On July 23, Dr. Kaifu Li, chairman and CEO of Innovation workshop, was invited to participate in the “deep tech benefits mankind” activity sponsored by sginnovation, and had a dialogue with Professor yoshua bengio, co founder of element AI, to discuss the future development of artificial intelligence. Professor yoshua bengio is one of the three inventors of deep learning and the winner of ACM Turing Award in 2019. He has long been committed to promoting the rational use of AI, especially in solving environmental issues with AI. For example, he used the AI algorithm concept of new drug research and development, and extended its application in exploring new low pollution materials to combat the arduous challenges of global climate and environmental change, showing that technology can make the world a better place. In the novel coronavirus pneumonia, Dr.
and Dr. Yoshua Bengio discussed the significance of AI to human society, especially in the post new crown pneumonia era. How can AI help the future economic society to become more flexible, livable and sustainable in the era of Yoshua. < p > < p > they believe that AI is a once-in-a-lifetime opportunity for humans to truly free themselves from repetitive tasks. With the help of AI, we will hope to build a wise, rational and inclusive society and build a virtuous circle between human society and AI. < / P > < p > “although the next research is heavy, the new exhibition is exciting. Especially in the field of deep learning, I call it “deep learning 2.0.” ——Yoshua bengio < / P > < p > yoshua bengio: the first question I have a lot of resonance. In my opinion, one of the major limitations of machine learning at present is the generalization ability of learning system. The problem of < / P > < p > seems to be unsolved, but at present, we have found several breakthrough points and ideas, mainly referring to the human consciousness processing mechanism, rapidly reorganizing the scattered knowledge accumulation. < / P > < p > although the combination of these knowledge does not necessarily follow the distribution of training data, we can still get some advantages in the direction of reorganization, so as to better summarize the training distribution. < / P > < p > although the following research is arduous, the new exhibition is exciting. Especially in the field of deep learning, I call it “deep learning 2.0”, which can absorb the inductive tendency of human beings and generalize the data distribution algorithm. Li Kaifu: I’ll take Professor bengio’s point of view to say a few more words. I have been studying conversational AI since I was a college student. At present, the man-machine interface, which I call delegation interface, is mostly based on direct operation, such as keyboard, mouse, multi-point touch and so on. However, language is the most basic and natural way of human communication. To AI speech recognition, natural language understanding, has always been our goal. < / P > < p > for example, when we used search engines, we used to search web pages by entering keywords. Later, Google brought a new breakthrough, intelligent question and answer function based on deep learning can directly let the machine “say” the answer. < / P > < p > but we should not stop here, but should continue to work towards the next goal: to improve the understanding and execution ability of the machine to human command intention through further research of deep learning. < / P > < p > for example, can we send an instruction directly to Amazon Alexa: “give my mom a birthday present.”. After that, it will automatically sort out, browse gifts and arrange delivery. It knows my personal preferences, the price range I can accept, who my mom is, where I live, and what gifts I want. < / P > < p > yoshua bengio: I’d like to share a brief introduction to the industrial application of AI. I think it’s a very difficult issue. The difficulties come from two aspects: one is the social aspect, the other is the technical aspect. < / P > < p > in the social aspect, from basic scientific research to the final product development stage, it is necessary to jointly create a culture so that researchers can have research freedom and make real breakthroughs. In terms of technology, we need some software tools to make the transformation process from R & D to production as efficient and fast as possible. < / P > < p > for example, autonomous driving will completely change the transportation industry; Alexa is changing the speaker industry to some extent; and new Internet insurance applications, such as lemonade in the United States and shuidi company in China, are likely to subvert the insurance industry. < / P > < p > these industries already have certain conditions, and it is very expected that industry experts can bring subversive influence through AI. When AI is combined with disruptive innovation in the industry, there will be an opportunity to defeat industry giants and restructure the industry pattern. < p > < p > PwC estimates that AI will bring a net wealth increase of US $15 trillion to the world in 2030, mainly from the combination of traditional industries and AI. Due to the large scale of traditional industries, only a few percentage points can generate huge wealth. However, the difficulty lies in the fact that some traditional enterprises do not know anything about AI at present. They think that AI is the imagination of science fiction, and they can not see the immediate benefits. In addition, the technical tools are too difficult to use, which makes their IT departments unable to control. < / P > < p > therefore, we should help traditional industries accept and recognize the benefits of AI through training. At the same time, the AI enterprises we invest in or companies like element AI need to help traditional enterprises find simple and easy-to-use tools, so that they can cross the technology gap and use them immediately. < / P > < p > “privacy must be considered in the context of public health or personal health. During a public health crisis, the state should balance respect for rights and necessary prevention and control measures, so as to effectively control the spread of disease.” ——Kaifu Lee < / P > < p > Li Kaifu: I’d like to talk about a few examples of personal experience. Social isolation during the outbreak has spawned many AI applications, such as delivery robots in hospitals. The same is true for people with < / P >. When I returned to my home in Beijing for isolation a while ago, I didn’t see a single person in my apartment building. All the things were handed over to a robot, including online shopping packages and food delivery, which really realized zero contact and minimized the risk. The second example is the combination of AI and healthcare. We invest in the AI medical enterprise insilico medicine, which mainly uses the generated chemical warfare neural network to develop new drug small molecules. During the outbreak, they developed new drug molecules that inhibit the main protein components responsible for replication in the virus through the AI platform in a few weeks. < / P > < p > the last example may be controversial, which is contact tracking. Many countries in the world have successfully established contact tracking system and effectively controlled the spread of the epidemic. But in the United States, Europe and other places, this practice is seen as a violation of privacy. < / P > < p > in this regard, I fully understand and respect countries that value privacy, but I think privacy must be considered in the context of public health or personal health. During the public health crisis, the state should balance the respect for rights and the necessary prevention and control measures, so as to effectively control the spread of the disease. When the epidemic is over, it will return to normal. < / P > < p > we all don’t want to repeat the epidemic. I expect that in the future, AI will be used to prevent the occurrence and spread of epidemics. The hospital will widely use sensors and wearable devices to collect epidemic information, report potential hazards in time, and curb the trend of exponential growth of the epidemic in the early stage, so as to better deal with the crisis and avoid losing control again. < / P > < p > in the field of chemistry and biology, there are so many combinations of tests that it is impossible to study them one by one. So we need a reasonable search strategy, which is what I’m involved in now. We hope to shorten the research time with AI, and develop new antiviral drugs by recombining existing drugs. < p > < p > in the aspect of contact tracking, most of the existing contact tracking methods do not use AI, only simple testing method: if someone’s test result is positive or confirmed to be infected, then all people who have contacted with them should be isolated. But the infection began before the test was positive and then quarantined. < p > < p > our research shows that if we can predict the infectivity and infectivity of an individual in advance with the help of machine learning, and through some fuzzy data analysis, we can greatly save the waiting time and know the virus carriers as soon as possible, so as to suppress the spread of the virus. < / P > < p > of course, privacy issues will inevitably arise. There is an interesting contradiction between privacy protection and machine learning needs. Privacy protection needs to reduce data exchange as much as possible, while machine learning needs to collect as much data as possible. < / P > < p > many countries are very worried that the abuse of contact tracking will infringe privacy, so many privacy protection technologies have been promoted. The good news is that the two can coexist. < / P > < p > “if we can make good use of the capabilities of artificial intelligence, we can quickly find better new materials to replace the toxic materials such as carbon and batteries that cause long-term pollution to the earth today.” ——Yoshu bengio < / P > < p > Li Kaifu: I will give some examples from the perspective of Innovation workshop. Innovation workshop is a venture capital company, and we hope AI can be applied reasonably. Professor bengio may add a little more on climate change. < / P > < p > in AI the future, I describe a blueprint for the coexistence of humans and AI: AI takes on the task of optimizing routine work, allowing humans to focus on tasks that require creativity and compassion. In the future, doctors will become compassionate caregivers, deeply care for patients and communicate with them. AI can be used to analyze radiation results, MRI, CT reports, propose various possible diagnosis and treatment results, recommend drugs specifically, and assist scientists to develop new drugs. The same is true of the education industry. We’ve invested in a lot of online education companies and found that AI In the teachers’ routine work, they are very good. They can arrange homework according to the characteristics of the students. They can save time, let them focus on guiding the children’s ability and soul, carry out personalized education, and help them cultivate their creativity, team cooperation, communication ability and compassion. Therefore, health care and education are not only areas where AI can show its advantages, but also valuable investments. At present, these two areas are booming, and we have invested a lot of energy and money. < / P > < p > yoshu bengio: I totally agree with Dr. Lee. The progress of AI technology can benefit most people. We need to devote a lot of energy to such projects. < p > < p > we use AI algorithms similar to new drug research and development to generate, synthesize, and evaluate various new material technologies, including carbon recovery and batteries, to combat the formidable challenges of global climate change. < / P > < p > under normal circumstances, the research and development of these new materials takes a long time, often more than ten years, even longer than the development of new drugs. But if we can make good use of the ability of artificial intelligence, we can find better new materials more quickly to replace the toxic materials such as carbon and batteries which cause long-term pollution to the earth.