US Should Allow AI Labs to Continue Certain China Operations, Experts Say

The U.S. should allow research labs working on sensitive technologies, including artificial intelligence, to continue operations in China despite new export controls limiting their activities, technology policy experts said in a report this week. They also said the U.S. should create a new research security institution to help academia and industry work through “ethically or geopolitically difficult questions” on research security.

The report, written by the Brookings Institution and the Carnegie Endowment for International Peace, comes about three months after the Commerce Department published new export restrictions meant to prevent China from acquiring advanced computing chips and manufacturing advanced semiconductors, which could help it develop AI tools (see 2210070049). The new controls will “constrain the involvement of U.S. companies and individuals in transmitting this information or technology as well as the hiring of foreign researchers” in China, the report said, but there are still ways the labs can continue to operate legally.

“We recommend that U.S. research labs be allowed to continue operating in China,” the report said, “but that they operate within constraints meant to minimize the ethical and strategic concerns involved.”

Those constraints should include restrictions on U.S. AI labs forging “partnerships” with Chinese military institutions or recruiting from those entities, the authors said. The labs should also be allowed to conduct only research that’s “intended for open publication, thereby preventing knowledge transfer through proprietary or nonpublic work.” Although this likely won’t “stop research advances from contributing to China's surveillance state, it would limit any additional transfers due to that research being done in China,” the report said.

The authors noted some downsides to allowing U.S. researchers to continue to run labs in China, including the possibility the research could “contribute to oppressive technology deployment.” But the report also pointed to several benefits, including “access to foreign technical talent that powers American businesses.”

The “transition to fully open and online research publication has also meant that fundamental AI research can contribute to China's surveillance state regardless of whether it is done in Boston or Beijing,” the report said, arguing that today, “transfers of knowledge often flow in reverse, from the China-based lab back to the U.S. company and the global research ecosystem.”

The authors also make the case that improved U.S. research will aid U.S. innovation and allow it to out-compete China. Although there “was a time when the U.S. could have significantly slowed the growth of China's AI capabilities by eliminating these overseas labs,” the report said, “that time has likely passed.”

Recent export controls on semiconductors and chip technology "have laid bare the critical role of countries such as Japan and Korea,” the report said. The U.S. government has been able to “force foreign compliance through administrative measures,” including through the use of the foreign direct product rule, “but these mechanisms may be made moot if foreign manufacturers engineer U.S. technology out of their supply chain.”

The U.S. can also take steps to make sure U.S. researchers better protect sensitive technologies, including through the creation of a public-private research security institution, the report said. This institution would allow academia, industry, civil society and government to share information and best practices on “protecting research integrity from all threats, including those originating in” China. “It would provide a venue for many public and private sector actors to share information and engage with the U.S. government outside of the highly fraught circumstances of punitive law enforcement actions.”

The authors said the body could receive “initial support and ongoing participation” from government agencies but would be led by universities, companies and labs. “This will be key to assuaging some of the mutual suspicions that arise when government actors attempt to dictate terms or punish participants in the research community,” the report said.

The U.S. can also use a similar model on a multilateral basis, the authors said. “Each country could seek to create its own research security institution, one that accounts for the nuanced needs and constraints of their own domestic environment,” the report said, suggesting the Five Eyes partners -- Australia, Canada, New Zealand, the U.K. and the U.S. -- “could seek to create a transnational research security institution, one that leveraged the experience of researchers across several countries to create the most comprehensive information-sharing platform for making security- and ethics-informed research decisions.”

But the authors stressed “it will be crucial for governments to remember that their role is to support and contribute to -- but not control -- the decisions made by researchers.” The U.S. already has regulations “for punishing those who directly violate laws around intellectual property, sanctions, or export controls. To gain the needed buy-in from industry and the research community, the institution would need to be a cooperative endeavor in which all partners bring their unique knowledge and expertise to bear on shared issues of concern.”