Ever since the popularisation of Open AI’s ChatGPT in late November 2022, it has been virtually impossible to remain unaware of the transformative potential brought about by artificial intelligence (AI). Particularly, the ongoing improvement of large language models (LLMs) has provided AI with unprecedented generative capabilities. Generative AI (GenAI) can produce text, multimedia content and contribute to new software development. The technology paved the way for further innovation and new economic development opportunities. However, GenAI has also proven to be a potential source of threats arising in various forms if not regulated properly. Spreading false or undesired information is only one of a few examples.
Therefore, it should not come as a surprise that rapid GenAI innovation has gone hand in hand with increasing regulation from institutions across the world. These regulations are useful documents showing a country’s attitude toward GenAI. How should GenAI be understood? Should its development be a priority of governments’ policies? What could be the risks associated with GenAI? This article attempts to identify China’s stance on these issues.
China’s AI in Context
China is now strongly encouraging the use of AI in several sectors, but just a few years ago, the domestic scenario was vastly different. Beijing started tightening AI regulation in 2021, when the Cyberspace Administration of China (CAC) published the “Guiding Opinions on Strengthening the Comprehensive Governance of Network Information Service Algorithms” (Cyberspace Administration of China, 2021). At the time, artificial intelligence was primarily a matter for experts in the field, and people around the world were only beginning to understand its potential. It was not until the following year, when the release of ChatGPT popularised GenAI and made its capabilities known to the public, that companies everywhere rushed to develop their own GenAI models. However, such a frenzy was absent among major Chinese tech companies. To this day, ChatGPT has never been allowed in China, and Baidu’s Ernie Bot – its most serious Chinese competitor – remained in trial mode or for limited business use only for months. Only at the end of August 2023, about two weeks after the CAC published new GenAI regulations, Baidu, Sensetime, Baichuan and Zhipu AI received Beijing’s approval to provide their GenAI services to the public (Huang, Zhang, Zheng, 2023). Not surprisingly, Article 4, subsection 1 of the regulation stood out in debates at that time, as it states that GenAI service providers and its users should adhere to “socialist values” (Cyberspace Administration of China, 2023).
After the 2023 regulation, Beijing has steadily promoted the use of AI, and GenAI has also gradually developed both among businesses and the public. The latest major development in such a trend took place when the term “AI Plus Initiative” became a buzzword of this year's Two Sessions, showing just how much China is now counting on AI to boost its economic growth (Global Times, 2024). The results of the leadership’s new stance on AI seem encouraging. On 16 April, Baidu’s co-founder and chief executive Li Yanhong claimed that Ernie Bot had attracted more than 200 million users, 85,000 of which are enterprise clients (Deng, 2024). Similarly, a few days later, Beijing-based company Baichuan joined the club of Chinese AI unicorns, namely start-ups valued at more than $1 billion, together with Zhipu AI, Moonshot and MiniMax. These giants have recently become publicly known as the “four new AI tigers of China” (Jiang, 2024a).
China’s vision of AI
Beijing’s new AI Plus initiative aims to integrate AI into products and services of existing industries, most notably those of the real economy. A focus is particularly placed on areas regarded as key for economic growth, like high-end manufacturing and healthcare. The government’s work report delivered by Li Qiang at the second session of this year’s National People’s Congress regarded industrial system modernisation and new quality productive forces development as major tasks of 2024. To achieve this, the government vowed to boost digital economy innovation through research and development, advancing big data and AI application (Li, 2024). No specific mention of its generative capabilities was made.
Like elsewhere in the world, the Chinese leadership is well aware of the risks brought about by integrating machine intelligence into existing socioeconomic structures. The tension between opportunities and risks is evident in the “White Paper on AI-Generated Content (AIGC)” published by the China Academy of Information and Communication Technology, a think tank under the Ministry of Industry and Information Technology. The paper, promulgated just a few months before the release of ChatGPT, assessed positively AI’s domestic impact across sectors, such as in e-commerce and healthcare. However, it also reminded readers that the state-of-the-art technology had yet been refined to the best of its abilities: current usage may pose a threat to internal information and network security (China Academy of Information and Communication Technology, 2022). The Chinese leadership is well aware that GenAI, if not handled properly, may run counter to its efforts to control public information.
Beijing's efforts to boost AI should also be fathomed as a new aspect of competition between the US and China. Huang Rihan and Yao Haolong (2023) stated that ChatGPT will affect countries’ domestic security, linking ChatGPT’s deep data retrieval and serialisation capabilities to the risks of disrupting state confidentiality. The authors argued that ChatGPT’s abilities may render countries with less technological prowess unilaterally transparent to more technologically developed ones (Huang, Yao, 2023). According to this line of thinking, the US's AI leadership is a tool for Washington to maintain its global hegemony. Such views have gradually gained a foothold among the Chinese leadership, especially since the Biden administration began restricting high-end chip exports to China (Allen, 2022). The latest manifestation of the US-China tech race took place at the Digital Silk Road (DSR) Forum in April 2024. At the forum, hosted by the World Internet Centre (WIC), vice-chairman of the 14th Chinese People’s Political Consultative Conference (CPPCC) National Committee Wang Yong called on DSR members to cooperate in advancing technological sharing and opposing blockades and decoupling tendencies (Jiang, 2024b).
What’s next?
China’s success in digital technologies has often been attributed to regulators’ initial tolerance or experimentation. With GenAI, Beijing is at a crossroads. Too much tolerance may result in chatbots returning undesired information to users. On the other hand, overly curbing LLMs development in the name of internal security is likely to hinder technological innovation. This is because the bigger the dataset on which the LLM is trained, the more accurate the output it generates. But if the dataset contains undesired data leading to the risk of the LLM returning taboo information, the developer is left with two options. The first is to enhance post-training control, which is something Western countries are also regulating on, as even their superior LLMs still struggle to avoid harmful content being returned to users. The second and more efficient way seems to be to better select the training material, although at the expense of performance, as the size of datasets are smaller and LLMs are therefore less accurate.
Ever since GenAI became popular, Beijing has always shown great determination in the attempt to tame this technology. Ernie Bot systematically dodged sensitive questions before its release to the public and, to a lesser degree, has continued to do so afterwards, whereas other companies unable to comply with regulations fared a lot worse (Feng, 2023; Tho and Gan, 2023; Ye, 2023). Given the leadership’s heightened focus on perceived threats to national security, GenAI development in China will likely continue to be slowed down to some extent, as companies will have to strike the right balance between pushing LLMs’ training and adhering to the values of the Chinese Communist Party (CCP).
A new LLM dataset approved by the CCP was released by CAC in 2023 in order to train language models in accordance with CCP values. However, the size of the dataset is still significantly smaller compared to that used by Western competitors (The Decoder, 2023). Similarly, in May 2024 the China Cyberspace Research Institute, operating under the CAC, launched a LLM based on “Xi Jinping Thought on Socialism with Chinese Characteristics for a New Era.” Demonstrations of the LLM at work involved questions about “new productive forces”, a buzzword from the annual meeting of the National People Congress held a few months before, suggesting that that the model is being trained on official documents and that the party is trying to extend its propaganda to the realm of GenAI (Zhuang, 2024).
Beijing will also exert its influence in international GenAI regulation. Recently, Baidu, Ant Group and Tencent Holdings have joined forces with OpenAI and Microsoft to set up the first international governmental standards in the field of artificial intelligence. During a side event at the United Nations Science and Technology Conference organised by the World Digital Technology Academy (WDTA), the companies released the “Generative AI application security testing and validation standard” and the “Large Language Model Security Testing Method.” The WDTA has also published the “Generative AI Talent Development Framework,” a document that assesses the impact of AI as well as its applications and connected risks. On the one hand, it is reported that AI will somewhat contribute to global economic growth and that the economies of China and North America will benefit the most from it in the following years. On the other hand, the report mentions risks associated with misinformation, privacy, social engineering, and intellectual property (World Digital Technology Academy, 2024). It is a new step towards designing global AI regulations and an opportunity for China and the US to shape their own narrative on GenAI.
Conclusion
China is currently lagging behind in GenAI, focusing instead on integrating AI into the existing economy, where the gap with the West is instead narrowing at a remarkable pace. At the same time, the participation of Baidu, Ant Group, and Tencent Holding in releasing the first international GenAI regulations has demonstrated China’s awareness of the importance that the generative potential of AI will have in the future. This falls within Beijing’s attempt to affirm its global vision across a range of specific core technology sectors, most notably through the DSR project, and suggests that Beijing is committed to develop its own GenAI models, although at a slower pace for now.
ABOUT THE AUTHOR
Pietro Damiani is a recent master graduate from Ca’ Foscari University of Venice, where he achieved a master’s degree in China Studies. He spent time in Beijing and was an exchange student at Jilin International Studies University. He is passionate about Chinese economic History, particularly in a global context.
This article was edited by Kalos Lau and Mattia Sossio.
BIBLIOGRAPHY
Allen, G. C., 2022. Choking off China’s Access to the Future of AI. Washington, Center for Stratetic and International Studies. Available from: https://www.csis.org/analysis/choking-chinas-access-future-ai.
China. China Academy of Information and Communication Technology, 2022. White Paper on AI-Generated Content (AIGC) [Online]. Beijing: China Academy of Information and Communication Technology. Available from: http://www.caict.ac.cn/english/research/whitepapers/202211/t20221111_411288.html .
China. Cyberspace Administration of China, 2021. Guiding Opinions on Strengthening the Comprehensive Governance of Network Information Service Algorithms. Beijing: Ministry of Education of the People’s Republic of China. Available from: http://www.moe.gov.cn/jyb_xxgk/moe_1777/moe_1779/202109/t20210929_568182.html.
China. Cyberspace Administration of China, 2023. Interim Measures for the Management of Generative Artificial Intelligent Services. Beijing: Cyberspace Administration of China. Available from: https://www.cac.gov.cn/2023-07/13/c_1690898327029107.htm.
Deng, I., 2024. Baidu claims 200 million users for Ernie chatbot, in sign that generative AI investments are paying off. South China Morning Post [Online], 6 April. Available from: https://www.scmp.com/tech/tech-trends/article/3259205/baidu-claims-200-million-users-ernie-chatbot-sign-generative-ai-investments-are-paying?campaign=3259205&module=perpetual_scroll_0&pgtype=article.
Feng, C., 2023. ChatGPT vs Ernie Bot: Baidu’s AI product has an issue with politics but is adept at grabbing up-to-date information. South China Morning Post [Online], 26 March. Available from: https://www.scmp.com/tech/article/3214782/chatgpt-vs-ernie-bot-baidus-ai-product-has-issue-politics-adept-grabbing-date-information.
Global Times, 2023. China to Launch AI Plus Initiative: Government Work Report. Global Times [Online], 5 March. Available from:https://www.globaltimes.cn/page/202403/1308210.shtml.
Huang, J., Zhang, J., Zheng, S., 2023. Baidu, Sense Time Among First Firms to Win China AI Approval, Bloomberg [Online], 31 August. Available from: https://www.bloomberg.com/news/articles/2023-08-30/baidu-among-first-firms-to-win-china-approval-for-ai-models.
Huang, R. 黄日涵, Yao, H. 姚浩龙, 2022. Bei chongsu de shijie? ChatGPT jueqi xia rengong zhineng yu guojia anquan xin tezheng 被重塑的世界？ChatGPT 崛起下人工智能与国家安全新特征 [A reshaped world? New characteristics of AI and national security unders the rise of ChatGPT]. Guoji anquan yanjiu 国际安全研究, 41(4), pp. 82-106, doi: 10.14093/j.cnki.cn10-1132/d.2023.04.004.
Jiang, B., 2024a. China’s 4 new ‘AI tigers’ – Baichuan, Zhipu AI, Moonshot AI and MiniMax – emerge as investor favourites. South China Morning Post [Online], 19 April. Available from: https://www.scmp.com/tech/big-tech/article/3259499/chinas-four-new-ai-tigers-baichuan-zhipu-ai-moonshot-ai-and-minimax-emerge-investor-favourites?campaign=6da08e9c-fe7d-11ee-a9d4-36b302c79d58&module=AI_Recommended_for_you_In-house&pgtype=section.
Jiang, B., 2024b. Officials promote China-led cyber governance to Belt and Road members during the Digital Silk Road forum in Xian. South China Morning Post [Online], 16 April. Available from: https://www.scmp.com/tech/tech-trends/article/3259176/officials-promote-china-led-cyber-governance-belt-and-road-members-during-digital-silk-road-forum.
Li, Qiang. 2024. Report on the Work of the Government [online]. Beijing, Xinhua. Available from: https://english.news.cn/20240312/0eb7388829f1400cb2f50a49f1373932/c.html.
The Decoder, 2023. CCP releases politically approved LLM dataset with 50 billion tokens. The Decoder [Online], 29 December. Available from: https://the-decoder.com/ccp-releases-politically-approved-llm-dataset-with-50-billion-tokens/
Tho, M. and Gan, N., 2023. We asked GPT-4 and Chinese rival ERNIE the same questions. Here’s how they answered. CNN [Online], 15 December. Available from: https://edition.cnn.com/2023/12/15/tech/gpt4-china-baidu-ernie-ai-comparison-intl-hnk/index.html.
World Digital Technology Academy. Generative AI Talent Development Framework [Online]. Geneva: World Digital Technology Academy. Available from: https://wdtacademy.org/publications/GenerativeAiTalentDevelopmentFramework.
Ye, J., 2023. Shares in China's iFlyTek tumble after reports AI-powered device criticised Mao. Reuters [Online], 24 October. Available from: https://www.reuters.com/technology/shares-chinas-iflytek-tumble-after-reports-ai-powered-device-criticised-mao-2023-10-24/.
Zhuang, S., 2024. China rolls out large language model AI based on Xi Jinping Thought. South China Morning Post [Online], 21 May.
Comments