China, US need every space for cooperation in AI

The third China-US Track 1.5 Dialogue was held in Beijing in late August, with participants discussing constructive strategic stability and practical cooperation between the two countries. Artificial intelligence was among the areas identified as requiring greater communication and cooperation.

That may sound surprising at a time when the technology relationship between the world's two largest economies is becoming more contentious. China and the United States are competing over advanced chips, computing power, industrial supply chains and the rules that will shape the future of AI. Yet AI presents an unusual dilemma: the faster the technology develops, the harder it becomes for either country to manage its risks in isolation.

China's rapid progress is already reshaping the global AI development landscape. The rise of open-source models in particular is challenging a market long dominated by a small number of US companies and their expensive, proprietary systems. In July, China released the Kimi K3 model — an open-weight model from Chinese start-up Moonshot AI. That Microsoft, Amazon and Google were exploring business arrangements involving Kimi K3, while some US companies were testing the model because of its cost advantages, suggests that Chinese AI is no longer simply trying to catch up. It is beginning to influence technology choices and business models in the US market.

That shift is changing the nature of the competition. When the US enjoyed a clear lead in advanced models, computing, chips and development ecosystems, its China policy focused largely on restricting the transfer of critical technologies and hardware. As Chinese models improve, particularly through lower-cost and more accessible open-source development, Washington faces a different question: not simply how to prevent China from obtaining US technology, but how to respond when Chinese technology itself begins to spread globally.

This has also sharpened US concerns over how Chinese companies develop their models, including questions about intellectual property and national security. The US government has increasingly focused on whether Chinese enterprises can obtain capabilities from leading US models through techniques such as distillation. After the release of Kimi K3, debate in Washington over whether further restrictions on Chinese AI models were necessary intensified.

The US has already built an extensive system of export controls aimed in part at slowing China's access to advanced AI capabilities. Restrictions have expanded beyond chips and semiconductor equipment to investment screening, entity listings and efforts to prevent restricted computing resources from reaching China through third countries. Washington has also worked with allies to tighten those channels.

But the more consequential development may be the spread of the rivalry beyond bilateral competition and into the international system.

In July, China announced the creation of the World Artificial Intelligence Cooperation Organization, with 29 founding members; it later expanded to 38 members. At the same time, the US State Department planned to ask the 35 countries that had signed the AI Opportunity Statement not to join initiatives that conflict with the US framework. Clearly, participation in the US Department of State's Pax Silica initiative would entail commitments, not simply membership.

Yet few countries are likely to welcome a choice forced upon them. AI depends on chips, models, capital, talent, data centers, energy and markets. At a time when the technology is evolving rapidly, governments in the developing world and middle powers are more likely to seek access to technology and opportunities for growth than abandon one ecosystem for another.

Turning AI governance into another arena for geopolitical alignment would narrow those countries' choices and risk dividing the global AI ecosystem into competing camps. It could also raise the cost of innovation while making the already difficult task of managing AI's global risks even harder.

This is why the China-US dialogue on AI matters, even amid intensifying rivalry. The apparent contradiction is revealing. Washington may worry that China's technological progress will erode the US' lead. But both sides also face risks that neither can fully manage alone. Competition, in other words, is not the whole relationship.

The opportunity for cooperation is narrow, but it has not disappeared. The most realistic goal is not a sweeping framework for cooperation covering every aspect of AI. It is to identify a few issues on which both sides have a basic interest in reducing risks, and to build durable principles around them.

Strategic stability should be one such area. If AI competition increasingly enters the military sphere without reliable channels of communication, the risks of miscalculation, false alarms and escalation will grow. China and the US should therefore continue discussions on the responsible development and use of AI, including a clear commitment that decisions involving nuclear weapons must remain under human control.

A second opening may be found in a less politically charged domain: protecting children and adolescents in the age of AI. Generative AI is rapidly moving into education, social interaction, entertainment and emotional companionship. Young people may therefore become the first generation to experience its effects throughout their lives. Exposure to harmful content, deepfakes, AI-enabled fraud, privacy breaches and manipulative algorithms is already a concern shared by societies in both countries.

These issues lack much of the geopolitical symbolism attached to chips, models and industrial policy. That is precisely what makes them useful starting points.

The recent presidential engagement offered an opportunity to turn that possibility into something more concrete. If China and the US can reaffirm the importance of AI governance, establish sustained government-to-government dialogue and develop clearer understandings on responsible AI and strategic stability, they may create a few anchors strong enough to withstand political storms.

For China and the US, that may be the most realistic ambition: not ending competition, but ensuring that competition does not eliminate every space for cooperation.

Wang Dong is a tenured professor at the School of International Studies, a jointly appointed professor at the Institute for Artificial Intelligence, and executive director of the Institute for Global Cooperation and Understanding, Peking University. Zhang Xueyu is a research assistant at the Institute for Global Cooperation and Understanding at Peking University.

The views don't necessarily reflect those of China Daily.