Artificial Intelligence (AI) is a semantic umbrella term widely referred to as “a system’s ability to interpret external data correctly, to learn from such data, and to use those learnings to achieve specific goals and tasks through flexible adaptation” (Haenlein and Kaplan 2019). It uses databases, software, and large language models (LLM) to build an original response, improving itself through feedback. Generative AI converts “simplified representation of their training data and draw[s] from it to create a new work that’s similar, but not identical, to the original data” (IBM Research 2023). Video, text, audio, software code, simulations, and other data are all probable outputs. AI uses mathematics, linguistics, and psychology to learn and generate truthful and statistically significant answers. AI’s learning capacity and speed, being connected to the Internet, has multiplied after the introduction of real conversations in 2022. With these new iterations and processes, the algorithm is self-strengthening, capable of learning and mastering almost all disciplines (Domingos 2015). Generative AI produces new and increasingly accurate products and services, answering complex questions based on access to diverse sources, operating not in a vacuum but by following set patterns learned from data libraries.

AI's strategic nature will re-shape our current understanding of power and influence, and subsequently, diplomacy. In security, AI will affect the base conceptions of national security and its premises. In economy, financial resources, R&D investments, and training, it will build up “an opportunity to leapfrog foreign competitors” (Fischer 2018, p. 2). In the balance of power, AI can affect Sino-American competition, weaken democracy, promote nationalistic responses, increase private power, and impact military power and capabilities (Franke 2021). Generative AI technologies intensify the challenges as it can be used for analytical roles (data management and cognitive tasks), predictive uses (policymaking), and operational practice (military and human security). Having agency, AI poses a significant challenge in decision-making, regulation, competition, and human rights (Cummings et al. 2018). In diplomacy, generative AI affects the diplomatic profession due to its inherent drawing on both correct and incorrect information sources, potentially decreasing public confidence in multilateral institutions (Manor 2023). Other challenges include privacy and personal data regulation, intellectual property rights, and the cooperation with corporations and private actors. AI narrative builds an aura of innovation and technological reputation for public diplomacy purposes.

This paper explores the discourses around the development of generative AI technologies in Chinese diplomacy. The topic's impact is relevant because (1) in the growing geopolitical rivalry, China represents a distinct set of values that attracts a growing number of actors in the so-called Global South. Its response to technological development will produce standards that other countries will imitate. (2) China is a norm and policy entrepreneur in multilateral organizations. Its agenda favors the formation of new coalitions and strategic alliances (e.g., BRICS +) that have broader implications for international order.

It is possible to identify four key topics that shape the diplomatic debates around generative AI technologies in China from available policy documents and expert analyses from both Chinese and non-Chinese sources. China has been closely observing the emergence of ChatGPT (Wang and Zhu 2023). Wang and Zhao (2023, p. 7) believe that generative AI can pose a threat to China’s ideological security, which in their view, is “an extremely important task for the [Chinese Communist] Party.” In this respect, Wen (2023) considers that to prevent the ideological risk posed by ChatGPT, as having been developed by a capitalist society, China should promote its industry to consolidate the country’s dominant position as a representative of socialism. Cheng (2023b, a) argues that China should strive to balance both security and development in the management of generative AI. At an international level, Qi (2023) believes that China needs to actively promote international cooperation, global governance, and regulation of AI technologies. In the public diplomacy field, Ou and Liu (2020) identify at least three challenges to the application of AI technologies: (1) the difficulty in processing massive amounts of data and discriminating between truth and “distortions of reality”; (2) the enormous power concentrated in the hands of non-state actors such as tech giants and the impact thereof on public diplomacy; and (3) the qualitative limitations of “artificial” intelligence technologies and the “emotional” intelligence of human beings.

(Techno-) socialism with Chinese characteristics

In 2015, the Chinese government announced the “Made in China 2025” plan, a state-led industrial policy oriented to global high-tech manufacturing and standard regulation. Robotics, semiconductor manufacturing, electronic vehicles, energy, and artificial intelligence are part of the commitment to use technology to leverage economic transformation. The Belt Road Initiative or the Asian Infrastructures Investment Bank completes the economic statecraft. The unique proposal aspires to realize modern state capitalism where government, private initiatives, and technological development converge (van Apeldoorn and de Graaff 2022). The Chinese strategy articulates the economy, sets priorities, and uses resources to achieve policy objectives (Kurlantzick 2016). The policy aims to reduce dependencies on external markets, promote its own technology, and lead in high-value-added industries (Du 2023). Combining financial support with control of private initiatives, the policies include subsidies, taxes, preferential treatment, access to data, public procurement strategic investments, state-owned companies, export restrictions, forced transfer agreements, and the defense of national champions. In the technological context, private initiatives adapt to Chinese characteristics (Lei 2023). The government narrative posits technosocialism as an alternative to the “neoliberal trap” of Western economies (Cheng 2023b, a).

China’s strategy takes shape in its public policy making. The Cyberspace Administration of China (CAC) holds that AI must abide by the “core values of socialism with Chinese characteristics.” This is explicitly stated in two relevant documents: “Administrative Provisions on Deep Synthesis for Internet Information Services” (Cyberspace Administration of China 2022) and “Interim Measures for the Management of Generative Artificial Intelligence Services” (Cyberspace Administration of China 2023). These interventionist regulations aim to control the AI industry through the identification of legitimate sources. The narrative of intellectual property rights prevents the free use of non-authorized data, platforms, or algorithms. This regulation is particularly relevant for companies offering services to public institutions.

Chinese strategic planning has been successful, but it fails to address the upgrade bottleneck (Cao 2023) and the regulatory environment. Even though Chinese technology has proven to be leading in areas such as facial recognition, surveillance, or biomedicine, innovation under state control and regulatory hurdles is unfriendly to the AI market. This is not a problem of technical incapacity but of the desire to control the industry's growth and the services offered. Considering this, investors face a conundrum. While they do not want to offend the government and harm other economic activities, they cannot develop at the same speed as European and US competitors. Moreover, they cannot guarantee a quality product if they cannot work with open databases (Davidson 2023). Constrained by design, the regulatory approach increases costs and slows down a local alternative. Netease.com, JD.com, Tencent, Huawei, Kuaishou, and Qihoo 360 are progressing but have yet to advance any real technological solution (Huang 2023). However, without clear compliance, companies follow Weibo or Bytedance’s model, platforms responsible for posting and surveilling content. Likewise, the prohibition of the use of public sources (Reddit or Wikipedia) has restricted the ability to learn, handle real language, and systematically update knowledge. Dcard, Baidu PostBar, Tieba, and Zhihu are the local services where LLM can learn everyday language. However, those are controlled and supervised.

In the absence of anonymity, standardized language provides fewer active groups. Baidu has unveiled the Ernie chatbot, which is aimed at intelligent search, electric and automated driving, and smart assistance. These sectors are removed from politics, as there is no requirement for freedom of expression. Therefore, it is relevant to understand which sectors are preferred and receive state support. Chinese companies are facing a trade-off. Domestically, they are moving toward B2B solutions (search, electric vehicles, consumer services, vertical services) that use mathematical logic and rationality. At the same time, they are abandoning the B2C race (conversational chatbot) that requires access to sources and social sciences (history, politics, current affairs). This model limits the effectiveness of "Made in China 2025" and generates new security problems at the domestic and international levels.

The securitization of AI

As argued by Buzan, Waever, and de Wilde (1997), the attribution and classification as a "national security" threat anticipates political action and behavior. In the Chinese context, the first level of securitization connects to the “comprehensive national security” advanced by President XI a decade ago (Xi 2014), which has now been expanded to include new key areas such as AI (Chen 2022). We identify two impacts in the diplomatic domain. AI has dual-use in civilian and military technologies, seen as a matter of national security and a strategic asset where defense, deterrence, public communication, and economics converge. In line with the agenda set by the CPC National Congress, the Foreign Affairs Minister considers science and technology to be key elements in China's strategy to develop domestically and increase its influence internationally (Xinhua 2017). Against this view, ChatGPT is perceived as a possible risk to internal domestic stability as it “can not only become a tool for the powerful to promote their political intentions, but also provide space for a small number of extremists to propagate their ideas, and even provide space for ideological infiltration and interference by Western countries in China” (Zhang 2023, p. 8). In this sense, the development of AI also concerns the military. In fact, the People’s Liberation Army (PLA) recognizes the role of AI in warfare and how it affects training, command chains, decision-making, intelligence analysis, and future capabilities. Private companies are needed to achieve this goal and lead technological competition (Kania 2017). That being said, the Foreign Affairs Ministry and the PLA pose distinct or at least nuanced positions on technological developments. Both actors do not always share priorities, creating a clash of strategic cultures (Allen 2022).

The second level of securitization is closely linked to economic security and global rivalry. The recurrent argument focuses on the dual-use nature of processors, whereby the military aspect constitutes a serious concern for the United States government. Restrictions on the export of technology such as quantum computing, advanced chips, and AI components have limited its development in the Chinese market. Taiwan, Japan, and South Korea produce 90% of the semiconductor industry (Asia Report 2023). In September 2022, the White House allocated $52 billion to the Chips and Science Act to support semiconductor research and manufacturing and limited the sale and export of high-end processors and data center graphic processing units to China. Chinese diplomacy considers the decision a “technological blockade” from the “technological hegemon” (Langley and Waters 2022). Chinese diplomacy points out that the restrictions hinder global free trade. and claims that Western countries want to impose conditions and dependencies on other countries. The “Made in China 2025” plan cannot be implemented without open access to semiconductors. The Chinese government accuses the move of having “undermined the legitimate rights and interest and the stability of global industry and supply chain” (Langley and Waters 2022). However, China’s regulations appear to contradict its own global trade narrative. Domestic regulations impose three artificial barriers: (1) limiting the inflow of foreign capital, (2) promoting the local market and the “core values of socialism,” and (3) controlling the exports of chipmaking metals through licenses (Hunter and Cang 2023). Local generative AI may not be compatible with American standards, so global companies need to develop alternative solutions for the Chinese market. The consequence is a rise in technological decoupling.

The third element of securitization relates to reputation. The technology race is not about chips and semiconductors but about the ability to deliver products, services, and solutions fit for the digital society. Lack of market access and fear of freedom of expression lead to a very limited policy response. If the industry becomes obsolete, the delay in creating a Chinese service negatively impacts the country's reputation as a technology benchmark, while American companies lead the global tech innovation race.

Domestic policy: soft censorship?

Domestically, discussions on AI regulation are also framed by the public interest, i.e., censorship is justified to fight against misinformation and the goal of maintaining social stability. The public interest approach connects with “social morality and ethics” (Ye 2022). In fact, the blocking of ChatGPT has been justified by “violating relevant laws and regulations” (Davidson 2023). Other arguments recall that generative AI can “produce large amounts of misleading information,” is built on “biased training data,” causes “data leakage,” and affects “intellectual property rights” (Zhi 2023).

Creating original content generates a new information activity environment, so the CAC is committed to transparency and accountability. The regulation uses general principles but with Chinese characteristics. This means, among other items, “deleting irrelevant content” that harms the political system or the country’s strategic interests (Cyberspace Administration of China 2023). The commitment to updating the algorithm, labeling generated content, and identifying the creators of generative videos pursue the same ends. Sources, ideas, and references that do not agree with the official version are subject to censorship. A total of 41 companies have joined the process, bringing their activity under the umbrella of relevant national security actors in the fight against disinformation (Cai 2023). Alibaba, Baidu, and Tencent are committed to collaborating on control and security tasks.

The restrictions amount to a new type of censorship. It consists of the incentive of automated responses, the concealment of the databases that build the response, the refusal to respond to disputes, and the transfer of responsibility on companies (which prefer to remove content rather than discuss with the government), among other measures. These actions standardize the responses. International companies cannot operate, and local services respond in English that they are programmed for “politically sensitive content about the Chinese government or Communist Party of China” (Zheng 2023). The list of restricted activities or topics is unavailable in the local market version. There is also no definition of “false information,” “true and accurate,” or “discriminatory.” Without a definition, soft censorship is consolidated.

Conclusions

Generative AI consolidates the convergence between private economic actors and political or military interests in the diplomatic field in the world in general, as well as in China in particular. The new great game is digital both in the manufacturing of goods (supply chains, foreign markets) and in the provision of services (information management). At this point, it is surprising that technological diplomacy has not materialized in appointments. Chinese diplomacy has not appointed diplomats in top positions to defend its interests in Silicon Valley or Brussels. On the same page, neither the European Union nor the United States have granted such status to their representatives in the Chinese market. In economic terms, technosocialism aims to control AI developments. Thus, given the regulatory conditions, venture capital is not interested in investing. Consequently, local brands accelerate decoupling from global AI, creating the conditions for maintaining the “Chinese characteristics.” From a diplomatic perspective, the lack of access makes it difficult to penetrate Chinese society, affecting potential track II tools (culture, science, public opinion). Generative AI content will be based only on local, government-approved, or government-friendly content.

The second conclusion points to the weakness of Chinese perceived tech sophistication. Controlling emerging technologies and the market means credibility, particularly in targeted markets and Belt Road Initiatives participants. China's competitiveness is limited in the current competition for semiconductors and chips, and the emerging ‘new technology diplomacy’ (Feijóo et al. 2020) is weakened. If China does not consolidate its technological capacity, it will lose relevant market positions, and it will be difficult to exert its current political and diplomatic influence. An actor with a global reputation, an alternative to the European or U.S. models needs a generative AI sector to foster innovation, exports, and public diplomacy. Hence, the economic problem affects global reputation and, ultimately, the execution of any diplomatic strategy.