AI slowdown debate opens new front ahead of US-China summit

Informal talks precede Thursday summit, but AI safety agreement remains elusive

China bristles at 'distillation' pressure; the two sides see AI risks differently

US: technological edge over China must hold even if development slows

China: safety issue being turned into zero-sum competition

US President Donald Trump (left) and Chinese President Xi Jinping arrive for a state dinner at the Great Hall of the People in Beijing on Monday (local time). [AP]

The debate over slowing AI development — born of fears that artificial intelligence could escape human control and threaten humanity's survival — has emerged as a new front in the US-China technology rivalry. President Donald Trump pushed back sharply, saying China would be "the only one cheering," while Beijing countered that claims of an AI safety threat lack any basis and accused Washington of pulling up the ladder behind it.

The Financial Times reported Thursday that while the United States and China both recognize the risks surrounding the pace of AI development, conflicting interests have prevented the two countries from finding common ground.

According to the Financial Times, Chinese think tanks and technology companies have been holding informal consultations with their American counterparts on managing AI risks ahead of the US-China summit on Thursday (local time). However, the paper said a shared sense of urgency — that neither side can afford to fall behind the other in AI capability — is making it difficult to establish common safety guardrails.

Amodei: China leading in AI would pose 'serious risk' to US

Anthropic CEO Dario Amodei participates in a panel discussion at the Vivatech technology startup and innovation fair at the Porte de Versailles exhibition center in Paris in May 2024. [AFP]

Xiao Qian, deputy director of the Institute for AI International Governance at Tsinghua University, who has taken part in the informal consultations with the American side, said Chinese AI safety experts want to work with the United States to manage the risks posed by frontier AI — but added that Anthropic CEO Dario Amodei's essay "gave the impression that the space for cooperation is far narrower than expected."

Earlier, on Sept. 12, Amodei said "China taking the lead in AI would pose a serious risk to the United States and the world," and called on the US government and companies to work together to impose restrictions on China.

China's Global Times criticized Amodei's remarks, saying "commercial interests underlie" the argument and that it risks "turning a global challenge that major AI powers should address together into a geopolitical zero-sum problem."

Against that backdrop, the prevailing view is that the chances of any meaningful agreement on AI emerging from the summit are slim. People familiar with internal US government deliberations said expectations are very low that Trump and Chinese President Xi Jinping will produce a substantive agreement on AI.

'Distillation': the technology at the heart of the US-China AI rivalry

The technique known as "distillation" — training one AI model using the outputs of a more powerful one — has also become a source of friction between the two countries.

The US government contends that Chinese companies have been covertly accessing advanced American AI models, harvesting their responses and reasoning processes in bulk, and using that data to train their own systems. China rejects this as baseless, arguing that such claims dismiss the independent technological progress its companies have achieved on their own.

Chinese experts said the very act of raising such allegations adds political weight to AI safety cooperation. Sun Chenghao, a researcher at Tsinghua University's Center for International Security and Strategy, said "attributing the advances of Chinese models to distillation from American ones simultaneously creates security concerns and political pressure around high-technology cooperation."

Diverging views on AI risk: US fears loss of control; China worries about sensitive information

China's five-star red flag (left) and the US Stars and Stripes flutter side by side. [AFP]

The two countries also view AI risks through fundamentally different lenses. In the United States, the central debate is whether AI could eventually escape human control and threaten humanity's survival — and how to prevent that. Chinese authorities, by contrast, are more focused on present-day threats such as hacking and data theft, as well as curbing the spread of politically sensitive information.

The Financial Times said that even within China's leading AI companies, discussion of "existential risks" such as human extinction remains limited. A person familiar with Chinese regulators' thinking said "China is still behind in AI, so introducing strong regulations now does not serve China's interests."

However, as the United States and China struggle to find common ground on AI regulation, there are signs that both sides are looking to focus on areas where agreement may be more within reach.

The Financial Times said options under discussion include creating a shared evaluation framework for assessing the safety of AI models, exchanging information on cases where AI has been used maliciously, and establishing formal communication channels through which the two governments could immediately alert each other if an advanced AI system behaves in unexpected or harmful ways.

Debriefing: The Korea Herald's international desk unpacks the stories behind the biggest global issues.

yckim6452@heraldcorp.com