Academician E Weinan： Reconstructing the scientific research system in the AI era is the best opportunity for China’s scientific and technological innovation
Academician of Chinese Academy of Sciences, President of Beijing Institute of Scientific Intelligence, and Director of Peking University International Machine Learning Research Center, Ewinan was interviewed. Beijing News reporter Luo Yidan/photo
On May 30th, E Weinan, academician of Chinese Academy of Sciences, president of Beijing Institute of Scientific Intelligence and director of Peking University International Machine Learning Research Center, gave a speech entitled "Reconstruction of scientific research system in AI era" at the parallel forum of Zhongguancun Forum, and was interviewed by the Beijing News reporter.
E Weinan told the Beijing News reporter that AI can have a great impact on scientific research, but the creative work that scientists are engaged in cannot be replaced by AI. In his view, the main task at present is to establish "platform research" under the impetus of "AI for Science" (scientific intelligence), which can be compared to the "Android model" of scientific research, and then produce more new methods and tools. "I don’t think this is just what China has to do. We should work together internationally to conduct research."
It is worth noting that on May 30th, the "Literature Knowledge Base Based on Large Language Model and Vector Database-Science Navigator" jointly developed by Beijing Institute of Scientific Intelligence, computer network information center and Moqi Technology was officially released. In this regard, Ewinan said that this is an important step in the infrastructure construction of "AI for Science".
Ewinan believes that "AI for Science" system can greatly enhance scientific research ability and strengthen the connection between scientific research and industry, which is the best opportunity in the whole history of scientific and technological innovation in China.
AI provides an effective means to solve the bottleneck problem of scientific research. Bring great benefits to scientific researchspace
In his speech, Ewinan mentioned that "scientific research includes two modes: data-driven and basic principle-driven. The bottleneck of data-driven is that we lack effective data and data analysis methods in most cases. However, using basic principles to solve practical problems, although the basic principles themselves are clear, it is difficult to express the mathematical model of the basic principles. The result is that simple problems can be solved well, such as using structural mechanics to help build houses, but complex problems, such as material design and drug research, can only be solved by experience and trial and error at present, because these complex problems are facing the disaster of dimension: with the increase of the degree of freedom of the problem (that is, the dimension in mathematics), the complexity increases exponentially. And this difficult AI can help us solve it. "
In his view, deep learning provides a basic tool to solve the dimension disaster. This is not only for science, but also general, which is the starting point of AI for Science. "AI provides new tools. In terms of data-driven, AlphaFold2 (no official Chinese name yet) is a typical tool to develop new data-driven by AI method, and it is also the most typical successful case, which basically solves the most basic problem in structural biology, protein structure. At the level driven by basic principles, the best example may be DeePMD (Deep Potential Molecular Dynamics): combining machine learning and physical modeling, it can handle hundreds of millions or even billions of tools that can only handle thousands of atoms in the past through the blessing of artificial intelligence. "
"From my own research experience, I have been trying to solve a series of problems in materials science, chemistry and other disciplines for many years. I am aware of the core difficulties, and artificial intelligence can help us, which is also the most basic starting point of AI for science. " Ewinan said.
Ewinan emphasized that the productivity change brought by the ability of AI blessing will inevitably bring about the change of production relations. "Although we have done scientific research for so many years, we have also made many scientific research achievements, but if you think about it carefully, the current scientific research model is still biased towards workshop style. Every laboratory and every research group are basically self-sufficient, and the cycle is very long, and the efficiency needs to be improved. With the artificial intelligence method just mentioned, it will inevitably bring a new generation of infrastructure tools and improve scientific research efficiency. "
In an interview, he said that in addition to scientific research, the core points of many problems in the industry are also related to algorithms. For example, industrial software is based on algorithms, and drug design also needs algorithms, so AI can also have a direct impact on the industry. "As for the impact of AI for Science on the manufacturing of the real economy, we are confident that in about 5 to 10 years, industrial design software in micro-fields will be produced in industries such as materials and drugs to make it more systematic, just like building cars now, they will also be. Driven by AI for Science, the micro-level manufacturing industry can also become an assembly line operation mode. "
In an interview, Ewinan said that the space that AI brings to scientific research is huge, and it provides some platforms and solves many problems that scientists could not solve before. The difference is that scientists not only emphasize solving problems, but also need to know the reasons, so I hope AI can also tell scientists why they can solve these problems and how to solve them.
Publishing literature knowledge base based on large model and vector database is an important step for AI for Science.
Ewinan also proposed whether all the literature and experimental data can be turned into a knowledge base or database, and then natural language dialogue or other search and query methods can be used to help scientists improve the efficiency of reading literature and find the needed knowledge more conveniently.
For the Science Navigator released on the same day, E Weinan said that the knowledge base allows researchers to search, read, analyze and manage documents through dialogue and questioning, which greatly improves efficiency, further helps researchers improve scientific research productivity and releases more time and energy. It is one of the important infrastructure construction of AI for Science.
"Today’s release is the first version, and we hope that it will become a big platform for integrating literature and experimental calculation data in the future, and everyone can use it."
"Generally speaking, such a new scientific research system will greatly enhance scientific research capabilities on the one hand, and greatly accelerate the connection between scientific research and industry on the other. AI for Science is the best opportunity in the whole history of scientific and technological innovation in China, which can completely change the pattern of scientific research and industrial innovation, and it is highly consistent with the national policy of national development. In this regard, China also has certain first-Mover advantages, highly forward-looking design and broad consensus. We need to make use of these advantages, concentrate our strength and resources, land as soon as possible, and take the lead in stepping out of the new scientific research paradigm of platform+vertical integration. "
Reporter contact email: luoyidan@xjbnews.com
Beijing News reporter Luo Yidan Editor Wang Jinyu Proofread Wang Xin