Key facts
- Over 1,100 AI employees signed a petition urging the US government to pace the industry.
- The performance gap between leading US and Chinese AI models has narrowed to 39 points from over 1,300.
- China launched the World Artificial Intelligence Cooperation Organization (WAICO) with 29 founding members, excluding major Western democracies.
- Chinese AI models must undergo government safety and content reviews before release.
- US Secretary of State Marco Rubio directed diplomats to counter arguments for 'AI sovereignty'.
While a significant number of AI professionals in the West have called for a slowdown in the industry's pace, China has shown no inclination to follow suit. This divergence in approach comes as China has rapidly closed the performance gap with leading AI models from the US and other Western countries. A recent Stanford AI Index report indicated that the performance difference between top US and Chinese models has shrunk dramatically, with China also surpassing the US in AI research, patents, and robotics.
Instead of joining a Western-led effort to regulate AI development, China has been actively building its own framework. In July, Beijing launched the World Artificial Intelligence Cooperation Organization (WAICO) in Shanghai. This new intergovernmental body includes countries like Russia and Brazil among its 29 founding members but notably excludes the US, UK, and EU, signaling a move towards establishing parallel global AI governance standards.
Despite this parallel track, China has not entirely disengaged from Western safety initiatives. Chinese officials have signed the Bletchley Declaration, an international agreement on AI safety, and participated in dialogues with Washington and co-sponsored a UN resolution on the topic. However, this cooperation appears to be supplementary to China's drive for independent standards and alliances.
China's public stance emphasizes openness, with its AI industry consistently favoring open-source and open-weight models. This approach, where model designs and code are made publicly available, is contrasted with closed models like ChatGPT. Proponents, including Nvidia CEO Jensen Huang, argue that open models enhance safety by allowing external researchers to identify and fix vulnerabilities, preventing a single point of failure.