U.S. Grapples With Potential Threats From Chinese AI

Biden administration officials have been preparing a new executive order for months that will restrict U.S. investment into some geopolitical rivals, namely China. Their goal is to prevent U.S. private equity and venture capital from contributing to China’s development of cutting-edge technology that could aid Beijing’s military.

Washington’s efforts to preserve American technological superiority over China, including by banning the export of some advanced semiconductors last year, has been a defining issue in its relationship with Beijing. Secretary of State Antony Blinken will travel to Beijing this weekend, where he will likely face complaints from Chinese officials who view moves like the investment restrictions as American attempts to hold back Chinese economic growth.

U.S. officials, meanwhile, cast the coming capital controls as a targeted step aimed at what they view as national-security threats. They have sought to craft the investment rules so that they don’t endanger broader investment and trade flows between the world’s two largest economies.

But distinguishing between technology that Beijing could use to advance its military and technology that Chinese companies use for everyday commercial purposes has proven difficult for the Biden administration, particularly when it comes to AI, according to people familiar with the deliberations.

The quandary is among the unresolved questions about the executive order, which is expected to prohibit investments in advanced semiconductors and quantum computing as part of a one-year pilot program, the people said. Investments in some forms of AI could be banned under the rules or simply subject to new disclosure requirements, the people said.

“AI is in many ways a meaningless category. It encompasses everything from Netflix recommendation algorithms to autonomous weapon systems and a bunch of stuff in between,” said Martin Chorzempa, who studies capital and technology controls at the Peterson Institute for International Economics. “It’s extremely hard to define.”

A spokesman for the National Security Council said the U.S. was making progress on the executive order, which investors and lobbyists have been expecting to be released for months.

“This is a complicated process that we want to make sure we get right, and that takes some time,” the spokesman said.

While some forms of AI are developed to accomplish specific functions, many AI companies focus on building general-purpose systems that can be trained to perform all sorts of tasks.

Global investment in AI startups is booming thanks to rapid consumer adoption of ChatGPT and other so-called generative AI tools, which can instantly create text, images, videos, and computer code based on written prompts.

The technology is expected to have wide-ranging commercial uses. But AI-models designed for computer coding could easily be used for hacking, while models intended to help create pharmaceutical drugs could also produce new chemical weapons, for example, researchers say.

“If you are using AI to generate imagery, that could be used to run a war simulation or it could be used for a game,” said Daniel Castro, a vice president at the Information Technology and Innovation Foundation think tank.

Other forms of AI can pose similar challenges. In 2016, a venture capital arm of Qualcomm, the U.S. mobile-phone chip maker, invested in 7Invensun, a Beijing-based startup that makes an AI-powered tool for tracking eye movement, according to Georgetown University researchers who wrote a recent report documenting U.S. venture investments in AI in China.

7Invensun’s technology could be used in virtual reality goggles or other consumer products, but it could also find its way into military or security applications, including infrared facial-recognition cameras or the training of fighter pilots. The company has disclosed working with a Chinese state-owned defense firm and China’s Air Force Aviation University, wrote the Georgetown researchers, Emily Weinstein and Ngor Luong.

Representatives of Qualcomm and 7Invensun didn’t respond to requests for comment.

Any policy banning investment such as Qualcomm’s would be challenging to articulate and enforce. Simply examining the code of a given AI system wouldn’t necessarily reveal its capabilities, since the systems can be rapidly trained on fresh data to accomplish new tasks. Advanced AI systems can also involve extraordinarily complex computing processes that yield unexpected results, making it hard to place limits on their output.

“These systems are just really hard to make reliably safe. They’re essentially black boxes,” said Tim Fist, a fellow at the Center for a New American Security.

The Georgetown report found that U.S. investors were involved in 401 transactions in Chinese AI companies between 2015 and 2021, with investments from exclusively American investors amounting to $7.45 billion in that time period.

But the specter of limitations on U.S. investment in advanced technology in China is already discouraging some U.S. firms from the market. Storied venture-capital firm Sequoia recently split off its China business amid the tensions between Washington and Beijing. Overall net foreign-direct investment in China will reach its lowest levels in almost 20 years in 2023, according to the Institute of International Finance.

A spokesman for the Chinese Embassy in Washington said the U.S. and China should pursue healthy economic competition.

“Discriminatory restrictions targeting companies of certain nationalities run counter to the basic principles of international economy and trade,” the spokesman said. “China will pay close attention to relevant developments and resolutely safeguard its own rights and interests.”

The Biden administration’s previous export ban on advanced semiconductors will be one way to try to prevent AI companies from accessing the computing power necessary to develop the most sophisticated models. The export ban has pushed some Chinese AI companies to try to develop advanced AI without cutting-edge chips. The U.S. has also previously banned the export of AI technology used to automate geospatial imagery.

Still, Biden administration officials are concerned that U.S. investors could transfer valuable knowledge and expertise to Chinese startups, allowing them to develop their own versions of the advanced technology, including semiconductors. U.S. venture-capital firms often give companies they invest in access to industry knowledge and contacts that may not be available elsewhere.

“There’s never going to be any investment in China that doesn’t pose any risk,” said Weinstein, who is a fellow at the Center for Security and Emerging Technology. “What the administration needs to decide is how much risk they can take on and draw the line there.”

Write to Andrew Duehren at [email protected] and Ryan Tracy at [email protected]

Biden administration officials have been preparing a new executive order for months that will restrict U.S. investment into some geopolitical rivals, namely China. Their goal is to prevent U.S. private equity and venture capital from contributing to China’s development of cutting-edge technology that could aid Beijing’s military.

Washington’s efforts to preserve American technological superiority over China, including by banning the export of some advanced semiconductors last year, has been a defining issue in its relationship with Beijing. Secretary of State Antony Blinken will travel to Beijing this weekend, where he will likely face complaints from Chinese officials who view moves like the investment restrictions as American attempts to hold back Chinese economic growth.

U.S. officials, meanwhile, cast the coming capital controls as a targeted step aimed at what they view as national-security threats. They have sought to craft the investment rules so that they don’t endanger broader investment and trade flows between the world’s two largest economies.

But distinguishing between technology that Beijing could use to advance its military and technology that Chinese companies use for everyday commercial purposes has proven difficult for the Biden administration, particularly when it comes to AI, according to people familiar with the deliberations.

The quandary is among the unresolved questions about the executive order, which is expected to prohibit investments in advanced semiconductors and quantum computing as part of a one-year pilot program, the people said. Investments in some forms of AI could be banned under the rules or simply subject to new disclosure requirements, the people said.

“AI is in many ways a meaningless category. It encompasses everything from Netflix recommendation algorithms to autonomous weapon systems and a bunch of stuff in between,” said Martin Chorzempa, who studies capital and technology controls at the Peterson Institute for International Economics. “It’s extremely hard to define.”

A spokesman for the National Security Council said the U.S. was making progress on the executive order, which investors and lobbyists have been expecting to be released for months.

“This is a complicated process that we want to make sure we get right, and that takes some time,” the spokesman said.

While some forms of AI are developed to accomplish specific functions, many AI companies focus on building general-purpose systems that can be trained to perform all sorts of tasks.

Global investment in AI startups is booming thanks to rapid consumer adoption of ChatGPT and other so-called generative AI tools, which can instantly create text, images, videos, and computer code based on written prompts.

The technology is expected to have wide-ranging commercial uses. But AI-models designed for computer coding could easily be used for hacking, while models intended to help create pharmaceutical drugs could also produce new chemical weapons, for example, researchers say.

“If you are using AI to generate imagery, that could be used to run a war simulation or it could be used for a game,” said Daniel Castro, a vice president at the Information Technology and Innovation Foundation think tank.

Other forms of AI can pose similar challenges. In 2016, a venture capital arm of Qualcomm, the U.S. mobile-phone chip maker, invested in 7Invensun, a Beijing-based startup that makes an AI-powered tool for tracking eye movement, according to Georgetown University researchers who wrote a recent report documenting U.S. venture investments in AI in China.

7Invensun’s technology could be used in virtual reality goggles or other consumer products, but it could also find its way into military or security applications, including infrared facial-recognition cameras or the training of fighter pilots. The company has disclosed working with a Chinese state-owned defense firm and China’s Air Force Aviation University, wrote the Georgetown researchers, Emily Weinstein and Ngor Luong.

Representatives of Qualcomm and 7Invensun didn’t respond to requests for comment.

Any policy banning investment such as Qualcomm’s would be challenging to articulate and enforce. Simply examining the code of a given AI system wouldn’t necessarily reveal its capabilities, since the systems can be rapidly trained on fresh data to accomplish new tasks. Advanced AI systems can also involve extraordinarily complex computing processes that yield unexpected results, making it hard to place limits on their output.

“These systems are just really hard to make reliably safe. They’re essentially black boxes,” said Tim Fist, a fellow at the Center for a New American Security.

The Georgetown report found that U.S. investors were involved in 401 transactions in Chinese AI companies between 2015 and 2021, with investments from exclusively American investors amounting to $7.45 billion in that time period.

But the specter of limitations on U.S. investment in advanced technology in China is already discouraging some U.S. firms from the market. Storied venture-capital firm Sequoia recently split off its China business amid the tensions between Washington and Beijing. Overall net foreign-direct investment in China will reach its lowest levels in almost 20 years in 2023, according to the Institute of International Finance.

A spokesman for the Chinese Embassy in Washington said the U.S. and China should pursue healthy economic competition.

“Discriminatory restrictions targeting companies of certain nationalities run counter to the basic principles of international economy and trade,” the spokesman said. “China will pay close attention to relevant developments and resolutely safeguard its own rights and interests.”

The Biden administration’s previous export ban on advanced semiconductors will be one way to try to prevent AI companies from accessing the computing power necessary to develop the most sophisticated models. The export ban has pushed some Chinese AI companies to try to develop advanced AI without cutting-edge chips. The U.S. has also previously banned the export of AI technology used to automate geospatial imagery.

Still, Biden administration officials are concerned that U.S. investors could transfer valuable knowledge and expertise to Chinese startups, allowing them to develop their own versions of the advanced technology, including semiconductors. U.S. venture-capital firms often give companies they invest in access to industry knowledge and contacts that may not be available elsewhere.

“There’s never going to be any investment in China that doesn’t pose any risk,” said Weinstein, who is a fellow at the Center for Security and Emerging Technology. “What the administration needs to decide is how much risk they can take on and draw the line there.”

Write to Andrew Duehren at [email protected] and Ryan Tracy at [email protected]