Introduction
Artificial intelligence is regarded as a hallmark of a new industrial revolution. Advanced economies and many technology companies have invested heavily in research and deployment. China is working to build an early advantage in AI development. The report of the 20th National Congress called for promoting the integrated development of strategic emerging industries and fostering a new generation of information technology, artificial intelligence, biotechnology, new energy, new materials, high-end equipment, and green technologies as new growth engines. What is artificial intelligence? How will it change our lives? How can organizations gain an early position in this technological shift?
What is artificial intelligence
Artificial intelligence has a history of roughly six to seven decades. It refers to machine-made systems that mimic human intelligence, typically manifested via robots, computers, and similar platforms. Two key concepts are general AI, or strong AI, and narrow AI, or weak AI. General AI denotes systems whose functions and capabilities match or exceed human abilities. Narrow AI refers to systems that perform a single task, potentially outperforming humans on that task, such as face recognition or speech recognition systems.
History of AI development
In 1956, ten young scholars gathered at Dartmouth College for a summer research workshop. Over two months they discussed what AI should do and how to pursue it, identifying research topics such as automatic computation, programming languages, neural networks, and computational complexity. That report is widely considered an early comprehensive description of AI, and 1956 is often regarded as AI’s founding year.
From 1956 to 1976, the first phase of AI focused on simulating higher-level brain functions such as logical reasoning rather than low-level signal simulation. A notable achievement from this period was automatic mathematical theorem proving, including algebraic and geometric theorem proofs completed by Chinese scholars Wang Hao and Wu Wenjun. Researchers later realized that simulating the human brain alone was insufficient, since goals like beating chess champions or composing music had not been achieved. AI declined from its initial peak and researchers shifted to more practical systems.
Researchers began developing socially impactful systems such as medical diagnostic expert systems and fault diagnosis systems to perform tasks previously done by doctors or specialists. Neural-system-inspired approaches proved effective in character recognition and handwriting recognition, later applied to machine vision and automated sorting systems. Modern speech recognition similarly follows neural network-based techniques.
These two technical routes led to a flourishing period for expert systems over approximately 30 years. However, expert systems did not sustain their early prominence because field deployments often underperformed compared with demonstrations. Interest waned until 2006, when three influential papers signaled a renewed AI surge.
In 2006, Geoffrey Hinton, Yann LeCun, and Yoshua Bengio published influential work showing deep neural networks were capable of large-scale learning. Progress continued, and in the early 2010s, Fei-Fei Li and others built a very large image database and used it in competitions where systems were judged by error rates. Error rates dropped from 28% in 2010 to 16% in 2012, when a system based on Hinton’s published techniques defeated other competitors. By 2015, error rates fell to 3.6%, below the average human error of 5% on image classification. A major advance that year was the residual network proposed by Kaiming He and his team, leading to state-of-the-art performance in image tasks and influencing other domains such as game-playing systems. These developments demonstrated that deep networks were ready for broad applied and industrial research where narrow AI is appropriate.
Characteristics and growth of AI
AI is progressing from perception toward cognition. Intelligence can be categorized as perception intelligence, cognitive intelligence, and decision intelligence. Perception intelligence connects directly to sensory inputs such as vision and hearing. Current face recognition, speech recognition, machine translation, medical diagnosis, and defect detection have matured considerably. AI is gradually transitioning from general perception toward higher cognitive capabilities.
There are two potential breakthrough routes for scaling toward cognitive or general AI: brain-inspired computing and quantum computing. Current computers have much lower energy efficiency compared with the human brain. Brain-inspired architectures could improve energy efficiency for information processing. Quantum computing also offers potential energy-efficiency gains, but both directions remain uncertain and require further research and development.
Human-machine hybrid intelligence offers practical advantages. Human-machine collaborative models, where machines perform tasks they are good at and humans intervene when needed, represent an important technical pathway for current AI development.
Application-driven development helps technical progress. Earlier, technology was often developed first and later applied. Now applications often drive technical advances. For example, the deep neural network surge accelerated when Hinton’s student applied the technique to image recognition competitions. Application scenarios can reveal the practical value of new methods.
Precaution and governance are essential to prevent loss of control. AI has social attributes. Concerns about AI dominating or controlling humans must be taken seriously. Research and legislation should clarify the roles and limits of AI systems, regulated through legal and ethical frameworks to ensure safe, controllable deployment.
China's AI advantages and gaps
Since 2013, many governments have studied AI’s potential social and economic impacts and released national AI strategies. In this global technological shift, where are the opportunities and what are the strengths and weaknesses?
China has become one of the major AI powers. Since the reform and opening period, China has invested substantially in basic research, creating a strong foundation for AI development. Some core technologies, such as face and speech recognition, are globally competitive. AI has deeply penetrated many industries in China, and its application and deployment are often more widespread than in other countries. An initial innovation ecosystem has emerged, including autonomous driving efforts, city-scale analytics, intelligent healthcare, speech platforms, and companies focusing on image and video processing. Hardware and industrial platforms from companies such as Huawei, Cambricon, and Hikvision have also contributed. Global AI indices place China among the leading countries, following the United States, with progress in talent, education, and patent output. The overall global AI landscape remains led by the US and China in a tiered distribution.
Four major advantages support AI development in China. First, policy support: AI has been designated a national priority. Second, vast data resources: China’s population and scale of mobile usage generate large datasets across consumption, travel, healthcare, tourism, and logistics. Third, application scenarios: as a developing market, many infrastructure gaps create deep scenarios for AI intervention in areas such as urban-rural infrastructure, healthcare, education, and public services. Fourth, a large pool of young technical talent: higher education enrollment rates and a strong proportion of science and engineering students provide significant talent reserves. Funding agencies have established AI disciplines and funded foundational, exploratory, and applied research.
There are also notable weaknesses that require strengthening. First, original basic theory and algorithm research remains relatively weak. Second, R&D capability for high-end devices is limited; for example, domestic GPU development for deep training lags behind. Third, there is a shortage of influential open-source AI platforms. Fourth, high-end talent is insufficient; the number of top-tier AI experts is significantly lower than in the United States, creating a shortage in strategic talent areas.
How to plan for the future
Major technological shifts have already transformed society. Over the past 30 years, computing power has increased by roughly one millionfold, storage capacity by a millionfold, and communication speed by about a millionfold. These changes have dramatically altered how people live and work. The next industrial revolution may occur around 2030 to 2040, with AI as the central technology for the coming century.
Comprehensive planning is required, including national strategy, talent centers, infrastructure, and legal safeguards. Key measures include elevating AI development to a national strategic priority, establishing a robust national R&D system, accelerating talent cultivation to create national centers of excellence, strengthening intelligent infrastructure and data openness while ensuring data security through regulation, advancing research on AI law and ethics to guide safe development, and deepening international cooperation to participate in global governance and standards.
Three practical levers are critical: data handling, open platforms, and application scenarios. Data is fundamental: without well-organized, cleaned, and secured data, even narrow AI cannot function effectively. Second, developing strong open-source platforms aligns with China’s position as a major AI nation. Third, nurturing and enabling application scenarios is essential. Many current projects are investment-driven; governments should identify priority areas where intervention is necessary and let the market decide where appropriate. Targeted government support can accelerate maturation of strategic applications. After addressing domestic gaps, synchronizing with global AI developments is important to meet the next industrial transformation.
About Gao Wen
Gao Wen is an academician of the Chinese Academy of Engineering, a chair professor at Peking University, and director of the Pengcheng Laboratory. He has witnessed and participated in the rapid development of China’s computing industry and the growth of AI. He has led more than 20 national research projects in AI, video coding and analysis, and computer vision. His team at the National Engineering Laboratory for Digital Video Coding developed a second-generation source coding standard system through independent innovation, reaching international standard levels. In 2021, he and his project team received a national technology invention award for work on ultra-high-definition video coding. The AVS ultra-high-definition video coding national standard led by his team was adopted by an international ultra-HD industry alliance.
