U.S. President Trump is about to have a summit with visiting Chinese President Xi Jinping, and the two sides are expected to discuss a series of thorny issues ranging from trade to Taiwan. At present, one point that both parties may agree on is: in the field of AI, no one can afford to put on the brakes.
After researchers at Anthropic, an upstart American artificial intelligence (AI) company, suggested earlier this month that AI may destroy humanity before 2030, AI giants including company CEO Dario Amodei, OpenAI CEO Sam Altman, xAI founder Elon Musk, and Google DeepMind CEO Demis Hassabis have all spoken out, arguing that it is necessary to limit the speed of AI development.
Please read on...
Trump holds the opposite view, even calling the statement that "AI will take over the world" a hoax; Huida CEO Jen-Hsun Huang also believes that AI can be safely managed, advocates that new regulatory norms are not needed, and feels that "doomsday theories" have been over-hyped.
U.S. financial news network CNBC reported that before the Trump-Xi summit on the 24th, U.S. Treasury Secretary Scott Bessent discussed with Chinese Vice Premier He Lifeng last weekend the prospects of restarting the "U.S.-China Artificial Intelligence Dialogue," which included establishing a communication channel to handle matters "involving national security levels."
Aalok Mehta, director of the Wadhwani AI Center at the Center for Strategic and International Studies (CSIS), told CNBC that the most likely areas for the United States and China to reach consensus are to develop a common definition and architecture for AI security for advanced models and to establish an emergency communication mechanism to discuss emergencies.
Chris McGuire, senior fellow for China and emerging technologies at the Council on Foreign Relations (CFR), told CNBC that it is almost certain that the two sides will not reach an agreement to slow down the development of AI, but if a communication channel can be established for national security-related matters, it will be a positive step.
Despite various restrictions imposed by the United States, there is news that some Chinese companies can still obtain the computing power of high-end chips remotely through data centers in Southeast Asia.
Melanie Hart, senior director of the Atlantic Council's Global China Center, said that although U.S. Trade Representative Jamieson Greer said that chip controls were not on the agenda for last weekend's U.S.-China meeting, China may still push for loosening them.
"China is asking the United States to accept China's AI models and relax U.S. restrictions on China's access to chips," she said.
A Bloomberg News analysis believes that while Silicon Valley is still debating the risk of AI destroying humanity, Washington and Beijing seem to be more worried about global technological dominance, and neither country has shown any intention to slow down the development of AI.
Elly Rostoum, a senior researcher at the Science and Technology Policy Program at the Center for European Policy Analysis (CEPA), a non-profit think tank in Washington, and a former U.S. intelligence analyst, said that neither side wants to slow down development, but both the United States and China understand that the problem is serious enough to warrant dialogue, and relevant companies must also dialogue.
Analysis says that both the United States and China hope to continue to promote the development of AI, which means that the Trump-Xi meeting on the 24th is more likely to formulate clear-scope security protection measures on AI issues, rather than restricting the speed of AI development; at the same time, both sides are also tentatively seeking to build consensus on key risks.
In order to narrow the gap with the United States, China has regarded the opening of weighted AI as a strategy to catch up with the United States domestically and expand its influence overseas.
Emerging models such as DeepSeek's R1 and Moonshot's Kimi K3 can be downloaded and adjusted by developers around the world, allowing China's AI labs to learn from each other's results and further improve model performance with limited computing resources. These models have also won support overseas, helping to integrate Chinese software into various workflows and products, paving the way for future commercialization.
Z.ai assisted Malaysia in establishing sovereign AI infrastructure; Humain, supported by the Saudi Arabian government, entrusted Chinese startup MiniMax Group to use the latter's open model to build a national AI platform.
China's open weight model is popular because it combines advantages such as cost, flexibility, and performance. However, it threatens major U.S. AI companies that adopt closed-source models to charge customers for the use of proprietary models. This may lead to the United States further tightening export controls, thereby restricting China's access to the computing resources needed to maintain AI competitiveness. (Compiled by: Chen Yiwei) 1150924
