China's AI Implementation Gap
Will China become the leader of the fourth industrial revolution? Debates about China's capabilities in artificial intelligence (AI) tend to center on its innovation capacity—a state's ability to produce new-to-the-world advances. This essay argues that more attention should be paid to China's ability to spread and adopt technical breakthroughs across a wide range of productive processes (diffusion capacity), which is especially important for general-purpose technologies that have a wide variety and range of uses like AI. The essay finds that China faces a significant "AI implementation gap," drawing on evidence from the limited adoption of other information and communication technologies, regulatory obstacles, and a shortage of skilled AI practitioners. Resolving this gap will be central to China's long-run prospects of sustaining its economic rise.
The arrival of ChatGPT, a remarkably powerful AI model that can understand and generate human-like text, has led many to declare that we are at the threshold of the next industrial revolution. Alongside these latest technological developments, China’s rise continues to be a key geopolitical trend as governments around the world manage the return of great power competition. At the intersection of these twin currents, technological competition between the U.S. and China has become an inescapable topic of debate. Who will lead the way in the next industrial revolution?
To answer this question, leading thinkers and policymakers around the world have tried to analyze China’s progress in AI. Unfortunately, their assessments have been shaped by an innovation-centric view of technological leadership. In contrast, this essay argues that China faces a significant “AI implementation gap” between the initial generation of new AI advances and widespread adoption of these technical advances throughout the entire economy.
The essay proceeds in three parts. First, it highlights that discussions about China’s AI capabilities tend to center on innovation capacity, or China’s ability to generate new-to-the-world AI breakthroughs. Second, it posits that when it comes to general-purpose technologies like AI, more attention should be paid to China’s ability to spread and adopt AI productive processes (diffusion capacity). Finally, it presents evidence that China faces significant barriers to diffusing AI across a wide range of application sectors.
Innovation-centric Assessments of China’s AI Capabilities
Debates about U.S.-China competition in AI are preoccupied with leading-edge innovation. The guiding question seems to be: Which country will produce the next AI breakthrough? China’s advances in AI are often framed as an opportunity for an emerging economy to catch up with and leapfrog over the advanced countries in key segments, like AI chips or autonomous vehicles. The Chinese government has also adopted this innovation-centered mindset. China’s national AI development plan outlines its ambition to become the world’s leading center of AI innovation by 2030.[1] Outside of China, analysts describe the country’s AI strategy as aimed at seizing “the commanding heights” of next-generation technologies, reflecting the belief that competition in AI will be over global market shares in strategic sectors.[2]
Innovation-centric views of China’s AI capabilities paint an overly optimistic picture of China’s challenge to U.S. technological leadership. For instance, a report titled “Is China Beating the U.S. to AI Supremacy?,” authored by Professor Graham Allison, director of Harvard Kennedy School’s Belfer Center for Science and International Affairs and Eric Schmidt, former CEO of Google, emphasizes China’s growing competencies in AI-related R&D investments, leading AI start-ups, and valuable internet companies.[3]
Consider also the viewpoint of the National Security Commission on Artificial Intelligence (NSCAI), an important body created by Congress to provide recommendations to the U.S. government on AI policy. In its final report, the NSCAI suggests that China is poised to overtake the United States in its capacity to develop novel AI advances, citing shares of top-cited, breakthrough papers in AI and investments in start-ups.[4]
Other comparisons of U.S. and Chinese AI capabilities come to the opposite conclusion, but they still rely on an innovation-centric perspective. For instance, two Oxford scholars, Carl Frey and Michael Osborne, compare claims that China will soon surpass the United States in AI capabilities to overestimates of Japan’s technological leadership in information and communications technologies in the 1980s. In their view, just as Japan failed to overtake the U.S. in technological leadership due to its inability to produce radical breakthroughs in computing, China’s technological challenge will fade because of its inability to generate foundational innovations in AI. In fact, they claim that the prospects are even bleaker this time around: “China, if anything, looks less likely to overtake the United States in artificial intelligence than Japan looked to dominate in computers in the 1980s.”[5] Again, the common thread in all these assessments is a preoccupation with which country will develop new-to-the-world breakthroughs.
Why Diffusion is Key to China’s AI Prospects
Not all technologies are created equal. Diffusion capacity matters more for AI than other emerging technologies due to the simple fact that a general-purpose technology (GPT) has the potential to transform many more industries. Recognized by economists and economic historians as “engines of growth” that can usher in waves of productivity growth, GPTs are defined by three characteristics. First, they offer great potential for continual improvements. While all technologies offer some scope for improvements, a GPT is often connected to a major research field that produces sustained adaptations. Second, GPTs acquire pervasiveness. As a GPT evolves, it will find a wide variety and range of uses. Third, GPTs have strong technological complementarities. In other words, the benefits from AI will depend on innovations in application industries. For example, electricity’s contribution to productivity growth was unlocked by complementary innovations in power generation, factory reorganization, and household appliances (e.g., refrigerators and washing machines).[6]
Among the emerging technologies that are flashpoints in U.S.-China technological rivalry, AI stands out as a GPT. Recent breakthroughs in deep learning have improved the ability of machines to learn from data in fundamental ways that can be applied across hundreds of domains, including medicine, transportation, and other enabling technologies such as biotechnology and robotics. This is why AI is often called the “new electricity”—a comparison to the prototypical GPT. Using job posting data as an early indicator of general-purpose potential, one study finds that machine learning technologies are more likely to be GPTs than other technologies such as 3D printing, nanotechnology, and blockchain.[7] It is no surprise that economists deem AI to be “the most important general-purpose technology of our era.”[8]
Based on historical case studies of past industrial revolutions, my book Technology and the Rise of Great Powers argues that diffusion capacity in GPTs is central to competition among the great powers. Instead of comparing which rising power first introduces major innovations in new industries, the book demonstrates that states that attained technological leadership were more successful than their rivals in adopting and diffusing GPTs at scale.[9] In fact, during the second industrial revolution (1870–1914), the U.S. owed its economic rise not to dominance of the most important innovations in new sectors such as electricity and chemicals but rather to its effectiveness at adapting machine tools across almost all branches of industry. The diffusion of interchangeable parts manufacturing was a key factor in America’s ability to sustain economic growth at higher rates than either Britain or Germany.[10]
Notably, this diffusion- and GPT-centered lens resonates with Chinese leaders and thinkers. In July 2018, in an address to the BRICS summit on the historical stakes in the fourth industrial revolution, Chinese president Xi Jinping stated, “From the mechanization of the first industrial revolution in the 18th century, to the electrification of the second industrial revolution in the 19th century, to the informatization of the third industrial revolution in the 20th century, rounds of disruptive technological innovation have … fundamentally changed the development trajectory of human history.” Referencing recent breakthroughs in cutting-edge technologies like AI, Xi proclaimed, “Today, we are experiencing a larger and deeper round of technological revolution and industrial transformation.”[11] After this speech, Chinese analysts and scholars added their own interpretations of the relationship between technological revolutions and power shifts. One commentary, published on the website of Study Times, an authoritative publication of the Central Party School, makes the connection between new technologies, geopolitical connections, and widespread productivity growth: “Britain seized the opportunity of the first industrial revolution and established a world-leading productivity advantage. … After the second industrial revolution, the United States seized the dominance of advanced productivity from Britain.”[12]
China’s AI Implementation Gap
The above GPT diffusion framework helps ground debates about China’s AI progress. Since OpenAI released ChatGPT in November 2022, much of the analysis about U.S.-China technological competition has been preoccupied with which country can develop the next breakthrough in foundation models, with particular attention to China’s ability to produce its own version of ChatGPT.[13] In the first half of 2023, Chinese labs developed and released over 100 large language models (LLMs)—a surge that garnered the moniker the “Hundred Model War.”[14] In recent months, some of these Chinese labs have released AI models that register stronger performance on certain benchmarks than their strongest Western competitors.[15] These evaluation systems include those that test models on a wide range of scenarios and prompts (such as Hugging Face’s OpenCompass leaderboard tool) as well as those that rank models based on human preferences (such as the Chatbot Arena platform developed by UC Berkeley researchers).
Yet, even as China’s AI ecosystem has become more competitive in terms of pioneering novel advances (innovation capacity), it continues to encounter an implementation gap. According to a report by the Chinese technology media outlet Leiphone, there exists an “impassable chasm between technology and implementation in the large model era. …Large models are deemed a new generation of infrastructure, but on the industrial side, many companies do not really use large models.[16] Among the numerous adoption challenges cited in the report, some of the most noteworthy obstacles include: the poor performance of pre-trained large models for specific industrial applications; the high labor and usage costs associated with fine-tuning large models; and the inability to verify and compare the effects of all the different models against one another. These challenges reflect the protracted, gradual trajectory of GPT diffusion outlined in the previous section.
It should be noted that this AI implementation gap is not unique to China. According to an April 2024 survey conducted by Reuters Institute for the Study of Journalism across six countries (Argentina, Denmark, France, Japan, the UK, and the United States), intensive use of generative AI services is still very limited.[17] Even when it comes to the diffusion of ChatGPT—hailed as the fastest-growing app in history—only 1 percent of Japanese respondents use it on a daily basis and 7 percent of American respondents use it on a daily basis. Many of the respondents from these six countries reported that they have only used generative AI tools once or twice, thus underscoring the need to differentiate between casual experimentation and serious adoption.
Evidence from China’s diffusion of information and communications technologies (ICTs) provides further evidence that China’s diffusion capacity may trail its innovation capacity. While China has been successful at large-scale deployment in a few key domains—consumer-facing applications like mobile payments and high-speed rail—these achievements do not characterize the overall trend in ICTs. Chinese businesses have been slow to embrace digitization, as measured by adoption rates of digital factories, industrial robots, smart sensors, and key industrial software.[18] According to a 2020 report on the state of digital intelligentization (i.e., the use of data-based predictions to improve businesses processes, such as personalized recommendations, generative designs, and optimization of supply-chain management), less than one-tenth of China’s small- and medium-sized enterprises have initiated such projects.[19]
These findings are confirmed by influential indexes that assess national scientific and technological capabilities. The International Telecommunication Union’s ICT Development Index provides a composite measure of the level of access to and use of ICTs in countries around the world. On this metric, China ranks 83rd in the world, trailing the U.S. by 67 places.[20] China also significantly trails the U.S. in an influential index for the adoption of cloud computing, which is essential to implementing AI applications. In 2018, U.S. firms averaged a cloud adoption rate of over 85 percent, more than double the comparable rate of Chinese firms.[21]
Since cloud-based services will serve as important diffusion channels for AI, it is worthwhile to further investigate roadblocks to a more widespread adoption of cloud computing across the Chinese economy. In 2013, only 5 percent of Chinese small and medium-sized enterprises used cloud hosting services.[22] While the Chinese government has supported cloud computing development through infrastructure spending, subsidies, and other financial incentives, its cyber-control policies have impeded diffusion. Data localization measures and internet censorship protocols have made it difficult for small businesses to access not only services supplied by foreign cloud computing platforms (e.g., Dropbox cloud storage) but also objects hosted on servers outside China. Also, due to concerns about data security, large state-owned enterprises have been hesitant to outsource data center services to the cloud.[23]
China treads a similar tightrope between information control and unencumbered diffusion, especially in managing the output of generative AI models. Strict censorship rules on sensitive political content are a poor match for LLMs that follow a user’s prompts to generate text on any subject. For instance, China’s draft regulations on generative AI, issued in April 2023, require that LLM developers verify the veracity of the training data that underlie these models. This poses a serious impediment to the capacity of Chinese AI labs to release their services out to the general public, since these LLMs are often trained on massive collections of data scraped from the internet.[24] Although the Chinese government has relaxed some of these provisions, its overall approach to controlling AI may hinder the spread of generative AI technology throughout the entire economy.
Another key element in China’s AI implementation gap is a talent bottleneck. Despite graduating more students in STEM fields, China’s talent pool of AI practitioners is less than half the size of the U.S. pool.[25] According to a report by China’s Ministry of Human Resources and Social Security, China currently faces a shortage of 5 million AI talents.[26] To be sure, China has made important investments in enhancing AI education. In 2018, the Chinese Ministry of Education approved the creation of an AI major, which was quickly adopted by universities throughout the country.[27] However, as pointed out by a white paper co-authored by Baidu and Zhejiang University’s Institute of China’s Science, Technology, and Education Policy, many Chinese universities lack the computing and data resources required to give students practical experience in AI development.[28]
When it comes to disseminating AI advances across the entire economy, robust linkages between academic and industry settings are especially crucial. The U.S. has built a strong connective tissue in this respect. Based on data for the years 2015 to 2019, the U.S. was the world leader in its number of academic-corporate hybrid AI publications—publications co-authored by at least one researcher from industry and one researcher from academia. This more than doubles the number of Chinese hybrid AI publications.[29] Indeed, China’s official state news agency has highlighted the lack of technical exchanges between universities and industry as one of five key weaknesses in China’s AI talent ecosystem.[30] In sum, when it comes to its AI diffusion capacity, China faces a sizable implementation gap.
Conclusion: AI Diffusion and China’s Productivity Slowdown
The significance of China’s AI diffusion capacity extends beyond great power competition over emerging technologies. Put simply, a general-purpose technology like AI presents an opportunity for China to boost productivity, which is the key variable that will empower long-term economic growth. Maintaining high growth rates also undergirds the Chinese Communist Party’s “performance legitimacy,” a crucial source of its political authority.[31]
Notably, China has struggled to sustain productivity growth in recent decades. Held back by inefficient infrastructure expenditures, China’s aggregate total factor productivity growth declined from 2.8 percent in the decade prior to the global financial crisis to 0.7 percent in the decade after the crisis (2009–2018).[32] If calculated using alternative estimates from The Conference Board Total Economy Database, China’s total factor productivity actually declined from 2010 to 2017, averaging a negative growth of 0.5 percent over this period.[33] It should be noted that this period’s slowdown in productivity also applies to the advanced and emerging economies other than China. Economists attribute some of this decline to the waning effects of the benefits of previous general-purpose technologies, including the computer and the internet.
Thus, effective diffusion of AI—possibly this era’s GPT— will be central to China’s prospects of maintaining high rates of productivity growth. China’s leaders worry about getting stuck in the “middle-income trap,” a condition whereby an economy is unable to advance to high-income status after it exhausts export-driven, low-cost manufacturing advantages.[34] Given the diminishing effects of other growth drivers, including urbanization and demographic trends, improvements in technology and total factor productivity will become more important determinants of future growth.[35] For example, recent speeches by Chinese Communist Party leaders and government ministry reports emphasize that the wide-ranging diffusion of digital technologies like AI are central to the development of “new quality productive forces”—a buzzword first coined by President Xi Jinping in September 2023.[36]
In sum, it is all too easy to be seduced by the latest breakthroughs in fast-moving technological fields like AI. When it comes to assessing whether China can overtake the U.S. as an AI superpower, the humble undertaking of diffusion must take center stage. Technological leadership in the fourth industrial revolution will rest on which country will be able to overcome the AI implementation gap.
About the Contributor
Jeffrey Ding is Assistant Professor of Political Science at George Washington University. His book Technology and the Rise of Great Powers, published by Princeton University Press, investigates how past technological revolutions influenced the rise and fall of great powers. Ding’s research has been published in European Journal of International Relations, Foreign Affairs, International Studies Quarterly, Review of International Political Economy, and Security Studies. He received his PhD in 2021 from the University of Oxford, where he studied as a Rhodes Scholar, and he earned his B.A. in 2016 from the University of Iowa.
Notes
[1] Graham Webster, Rogier Creemers, Paul Triolo, and Elsa Kania, “China’s Plan to ‘Lead’ in AI: Purpose, Prospects, and Problems,” New America, August 1, 2017. http://newamerica.org/cybersecurity-initiative/blog/chinas-plan-lead-ai-purpose-prospects-and-problems/.
[2] Daniel Araya, “Who Will Lead in the Age of Artificial Intelligence?” Forbes, January 1, 2019. https://www.forbes.com/sites/danielaraya/2019/01/01/who-will-lead-in-the-age-of-artificial-intelligence/.
[3] Graham Allison, and Eric Schmidt, “Is China Beating the US to AI Supremacy?” Belfer Center for Science and International Affairs, Harvard Kennedy School, August 2020. https://www.belfercenter.org/publication/china-beating-us-ai-supremacy.
[4] National Security Commission on Artificial Intelligence, “The Final Report,” NSCAI, March 2021. https://www.nscai.gov / 2021-final-report/.
[5] Carl Benedikt Frey, and Michael Osborne, “China Won’t Win the Race for AI Dominance,” October 7, 2020. https://www.foreignaffairs.com/articles/united-states/2020-06-19/china-wont-win-race-ai-dominance.
[6] The above discussion mainly draws from Bresnahan and Trajtenberg (1995) and Lipsey, Carlaw, and Bekar (2005).
[7] Avi Goldfarb, Bledi Taska, and Florenta Teodoridis, “Could Machine Learning Be a General Purpose Technology? A Comparison of Emerging Technologies Using Data from Online Job Postings,” Social Science Research Network, May 8, 2021. https://doi.org/10.2139/ssrn.3468822.
[8] Erik Brynjolfsson, and Andrew McAfee, “The Business of Artificial Intelligence,” Harvard Business Review, July 18, 2017. https://hbr.org/2017/07/the-business-of-artificial-intelligence.
[9] Jeffrey Ding, Technology and the Rise of Great Powers: How Diffusion Shapes Economic Competition (Princeton, NJ: Princeton University Press, 2024).
[10] Jeffrey Ding, "The Rise and Fall of Technological Leadership: General-purpose Technology Diffusion and Economic Power Transitions," International Studies Quarterly 68, no. 2 (2024): sqae013. https://doi.org/10.1093/isq/sqae013.
[11] Rosh Doshi, The Long Game: China’s Grand Strategy to Displace American Order, Bridging the Gap (NY: Oxford University Press, 2021).
[12] Ibid.
[13]Helen Toner, Jenny Xiao, and Jeffrey Ding, "The Illusion of China’s AI Prowess," Foreign Affairs, June 2, 2023. https://www.foreignaffairs.com/china/illusion-chinas-ai-prowess-regulation-helen-toner.
[14] https://chinamediaproject.org/the_ccp_dictionary/hundred-model-war/.
[15] Richard Fletcher, and Rasmus Kleis Nielsen, “What Does the Public in Six Countries Think of Generative AI in News?” Reuters Institute for the Study of Journalism, 2024. https://doi.org/10.60625/RISJ-4ZB8-CG87.
[16] Jeffrey Ding, “ChinAI #236: The LLM Implementation Gap,” September 11, 2023. https://chinai.substack.com/p/chinai-236-the-llm-implementation.
[17] https://reutersinstitute.politics.ox.ac.uk/what-does-public-six-countries-think-generative-ai-news.
[18] Alibaba Research Institute, “From Connected to Empowered: Smart+ Assisting the High-Quality Development of China’s Economy” [从连接到赋能：‘智能+’助力中国经济高质量发展], March 11, 2019; Synced [机器之心]. “Market Research Report on Supply and Demand for Digital Intelligentization Solutions for China’s Small and Medium Enterprises” [中国中小企业数智化解决方案供应市场研究报告2020], October 2020; Techxcope [战略前沿技术], “Innovation Is More than Invention: Detailed Explanation of the German Industry-University-Research Systems Big Four” [创新不止于发明：德国产学研体系四大金刚详解], November 18, 2020.
[19] Synced, Market Research Report on Supply and Demand for Digital Intelligentization Solutions for China’s Small and Medium Enterprises 2020” [中国中小企业数智化解决方案供应市场研究报告2020].
[20] International Telecommunications Union, “Measuring the Information Society Report 2017,” 2017. https://www.itu.int/en/ITU-D/Statistics/Pages/publications/mis2017.aspx.
[21] J. Wang, and X. Chen, “Information Security: A Comparative Study on Cloud Security in China and the United States, What Gaps Exist in Chinese Cloud Security? [信息安全：中美云安全产业对比研究, 国内云安全公司空间几何?], DongXing Securities (blog), June 29, 2020. http://www.xcf.cn/article/2a6265f3bc2f11eabf3cd4c9efcfdeca.html.
[22] Nir Kshetri, "Institutional and Economic Factors Affecting the Development of the Chinese Cloud Computing Industry and Market," Telecommunications Policy 40, no. 2–3 (2016): 116–129.
[23] Ibid.
[24] Toner, Xiao, and Ding, "The Illusion of China’s AI Prowess."
[25] Jeffrey Ding, "Deciphering China’s AI Dream," Future of Humanity Institute Technical Report, 2018.
[26] “China Artificial Intelligence Talent Training Report,” CSET Translation, August 25, 2022, https://cset.georgetown.edu/publication/china-artificial-intelligence-talent-training-report/.
[27] Jeffrey Ding, "China’s Current Capabilities, Policies, and Industrial Ecosystem in AI," testimony before the US-China Economic and Security Review Commission Hearing on Technology, Trade, and Military-Civil Fusion: China’s Pursuit of Artificial Intelligence, New Materials, and New Energy (2019).
[28] “China Artificial Intelligence Talent Training Report.”
[29]Daniel Zhang, et al., “The AI Index 2021 Annual Report, ” Stanford Human-Centered Artificial Intelligence Institute, 2021.
[30] Xinhua, “News Analysis: Examining the Five Shortcomings of China’s AI Talent System” [新闻分析：透视中国人工智能人才体系五大短板], August 28, 2019. http://www.gov.cn/xinwen/2019-08/28/content_5425310.htm.
[31] Minxin Pei, “China: Totalitarianism's Long Shadow,” Journal of Democracy 32, no. 2 (2021): 5–21.
[32]Loren Brandt, et al., “China’s Productivity Slowdown and Future Growth Potential,” Policy Research Working Paper, World Bank Group, June 2020.
[33] The Conference Board, “Total Economy Database Summary Tables,” July 2020. https://www.conference-board.org/retrievefile.cfm?filename=TED_SummaryTables_Charts_july20201.pdf&type=subsite.
[34] Juzhong Zhuang, Paul Vandenberg, and Yiping Huang, “Growing Beyond the Low-Cost Advantage: How the People’s Republic of China Can Avoid the Middle-Income Trap,” Asian Development Bank, 2012.
[35] Xiaodong Zhu, “Understanding China’s Growth: Past, Present, and Future,” Journal of Economic Perspectives 26, no. 4 (2012): 103–24.
[36] https://datainnovation.org/2024/05/explaining-chinas-focus-on-new-quality-productive-forces/.
Photo credit: Alenoach, Public domain, via Wikimedia Commons
