The Evolution of AI Security in China: A New Engagement Between Science and Politics | Generative ai tools for images | Generative ai chatgpt | Google generative ai course | Turtles AI

The Evolution of AI Security in China: A New Engagement Between Science and Politics
Growing Interest in AI Risks in China and Regulatory Prospects at National and International Levels
Editorial Team4 December 2024

 

Over the past two years, China’s approach to AI safety has undergone a significant shift. While concerns about catastrophic risks from powerful AI were limited in the past, discussions about AI safety are now becoming increasingly prominent in both the scientific community and Chinese policymaking. With the emergence of new research, public statements, and government documents, China appears to recognize the need to manage the risks associated with developing advanced AI.

Key Points:

  • Growing concerns about AI risks in China.
  • The importance of China in global AI governance and security.
  • The influence of Chinese scientists and their engagement in AI security research.
  • Recent policy statements and the initiative for AI oversight.

Over the past two years, China has shown a significant shift in its approach to AI safety, a topic that previously did not seem to occupy a central place in the country’s public and scientific debates. Until recently, concerns about the catastrophic risks posed by very powerful AI systems, such as general-purpose AI (AGI), were predominantly prevalent in Western scientific circles. Scientists and AI experts from the US and UK had warned about the potential future threats these systems could pose, but China seemed to ignore or downplay these concerns. However, recent developments suggest that China is starting to take these risks seriously, although not all aspects of its strategy to address them are clear. A significant sign of this evolution is the Chinese leadership’s adoption of policy measures to create oversight systems to ensure AI safety. In a document released by the Chinese Communist Party (CCP) in July 2024, it specifically calls for the establishment of a supervisory and regulatory system to monitor AI risks. This policy signal adds to a growing body of scientific research exploring how to prevent the dangers of developing advanced AI.

The emergence of these concerns began around 2020, when some of China’s leading scientists and researchers began to speak publicly about the risks of developing AGI, such as Wen Gao, director of one of the most prestigious AI research centers and a professor at Peking University. In 2021, Gao co-authored a paper exploring the dangers of AGI being self-sufficient and the possibility that the technology could unleash an uncontrolled intellectual explosion. Another important signal of change came in 2023, when a number of prominent Chinese scientists began to actively participate in international discussions on AI safety, such as Turing Award winner Andrew Yao. Yao has publicly emphasized the need to address the extreme risks of AI, co-authoring a major joint paper with international scientists, which expressed concern about the potentially devastating effects of rapid AI evolution.

In the context of government policies, the September 2021 document “Ethical Standards for the Next Generation of AI,” issued by the Expert Committee on AI Governance, included for the first time a specific recommendation on the need to ensure that AI remains under human control at all times. Although this recommendation was not binding, it was a clear sign of how security concerns were finding their way into public debate and policy decisions. However, the first regulations on the subject, issued by the Cyberspace Administration of China (CAC) starting in 2022, mainly focused on regulating content and algorithms, without addressing the more serious issues related to the future security of AI. Only with the arrival of ChatGPT, in the last quarter of 2022, did the discussion intensify. In early 2023, China has been an active participant in the global debate on AI safety, as evidenced by the forum held at the Beijing Academy of Artificial Intelligence and a series of statements by leading figures, including internationally renowned scientists such as Yao.

2023 also saw the participation of Chinese scientists and officials at the first AI Safety Summit in the UK, where China signed the Bletchley Declaration, an international agreement recognizing the need to work together to reduce the risks associated with advanced AI. Although the declaration did not impose any formal binding commitments, China’s signing of the declaration showed a symbolic commitment to international cooperation on AI safety. Starting in 2024, the topic of AI safety has emerged even more prominently at the World AI Conference in Shanghai, where both technical and regulatory aspects were explored, marking a further step towards establishing clear safety policies. Here, the Shanghai AI Lab released a report proposing to treat AI safety research results as global public goods, advocating for international funding and dissemination.

In July 2024, the CCP adopted new language to address AI risks. The Third Plenary Final Declaration included an explicit reference to the need to establish a supervisory system to ensure AI safety, explicitly citing threats to public safety. Although the document raises many questions about the scope and objectives of such measures, it is significant that China is now treating AI safety as a priority. The proposed supervision comes in the context of global and public safety risks, such as natural disasters or health emergencies, but it is not yet clear whether this is an approach aimed solely at regulating general AI or also at risks related to advanced AI models. China seems intent on shaping AI safety governance that can avoid the catastrophic scenarios envisioned by international scientists and policymakers.

The implementation of these ideas will depend on future choices regarding AI safety institutions.