A.M.Turing Award winner and AI pioneer Raj Reddy: says goodbye to 996; AI makes mankind work only one hour a day
Introduction: Raj Reddy is not only a pioneer in the AI industry, but also an academic mentor of many great talents. Recently, while participating in the activity jointly organized by the International Knowledge Centre for Engineering Sciences and Technology under the Auspices of UNESCO (IKCEST), China Knowledge Centre for Engineering Sciences and Technology (CKCEST), Baidu and Xi’an Jiaotong University (XJTU), Raj Reddy, tenured professor of Carnegie Mellon University (CMU), shared with us many interesting new ideas.
Raj Reddy is a Member of the National Academy of Sciences and the National Academy of Engineering, US, a Foreign Member of the Chinese Academy of Engineering and a tenured professor at the CMU School of computer science, where he served as dean for a long time. In 1994, together with Edward Feigenbaum, Raj Reddy was awarded the A.M. Turing Award for his decisive contribution in large-scale AI system research. He also established the computer science experimental class at Tsinghua University (often referred to as the "Yao Class") and mentored quantum computer pioneer Professor Yao Qizhi and other talents like Li Kaifu, Shen Xiangyang and Hong Xiaowen, etc.
Besides, Raj Reddy has launched the first robotics institute in the US, the CMU Robotics Institute, which is also the most prominent robotics research center in the world. In 1979, Raj Reddy and others initiated and founded the Association for the Advancement of Artificial Intelligence (AAAI).
Jointly organized by IKCEST, CKCEST, Baidu and XJTU, the 5th Baidu & XJTU Big Data Competition and The First IKCEST “Belt and Road” International Big Data Competition started recently. Professor Reddy came to Beijing as the IKCEST councilor and chairman of the advisory committee, and we had the opportunity to talk with the AI pioneer. Professor Reddy reminisced the development history of AI and held great expectation towards future technology.
The history of artificial intelligence
Synced: In 1994, you and Mr. Edward Feigenbaum won the A.M. Turing Award for your contribution of large-scale artificial intelligence ("AI") system, which defined and demonstrated the significance of AI before the conception of deep learning. Could you share with us the development of AI over the years?
Raj Reddy: We had developed certain systems which could study and classify different tasks back then, but none of them could deal with large amounts of data. Geoffrey Hinton, father of deep learning, conducted a series of research concerning backpropagation between 1981 and 1986 and published his findings in 1987. He believed the method would succeed; however, no one was able to prove that. In fact, the method to some extent imitated the working pattern of human cranial nerve.
However, deep learning cannot be proven workable for quite a long time, from 1980s to 2010. Hinton's idea was right from the beginning, we nevertheless did not have enough computing power to prove it.
And now, after nearly 30 years, we've finally built a computer with tens of thousands of times more hashrate, more memory and more data volume. Today's computing power is gradually approaching the ability of human brain and AI algorithm can do many things.
So here is the difference: human brain cannot make the best of big data, but neural network based on deep learning system can. That's the most fascinating part of AI. We are now approaching the new phase of scientific development. And all of these cannot be achieved without the development of deep learning in computer science.
Synced: How has winning the Award affected your life?
Raj Reddy: Actually it hasn't done much, although my life did change a little bit. Winning the Award means people will know that someone has brought up a new technology, which is a good thing. They might say you truly did a great job, but I don’t really care about that, to be honest.
Synced: How do you spend your leisure time?
Raj Reddy: Many people will choose to exercise or take a rest outdoors when they feel stressed at work, but I am not one of them. I like reading. There are hundreds of e-books on my cell phone.
Synced: Could you tell us more about the Robotics Institute of Carnegie Mellon University (CMU)? We all know that you initially established it.
Raj Reddy: We established Institute for Computer Research of CMU in 1979. We had realized that the academic field needed to undertake detailed research regarding automated system back then. We began to study autonomous vehicles in 1984, as well as drone and unmanned ship; however, progress in the latter two directions was less promising.
We launched a fully self-driving car in 1995, but the system we used was immature. It was not true deep learning. We used a great amount of data to train the algorithm, so that it can learn how to drive. The result was satisfying, the car moved.
It was a milestone. A car drove from Washington D.C. to San Diego, California. No one controlled the steering wheel most of the time except for only a few conditions when the car could not make a judgment. 99% of the time of the whole journey, the car navigated itself.
We kept challenging to drive across the United States automatically from 1995 to 2007. The first team that accomplished the task was led by Sebastian Thrun from Stanford University. It is no surprising that he was a member of CMU Robotics Institute ten years ago.
Today, many people and companies such as Google, Apple and Baidu conduct research concerning automatic driving technology. In fact, all the automatic driving technologies used today come from CMU.
Besides, we also studied robotic technology in manufacturing industry, hoping to realize the conception of light-off manufacturing, in which all the tasks will be carried out by robots. I've mentioned automatic driving above. Search-and-rescue robot is our third direction, for instance, we make robots responsible for search and rescue works during nuclear accident. Since working under radiation conditions is too dangerous for human being, robots can take over their responsibility. Such dangerous working environment includes the Chernobyl, Fukushima and other nuclear accident sites.
Today, the range of study of Robotics Institute is wider. Besides the above-mentioned three directions, we also study computer vision, making robots acknowledge surrounding environment. It is a very important technology.
Synced: Nowadays, many colleges and universities have set up AI programs at undergraduate level, and Chinese education authorities have also approved application from 35 colleges and universities to set up AI programs this year. What do you think of the trend?
Raj Reddy: It is not a bad thing of course. The emergence of a new field, whether robotic, biological, engineering or CRISPR, indicates an important scientific and technological development. Many new technologies surface these days; 3D print is one of them and is in great demand. However, even if you set up these programs 30 years ago, students would not buy them. They would prefer those programs that would make them easily find jobs after graduation. Thus, development-orientated education is the most desirable way to introduce new programs.
However, the problem we encounter now is the shortage of capable professors. We need to launch a Professor Training program, bringing together the educators and spending several months in training them in unprecedented new ways, so that they can better educate their students. Today's students are so smart, and they learn fast. They will make considerable progress with better leaning methods. Let’s see what the future holds.
This big data competition embraces more realistic scenarios, extensive data and computing power, serving as a rare opportunity to the students.
Apart from the competition, technological firms have brought their own experiences in AI R&D and implementation to colleges and universities, which will further benefit China's AI education at undergraduate level.
Synced: What suggestion do you have for young AI researchers? What do you want to say to them?
Raj Reddy: Thinking independently is the most important thing, yet you still need to follow your supervisor's advice. You should work hard on basic principles, which refer to traditional programming of the computer science course. They are essential. People learn from their experiences; you also need to learn from yours when programming.
Synced: For young researchers, what are the characteristics of being engaged in academic field and the industry? What's your advice on how to better integrate production, education and scientific research?
Raj Reddy: People make their own choices. Someone just wants to solve the practical problems in industry, which means, to the most extent, they tend to only deal with daily new problems instead of promoting the development of new technologies.
Being engaged in the academic field, you will face brand-new directions and unsolved difficulties. You might succeed and might not. But failure should not make you frustrated. That's how research works.
Undoubtedly speaking, in the AI field, the combination of academic circle and industry becomes all the more important. Many education organizations have launched cooperation programs with enterprises. You can study at colleges or universities at first and then work in a company for a while. Today's research is comprehensive and those methods prevailing 20 years ago are not workable anymore. Working in a company is the right choice, so is cultivating oneself in the industry. You can study in school for one year, then apply for an internship in a company and come back to school for further study after a few years.
Now some American universities have carried out this way of learning. But what I am going to say is how to become a qualified doctoral student. You cannot solve all puzzles of mankind within only six years of doctoral studies, so we must make some breakthroughs in small sub-fields.
Synced: If you are a young AI newcomer, which specific research direction will you choose?
Raj Reddy: Deep learning needs the support of big data; however, human being can find certain rules and make relevant conclusions through an example. Currently, we are not able to build a similar system, for there are so many problems remaining unsolved. I think I will start from this point.
Synced: Is automatic driving one of the biggest challenges AI faces right now?
Raj Reddy: I do not think automatic driving is a difficult challenge. As far as I am concerned, the problem has been well solved, thanks to the existing technology. The real question is people's attitudes towards it. We do not care too much about a car accident, as long as a human being takes the driver seat. However, when a car's automatic driving system goes wrong, it instantly becomes big news that people all over the world will know.
From a fundamental perspective, we shall not expect a computer to drive much well than a human being. Human being makes mistakes, so does computer, because both of them learn similar knowledge. Thus, the computer's mistakes need to be tolerated and its problems to be solved.
A perfect computer does not exist. And that's the problem right now. True perfection only exists in mathematics, physics and so on. Even among these theories, the theoretical basis of physics and mathematics are constantly evolving.
In my opinion, neural network field would not have encountered too many significant changes. Enhanced computing power has allowed us to do things we could not achieve before, exploring new knowledge that couldn’t be discovered by depending solely on the human brain. This is an exciting process.
Automatic driving, therefore, is not the biggest challenge faced by AI.
Synced: AI still has its limits. What might be the possible directions for future breakthroughs?
Raj Reddy: When humans conduct a research, they will use data and accept ambiguous interpretations with certain error rates. It means that we also make mistakes and computers must know how to deal with it. It is not a rare situation, but the results are far from being accepted.
The results need to be further processed by interaction and natural language processing. But most systems do not contain relevant modules to do natural language processing now, so if the situation becomes complicated in the end, we must extract useful information. Map is the best example on this respect. Human being invented maps 3000 years ago, as the results of the extraction of information like landmarks and roads. Humans can understand the relations between maps and the real view.
Now, we are trying to make computers do such extraction. Human being can extract information and we call it intelligence. Humans can also study, analyze data, tolerate errors, understand ambiguous explanations, communicate by means of languages and extract signals.
Not a single system can do such things today. It might take 100 years for AI researchers to accomplish these tasks. They would have to keep responding to such challenges for a long time in the future.
Social influence of artificial intelligence
Synced: You come to Beijing to participate in the first IKCEST International Big Data Competition. What do you think of the increasing competitions of big data?
Raj Reddy: It is a very interesting image data competition. If you want to find a railway station on a satellite map, you need to search a lot about the time schedule of the train in the process. If these two kinds of information can be further organized, as human brain always works, we do not need to take it as a railway station, but a location that many people will go to instead. Under such circumstance, we can do a lot more interesting things.
Universities are the cradle of AI talents. It is admitted that schools lack resources of data, computer hardware and application scenario. Such competition can seek for possible solutions. Data of remote sensing and user behavior, as well as the extensive computing power resources adopted in the competition are difficult for colleges and universities to obtain, and that is precisely the advantage and value of enterprises like Baidu to participate in.
Synced: Could you please envision the future of AI in education?
Raj Reddy: Unlike many other industries, AI requires a high degree of integration between industry and academic field. Speaking of AI, not a single entity can step aside, whether enterprises, colleges and universities or research institutions. With the deepening integration among industry, education and scientific research, AI technology will infiltrate every corner of our daily life, which indicates the overall development of AI in the future and the impetus it provides for the development of macro economy.
Synced: What do you think of China's AI development and those technological companies developing AI in China like Baidu?
Raj Reddy: AI has witnessed a rapid development in China in recent years, whose current form is AI-based service. Ordering food, express delivery and online shopping all contain AI element. Today, China's domestic society is cashless. Young generation uses Wechat and Alipay to pay instead of cash. Even Americans are still using a great amount of cash today. Other countries are not as advanced as China.
China was underdeveloped a few decades ago, but now she is more advanced than any other countries, which is very amazing.
Synced: At the convention held in Beijing last October, you hoped that the state would distribute free smartphone to its people. How does new technology serve the society?
Raj Reddy: In my point of view, the increasing smartphone coverage will cause network effect, which means, with more and more people using the network, the economic development level will be rapidly enhanced. Nowadays, less than 50% of people in the country own a smartphone, and some still use a feature phone. But a feature phone does not support many apps. Making calls only takes 2% of a phone’s function.
As far as I know, Chinese government spends about ten thousand RMB on each child every year. The figure is a little less in India and about five times higher in the United States. Seeing that number, you have to admit that additionally offering a cheap smartphone wouldn’t be a heavy burden. The benefits of a fully connected population through network far outweigh the costs and the resulting tax increases are also much bigger than the costs.
Synced: Will the future of mankind be a small group of superman and AI who accomplish all the tasks with most of the population being at a loose end?
Raj Reddy: I don’t mind if human being sits around. I thought this is what’s going to happen in the future: individual’s productivity has increased ten times and each person’s working days might be reduced to only an hour.
We all want to have what we do not have. When I was growing up, about 60 years ago, I never wore shoes until I was 15. Most people in rural India did not have enough clothes to wear. And it is a lot different today.
It is not because we couldn’t afford it, but because we lacked productivity; demand, therefore, became the highest. Today, everyone has various demands, including physical demands and spiritual demands. Humans need water, energy, food, medical care, education, sufficient communication, etc.
Even now these demands are not being satisfied. Perhaps only 10% of people have met all their demands. Thus, people need to work hard to improve their productivity.
However, with the development of technology, people might only have to work 20 hours per week, even ten hours. We do not need to do too much work anymore. On the other hand, economists indicated five years ago that even if we “pretended” to work 40 hours per week, our actual working time would be 15 hours only. We would use the rest of the time reading, making coffee or chatting. If you made an accurate calculation, the actual working time would be about 15 hours per week. He also postulated that this figure would reduce to 5-7 hours in the next 20 years.
It means that we would only spend an hour a day on working. Productivity development will also liberate many people from work.
Synced: How many years later will AI reach the level of human intelligence?
Raj Reddy: It is hard to say, maybe 1000 years? Or even more.
Human being invented characters and numbers 5000 years ago, starting the first information revolution. However, only a small group of smart people could learn during that time. Thanks to the invention of printing 1700 years ago, people were able to save knowledge in the form of books and more people could gain knowledge by reading. This process has achieved the democratization of knowledge, so that everyone can learn.
However, knowledge remained in books during this period. With the emergence of computer, we are embracing the third information revolution, when knowledge becomes dynamic. That is the significance of AI: if you hold certain knowledge and information, you can use them to generate new knowledge and that is how automatic driving came true. Computers can accomplish such tasks automatically and AI agents can accumulate knowledge automatically. Today, AI can read news and make abstracts, providing us with the tips we need.
Such things were beyond imagination in the past. Computer and AI make knowledge dynamic for the first time.