US, China meeting this week to talk AI safety, risks

The United States and the People’s Republic of China will meet in Geneva on Tuesday to discuss ways to manage societal risks around artificial intelligence, as the two superpowers mull whether and how to restrict the use of a technology that both consider essential to their national and economic futures.

The meetings were agreed to as part of an earlier meeting between U.S. President Joe Biden and Chinese President Xi Jinping last fall, and the discussions will focus on the risk and safety of AI-enabled systems. A senior Biden administration official told reporters in a briefing Friday that there are no formal deliverables or agreements expected, and that the meeting was an opportunity for “an exchange of views on the technical risks” presented by AI systems and to “directly communicate on prospective areas of concern.”

The administration has expressed concern that the Chinese government is rapidly deploying AI capabilities across civilian, military and national security sectors, “in many cases in ways that we believe undermines both U.S. and allied national security,” the official said.

The talks will not involve discussions of potential collaboration or information-sharing between the U.S. and China on AI research, the official said.

Domestically, the Biden administration has issued a sweeping executive order governing how the federal government utilizes large language models and other machine learning tools, while attempting to infuse best practices around the ethical and responsible use and deployment of AI systems in voluntary agreements with industry.

Earlier this month, the State Department unveiled a new international strategy for cyberspace, which puts an emphasis on building international coalitions to influence global digital policy.

“We are in a competition to shape the rules of the road, but also to explore if some of the rules can be embraced by all countries,” a second administration official said.

China too has made investment in AI a top national priority, though researchers inside and outside the country believe U.S. export restrictions on high-end computer chips may be stifling efforts by Beijing to keep pace with its western counterparts. Beijing has responded to these restrictions by amassing tens of billions of dollars that are targeted towards investments in their own semiconductor manufacturing industry.

The talks could begin a process for putting high-level guardrails around the most troubling uses of AI, such as its incorporation into the decision chain for launching nuclear weapons.

Earlier this month, Paul Dean, principal deputy assistant secretary for the State Department’s Bureau of Arms Control, Deterrence and Stability, called for China and Russia to match a U.S. commitment that decisions about whether to launch a nuclear weapon “would only be made by a human being.”

Asked whether the U.S. would press for China to make a similar commitment, the administration official said they expect to discuss “the full range” of risks related to AI but do not want to “pre-judge” China’s receptiveness this early in the process.

It’s not clear to what extent issues like the cybersecurity risks posed by AI systems or the hacking and intellectual property theft of AI technologies will be discussed between the two sides. An administration official said the two countries are holding separate talks on cybersecurity matters and declined to comment on whether the U.S. intended to convey any message surrounding the theft of domestic AI technologies.

Asked whether deepfakes — false audio, video, imagery or text generated by AI — and reports of Chinese interference in American elections would be broached, a third official said it was not on the agenda for this early round of meetings, but that if discussions are productive “we’ll certainly continue to express those concerns and warnings about activity in that space.”