work of the Institute of Computational Linguistics after he went abroad.
Liu KaiyingHe is a well-known scholar in the field of natural language processing and a professor at Shanxi University. He passed away in March 2022.
Chapter 5, "The Spring of Computational Theory and the Rise of Computational Linguistics".
Li SanliHe is an academician of the Chinese Academy of Engineering, the former director of the Department of Computer Science of Tsinghua University, and an academic leader of microcomputer structure in China. He passed away in March 2022.
Chapter 7, "Artificial Intelligence Research at Tsinghua University", mentions the role played in the early construction of artificial intelligence at Tsinghua University.
Sun JianHe passed away in June 2022.
In Chapter 6, "Computer Vision", Sun Jian is mentioned when Zheng Nanning is mentioned.
Liu ZhuoThe originator of machine translation in China and the pioneer of NLP. He passed away in September 2022.
Chapter 5, "The Spring of Computational Theory and the Rise of Computational Linguistics".
Uncle YangMechanical engineering expert, educator, former president of Huazhong University of Science and Technology. He passed away in November 2022.
In Chapter 9, "The Twists and Turns of Robots", it is mentioned that Yang Shuzi has contributed to the field of robotics at Huazhong Institute of Technology (Huazhong University of Science and Technology) as an academic leader.
Tu XuyanHe is a famous scientist in the field of automatic control and artificial intelligence, the main founder of the Chinese artificial intelligence discipline, and the former chairman of the Chinese Artificial Intelligence Society. He passed away in March 2023.
Chapter 2, "Preparation for the Chinese Society of Artificial Intelligence" and Chapter 4, "The Blowout of Expert Systems", both mention Tu's work and prominent position in the early group of AI researchers.