In the past three years, Shen Nanpeng, global managing partner of Sequoia Capital, was named "the best venture capitalist in the world" by Forbes, becoming the first venture capitalist in China to win this honor.

Li Feifei is the first professor of Sequoia Capital and co-dean of the People-oriented Artificial Intelligence Institute of Stanford University. According to public information, Li Feifei was selected as one of the "Top 100 Thinkers in the World" and won the "Global Influence Chinese Award". He is also an academician of the National Academy of Engineering and the National College of Medicine.

This time, Shen Nanpeng connected with Li Feifei to discuss the theme of "illuminating the dark space of medical care with AI".

China Investment Network extracted the wonderful views of Shen Nanpeng and Li Feifei at the first time.

Shen Nanpeng:

1. In China, we saw the application of artificial intelligence in diagnostic medicine. For example, in the diagnosis of lung and chest radiograph, more and more people see that artificial intelligence can help doctors, even very experienced doctors, and enable hospitals and the whole doctor group to share their previous knowledge accumulation.

Li Feifei:

The application of 1 AI smart sensor allows us to monitor the patient's turning and moving in real time, which is very important for nursing and medical treatment. As can be seen from this small example, AI can play a great role.

2. When we founded Hai about two or three years ago, we deeply realized that AI is not only a technical field, but also involves profound sociological and ethical issues. In this institution, we have made very detailed arrangements from research to education to policy research.

3. The purpose of AI index is to report the development of global AI fairly, justly and comprehensively. It records the changes and influences of AI from scientific research and education to industry, industry and commerce. 2020 is definitely a very interesting year, because there are some new trends in the development of AI in COVID-19. For example, we found that the application of artificial intelligence in drug R&D and design has changed significantly in 2020, which will have a huge impact.

The following is the full text of the dialogue, edited by China Investment Network.

Shen Nanpeng: Congratulations on your election to the National Academy of Engineering and the National Academy of Medical Sciences in 2020. Last year, you published an article in Nature magazine, using ambient intelligence to illuminate the dark space of medical care. So I want you to explain, what is the meaning of ambient intelligence? How does it illuminate the dark space of our medical treatment?

Li Feifei: First of all, thank Nan Peng and Sequoia for their invitation. It spans the Pacific Ocean, but I want to say good morning to everyone!

On this premise, my collaborators and I discovered ten years ago that artificial intelligence actually brought a new opportunity. It can help us collect such information through the perceptron. Most importantly, it not only collects environmental information and human behavior information, but also carries out intelligent analysis to let us know whether the patient's condition has changed and whether the behavior of doctors and nurses has affected the patient's rehabilitation behavior, which is very important.

Driverless driving has given me the greatest inspiration. It was in Silicon Valley. Ten years ago, Silicon Valley was the birthplace of driverless driving. As the director of Stanford artificial intelligence laboratory at that time, we found that the unmanned technology was produced by the integration of perceptron, AI algorithm and the whole system. This idea was put into the medical scene by us, so we came up with the idea of being a ambient intelligence.

Shen Nanpeng: In today's medical application scenario, where can this be applied? In China, we have seen the application of AI in diagnostic medicine. For example, in the diagnosis of lung and chest radiograph, more and more people see that artificial intelligence can help doctors, even very experienced doctors, and enable hospitals and the whole group of doctors to share their previous knowledge accumulation. What important applications do you see in other scenarios? What specific scenes do you think can be broken in the future?

Li Feifei: This is a very good question. What we have been committed to is not the substitution of human beings, but the promotion of human beings. We used many scenes in Nature magazine, such as scenes in hospitals and scenes at home.

We found that the most important point is mobility. The mobility of patients is very important for the prevention of bedsores and in ICU, but how to measure the mobility? If you put a sensor under the bed, it is actually difficult to detect. Now there is a way to invite others to see it. For example, let the nurse record the patient's rounds in the electronic medical record every two hours. But this is a very inaccurate and rough record. If AI smart sensor is used, we can detect the patient's turning and activity in real time. This information is very important for nursing and medical care. This small example shows that it can produce great results.

For example, some chronic diseases of the elderly, in fact, if treated in time, some problems that can be solved with antibiotics do not need to go to the emergency room. But how can we find out whether the old man is infected at first, or whether his heart rate and breathing have changed? Or he hasn't moved much this day, and his diet and sleep have changed. You can even see that he doesn't usually do his usual social activities. Where does this information come from? There are usually only two. The first one is the nursing staff, whether they are family members or domestic nursing staff, but this information is very inaccurate and unsustainable.

The other is wearable devices, which I think is a very promising technology. But wearable devices also have their own problems, especially for the elderly, wearable devices are not particularly popular. Moreover, many behavioral problems of the elderly are impossible to see like eyes. Through equipment and sensors, we can observe and continuously observe the behavior changes of the elderly and important medical information, which can be sent to family members and medical staff in time. Just like an old man with chronic diseases, he may only need antibiotic intervention once, and he won't have to go to the emergency room or hospital after two weeks.

Shen Nanpeng: The application of artificial intelligence in medical industry may be a long-term trend. The other is COVID-19, which is a short-term event, but how does this event promote the innovation of the medical system? When we human beings encounter such a disaster, of course, on the one hand, we have to solve the short-term pain, on the other hand, we also use this opportunity to promote the innovative utilization of the medical industry. What kind of experience can I share with you?

Li Feifei: COVID-19 has a profound influence on everyone here, whether it is personal, life or career.

Speaking of specific technical points, I think there are the following points.

First, telemedicine. As a patient who has lived in the United States for a long time and sometimes can only talk to my doctor remotely, I have been wondering why our telemedicine has not been widely used. Therefore, COVID-19 quickly popularized the remote application. So I think it promotes the development of all aspects of the whole ecology related to telemedicine.

The public health crisis you mentioned is "health information". Many people say that COVID-19 is not the first epidemic, but the first information scholar. Infodemic also means that all kinds of true and false information are spreading at high speed. This is a matter that has a far-reaching impact on technology and society. Many of my medical college colleagues have seen that technology has played a good role as well as a bad role in information dissemination, and the Internet has brought rapid information transmission. But AI artificial intelligence also brings the transmission of mis and dis information. Therefore, COVID-19 has had a far-reaching influence in all aspects.

Li Feifei: This is indeed a very important issue. As a scientist and technician, I have changed and grown a lot since I entered the scientific field 20 years ago. I didn't expect that the science I love so much will eventually become the driving force for changing society.

In this process, when we established HAI about two or three years ago, we realized a very profound problem. AI is not only a technical field, but also involves sociology and ethics. In this institution, we have very important fields, from research to education to policy research.

The first is economics, which is a social science, but also a very important subject closely related to people. Especially for the changes in the digital economy and human capital market, AI has brought many changes. So now we have several top economists in the world to promote this research.

Another important direction is the law. The law involves ethics, but the law itself faces AI, from driverless to medical care, to the government itself. Any decision that AI participates in is actually challenging some basic assumptions of past laws. Professors in our law school have participated in a lot of HAI's work. On the one hand, it is to see how the government can apply AI technology to make the government run more efficiently. But on the other hand, I am also thinking about how to make good policies and laws. On the one hand, I will continue to promote innovation, on the other hand, I will face many problems brought by the new AI.

Shen Nanpeng: Can you give me an example? Even if it has not been realized, how can artificial intelligence interact with artists, musicians and painters?

Li Feifei: Of course. About two years ago or a year and a half ago, the world's most famous auction house auctioned the first AI painting, which is the world's first art work generated by an algorithm, and then sold it at a high price, whether it is social art or music art. In fact, AI algorithm can produce very interesting works. This poses a challenge to human artists. What is the role of human artists? My AI can continuously produce Van Gogh's starry sky. If the human audience also loves the works of art created by AI, what do human artists represent? Is it sincere or other expression? So now there are many explorations on how to open the artistic space, because with such an algorithm, human expression and human emotion can continue to exist and develop in such a space. This is an example.

So we invited a professor of law, a professor of ethical philosophy and two professors of ethical biology. The four of them set up this committee to interact with us in real time at high frequency, helping us to think about the research direction, how to promote technology on the one hand and respect universal values and humanity on the other. Let science create benefits instead of accidentally hurting patients or medical staff.

: Hai also released the world's first artificial intelligence index report. Can you share this? This is a very forward-looking move.

Li Feifei: It may indeed be the first project in the world, led by a senior professor of Stanford Artificial Intelligence Laboratory in 20 17. Therefore, after the merger of the AI index project in 20 19, Shanghai continued to support this project, so this is our AI index for the fourth year.

The purpose of this AI index is to report the global AI process fairly, justly and comprehensively. It is some influences or changes from research and education to industry, industry and commerce. 2020 is definitely a very interesting year, because there are some new trends in COVID-19. For example, the first one, in 2020, the application of artificial intelligence in drug research and development design has undergone significant changes, which is a huge impact.

The second is that industrialization continues to develop strongly, and AI is becoming more and more industrialized. It represents that many doctoral students and even professors have begun to enter this industry. There is also AI, which still has great diversity challenges. The AI population is still dominated by men, and this challenge continues and has not been well solved.

Shen Nanpeng: I think everyone will be looking forward to it. In the future, the annual index report will guide this industry. Let's go back to your earliest work As an outstanding expert of artificial intelligence in the world and a leading figure of artificial intelligence in China, I want to share it with you. How did you do the ImageNet project at that time? What kind of promotion and revolutionary guidance does this bring to the whole artificial intelligence deep learning? What was the initial intention of doing this at that time?

Since 20 10, ImageNet has held an academic ImageNet Challenge every year. This competition requires the use of AI algorithm to classify 1000 pictures of1000 items. In 20 12, Professor Geoff Hinton of Canada and his students used a traditional algorithm called convection neural network and won the first place in our ImageNet challenge. It can be said that this is a historic event, which is equal to the "second spring" of neural network algorithm, which has opened the revolutionary development of deep learning and brought great changes to the past decade.

From 20 12, why should they participate in ImageNet and why should I do ImageNet? Back around 2006, AI was still a small field in computer science. I am a young professor who just graduated from a doctor's degree, and I have been thinking about what is the "North Star" in the field of AI. Polaris is the pursuit of people who do science. I come from physics. What I value most is where the most important question is. For me, the most important Polaris is visual learning. The ability to recognize thousands of things is the most important ability. If we humans don't have this ability, we can't do anything else. We can't go shopping or shopping.

From this point of view, I think we may have gone the wrong way before. We used to try to adjust the model parameters and look at one or two kinds of objects. We changed our thinking and used big data to promote the learning of visual intelligence. Actually, I thought of a dictionary. At that time, the biggest kind of visual objects probably came from dictionaries. This dictionary is specially called WordNet. It contains 80,000 noun symbols, but some nouns are not objects. For example, nouns like anger do not represent an object.

So I extracted nouns from 20,000 to 30,000 objects. Fortunately, 2007 was also a period of rapid development of our Internet. With the internet and data sources, our laboratory has done a lot of work for three years, and finally collected more than one billion pictures into a data set of 65.438+0.5 million pictures. At that time, our initial intention was to extract this Polaris from the data set. This is the original story of ImageNet.

Shen Nanpeng: I think this may be a story that will be written into a textbook, because it really led the development of a lot of artificial intelligence.

Li Feifei: You and Sequoia have always had a very keen sense of cutting-edge technology. I am a scientist myself. I especially want to ask you, as a global managing partner of Sequoia, what is your judgment on the development and application of AI in medical care in the next decade?