This article was published in the Bulletin of the Chinese Academy of Sciences, No. 5 "Scientific Observation". This article is compiled from the expert views in the third session of the "Songshan Lake Scientific Dialogue" in the special event of the 2023 Editorial Board of the Proceedings of the Chinese Academy of Sciences on April 25, 2023, and the author is the host and guest of the dialogue, and the signature method is in the order of the last name stroke.
BAO Yungang1, LIU Miao2,3, LU Pinyan4, QIU Xipeng5, XU Jiang6
1 Institute of Computing Technology, Chinese Academy of Sciences
2 Institute of Physics, Chinese Academy of Sciences
3 Matsuyama Lake Materials Laboratory
4 School of Information Management and Engineering, Shanghai University of Finance and Economics
5 School of Computer Science and Technology, Fudan University
6 The Hong Kong University of Science and Technology (Guangzhou) Microelectronics Domain
The 20th National Congress of the Communist Party of China pointed out the direction for the new generation of information technology industry, and it is necessary to promote high-quality development as the theme to build a new growth engine of the new generation of information technology industry. The market in the field of information technology is huge, the technology competition is fierce, and it is more monopolistic and technological exclusivity than traditional industries. Academician Li Guojie pointed out that the main factor determining the victory or defeat of the market is not a single technology, but an advantageous information technology system. Once a technical system dominates, it is difficult for the latecomers to catch up or replace in the same track, and it is difficult for the leaders of the original track to continue their success in the new "blue ocean". Therefore, the formation of a new technical system must seize the valuable opportunities when new applications emerge. With the rapid progress of the new round of scientific and technological revolution and industrial transformation, the scientific research paradigm has undergone profound changes, and grasping the current opportunity period of information technology development is of great significance for accelerating the construction of a new development pattern and accelerating the construction of a scientific and technological power.
1 The development and application of artificial intelligence has become the main trend in the forefront of information technology development
The rapid development of artificial intelligence is having a significant impact on human society
In the annual "10 Breakthrough Technologies" released by MIT Technology Review in the past 5 years, technologies related to artificial intelligence (AI) are frequently selected. For example, "AI for making graphics" and "chip design that changes everything" in 2023; 2022 "End Password", "AI Protein Folding", "PoS Proof of Stake", "AI Data Generation"; 2021 "GPT-3", "Data Trust", "Digital Contact Tracing", "TikTok Recommendation Algorithm", "Multi-skilled AI", etc. From these selected breakthrough technologies, it can be seen that artificial intelligence plays an important role at the application side and infrastructure level. On the application side, artificial intelligence plays an important role in many fields (such as biology, graphics, etc.), among which AI assistants are also widely used. At the infrastructure level, the infrastructure that supports AI applications faces many challenges. For example, how to better generate, manage, and protect data, and how to provide sufficient computing power. In general, the emerging problems of artificial intelligence are mainly concentrated on the application side and infrastructure side, which is also the current key research direction.
At present, there are huge development opportunities in the field of information technology. With the advancement of new data mining and application algorithms, data-driven tools have significant implications for many basic scientific research. For example, the rapid development and application of artificial intelligence has a huge role in promoting natural sciences (such as life science, material science, etc.). But at the same time, the development of information technology is also facing great challenges, and Moore's Law is gradually failing. On the one hand, the development of information technology requires more computing power; On the other hand, some new computing methods, such as quantum computing and optical computing, have re-challenged the understanding of the nature of computers. For example, when redesigning quantum computer systems, one will find that the nature of computers is not fully understood. This is a challenge in the field of information technology, especially in the field of computers, and at the same time a great theoretical opportunity.
From the perspective of the development of the entire field of information technology, the iterative nature of artificial intelligence is similar to Moore's Law - as the amount of data, training volume, and model size increases, its capabilities continue to increase, and the growth still does not stop. This rapid iteration is not only helpful for the improvement of the capabilities of artificial intelligence itself, but also has a huge help to the development of traditional industries and disciplines, because with more data, the scientific research process can be accelerated. Therefore, the development of artificial intelligence has a positive impact on society as a whole.
The integration between academia and industry in the field of information technology still needs to be strengthened
There is a close connection between the information technology community and the industry, but it is still an indisputable fact that there is a certain gap between the issues that the academic community is concerned with and the problems that the industry is actually concerned about. At present, there is an urgent need to extract key issues from industrial applications and feed these problems back to the academic community for research.
Taking theoretical computer science as an example, theoretical computer science mainly studies the feasibility of computing, which is to a large extent the basis of computer science and information science; Theoretical computers can be seen as fundamental laws in the world of computers and information processing, similar to objective laws in physics. Before the creation of computers, theoretical computers existed as branches of mathematics, focusing on the basic theory, algorithms and complexity of computing, and many epoch-making innovations such as computer invention and manufacturing and algorithm applications were based on it. Therefore, theoretical computer science plays an important role in the field of computer science and is at the forefront of the intersection of computer science and other disciplines. Theoretical computer science is a methodological discipline, its value is contained in computational thinking, with strong universality, can intersect in computer, economics, natural science, engineering and other fields, can put forward new problems, new perspectives and solutions in many complex problems and scenarios that originally lack language description or are difficult to solve, which is not only meaningful for computing itself, but also important for the physical world and human society.
The matching between theoretical computers and applications is a very important frontier. In the past, theoretical computers were first theories and then computers; The behavior of computers is set by humans, and humans can fully understand them. However, as the complexity of computers becomes higher and higher, including the complexity of ChatGPT models, etc. has become closer and closer to natural science, some experiments need to be done to explore and verify their theories, which are inconsistent with the original theory. Therefore, how to better match theory and application is a very important frontier direction. When the theory does not match the application, it is necessary to develop a new theory or better present the abstraction of the theoretical application. For example, in deep learning, the application is very cutting-edge, but the theoretical understanding is not enough. Therefore, it is necessary to explore how to bring theory and application closer together. This direction is closely related to both basic and applied research, and is also driven by curiosity. In 2020, Huawei established a theoretical computer laboratory with research interests related to algorithm complexity in industrial applications. Theoretical computer science is very useful in industrial applications, especially in Huawei's information and communication technology (ICT), optical, speedrun, chips, systems, applications, and cloud services.
At present, Chinese enterprises have a lot of room for exploration in high-tech, but they are holding back due to greater risks at the same time. Although enterprises are the main body of technology transfer into products, they involve more basic research investment in exploring new technologies, and enterprises face greater risks, and many enterprises are unwilling to bear this risk. This has also led to certain shortcomings in complex chip design and electronic design automation (EDA) tools.
Taking processor chips as an example, processor chips require software and hardware cooperation, which is a relatively complex design of chips. In recent years, the design of "open source chips" (applying the model of open source software to processor chip design) represents a new direction in the field of processor chips. For example, the architecture of RISC-V's new processors has attracted global attention. Like the 5G standard in the field of communications, global forces can be united to build a chip ecosystem and jointly develop standards, and countries can compete at the product level. In the future, the design of chips hopes to give full play to China's advantages of large market scale and many technical talents through a more open way. In recent years, the "Xiangshan" open source chip project initiated by the Institute of Computing Technology of the Chinese Academy of Sciences has attracted domestic and foreign enterprises to participate in joint development.
Talent training is an important power reserve for the development of information technology
The development of science and technology in mainland China has entered a new stage, and innovation has become the primary driving force for development. Cultivating young scientific and technological talents with outstanding innovation ability in the local area is the driving force for the sustainable development of the mainland's scientific and technological undertakings.
Today, mainland research institutes and universities have invested a lot of resources in basic theory and core technology to cultivate talents and make every effort to tackle tough problems. For example, in August 2019, the University of Chinese Academy of Sciences launched the "One Core for Life" program, which allows undergraduates to participate in the whole process of processor chip design from design to production and operation, cultivate processor chip design talents with solid theoretical and practical experience, improve the scale of training of mainland processor chip design talents, shorten the cycle of talents from the training stage to the front line of scientific research and industry, and cultivate more chip talents in short supply in the country. The program has been carried out for 5 phases, with more than 2,000 students participating, and a large-scale high-quality chip design talent training program has been initially formed. The Hong Kong University of Science and Technology (GZ) Microelectronics Division has formed a team to establish a series of central research facilities, including the Materials and Devices Microfabrication Laboratory and the EDA Research Center, aiming to cultivate talents and promote the output of more original technologies in the field of microelectronics. Recently, the team has made some progress and original discoveries in the fields of photoelectric fusion chips, verification, and multiprocessor high-speed simulation.
In the rapid development of the field of information technology, there may be some changes in the way the scientific community cooperates. The organizational structure of previous schools and institutes may need to be adjusted to accommodate the range of changes brought about by AI. These changes will also affect students' curriculum and educational styles. The impact of artificial intelligence has made the existing education system seem to be unable to keep up with the trend of technological progress. Therefore, it is necessary to constantly adjust the education method and organizational structure to adapt to the new trend of scientific and technological development.
2 Information technology will accelerate the development of natural science research
Artificial intelligence has become an important driving force for the acceleration of basic scientific research
In fact, information technology has gradually penetrated into natural science research. For example, methods such as databases and artificial intelligence have become essential tools in daily scientific research.
In the field of data science, Turing Award winner Jim Jim Gray proposes the fourth paradigm, a data-driven scientific research paradigm followed by experimental observations, theoretical deductions, and computational simulations. In recent years, the fifth paradigm has been proposed, and this new scientific research paradigm is immersive embodied research with intelligence as the research goal, based on the ontological understanding of data science. It can be guessed that the "fifth normal form", like the fourth normal form, will be aimed at data; The difference is that the "fifth paradigm" focuses more on the interaction between people, machines and data, emphasizing the integration of human decision-making mechanism and data analysis, reflecting the organic combination of data and intelligence.
The combination of artificial intelligence and scientific research can help researchers improve the efficiency and accuracy of scientific research. For example, researchers with cross-cutting backgrounds in fields such as mathematics, statistics, physics, and computer science can combine AI with high-performance computing to provide powerful tools for molecular dynamics simulations and first-principles simulations. Through the acceleration of supercomputers, researchers have made great breakthroughs in simulating atoms, the simulation scale has increased from the previous million-level to billion-level, and the simulation time has been increased to the nanosecond level, which is of great help to physics and materials science research. With the continuous combination of science and artificial intelligence, there will be more breakthroughs in the future.
In the future, data will become the basic resource of scientific research, and database will become an important scientific infrastructure, nourishing the growth of various disciplines like a big scientific device. The Materials Science Database (https://Atomly.net) recently released by Songshan Lake Materials Laboratory and the Institute of Physics of the Chinese Academy of Sciences uses high-throughput computing and information technology to bring high-quality scientific data to mainland researchers. With the help of information technology, the development of material science has entered the era of "big data + artificial intelligence", breaking the monopoly position of foreign countries in this field, creating high-quality basic data and tools to widely support the development of material disciplines in the mainland, and has begun to play an effective role and effectively promote the development of the field.
Artificial intelligence has had a huge impact on the entire field of science, including in education. In learning and research, the application of new technologies brings new challenges and opportunities. For example, when using a technology such as GPT, there is a need to weigh its pros and cons to ensure that it is used legally and fairly. From the perspective of information itself, information processing is the foundation of modern society, including information collection, sorting, processing, storage, encryption and other links. Artificial intelligence is of great help to the collation and collection of information, but in terms of deep mining of information, because artificial intelligence itself is based on statistical methods, it still has challenges for the discovery of things themselves, and more research and breakthroughs are needed. Therefore, when applying artificial intelligence, it is necessary to fully consider its limitations, and at the same time, it is necessary to actively explore new technologies and methods to continuously promote the development of artificial intelligence.
In the process of the development of artificial intelligence and theoretical computers, it is necessary to pay attention to their positive and negative effects on scientific research. Positive impact: On the one hand, artificial intelligence will lower the threshold of basic scientific research, so that more people can participate in this process; On the other hand, from the national scale to the global scale, AI will also make large-scale collaboration easier, and cooperation between scientists will be more convenient. Negative effects: For example, misuse of data tools can lead to a large number of articles, and there are already cases where machines generate data and articles and then process them with machines, which can lead to problems such as information loss and materialization. Therefore, when using artificial intelligence and theoretical computers for scientific research, we need to carefully weigh the pros and cons and maintain a focus on the quality and accuracy of data and research results.
Artificial intelligence promotes the development of interdisciplinarity and interdisciplinarity
As a tool for natural science research, information technology has been widely used and achieved remarkable results. In addition, the development of computer science has also had a profound impact on natural science research. For example, the problem of NP completeness in theoretical computers, originally posed only for computational complexity, has now been widely used in fields such as physics, chemistry, and biology. These disciplines use NP complete problems to illustrate their regularities and complexity as a tool for describing complexity. If a substance is NP-complete, its laws are more chaotic; Conversely, if it is not NP-complete, there may be some internal laws. This interdisciplinary application fully reflects the important role of computer science in natural science research. Looking at the computer field from different perspectives, such as the combination with physics, economics and other disciplines, you can find the advantages of computer science in terms of profundity and universality. What makes the computer field so unique is its rich color and diversity.
The development of information technology and computing has brought new essential concepts and measurement methods to natural science research, which is very beneficial to the development of science. For example, it is proved that the complexity of the assignment function calculation in a statistical physical system coincides exactly with the phase transition line of the physical system, which shows that the concept of computational complexity is intrinsically related to the phase transition of the physical system. Such interdisciplinary developments not only use information technology as a tool, but also provide new essential concepts and methods for scientific research. Therefore, it is necessary to explore more interdisciplinary research, integrate knowledge and methods in different fields, and promote the development of scientific research. Another example is the development of communication technology, which was based on Shannon's information theory and Maxwell's electromagnetic field theory in the early days, but when these two theories are combined and electromagnetic fields are used as information carriers, the boundary between information and physics can be broken and communication efficiency can be improved. This connection is very important and can bring more innovation and breakthroughs in the field of information and communication.
The fifth paradigm is an important model in current scientific research, especially "AI for Science". Now, people think more that AI can help us discover scientific laws and come up with hypotheses. In particular, today's large language models can read a variety of literature, including computer science, information science, and some traditional disciplines (such as physics and chemistry), and these concepts are common in their semantic space. Therefore, AI is like a generalist, able to facilitate communication and interaction between different disciplines, thus discovering common ground between them and coming up with relevant examples. This interdisciplinary exchange can lead to new scientific discoveries and allow us to better understand the relationship between AI and science.
As a tool for researchers, AI output also needs to be treated with caution
The role of artificial intelligence in scientific research is very obvious, it can help scientists to explore and study better. Take ChatGPT, for example, which can provide researchers with better ideas and directions, even more perfect than the ideas of the scientists themselves. In addition, the output of ChatGPT has a certain randomness, which can be used as a brainstorming tool to provide new ideas and ideas. However, users also need to be aware of the hyperparameters behind AI models and the possible biases and errors in the outputs they produce. Therefore, when using AI models for scientific research, it is necessary to be careful with their outputs, while maintaining a focus on the quality and accuracy of scientific research.
3 Information technology leads the industry to digital and intelligent transformation
Artificial intelligence brings benefits to industry and downstream applications
Artificial intelligence, data-driven, and computing power have caused huge changes in the scientific world, and this change has quietly happened in industry. With the continuous development of information technology, the industry is also gradually turning to digitalization and intelligence. In this environment, the development of industry-university-research relations and the changes in the industry deserve special attention.
The development of artificial intelligence has brought obvious benefits to industry and downstream applications. As the versatility and capabilities of AI models continue to increase, the development cost of downstream applications has been greatly reduced. Historically, applying AI to traditional industries required specialized data collection, labeling, and debugging, which was costly. Now, AI models are more versatile and intelligent, and everyone can train with their own data; The understanding ability of AI models has also improved dramatically, allowing them to communicate and correct according to the user's wishes. The wide application of this AI model greatly reduces the development cost of downstream applications, and users only need to debug and modify and achieve the desired effect through simple interfaces and prompts, which greatly reduces the cost and consumption of computing power. Therefore, the prospects of AI models in industry and downstream applications are very broad.
The development prospects of artificial intelligence in the field of natural sciences are very broad, and more breakthroughs and progress will be made. In the next 3-5 years, artificial intelligence is expected to make major breakthroughs in the fields of biology, physics, weather forecasting, and EDA integrated circuit design tools. At present, there are already some preliminary results in these areas. For example, in biology, machine learning is already being used to accelerate simulation; There are also many applications in the field of physics and quantum physics; In the field of EDA IC design tools, machine learning has shown great effectiveness and efficiency. However, at present, these application scenarios still need to be further industrialized before they can be transmitted to products.
Trends in the field of information technology
The open source model has already had a huge impact in the field of information, and the impact will be even more profound in the next 3-5 years. Open source software has already had a huge impact on the information field, and now this model is penetrating the hardware field. For example, in the field of chips, the trend of open source chips will gradually increase. In addition, although the open source pre-trained model is still only a small model, it is possible to gradually grow from a small model to a large model as more and more people join in. The open source model will bring a series of far-reaching impacts, including the way data is open source, how data is shared and exchanged. The impact of this model is not only at the individual technical level, but also on the change of the entire technology research and development model and production relations. Therefore, we must not only pay attention to the impact of the open source model, but also constantly adapt to this change in order to better promote the development of information technology.
Information technology is a field that can continue to grow exponentially, and this exponential growth trend does not exist in every field of research. Taking the development of aircraft engine thrust as an example, from the invention of the first aircraft by the Wright brothers in 1911 to the 50s of the 20th century, the development of aircraft engine thrust increased exponentially during this period; This growth used to raise hopes for a moon landing, but then the trend basically stalled. In contrast, since the 60s of the 20th century, Moore's Law in the field of information technology has been developed for nearly 60 years, although the development of Moore's Law is about to stagnate from reality, but the data field has shown an exponential growth, which has brought new vitality to the field of information technology. This growth trend is not only reflected in hardware, but also in data volume and other dimensions. This is a distinctive feature of the field of information technology, and we need to seize this opportunity, continuously play its advantages, and promote the integration of information technology with other fields.
Human society has entered the information age, but people's understanding, mastery and application of information technology are far from enough. We've seen many amazing use cases (like ChatGPT) in recent years, but these are just the tip of the iceberg, and there will be more new technologies and breakthroughs in the future. From the perspective of information technology, the current is the golden age of academic research, although the United States and Western countries are trying to "decouple" from China, which forces us to do our own original technology and find other ways out, which presents us with new challenges, but it is a rare opportunity. In a new field, there is never a shortage of new ways out. Therefore, we must firmly grasp this opportunity, actively explore and carry out research, and strive to promote the development of information technology.
Information science and information technology are common technologies, and they are the underlying key technologies for different scientific research and industrial applications. Therefore, the importance of information technology is self-evident, it is helpful for scientific research and industrial applications. Information technology has been widely used in various fields, such as medical care, finance, transportation, etc., bringing convenience to people's life and work. At the same time, the development of information technology has also promoted the innovation and transformation and upgrading of various industries. In the future, with the continuous development of information technology, it will continue to play a more important role and make greater contributions to the progress and development of mankind.
Bao Yungang is a researcher and deputy director of the Institute of Computing Technology, Chinese Academy of Sciences, and a professor and deputy dean and professor of the School of Computer Science and Technology, University of Chinese Academy of Sciences. His main research area is computer system architecture.
Miao Liu is a distinguished researcher at the Institute of Physics, Chinese Academy of Sciences and Songshan Lake Materials Laboratory. Main research areas: develop the underlying methods and databases in the direction of "artificial intelligence + material science", and lead the development of Atomly materials science database and platform.
Lu Pinyan is a professor at the School of Information Management and Engineering, Shanghai University of Finance and Economics, and the director of the Research Center for Theoretical Computer Science. Mainly engaged in theoretical computer and interdisciplinary research.
Qiu Xipeng is a professor at the School of Computer Science and Technology, Fudan University. Mainly engaged in natural language processing, deep learning and other research directions.
Xu Jiang is a professor and director of the Microelectronics Division of the Hong Kong University of Science and Technology and the Hong Kong University of Science and Technology (Guangzhou). Mainly engaged in the research of integrated circuits.
Article from:Bao Yungang, Liu Miao, Lu Pinyan, et al. Some thoughts on information technology driving future industries. Bulletin of Chinese Academy of Sciences, 2023, 38(5): 766-772
DOI:10.16418/j.issn.1000- 3045.20230515003