Trump Replaces "Artificial Intelligence" with "Superintelligence," but the Three Paradoxes of AI Development Remain Hard to Conceal

[[**The Trump administration's move to replace "artificial intelligence" with "superintelligence" reflects three paradoxes currently facing AI development: the productivity paradox, the Polanyi paradox, and the innovation paradox. The automation orientation dominated by a handful of tech giants exacerbates inequality; the article argues for advancing inclusive AI through the "democratization of innovation," so that innovation resources are combined with the diverse needs of the economy and society.**

**Key Points**

**1. The three paradoxes reveal the non-inclusive predicament of AI**

The productivity paradox refers to the rapid expansion of the ICT industry failing to drive efficiency gains in other industries; the Polanyi paradox shows that AI currently only replaces codifiable routine tasks and cannot be integrated into key links requiring tacit knowledge; the innovation paradox points out that the automation orientation dominated by giants ignores the actual needs of developing countries and small and medium-sized enterprises.

**2. The automation technology path is locked in by a few giants**

Historically there have been two automation paths: workers supervising machines and machines replacing workers. But with the combined forces of the military, management, and the tech community, the "programming-control" path of replacing labor became mainstream, sidelining frontline workers from participating in innovation. Today's tech giants further entrench this direction through their advantages in computing power, data, and talent.

**3. "Democratization of innovation" is the way to break the deadlock**

Through open-source models, shared science and innovation platforms, and public funding support, small and medium-sized enterprises, researchers, and frontline workers can obtain innovation resources. In practice, optimizing intelligent products by combining open-source models with industry experience (such as agricultural forecasting and mold production) has proven capable of breaking the giants' monopoly and exploring more inclusive technology paths.

**4. China's "AI+" initiative has demonstrative significance**

This initiative promotes deep integration of AI with six key fields, establishes pilot application bases, supports open-source communities, and advocates industry implementation oriented toward typical application scenarios. This path differs from the US model led by tech giants and is expected to inspire diverse actors to jointly participate in innovation, so that development outcomes benefit society more equitably. 2026-10-06 11:36 Trump replaces "artificial intelligence" with "superintelligence," failing to conceal the three most awkward truths of AI development Culture Vertical and Horizontal Quick View

This article is from the WeChat public account: Culture Vertical and Horizontal, author: Jiang Yuhao

[Introduction] On September 26, China and the US announced an eight-point consensus on outcomes, proposing to establish a China-US AI dialogue mechanism to exchange views on the risks and benefits of artificial intelligence, with the next dialogue scheduled for November this year. On September 29, US President Trump signed an executive order requiring the federal government to replace "artificial intelligence (AI)" with "superintelligence (SI)" in official documents and to propose a federal statutory definition of "superintelligence" within 60 days. Many analysts believe this move is aimed at advancing AI regulatory legislation at the federal level and further consolidating the uneven regulatory standards of various states into federal control. On the same day, Trump signed a "voluntary" AI safety agreement called the "Frontier Responsibility Joint Commitment" with companies including OpenAI, Anthropic, Meta, Google, and Nvidia, pledging to strengthen AI system safety management through independent assessments and other means.

But regarding AI, perhaps we need to consider more fundamental questions: In what direction will AI develop? When innovation resources are concentrated in a few tech giants and the technology path leans toward automation and labor replacement, can the productivity and benefits brought by AI reach broader industries, workers, and social groups? How can more people participate in shaping AI, and how can AI's development outcomes benefit society more broadly?

Jiang Yuhao's article starts from the "productivity paradox," the "Polanyi paradox," and the "innovation paradox," pointing out that current AI development is not merely a matter of technical capability: new technologies have not yet been widely translated into productivity gains across industries, applications remain limited to codifiable and repeatable tasks, and the innovation direction led by giants may not respond to the diverse actual needs of the economy and society. The problem lies not only in what AI can do, but also in who has the ability to define problems, allocate resources, and decide how technology is implemented.

Therefore, more enterprises, researchers, and frontline workers should be able to obtain innovation resources and participate in technical improvement and direction exploration in specific applications. Open source, shared platforms, public support, and worker participation are all possible paths to move AI from the innovation agenda of a few giants toward broader social needs.

This article was originally published in Culture Vertical and Horizontal, Issue 5, 2026, and represents only the author's views for readers' reference.

From giant dominance to the "democratization of innovation":

Artificial intelligence moving toward inclusiveness

In contrast to the feverish rapid iteration on the supply side, artificial intelligence (AI) has not yet produced ideal results on the demand side. More seriously, the current AI technology path, which leans toward labor-saving automation, is exacerbating the wealth gap between people with different skills and between developing and developed countries. In April 2025, the United Nations Conference on Trade and Development (UNCTAD) published a report advocating a policy program for "inclusive artificial intelligence." In recent years, China has implemented the "AI+" initiative to promote the widespread application of new technologies, and its governance philosophy and policy arrangements have attracted great attention at home and abroad, being regarded as an AI development path distinct from the US paradigm led by a few tech giants. Against this background, this article addresses the problems existing in the actual development of artificial intelligence and, based on its innovation characteristics and practical exploration, analyzes the possibility of developing it through widespread application into a technological tool conducive to inclusive growth.

▍The development problems of artificial intelligence: an analysis of the triple "paradoxes"

Since the rise of mechanization technology, "technological anxiety" triggered by technological progress has always permeated discussions of all kinds. The optimistic narrative of economic history points out that in the long run, even in economies with high levels of automation, there are still a large number of useful jobs that humans can perform.[1] But this assertion is insufficient to respond to the real problems in the midst of dramatic transformation. In the development of artificial intelligence technology, three "paradoxes" are notably present, posing enormous challenges to this.

First, the "Productivity Paradox": the non-inclusiveness of artificial intelligence.

An important indicator for measuring whether technological progress is beneficial to society is productivity. In 1987, Nobel laureate in economics Robert Solow proposed that "you can see the computer age everywhere but in the productivity statistics," sparking debate around information and communication technology (ICT). Entering the 21st century, ICT saw major development, but the growth rate of productivity in the United States fell into sustained stagnation, which led academia to focus on discussing the "productivity paradox" of ICT: does large-scale investment in the ICT industry help improve production efficiency? Robert Gordon pointed out that ICT investment from 1994 to 2004 briefly pushed up growth in manufacturing capacity, but afterward the economy's average annual total factor productivity once again fell into long-term stagnation. Gordon thus questioned the "revolutionary" nature of ICT: compared with the Second Industrial Revolution, which transformed every aspect of the economy and society, the scope of impact of emerging technological products such as personal computers and the Internet was relatively limited.[2]

In recent years, the development of a new generation of ICT, such as artificial intelligence, has had very different impacts across industries: a UNCTAD report points out that, whether in developed or developing countries, enterprises that use artificial intelligence to improve production efficiency are mainly concentrated in the ICT industry. Some research points out that in the first half of 2025, US GDP growth was mainly due to investment in data centers and information processing technology; if categories related to new technologies are excluded, the annualized GDP growth rate was only 0.1%.[3] In China, the expansion of the ICT industry has failed to drive efficiency improvements in other industries, and the service sector group that intensively uses ICT contributed -0.5% to total factor productivity.[4] In other words, China and the United States currently both have an obvious "productivity paradox": AI-related industries have developed rapidly, but the supply of new technologies has failed to bring substantial improvements in production efficiency to many industries.

The "productivity paradox" raises a question: does the non-inclusiveness of new technologies stem from the fact that most enterprises do not know how to use them, or from the fact that artificial intelligence is not "revolutionary" enough? The second paradox discussed in this article shows that existing advanced artificial intelligence technologies are still difficult to deeply integrate into enterprises' key business activities, and this is not even a problem that can be solved by waiting for technology to adapt or by the rapid development of technology.

Second, "Polanyi's Paradox": the actual utility of artificial intelligence.

Economist David Autor once drew on Michael Polanyi's distinction between explicit knowledge and tacit knowledge to characterize the actual utility of ICT: computers can more easily replace humans in completing routine tasks that follow explicit, codable processes, but have greater difficulty replacing humans in completing non-routine tasks that require flexibility, judgment, and common sense, the latter involving knowledge that can only be understood tacitly. Changes in jobs and corresponding wages in the United States from 1979 to 2012 confirmed the existence of Polanyi's paradox: after the development of ICT, the trend of middle-skill jobs belonging to routine tasks being replaced was relatively obvious, while both high-skill and low-skill jobs that require more tacit knowledge that cannot be put into words increased.[5]

In recent years, generative artificial intelligence technology has developed rapidly, and the impact of new technologies on human work has been strengthening. A data analysis of 5,172 customer service personnel pointed out that generative artificial intelligence improved employee work efficiency, and new employees with insufficient experience and lower skill levels also improved their work capabilities through AI assistance.[6] In addition, an annual survey report released by the Wharton School and GBK Collective in October 2025 pointed out that large enterprises are rapidly narrowing the application gap in generative artificial intelligence, with repetitive tasks becoming the main application scenario, and "AI replacing people" concentrated in positions such as data analysis, document/meeting minutes generation, document editing/presentation creation, and brainstorming.

However, the above two studies also show that artificial intelligence technology has not broken through Polanyi's paradox: when enterprises apply generative artificial intelligence, the affected work mostly belongs to obviously repetitive routine tasks, while jobs that require more tacit knowledge have not been profoundly affected. This determines the application limits of frontier technology. In fact, tacit knowledge constitutes the underlying logic of the production environment, and rashly breaking this ecosystem will cause unpredictable losses. For example, in the field of industrial manufacturing, traditional OT control systems, under long-term operation, can strictly design and provide expected stable behavior according to deterministic attributes. The environment thus created cannot be adapted to by agents that are still mainly probabilistic and unable to obtain complete structured data for training. Therefore, even with super-strong language processing and a certain degree of generalization capability, generative artificial intelligence is still difficult to integrate into the key core links of the production field.

However, growing technological anxiety is fueling radical artificial intelligence strategies. Large enterprises frequently report significant layoffs for the deployment of AI strategies; Lenovo Group's "Global CIO Report: Entering the New AI Economy Era" notes that among enterprises surveyed in 2025, AI spending grew twofold year-on-year, with data science/business intelligence becoming the largest investment area, and more than 40% of funds flowing to generative artificial intelligence; yet 37% of respondents expressed doubts about whether AI can deliver value. Under intense technological anxiety, financial, policy, and marketing bubbles surrounding frontier technologies keep emerging, while more important questions—such as what the actual utility of the technology is, in which fields it can raise production efficiency, and how technological progress can help economic and social development—are seldom discussed.

Third, the "Innovation Paradox": the direction of artificial intelligence innovation.

In 2017, the World Bank published a research report titled "The Innovation Paradox," analyzing why developing countries participating in global value chains are unable to apply the various new knowledge, new technologies, and new organizational models spillover from developed countries. Rodrik further pointed out that new technologies created by developed countries are mostly automation- and skill-biased, and do not fit the labor-intensive comparative advantage of developing countries.[7] This focus reveals a deeper question behind the "innovation paradox": what innovation direction can actually promote economic and social development?

At present, the mainstream direction of artificial intelligence innovation is not conducive to the overall development of the economy and society. In terms of function, current artificial intelligence is oriented toward automation and surveillance; this technological route, driven rapidly forward by a handful of tech giants, has made "replacing human work" the mainstream imagination of the prospects of new technology development.[8] However, this automation-oriented technological route ignores the diverse needs of economic and social development. For example, for quite a few enterprises, saving labor costs may not be an urgent priority; as global value chains come under shock, new technological needs are emerging: current communication technologies are mainly used to coordinate the cross-border outsourcing costs of production tasks within products, but the globalization model based on the division of production tasks within products is now difficult to sustain, and enterprises increasingly need new information technologies that can assist workers in completing multiple production tasks.[9] Compared with the rapid development of automated intelligent technologies, such technologies are far too scarce.

In short, the analysis of the three paradoxes of artificial intelligence development both answers the question of why artificial intelligence remains non-inclusive today and inspires a new approach to coping with current technological anxiety: the rapidly expanding information and communication technology industry has not driven efficiency improvements in many other sectors, but this is not necessarily because artificial intelligence technology is insufficiently "revolutionary"; rather, its innovation direction is "incorrect." The path to alleviating the increasingly severe transformation crisis lies not only in redistribution and improving social security systems, but perhaps more importantly in exploring more inclusive technological routes and promoting diversified development.

▍Which kind of automation? How does it become a "problem"?

Before discussing how to explore new technological routes, it is necessary to answer a question: since the 17th century, the standardization and automation of production processes have significantly raised productivity and created astonishing wealth for human society. Does opposing the automation orientation of artificial intelligence amount to negating this technological pursuit? This depends on how "automation" is understood.

The term "automation" clearly entered the industrial field in the 1940s–1950s, when Ford Motor Company management used the concept to describe the automatic entry and exit of raw materials and parts into and out of machine tools. But the pursuit of automation technology had already appeared when mechanization arose, and it advanced by leaps and bounds in the second half of the 20th century with the successive rapid development of electronic and information technologies.

There are roughly two kinds of automation: first, automation technology in which workers supervise and maintain machine operation and control the production process; second, automation technology that treats workers as auxiliaries to machine production, having human labor complete operations on the assembly line. In the former, workers watch over and manage machines, using their experience to push machines closer to specific production requirements; in the latter, workers serve merely as the eyes and hands assisting machine operation during the period before machines are fully developed, and are inevitably marginalized as technology further develops. At almost every historical stage of technological progress, both kinds of automation technology have existed simultaneously, and as long as they are used "according to local conditions," both can bring improvements in industrial productivity. In the 18th century, given its economic and social characteristics of labor shortage and relatively low energy extraction costs, Britain invented a series of labor-saving, low-cost energy-using technologies such as the spinning jenny, creating conditions for centralized, assembly-line production methods; while in the early 19th century, to adapt to the characteristics of its cotton textile industry being widely distributed in rural areas, France invented the Jacquard loom, which used punched paper tape programming to control product standards, facilitating farmers' independent decentralized work while also meeting the needs of standardized production, likewise bringing a major breakthrough in the cotton textile industry. Interestingly, with the development of technologies such as electricity and steam engines, the same kind of automation technology can both save labor and enhance human labor. Fei Xiaotong found in the 1940s that electric power and internal combustion engines could distribute urban-concentrated production lines to the countryside, while simultaneously raising the level of machine production in rural industry and the handicraft capabilities of farming households.[10]

However, the invention of technology and its application do not simply follow technical logic; rather, they are comprehensively shaped by multiple complex factors. Driven by management and users' strong demand to control uncertainty in the production process, automation technology increasingly moved toward a development path that emphasized saving labor and excluding frontline workers from participating in the technological innovation process. Research in the history of technology reveals that in the 1950s and 1960s in the United States, the field of industrial automation technology experienced a period in which two technological routes were explored in parallel: one was the "record—playback" method of recording the motions of skilled machinists and issuing instructions to machines based on the recordings; the other was the "programming—control" method in which engineers pre-wrote programs to control the production process. The former depended on skilled workers, while the latter better met the needs of management and the scientific and technological community to control technical deviations and manage production order. As the military's large-volume, standardized product orders became the largest market demand, a situation formed in which the military funded and determined the direction of technological development; the scientific and technological community merely provided scientific justification for military needs, while management promoted it to the enterprise production level. Under the combined effect of these three forces, the technological route that did not rely on workers' practical experience became the main direction of automation development.[11]

Subsequent changes in economic and social conditions accelerated the above-mentioned progress of automation. In 1952, information technology expert John Diebold redefined the term "automation," proposing that computer technology is "built on the mathematical formulas of the basic theory of communication and control, and can manufacture machines that can control themselves and program themselves." By simply switching between tape programs, a single machine could run multiple application programs, and the innovation of stored programs enabled each tape program to include a series of different instructions for the machine to execute. The ultimate result was that the number of personnel required to complete multiple tasks within an organization could be greatly reduced.[12] Entering the 1970s and 1980s, digital automation technology made further major breakthroughs, especially the development of microprocessors, which combined all the key components of a computer's central processing unit into a single, tiny, and inexpensive unit, thereby inventing inexpensive, fast, and reliable minicomputers and making it easier to automate clerical work and routine work in manufacturing.[13] IT experts and corporate information officers once became so-called "dictators" in the business world ("software defines everything"), and workers completely lost their place in the production process and innovation system. Changes at the policy level also intensified this trend: in the context of deindustrialization, the government stopped funding research that created new work tasks, and the labor market institutions that had previously helped create and sustain "good jobs" were simultaneously weakened. A small number of giants with automation as their business model dominated the economy, and government tax policy tilted even more toward capital and automation.[14]

In the past decade or so, a new generation of artificial intelligence has pushed automation to an entirely new stage. On the one hand, the deep learning model of neural networks has made major breakthroughs, embodying knowledge in the weight values of the connections among neurons. This knowledge production mechanism, which so far cannot be reliably explained, excludes the participation and diagnosis of individual experience in deep learning programs.[15] On the other hand, machine learning is applied to large datasets to complete prediction-based tasks at low cost and high efficiency, replacing the on-site differences among millions of workers performing specific jobs with technologies such as pattern recognition, object classification, and statistical prediction. Management thus obtained a "scientific" basis for making decisions without having to delve deeply into the production and operation process.[16]

It can be clearly stated that what needs to be opposed is the "non-inclusive" orientation of automation, which takes replacing human labor as its main function and excludes the participation of concrete production experience in the innovation process. However, if it can be combined with the diverse development needs of the economy and society, even current automated intelligent technology can likewise play a role in enhancing rather than replacing labor. The UNCTAD report records that Beijing Normal University, South China Agricultural University, and others used open-source datasets and deep neural network technology, combined with farmers' production experience, to develop intelligent products that improve the accuracy of crop yield predictions. On the enterprise application side, large language models currently still find it difficult to produce precise and specialized output. Some enterprises have explored an application method of "foundation model + retrieval-augmented generation technology (RAG)," which on the one hand breaks through the performance limitation that large models are difficult to deploy locally, and on the other hand places greater emphasis on the importance of structuring the experience of engineers and skilled workers. These cases use intelligent technology to assist the labor production process, enabling advanced artificial intelligence technology to serve the solution of concrete development problems.

However, a small number of technology giants dominate the frontier of artificial intelligence innovation, causing technical elites around the world to conduct exploration almost entirely around the established mainstream route and rarely to proactively integrate the diverse development needs of the economy and society. Some technology elites even claim that by combining learning models with neuro-symbolic models they can surpass various large language models and break through the "Polanyi paradox," thereby completely replacing human labor.[17]

The reason a small number of technology giants can dominate the direction of artificial intelligence innovation is related to the innovative characteristics of artificial intelligence. The boundary between basic research and applied research in artificial intelligence is very blurred. Compared with university scientific research and other laboratory research supported by the government, technology giants are better able, by virtue of massive computing power, ultra-large-scale data sets, and numerous technical experts, to carry out frontier exploration that is highly general-purpose, interdisciplinary, and more attractive to financial capital, thereby producing the dominant effect that "dominance over the application end also gives it the power to shape the direction of basic research."[18] Acemoglu points out that as public funding for scientific research has fallen sharply, technology giants set the research agenda in three important ways. First, technology giants directly influence the curriculum settings and artificial intelligence research fields of top universities through academic relationships and charitable institutions; second, technology giants have a great say in Washington and in research funding agencies, and the latter two determine the direction of academic research; third, students aspiring to fields related to artificial intelligence usually hope to be hired by technology giants, or to join startups that develop technologies that can be sold to technology giants, which makes it necessary for universities to respond to students' practical needs by cultivating connections with technology giants.[19]

What needs further consideration is how to challenge the development trend of automation technology in recent decades and the current situation in which a small number of technology giants occupy the dominant position in artificial intelligence innovation, and explore more inclusive artificial intelligence?

▍Advancing the exploration of inclusive artificial intelligence with the concept of "democratizing innovation"

The mainstream technology route of artificial intelligence dominated by a small number of technology giants does not pay attention to the diverse development needs of the economy and society. The rapid expansion of the information and communication technology industry, in which a small number of large enterprises enhance their own ability to process complex information through frontier technologies, is merely what Unger calls "pseudo-vanguardism":[20] the information and communication technology industry and large enterprises possess policy advantages, scale advantages, and technological advantages for coping with the high costs of artificial intelligence infrastructure, and these advantages are difficult to diffuse to other industries and enterprises. Breaking the monopoly of a small number of industries and enterprises over innovation resources and incentivizing diverse development actors to jointly carry out innovative research is the necessary path for exploring more inclusive new technology routes.

Existing research has systematically theorized the experience of the diffusion of innovation resources. Hippel analyzed the activities of product improvement and development carried out by users in many fields, pointing out that the capabilities and information required for innovation are in fact widely dispersed, and that the model of concentrating resources on a small number of predetermined innovators is inefficient. Only when the cost of design and prototyping falls substantially can innovation resources diffuse widely and reduce the difficulty of allocating innovation resources. Moreover, "the net result is and should be the democratization of opportunity."[21] What the diffusion of innovation resources expects is precisely to realize the "democratization of innovation": enabling diverse development actors to obtain resources and jointly explore more inclusive artificial intelligence technology routes.

As far as artificial intelligence is concerned, the "democratization of innovation" can closely connect innovation resources with various production systems and produce unexpected innovative effects. First, it promotes the implementation of relatively mature artificial intelligence technologies, as stated by Academician Li Guojie of the Chinese Academy of Engineering: "We must recognize the limitations of artificial intelligence and must not blindly pursue artificial intelligence that can solve all problems. The key is still to implement relatively general artificial intelligence technologies in various industries according to actual needs, so that artificial intelligence technologies within a certain scope can produce practical results."[22] Second, it stimulates various actors to innovate in application, and even to rethink the technology development route. Artificial intelligence is a collection of multidisciplinary knowledge and multiple technological inventions, and a product of the coordinated evolution of breakthroughs in basic principles and tackling of key technologies. Improvement in any link from scientific research to product application may create new possibilities for exploration.[23]

In this regard, there was once a highly instructive policy design: in October 2019, the U.S. National Science Foundation, together with multiple departments, launched the National Artificial Intelligence Research Institutes program, which not only emphasized that the establishment of each institute must undertake the task of "advancing basic research and use-inspired research in an integrated manner," but also deployed them in multiple regions with different development conditions, striving to make artificial intelligence research cover diverse development fields—including application directions in new information and communication technology industries such as big data, machine learning, and algorithm optimization, in which developed regions lead, as well as fields such as agriculture, education, and networks that have a broad application foundation in less developed regions.[24] However, after the Trump administration sharply cut federal research spending, attention to the program declined. China's "Artificial Intelligence+" action opinion emphasizes measures such as promoting the broad and deep integration of artificial intelligence with six key fields, planning and building national artificial intelligence application pilot bases, and supporting the construction of artificial intelligence open-source communities, as well as issuing plans to promote the deep integration of artificial intelligence with specific industries according to typical application scenarios, and so on. It likewise adheres to an exploration path for advanced artificial intelligence technologies oriented toward diverse development needs. To date, many places have built open-source community platforms, taking a solid step in promoting the popularization and application of artificial intelligence and diverse exploration.

In any case, the primary purpose of "innovation democratization" is to enable diverse development actors to access innovation resources more conveniently and to innovate in the process of application. The effects of the diffusion of innovation resources can be seen preliminarily from the following two cases. Several researchers in South China with backgrounds in pulp and paper, food, and other specialties founded a technology company, using open-source models to design and operate process large models for traditional industries, combining professional knowledge and "veteran workers' experience" to improve model performance. Measured data from their intelligent products show that the accuracy of recognizing industry terminology and metaphors reaches 95%, and the accuracy of answering professional questions reaches 90%. In another case, an enterprise that initially engaged in mold production has in recent years, with the help of the local government's vigorous introduction of scientific research resources, transformed into a technology enterprise. Using foundation models, it developed a set of "integrated data collection and inference solutions." Based on its familiarity with mold production processes and techniques and the large model's data-processing capabilities, it built a full-chain system from collection and training to inference and control, thereby to some extent overcoming the problem that non-standardized, discrete industries such as mold production find it difficult to establish high-quality, real-time, accurate, and complete data sets. The successful experience of these two cases can be summarized in two points: the open-source route makes it possible for many users to access various artificial intelligence models at low cost and conveniently; and in the application of advanced technologies, industry practical knowledge and needs are taken as the core to optimize intelligent models.

At the same time, "innovation democratization" also puts forward a series of reform requirements for the existing innovation system, to ensure that innovation resources serve diverse and specific economic and social development needs. First, break through the existing barriers between departments, build shared science and technology innovation platforms that support the development of various enterprises, and promote closer collaboration among researchers, entrepreneurs, science and technology financial institutions, and frontline workers, jointly carrying out innovation activities to solve specific development problems. Second, establish public funds to support research institutions and enterprises in forward-looking research and development of new technologies to respond to drastic changes in the development environment. New intelligent technologies such as 3D printing may, at a new technological level, advance production methods of "flexible specialization" and "teamwork," helping enterprises respond to the drastic restructuring of globalization models. For this to ultimately be implemented, frontline workers must participate deeply, and therefore it is worth continuously supporting with public resources. [25] Third, enhance the voice of skilled workers, engineers, and others in the innovation system, so that innovation needs and practices endogenous to the production system receive fuller attention. For example (for enterprises receiving fiscal subsidies and research awards and subsidies), mandatorily increase the obligation to explain layoff plans, establish technological innovation projects that require frontline worker participation, incorporate data rights into the worker rights protection system, and promote broad worker participation in workforce development and transition plans, forming a "worker-centered AI development approach": planning the path of artificial intelligence development around the improvement of workers' living standards and career growth.

In summary, "innovation democratization" is support for diverse actors to participate in the innovation process. Reconstructing the innovation system with this concept as the core can break through the obstacles constraining the diffusion of innovation resources, preferential policies, and technological capabilities. (Figure 2) Only by guiding the reconstruction of the innovation system with the concept of "innovation democratization" can the innovation agenda cover a broad production system, prompt abundant innovation resources to combine with the diverse development needs of the economy and society, and create conditions for realizing inclusive artificial intelligence that contributes to the development of diverse actors.

▍Conclusion

The breakthrough technological capabilities of artificial intelligence are currently manifested more in creating the wealth gap and technological anxiety. A small number of technology giants dominate the direction of innovation, making the non-inclusive orientation toward automation an unchallenged mainstream technological route. This mainstream route must be changed, and artificial intelligence must be urged to become a technological force conducive to inclusive economic and social growth. Taking the series of issues in technological development, institutions and mechanisms, and so on targeted by this line of thinking as a reference, observing China's artificial intelligence policy, one can appreciate that its significance lies not only in emphasizing the importance of "artificial intelligence industrial application," but also in its potential to unleash the enormous potential of "innovation democratization": combining innovation resources with the diverse development needs of the economy and society, and ensuring that diverse actors jointly explore new technological routes. Efforts implemented accordingly can ensure inclusiveness in the advanced technology field of artificial intelligence, so that "the fruits of modernization are shared more and more fairly among all the people."

Notes (swipe up to view)

[1] Joel Mokyr, Chris Vickers, and Nicolas L. Ziebarth, "The History of Technological Anxiety and the Future of Economic Growth: Is This Time Different?" Journal of Economic Perspectives, Vol. 29, No. 3, 2015, pp. 31–50.

[2] Robert Gordon, The Rise and Fall of American Growth, translated by Zhang Linshan, Liu Xianwei, Sun Fengyi, et al., CITIC Publishing Group, 2018 edition, pp. 554, 551, 516–518.

[3] Unusual Whales, "Without data centers, GDP growth was 0.1% in the first half of 2025," Unusual Whales, October 18, 2025.

[4] Wu Xiaoying, "AI: A New 'Solow Paradox'?" Luohan Academy Viewpoint WeChat public account, July 12, 2025.

[5] David H. Autor, "Polanyi's Paradox and the Shape of Employment Growth," NBER Working Paper No. 20485, September 2014.

[6] Erik Brynjolfsson, Danielle Li, and Lindsey Raymond, "Generative AI at Work," The Quarterly Journal of Economics, Vol. 140, No. 2, 2025, pp. 889–942.

[7] Dani Rodrik, "New Technologies, Global Value Chains, and Developing Economies," NBER Working Paper No. 25164, October 2018.

[8][14][16][19] Daron Acemoglu, Redesigning AI: Work, Democracy, and Justice in the Age of Automation, Boston Review Forum, 2021, pp. 20–21; p. 19; p. 20; p. 31.

[9][25] Cui Zhiyuan, "Dual Circulation, Global Value Chains, and Additive Manufacturing: Preliminary Theoretical Reflections on a New Development Strategy," Twenty-First Century, August 2021 issue.

[10] Fei Xiaotong, "Rural Reconstruction," in The Complete Works of Fei Xiaotong, Vol. 5, Inner Mongolia People's Publishing House, 2009 edition, pp. 85–92; Cui Zhiyuan, "A New Exploration of Chinese Rural Industry: Flexible Specialization and the 'Möbius Strip' Organization," in Cui Zhiyuan, The Second Emancipation of the Mind and Institutional Innovation, Oxford University Press, 1997 edition.

[11] David F. Noble, Forces of Production: A Social History of Industrial Automation, translated by Li Fenghua, China Renmin University Press, 2007 edition, pp. 100, 225–226.

[12] Alfred D. Chandler and James W. Cortada, eds., A Nation Transformed by Information: How Information Has Shaped the United States from Colonial Times to the Present, translated by Wan Yan and Qiu Yanjuan, Shanghai Far East Publishers, 2011 edition, p. 229.

[13][15] Michael Wooldridge, The Complete History of Artificial Intelligence, translated by Xu Shu, Zhejiang Science and Technology Publishing House, 2021 edition, p. 250.

[17] Aaron (Youshen) Lim, "Expanding AI Power and Value Through Neural Networks (Part 1): Navigate the Structure and Complexity of Neural Networks," Medium, October 26, 2025.

[18] Nur Ahmed, Muntasir Wahed, and Neil C. Thompson, "The Growing Influence of Industry in AI Research," Science, Vol. 379, No. 6635, 2023, p. 885.

[20] Roberto M. Unger, The Knowledge Economy, Verso, 2019, pp. 57–58.

[21] Eric von Hippel, Democratizing Innovation: How User Innovation Improves Firms' Innovation Efficiency, translated by Chen Jin and Zhu Zhaohui, Intellectual Property Publishing House, 2007 edition, pp. 123–124.

[22] Li Guojie, "Reflections on AI Development Paths Triggered by DeepSeek," Science & Technology Review, 2025, No. 3.

[23][24] Jiang Yuhao, Zhang Xinyi, and Dai Mingjie, "Open Governance and Innovation Direction Regulation: A Study of Institutional Mechanisms for Advancing General Artificial Intelligence Development," Bulletin of the Chinese Academy of Sciences, 2024, No. 4. AI Venture Capital Daily Channel: Frontier Technology. This content comes from the internet and the original link; the views represent only the author and do not represent the position of Huxiu. For copyright issues, please contact hezuo@huxiu.com, and we will verify and handle it promptly. Everyone who is changing the world and wants to change the world is on the Huxiu APP [ ]