Less than a hundred years after its birth, computer science has had a tremendous impact on human life. What is its future development direction? On November 7, the world's top scientists in the fields of computer science, artificial intelligence and information technology "sat around the round table π" at the scientific round table meeting of the world's top scientists to discuss "new breakthroughs in the future of digital intelligence". John Hopcroft, the chair of the round table and winner of the 1986 Turing Prize, said: "We are at the beginning of the information age, not the end." Artificial intelligence should not be satisfied with behavior imitation through training. Artificial intelligence has been able to imitate many human behaviors, and it is not too much to describe it as "all-round development of music, chess, calligraphy and painting". Lano Bloom, winner of the 2004 U.S. President's Award for Excellence in Science, Mathematics and Engineering Guidance, pointed out at the meeting: "Artificial intelligence is like an 'integrator' that can be combined with many fields, which is worth thinking deeply." In the future, AI should pay more attention to "decision-making" than imitation. Michael I. Jordan, the winner of the "Intelligent Science or Mathematics Award" of the first Top Scientists Association Award (hereinafter referred to as the Top Science Association Award), said that although "imitating human behavior" is the classic definition of AI, in-depth analysis is the key. For example, AI can analyze massive data with game theory; It can also predict the structure of protein folding, which is impossible for human beings. On November 6, Jordan also said at the media meeting of the first winner of the Top Science Association Award: "The significance of AI is not to let robots dance." Jordan's theoretical framework is the basis for the application of machine learning. He has successively conducted research in psychology, mathematics, cognitive science and other fields, and successfully connected computer science and statistics. He believes that both individuals and groups are intelligent subjects and can produce decision-making behavior. "All restaurants in Shanghai can also be an intelligent collection. If the weather changes or the urban pattern is updated, this intelligent experience will make relevant responses. Importantly, it is also a very robust intelligent collection." Jordan said, "If we collect huge amounts of data through individuals, we can open up a new world in economics." Robustness is an important evaluation index of machine learning model, which mainly tests whether the model is stable when facing certain changes. In the future, data will be "connected into a line". Human society is composed of extremely complex cultural connections, which cannot be described by single data. "In the future, computer science will" connect the dots into a line "and make structural prediction and analysis through interaction between subjects, so that human beings can have a better and more efficient life." Jordan said. So, in a dynamic society, how can machines "improve themselves"? This is a long-term concern of Long Mingsheng, an associate professor of Tsinghua University's School of Software. "The 'error detection rate' may be a missing link in AI research. More than 10 years ago, machine learning technology was always defeated by practical problems, which prompted me to think about the shortcomings of machine learning." However, Long Mingsheng said frankly that in his research process, the training of artificial intelligence often fails due to changes in the environment and deployment. Artificial intelligence is in urgent need of some principles and methodologies to form a system. "Robustness