This article is reprinted from the WeChat public account: Chuangyi Think Tank, Wisdom World, source: China People’s Congress Network, China Information Technology Weekly, etc.,. Copyright belongs to the original authors and publishing media. If there is any infringement, please contact us immediately, and we will handle it promptly.
| Editor’s Note
The field of artificial intelligence has recently been experiencing explosive growth led by generative artificial intelligence large models. The choice of development path for artificial intelligence is crucial for our country, as it relates to sustainable development and the ultimate international competitive landscape.
Recently, Academician Sun Ninghui lectured to national and vice-national leadership teams on the topic of “The Development of Artificial Intelligence and Intelligent Computing,” briefly introducing the history of computer development and intelligent computing, raising concerns about the security risks of artificial intelligence, and analyzing the difficulties faced by China’s intelligent computing development, delving into how China can choose a path for developing intelligent computing.
In the face of the complex situation of Sino-U.S. technological competition, Academician Sun conducted an in-depth analysis of China’s development challenges in the field of intelligent computing, including insufficient core AI capabilities, bans on high-end computing products, a weak domestic intelligent computing ecosystem, insufficient penetration of AI development frameworks, and high costs and barriers when applying AI in various industries. Based on this, he proposed strategic choices for China’s development of intelligent computing, including building an open-source and open technology system, promoting the infrastructure of data and computing power, and the high-quality development path of AI empowering the real economy.
01
About the Academician
Sun Ninghui, academician of the Chinese Academy of Engineering, researcher at the Institute of Computing Technology, Chinese Academy of Sciences, and chairman of the Academic Committee. He also serves as the dean of the School of Computer Science and Technology at the University of Chinese Academy of Sciences, head of the strategic research expert group for the development roadmap in the field of information technology at the Chinese Academy of Sciences, and has served as the director of the Institute of Computing Technology and the National Intelligent Computer Research and Development Center, as well as vice president of the China Computer Federation. An expert in high-performance computing, his research areas include computer architecture and high-performance computing. He has received one first-class and four second-class National Science and Technology Progress Awards, the Outstanding Science and Technology Achievement Award from the Chinese Academy of Sciences, and honors such as the “China Youth Science and Technology Award” and “Top Ten Outstanding Youths in China”.
Academician Sun Ninghui is a leading academic figure in general-purpose high-performance computing. Previously, high-performance computers were only used in important fields, but his cluster technology has made high-performance computers common high-end computing equipment across various industries. He was the first to conduct cutting-edge basic research on high-throughput computers internationally, achieving several pioneering academic results at an international level, pushing China’s architectural academic research to enter the ranks of the world’s top.
02
Full Speech
The Development of Artificial Intelligence and Intelligent Computing
Sun Ninghui
Chairman, Vice Chairmen, Secretary General, and all members:
The field of artificial intelligence has recently been experiencing explosive growth led by generative artificial intelligence large models. On November 30, 2022, OpenAI launched an AI conversational chatbot, ChatGPT, whose outstanding natural language generation capabilities attracted widespread global attention, surpassing 100 million users in just two months. This sparked a wave of large models both domestically and internationally, with various large models such as Gemini, Wenxin Yiyan, Copilot, LLaMA, SAM, and SORA emerging like mushrooms after rain, making 2022 known as the Year of Large Models. The current information age is rapidly entering the development stage of intelligent computing, with breakthroughs in artificial intelligence technology emerging continuously, gradually empowering various industries, and making artificial intelligence and data elements typical representatives of new productive forces. General Secretary Xi Jinping pointed out that the new generation of artificial intelligence should be used as a driving force to promote leapfrog development in science and technology, optimize industrial upgrades, and achieve an overall leap in productivity, striving for high-quality development. Since the 18th National Congress of the Communist Party of China, the Party Central Committee, with Xi Jinping at its core, has attached great importance to the development of the intelligent economy, promoting the deep integration of artificial intelligence and the real economy, injecting strong momentum into high-quality development.
1.Overview of Computing Technology DevelopmentThe history of computing technology can be roughly divided into four stages: the emergence of the abacus marks the first generation—the mechanical computing era; the second generation—electronic computing is marked by the advent of electronic devices and electronic computers; the appearance of the internet brings us into the third generation—network computing; currently, human society is entering the fourth stage—intelligent computing.
Early computing devices were manual and semi-automatic computing devices. The history of human computing tools began with the Chinese abacus in 1200 AD, followed by the invention of Napier’s bones (1612) and the wheel-based calculator (1642), leading to the birth of the first automatic computing device— the step calculator in 1672, which completed basic arithmetic operations automatically.The mechanical computing period already saw some basic concepts of modern computers. Charles Babbage proposed designs for the Difference Engine (1822) and the Analytical Engine (1834), supporting automatic mechanical computation. During this period, the concepts of programming and programs were basically formed. The concept of programming originated from the Jacquard loom, which controlled printed patterns using punched cards, eventually evolving into a form of storing all mathematical calculation steps through computational instructions; the first programmer in human history was Ada Lovelace, Byron’s daughter, who wrote a set of instructions for solving Bernoulli numbers for Babbage’s Difference Engine, which is also the first computer algorithm program in human history, separating hardware from software and introducing the concept of programs for the first time.It wasn’t until the first half of the twentieth century that four scientific foundations of modern computing technology emerged: Boolean algebra (mathematics), Turing machines (computational models), von Neumann architecture (architecture), and transistors (devices). Among these, Boolean algebra is used to describe the underlying logic of programs and hardware like CPUs; Turing machines are a universal computational model that transforms complex tasks into automated computations without human intervention; the von Neumann architecture proposed three basic principles for constructing computers: using binary logic, storing and executing programs, and that computers consist of five basic units: arithmetic units, controllers, memory, input devices, and output devices; transistors are semiconductor devices that form the basic logic circuits and storage circuits, the “bricks” for building modern computers. Based on these scientific foundations, computing technology has developed rapidly, forming a large-scale industry.From the birth of the world’s first electronic computer ENIAC in 1946 to today in the 21st century, five categories of successful platform 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 scientific and engineering computing problems for core national departments; the second type is enterprise computing platforms, also known as servers, used for enterprise-level data management and transaction processing; currently, computing platforms of internet companies like Baidu, Alibaba, and Tencent all belong to this category; the third type is personal computer platforms, appearing in the form of desktop applications, allowing people to interact with personal computers through desktop applications; the fourth type is smartphones, characterized by mobility and portability, which connect to data centers via networks, primarily serving internet applications, distributed across data centers and mobile terminals; the fifth type is embedded computers, integrated into industrial equipment and military devices, ensuring the completion of specific tasks within a defined time through real-time control. These five types of devices almost cover all aspects of our information society, while the long-sought sixth type of platform computing system centered around intelligent computing applications has yet to be formed.The development of modern computing technology can be roughly divided into three eras. IT1.0, also known as the electronic computing era (1950-1970), is characterized by being “machine-centered”. The basic architecture of computing technology was formed, and with advancements in integrated circuit technology, the scale of basic computing units rapidly shrank, with transistor density, computing performance, and reliability continually improving, leading to widespread applications of computers in scientific engineering computing and enterprise data processing.IT2.0, also known as the network computing era (1980-2020), is characterized by being “human-centered”. The internet connects the terminals used by people with backend data centers, and internet applications interact with people through intelligent terminals. Internet companies like Amazon proposed the concept of cloud computing, encapsulating backend computing power as a public service for third-party users, forming the cloud computing and big data industries.IT3.0, also known as the intelligent computing era, began in 2020, adding the concept of “things” compared to IT2.0, meaning that various edge devices in the physical world are digitized, networked, and intelligent, achieving a triadic integration of “human-machine-thing.” In the intelligent computing era, in addition to the internet, there is also data infrastructure that supports various terminals through edge-cloud integration to achieve the Internet of Everything, with terminals, things, edges, and clouds all embedding AI, providing intelligent services similar to large models like ChatGPT, ultimately realizing that wherever there is computation, 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.Overview of Intelligent Computing Development
Intelligent computing includes artificial intelligence technology and its computing carriers, which have roughly gone through four stages: general computing devices, logical reasoning expert systems, deep learning computing systems, and large model computing systems.
The starting point of intelligent computing is the general automatic computing device (1946). Scientists like Alan Turing and John von Neumann initially hoped to simulate the process of the human brain processing knowledge, inventing machines that think like the human brain. Although this was not achieved, they solved the problem of automation in computation. The emergence of general automatic computing devices also promoted the birth of the concept of artificial intelligence (AI) in 1956, and all subsequent developments in artificial intelligence technology have been based on new-generation computing devices and stronger computing capabilities.
The second stage of intelligent computing development is the logical reasoning expert system (1990). Scientists from the symbolic intelligence school, such as Edward Albert Feigenbaum, aimed to automate logical reasoning capabilities, proposing expert systems capable of conducting logical reasoning with knowledge symbols. Human prior knowledge enters the computer in the form of knowledge symbols, enabling the computer to assist humans in making certain logical judgments and decisions in specific domains. However, expert systems heavily rely on manually generated knowledge bases or rule bases. A typical representative of this type of expert system is Japan’s fifth-generation computer and China’s 863 program-supported 306 intelligent computer. Japan adopted dedicated computing platforms and knowledge reasoning languages like Prolog to complete application-level reasoning tasks in logical expert systems; China took a different technological route, based on general computing platforms, transforming intelligent tasks into artificial intelligence algorithms, integrating both hardware and system software into general computing platforms, giving rise to key enterprises such as Shuguang, Hanvon, and iFlytek.
The limitations of symbolic computing systems lie in their explosive computational time-space complexity, meaning that symbolic computing systems can only solve linear growth problems and are unable to tackle high-dimensional complex spatial problems, thus limiting the size of problems they can handle. Additionally, since symbolic computing systems are based on knowledge rules, we cannot enumerate all common sense using exhaustive methods, greatly restricting their application range. With the arrival of the second AI winter, the first generation of intelligent computers gradually exited the historical stage.
It wasn’t until around 2014 that intelligent computing progressed to the third stage—deep learning computing systems. Represented by Geoffrey Hinton and others from the connectionist school, they aimed for automation of learning capabilities, inventing new AI algorithms such as deep learning. Through the automatic learning of deep neural networks, the model’s ability for statistical induction has greatly improved, achieving significant breakthroughs in applications such as pattern recognition, with some recognition accuracies even surpassing human capabilities. Taking facial recognition as an example, the entire training process of the neural network is akin to a process of adjusting network parameters, inputting a large number of labeled facial image data into the neural network, and then adjusting parameters among networks to make the output probability of the neural network approach the true results. The higher the probability of the neural network outputting the true situation, the larger the parameters, thereby encoding knowledge and rules into network parameters, allowing for learning a wide range of common sense as long as the data is sufficient, greatly enhancing universality. The applications of connectionist intelligence are more widespread, including speech recognition, facial recognition, and autonomous driving. In terms of computing carriers, the Institute of Computing Technology of the Chinese Academy of Sciences proposed the world’s first deep learning processor architecture in 2013, and internationally renowned hardware manufacturer NVIDIA has continuously released several leading general-purpose GPU chips, which are typical representatives of deep learning computing systems.
The fourth stage of intelligent computing development is the large model computing system (2020). Driven by large model technology in artificial intelligence, 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 today’s text generation, 3D digital human generation, image generation, speech generation, and video generation. A typical application of large language models in dialogue systems is OpenAI’s ChatGPT, which uses the pre-trained base large language model GPT-3, incorporating 300 billion words of training data, equivalent to the total amount of English text on the internet. Its basic principle is to train the model by giving it an input and predicting the next word, improving prediction accuracy through extensive training, ultimately allowing for real-time dialogue with the model by asking questions. On the basis of the base large model, it is then fine-tuned with supervised instruction prompts, gradually teaching the model how to engage in multi-turn conversations with people; finally, reinforcement learning iterations are conducted through manually designed and automatically generated reward functions to gradually align the large model with human values.
The characteristics of large models are defined by their “largeness,” which has three layers of meaning: (1) large parameters, with GPT-3 having 170 billion parameters; (2) large training data, with ChatGPT using approximately 300 billion words and 570GB of training data; (3) large computing power requirements, with GPT-3 requiring tens of thousands of V100 GPUs for training. To meet the explosive demand for intelligent computing power generated by large models, both domestic and international efforts are underway to build large-scale, costly new intelligent computing centers, and NVIDIA has also launched a large model intelligent computing system using 256 H100 chips and 150TB of massive GPU memory.
The emergence of large models has brought about three transformations. The first is the technological scaling law, which states that the accuracy of many AI models rapidly improves once the parameter scale exceeds a certain threshold, although the reasons for this in the scientific community are not yet fully understood and remain contentious. The performance of AI models is in a “logarithmic linear relationship” with the model parameter scale, dataset size, and total computing power, thus increasing the model scale can continuously improve model performance. Currently, the most advanced large model GPT-4 has reached trillions to tens of trillions of parameters and continues to grow; the second is the explosive growth of computing power demand in the industry, where training large models with hundreds of billions of parameters typically requires training on thousands to tens of thousands of GPU cards for 2-3 months, leading to a rapid increase in computing power demand that has driven related computing enterprises to develop at super speed, with NVIDIA’s market value nearing $2 trillion, a phenomenon never seen before in chip companies; the third is the impact on the labor market in society, as noted in a report by Peking University’s National Development Research Institute and Zhilian Recruitment, which pointed out that among the 20 occupations most affected, accounting, sales, and clerical work are at the forefront, while physical labor jobs that require interaction with people and providing services, such as human resources, administration, and logistics, 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 the human perspective, human intelligence is naturally multimodal, possessing eyes, ears, noses, tongues, bodies, and mouths (language). From the AI perspective, visual and auditory senses can also be modeled as sequences of tokens, using the same methods as large language models for learning, and further aligning with the semantics in language to achieve intelligent capabilities aligned across modalities.
The second frontier direction is video generation large models. On February 15, 2024, OpenAI released the video generation model SORA, significantly increasing video generation length from a few seconds to one minute, with notable improvements in resolution, visual realism, and temporal consistency. The greatest significance of SORA is that it possesses basic characteristics of a world model, meaning the ability to observe and predict the world. A world model is built on fundamental physical knowledge (e.g., water flows downhill) and can observe and predict what will happen in the next second. Although SORA still faces many issues to become a world model, it can be said that SORA has learned the ability to imagine visuals and predict events one minute into the future, which are foundational features of a world model.
The third frontier direction is embodied intelligence. Embodied intelligence refers to intelligent agents that have bodies and can interact with the physical world, such as robots and unmanned vehicles. They process various sensor data inputs through multimodal large models and generate motion commands to drive the agents, replacing traditional rule-based or mathematically formula-driven movement methods, achieving deep integration of the virtual and real worlds. Therefore, robots with embodied intelligence can integrate the three major schools of artificial intelligence: connectionism represented by neural networks, symbolicism represented by knowledge engineering, and behaviorism related to cybernetics, which is expected to bring about new technological breakthroughs.
The fourth frontier direction is AI4R (AI for Research) becoming the main paradigm for scientific discovery and technological invention. Currently, scientific discovery mainly relies on experimentation and human intelligence, involving bold conjectures and careful verification by humans, with information technology serving merely as auxiliary and verification tools. Compared to humans, artificial intelligence has significant advantages in memory, high-dimensional complexity, panoramic views, reasoning depth, and conjecture. The question remains whether AI can take the lead in scientific discoveries and technological inventions, greatly enhancing 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 large AI models have access to all data and possess a god-like perspective, they can look several steps ahead compared to humans. If a leap from inference to reasoning can be achieved, artificial intelligence models could potentially possess the imagination and scientific conjecture capabilities akin to Einstein, significantly enhancing the efficiency of human scientific discoveries and breaking through the boundaries of human cognition. This is the true disruption.
Lastly, general artificial intelligence (AGI) is a highly challenging and contentious topic. There was once a bet between a philosopher and a neuroscientist: whether researchers would be able to uncover how the brain achieves consciousness in 25 years (by 2023). At that time, there were two schools of thought regarding consciousness: one proposed the integrated information theory, which posits that consciousness is formed by specific types of neuron connections in the brain, while the other suggested that consciousness arises when information is transmitted through interconnected networks to brain regions. In 2023, adversarial experiments conducted across six independent laboratories yielded results that did not fully match either theory, with the philosopher winning and the neuroscientist losing. This bet illustrates humanity’s enduring hope that artificial intelligence can comprehend the mysteries of human cognition and the brain. From a physics perspective, physics involves a thorough understanding of the macroscopic world before embarking on understanding the microscopic world starting from quantum physics. The intelligent world, like the physical world, is also a research object of immense complexity, and AI large models still study the macroscopic world through data-driven methods, improving machine intelligence levels, but understanding the intelligent macroscopic world is insufficient. Directly seeking answers in the microscopic world of the neural system is challenging. Since its inception, artificial intelligence has carried humanity’s various dreams and fantasies about intelligence and consciousness, inspiring continuous exploration.
3.Security Risks of Artificial Intelligence
The development of artificial intelligence has promoted technological progress in today’s world, but it also brings many security risks that need to be addressed from both technological and regulatory perspectives.
Firstly, there is a flood of false information on the internet. Here are several scenarios: First is digital avatars. AI Yoon is the first official candidate synthesized using DeepFake technology, modeled after the South Korean National Power Party candidate Yoon Suk-yeol, created by a local DeepFake technology company using 20 hours of audio and video clips of Yoon, as well as over 3,000 sentences specifically recorded for researchers. AI Yoon quickly gained popularity online, but the content expressed by AI Yoon was written by the campaign team, not the candidate himself.
Second is forged videos, especially fabricated videos of leaders that can cause international disputes, disrupt election orders, or trigger sudden public opinion events, such as the forgery of Nixon announcing the failure of the first moon landing or the forgery of Ukrainian President Zelensky announcing “surrender,” leading to a decline in social trust in the media industry.
Third is forged news, primarily generated through fake news for illegal profit, using ChatGPT to generate trending news and earn traffic. As of June 30, 2023, there were 277 global websites generating fake news, severely disrupting social order.
Fourth is face and voice swapping, used for fraud. For example, due to AI mimicking the voice of a corporate executive, an international company in Hong Kong was defrauded of $35 million.
Fifth is generating inappropriate images, particularly targeting public figures, such as the production of pornographic videos of film stars, causing negative social impacts. Therefore, there is an urgent need to develop technologies for detecting forgery of internet false information.
Secondly, AI large models face serious credibility issues. These issues include: (1) factual errors that “speak seriously nonsense”; (2) narratives based on Western values, outputting political biases and erroneous statements; (3) susceptibility to inducement, outputting incorrect knowledge and harmful content; (4) data security issues exacerbated, with large models becoming significant traps for sensitive data, as ChatGPT incorporates user inputs into its training database for improvement, allowing foreign entities to utilize large models to access Chinese language data that may not be available through public channels, gaining knowledge that we may not even possess. Therefore, there is an urgent need to develop safety regulation technologies for large models and our own credible large models.
Besides technological measures, the security guarantees of artificial intelligence require relevant legislative work. In 2021, the Ministry of Science and Technology released the “Ethical Norms for the New Generation of Artificial Intelligence,” and in August 2022, the National Information Security Standardization Technical Committee released the “Information Security Technology Machine Learning Algorithm Security Assessment Specification.” Between 2022 and 2023, the Central Cyberspace Affairs Commission successively released regulations on “Algorithm Recommendation Management for Internet Information Services,” “Deep Synthesis Management for Internet Information Services,” and “Management Measures for Generative Artificial Intelligence Services.” European and American countries have also successively introduced regulations, with the European Union issuing the “General Data Protection Regulation” on May 25, 2018, and the United States publishing the “Blueprint for the Artificial Intelligence Bill of Rights” on October 4, 2022, while the European Parliament passed the EU “Artificial Intelligence Act” on March 13, 2024.
Our country should accelerate the introduction of the “Artificial Intelligence Law,” build an artificial intelligence governance system, ensure that the development and application of artificial intelligence adhere to common human values, promote harmonious and friendly human-machine relationships; create a policy environment conducive to the research, development, and application of artificial intelligence technologies; establish reasonable disclosure mechanisms and audit assessment mechanisms to understand the principles and decision-making processes of artificial intelligence mechanisms; clarify the safety responsibilities and accountability mechanisms of artificial intelligence systems, traceable responsible entities, and remedies; and promote the formation of fair, reasonable, open, and inclusive international governance rules for artificial intelligence.
4.Difficulties in China’s Intelligent Computing Development
The technology and intelligent computing industry are at the forefront of Sino-U.S. technological competition. Although our country has made significant achievements in recent years, it still faces numerous development difficulties, especially those arising from the U.S. technology suppression policies.
The first difficulty is that the U.S. has long been leading in core AI capabilities, while China is in a tracking mode. China lags behind the U.S. in the number of high-end AI talents, innovation of basic AI algorithms, capabilities of foundational large models (large language models, text-to-image models, text-to-video models), training data for foundational large models, and training computing power for foundational large models, and this gap is expected to persist for a long time.
The second difficulty is the ban on high-end computing products, with high-end chip processes long being restricted. High-end intelligent computing chips like A100, H100, and B200 are banned from sale to China. Companies such as Huawei, Loongson, Cambricon, Shuguang, and Haiguang have all been placed on the Entity List, and their advanced chip manufacturing processes are restricted. The domestic processes capable of meeting large-scale production are 2-3 generations behind international advanced levels, and the performance of core computing power chips lags 2-3 generations behind international advanced levels.
The third difficulty is the weak domestic intelligent computing ecosystem and insufficient penetration of AI development frameworks. The NVIDIA CUDA ecosystem is complete and has formed a de facto monopoly. The domestic ecosystem is weak, specifically manifested in: (1) insufficient R&D personnel; the NVIDIA CUDA ecosystem has nearly 20,000 developers, which is 20 times the total number of personnel in all domestic intelligent chip companies; (2) insufficient development tools, with CUDA having 550 SDKs, which is hundreds of times that of domestic related companies; (3) insufficient funding investment, with NVIDIA investing $5 billion annually, which is dozens of times that of domestic related companies; (4) the AI development framework TensorFlow occupies the industrial market, and PyTorch occupies the research market, while domestic AI development frameworks like Baidu’s PaddlePaddle have only 1/10 the number of developers compared to foreign frameworks. More seriously, domestic companies are fragmented, unable to form a concerted effort, with related products at each level, but lacking deep adaptation between layers, making it impossible to form a competitive technological system.
The fourth difficulty is that the costs and barriers of applying AI in industries are high. Currently, AI applications in our country are primarily concentrated in the internet industry and some defense sectors. Promoting AI technology applications across various industries, especially migrating from the internet industry to non-internet industries, requires extensive customization work, making migration difficult and incurring high one-time usage costs. Lastly, the number of talents in the AI field in our country is also significantly insufficient compared to actual demand.
5.How China Can Choose a Path for Developing Intelligent Computing
The choice of development path for artificial intelligence is crucial for our country, as it relates to sustainable development and the ultimate international competitive landscape. Currently, the cost of using artificial intelligence is extremely high; for example, the Microsoft Copilot suite requires a monthly fee of $10, ChatGPT consumes 500,000 kilowatt-hours of electricity daily, and the price of NVIDIA’s B200 chip exceeds $30,000. In general, our country should develop affordable, safe, and trustworthy artificial intelligence technologies, eliminate information poverty for our population, and benefit countries along the Belt and Road; empower various industries at a low threshold to maintain the competitiveness of our advantageous industries and significantly narrow the gap for relatively backward industries.
Choice 1: Should we pursue a closed-source system compatible with the U.S. dominant A system or an open-source system?
The intelligent computing industry is supported by an intricately coupled technological system, a whole that closely connects materials, devices, processes, chips, whole machines, system software, and application software through a series of technical standards and intellectual property. There are three paths for our country to develop an intelligent computing technology system:
First, to catch up and be compatible with the U.S.-led A system. Most of our internet companies follow the GPGPU/CUDA compatible path, and many startup companies in the chip field also strive to be compatible with CUDA in ecosystem building, which is a more realistic path. Due to U.S. restrictions on our processes and chip bandwidth in terms of computing power, the domestic ecosystem is fragmented and it is difficult to form a unified ecosystem, severely limiting ecological maturity. Additionally, the lack of high-quality Chinese data makes it difficult for followers to narrow the gap with leaders, and at times may even widen it further.
Second, to build a dedicated closed B system. In specialized fields such as military, meteorology, and judiciary, we can build enterprise closed ecosystems based on domestically mature processes to produce chips, focusing more on specific vertical large models rather than foundational large models, and training large models using domain-specific high-quality data. This path is easier to form a complete and controllable technological system and ecosystem. Some large backbone enterprises in our country are following this path, but its downside is that it is closed, unable to gather most domestic forces, and is difficult to achieve globalization.
Third, to globally co-build an open-source C system. We can break ecological monopolies with open-source, lowering the threshold for enterprises to possess core technologies, enabling every enterprise to create its own chips at low cost, forming a vast ocean of intelligent chips to meet ubiquitous intelligent needs. By using openness to form a unified technological system, our enterprises can join forces with global powers to co-build a unified intelligent computing software stack based on international standards. This would create a pre-competitive mechanism for enterprise competition, sharing high-quality databases and open-source general foundational large models. In the global open-source ecosystem, our enterprises have greatly benefited during the internet era; while we were primarily users and participants, in the intelligent era, our enterprises should become major contributors to the RISC-V+AI open-source technology system, becoming leading forces in global open sharing.
Choice 2: Should we focus on algorithm models or new infrastructure?
Artificial intelligence technology must empower various industries, exhibiting a typical long-tail effect. Our 80% of small and medium-sized enterprises need low-threshold, low-cost intelligent services. Therefore, our intelligent computing industry must be built on new data space infrastructure, with the key being that our country should take the lead in achieving comprehensive infrastructure for intelligent elements—data, computing power, and algorithms. This task can be compared to the historical role of the U.S. information highway plan in the early twentieth century (i.e., the construction of information infrastructure) for the internet industry.
The core productive force of the information society is cyberspace. The evolution of cyberspace is from a computing space formed by unidirectional connections of machines to an information space formed by bidirectional connections of humans and machines, and then to a data space formed by triadic connections of humans, machines, and things. From the perspective of data space, the essence of artificial intelligence is the refinement of data, and large models are products of deep processing of all internet data. In the digital age, what is transmitted over the internet is information flow, which is a structured abstraction of data processed roughly by computing power; in the intelligent age, what is transmitted over the internet is intelligent flow, which is a modeled abstraction of data processed deeply and refined by computing power. A core feature of intelligent computing is processing massive data items using numerical computation, data analysis, artificial intelligence, and other algorithms in a computing power pool to obtain intelligent models, which are then embedded into various processes in the information world and physical world.
The government of our country has proactively laid out new infrastructure, seizing opportunities in global competition. First, data has become a national strategic information resource. Data has dual attributes of resource elements and value processing, with the resource element attributes of data including production, acquisition, transmission, aggregation, circulation, trading, ownership, assets, security, etc. Our country should continue to strengthen the construction of national data hubs and data circulation infrastructure.
Second, AI large models are a type of algorithmic infrastructure in the data space. Using general large models as a base, we can build the infrastructure for the research and application of large models, supporting a wide range of enterprises to develop domain-specific large models, serving 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 pioneering role in promoting the infrastructure of computing power. The Chinese solution for the infrastructure of computing power 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 power infrastructure needs to achieve “two lows and one high,” meaning on the supply side, significantly reducing the total costs of computing devices, computing equipment, network connections, data acquisition, algorithm model invocation, power consumption, operational maintenance, and development deployment, making high-quality computing services affordable for small and medium-sized enterprises, encouraging them to develop computing network applications; on the consumer side, significantly lowering the threshold for users to use computing power, ensuring that public services accessible to the masses are easy to obtain and use, ready to use like water and electricity, and easily customizable like writing web pages, developing computing network applications. In terms of service efficiency, China’s computing power services should achieve low entropy and high throughput, where high throughput refers to the ability to provide high concurrency services while ensuring a high satisfaction rate for end-to-end service response times; low entropy refers to ensuring system throughput does not sharply decline under high concurrency loads where resource competition becomes disorderly. Ensuring “ample computation” is particularly important for China.
Choice 3: Should we focus on empowering the virtual economy or strengthen the real economy? The effectiveness of “AI+” is a touchstone for the value of artificial intelligence. After the subprime mortgage crisis, the manufacturing value added in the U.S. dropped from 28% of GDP in 1950 to 11% in 2021, and the proportion of employment in manufacturing declined from 35% in 1979 to 8% in 2022, indicating that the U.S. tends to favor the virtual economy with higher returns while neglecting the real economy, which has higher investment costs and lower economic returns. China tends to develop the real economy alongside the virtual economy, placing greater emphasis on the development of equipment manufacturing, new energy vehicles, photovoltaic power, lithium batteries, high-speed rail, and 5G in the real economy.
Correspondingly, AI in the U.S. is primarily applied to the virtual economy and IT infrastructure tools, reflecting a trend of “detaching from reality.” Since 2007, Silicon Valley has continuously hyped virtual reality (VR), the metaverse, blockchain, Web 3.0, deep learning, and large AI models, reflecting this trend.
China’s advantage lies in the real economy, with the most complete global industrial categories in manufacturing, characterized by diverse scenarios and abundant private data. We should select several industries to increase investment, forming paradigms that can be promoted across all industries at low thresholds, such as choosing equipment manufacturing as a representative industry to continue our advantages and selecting the pharmaceutical industry as a representative industry to rapidly narrow the gap. The technical challenge of empowering the real economy lies in the integration of AI algorithms with physical mechanisms.
The key to the success of artificial intelligence technology is whether it can significantly reduce the costs of an industry or product, thereby expanding user numbers and industry scale tenfold, achieving transformational effects similar to those of the steam engine on the textile industry or the smartphone on the internet industry.
Our country should carve out a high-quality development path for artificial intelligence to empower the real economy.
6. Notes①Pattern recognition refers to the use of computational methods to classify samples based on their characteristics, involving the automatic processing and interpretation of patterns through mathematical methods by computers, with image processing, computer vision, speech language information processing, brain networks, and brain-like intelligence as major research directions.②Token refers to symbols used to represent words or phrases during the natural language processing process. Tokens can be individual characters or sequences composed of multiple characters.③General artificial intelligence refers to a type of artificial intelligence that possesses intelligence comparable to or exceeding that of humans. General artificial intelligence can perform basic cognitive abilities such as perception, understanding, learning, and reasoning similarly to humans and can flexibly apply, learn quickly, and think creatively across different fields. The research goal of general artificial intelligence is to seek a unified theoretical framework to explain various intelligent phenomena.④Chip manufacturing processes refer to the processes used to manufacture CPUs or GPUs, specifically the size of transistor gate circuits, measured in nanometers. The most advanced manufacturing processes currently in mass production are represented by TSMC’s 3nm. More advanced manufacturing processes can integrate more transistors into CPUs and GPUs, providing more functions and higher performance while being smaller and cheaper.⑤CUDA is a parallel computing platform and programming model developed by NVIDIA, including the CUDA instruction set architecture and the parallel computing engine within GPUs. Developers can use C language to write programs for the CUDA architecture, which can run at super high performance on CUDA-supported processors.⑥RISC-V (pronounced “risk-five”) is an open general-purpose instruction set architecture initiated by the University of California, Berkeley, allowing anyone to freely use, design, manufacture, and sell chips and software based on the RISC-V instruction set, compared to other paid instruction sets.⑦Long-tail effect refers to the phenomenon where products or services that originally received little attention but have a large variety accumulate total revenues that exceed those of mainstream products due to their sheer volume. This effect is particularly evident in the internet field.⑧High concurrency typically refers to a design that ensures the system can handle many requests simultaneously and in parallel.
Scan to follow us