Virtual reality (VR) and artificial intelligence (AI) have opened the door to “hyper-reality” for mankind and are leading the next wave of technological changes. The communication between humans and intelligent machines will become real if humans can directly communicate with robots rather than interact with computer programs. What interesting changes will be brought about if VR and AI are applied at the same time to humans’ everyday life?
Andrew Yan: Good afternoon, everyone. In this session, we are going to discuss the impact of AI and VR on us. I’m an investor but I majored in aircraft design for my undergraduate study. Back then, when we talked about VR, we meant simulation technology, a pilot sitting in a driving simulator, for example. So VR has been around us for a long time. The four panelists with us today have been in the industry for many years. On my left is Mr. Edward Suning Tian. We have been friends for over 20 years. He is an investor but more importantly he is the founder of AsiaInfo. Mr. Tian started his own business about 20 years ago when I was working in the United States and the company has been growing in size ever since. I believe AsiaInfo is bound to be one of the world’s leading providers of software and IT services to the telecommunications industry. I have no doubts about that and probably it will not take a long time. There is Kai Yu, founder of Horizon Robotics. Previously employed by Baidu, Yu is now in the robotics business. Ying Huang specializes in big data and AI at Lenovo Corporate Research & Development. Yun Xie is the Chief Scientist of Digital China. I was the largest shareholder of Digital China for more than a decade until this year.
So we have with us today researchers and experts on robotics. We are going to talk about three issues and there will be a 15-minute Q&A session in the end.
About the first issue. Big data and AI are actually nothing new to us. We all know that something big happened in the first half of this year – AlphaGo beat a Go genius of the world. It happened ten years earlier than academia had predicted. VR has developed for several decades but it was suddenly brought into the spotlight this year. Why did all these happen this year? Let’s begin with Edward Suning Tian.
The Emergence of AI
Edward Suning Tian: I think new technologies have shown great vitality for evolvement ever since the Internet was born. AI derives from cloud computing that emerged five or six years ago. The mobile Internet brought us a vast mass of data and thus great computing power so AI came into being. We can see how computer has evolved during this process so I think the main factor here is cloud computing that has paved the way for computing and the training of data processing makes brains smarter.
Yun Xie: In my view, we should thank Intel for making integrated circuits featuring a high level of integration, large processing capacity and very low cost. Let’s take a look at two facts. I was in college when the People’s Republic of China was celebrating its 35th birthday in 1984. For the first time, a dozen cameras were installed on the utility poles in the Tian’anmen Square. A camera would cost more than RMB 10,000 at that time. High-definition cameras are very cheap today so a huge mass of data are captured. This is one fact. Another fact is about processing capacity. The processor speed of mobile phones we use today is several billion instructions per second but the speed of China’s first ever supercomputer Yinhe-I was just 100 million instructions per second. It was developed in 1980 with countless components from the United States. This fact tells us that the processing capacity we have today is staggering. That’s why we are talking about big data. VR also requires exceptional processing capacity.
Ying Huang: I agree with them both. We now have an enormous mass of data, which is also a marked change. Take image recognition as an example. To teach a robot how to identify an image, we could find hundreds or thousands of pictures at most in the past but now it is easy to get hundreds of thousands of or even several million pictures. As you know, Google uses several million pictures to teach a robot how to identify a cat. Without such a large quantity, it would be impossible to do that despite great computing power. Besides, the entire framework of deep learning was improved by Prof. Geoffrey Hinton at the University of Toronto in 2007 so that the algorithms can be used to process a huge mass of data. It also makes deep learning operable. So I would say there are three factors, i.e. a broad mass of data, great computing power, and improved algorithms.
Andrew Yan: You did make some important points. AI is closely connected with the development of big data while the latter is clearly related to the computing power of computers. Two factors are essential to big data. One is a huge mass of data; the other is algorithm. Kai Yu is a practitioner. You left Baidu and started your own robotics business at the end of last year. What were you thinking back then?
Kai Yu: Thank you, Mr. Yan. I have been working on AI for 20 years and have witnessed the whole process of AI’s change from something very far to something very good. I joined Baidu in 2012 and founded the Institute of Deep Learning (IDL). I spoke at Tsinghua University on several occasions. Deep learning was unknown to most people at the time, even to university professors but now everybody is interested in it. As the three panelists said just now, three factors come into play. The first is data which can be generated anywhere anytime. The mass of data makes many impossible things possible, such as autonomous driving. The second is computing. Intel has officially announced that Moore’s Law is dead despite its efforts to advance it over the years.
Earlier we talked about a hi-tech company Nvidia whose stock price has soared from USD 20 in September last year to USD 100 today, up four times in about one year. One of the company’s founders is Chinese American. In the beginning, the company focused on designing GPUs for the gaming market. Back in 2012, we at Baidu came to realize that GPUs are suitable for the algorithms of Deep Neural Networks (DNN) but Google didn’t. I saw an interview with Andrew Ng, Chief Scientist of Baidu. When asked about why he joined Baidu, one of the reasons he gave is that he finds Baidu is more flexible than Google and one can buy GPUs from Baidu as many as he wants. Do you know how sought-after are GPUs when they have wide application? Chinese companies would send someone to the United States to buy GPUs which are in short supply now. If you ask me whether it is wise to buy Nvidia’s shares now, I would say it is as far as the technology is concerned. The value of the company rose from USD 10 billion to USD 50 billion and will surely exceed USD 100 billion.
Computing power bears more on AI than big data. Take a look at the historic turning point marked by AlphaGo. It is actually about analysis of “small” data instead of big data. The computer is trained to reach Level II of a professional Go player by analyzing the competition data of many players on a website. The leap from Level II to Level III is not based on analysis of big data but GPU’s computing power. So I would say there will be a shift from big data to “big computing”. Whoever has greater computing power has better chances of making money.
Andrew Yan: You talked about the relevance of AI’s emergence to computing power and data processing but I think the advent of mobile Internet is also very important. The biggest change that has occurred during the transition from traditional to mobile Internet is that a large number of people who were not Internet users such as taxi drivers, police officers, soldiers, government officials, chefs, etc. now rely heavily on the Internet. For example, if you use the App Didi Dache to take a taxi, you will find the driver has two or even three mobile phones. I even found a driver in Hong Kong who uses six mobile phones with six different Apps for different groups of people. The penetration of mobile Internet has provided strong support for the collection of data.
Everybody knows that AlphaGo has beat a Go genius but we should also know that IBM Watson does better than over 95% of doctors and lawyers in terms of making diagnoses and giving legal advice. In other words, if someone goes to the hospital, there is a 95% probability that Watson’s diagnosis is more accurate than the doctor’s. Then comes the question: how much will our everyday life and professions be influenced by highly developed AI? I would like to hear your thoughts on this issue.
The Changes Brought by AI
Edward Suning Tian: Let’s take a look back at the history of AI. The advent of computers makes the job of typists less important. Many workflows have been automated but what cannot be automated is the knowledge and experience that are most important to humans’ life. That’s why we see the practices of lawyers, doctors and teachers that require the build-up of knowledge and experience are standardized but not automated. In the future, AI means what used to be done based on experience can be done better with the help of robots. I’ll give you an example. I invested in a company called LinkDoc two years ago. What this company does is to collect chest CT scans. To date, it has collected 300,000 scans that show various information about early signs of lung cancer. The founder’s father is a very good lung cancer doctor. As he told me, it took 20 years of training for a doctor to develop the ability to tell the early signs of lung cancer but now based on deep learning of these 300,000 chest CT scans machines can make diagnoses as good as those given by excellent doctors.
With growing computing power and more and more data to process, at least many early diagnoses can be done by machines today, which is more efficient and accurate and less expensive. Such an application has profound influences. The greatest contribution of Industrial Revolution is the prolonging of human life thanks to the invention of antibiotics and new chemicals for boosting immunity and other purposes. Information technology has been developing for four or five decades but its impact on human life is much less compared with that exerted by industrialization. In the 20 or 30 years to come, robots, AI in particular, will have fundamental impact on our life. Machines will take the place of humans in doing jobs generating a mass of data and requiring experience and knowledge that could not be passed on, such as healthcare, education and transportation.
Ying Huang: Speaking of AI, people may think of robots or humans-related intelligence. In fact, AI will probably produce greater impacts on our life or other aspects. The key element is that AI has a special quality – it can process a vast mass of data, translate data into insights, and find solutions to problems that cannot be solved otherwise. AlphaGo is an example. We all know that it has beat a Go master based on training. After that, similar algorithm is applied to Google’s data center for power saving. As a rule, it would be amazing if power use of a data center can be reduced by 5% or 10%. As we know, the cost of data center in power use is considerable. AlphaGo helps Google cut power use by 40%. It is usually very hard for experts to find a good solution since many things are at stake and the amount of data is too huge but AlphaGo did it by combining deep learning and reinforcement learning.
AlphaGo is also used to predict the probability of blindness caused by diabetes. Experienced doctors may make a less than 10% prediction but AlphaGo can give a 90% or higher. What AI can offer is something that cannot be reached by a human even after working hard for a lifetime so AI can do better or far better than humans in a particular field. This will make our life better. There is no need to worry about being defeated or taken over by robots someday. On the contrary, they will serve humans better and help improve the quality of our life.
Andrew Yan: About AlphaGo, I want to add something. An engineer at Google told me something astonishing. AlphaGo has read more than 25 million books so it is impossible for any human to beat it. This is self-learning.
Kai Yu: It is self-improvement based on virtual data generated in a virtual world. This is quite appalling to some extent.
Andrew Yan: Digital China has been doing something over the last seven or eight years that may influence every Chinese’s life. This is a project called “Digital City”. Mr. Xie has been running this project. Tell us something about it.
Yun Xie: Computer is in essence a machine of logic programming so any problem in our daily life that can be changed to an issue of logic programming can be solved by computer. Besides, with growing processing capacity, computers can surely do better than humans. Computers will take the place of humans in addressing problems that boil down to logic. For example, a simple application is the writing of a news report which does not require creativity but basic logic. As far as “Smart City” is concerned, I think we are trying to leverage the great power of computers to serve humans in an all-round way. We are trying to develop the killer applications. You may buy everything at Walmart but that only addresses one of your many needs. We are seeking to extend the scope and provide comprehensive solutions for different clients.
Will AI Replace Humans?
Andrew Yan: The development of AI has also provoked philosophical thinking, including fears about humans’ survival. Some philosophers and scientists have expressed their concern that AI may drive the self-alienation of humans in the future. In other words, the AI we have created may turn into formidable adversaries that will eventually wipe out humans. This issue has grabbed much attention. Another question has also been raised. Will humans be still needed as AI develops and what’s the biggest difference between humans and AI? I’m a venture capitalist so I’m more concerned about whether AI can be creative. If AI can be creative, there will be no room for artists or venture capitalists. We have seen some challenges. For example, many people are working as financial advisors. If we apply AI in the finance department, I’m afraid many people will lose their jobs because AI will surely do a better job than humans in this regard. As we know, Google’s driverless cars will be put into operation in 2018. Germany does very well in autonomous driving. I was shocked by their autonomous parking technology in my recent visit to Germany. Autonomous driving is developed based on a huge mass of data. There is one accident about every 100,000 kilometers in the world. By accident I mean death or injury. If autonomous driving becomes a reality, we may only see one accident every 10 million kilometers. When that day comes, the car insurance industry may face a real challenge. Can you share with us your thoughts about the challenges that AI will pose to our future, our life and our professions? Will there still be artists? Will we lose our jobs?
Kai Yu: Speaking of autonomous driving, it is actually about embedded AI solutions. A key issue here is to enable decision-making based on computing without access to the Internet. Suppose a child suddenly appears in the road, if decision-making requires access to the cloud, the response will probably be delayed by network instability. Given that, a main direction of autonomous driving is embedded AI solutions. The German companies you mentioned earlier and some Japanese companies are all our clients. There is no doubt that machines will do better than humans in the area of autonomous driving and many other areas. When it comes to making rational decisions, machines do better because humans have emotions.
Artists will not disappear because art is not about objectivity but about resonance. Take piano as an example. I heard a story. A Russian pianist lost all of his loved ones during World War II. He held a concert on the ruins of war and his music moved everyone present to tears. Can you imagine a robot playing the piano or being moved to tears? Art is not rational. AI can do very well in all areas of rationality.
Edward Suning Tian: I’ve also been thinking about this issue. Humans have invented so many powerful things, from steam engine to atomic bomb. The history of human civilizations tells us that we can always make better use of our goodness, which remains the biggest feature of humans. As for the impact on professions, five years ago when cloud computing had just emerged, Deputy Governor of the People’s Bank of China Wen Xie told me why commercial banks are still needed since cloud computing is so powerful. The central bank needs commercial banks because it is impossible for it to manage so many accounts. If every account can be directly managed, the central bank can directly manage every client. I thought it was brilliant. Now we have powerful cloud computing and software but his goal is yet to be reached. There is still a long way to go before the technology is fully applied to the industries. Many factors are at stake, such as expertise, change of mentality among industry professionals, and change of regulation, about which investors or technology entrepreneurs seldom think. In conclusion, I think AI is an important means to advance the progress of mankind. Many workers will be replaced but humans are good at learning. We will learn to get along with AI and create a better life.
Ying Huang: There is some misunderstanding about AI. In fact, most of the things that AI does are based on simulations of human brains. We talked about deep learning earlier. Artificial neural network is a very hot topic right now. It is actually a simulation of human brain’s neural network. Since computers have great computing power and can do very well in a particular aspect, you may feel that they are more powerful than humans. Many countries including China, the U.S. and those in Europe are doing a human brain project which is designed to make computers operate like human brains in five to ten years. We are still very far from this goal. It is even impossible to simulate a mouse’s brain. Under such circumstances, I think it is still early to draw the conclusion that AI will beat human brains. I do believe that in some fields computers have more information and do better than humans but it does not mean that computer programs are better than humans. I still think AI is a tool for humans and will not defeat humans in the foreseeable future.
Yun Xie: First of all, AI will surely change human society in a radical way. I would like to clarify a concept. I do not think we are experiencing some industrial revolution. We are actually making the shift to a new form of civilization from industrial civilization. Mankind experienced the shift from primitive to agricultural and then to industrial society. Now we are moving towards a brand-new information civilization. All aspects of human society will undergo changes during such a process. I agree with Mr. Yan’s view that new professions will come up when old ones die.
I would like to say something about how far AI can go from a scientific point of view. We are not able to construct a set of rigorous theories to explain how intelligence comes into being. We cannot explain how human brains work. In fact, it is impossible to build a system of theories independent of human brains so that we can create a system that has similar functions like a human brain. In the foreseeable future, it is impossible for machines to reach the level of intelligence that humans have. But I do agree with Mr. Huang on the point that machines do better than humans in solving many problems.
Andrew Yan: It is such a relief to hear all you have said. It seems I do not need to change my profession and be an artist, at least for the next few years. Maybe the safest job to do is to be an artist. After all, computers cannot beat humans when it comes to non-rational thinking. We have discussed a lot today and it is apparent to everyone that technology advances have changed our way of life before we know it. For example, more than half of the people here obtain information or communicate with friends with mobile phones, which was unthinkable five or six years ago.
This morning we talked about one American company called “Impossible Food”. It can already make eggs in laboratory. I cannot tell the difference between artificial and real eggs from taste. The company has also made artificial beef but I was afraid to try it. Now 40% of the farmland is used to grow plants most of which are to feed cows and pigs. If all the meat can be directly made in the future, I believe most farmland will be turned into gardens.
Over the past 20 years, many new professions have been created and many have died. Programmer was not even a profession when I was in college but it is now. Technology is changing the life of everyone but we may not be aware of that.
Q&A
Student 1: Thank you. My name is Yifan Xu. I’m a medical school student. I have one question for Dr. Yu. As we know, medicine is not simply about making rational judgments. It also involves moral or ethical issues. As you said earlier, machines will surely beat humans from the perspective of technology but they cannot address issues beyond the boundaries of rationality. In fact, this is just one aspect. There is also the issue of ethics. How do we deal with that? Robots will find wider application in our life. How should we respond to the challenge?
Kai Yu: In the area of medicine, the decisions concerning ethics should be left to humans to make. This is called “AI-based assistant decision-making system” or man-machine collaboration. In many cases, the machine makes the judgment and we do the following. It has to be like this. Many diagnoses are already made by machines but in some special cases that involve ethical issues the job is still done by humans. Machines will do the regular work while humans will do something more creative but machines’ capacity to make rational decisions will continue to grow.
Andrew Yan: In fact, the medical, legal, and consulting professions involve the gathering of data. Many conclusions are drawn based on big data. People in these professions are most likely to be replaced by AI.
Kai Yu: Speaking of ethics, take autonomous driving as an example. Suppose you have to choose between hitting an elderly person and hitting a child. In this case, the machine can calculate the probability of death or injury for the elderly person and the child but it cannot decide on who to hit. That is the boundary between humans and machines.
Student 2: Mr. Yu, you mentioned earlier that some of your products have been exported to European automakers. What are the products? I think this is interesting because we usually need to import components from Europe. Can you talk more about those products?
Kai Yu: Our clients include BMW and some Japanese automakers. They are very interested in China’s intellectual resources for the development of AI. Why is that? Because the world’s two largest talent pools for AI are the United States and China. We have the largest Internet companies that have big data, big computing platforms, and all the applications. China is actually ahead of Europe and Japan in terms of talent development for AI. Due to China’s special industry background, we have some rare resources in some fields. Members of the team are all former employees of Baidu’s Institute of Deep Learning. We did not do much publicity. The companies came to us, asking about possibility of working together on autonomous driving.
Student 3: My name is Jiahe Huang. My question is about recommendations by search engines, shopping sites, etc. according to user preferences. Our preference information is collected by the sites without our knowledge. I think this is to some extent violation of privacy. What do you think of the balance between respect for privacy and the collection of data?
Edward Suning Tian: I think we talked about this new challenge. Before the Industrial Revolution, no clear definition of property rights would be found and there were no banking, property rights, and insurance systems. I believe there will emerge digital property rights, digital insurance and digital banks in the future. Everything will be in place, be it regulatory framework, ethical standards or legal framework, as along as digital assets are recognized.
Andrew Yan: Your question is not that complicated. You will receive recommendations after buying books at Amazon or shopping at Taobao several times. The core of AI is the ability to learn by analyzing data. But privacy is indeed closely related to ethics. Any other questions?
Student 4: I’m Sen Li. Mr. Yu, I have a question for you. Machine learning is usually to design a specific algorithm or model for a particular field. What’s your take on the universal AI model? In deep learning, the training of neural network is still weighted training. It is rather hard to design a networked structure or approach with machines but human brains do have such variability. This is still about shifting from particular way of training to universal model. What’s your comment on that?
Kai Yu: This is a very profound issue concerning memory learning and the essence of AI. I was actually discussing this with Prof. Zhao when I was in Hangzhou the day before yesterday. Now AI is not universal. Give you a simple example. AlphaGo won in the 19*19 mode but would probably not if another game in the 21*21 mode were held the next day. Humans are much better than machines in infering other things from one fact. Now machines can address specific problems in a specific context. The question is how to make them more universal. I think we should change the reliance on data and allow computers to create and go beyond the context. I read an article in Science which is groundbreaking. It was an article published in 2006 about deep learning, or small-sample data learning which I believe is the path to universal AI. It is to imagine how to solve problemes in a new context or environment. About network learning. It is a typical issue in machine learning. It is so hard that very few people would touch it. It involves discrete optimization which is actually an issue of combination. I think future efforts should focus on addressing low power consumption and computing power.
Student 5: I’m a student from Qingdao No.2 High School. My question is about chemistry. Since AI has great capacity for processing data or making predictions, is it possible to predict the movement of a particular molecule? Do we have the technology now? If not, what’s the challenge in developing it?
Ying Huang: I will give you an example. Founder of deep learning Prof. Geoffrey Hinton has a student who knows nothing about chemistry or pharmaceutical studies. He took part in a contest organized by the famous US pharmaceutical manufacturer Merck. The contestants were asked to find the best drug to treat a disease. Prof. Hinton’s student won the contest using the methods of deep learning. Deep learning has its strength in finding the patterns from a mass of data so AI should be able to help scientists find better molecular structures or drugs. It is absolutely possible.
Andrew Yan: I’m an outsider but I have read some books out of curiosity. Will AI pose challenges to mankind or even eventually lead to the doom of the human race? I think we are still very far from this day. Now AI relies heavily on data and scenarios. Human brains are capable of generating new data based on past data but computers still have a long way to go in this regard. I agree with what the panelists said. AI will not become the enemy of humans or wipe out the profession of artist. I’m glad to know that because my job is safe to keep now.
Student 6: I’m a native of Hangzhou. Now I’m working on my graduate program in Beijing and I find smog is a serious problem here. My question is about how AI can help us solve problems around us, like looking for a lost senior or child. Mr. Xie talked about digital city or smart city. Is it possible to add some new functions to transportation or other areas that can help us more?
Yun Xie: Smart city involves a wide range of aspects but the most important is that the whole city operates in a transparent way. As the network improves, I cannot promise that all the problems will be solved but marked improvements will be made. We will see fast progress in AI-related recognition, human recognition, etc.
Andrew Yan: I think the time is up. I will sum up briefly. It is good that AI is changing our life unconsciously and bringing us improvements. However, many of the problems cannot be solved by AI alone or it will take a much longer time to solve them. I often think about the possibility of capturing corrupt officials with the help of AI or taking away their chances of promotion. We are still far from that but the discussion can go on. In short, I’m confident that AI will lead us to a better future. Thank you.