Cornell Professor John Hopcroft on Artificial Intelligence

Turing Award Winner: Humanity is entering the next era of information science, society will change

November 6, 2019

Author: Song Yu Lili Association. Editor: 邸利会. This article is a share of a WeChat post and was translated into English by WeChat. Source: https://mp.weixin.qq.com/s/ukK-8_u-pJBYbc_HWq5TEA.

Computer science is a popular profession at the moment, and it is also a young discipline with a history of more than 60 years. As a pioneer in computer science, John Hopcroft witnessed the establishment, growth, and prosperity of this discipline. In 1987, Cornell University professor Hopcroft and Princeton University professor Robert Tarjan won the highest award in the computing field for their Turing Award for "the fundamental contribution of algorithm design, analysis, and data structure." The 80-year-old Hopcroft is not only academically accomplished but also a model for teaching and educating people and serving the society. As early as a few years ago, he came to China to make suggestions and work hard to help China improve its education. In September 2016, he was awarded the "Chinese Government Friendship Award". On November 1, 2019, Dr. Zhang Hongjiang, Chairman of Zhiyuan Research Institute and Hopcroft discussed "Artificial Intelligence: Strategy, Research and Education". In the 40-minute conversation, Hopcroft shared his deep views on artificial intelligence, higher education in China, and talent development. "Intellectuals" compiled and excerpted the conversation between the two, as well as Hopcroft's personal biography, to readers.

1. The true mission of the university is to produce the next generation of talent

Zhang Hongjiang: You won the Turing Award in 1986. Many of the audience present were not born at the time. Would you please talk about the situation at that time?

John Hopcroft: At that time, early computer science was very simple. At that time, to decide whether an algorithm is OK or not, just run it in the computer and measure the time it takes to solve a problem. If there is another researcher who does this problem, propose another algorithm to solve the problem. It takes less time, but it doesn't really matter why (the time spent is small) . The second researcher used the same data as the first, but if it used some random data, it actually was slower. I think we have to use mathematical methods to determine why an algorithm is slower. I proposed the worst-case asymptotic analysis method, and then I developed some techniques, such as divide and conquer, to develop this form of optimal algorithm. After doing this, the teaching method of the previous algorithm was transformed into a real science.

Zhang Hongjiang: It’s good. When you teach computer science at Princeton, there is actually no such class. Where do you find information about teaching?

John Hopcroft: I think I should mention that when I got a (undergraduate, master, doctoral) degree, I didn't have a computer science major. My degree is electronic engineering. I am also employed by Princeton's electrical engineering department, but I tend to like electronics. On the other side of the computer science discipline of engineering, I prefer mathematics a bit. At that time, there were a few articles, and I developed these courses on the basis of these articles. Later, this course was widely adopted by other universities and became the theoretical class of that era. In 1967, I went to the newly established Department of Computer Science at Cornell University and started to develop my career in a department that is more relevant to my interests. This is also a better choice. I also have the opportunity to help build a world. Leading computer science department.

Zhang Hongjiang: What do you think about deep learning?

John Hopcroft: Artificial intelligence is like a toolbox with tools in it, and deep learning is an important part of these tools, at least for today. However, to some extent, deep learning is only pattern recognition in high-dimensional space. For example, if you train a neural network to sort pictures, give a picture that looks like a bicycle, but can't ride (the function without a bicycle), the model will still be considered a bicycle. In fact, I don't think this is kind of intelligence, because in the image classification, I can't extract the function from the objects in the picture. I think I have to wait for years to reach this point. When we really understand the function, we will enter into the next information revolution.

Zhang Hongjiang: From the perspective of the algorithm, what do you think is the problem of data and parameters?

John Hopcroft: Maybe I can talk about how I think about doing research. There are two kinds of research in the world, one is basic research and the other is applied research. The current state of artificial intelligence I have seen is applied research: we develop technology to solve important problems, and it is very successful. It is also important to promote the country's economic development, but the basic research is fundamentally different. Why do basic research? Basic research is purely because researchers are interested in the issues they study, and it is as simple as that. In the United States, there is no applied research in the university. We feel that this is not our mission. Our mission is to focus on basic research. Maybe I should change a word without a mission. The real mission of the university is to produce the next generation of talent. When we hired faculty members, they had a 40-year career, and we wanted to hire faculty members who remained active throughout their careers. The point we like is that it is curiosity. We hire someone, I hope, not to see what they are doing but to see if he is interested in it and to be active throughout his career. You may say that the United States is not blindly funding the basic research of the university. Is there no target? It doesn't sound like a good investment, but in reality, this may be the best investment the US has made. Because thousands of researchers are exploring in different directions, many people have no influence, but sometimes some people have done something to create a whole new industry, creating billions of jobs, billions of money. I think this is a very good investment, and I hope to stick to this.

2. Long-term investment in artificial intelligence

Zhang Hongjiang: Will there be progress in deep learning?

John Hopcroft: Deep learning has proven to be very effective over the years, but the tricky part is that we can't explain why. In this way, if you are a faculty member, you will encounter many difficulties in teaching, because what you teach students now is an experimental discipline. It is urgent to develop a theory that can explain deep learning, and why it is useful. One difficulty is that, for pattern recognition, if you have a cat series or a dog series, you don't have a mathematical definition of the series, then maybe the researcher takes a step back and creates a mathematically well-defined Categories that help prove the theorems related to deep learning. One of the reasons why I care about the development of theory is that it makes it easier to teach.

Zhang Hongjiang: What do you think of unsupervised learning?

John Hopcroft: It's important to understand how to do unsupervised learning or how to learn from a picture. I can tell a story about my three-year-old daughter. I have a book that reads literacy. I used to sit on the sofa with her and look at the picture and recognize what it is. Once I walked with her on the street, she pointed to, Dad, fire truck! She actually only saw the map of the fire truck once and recognized it. I always wanted to understand why, maybe people have learned how to learn. She had seen millions of pictures before looking at the picture, and she learned how to learn from a picture. This means that it is important to take care of multiple disciplines, such as understanding how the human brain works. There have been a lot of research on the human brain in the past 25 years, and perhaps knowing how the human brain learns will help the machine learn.

Zhang Hongjiang: After learning how the human brain works, can it be transformed into a mathematical model or algorithm?

John Hopcroft: I know very little about how the brain works. I have read some studies. It seems that the distance between the neurons in the brain is relatively long at the beginning, and then there is a transition phase in which the neurons are close. The other thing is that in the first five years of the brain, the nerves will create new connections, and then there will be a transition period, and new neural connections will not be of much use. You can learn from these and see if it can be used to train neural networks. Basic research has a feature. Thousands of people try various ideas, but often have little value, but one or two people will have very important rewards.

Zhang Hongjiang : Which two people are there?

John Hopcroft: This is not known in advance. I can talk a little about the US funding. When I submitted the application, the funding agency did not ask me to conduct research according to the application. The reason why they funded was to create the next generation of talent. If I feel that other research is important, they are happy. This can be traced back to the 1960s. When I finished the research project, I would not write a report, but the university would write a report explaining how the money was spent. The funding agencies also do not specify which directions will be rewarded. They judge the direction according to the next generation of talents.

Zhang Hongjiang: I am very curious. If you are a money manager, which direction will you put money into artificial intelligence?

John Hopcroft: To decide the direction, I think there are two things that are worth considering. One of the things is short-term. You want to increase the gross national product of China or Beijing and create jobs. Then you may have to help the company do things like face recognition, picture recognition, machine translation, all these important things. Things. But you still have to spend some money to produce the next generation of talent, but also focus on the long-term.

Zhang Hongjiang: Almost every major country now has its own AI strategy. What do you think the two US governments have done right in the development of artificial intelligence in recent years?

John Hopcroft: Maybe I should say that when I started my career, the government didn't even believe that computer science was important. The government seeks the advice of some physicists who tend to think that computer science is training programs. It seems now that it is very difficult to develop a new field. The old field didn't want to give up the resources they had, but the United States did a good job in this area and increased the amount of computer science funding in the National Science Foundation. The US Advanced Research Projects Agency also carried out very important funding. But as far as the AI ​​strategy you are talking about, from the US version, it is something very obvious. If you ask non-scientists what to do, it is a long-term investment in artificial intelligence. After I read it, there was no strategy to find out what was insightful.

Zhang Jiang: the United States will always lead in the artificial intelligence it?

John Hopcroft: I am not sure if the US can stay ahead of artificial intelligence because I don't think the investment is enough. I really think that when we enter an era of information revolution, the nature of society will change. I want to say that students tend to vote with their feet, and students realize that the future is the age of information technology. I believe that 10% of the majors in most American universities today are related to computer science, but the funds are not allocated according to this trend.

Zhang Hongjiang: What do you think about the development of artificial intelligence in China and the United States?

John Hopcroft: In this respect, China actually has many advantages. Taking mobile phones as an example, when the mobile network just appeared, almost all households in the United States have installed cable broadband, so we have not paid enough attention to it. China has invested a lot of resources to develop this industry, so it is much more than the United States. Leading in the field, for example, the Chinese do almost everything with their mobile phones, and the United States has not yet reached this level. I think in some places, China does not have the perfect facilities in the United States, and the rate of development is much faster.

Zhang Hongjiang: Autopilot faces some security problems, but some people worry that work is being replaced. What do you think?

John Hopcroft: It is very difficult to drive on the streets of the city. There are pedestrians, bicycles and cars. However, if there is an intercontinental highway in the United States, the car will drive in the lane and will not change lanes. It is easier to drive automatically. I imagined that the driver drove the car to the highway junction, and then the vehicle automatically drove, and then the driver took over when the speed was high. This problem is much simpler.

Zhang Hongjiang: I also agree that it is easy to drive a truck automatically, but why not do it?

John Hopcroft: I don't know the answer, perhaps because there may not be much "money" compared to carrying passengers in the city. In automatic driving, road coordination is a good development direction, perhaps allowing the road to transmit signals to the vehicle, then passing the vehicle to each other, sealing the road, not letting the animals walk through, etc. But we don't seem to be moving in this direction, but trying to solve a very, very difficult general problem (let the car drive autonomously) . China can easily set up an “automatic driving lane” for testing, but it is almost impossible in the United States.

Zhang Hongjiang: Now we see that especially in China, many universities have set up artificial intelligence colleges. What is your opinion?

John Hopcroft: I don't know why I have to create so many artificial intelligence colleges, but it may also be a government-restricted computer science-related major, but too many students want to learn this, so the university will establish artificial intelligence, software engineering and other majors to solve. these questions. In my opinion, we are entering a new era of information, and the university will want to establish a department like information science, like the previous engineering, science, and art. Because the profession in this field is too big, 5 or 6 departments are needed, and perhaps the School of Information Science can include these. This will be a big engineering, I think many schools are doing the right thing.

Zhang Hongjiang: But artificial intelligence also has ups and downs. If the winter of artificial intelligence comes, will these schools cut down on artificial intelligence?

John Hopcroft: Things change from time to time as things change. Take the electrical engineering that I studied and studied. It used to be a science about power generation and energy transmission. But in recent decades, radio and computers have been integrated into it. Electrical engineering is so successful because They are constantly changing, and artificial intelligence needs to do the same. If artificial intelligence is done the same as today, it may be problematic. I just believe that we are entering a society where information is critical. The Industrial Revolution has liberated our bodies. What we are doing now is automation intelligence. I don't think this process will stop.

3. Research publications are hurting science

Zhang Hongjiang: In addition to the Turing Award, you also have an important honor as the "Chinese Government Friendship Award." You have also given advice to our Prime Minister on education, and devoted a lot of time and energy to higher education in China. Why do you do this?

John Hopcroft: All along, I want to do something that will make the world better and benefit as many people as possible. Before I came to China, I worked in 15 different countries (Colombia, Brazil, Mexico, India, etc.) and helped several faculty and students, but never had the opportunity to influence their education system. The Chinese Ministry of Education has invited me to China, which is very different from the previous countries. The government hopes to improve the undergraduate teaching and education, and I will have an opportunity to have an impact. The governments of other countries are concentrated on other things. Efforts to improve education are not their biggest concern. I think the advantages of these countries are energy and materials resources, but these other countries also have talents in top countries. This is the promotion of national progress. Power. China, I think I know this. If you talk to the Chinese government, they will say that we must improve the education of undergraduates and enhance their talents so that the company can continue to expand and continue to make profits. This is an opportunity for China, which is why I am going to do this.

Zhang Hongjiang : What advice do you have for higher education?

John Hopcroft: I should mention that 20 years ago, children had to have a degree to get a job. The number of college students increased by 1 million a year. To continue to develop, China must be able to accommodate more students. 50 new universities are to be established each year. After being able to accommodate these students, we must improve the quality of the students. I think one of the obstacles is that in China, a principal is only about five years old because they are government civil servants and will transfer to another job after the term ends. In order to prove their achievements, he has to stare at the university international ranking indicator. However, these ranking measures are based on research and the number of articles published, and have little to do with the purpose of the university, that is, to create the next generation of talent. Therefore, I think the most important thing that China can do is to change the evaluation dimension of college principals. The principals I have talked to are world-class people, knowledgeable and talented. It is good to use undergraduate education as an indicator to evaluate their work. Another interesting thing is that I think the quality of first-year students in China's top schools is better than that of American universities. If Tsinghua University, Peking University and Shanghai Jiaotong University can improve undergraduate education, they can surpass Stanford, Berkeley and MIT.

Zhang Hongjiang: You have a great passion for education. Why?

John Hopcroft: I enjoy education and research very much. Maybe I can give some advice to young people who are just starting a career. If you want to be successful in your career, you must do what you like; if you are teaching students, don't just teach them what they have learned in the classroom, there are other skills. Narrow technical education can bring them the first job, but you should give them a broader education, history, sociology, etc., to promote their success and enjoy life.

Zhang Hongjiang: As a young faculty member, how to balance the publication of articles and education?

John Hopcroft: If an undergraduate finds me and says that I want to be able to do some research, I will advise him that research may not be the most time-consuming thing for you at this stage, unless you want to enter a PhD program. Because if you want to enter a good doctoral program, maybe you have to publish. In 1964, Princeton University hired me without a paper. Today I am afraid that even graduate students can't get in. Because they are concerned about whether I have been active throughout my career for a long time, not the research I have done.

Zhang Hongjiang: In the field of computers, many of us are first to become IEEE Fellow. After that, we will continue to do well and become ACM Fellow. But you can get the Turing Award first, then have other honors and titles. What are you doing? Good advice for young people?

John Hopcroft: Complete the minimum amount of papers you are asked to publish, and then focus your energy on those basic research. If you make a basic discovery, you will get the recognition you deserve. In fact, I think that the high amount of research published in the world today is hurting the development of science. If you study a new subject, there are too many articles to read, which invisibly increases the time for people to screen and retrieve.

A pioneer in computing

In 1939, John was born in Seattle. At that time, Seattle was still a small city with a population of just over 500,000. On the other side of Lake Washington, where Microsoft and other companies gathered today, it was still a forest. When I was a child, John often walked in the woods with his friends and saw a path that stretched into the depths of the woods. He was always curious and he often had to find out. John's family conditions are very poor, his parents have no culture, and he has not graduated from high school. In his early years, his father sneaked into the United States from the Canadian border, holding half of the minimum wage and living a frugal life. Fortunately, John's parents have a good relationship, never said that the other party is not good, and for the sake of John, I hope he can live a better life in the future, and hope that he can receive education and go to college. Parents are particularly caring for John, teaching him to swim and do all sorts of things. In reviewing this journey, John said that this early care and stable environment may be beneficial to children's brain development, and many of his successes are attributed to this. John’s best subjects after school are math, and the worst is history. The history lesson at that time was to list who was the general, what was hit, and when to play, but rarely explained why the world is so moving. In high school, John is still very focused on his studies, but he is not so nervous. The school is out of school after 3 pm, and he is free to control in the afternoon and evening. One of John's most impressed high school teachers is algebra and a football coach (that is, John's mathematics is taught by physical education teachers). But the teacher is particularly concerned about the students, especially the success of the students, and John is trying hard to prevent the teacher from being disappointed. In the interview, John mentioned several times that the most important thing for a good teacher is to care if the student can succeed. After graduating from high school, John had planned to go to the University of Washington to go to school, but a faculty member of the school told John that your school was not qualified and did not recognize it. John turned around and applied for the admission of Stanford University on the advice of the teacher. At Stanford, John finished his master's degree in 3 years, and he graduated from the doctor at the age of 24. At that time, he didn't need to spend so much time studying. However, at that time, Stanford did not have a computer science department. John entered electronic engineering and learned some wires and vacuum tubes. At that time, the physics teacher at the University of Washington had a computer program that didn't work. Ask John to go to the bug, which was his first contact with the computer. To this day, John still clearly remembers that it was an IBM 650, and there was something like a drum. At the time, there was no programming language, and John used only 10 symbols in assembly language. He also had a programming class. As a student in the electrical engineering department, John also took an experimental class, but he quickly realized that he was not good at playing with various physical devices. He is better at metamorphosis or theoretical things, such as information theory. After graduating from Dr. Stanford, John had planned to take a teaching position at the University of Washington. One day, he passed the door of Bernard Widrow, and Bernard was calling Edward J. McCluskey of Princeton University. McCluskey asked if there were qualified PhD graduates to be instructors. John went to Princeton after the interview. John later said that his life was often accidental and he did not make any special plans. When John applied for Princeton, Princeton also had no computer science department. He entered the Department of Electronic Engineering and was hired without an article, which is almost unimaginable today. At that time, the Department of Computer Science just appeared. In 1964, the Purdue University in the United States established the first one. After that, Stanford University and Cornell University also established the Department of Computer Science, and they are recruiting teachers. In Princeton, John opened the first computer science class. At that time, McCluskey asked him to teach, but there was no teaching material. On the basis of several papers, John wrote a book that will be used for decades in the future. Every computer science department will be used, which has a great impact on the field of computer science. Classic textbook. John wrote this textbook not only to summarize the research results at that time, but to write about the things he created and developed, almost setting the benchmark for the industry. There were not many students at the time, only six people, but they have achieved outstanding achievements in the future. After spending two and a half years at Princeton, John went to Cornell University in 1967. Until today, he is a professor at the school. The reason for leaving, John said, was that Princeton's Department of Electrical Engineering was relatively mature. There were ten qualified candidates when there was a professor's vacancy. The forward-thinking McCluskey tries to push people in computing, but because there is no subject setting, it is much harder, and it is hard to say that computer science is growing, it is best to enter a person in this area. There was some chance that John left Princeton when he hosted a series of seminars, but the budget was only enough for two outsiders, one of whom was Juris Harmanis of Cornell University. Juris Harmanis was the creative director of the Department of Computer Science at Cornell University in 1965. From Harmanis, John learned that Cornell is recruiting and assistant professors are 50% more expensive than themselves. John decided to go to Cornell. He later said that it is better to go to a department that understands what he is doing than to stay in a department that struggles to gain recognition. After Cornell, John turned to algorithmic research from the previous formal language and automata theory. He realized that computer science is a very broad field, and algorithms are especially important. He has studied algorithms such as divide and conquer, deep search, etc., which we all study today. During his vacation at Stanford, he met Bob Tarjan and shared an office with him. At that time, Bob was a doctoral student and was studying whether a graph was a floor plan. Together they developed a linear algorithm. This result is also considered to be one of the most important achievements of the two people's cooperation. John said that when someone stays in the office, it is not well isolated. Others have different opinions on the issues you are studying. Through discussion, the ideas will gradually become clear. At Cornell, John co-authored another book on very well-known algorithmic design and analysis. For the other two authors, Alfred V. Aho and Jeffrey D. Ullman, John said that they actually made a huge contribution, but the Turing Award seems to be only one, not fair to them, but helpless Is the truth. From 1964 to Ph.D. to 1974, John's first decade can be said to be fruitful. That is, before the age of 35, John has made quite a success. And a person often does the most important job early in his career. John analyzes that it may be because there are more time in youth. In addition, young faculty members are often older than graduate students, and they are in the career establishment period. The relationship is more suitable. Later, when John became "old", when a student asked him to be a mentor, he also suggested that they go to an assistant professor because the teacher-student relationship would change subtly. He might give some advice, but maybe not. Will "give together and suffer." After becoming a fame, John, who was a dean of the department at the age of 48, became a deputy director and director of the School of Engineering. John said that in fact he didn't want to be an executive before, but there were no more senior people in the department. However, after that, he felt that it was quite good and could exert a greater impact. John’s administration is also very impressive. Considering that the school is made up of departments, he has delegated a lot of power to the head of the department - after giving a certain budget, how many teachers to hire and how much salary to pay will be calculated by the department, as long as it is within the budget. How can different systems divide the cake? John noticed that some department heads are better than other department heads, and outstanding department heads will also recruit outstanding faculty members, so in terms of funding, John will be inclined accordingly. In addition, John will also make a comprehensive consideration based on how many classes the teacher teaches and how many students. It can be seen that he has always attached importance to teaching. In 1992, the 53-year-old John was appointed by the then President Bush of the United States as a member of the American Science Council, and the American Science Council administered the National Science Foundation. John said that being so young in such a young position is a good example of how lucky he is in an emerging field. He taught the first computer science classes, which made him a more senior computer scientist at a young age. When the US government looked for the most senior computer scientist, no one was ahead of John. John said, "If I am in the field of high-energy physics, I am afraid that it will be my turn to wait for senior faculty to retire today. I just want to say that computer science is constantly changing. A young person should not stick to the old field. Moving to a new direction will quickly become a senior person." Eight years later, John, who ended his administrative work, returned to the position of professor. After a year of concentrated research, John was able to pick it up again and return to the path of research. He started doing social networking and then started machine learning until now. In fact, he became an artificial intelligence researcher. Since 2002, John has further extended his footprints to Brazil, Chile, Colombia, India, Mexico, Saudi Arabia, Vietnam and other countries, and of course China. On September 29, 2016, John accepted the "Chinese Government Friendship Award" at the Great Hall of the People. In the past few years, John has helped Chinese universities improve their undergraduate education, improve evaluation mechanisms and recruit, and train students. He used to lecture in Shanghai Jiaotong University and Peking University during the summer and winter vacations, recruiting students to Cornell for internships and research. He also serves as a consultant to advise on undergraduate education in China. The most recent thing is that in May 2017, he was appointed as a visiting professor at Peking University, director of the Frontier Computing Research Center, and presided over the opening of the Turing class. Also in China, John feels that his role and educational philosophy have been well developed compared to other countries he has been to before. He likes to stay in China and help China. The story of him and China continues.