US-China AI Decoupling Would Be 'Mistake,' Former DOD Official Says

The U.S. should work with China in select artificial intelligence areas instead of imposing sweeping export controls that create financial incentives for companies to “design-out” U.S. technology, Paul Scharre, vice president and director of studies at the Center for a New American Security, said in an April 18 opinion article for Time Magazine. While current U.S. restrictions on semiconductors exports to China are “narrowly targeted,” he said they will “de facto grow over time as chips advance and the threshold for export controls remains the same.”

Scharre, a former Defense Department official, argued for a “better strategy” of keeping Chinese commercial companies “dependent on foreign chips built using U.S. technology” while restricting chip exports that may be used for military applications or human rights abuses. “Keeping China dependent on U.S. technology would retain the kind of leverage the U.S. government used against Huawei, a powerful tool in the U.S.-China technology competition.”

Scharre said the U.S. is on a “steady path towards selective decoupling” of U.S. and China AI “ecosystems,” which he said would be a “mistake.” The U.S. “benefits disproportionately from talent flows and China’s dependence on American hardware,” and should “maintain these linkages even while severing others.”

“The U.S. is in a long-term technology competition with China, but pure decoupling is a weak strategy,” Scharre said. “The best strategy for staying ahead of China in AI includes working with China when the U.S. benefits more.”

Technology policy experts have urged the U.S. to allow research labs working on sensitive technologies, including AI, to continue operations in China despite new export controls limiting their activities (see 2301100056). Researchers with Georgetown University’s Center for Security and Emerging Technology have also said U.S. export controls on AI may not be the right strategy to hinder Chinese progress in certain AI subfields (see 2207270005). Others have said controls could spur Chinese AI researchers toward “subfields that are less computationally demanding” and lead them to develop “new competitive advantages” in those areas (see 2304130010).