Chairman, Vice Chairmen, Secretary General, and Members:
In recent years, the field of artificial intelligence is experiencing an explosive development led by generative artificial intelligence models. On November 30, 2022, OpenAI launched an artificial intelligence chatbot called ChatGPT, which attracted widespread attention worldwide for its excellent natural language generation ability. Within two months, it surpassed 100 million users and immediately sparked a wave of large-scale modeling both domestically and internationally, Gemini、 ERNIE Bot, Copilot, LLaMA, SAM, SORA and other large models have sprung up, and 2022 is also known as the first year of large models. The current information age is accelerating into the development stage of intelligent computing, and breakthroughs in artificial intelligence technology are emerging one after another, gradually empowering thousands of industries and promoting artificial intelligence and data elements to become typical representatives of new quality productivity. General Secretary Xi Jinping pointed out that the new generation of artificial intelligence should be regarded as a driving force to promote technological leapfrog development, industrial optimization and upgrading, and overall productivity leap, and strive to achieve high-quality development.
1、 Introduction to the Development of Computing Technology
The development history of computing technology can be roughly divided into four stages. The emergence of the abacus marks the first generation of human beings - the era of mechanical computing. The second generation of electronic computing is marked by the emergence of electronic devices and computers. The emergence of the Internet has enabled us to enter the third generation - network computing. At present, human society is entering the fourth stage - intelligent computing.
Early computing devices were manual auxiliary computing devices and semi-automatic computing devices. The history of human computing tools began with the Chinese abacus in 1200 AD, followed by the Napier chip (1612) and the rolling wheel adder (1642). In 1672, the first computing device that automatically completed four arithmetic operations, the stepper calculator, was born.
Some basic concepts of modern computers have already emerged during the era of mechanical computing. Charles Babbage proposed the design concepts of differential machines (1822) and analytical machines (1834) to support automated mechanical calculations. During this period, the concepts of programming and programming were basically formed. The concept of programming originated from the Jacquard machine, which controlled printing patterns through punched cards, and eventually evolved into storing all mathematical calculation steps in the form of calculation instructions; The first programmer in human history was Ada, the daughter of the poet Byron. She wrote a set of calculation instructions for solving the Bernoulli sequence for the Babbage differential engine. This set of instructions was also the first computer algorithm program in human history, which separated hardware and software and introduced the concept of programming for the first time.
Until the first half of the 20th century, four scientific foundations of modern computing technologies emerged: Boolean algebra (mathematics), Turing machines (computational models), von Neumann architecture (architecture), and transistors (devices). Among them, Boolean algebra is used to describe the underlying logic of programs and hardware such as CPUs; Turing machine is a universal computing model that transforms complex tasks into automated processes that do not require human intervention; The von Neumann architecture proposed three basic principles for constructing computers: the use of binary logic, program storage and execution, and the computer being composed of five basic units: arithmetic unit, controller, memory, input device, and output device; Transistors are semiconductor devices that make up basic logic circuits and storage circuits, and are the "bricks" that build the tower of modern computers. Based on the above scientific foundation, computing technology has developed rapidly and formed a large-scale industry.
From the birth of the world's first electronic computer ENIAC in 1946 to today's 21st century, five successful platform based computing systems have been formed. Currently, various types of applications in various fields can be supported by these five types of platform computing devices. The first type is high-performance computing platforms, which solve the scientific and engineering computing problems of core national departments; The second category is enterprise computing platforms, also known as servers, which are used for enterprise level data management and transaction processing. Currently, computing platforms of Internet companies such as Baidu, Alibaba and Tencent belong to this category; The third type is personal computer platforms, which appear in the form of desktop applications, through which people interact with their personal computers; The fourth category is smart phones, which are mainly mobile and portable. Mobile phones are connected to the data center through the network, mainly Internet applications, and they are distributed in the data center and mobile terminals; The fifth type is embedded computers, which are embedded in industrial and military equipment to ensure the completion of specific tasks within a certain time frame through real-time control. These five types of devices almost cover all aspects of our information society, and the sixth type of platform based computing system centered on intelligent computing applications that people have long pursued has not yet formed.
The development of modern computing technology can be roughly divided into three eras. IT1.0, also known as the era of electronic computing (1950-1970), is characterized by a focus on "machines". The basic architecture of computing technology has been formed. With the advancement of integrated circuit technology, the scale of basic computing units is rapidly shrinking, and transistor density, computing performance, and reliability are constantly improving. Computers have been widely used in scientific engineering calculations and enterprise data processing.
IT2.0, also known as the era of network computing (1980-2020), is centered around "people". The Internet connects the terminals used by people with the background data center, and Internet applications interact with people through intelligent terminals. Internet companies, such as Amazon, put forward the idea of cloud computing, encapsulating the computing power of the background into a public service to rent to third-party users, forming a cloud computing and big data industry.
IT3.0, also known as the era of intelligent computing, began in 2020. Compared with IT2.0, it added the concept of "things", which refers to various end devices in the physical world that are digitized, networked, and intelligent, achieving the tripartite integration of "human machine object". In the era of intelligent computing, in addition to the Internet, there is also data infrastructure, which supports all kinds of terminals to realize the interconnection of everything through the end-to-end cloud. The terminals, the physical end, the edge, and the cloud are all embedded in AI, providing large model intelligent services similar to ChatGPT. Finally, where there is computing, there is AI intelligence. Intelligent computing has brought about massive amounts of data, breakthroughs in artificial intelligence algorithms, and explosive demand for computing power.
2、 Introduction to the Development of Intelligent Computing
Intelligent computing includes artificial intelligence technology and its computing carriers, which have roughly gone through four stages: general-purpose computing devices, logical reasoning expert systems, deep learning computing systems, and large-scale model computing systems.
The starting point of intelligent computing was the universal automatic computing device (1946). Scientists such as Alan Turing and John von Neumann initially hoped to simulate the process of human brain processing knowledge and invent machines that think like the human brain. Although they were unable to achieve this, they solved the problem of computational automation. The emergence of general automatic computing devices also led to the birth of the concept of artificial intelligence (AI) in 1956. Since then, all developments in AI technology have been based on a new generation of computing devices and stronger computing capabilities.
The second stage of the development of intelligent computing was the logical reasoning expert system (1990). E. Scientists from the Symbolic Intelligence School, such as Edward Albert Feigenbaum, proposed an expert system that can perform logical reasoning on knowledge symbols, with the main goal of automating logic and reasoning abilities. The prior knowledge of humans enters computers in the form of knowledge symbols, enabling computers to assist humans in making certain logical judgments and decisions in specific fields. However, expert systems heavily rely on manually generated knowledge bases or rule bases. The typical representatives of such expert systems are Japan's fifth generation computers and the 306 intelligent computer theme supported by China's 863 program. Japan uses dedicated computing platforms and knowledge reasoning languages such as Prolog to complete application level reasoning tasks in logical expert systems; China has taken a different technological route from Japan, based on a universal computing platform, turning intelligent tasks into artificial intelligence algorithms, integrating hardware and system software into the universal computing platform, and giving birth to a group of backbone enterprises such as Shuguang, Hanwang, and iFlytek.
The limitation of symbolic computing systems lies in their explosive computational spatiotemporal complexity, which means that symbolic computing systems can only solve linear growth problems and cannot solve high-dimensional complex space problems, thereby limiting the size of the problems they can handle. At the same time, because the symbolic computing system is built on knowledge rules, we cannot exhaustively enumerate all common sense, which greatly limits its application scope. With the arrival of the second AI winter, the first generation of intelligent computers gradually exited the historical stage.
Until around 2014, intelligent computing advanced to the third stage - deep learning computing systems. The Connected Intelligence School, represented by Geoffrey Hinton and others, aimed to automate learning abilities and invented new AI algorithms such as deep learning. Through the automatic learning of deep neural networks, the statistical induction ability of models has been greatly improved, and significant breakthroughs have been made in application effects such as pattern recognition. In some scenarios, the recognition accuracy even exceeds that of humans. Taking facial recognition as an example, the entire training process of the neural network is equivalent to a process of adjusting network parameters. A large amount of annotated facial image data is input into the neural network, and then the parameters between the networks are adjusted to make the probability of the output results of the neural network infinitely close to the real results. The higher the probability of the neural network outputting the true situation, the larger the parameters, thus encoding knowledge and rules into the network parameters. This way, as long as there is enough data, it can learn a large amount of common sense, greatly improving its generality. The application of connected intelligence is more extensive, including speech recognition, facial recognition, autonomous driving, etc. In terms of computing carrier, the Institute of Computing Technology of the Chinese Academy of Sciences proposed the world's first deep learning processor architecture in 2013, and the internationally renowned hardware manufacturer NVIDIA has continuously released a number of general-purpose GPU chips with leading performance, which are typical representatives of deep learning computing systems.
The fourth stage of the development of intelligent computing is the large-scale model computing system (2020). Driven by artificial intelligence big model technology, intelligent computing has reached new heights. In 2020, AI shifted from "small model+discriminative" to "large model+generative", upgrading from traditional facial recognition, object detection, and text classification to current text generation, 3D digital human generation, image generation, speech generation, and video generation. A typical application of the big language model in the field of dialogue system is the ChatGPT of OpenAI. It uses the big language model GPT-3 of the pre training base and introduces 300 billion words of training corpus, which is equivalent to the sum of all English words on the Internet. The basic principle is to train the model by giving it an input to predict the next word, improve prediction accuracy through extensive training, and ultimately ask it a question. The large model generates an answer and engages in real-time conversation with the person. On the basis of the base model, some prompt words are given to it for supervised instruction fine-tuning. Through human, the model gradually learns how to engage in multiple rounds of dialogue with humans; Finally, reinforcement learning iterations are carried out through manually designed and automatically generated reward functions to gradually align the large model with human values.
The characteristic of a big model is to win with "big", which has three layers of meaning: (1) large parameters, GPT-3 has 170 billion parameters; (2) The training data is large, with ChatGPT using approximately 300 billion words and 570GB of training data; (3) The computing power demand is high, and GPT-3 requires approximately tens of thousands of V100 GPUs for training. In order to meet the explosive demand for intelligent computing power in large models, both domestic and foreign companies are building new intelligent computing centers on a large scale with huge costs. Nvidia has also launched a large model intelligent computing system composed of 256 H100 chips and 150TB of massive GPU memory.
The emergence of large models has brought about three changes. One is the scaling law in technology, which states that the accuracy of many AI models rapidly improves when the parameter size exceeds a certain threshold. The reason for this is not yet very clear in the scientific community and is highly controversial. The performance of AI models has a logarithmic linear relationship with three variables: model parameter size, dataset size, and total computing power. Therefore, increasing the size of the model can continuously improve its performance. At present, the parameter count of the most advanced large-scale model GPT-4 has reached the order of trillions to trillions, and is still growing; The second is the explosive growth of computing power demand in the industry. Training large-scale models with billions of parameters usually requires 2-3 months of training on thousands or even tens of thousands of GPU cards. The sharp increase in computing power demand has driven the rapid development of related computing power enterprises. Nvidia's market value is close to $2 trillion, which has never happened to chip companies before; The third is the impact on the labor market in society. The report "Potential Impact of AI Models on China's Labor Market" jointly released by the National Development Research Institute of Peking University and Zhilian Recruitment points out that among the 20 most affected professions, accounting, sales, and clerical work are among the top. Physical labor jobs that require dealing with people and providing services, such as human resources, administration, logistics, etc., are relatively safer.
The technological frontier of artificial intelligence will develop in the following four directions. The first frontier direction is multimodal large models. From a human perspective, human intelligence is naturally multimodal, with eyes, ears, nose, tongue, body, and mouth (language). From an AI perspective, vision, hearing, and other aspects can also be modeled as sequences of tokens ②, which can be learned using the same methods as large language models and further aligned with semantics in language to achieve intelligent multimodal alignment.
The second frontier direction is the generation of large models for videos. OpenAI released the SORA video model on February 15, 2024, which significantly increased the video generation time from a few seconds to one minute, and showed significant improvements in resolution, image realism, and temporal consistency. The greatest significance of SORA is that it possesses the basic characteristics of a world model, namely the ability of humans to observe the world and further predict it. The world model is built on the basic physical knowledge of understanding the world (such as water flowing downwards), and then observing and predicting what events will happen in the next second. Although there are still many problems for SORA to become a world model, it can be considered that SORA has learned visual imagination and minute level future prediction ability, which are the fundamental features of world models.
The third frontier direction is embodied intelligence. Embodied intelligence refers to intelligent agents that have a body and support interaction with the physical world, such as robots, unmanned vehicles, etc. By processing various sensor data inputs through multimodal large models, the large models generate motion instructions to drive the intelligent agents, replacing traditional rule-based or mathematical formula based motion driving methods, achieving deep integration of virtual and reality. Therefore, robots with embodied intelligence can gather the three major schools of artificial intelligence: connectionism represented by neural networks, symbolism represented by knowledge engineering, and behaviorism related to control theory. These three schools can simultaneously act on an intelligent agent, which is expected to bring new technological breakthroughs.
The fourth frontier direction is AI4R (AI for Research) becoming the main paradigm for scientific discovery and technological invention. The current scientific discoveries mainly rely on experiments and human brain intelligence, where humans make bold guesses and carefully verify. Information technology, whether it is computation or data, only plays a role in assisting and verifying. Compared to humans, artificial intelligence has significant advantages in memory, high-dimensional complexity, full field of view, deep reasoning, and speculation. Can AI be used as the main tool for scientific discoveries and technological inventions, greatly improving the efficiency of human scientific discoveries, such as actively discovering physical laws, predicting protein structures, designing high-performance chips, and efficiently synthesizing new drugs. Because artificial intelligence models have full data and a god's perspective, with the ability of deep learning, they can take more steps forward than humans. If they can achieve a leap from inference to reasoning, artificial intelligence models have the potential to possess Einstein like imagination and scientific conjecture ability, greatly improving the efficiency of human scientific discovery and breaking down human cognitive boundaries. This is where the true subversion lies.
Finally, Artificial General Intelligence (AGI) is a highly challenging and controversial topic. A philosopher and a neuroscientist once made a bet: Will researchers be able to reveal how the brain achieves consciousness in 25 years (i.e. 2023)? At that time, there were two schools of thought about consciousness, one called the integrated information theory and the other called the global network workspace theory. The former believed that consciousness is a "structure" formed by the connection of specific types of neurons in the brain, while the latter pointed out that consciousness is generated when information spreads to brain regions through interconnected networks. In 2023, adversarial experiments were conducted in six independent laboratories, and the results did not fully match either theory. Philosophers won, while neuroscientists lost. Through this bet, it can be seen that people always hope that artificial intelligence can understand the mysteries of human cognition and the brain. From the perspective of physics, after gaining a thorough understanding of the macroscopic world, physics began to understand the microscopic world from the perspective of quantum physics. The intelligent world, like the physical world, is a research object with enormous complexity. AI big models still use data-driven methods to study the macro world, improve the intelligence level of machines, and have insufficient understanding of the intelligent macro world. It is difficult to directly search for answers in the micro world of the nervous system. Since its inception, artificial intelligence has been carrying various dreams and fantasies of human intelligence and consciousness, and has also inspired people to constantly explore.
3、 Security risks of artificial intelligence
The development of artificial intelligence has promoted technological progress in today's world, but it has also brought many security risks that need to be addressed from both technical and regulatory perspectives.
The first is the proliferation of false information on the Internet. Here are several scenarios: one is the digital avatar. AI Yoon is the first official "candidate" synthesized using DeepFake technology. This digital figure is based on Yoon Suk yeol, the candidate of the National Power Party in South Korea. With the help of Yoon Suk yeol's 20 hours of audio and video clips, as well as more than 3000 sentences recorded specifically for researchers, a local DeepFake technology company created a virtual image AI Yoon, which quickly became popular on the internet. In fact, the content expressed by AI Yoon was written by the campaign team, not the candidate themselves.
The second is the forgery of videos, especially those of leaders, which can cause international disputes, disrupt election order, or trigger sudden public opinion events, such as the forgery of Nixon's announcement of the failure of the first moon landing and the forgery of Ukrainian President Zelensky's announcement of "surrender". These actions have led to a decline in social trust in the news media industry.
The third is news forgery, mainly through the automatic generation of fake news to seek illegal benefits, using ChatGPT to generate hot news and earn traffic. As of June 30, 2023, there are 277 websites worldwide that generate fake news, seriously disrupting social order.
The fourth is changing faces and voices, used for fraud. A Hong Kong international company was defrauded of $35 million due to AI voice imitating the voices of corporate executives.
The fifth is to generate indecent images, especially for public figures. The production of pornographic videos by film and television stars has caused negative social impact. Therefore, it is urgent to develop the forgery detection technology of false information on the Internet.
Secondly, AI big models face serious credibility issues. These issues include: (1) factual errors of "serious nonsense"; (2) Narrating with Western values, exporting political biases and erroneous statements; (3) Easy to be induced, outputting incorrect knowledge and harmful content; (4) The issue of data security has worsened, and big models have become important traps for sensitive data. ChatGPT incorporates user input into the training database to improve ChatGPT. The US can use big models to obtain Chinese language corpus that cannot be covered by public channels, and master "Chinese knowledge" that we may not even have mastered ourselves. Therefore, there is an urgent need to develop large-scale model security supervision technology and one's own trustworthy large-scale model.
In addition to technological means, the security of artificial intelligence requires relevant legislative work. In 2021, the Ministry of Science and Technology issued the Code of Ethics for the New Generation of Artificial Intelligence. In August 2022, the National Technical Committee for Information Security Standardization issued the Code for Security Assessment of Machine Learning Algorithms for Information Security Technology. From 2022 to 2023, the Central Cyberspace Office successively issued the Regulations on the Management of Internet Information Service Algorithm Recommendations, the Regulations on the Management of Internet Information Service Deep Synthesis, and the Measures for the Management of Generative Artificial Intelligence Services. European and American countries have also introduced regulations. On May 25, 2018, the European Union issued the General Data Protection Regulation. On October 4, 2022, the United States released the Blueprint for the Artificial Intelligence Bill of Rights. On March 13, 2024, the European Parliament passed the European Union's Artificial Intelligence Act.
China should accelerate the promulgation of the Artificial Intelligence Law, establish an artificial intelligence governance system, ensure that the development and application of artificial intelligence follow the common values of humanity, and promote harmony and friendship between humans and machines; Create a policy environment conducive to the research, development, and application of artificial intelligence technology; Establish a reasonable disclosure mechanism and audit evaluation mechanism, understand the principles of artificial intelligence mechanisms and decision-making processes; Clarify the security responsibilities and accountability mechanisms of artificial intelligence systems, trace the responsible parties and remedy them; Promote the formation of fair, reasonable, open and inclusive international artificial intelligence governance rules.
4、 The development dilemma of intelligent computing in China
Artificial intelligence technology and intelligent computing industry are at the forefront of technological competition between China and the United States. Although China has made great achievements in the past few years, it still faces many development difficulties, especially those brought about by the US technology suppression policy.
The first dilemma is that the United States has long been in a leading position in AI core capabilities, while China is in a tracking mode. There is a certain gap between China and the United States in terms of the number of high-end AI talents, innovation in AI basic algorithms, AI base big model capabilities (big language models, text image models, text video models), base big model training data, and base big model training computing power, and this gap will continue for a long time.
The second dilemma is the ban on the sale of high-end computing products and the long-term stagnation of high-end chip technology. High end intelligent computing chips such as A100, H100, B200, etc. are prohibited from being sold to China. Huawei, Loongson, Cambrian, Shuguang, Haiguang and other companies have all entered the physical list. Their advanced chip manufacturing processes are limited, and the domestic process nodes that can meet mass production are 2-3 generations behind the international advanced level. The performance of core computing chips is 2-3 generations behind the international advanced level.
The third dilemma is the weak domestic intelligent computing ecosystem and insufficient penetration rate of AI development frameworks. NVIDIA CUDA ⑤ (Compute Unified Device Architecture) has a complete ecosystem and has formed a de facto monopoly. The domestic ecosystem is weak, manifested in: firstly, there is a shortage of R&D personnel. The NVIDIA CUDA ecosystem has nearly 20000 developers, which is 20 times the total number of personnel in all domestic smart chip companies; Secondly, there is a shortage of development tools. CUDA has 550 SDKs (Software Development Kit), which is hundreds of times more than related domestic enterprises; Thirdly, there is insufficient capital investment. Nvidia invests $5 billion annually, which is dozens of times more than related domestic companies; Fourthly, the AI development framework TensorFlow occupies the industrial market, PyTorch occupies the research market, and the developers of domestic AI development frameworks such as Baidu PaddlePaddle are only 1/10 of those of foreign frameworks. What is even more serious is that domestic enterprises are surrounded by mountains and cannot form a joint force. Although there are related products in each layer, such as intelligent applications, development frameworks, system software, and intelligent chips, there is no deep adaptation between each layer, which cannot form a competitive technological system.
The fourth dilemma is that the cost and threshold of applying AI to the industry remain high. At present, AI applications in China are mainly concentrated in the Internet industry and some national defense fields. When AI technology is popularized and applied in all walks of life, especially when migrating from the Internet industry to non Internet industries, a lot of customization work needs to be carried out. The migration is difficult and the cost of single use is high. Finally, the number of talents in the field of AI in China is clearly insufficient compared to the actual demand.
5、 How to choose the path of developing intelligent computing in China
The choice of path for the development of artificial intelligence is crucial for China, as it relates to the sustainability of development and the ultimate international competitive landscape. The current cost of using artificial intelligence is very high, with Microsoft Copilot suite paying a monthly usage fee of $10, ChatGPT consuming 500000 kilowatt hours of electricity per day, and Nvidia B200 chip prices exceeding $30000. In general, China should develop affordable, secure and credible AI technology, eliminate the information poor in China, and benefit the "the Belt and Road" countries; Empowering various industries with low barriers to entry, enabling China's advantageous industries to maintain competitiveness, and enabling relatively backward industries to significantly narrow the gap.
Option 1: Should we adopt a unified technology system that is closed source and closed, or open source and open?
The intelligent computing industry is supported by a tightly coupled technological system, which refers to a series of technical standards and intellectual property rights that closely link materials, devices, processes, chips, complete machines, system software, application software, and other related technologies. There are three paths for the development of intelligent computing technology system in China:
One is to catch up with the A system led by the United States. Most Internet enterprises in China follow the GPGPU/CUDA compatible path, and many start-ups in the chip field try to be compatible with CUDA in terms of ecological construction, which is more realistic. Due to the limitations imposed by the United States on China's technology and chip bandwidth in terms of computing power, it is difficult to form a unified domestic ecosystem in terms of algorithms, and the maturity of the ecosystem is severely limited. In terms of data, there is a lack of high-quality Chinese data, which makes it difficult to narrow the gap between catch-up and leading players, and sometimes even further widens it.
The second is to build a dedicated closed B system. Building a closed ecosystem for enterprises in specialized fields such as military, meteorology, and justice, producing chips based on mature domestic processes, paying more attention to specific vertical models in specific fields compared to base models, and training large models using domain specific high-quality data. This path is easy to form a complete and controllable technological system and ecology. Some large backbone enterprises in China are taking this path, but its disadvantage is that it is closed, unable to gather the majority of domestic forces, and difficult to achieve globalization.
The third is to jointly build an open and open-source C system globally. Breaking the ecological monopoly with open source, lowering the threshold for enterprises to possess core technologies, allowing every enterprise to make their own chips at low cost, forming a vast ocean of intelligent chips, and meeting the ubiquitous demand for intelligence. By forming a unified technological system through openness, Chinese enterprises and globalization forces have joined forces to jointly build a unified intelligent computing software stack based on international standards. Establish a pre competition sharing mechanism for enterprises, share high-quality databases, and share open-source universal base models. For the global open source ecosystem, Chinese enterprises have gained a lot in the Internet era, and they are more users and participants. In the intelligent era, Chinese enterprises should become more major contributors to the RISC-V ⑥+AI open source technology system, and become the leading force of global open sharing.
Option 2: Should we focus on algorithmic models or new infrastructure?
Artificial intelligence technology should empower various industries and have a typical long tail effect. 80% of small and medium-sized enterprises in our country require low threshold and low-priced intelligent services. Therefore, China's intelligent computing industry must be built on a new data space infrastructure, with the key being that China should take the lead in achieving comprehensive infrastructure of intelligent elements such as data, computing power, and algorithms. This work can be compared to the historical role of the American Information Highway Program (i.e., the construction of information infrastructure) in the Internet industry at the beginning of the 20th century.
The most essential productivity of the information society is cyberspace. The evolution process of cyberspace is: from the computing space composed of machine unary connections, to the information space composed of human-machine information binary connections, and then to the data space composed of human-machine object data ternary connections. From the perspective of data space, the essence of AI is that data is tempered into steel, and the big model is the product of in-depth processing of the full amount of Internet data. In the digital era, information flow is transmitted on the Internet, which is a structured abstraction after rough processing of data; In the intelligent era, what is transmitted on the Internet is an intelligent stream, which is a model-based abstraction after deep processing and refining of data by computing power. A core feature of intelligent computing is the use of algorithms such as numerical computation, data analysis, and artificial intelligence to process massive data pieces in a computing power pool, obtain intelligent models, and embed them into various processes in the information and physical worlds.
The Chinese government has proactively laid out new infrastructure and seized the opportunity in competition among countries around the world. Firstly, data has become a national strategic information resource. Data has dual attributes of resource elements and value processing. The resource element attributes of data include production, acquisition, transmission, aggregation, circulation, transaction, ownership, assets, security, and other links. China should continue to increase efforts to build a national data hub and data circulation infrastructure.
Secondly, AI big models are a type of algorithmic infrastructure in the data space. Based on the universal large model, we will build the infrastructure for the research and application of large models, support the development of specialized large models in the field of enterprise research and development, and serve industries such as robotics, autonomous driving, wearable devices, smart homes, and intelligent security, covering long tail applications.
Finally, the construction of a nationwide integrated computing power network has played a leading role in promoting the infrastructure of computing power. The Chinese plan for computing infrastructure should significantly reduce the cost and threshold of using computing power, while providing high-throughput and high-quality intelligent services to the widest range of people. The Chinese solution for computing infrastructure needs to have "two lows and one high", which means on the supply side, significantly reducing the total cost of computing devices, computing equipment, network connections, data acquisition, algorithm model calling, power consumption, operation and maintenance, and development and deployment, so that the majority of small and medium-sized enterprises can afford high-quality computing services and have the initiative to develop computing network applications; On the consumer side, the threshold for using computing power for a large number of users must be significantly reduced. Public services aimed at the general public must be easy to obtain and use, like water and electricity, ready to use, easy to customize computing power services like writing web pages, and develop computing power network applications. In terms of service efficiency, China's computing power services aim to achieve low entropy and high throughput, where high throughput refers to the high response time of end-to-end services while achieving high concurrency; Low entropy refers to ensuring that system throughput does not sharply decrease in the event of disorderly resource competition in high concurrency loads. Ensuring that 'it counts' is particularly important for China.
Option three: Does AI+focus on empowering the virtual economy or on empowering the real economy?
The effectiveness of "AI+" is the touchstone of the value of artificial intelligence. After the subprime mortgage crisis, the proportion of added value of the US manufacturing industry to GDP decreased from 28% in 1950 to 11% in 2021, and the proportion of employment in the US manufacturing industry decreased from 35% in 1979 to 8% in 2022. This shows that the US tends to favor the virtual economy with higher returns and neglects the real economy with high investment costs and low economic returns. China tends to promote the synchronous development of the real economy and the virtual economy, placing greater emphasis on the development of real economies such as equipment manufacturing, new energy vehicles, photovoltaic power generation, lithium batteries, high-speed rail, and 5G.
Correspondingly, AI in the United States is mainly applied in the virtual economy and IT infrastructure tools, and AI technology is also "moving away from reality and towards virtuality". Since 2007, Silicon Valley has continuously hyped up virtual reality (VR), metaverse, blockchain Web3.0、 Deep learning, AI big models, etc. are reflections of this trend.
Our country's advantage lies in the real economy, manufacturing industry, which has the most complete global industrial categories and systems, characterized by multiple scenarios and private data. China should select several industries to increase investment and form a paradigm that can be promoted across the entire industry with low barriers to entry, such as choosing equipment manufacturing as a representative industry for continuing advantages, and choosing the pharmaceutical industry as a representative industry for rapidly narrowing the gap. The technical difficulty in empowering the real economy lies in the integration of AI algorithms and physical mechanisms.
The key to the success of AI technology is whether the cost of an industry or a product can be significantly reduced, thus expanding the number of users and the scale of the industry by 10 times, producing a transformation effect similar to that of steam engines for the textile industry and smart phones for the Internet industry.
China should embark on a high-quality development path of empowering the real economy with artificial intelligence that suits itself.
① Pattern recognition refers to the use of computational methods to classify samples into certain categories based on their characteristics. It is the use of mathematical methods by computers to study the automatic processing and interpretation of patterns, collectively referring to the environment and objects as "patterns". The main research directions include image processing and computer vision, speech and language information processing, brain network group, and neuromorphic intelligence.
② Token can be translated as a morpheme, referring to a symbol used in natural language processing to represent words or phrases. A token can be a single character or a sequence of multiple characters.
③ General artificial intelligence refers to a type of artificial intelligence that is comparable to or even surpasses human intelligence. General artificial intelligence can not only possess basic thinking abilities such as perception, understanding, learning, and reasoning like humans, but also flexibly apply, rapidly learn, and creatively think in different fields. The research goal of general artificial intelligence is to seek a unified theoretical framework to explain various intelligent phenomena.
④ The chip manufacturing process refers to the manufacturing process of CPUs or GPUs, that is, the size of transistor gate circuits, measured in nanometers. Currently, the most advanced technology for mass production internationally is represented by TSMC's 3nm. More advanced manufacturing processes can integrate more transistors inside the CPU and GPU, giving the processor more functionality and higher performance, smaller area, and lower cost.
⑤ CUDA is a parallel computing platform and programming model designed and developed by NVIDIA, which includes the CUDA instruction set architecture and the GPU's internal parallel computing engine. Developers can use C language to write programs for the CUDA architecture, which can run at ultra-high performance on processors that support CUDA.
⑥ RISC-V (pronounced as "risk five") is an open and universal instruction set architecture initiated by the University of California, Berkeley. Compared to other paid instruction sets, RISC-V allows anyone to design, manufacture, and sell chips and software using the RISC-V instruction set for free.
⑦ The long tail effect refers to the phenomenon where products or services with small sales but a wide variety, which were not previously valued, accumulate a total revenue exceeding that of mainstream products due to their large quantity. In the Internet field, the long tail effect is particularly significant.
⑧ High concurrency usually refers to designing a system that can handle many requests in parallel at the same time.
The lecturer is an academician of the CAE Member, a researcher of the Institute of Computing Technology of the Chinese Academy of Sciences, and the director of the Academic Committee
Article source: China National People's Congress website
Global
Hot News
沪公网安备31011702889749号
