Development of New Generation of Artificial Intelligence in China

Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=cjcc20
Journal of Contemporary China
ISSN: (Print) (Online) Journal homepage: www.tandfonline.com/journals/cjcc20
Development of New Generation of Artificial
Intelligence in China: When Beijing’s Global
Ambitions Meet Local Realities
Shaleen Khanal, Hongzhou Zhang & Araz Taeihagh
To cite this article: Shaleen Khanal, Hongzhou Zhang & Araz Taeihagh (09 Apr
2024): Development of New Generation of Artificial Intelligence in China: When
Beijing’s Global Ambitions Meet Local Realities, Journal of Contemporary China, DOI:
10.1080/10670564.2024.2333492
To link to this article: https://doi.org/10.1080/10670564.2024.2333492
© 2024 The Author(s). Published by Informa
UK Limited, trading as Taylor & Francis
Group.
Published online: 09 Apr 2024.
Submit your article to this journal
View related articles
View Crossmark data

Development of New Generation of Artificial Intelligence in
China: When Beijing’s Global Ambitions Meet Local Realities
Shaleen Khanal a
, Hongzhou Zhang b
and Araz Taeihagh a
a
National University of Singapore, Singapore; b
Nanyang Technological University, Singapore
ABSTRACT
How did China become one of the leaders in AI development, and will
China prevail in the ongoing AI race with the US? Existing studies
have focused on the Chinese central government’s role in promoting
AI. Notwithstanding the importance of the central government, a
significant portion of the responsibility for AI development falls on
local governments’ shoulders. Local governments have diverging
interests, capacities and, therefore, approaches to promoting AI. This
poses an important question: How do local governments respond to
the central government’s policies on emerging technologies, such as
AI? This article answers this question by examining the convergence
or divergence of central and local priorities related to AI development
by analysing the central and local AI policy documents and the
provincial variations by focusing on the diffusion of the New
Generation Artificial Intelligence Development Plan (NGAIDP) in
China. Using a unique dataset of China’s provincial AI-related policies
that cite the NGAIDP, the nature of diffusion of the NGAIDP is exam
ined by conducting content analysis and fuzzy-set Qualitative
Comparative Analysis (fsQCA). This study highlights the important
role of local governments in China’s AI development and emphasises
examining policy diffusion as a political process.
Introduction
Artificial intelligence (AI) is one of the most transformative technologies of today. While countries
worldwide strive to establish technological superiority in AI, China has emerged as one of the two
leaders in the global AI development.1
The country ranks first in terms of AI-related patents
generated, research articles published and research articles cited worldwide. It is also rapidly
developing an ecosystem that facilitates the application of AI systems in everyday life. Due to its
rapid progress amidst the rising geopolitical tensions between the US and China, AI has emerged as
the new frontier of the US and China rivalry.2
China’s remarkable progress could not have been possible without the government’s direct
involvement. Ever since the issue of the 2017 New Generation AI Development Plan (NGAIDP), the
Chinese central government has considered AI as a national strategic sector and targeted the
CONTACT Araz Taeihagh spparaz@nus.edu.sg Lee Kuan Yew School of Public Policy, National University of Singapore,
469B Bukit Timah Road, Li Ka Shing Building, Level 2, #02-10, Singapore 259771, Singapore
1
Hongzhou Zhang and Shaleen Khanal, ‘To Win the Great AI Race, China Turns to Southeast Asia’, Asia Policy (2024), p. 21.
2
Jing Cheng and Jinghan Zeng, ‘Shaping AI’s Future? China in Global AI Governance’, Journal of Contemporary China (2022), p. 1;
Nicholas Wright, ‘How Artificial Intelligence Will Reshape the Global Order’, Foreign Affairs 10, (2018).
JOURNAL OF CONTEMPORARY CHINA
https://doi.org/10.1080/10670564.2024.2333492
© 2024 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.
This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.
0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. The terms on which
this article has been published allow the posting of the Accepted Manuscript in a repository by the author(s) or with their consent.

development of AI as a national priority.3
Simultaneously, the Chinese government also launched the
National New Generation AI Innovation and Development Pilot Zone, integrated AI as a priority
sector in the fourteenth 5-year plan, and has been at the forefront of developing guidelines and
regulations for developing trustworthy AI.4
These efforts have been reflected in significant govern
ment investments in research and development and the commercialisation of AI within China.5
While the role of the central government in AI development in China cannot be overstated,6
existing studies have not adequately paid attention to the important role local governments have
played in this process. The dual system of responsibility characteristic of the Chinese political system
implies local governments play a crucial role in technology development.7
However, given the
unequal financial and technological capacities of various local governments, the inherent interests
of the local governments in promoting new and emerging technologies are likely to be very
different.8
Provinces with high economic and technological capacities are likely to be interested in
sustaining their competitive edge by investing in such technologies, while the priorities of less
economically developed economies are likely to be more divergent.9
Some governments are likely to
embrace new technologies as potential for lift-off, while others might opt for a safer choice of
promoting such technologies once they reach a certain degree of maturity and sophistication. Under
such circumstances, the interests, priorities and strategies of local governments and central govern
ment in AI development might be very different. This discussion becomes especially pertinent in the
case of AI because the current narrative on China and AI has focused on the strategic interests and
implications of the technology.10
This article addresses these gaps by comparing and contrasting central and local policy priorities
and examining the nature and form of diffusion of AI policy in China. Taking the 2017 NGAIDP as
a mandated central policy, it examines the responses of China’s provincial governments in imple
menting this policy. The findings show that local provinces allocate significant financial and policy
resources to advance AI development and, therefore, are important actors in the development of AI
within China. However, the findings of this article also show differences in the priorities of central and
local governments, where local governments prioritize local economies and do not share the
security-related ambitions that the central government attaches with AI. Finally, the findings also
shed light on the importance of variations in local governments’ capacities in their AI-related policy
design with politically and economically powerful provinces more prone to faster and localised
policies than others. Overall, the findings highlight the important role played by local governments
in AI development in China.
3
Fei Wu and others, ‘Towards a New Generation of Artificial Intelligence in China’, Nature Machine Intelligence 2, (2020), p. 312.
4
Huw Roberts and others, ‘The Chinese Approach to Artificial Intelligence: An Analysis of Policy, Ethics, and Regulation’, AI &
Society 36, (2021), p. 59.
5
Ashwin Acharya and Zachary Arnold, ‘Chinese Public AI R&D Spending: Provisional Findings’, (CSET 2019) CSET Issue Brief.
Accessed November 28, 2023. https://cset.georgetown.edu/publication/chinese-public-ai-rd-spending-provisional-findings/;
Ngor Luong, Zachary Arnold and Ben Murphy, ‘Understanding Chinese Government Guidance Funds’, (CSET 2021) CSET
Issue Brief. accessed November 28, 2023, https://cset.georgetown.edu/publication/understanding-chinese-government-
guidance-funds/.
6
Roberts and others (n 5); Wu and others (n 4).
7
Xufeng Zhu, ‘Mandate Versus Championship: Vertical Government Intervention and Diffusion of Innovation in Public Services in
Authoritarian China’, Public Management Review 16, (2014), p. 117.
8
JC Teets and W Hurst, ‘The Politics and Patterns of Policy Diffusion in China’, Local Governance Innovation in China (1st ed,
Routledge, 2004).
9
Cui Huang and others, ‘A Bibliometric Study of China’s Science and Technology Policies: 1949–2010’, Scientometrics 102, (2015),
p. 1521.
10
Gregory A Allen, ‘Understanding China’s AI Strategy: Clues to Chinese Strategic Thinking on Artificial Intelligence and National
Security’ (Center for a New American Security (CNAS) 2019); Jinghan Zeng, ‘Artificial Intelligence and China’s Authoritarian
Governance’, International Affairs 96, (2020), p. 1441.
2 S. KHANAL ET AL.

Review of Key Concepts and AI Development in China
Important Role of Local Governments in Chinese Politics
According to China’s constitution, the country follows a unitary system. Nevertheless, there are
significant discrepancies seen between the centre’s policy mandates and local realities.11
To explain
these discrepancies and local variations, scholars have utilised the concept of central-local relations.
Over the past decades, China study scholars have engaged in heated debates around how best to
theorise central-local relations. Broadly speaking, there are three main perspectives. The first per
spective is the localist school, which emphasizes the importance of local autonomy and argues that
after years of decentralisation China has become a de facto federalist country.12
The second
perspective is the centralist school, which stresses the importance of strong central control over
local governments.13
The third perspective is the tug-of-war school, which believes that local officials
continuously push the boundaries of what is considered permissible local actions.14
Specifically, the
fragmented authoritarian framework has become widely applied in studying China’s politics.15
According to the fragmented authoritarian framework, it would be necessary to include the
Chinese local governments to better understand the complicated nature of China’s AI development.
Meanwhile, over the years, the literature on central-local relations in China has focused on various
policy domains, such as fiscal reforms, fishery management, social welfare, economic development,
cultural policy and foreign policy.16
However, the issue of AI development has received scant
attention. This has left a serious void in the existing literature.
Policy Diffusion
Policy diffusion entails transfer of a government policy from one body politic to another, either in
terms of adoption, implementation or both.17
Existing studies on policy diffusion have been
dominated by US-based studies that have focused on the horizontal transmission of policies from
one local government to another in the forms of learning, competition, imitation and, to some
extent, coercion.18
In such studies, the mechanism and not the nature of diffusion is the focus of
attention. The consequence of ignoring the policy design during the diffusion process is associated
with the logical implication that diffusion leads to the harmonisation of policies across various
11
Hongzhou Zhang and Alfred M Wu, ‘Central—Local Relations in China: A Case Study of Heilongjiang’s GMO Ban’, The China
Quarterly (2023), p. 1.
12
Alfred M Wu, Lin Ye and Hui Li, ‘The Impact of Fiscal Decentralization on Urban Agglomeration: Evidence from China’, Journal of
Urban Affairs 41, (2019), p. 170; Yongnian Zheng, De Facto Federalism in China: Reforms and Dynamics of Central-Local Relations
(World Scientific, 2007).
13
Hongbin Cai and Daniel Treisman, ‘Did Government Decentralization Cause China’s Economic Miracle?’, World Politics 58, (2006),
p. 505; Meg E Rithmire, ‘China’s “New Regionalism”: Subnational Analysis in Chinese Political Economy’, World Politics 66, (2014),
p. 165.
14
Jae Ho Chung, ‘Changing Norms, Issue-Variance, and Unending Tugs of War’ in Assessing the balance of power in central-local
relations in China, ed. John Donaldson (Routledge, 2016); Yasheng Huang, Inflation and Investment Controls in China: The
Political Economy of Central-Local Relations during the Reform Era (3rd edition, Cambridge University Press, 1999).
15
Hongzhou Zhang, ‘The South China Sea Fishing Crisis: The Overlooked Role of Chinese Subnational Governments’, The Pacific
Review (2024), p. 1.
16
Ting Gong and Alfred M Wu, ‘Central Mandates in Flux: Local Noncompliance in China’, Publius: The Journal of Federalism 42,
(2012), p. 313; Yasheng Huang, ‘Central-Local Relations in China during the Reform Era: The Economic and Institutional
Dimensions’, World Development 24, (1996), p. 655; Huang (n 15); Genia Kostka and Jonas Nahm, ‘Central—Local Relations:
Recentralization and Environmental Governance in China’, The China Quarterly 231, (2017), p. 567; Genia Kostka and Chunman
Zhang, ‘Tightening the Grip: Environmental Governance under Xi Jinping’, Environmental Politics 27, (2018), p. 769; Mingjiang Li,
‘Central—Local Interactions in Foreign Affairs’, in Assessing the balance of power in central-local relations in China, ed. John
Donaldson (Routledge, 2016); Christine PW Wong, ‘Central—Local Relations in an Era of Fiscal Decline: The Paradox of Fiscal
Decentralization in Post-Mao China*’, The China Quarterly 128, (1991), p. 691.
17
N Goyal, A Taeihagh and M Howlett, ‘Whither Policy Innovation? Mapping Conceptual Engagement with Public Policy in Energy
Transitions Research’, Energy Research & Social Science 89, (2022), p. 102, 632.
18
Charles R Shipan and Craig Volden, ‘The Mechanisms of Policy Diffusion’, American Journal of Political Science 52, (2008), p. 840.
JOURNAL OF CONTEMPORARY CHINA 3

political domains. Indeed, studies on horizontal diffusion focus on understanding how policy
innovations arise and spread across various political bodies.19
However, policy diffusion can take various forms.20
For instance, diffusion can vary in terms of
diffusion of goals, intensity (frequency of transmission from source to destination), speed (the time
taken for transmission), breadth (the coverage of the policy being diffused) and direction (horizontal
vs. vertical).21
The dimension of coverage plays a vital role when discussing vertical diffusion (from
higher levels of government to lower), since diffusion under such conditions is normally mandatory
(or coerced). Therefore, drivers and diffusion mechanisms that apply to horizontal diffusion do not
necessarily apply to vertical diffusion since the topic of interest is not if diffusion takes place but how.
The question of how becomes especially more pertinent in authoritarian countries such as China,
where local governments and leaders are under severe pressure to comply with central govern
ments’ policy requirements but also have some level of discretion to redesign these policies as per
their own social and economic requirements.22
Many scholars have attributed this phenomenon to
‘political centralisation and economic decentralisation’.23
As such, local governments are, to some
extent, autonomous bodies responsible for creating economic and fiscal environments for inducing
growth that encourages experimentation and innovation within such local bodies.24
At the same
time, the extent of such experimentation is governed by the political and policy mandates of the
central government, where performances of the local leaders and local governments become
functions of such mandates and are closely monitored by the central leadership.25
This functional
requirement of following central-level policies but also having the flexibility to adapt them as per the
local needs provides fertile opportunities for local leaders to redesign diffused policies. As such,
policy innovations have been central features of Chinese local governance for many years.26
AI Policy in China
Although the earliest policies on AI in China can be traced back as early as 1956, the Chinese central
government only started paying serious attention to the technology since its 13th 5-year Plan.27
Earlier policies around 2015–2016, including the Made in China 2015 document, focused more on
intelligent and autonomous systems. The Internet Plus Action Plan 2015 was the first central-level
long-term plan that explicitly recognised AI and AI industries as one of the 11 emerging growth
sectors. While the government issued an AI Three-Year Activities and Implementation Program in
2016, it was only when the NGAIDP was implemented in 2017 that the country had its first AI-specific
long-term plan.28
The Plan identifies AI as a sector of strategic importance and provides specific
targets to be achieved by 2020, 2025 and 2030. The Plan declares that China will be able to close the
19
Jing Chen and Cui Huang, ‘Policy Reinvention and Diffusion: Evidence from Chinese Provincial Governments’, Journal of Chinese
Political Science 26, (2021), p. 723.
20
Jian Zhang and others, ‘Measurement on the Policy Diffusion: Evidence from China’s Policy Documents’, Technology Analysis &
Strategic Management 34, (2022), p. 71; Lili Li, Araz Taeihagh and Si Ying Tan, ‘What Factors Drive Policy Transfer in Smart City
Development? Insights from a Delphi Study’, Sustainable Cities and Society 84, (2022), p. 104, 008.
21
Zhichao Ba and others, ‘Spatio-Temporal Dynamics and Determinants of New Energy Policy Diffusion in China: A Policy Citation
Approach’, Journal of Cleaner Production 376, (2022), p. 134, 270; Zhang and others (n 21).
22
Chen and Huang (n 20).
23
Canfei He, Yi Zhou and Zhiji Huang, ‘Fiscal Decentralization, Political Centralization, and Land Urbanization in China’, Urban
Geography 37, (2016), p. 436.
24
Hongbin Cai and Daniel Treisman, ‘Did Government Decentralization Cause China’s Economic Miracle?’, World Politics 58, (2006),
p. 505; John Knight and Li Shi, ‘Fiscal Decentralization: Incentives, Redistribution and Reform in China’, Oxford Development
Studies 27, (1999), p. 5.
25
Xiaobo Zhang, ‘Fiscal Decentralization and Political Centralization in China: Implications for Growth and Inequality’, Journal of
Comparative Economics 34, (2006), p. 713.
26
Chen and Huang (n 20); Yunxiang Zhang and Sichen Wang, ‘How Does Policy Innovation Diffuse among Chinese Local
Governments? A Quali. . .: Business Source’, Public Administration & Development 41, (2021), p. 34.
27
Chao Yang and Cui Huang, ‘Quantitative Mapping of the Evolution of AI Policy Distribution, Targets and Focuses over Three
Decades in China’, Technological Forecasting and Social Change 174, (2022), p. 121, 188.
28
Roberts and others (n 5).
4 S. KHANAL ET AL.

gap in AI technology with the leading countries around the world by 2020; by 2025, China will
achieve breakthroughs in some dimensions of basic AI research; and by 2030, China will be a leading
country in AI development and application in the world.29
As an ambitious strategy set by the central
government, NGAIDP was intended to incentivise local government as well as private sectors to
develop AI.30
To be sure, before the release of NGAIDP, local governments and had already launched various
sub-national AI-related policy and projects. Nevertheless, NGAIDP, as the starting point of China’s AI
planning at strategic and highest level, ‘formalizes and definitively signals a national-level focus on
AI’.31
After its launch, more subnational policies were introduced and AI development projects were
launched with massive investment in the sector (several provinces have committed over 100 billion
yuan (14.7 billion USD) investment in AI).32
In fact, following the launch of the NGAIDP, provincial
governments across the country started to launch their own policy packages for the development of
AI within their territories.33
The findings of this research show that between 2017 and May 2023,
more than 6000 local-level policies that mention AI within their texts and more than 500 local-level
policies that directly refer to AI in their titles were launched. While many of these policies have been
implemented at municipal levels, the provincial governments still form the largest sources of policies
on AI. Such local government policies can take a broad spectrum of policy instruments ranging from
long-term plans to financial incentives for R&D and application to the introduction of awards and
awareness programs to increase the popularity of the technology among users.
Method and Data
Method: Content Analysis and Qualitative Comparative Analysis
This study focused on four forms of diffusion at the local government level: diffusion of goals and
diffusion of instruments in terms of policy intensity, breadth and speed. Taking the NGAIDP 2017 as
the basis of central government policy, it traced all the subsequent local government policies that
cited the NGAIDP 2017 as the ‘diffused’ policies and examined the nature of diffusion of the central
government plan. First, the study conducted qualitative content analysis to examine the substantial
contents of the major policy documents. These contents included policy goals and mention of
specific important topics. Subsequently, the study conducted a fuzzy-set Qualitative Comparative
Analysis (fsQCA) to perform a comparative and comprehensive analysis of the diffusion of policy
instruments in China. In light of a limited number of cases, fsQCA enables identification of necessary
and sufficient conditions to find the causal pathways that have led to policy diffusion in China.
Content Analysis
The study used the content analysis method adopted to systematically compare and contrast
Chinese central and local governments’ policies on new-generation AI development to understand
their priorities and goals. The analysis focuses on the qualitative and quantitative evaluation of policy
documents, specifically emphasising the frequency and context of key terms related to foreign
policy, national strategy, national security and defence, as well as quantifiable policy goals related
to AI.
The analysis began with a careful selection of policy documents issued by central and provincial
governments. Central government documents included strategic plans and official documents from
pivotal government bodies shaping AI development. Similar documents from the provincial
29
State Council. New Generation AI Development Plan (NGADP), (2017).
30
ibid.
31
Jeffrey Ding, ‘Deciphering China’s AI Dream’, Future of Humanity Institute Technical Report (2018).
32
Gregory C Allen, ‘Understanding China’s AI Strategy: Clues to Chinese Strategic Thinking on Artificial Intelligence and National
Security’, (Center for a New American Security, 2019).
33
Jinghan Zeng, ‘China’s Artificial Intelligence Innovation: A Top‐Down National Command Approach?’, Global Policy 12, (2021),
p. 399.

governments were gathered. The authors developed a coding scheme that classified references into
predefined categories such as ‘Xi Jinping’, ‘defence’, ‘international competition/national competi
tiveness’, ‘national security’ and ‘civil-military’. Each category has clear operational definitions to
guide the coding process, ensuring that each mention or reference is accurately categorised. For
instance, any explicit mention of ‘Xi Jinping’, ‘President Xi’ or ‘General Secretary Xi’ was coded under
the ‘Xi Jinping’ category. Similarly, terms like ‘military/defence’ and ‘national security’ were coded
under the ‘defence’ and ‘national security’ categories, respectively.
Qualitative Comparative Analysis
Existing studies on policy diffusion in China have primarily focused on the political constraints of
policy diffusion, often centering on the political relationship between the local and central govern
ments, the mobility of policy entrepreneurs and competition between provinces.34
However, follow
ing the triple helix model of innovation, the authors argue that, in addition to the political capacity of
the government, the capacity of the private sector and academia are also important drivers of policy
diffusion.35
The authors conceptualised the outcome variables as follows. To measure the existence of
diffusion, the authors took the existence of provincial AI-specific long-term or medium-term plans
that directly cite the NGAIDP. Since the NGAIDP itself is a long-term plan, direct diffusion has to take
place in the form of a long-term plan as well. Therefore, the authors conceptualised a dependent
variable diffusion which takes the value 1 if the province introduced a long-term AI plan citing the
NGAIDP following its launch.
To measure the speed of diffusion, the authors calculated the difference in days between the
introduction of the provincial long-term plan and the launch date of the NGAIDP for each province.
For an example, since the NGAIDP was introduced on 8 July 2017, and Anhui’s long-term plan was
introduced on 28 July 2020, 623 days were required for the diffusion to happen. The authors ranked
the provinces as per their diffusion speed, with the earliest adopters ranking at the top and the latest
adopters ranking at the bottom.
The authors measured the intensity of diffusion in terms of the number of subsequent policies
introduced on AI following the introduction of the provincial plan. As such, the greater the number
of policies introduced, the greater the intensity of policy diffusion. Finally, the authors measured the
breadth of policy diffusion in terms of the breadth of the nature of instruments introduced by the
provincial government. The authors classified policy instruments introduced in each provincial
government policy into one/more than one of the five categories introduced by Schneider and
Ingram,36
authority, capacity, financial incentive, learning and symbolic. Authority-based instruments
consist of granting permissions, prohibiting or mandating certain actions. They are mostly used in
hierarchical systems by senior levels to guide lower-level behaviour. Incentive tools are used to
incentivise (negatively or positively) actions. Positive incentives like tax breaks and negative incen
tives like charges or fees are soft incentives, while sanctions and the use of force carry stronger
implications. Capacity-based tools assume that although targets of policy tools are naturally included
to carry out desired outcomes, they do not have the financial or knowledge capacity to do so. Tools
here aim to address these gaps by providing knowledge or financial resources to required target
groups. These include training, education-based tools, grants, loans and subsidies, amongst others.
Symbolic and hortatory tools aim to appeal to the cultural and social beliefs of the target population
to convince them to comply with desired policy outcomes. Tools like public pronouncements,
advertisement campaigns and sloganeering, or even prioritising certain actions provide symbolic
meanings to the desired actions. Finally, learning tools are based on information gathering, program
34
Chen and Huang (n 20); Zhang and Wang (n 27).
35
Aase Marthe Johansen Horrigmo, ‘Why Study the Spread of Culture-Led Development Strategies?’, Regional & Federal Studies
22, (2012), p. 553; Modris Ozoliņš and others, ‘Institutional Attention to European Policy Agendas: Exploring the Relevance of
Instrumental and Neo-Institutional Explanations’, Tertiary Education and Management 24, (2018), p. 338.
36
‘Behavioral Assumptions of Policy Tools’, The Journal of Politics 52, (1990), p. 510.
6 S. KHANAL ET AL.

evaluations or institutional dialogues between stakeholders and allow the government to learn the
situation on the ground. The authors assume that the greater the types of policy instruments the
provincial government introduces, the greater the breadth of policy diffusion.
The data on policies for this research came from various sources. The authors utilised the pkulaw
database of Peking University, which hosts policy documents introduced by various levels of the
Chinese government. The keyword ‘artificial intelligence’ was used as the basis of the search, and all
documents with the keyword in their title were selected. The authors also searched the CNKI policy
document database using the same strategy. Finally, provincial governments’ websites were also
individually scraped in cases where no relevant data was found in the two databases. The authors
had two criteria for the inclusion strategy: first, the document must have been issued by the
provincial government, and second, the policy document must be concerned about AI. The dataset
consists of 580 policy documents introduced between 2017 and 2022 at the provincial level. Beyond
policy documents, relevant academic articles, industry reports, news articles and books published in
both Chinese and English were collected through open search to aid data analysis.
Given the total number of cases, the authors were limited by the number of conditional variables
that could be used. In this study, the authors used four AI-specific conditional variables, capacity of
the private sector, capacity of the academic sector, economic size of the provinces and the political
authority of the provincial government. The authors measured the private sector capacity using two
indicators: the development of the private sector in the province and the size of the province’s
economy. The role of the private sector in the technology policy process has been extensively
documented. Studies have shown that the private sector, and especially companies involved in big
tech, is now directly involved in working with the government to design and shape governments’
technology solutions.37
The existence of a strong private sector can lead to a ‘soft capture’ of the
government where the private sector can influence the policy process of the regulators.38,39
The authors also included academic sector capacity as a critical component of policy diffusion.
Oftentimes, academics and the university can serve as part of the epistemic communities that can
influence policy design and diffusion process.40
The academic sector, through its knowledge net
works, can establish its authority over the knowledge domain and thereby help establish common
understandings amongst various policy actors, including policymakers.41
The provincial authority is also likely to play an important role in policy diffusion. A greater extent
of political authority can imply two important things: a) given their authority, leaders will have some
form of discretion on policies and experimentation they can conduct within their jurisdiction;42
or b)
given their close relationship with the Central Committee in Beijing, they are expected to fulfill the
obligations as expected by the Centre expediently. As such, a greater extent of authority is expected
to lead to faster diffusion and a greater extent of experimentation within the provinces.
To measure the capacity of the private sector and the capacity of the academic sector, the authors
used data from China’s New Generation AI Technology Industry Region Competitiveness Evaluation
Index (NGAII). The Chinese Institute of New Generation Artificial Intelligence Development Strategies
produces the annual index under the Chinese Academy of Engineering and Tianjin Municipal
37
Per J Agrell and Axel Gautier, ‘Rethinking Regulatory Capture’ in Recent Advances in the Analysis of Competition Policy and
Regulation, ed. Joseph E Harrington Jr. and Yannis Katsoulacos (1st edn, ElgarOnline, 2012); Robert G Hollands, ‘Critical
Interventions into the Corporate Smart City’, Cambridge Journal of Regions, Economy and Society 8, (2015), p. 61.
38
Agrell and Gautier (n 37).
39
S Khanal, H Zhang, and A Taeihagh, ‘Why and how is the power of Big Tech increasing in the policy process? The case of
generative AI’, Policy and Society, (2024), puae01. https://doi.org/10.1093/polsoc/puae012.
40
David J Galbreath and Joanne McEvoy, ‘How Epistemic Communities Drive International Regimes: The Case of Minority Rights in
Europe’, Journal of European Integration 35, (2013). p. 169; Alexandru Rusu and Olga Löblová, ‘Failure Is an Option: Epistemic
Communities and the Circulation of Health Technology Assessment’ in Public Policy Circulation, ed. Tom Baker and Christopher
Walker (Edward Elgar Publishing, 2019). accessed June 21, 2023, https://china.elgaronline.com/view/edcoll/9781788119146/
9781788119146.00016.xml.
41
Céline Mavrot and Fritz Sager, ‘Vertical Epistemic Communities in Multilevel Governance’, Policy & Politics 46, (2018), p. 391.
42
Edmund J Malesky, ‘Straight Ahead on Red: How Foreign Direct Investment Empowers Subnational Leaders’, The Journal of
Politics 70, (2008), p. 97.

Government. The index measures the competitiveness of Chinese provinces along five major
indicators: the capacity of the private sector, the capacity of universities, the capital intensiveness
for the AI sector, the linkages between public, private and higher education sectors and the open
ness of the provinces to attract human capital and technology. The detailed description of condi
tional variables is as follows:
Capacity of the Private Sector. The authors used the NGAII index to measure the capacity of the
private sector, focusing on two indicators: enterprise scale and enterprise innovation capacity. For
enterprise scale, the authors looked at the number of AI enterprises within each province and the
market value of those enterprises. To measure enterprise innovation capacity, the authors consid
ered three indicators: the average number of patents, the number of basic and technology tier
enterprises within provinces and the number of technology-enabling relationships.
Capacity of the Academic Sector. The capacity of the academic sector was measured using two
primary indicators: the innovation capacity of AI universities and the innovation capacity of non-
university research institutions. The innovation capacity of AI universities was measured using four
sub-indicators: the number of universities with AI programs, the average number of articles pub
lished in domestic journals, the average number of articles published in international journals and
the average number of patents produced. The same measures were used to evaluate the capacity of
non-university research institutions.
Economic Capacity. The authors measured the economic capacity of the provinces using the per
capita income (GDPPC) of the respective provinces. The authors used the Statistical Yearbook of
China 2019 to measure the GDPPC of respective provinces.
Political Authority of the Provincial Government. The authors used the Regional Political Power
Index (RPPI) to gauge the political powers of the provincial governments.43
The RPPI measures the
degree of political authority of provinces by creating an index of the weighted score of the number
of central committee full members born, studied or worked in particular provinces, which in turn
allows political leaders to develop networks, alliances and political capital within those provinces
providing those provinces with greater degree of power.
For the purpose of conducting fuzzy-set QCA (fsQCA), the authors ranked all the provinces based
on their performances in terms of each of the outcomes/conditions with 1 being the best performer
and 30 being the worst for the various indices. The outcomes and the conditions (factors) were then
calibrated to range between 0 and 1 with 3 and 27 acting as inclusion and exclusion thresholds and
12.5 being the crossover point.44
Since the outcome instrument breadth is an ordinal variable with
values from 1 to 5, it was coded accordingly (0 if one 1 category of instrument was implemented, 0.33
if 2 categories implemented, 0.67 if 3, 0.9 if 4 and 1 if 5).45
Figure 1 illustrates the ranking of the
provinces based on the specified factors.
43
Damien Ma and Ruihan Huang. ‘Powerful Provinces: How Regional Political and Economic Power Correlate in China’, MacroPolo
(2021). https://macropolo.org/analysis/powerful-provinces-regional-political-economic-china/
44
Adrian Dușa, QCA with R: A Comprehensive Resource (Springer International Publishing, 2019). Accessed February 1, 2024.
https://www.bookdown.org/dusadrian/QCAbook/QCAbook.pdf; Alrik Thiem, ‘Membership Function Sensitivity of Descriptive
Statistics in Fuzzy-Set Relations’, International Journal of Social Research Methodology 17, (2014), p. 625.
45
Appendix A1 contains the calibrated values of indicators described in Table 2. Also note that since diffusion days is applicable to
only 24 observations, the exclusion threshold was changed to 22.
8 S. KHANAL ET AL.

Findings
Divergences Between Central and Local Priorities in AI Development
The State Council introduced a comprehensive development plan entitled ‘New Generation Artificial
Intelligence Development Plan’ (NGAIDP) on 28 July 2017. The NGAIDP set the national AI agenda
and provided strategic direction for AI development across the country. Shortly after the introduc
tion of NGAIDP by the State Council, policy documents on the new generation of AI were introduced
by most Chinese provincial governments. The authors examined the nature and content of the plans
to understand the diffusion of policy goals. In the NGAIDP, while introducing the background and
reasons for its strategic plans, the Chinese central government has prioritised the geostrategic and
security importance of AI development. For instance, in the background section, the strategic
importance of AI was mentioned before the discussion of AI as the new engine for economic
development. It states:
Artificial intelligence has become the new focus of international competition. Artificial intelligence is thought to
be the strategic technology leading the future, the world’s major developed countries regard the development
of artificial intelligence as the major strategy to increase national competitiveness and enhance national
security . . . China’s national security and international competition situation are more complex, so [China]
must look at the world, lay out the artificial intelligence development on the national strategic level, grasp
firmly the strategic initiative of international competition during the new stage of artificial intelligence
Figure 1. Provincial rankings of the four factors used for fsQCA.

development, create new competitive advantage, open up new spaces of development, and effectively protect
national security.
In the NGAIDP, as Figure 2 shows, the term ‘Competition/Competitiveness (as in phrases such as
‘International Competition’ (guoji jingzheng), ‘National Competitiveness (guojia jingzhengli),
“Technological Competition” (keji jingzheng), “Technological Power” (keji qiangguo))’ appears 14
times, ‘National Security (guojia anquan)’ appears 8 times and ‘Defense (guofang)’ is mentioned 11
times. These findings suggest that AI development at the central level is driven by a combination of
geopolitical concerns, national security interests and considerations related to national defence. On
the other hand, the Chinese local governments have collectively only made 11 mentions of
‘International Competition/Competitiveness’, 4 mentions of ‘National Security’ and 14 mentions of
‘Defense’ in their new AI policy documents. In the AI policy documents issued by many provincial
governments, such as Beijing, Shanghai, Jiangsu, Gansu and Heilongjiang, these terms are not
mentioned even once. Even more interestingly, while the central government cares most about
international competition and the need to win the global economic and tech rivalry, the provincial
governments are concerned about AI as a new focus of regional competition within China. For
instance, in the background section of Anhui province’s new generation of AI development plan, it
cited regional AI competition and the fact that some provinces have already introduced AI devel
opment plans or action plans to achieve an advantage in regional competition as a main reason for
its own AI plan. Furthermore, the relatively high number of mentions of ‘Defense’ and ‘Civil-Military’
(junmin) is primarily related to the application of AI in defense-related sectors, especially by those
provinces such as Shaanxi, Jilin and Sichuan, which have large military-related industries, rather than
due to concerns about national defense.
The limited reference to the terms ‘National Security’ and ‘International Competition’ goes in stark
contrast with the ubiquitous emphasis by local governments on AI as the new driver for their local
Figure 2. Divergence of drivers and concerns towards the development of AI.
10 S. KHANAL ET AL.

economic development in their AI policy documents. As Zeng46
points out, because a booming AI
industry is considered a key driving force of regional economies, subnational actors, including
provincial governments have shown great enthusiasm for supporting AI development. This is
evidenced by Table 1, which represents the AI core and associated sectors’ output targets set by
various levels of the Chinese government for the years 2020 and 2025.
As Table 1 shows, for 2020, the central government had set an output target of RMB 150 billion for
the core AI industries and RMB 1000 billion for all associated AI sectors. Even more ambitious targets
for both the core AI sectors and associated sectors were set for 2025 and 2030, respectively. In
contrast, the local governments collectively set much higher targets. Take the targets for 2020, for
example. Compared with the central government’s goals, the sum of the targets set by the local
governments was approximately RMB 432 billion for core AI sectors and around RMB 1,958 billion for
associated AI sectors. Table 1 also provides the targets set by individual provinces. Notably, Beijing
has set the highest target for the core AI industry for 2020, aiming for RMB 112 billion, which on its
own is comparable to the national goal of RMB 150 billion. Shanghai and Sichuan followed, each
aiming for RMB 100 billion and RMB 50 billion, respectively. Looking at the 2025 targets for AI sectors,
the Centre has set a higher target of RMB 400 billion for the core AI sectors and RMB 5,000 billion for
associated AI sectors. The local governments, collectively, have also set a higher target of RMB
704 billion for the core AI sectors and 4670 billion RMB for associated AI sectors. However, it is
important to note that not all provinces have introduced quantifiable goals, and only half of the 31
provinces are included in Table 1. Therefore, the total targets set by provincial governments would
be even higher if the other provinces also set their own targets for AI sectors. Consequently, analysis
of provincial long-term plans’ political and economic goals show a clear divergence in political
ambitions and prioritisation towards economic goals compared to the Centre’s long-term priorities.
In addition, the central government clearly acknowledges the lack of basic theory and core
algorithm development as significant gaps in China’s AI competency compared to other developed
Table 1. AI industrial development targets set by the Chinese central and local governments.
Entity
2020 (RMB billions) 2025 (RMB billions)
Core AI industry Associated AI industries Core AI industries Associated AI industries
Central government’s targets 150 1000 400 5000
Sum of local governments’ targets 432 1958 704 4670
Beijing* 112 373 300 1000
Shanghai** 100
Guangdong 50 300 150 1800
Sichuan 50 300 100 500
Zhejiang* 34 260
Chongqing* 34 130
Anhui 15 100 50 450
Jiangsu 100
Fujian 100
Shaanxi* 55
Hubei* 10 60 60 600
Hunan 8 80
Liaoning 6 40 16 20
Jilin 5 40 20 200
Heilongjiang 5
Guangxi 3 20 8 100
*These provinces did not directly list their output targets for 2020 or 2025; rather, they listed their output target by 2022, 2023 or
2024. The 2020 or 2025 numbers were obtained by using the annual growth rate.
**Shanghai does not use the term core AI industry in this Plan. Instead, it uses ‘important AI industry’..
46
Zeng, J, ‘Artificial intelligence and China’s authoritarian governance’, International Affairs, 96(6), (2020), p. 1441–1459. https://
doi.org/10.1093/ia/iiaa172.

countries. Thus, one of the priorities listed in the NGAIDP is to achieve major breakthroughs in basic
theories for AI. Therefore, in terms of key priorities for the central government, the country aspires to
achieve important progress in new-generation AI theories and technologies by 2020, realize sig
nificant breakthroughs in the basic theories of AI by 2025, and, by 2030, reach the world’s leading
level in AI theories. However, for the Chinese provincial governments, the centrality of their AI
development has clearly been on the industrial application of AI, as terms such as ‘service’, ‘applica
tion’ and ‘enterprise’ appear far more frequently, and far fewer references are made to basic theory or
core algorithms in provincial policy documents.47
Even for provinces where basic AI research or basic
47
W Song and H Xia, ‘A Quantitative Study on the Text of Local Government Artificial Intelligence Industry Policy’, Science
Technology Management and Research 39, (2019), p. 192.
Figure 3. New generation of AI development plan timeline.
12 S. KHANAL ET AL.

theories in AI are mentioned, the focus has clearly been on either cutting-edge (qianyan) or applied
(yingyong) basic AI theories or research. For instance, Yunnan stated that it intends to accelerate
applied basic research and technological innovation and strengthen cutting-edge basic theory
research. Similarly, Shanghai intends to improve cutting-edge basic research in AI.
Variations of AI Policy Diffusion at the Provincial Level
Within Chinese provinces, there are also great variations in the speed, type and content of their AI
policies. Firstly, as Figure 3 shows, these documents vary greatly in terms of the date of issuance. The
first local new AI plan was issued by Jiangxi just one and a half months after the NGAIDP. Subsequently,
between 2017 and 2019, 17 provinces or municipalities introduced their own version of the new
generation of AI development plans. In contrast, some other provinces, including Chongqing, Tianjin
and Hubei, introduced their AI policy in 2020, with the latest being Beijing in May 2023. As of May 2023,
the rest of the provinces and regions, including Inner Mongolia, Shanxi, Hebei, Hainan, Tibet, Xinjiang
and Ningxia, had not issued policy documents related to the new generation of AI development.
Table 2 provides a descriptive overview of the variables used for the fsQCA analysis. The variable
enterprise capacity describes the ranking of the province in terms of enterprise capacity, and
university capacity is concerned with the ranking of the province in terms of university capacity.
GDPpc provides the level of income per capita of the province in 2018. The Power indicator ranks
provinces as per the regional power parity index. Four outcome variables consist of the existence of
diffusion (Diffusion), the speed of transmission (Diffusion days), the variety of instruments introduced
(instrument breadth) and the number of policies introduced (policy intensity).
Table 2. Summary table of the relevant indicators.
Provinces
Enterprise
capacity
University
capacity GDPpc Power Diffusion
Diffusion
Days
Instrument
breadth
Policy
intensity
Anhui 6 15 10943 10 1 623 5 15
Beijing 1 1 28294 1 1 2123 4 13
Chongqing 12 10 13479 11 1 206 5 24
Fujian 8 20 18856 9 1 219 4 22
Gansu 19 17 6686 20 1 373 1 1
Guangdong 2 13 15151 8 1 368 4 16
Guangxi 19 24 7755 22 1 251 5 20
Guizhou 14 24 7779 14 1 70 2 4
Hainan 19 24 9902 29 0 3 6
Hebei 19 24 8474 4 0 4 9
Heilongjiang 16 7 7717 13 1 248 1 2
Henan 19 19 9233 19 1 685 4 12
Hubei 11 4 13686 15 1 1114 5 19
Hunan 13 8 10942 26 1 575 2 9
Inner
Mongolia
19 24 14343 25 0 2 8
Jiangsu 7 2 21467 12 1 286 2 9
Jiangxi 19 18 10544 21 1 81 2 6
Jilin 19 14 8229 17 1 154 3 10
Liaoning 10 11 10225 7 1 158 1 4
Ningxia 21 26 10374 30 0 3 3
Qinghai 20 25 9028 16 1 149 1 3
Shaanxi 19 6 12319 3 1 765 4 6
Shandong 5 16 12786 6 1 158 4 16
Shanghai 4 3 26747 2 1 412 4 31
Shanxi 17 21 10954 27 0 1 5
Sichuan 9 5 10077 24 1 435 3 14
Tianjin 15 12 17727 18 1 962 5 27
Xinjiang 18 23 10191 23 0 0 1
Yunnan 19 22 9176 28 1 834 0 0
Zhejiang 3 9 17617 5 1 136 3 15

The next set of outputs shows the results of fsQCA, where the outcome is the actual transmission
of the policy in the form of a long-term provincial plan (see Table 3). Although the test results for
necessary conditions are not shown here (see Appendix B for results of truth tables and necessary
conditions of all the fsQCA conducted), findings show that no necessary and non-trivial conditions
exist for the diffusion of long-term plans. Meanwhile, seven provinces—Hainan, Hebei, Inner
Mongolia, Ningxia, Shanxi, Tibet and Xinjiang do not have a long-term AI-based plan. The results
from the solutions table also show the trivialness of the results of sufficiency tests. There are two
sufficient conditions for the diffusion of long-term plans, but the cases often overlap across the two
conditions. Overall, the results show two paths through which policy diffusion is likely to take place.
The general findings show the importance of private sector capacity, academic sector capacity and
political authority in determining the success of long-term policy transmission and also show the
non-requirement of income (development) of provinces. The authors also tested if there are certain
pathways that lead to non-transmission of long-term policy, but the results were not conclusive.
The next dimension of policy diffusion the authors tested was the speed of transmission. The
authors ranked the provinces according to the speed of the transmission of the long-term policies
and defined the outcome variable speed according to this rank (Table 4). The authors found one
pathway for faster diffusion. The pathway was characterised by higher levels of political authority
and private sector capacity of the provinces and lower levels of academic sector capacity. Anhui,
Fujian, Guangdong and Shandong were the four cases that represented this pathway that led to
faster policy diffusion.
The authors measured the intensity of policy diffusion using two different indicators: a) the
number of policies introduced between 2017 (after the introduction of the central AI plan) and
2022 and b) the growth of policies in AI between the given period annually. Results from the
fsQCA show two different pathways to a higher intensity of policy diffusion. The authors did not
find any necessary conditions for a greater extent of the intensity of policies introduced to
support AI. The findings showed that, unlike the case of the speed of diffusion, the existence of
strong political authority predicts success in only one of the pathways. Additionally, the authors
Table 3. Paths of long-term plan diffusion.
Contributing
factors Path 1 Path 2
Private sector
capacity
Yes
Academic sector
capacity
Yes
Income per capita
Political authority Yes
Inclusion score 0.979 0.983
Proportional
reduction in
inconsistency
0.979 0.983
Case provinces Jilin; Heilongjiang; Tianjin; Shaanxi; Hunan, Sichuan;
Liaoning; Hubei; Beijing, Chongqing, Guangdong,
Jiangsu, Shanghai, Zhejiang
Guizhou; Anhui, Fujian, Shandong; Liaoning;
Beijing, Chongqing, Guangdong, Jiangsu,
Shanghai, Zhejiang
Table 4. Speed of transmission.
Contributing factors Path 1
Private sector capacity Yes
Academic sector capacity No
Income per capita
Political authority Yes
Inclusion score 0.856
Proportional reduction in inconsistency 0.739
Case provinces Anhui, Shandong; Fujian, Guangdong
14 S. KHANAL ET AL.

also find that strong economic capacity was present in both pathways of sufficient condition
(Table 5). Of the two pathways, the first path consisted of having strong private sector capacity,
economic sector capacity and political authority and consisted of cases of Fujian, Anhui and
Shandong. Meanwhile, Tianjin and Hubei, which represented path 2, were characterised by
strong academic sector capacity and economic sector capacity but relatively weaker political
authority.
The authors measured the breadth of diffusion using the variety of policy instruments as per the
categorisation, where the authors divided all policy instruments into five categories and ranked the
provinces as per the number of different categories of instruments they had applied for the
governance of AI. The authors do not find any of the four conditions necessary for the increased
breadth of instruments. Furthermore, the results showed that two primary paths lead to policy
diffusion in the form of increased breadth (Table 6). The first, as with the case of intensity, involved
the three provinces, Fujian, Anhui and Shandong, with lower levels of political authority and high
levels of private sector capacity and high per capita income. The second path was highlighted by
Shaanxi’s case that also involved higher levels of political capacity, income per capita and academic
sector capacity but lower levels of private sector capacity.
Discussion
This article examined how central and local government priorities in AI development align (or
misalign) and the role of local economic and political factors in influencing AI-specific policy diffusion
in China. Taking the case of the NGAIDP, the authors looked at more than 500 AI-specific provincial
policies to understand the central and local priorities in AI development and investigate the nature
of policy diffusion within China’s provinces. A number of crucial pointers emerge from the above
analysis.
First, as China emerges as a global power, the study of central and local relations in China is
hugely important, not only to China observers but also to the global policy community. This is
particularly the case in relation to global AI development and governance. Academic conversation
on AI development has largely focused on the role of the central government’s motivations, its
strategic alignment and its policy approach to advancing AI development in China. This is under
standable against the recent recentralization push under Xi Jinping, which has greatly undermined
the autonomy of the local governments. Nevertheless, this research reveals provincial governments’
Table 5. Intensity of transmission.
Contributing factors Path 1 Path 2
Private sector capacity Yes
Academic sector capacity No Yes
Size of economy Yes Yes
Political authority Yes No
Proportional reduction in inconsistency 0.895 0.806
Case provinces Fujian, Anhui, Shandong Tianjin, Hubei
Table 6. Breadth of diffusion.
Contributing factors Path 1 Path 2
Private sector capacity Yes No
Academic sector capacity No Yes
Size of economy Yes Yes
Political authority Yes Yes
Proportional reduction in inconsistency 0.860 0.806
Case provinces Fujian, Anhui, Shandong Shaanxi

important role in advancing AI within China. Specifically, the findings of this research reveal some
interesting differences between how central and local governments in China perceive and discuss AI.
The Chinese central government seems to take a more holistic view, emphasising the role of AI in
international rivalry, national security, national defense and economic development. In fact, as
Zeng48
argues, State Council’s NGAIDP is a clear attempt to securitise AI against the backdrop of
growing US–China tech rivalry and the wider geopolitical competition. Since the release of NGAIDP,
there is no doubt that provincial governments ‘have got onboard with the central government’s AI
plan, and are enthusiastically supporting its AI campaign’.49
As shown in Figure 3, some of the
provincial AI policies were introduced immediately after NGAIDP launch. Nevertheless, it should be
noted that the provinces jumped onto the bandwagon with the primary objective of boosting their
local economy. The provincial governments have shown little interest in the aspects of international
rivalry, national security and defence; suggesting the broader geopolitical landscape and the
ongoing global AI race against the US seems to be irrelevant in the local contexts. This could be
due to the more localised focus of these governments, which often tend to prioritise more immedi
ate or tangible benefits of AI, such as its potential to drive economic growth or improve public
services. Indeed, as Table 1 shows, while both the central government and local governments in
China have set ambitious targets for the development of the AI industry, reflecting the importance
they attach to AI in their future economic growth plans, the local governments have set particularly
high targets, indicating strong local-level support for the AI industry. These findings highlight the
complex and multifaceted nature of AI policy in China, with different government levels having
different strategic focuses. In particular, as pointed out by Ding,50
provincial governments have been
pursuing their own economic interests to stake out their claims to China’s AI dream. The above
suggests that the Chinese approach towards AI development should not be seen as a coherent, well-
coordinated national-concerted endeavour, and that the central government’s capacity to drive the
country’s AI development need not be overstated.51
Second, one of the key objectives of the central government’s new AI plan is to develop the
country into a global AI powerhouse by 2030 and win the global AI race, particularly against
the backdrop of the intensifying rivalry between China and the US in AI leadership. This study
shows that this goal has not been commonly shared by the Chinese local governments. The
Chinese provincial governments can be categorised into five types based on the extent of their
global ambitions in the development of AI. As Figure 4 shows, Beijing, the capital and leading
AI hub in China, has a clear and explicit ambition to become a global leader in AI development.
This is evident by the title of its 2023 AI policy document- Implementation Plan to Accelerate the
Construction of an Artificial Intelligence Innovation Hub with Global Influence (2023–2025). Its
detailed plans and specific goals reflect a strategic vision that aligns with the central govern
ment’s objective to become a world leader in AI. Beijing’s predominant role in China’s AI
landscape also places it in a strong position to compete globally. Shanghai and Guangdong
follow closely, reflecting their national leadership in AI and aspirations to become a global
leader. Their ambitions are likely fueled by their robust AI industries, which position them
favourably in the global AI race. For instance, in terms of long-term objectives for AI develop
ment, both Guangdong and Shanghai aim to reach international advanced level in terms of
their overall AI development by 2030. Following that, a group of provinces and cities, including
Anhui, Zhejiang, Jiangsu, Guangxi, Shaanxi, Tianjin, Sichuan, Liaoning, Fujian, Hubei and Hunan,
have aspirations to become global leaders in certain AI fields and theory development. This
specialisation strategy could help these provinces stand out in the global competition by
capitalising on their unique strengths and expertise. For instance, Shaanxi aspires to be an
48
Jinghan Zeng, ‘Securitization of Artificial Intelligence in China’, The Chinese Journal of International Politics 14, (2021), p. 417.
49
Jinghan Zeng, ‘Securitization of Artificial Intelligence in China’, The Chinese Journal of International Politics 14, (2021), p. 417.
50
Ding J, ‘Deciphering China’s AI Dream’, Future of Humanity Institute Technical Report (2018).
51
ibid.
16 S. KHANAL ET AL.

international leader in developing AI applications in the defence sector. Hubei aims to achieve
major breakthroughs in fields such as deep learning, brain-like intelligence, cross-media analysis
and reasoning, and autonomous unmanned intelligence and accomplish several leading
research outputs with international influence. Anhui, with its existing lead in perceptual
intelligence and cognitive intelligence, intends to focus on developing a few AI industry
clusters with international competitiveness. Meanwhile, many Chinese provinces show far lesser
global ambitions or have not articulated a new generation of AI policies, indicating a more
regionalised/localised or nascent approach to AI development. Among them, Gansu indicated
in its AI development plan that it intends to provide bilateral and multilateral cross-border and
cross-regional e-commerce and other services for the international markets of Central Asia and
West Asia, whereas Yunnan aims to actively utilize its location advantages to develop AI
products and services for South Asia and Southeast Asia. At the bottom of the pyramid are
the provinces which have not introduced their AI plans. Therefore, this diversity in global
ambition levels across provinces underscores the complexity of AI development in China. While
a few provinces are setting their sights on global AI leadership, mirroring the country’s grand
AI aspirations amid the China–US AI rivalry, most Chinese provinces either focus on local
economic development or are yet to define their AI strategy. This could indicate the varying
resources, industrial strengths and strategic priorities across different provinces. In the context
of the China–US AI rivalry, the provinces and cities, such as Beijing, Shanghai and Guangdong,
with global ambitions, are set to play key roles in driving China’s AI advancement. Their success
could bolster China’s standing in the global AI race. Nevertheless, the provinces with lower
rankings highlight areas where more work could be done to elevate AI development and
competitiveness on a broader scale.
This study also shows that provincial governments within the country are instrumental in
materialising national policies. For instance, the long-term AI plan of Guangdong province envisions
that at least a third of the national government’s target for the establishment of AI firms will take
place within Guangdong. Beyond specific targets, inter alia, provincial governments set aside
subsidies and investments, develop human resource capacity, incentivise research and develop
ment, and perhaps, most importantly, provide regulatory space for piloting the use of AI within their
domain. Although not part of the analysis in this article, several provinces had developed their AI
Beijing
Guangdong, Shanghai
Tianjin, Anhui, Hubei, Fujian,
Guangxi, Hunan, Jiangsu, Liaoning,
Shaanxi, Sichuan, Zhejiang
Chongqing, Gansu, Guizhou, Heilongjiang,
Henan, Jiangxi, Jilin, Qinghai, Shangdong,
Yunnan
Hainan, Hebei, Inner Mongolia, Ningxia, Tibet, Xinjiang
Figure 4. Categorisation of China’s provinces by global AI ambition.

development policies before the NGAIDP was even launched.52
Indeed, some studies even go as far
as to suggest that the current focus of scholars on the strategic approach of AI taken by the central
government vastly exaggerates the potential geopolitical risks given the current (in)capacity of
Beijing to align or coordinate its agenda with those of the local governments.53
Third, the results of this research show the importance of local capacities in policy diffusion in
China, and the authors find subtle differences in pathways to policy diffusion. In the case of diffusion
speed, provinces’ private sector capacity and political authority play an important role. Interestingly,
in the case of transmission speed, it is not just the presence of political capacity but also the absence
of academic and economic capacity that led to faster transmission speed. In the view of the authors,
the reasons could be that in the presence of economic capacity, governments are responding to the
administrative instructions of the central government to show political loyalty and simply ‘copy and
paste’ the central policy without much effort for localisation (lack of detailed plans for implementa
tion at the local level). However, in the case of diffusion intensity, the economic conditions of the
province become crucial.
In contrast, provinces with strong economic capacity and the private sector can design and tailor
their long-term plans according to the province’s needs and, therefore, take longer to implement
such policies. Indeed, when the authors examine the correlation between the number of days
between the introduction of national policy and provincial policy (speed of diffusion) and various
contributing factors, there is a clear positive relationship between these factors and the number of
days taken for diffusion. This suggests that provinces with higher capacities take a longer time to
implement long-term plans. This preliminary finding corresponds with the results of earlier studies,
which have found that provinces with less favourable economic conditions often tend to imitate
policies and, therefore, introduce central policies faster, while wealthier provinces (and even coun
ties) tend to ‘dilute growth targets and prioritise social goals’ of the central governments’ policies as
per their own needs.54
Fourth, despite the outcomes of the fsQCA, there are a few provinces whose diffusion perfor
mances are not covered here, primarily because they are outliers. To illustrate, Guangxi is, at least in
terms of policies, one of the best-performing provinces for designing AI policies. Guangxi ranks
amongst the top 5 in terms of speed of transmission and the number of policies introduced, and it
has introduced all five varieties of policies. However, none of the contributing factors explain the
performance of Guangxi since it has historically been one of the more disadvantaged regions in
China. The results show evidence of agency amongst such provinces, who, despite their lack of
existing capacities, have attempted to use AI development as a potential instrument to improve their
economic and political capital. Guangxi’s provincial leaders, including the former party secretary and
chairman Chen Wu, are enthusiastic supporters of technological development, especially digital tech
and growing AI in the province. In addition, they have been hoping to leverage AI to improve their
economic ties and cooperation with ASEAN to boost local economic development. In terms of local
technology development, the provincial government’s strategic planning is considered of critical
importance. Guangxi provincial leaders’ strong and exceptional interest in prioritising technology
development is evidenced by the fact that the earliest to formulate policy documents for digital
government construction at the local level was Guangxi, which in 2018 promulgated the Three-Year
Action Plan for Promoting Digital Government Construction in Guangxi (2018–2020). In 2019,
Guangxi specifically set up a digital government leadership group to coordinate the guidance and
deployment of Guangxi’s digital government reform and construction work and formulate develop
ment strategies, special plans and major policies related to the construction of digital Guangxi.
52
Roberts and others (n 5); Wu and others (n 4).
53
Zeng, ‘China’s Artificial Intelligence Innovation’ (n 33).
54
Yuen Yuen Ang, ‘Domestic Flying Geese: Industrial Transfer and Delayed Policy Diffusion in China’, The China Quarterly 234,
(2018), p. 420.
18 S. KHANAL ET AL.

Similarly, in regard to AI development, the initial focus for Guangxi has been to enhance its
AI-specific technology capacity. This has involved a range of policy initiatives aimed at fostering
AI growth within the region. Key steps include the launch of an AI-focused development plan
with ambitious goals over 2, 7 and 12 years in areas such as research, market expansion and
international engagement. Additionally, the province has rolled out various AI-supportive
policies. These encompass offering financial incentives for AI research and development by
businesses, the establishment of the China-ASEAN blockchain innovation centre in 2020,
collaborating with Huawei to create an AI innovation hub and implementing strategies to
bridge AI knowledge and educational gaps in government, educational institutions and the
market at large.55
Furthermore, Guangxi has capitalised on its geographic closeness to ASEAN countries. It has
embarked on numerous significant projects at both multilateral and bilateral levels to position
itself as a leading AI collaborator for ASEAN members. For instance, in 2019, Guangxi con
ducted the inaugural China-ASEAN Artificial Intelligence Summit, drawing participants from
ASEAN and China, including government officials and private sector entities. During the AI
summit, the then Guangxi Chairman, Chen Wu, said in his opening speech of the summit that
‘Relying on the China-ASEAN Expo, Guangxi will actively promote AI technology cooperation,
product trade, investment and industrial exchange with ASEAN countries’. This summit, having
completed its fourth session in 2023, exemplifies Guangxi’s commitment to AI collaboration.
Another major endeavour is the China-ASEAN Information Harbour, conceived as the Digital Silk
Road originating from Guangxi. This platform offers digital services and connectivity to clients
in China and ASEAN.
Additionally, the province organises the China-ASEAN Information Harbour Forum annually,
akin to the AI summit, to promote technical and economic partnerships in AI among partici
pants from member states and China. Guangxi also hosts other events like the Forum on China-
ASEAN Technology Transfer and Collaborative Innovation, now in its eleventh year, to further
technological cooperation.56
More evidently, in Guangxi’s new generation of AI development
plan issued in 2018, one of the guiding principles is Creating an ASEAN-oriented Artificial
Intelligence Innovation and Application Highland, and one of the key goals stated it clearly
that Guangxi aims to turn itself into an ASEAN-oriented base for AI science and technology
innovation.
On the other hand, there are subnational governments like Beijing, which are among the
leading regions in terms of AI development and possess strong capacities across all four
dimensions. Their successes in AI development are not reflected in terms of policy initiatives.
This could be explained by Beijing’s unique features. First, Beijing enjoys an unparalleled lead
in domestic AI research and industrial development. For instance, according to research con
ducted by AlShebli et al.57
in terms of AI research capacity, Beijing emerges as a clear outlier
not just domestically but internationally, as ‘the most impactful city since 2007, the most
productive since 2002, and the one housing the largest number of AI scientists since 1995’,
and it houses nearly 30% of the AI enterprises in China.58
The huge success of Beijing’s AI
development could have enabled the City Government to follow its own development path
ways rather than being limited by the national development plan. Thus, it is not surprising that,
rather than swiftly introducing its own local New Generation of AI development plan, on
18 February 2019, it established the Beijing National New Generation Artificial Intelligence
Innovation and Development Pilot Zone, which is China’s first National New Generation
55
Zhang and Khanal (n 2).
56
ibid.
57
‘Beijing’s Central Role in Global Artificial Intelligence Research’, Scientific Reports 12, (2022), p. 21,461.
58
Beijing Municipality Government, ’7 he 5000+ Beijingren gong zhineng chanye lingpao quanguo [7 and 1500+ Beijing’s
Artificial Intelligence Industry Leads the Nation] (Beijing Municipality Government, June 6, 2021). accessed December 9, 2023.
https://www.beijing.gov.cn/fuwu/lqfw/gggs/202106/t20210609_2409514.html.

Artificial Intelligence Innovation and Development Pilot Zone. At the launching of Beijing’s Pilot
Zone, Mr. Wang Zhijun, vice minister of the Ministry of Information Technology, delivered
a speech. In this speech, Wang fully affirmed Beijing’s efforts to create a national AI innovation
and application pioneer from the perspective of the national development strategy. Second,
Beijing enjoys a special political status. As compared to the other Chinese provincial leaders,
including some of those who govern other provincial-level cities, such as Tianjin, Shanghai and
Chongqing, Cai Qi, Beijing's Party Secretary between 2017 and 2022 (Cai was Beijing Mayor
between 2016 and 2017) has been widely considered to be one of Xi Jinping’s most trusted
confidants. Given his political ties with the paramount leaders, there is little need for him to
show his political loyalty constantly and explicitly to the central government by swiftly
responding to the central mandate, as seen in the case of the State Council New Generation
of AI Development Plan. Instead, Beijing can afford to focus more on AI development on the
ground while taking time to develop its local AI plan. Beijing introduced its new generation of
AI development as late as May 2023, after nearly 6 years of the national AI plan. In this plan,
Beijing set a very ambitious goal of building a core AI industry with a scale of RMB 300 billion
(USD 42.37 billion) by 2025.
Conclusion
In this study, the authors examined the role of provincial governments in AI development in China.
For this purpose, the authors conducted a detailed analysis of the diffusion of the 2017 NGAIDP.
Overall, the findings of this research show that the focus on central and provincial governments does
not necessarily align in terms of their AI development goals. While the central government prioritises
issues of national security, defence and global competition, provincial governments are more
concerned with the economic potential of AI. The authors also found that the private sector’s
capacity and the provincial leaders’ political authority are crucial variables that determine the
speed and quality of diffusion of AI policies. The findings show a complex interplay between
provincial governments’ economic and political capacities and their strategic priorities in determin
ing their role in promoting AI. This study also examined the role of provincial governments in AI
development in China by looking at the diffusion of the 2017 NGAIDP. The authors show that the
capacity of the private sector on the demand side and the political authority of the provincial leaders
are crucial variables that determine the speed and quality of the diffusion of AI policies.
Acknowledgments
Araz Taeihagh is grateful for the support provided by the Lee Kuan Yew School of Public Policy, National University of
Singapore.
Disclosure Statement
No potential conflict of interest was reported by the author(s).
Funding
This research/project is supported by the Lee Kuan Yew School of Public Policy, National University of Singapore and the
National Research Foundation Singapore under its AI Singapore Programme [Award Number: AISG3-GV-2021-002]. Any
opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not
reflect the views of the National Research Foundation, Singapore.
20 S. KHANAL ET AL.

ORCID
Shaleen Khanal http://orcid.org/0000-0002-8097-2461
Hongzhou Zhang http://orcid.org/0000-0002-2995-4217
Araz Taeihagh http://orcid.org/0000-0002-4812-4745
CRediT Authorship Statement
Conceptualization, A.T.; methodology, S.K., H.Z., A.T.; validation, A.T.; visualisation, S.K; investigation, S.K., H.Z., A.T.;
resources, A.T.; data curation, S.K., H.Z.; writing first draft, S.K. H.Z, A.T. writing—editing, A.T., H.Z., S.K.; supervision, A.T.;
project administration, A.T.; funding acquisition, A.T. All authors read and agreed to the published version of the
manuscript.

Appendices
Appendix A. Qualitative Comparative Analysis Variables
Summary table of the relevant indicators.
Provinces
Enterprise
capacity
University
capacity GDPpc Power Diffusion
Diffusion
Days
Instrument
breadth
Policy
intensity
Anhui 0.95 0.47276 0.37315 0.33158 1 0.15385564 1 0.64311
Beijing 0.99077 0.99077 0.99752 0.03824 1 0.19865304 0.9 0.52941
Chongqing 0.70392 0.82618 0.63282 0.39637 1 0.82380248 1 0.93755
Fujian 0.90479 0.23155 0.92576 0.2726 1 0.77935342 0.9 0.90359
Gansu 0.05 0.36696 0.06864 0.85404 1 0.46133473 0 0.03824
Guangdong 0.987 0.62706 0.76129 0.22065 1 0.53499553 0.9 0.69518
Guangxi 0.05 0.05 0.11074 0.90359 1 0.66842206 1 0.85404
Guizhou 0.54319 0.05 0.11191 0.58743 1 0.97083776 0.33 0.08443
Hainan 0.05 0.05 0.26317 0.9799 0 0 0.1391
Hebei 0.05 0.05 0.15054 0.08443 0 0.9 0.2726
Heilongjiang 0.41893 0.93073 0.10892 0.52941 1 0.72740171 1 0.05
Henan 0.05 0.2726 0.20459 0.82217 1 0.11767684 0.33 0.465
Hubei 0.77073 0.97435 0.65034 0.64311 1 0.03716982 0.33 0.82217
Hunan 0.62706 0.90479 0.37304 0.96007 1 0.2526 0.33 0.2726
Inner
Mongolia
0.05 0.05 0.70315 0.95 0 0.67 0.22065
Jiangsu 0.93073 0.987 0.97024 0.465 1 0.60363498 0 0.2726
Jiangxi 0.05 0.31791 0.3286 0.88103 1 0.96175922 0 0.1391
Jilin 0.05 0.54319 0.1358 0.74269 1 0.91556764 0.9 0.33158
Liaoning 0.82618 0.77073 0.29503 0.1762 1 0.86089553 0.9 0.08443
Ningxia 0.03291 0.03291 0.31047 0.98405 0 0.9 0.06513
Qinghai 0.0406 0.0406 0.1887 0.69518 1 0.93486964 0.67 0.06513
Shaanxi 0.05 0.95 0.52932 0.06513 1 0.08910947 1 0.1391
Shandong 0.96411 0.41893 0.57182 0.1391 1 0.86089553 0 0.69518
Shanghai 0.97435 0.98172 0.99563 0.05 1 0.38582135 0.67 0.98736
Shanxi 0.36696 0.19502 0.37442 0.96819 0 1 0.10879
Sichuan 0.87048 0.96411 0.28016 0.93755 1 0.31543038 0.9 0.58743
Tianjin 0.47276 0.70392 0.89166 0.78509 1 0.05 1 0.96819
Xinjiang 0.31791 0.13541 0.29157 0.92225 0 0.9 0.03824
Yunnan 0.05 0.16304 0.20007 0.97469 1 0.06695092 0 0.02916
Zhejiang 0.98172 0.87048 0.88769 0.1087 1 0.95 0.9 0.64311
Appendix B. Truth tables of the fsQCA
Table A1. Truth table regarding the existence of policy diffusion.
Enterprise
capacity
University
capacity GDPpc Power Out N Incl PRI Cases
1 1 1 1 1 6 0.975 0.975 Beijing, Chongqing, Guangdong, Jiangsu, Shanghai,
Zhejiang
1 0 1 1 1 3 0.949 0.949 Anhui, Fujian, Shandong
0 1 0 1 1 1 0.934 0.934 Heilongjiang
1 1 0 1 1 1 0.933 0.933 Liaoning
0 1 1 1 1 1 0.93 0.93 Shaanxi
1 0 0 1 1 1 0.921 0.921 Guizhou
0 1 0 0 1 1 0.909 0.909 Jilin
1 1 1 0 1 1 0.907 0.907 Hubei
1 1 0 0 1 2 0.9 0.9 Hunan, Sichuan
0 1 1 0 1 1 0.891 0.891 Tianjin
0 0 0 1 0 1 0.697 0.697 Hebei
0 0 0 0 0 9 0.688 0.688 Gansu, Guangxi, Hainan, Henan, Jiangxi, Ningxia,
Qinghai, Xinjiang, Yunnan
0 0 1 0 0 2 0.436 0.436 Inner Mongolia, Shanxi
22 S. KHANAL ET AL.

Table A2. Truth table regarding the speed of policy diffusion.
Enterprise
capacity
University
1 0 1 1 1 2 0.895 0.762 Fujian, Guangdong
1 0 0 1 1 2 0.862 0.661 Anhui, Shandong
0 1 0 1 0 2 0.799 0.554 Heilongjiang, Hunan
1 1 0 1 0 1 0.797 0.54 Chongqing
0 1 0 0 0 1 0.752 0.466 Shaanxi
1 1 0 0 0 2 0.736 0.431 Liaoning, Sichuan
1 1 1 0 0 1 0.732 0.375 Hubei
0 0 0 0 0 8 0.725 0.593 Gansu, Guangxi, Guizhou, Henan, Jiangxi, Jilin,
Qinghai, Yunnan
0 1 1 0 0 1 0.664 0.256 Tianjing
1 1 1 1 0 4 0.664 0.378 Beijing, Jiangsu, Shanghai, Zhejiang
Table A3. Truth table regarding the speed of policy diffusion.
Enterprise
capacity
University
1 1 1 0 1 1 0.91 0.772 Hubei
0 1 1 0 1 1 0.869 0.886 Tianjin
Zhejiang
0 1 1 1 0 1 0.739 0.435 Shaanxi
1 0 0 1 0 1 0.719 0.244 Guizhou
1 1 0 0 0 2 0.703 0.244 Hunan, Sichuan
0 1 0 1 0 1 0.635 0.105 Heilongjiang
0 1 0 0 0 1 0.634 0.087 Jilin
1 1 0 1 0 1 0.633 0.172 Liaoning
0 0 0 1 0 1 0.557 0.065 Hebei
Table A4. Truth table regarding the breadth of policy diffusion.
Enterprise
capacity
University
0 1 1 1 1 1 0.867 0.806 Shaanxi
1 1 1 0 0 1 0.847 0.746 Hubei
Zhejiang
0 1 1 0 0 1 0.832 0.697 Tianjin
1 0 0 1 0 1 0.711 0.551 Guizhou
0 0 0 1 0 1 0.701 0.538 Hebei
0 1 0 0 0 1 0.65 0.39 Jilin
1 1 0 0 0 2 0.642 0.446 Hunan, Sichuan
0 1 0 1 0 1 0.61 0.397 Heilongjiang
1 1 0 1 0 1 0.539 0.422 Liaoning

Table A5. Necessity of conditions in explaining the diffusion of policies.
Contributing factors InCLN RON CoVN
Enterprise capacity 0.544 0.96 0.95
University capacity 0.586 0.981 0.979
GDPpc 0.508 0.872 0.843
Power 0.554 0.93 0.919
~Enterprise capacity 0.456 0.721 0.673
~University capacity 0.414 0.716 0.636
~GDPpc 0.492 0.795 0.76
~Power 0.446 0.749 0.689
Table A6. Necessity of conditions in explaining speed of diffusion.
GDPpc 0.457 0.813 0.63
Power 0.609 0.771 0.676
~GDPpc 0.748 0.636 0.644
~Power 0.598 0.7 0.607
Table A7. Necessity of conditions in explaining the intensity of diffusion.
GDPpc 0.798 0.763 0.667
Power 0.716 0.728 0.599
~GDPpc 0.42 0.58 0.326
~Power 0.486 0.599 0.378
Table A8. Necessity of conditions in explaining the breadth of diffusion.
GDPpc 0.649 0.825 0.722
Power 0.635 0.815 0.755
~GDPpc 0.467 0.658 0.517
~Power 0.466 0.657 0.515
24 S. KHANAL ET AL.

Development of New Generation of Artificial Intelligence in China

More Related Content

Similar to Development of New Generation of Artificial Intelligence in China (20)

More from Araz Taeihagh (20)

Recently uploaded (20)

Development of New Generation of Artificial Intelligence in China