Donald Trump Wants to Choke Chinese Tech Growth. An Import Ban Could Do the Opposite.
Attempts to control how artificial intelligence develops and is used could backfire.
It is no secret that the Trump administration wants to curb China's growing global dominance. Imposing tariffs on Chinese imports has been a key element of Trump's economic agenda. But another important piece of the counter-China plan has gone less noticed: export controls on emerging technologies to Chinese markets.
Technology progresses when researchers and entrepreneurs are able to coordinate and learn from each other's work. But tensions can arise when governments believe that open inquiry will undermine national security. Few people want ISIS to get ahold of their own drones, for instance.
Governments therefore promulgate "export administration regulations" (EAR) to try and control what and how sensitive technological information can be shared. A multilateral treaty called the Wassenaar Arrangement harmonizes such efforts to prevent things like nuclear weapons from falling into the hands of hostile actors.
This is not to say that governments can't have ulterior motives in drafting EAR lists. In this case, certain "dual-use technologies" that have both peaceful and military applications are being targeted for export controls to curtail a geopolitical rival. Ironically, this strategy could ultimately redound to other countries' benefit.
Last August, the National Defense Authorization Act for Fiscal Year 2019 was signed into law. It contained the Export Control Reform Act of 2018 (ECRA), which granted the Bureau of Industry and Security (BIS) of the Commerce Department permanent authority to set export controls. (Previously, the BIS only had temporary authority through executive orders.) The ECRA directed BIS to issue a list of technologies that should fall under new export controls as well as "foreign persons and end-uses that are determined to be a threat to the national security and foreign policy of the United States."
BIS released their proposed list of technologies in a public notice of rulemaking last November. They are: biotechnology, artificial intelligence (A.I.), position navigation and timing (PNT) technology, microprocessors, advanced computing, data analytics, quantum information and sensing, logistics, 3D printing, robotics, brain-computer interfaces, advanced materials (e.g. functional textiles), and surveillance. The list may change by the time the final rules are issued, but the early draft provides a snapshot of what's at stake.
The BIS notice does not include a list of targeted countries. But we can expect that nations with strained US relations, like Iran and Russia, will get the axe.
China will undoubtedly be on the new technological export control list. Not only is the nation thought to have been involved in high-profile espionage campaigns against U.S. corporations and leadership, it is engaged in an industrial policy aimed at toppling American technology dominance, called Made in China 2025.
China's role in the future of A.I. is of particular concern. The People's Republic of China has poured billions of dollars into A.I. research. Such a king's ransom could lead to promising developments in the field at large, perhaps flavored by the particular political goals or cultural biases of the funders and builders.
This makes many in the West nervous. Our scientists and commentators are currently engaged in a difficult discussion about how to protect their most dear values. In particular, they are worried about A.I.'s capacity to further social bias or introduce uncontrollable weapons systems. They sometimes promote precautionary bans or limitations on A.I. technology to prevent such ills, although this approach could rob society of many benefits if too expansive. Adam Thierer, Raymond Russell, and I promote an alternative, called "permissionless innovation," which would allow us to keep "doomsday" scenarios like hostile A.I. in check while allowing the greatest possible space for development. (To the Trump administration's credit, they have shown willingness to embrace this approach so far.)
It is easy to sympathize with these concerns when it comes to the Chinese government. They have a less-than-stellar track record when it comes to Enlightenment values and technology, to put it mildly. Consider the "social credit" system, which is a massive surveillance campaign to track and punish disloyalty to the state. Perhaps more pointedly, the PRC has made great use of technology, including machine learning, to target its minority Uighur Muslim population for "re-education."
Then there's the mercantilist motivation. The fact that export controls would supplement the Commerce Department's ongoing trade war with China is surely no coincidence.
It appears that the Trump administration believes that export controls are a good path to control the threat of PRC A.I. domination. By limiting the ways in which U.S.-based scientists, and therefore foreign scientists collaborating with U.S. groups, can interact with their Chinese counterparts working on A.I., the Commerce Department hopes to clip China's wings. There is good reason to believe this won't work.
First of all, there is the problem with enforcement. Scientists chafe at any censorship of their work they view as unjust. They will subvert the efforts. The rather embarrassing debacle with the "Crypto Wars" is a classic example.
And not all technology hubs will follow suit. As a feature in the New York Times pointed out, the relevant factors of production are mobile. These export controls would limit beneficial scientific dialogue in addition to any shadier ventures. Companies and researchers frustrated by this virtual wall may just choose to move to a more hospitable locale, perhaps in Europe or Asia.
Thierer and Jennifer Huddleston Skees note that export controls on AI technology to China is just another classic example of what they call "global innovation arbitrage." Countries seeking to crack down on a technology rarely actually control production as they hope. Rather, capital and labor merely moves to a more permissive environment where it can continue less impeded.
Meanwhile, the goals the country hoped to advance—whether done in the name of "equality," or "hate speech," or "counter-terrorism," or whatever—are basically ignored everywhere else. Had domestic policymakers tried to work with technologists, rather than against them, they may have arrived at a policy environment that better addressed their concerns.
Could this be the case with export controls on A.I. technologies to China and other places? Perhaps Western values could be better promulgated through coordination. There are likely scientists in China that share our concerns. They may be better able to address them from within with the help of foreign scientists. And it may be easier to keep tabs on what the PRC is up to through regular communication.
What is clear is that U.S. export controls on emerging technologies will almost certainly fail to contain tech in the way our leaders want. Rather than freezing out China, export controls may ironically freeze us out as capital and labor move to less restrictive climates. This is far from a strong stance against potential Chinese malfeasance in technological development. It is a retreat.