Published by https://www.protocol.com/ on November 3, 2022
By Kate Kaye
In a keynote speech at an event in September, national security adviser Jake Sullivan portrayed the U.S.’s mission to recruit more tech talent, including researchers from China, as a race the country must win.
“China is doubling down on its STEM talent production and its STEM talent attraction, but attracting and retaining the world’s best STEM talent is an advantage that is the United States’ to lose, and we are determined not to lose it,” Sullivan said at the event held by the Special Competitive Studies Project, a group funded by former Google CEO and AI tech investor Eric Schmidt to promote U.S. government investment in AI and other emerging tech research and development.
As people inside the U.S. government and tech industry have pushed to retain and grow the number of AI researchers arriving from China, AI and other categories of tech built in China have come under fire by the U.S. government, U.S. tech leaders, and human rights groups, fomenting distrust of Chinese scientists and their research output in general.
Scientific researchers from China studying and working in the U.S. are bearing the brunt of the suspicion. A survey by the Asian American Scholar Forum of roughly 1,300 Chinese American scientific researchers in the U.S. who are involved in computer science and engineering, math, and other sciences, conducted between December 2021 and March 2022, found that 72% did not feel safe as an academic researcher, 61% had thought about leaving the U.S., and 65% were worried about collaborations with China.
COVID-19 travel restrictions stalled internship recruitment from China at a U.S. university where one U.S.-based AI professor born in China teaches. Now the professor, who spoke to Protocol on condition of anonymity for fear of government retribution, said political pressures have all but ended recruitment of intern candidates from China at his university.
“I don’t think they’ll be coming back,” he said, noting that working with people from China has become “very sensitive” and is “a little bit” discouraged by his university. “We want to identify good students for Ph.D.s; that’s the only motivation for us. So if [the university authorities] don’t encourage — or they discourage — we’ll just say, ‘Fine,’ we just [won’t do that],” he said.
Part of the U.S. government’s AI strategy is contingent on fostering AI advancement through federally funded research grants and a would-be national AI research hub. However, some scientists of Chinese origin employed by U.S. universities who have used federal grant money to conduct research in the past are reluctant to apply again: 45% of the AASF study participants.
Additional analysis by AASF showed a steady yearly uptick in the number of U.S.-based Chinese engineering and computer scientists who dropped U.S. academic affiliations and switched to China affiliations, from 175 in 2017 to 298 in 2021.
To ensure U.S. economic and tech research dominance over China, Big Tech and AI investors have linked arms with national security hawks, hoping to woo more of China’s top computer scientists to the U.S.
But people entrenched in AI research warn that turning a mission to attract Chinese STEM scholars into a battle for talent against China could backfire by alienating those researchers. And they fear that profit goals and anti-China rhetoric have clouded an otherwise richly collaborative environment that has helped advance AI research in the U.S. and globally, and fostered goodwill among Chinese students who historically have been eager to study in the U.S.
“You can’t have all the smart people in the world — the United States can’t,” said Nathan Myhrvold, who helped start Microsoft’s influential research lab in Beijing in 1998. “It’s a little bit like why I started the lab in China,” he said.
The conference circuit
When researchers at Meta AI and Reality Labs Research published a robotics and computer vision research paper in July, they cited one of the four top papers presented at this year’s renowned Computer Vision and Pattern Recognition conference in New Orleans. That top paper, deemed by CVPR chairs as the “Best Student Paper,” focused on object pose estimation. It came out of China’s Alibaba Group and Tongji University’s School of Automotive Studies in Shanghai. Researchers from MIT and universities in Brazil and Belgium have also cited the paper.
It was just one example of the AI research cross-pollination among China, the U.S., and the rest of the world on display at the most prestigious AI conferences. A March 2022 report from Stanford University’s Institute for Human-Centered Artificial Intelligence showed the largest number of cross-country AI research and development collaborations between 2010 and 2021 were among people from the U.S. and China.
At CVPR this year, researchers in Beijing shared work on a sports video data set called FineDiving and on deep learning using data with “Noisy Labels.” Others from China’s AI software maker SenseTime, which was added to the U.S. government’s list of sanctioned entities in 2019, submitted work on image coding. Research on 3D plane detection by collaborators from Hangzhou’s Zhejiang University, Penn State University, Northeastern University, and Adobe Research reflected cross-border partnerships.
Those papers were peer-reviewed for acceptance at the conference by researchers in the U.S. and across the globe from Apple, Facebook, Google, Carnegie Mellon University, and Microsoft Research Asia in Beijing.
At the smaller MobiSys conference, an event held annually by the Association for Computing Machinery, reviewers insisted on blind reviews to prevent bias for or against research from particular people, schools, or locations. “We don’t actually know the point of origin as [papers] come in as the reviewing committee,” Robert LiKamWa, an associate professor at both the school of Electrical, Computer and Energy Engineering and School of Arts Media and Engineering at Arizona State University, told Protocol at the June event in Portland, Oregon.
You can’t have all the smart people in the world — the United States can’t.
Reports of Chinese researchers stealing ideas and intellectual property from the West have fueled generalized stereotypes against them, but the accepted papers from Chinese researchers at AI-related conferences defy those preconceived notions. “We’re seeing a lot of Chinese scholars publish really excellent works these days,” LiKamWa said.
This year, travel restrictions prevented researchers in China from traveling to the U.S. to present 10 papers accepted to MobiSys, including work involving on-device machine learning by researchers from Peking University, the Institute for AI Industry Research at Tsinghua University, and China’s State Key Laboratory of Networking and Switching Technology. Rather than being presented by the researchers in person, the papers were presented by proxies.
“I can only imagine how much work they put into these papers, but they weren’t able to travel here,” LiKamWa said.
Global engagement is integral to how competition functions in much of AI, from mobile tech research to biotech.
“You don’t want to wall yourself out from your competitors,” said Abigail Coplin, assistant professor in Vassar College’s sociology department and Program on Science, Technology, and Society who studies biotech research and development in China. “You want to engage with them, you want to know exactly what they’re doing, hire their postdocs, hire people who were previously involved in that company, so that you stay up to date and at the cutting edge.”
‘Both sides had some concerns’
After the Trump administration launched its China Initiative to combat tech espionage originating in China in 2018, an MIT Technology Review analysis showed the program’s investigations increasingly focused on “research integrity” violations rather than on theft of trade secrets.
In January, when the Justice Department dropped a high-profile case against an MIT nanotechnology professor because the case failed to meet the DOJ’s burden of proof at trial, U.S. attorney Rachael Rollins said, “We understand that our charging decisions deeply impact people’s lives.”
President Biden’s Justice Department removed the China Initiative label in February and broadened the program’s efforts to rein in foreign technology espionage and intellectual property theft by Russia, Iran, and other countries in addition to China.
Several sources Protocol spoke with for this story said the effects of the U.S. government’s continued crackdown on tech from China are felt by Chinese academics and the people in the U.S. who have worked with them.
“In the past two years, basically they’ve stopped all of these types of international collaborations, at least from what I know,” said an AI researcher from Beijing who spoke to Protocol via video call in October and asked not to be named in this story for fear of government retribution.
Political pressures shut down a potential collaboration he hoped to have with a former research groupmate in the U.S., where the Beijing-based researcher had attained his Ph.D.
When China’s ride-hailing giant DiDi wanted to fund research on AI to optimize fleet management, the company asked the Beijing researcher if he’d like to participate in the project. The researcher thought his groupmate in the U.S. would be a great partner. But faculty members at the U.S. university were hesitant to get involved because the research would be funded by a Chinese company, he said. The U.S. researcher was reluctant, too. And thousands of miles away, DiDi itself was loath to work with a researcher in the U.S.
“Both sides had some concerns,” said the Beijing-based researcher, explaining that in the end, neither he nor his U.S. colleague took on the work, a project they had considered within the last two years. Today DiDi is under investigation by the U.S. Defense Department, which is assessing whether the company is a danger to national security.
LiKamWa said geopolitical tensions between the U.S. and China are affecting his students.
“It is stressful. There are personal issues that we encounter because these students — we welcome them into our family. They become part of our family, and we want to see them do well. It becomes difficult [when their travel is limited]. It becomes difficult to have them employed,” said LiKamWa.
Being at a U.S. university, I do worry about our losing that lead; especially being in the field of AI in particular, I worry about that a lot.
“We understand that we are a piece of the geopolitical equation” and that “there are real, serious national security threats,” in the mobile technology arena related to sharing wireless spectrum, he said.
The China Initiative inspired new requirements for two U.S.-based AI professors Protocol interviewed for this story, who have been accustomed to collaborating with AI researchers in China with less oversight in the past.
An AI professor who is focused on machine learning and computer vision at a U.S. university said he and others at the school have helped colleagues in China — including people from Microsoft Research Asia — by reviewing their research papers. Now the professor, who asked not to be named for fear of political retribution, said those sorts of collaborations have dwindled compared to three years ago, and his university requires him to report research partnerships with people in China or other countries.
Suspicious AI
Usama Fayyad, executive director at the Institute for Experiential AI at Northeastern University, said he understood why some people have described the mission to attract AI researchers from China as a race the U.S. must win.
“Being at a U.S. university, I do worry about our losing that lead; especially being in the field of AI in particular, I worry about that a lot,” said Fayyad. Ensuring the U.S. has the most top-tier AI researchers creates a virtuous cycle, he said, noting, “The more you do, the better you get at it.”
However, Fayyad said he hoped talent leadership won’t come at the cost of less collaboration with China. “My hope, No. 1, is that academic collaborations continue to be as open as they have been, and that China also opens up more, and more academically, and doesn’t start restricting its researchers from talking about the advancements they’re making.”
Schmidt himself has been a vocal advocate for bringing more people working in STEM to the U.S.
“To sustain the United States’ technology leadership in the face of China’s formidable economic and military challenge, U.S. President Joe Biden should launch an urgent drive to recruit and retain 1 million tech superstars from around the world by the end of his first term in office,” he wrote in a Foreign Policy opinion piece in July that called on the Biden administration to eliminate a cap on immigration that he said blocks entry to skilled scientists and engineers from countries such as India and China.
Schmidt has invested heavily to facilitate, educate, and cultivate tech talent, including in relation to software engineering and development, which is a key focus of his efforts to convince the Pentagon and U.S. government to procure more AI software. His nonprofit Schmidt Futures has given grant funding to people and organizations including U.S. universities such as MIT, Ohio State University, and Penn State.
In July, the organization announced it would give $40 million to establish software engineering centers at the U.K.’s University of Cambridge, Seattle’s University of Washington, Baltimore’s Johns Hopkins University, and Atlanta’s Georgia Institute of Technology through a new Virtual Institute for Scientific Software initiative.
By extension, financial support from Schmidt Futures inspires people with influential government roles to endorse the organization’s goals. After the nonprofit gave Georgia Tech $11 million as part of the software program, the chair of the School of Computer Science at the school, Vivek Sarkar, suggested a possible collaboration between Schmidt Futures and the U.S. Department of Energy during a meeting of the DOE’s Advanced Scientific Computing Advisory Committee, of which Sarkar is a member.
The alienation problem
But as Schmidt and others aim to attract people from China and keep them here, suspicion of those very researchers has inspired legislation that would defeat those goals. The Secure Campus Act, proposed last year by Republicans in the House and Senate, would prohibit Chinese nationals from receiving visas for graduate or postgraduate studies in STEM fields in the U.S., and it would ban Chinese nationals or people involved in China’s foreign talent recruitment programs from participating in federally funded STEM research.
“The Chinese Communist Party often sends its members to elite American universities where they gain highly sensitive skill sets, then return to China and use this knowledge to help the Chinese government. Our legislation will help secure American innovation by curtailing Beijing’s influence operations and preventing the CCP from utilizing the U.S. education system against us,” said Sen. Marsha Blackburn, a Tennessee Republican who has been a vocal critic of tech-related research collaborations between the U.S. and China.
In May, Blackburn and other Senate Republicans proposed legislation that would prohibit U.S. research agencies, universities, and corporations receiving federal funding from conducting STEM research with “Chinese entities of concern in areas of cutting-edge technology” that could help China’s People’s Liberation Army.
But some basic AI research or work without military implications could get swept up in efforts to ensure the U.S. does not facilitate China’s military AI advancements. “AI research is not like some of the military types of research that can be directly transferred to some of the national defense applications,” the AI researcher in Beijing told Protocol.
The researcher suggested that claims of the Chinese government’s control over AI research and development in the country were exaggerated. It would be “impossible” for the government to fund the thousands of research papers submitted by Chinese researchers to global AI conferences each year, he said. And he said it would be difficult for the government to intervene to fuse commercial tech with military tech. “I don’t think the government has the time or has the people to actually supervise all these products. There are so many companies and so many products, it’s just impossible,” he said.
In the long run, anti-China tech sentiment inside the U.S. government could counteract its tech talent recruitment goals, and it could be difficult to walk the fine line between welcoming more tech talent from China in order to combat the Chinese government’s AI ambitions and risking alienation of researchers from the country.
“If we get nationalistic in a way that stops graduate students from coming here to get educated,” Myhrvold said, “it would be a real tragedy.” He added, “If in fact you have such onerous conditions for Chinese graduate students coming here, we will be intellectually the poorer for it.”
The AI professor interviewed at the CVPR conference told Protocol that further restrictions on collaboration between the U.S. and China will have a negative impact on all sorts of AI and technological research.
“There’s no doubt: Not even [just] for computer vision, but all scientific collaboration, it’s always the case. If we really shut them down or separate, that basically will harm both sides, for sure,” he said.
The AI professor in Beijing lamented the potential for more blockades on working with people in the U.S. “I think it’s just a pity because you have very brilliant people [on both sides],” he said. “It greatly accelerates the research.”
View online