The AI Export Dilemma

Until now, I have resisted writing much about AI. I confess to being a doomer about it and wish it would just go away. That, of course, is not going to happen, so I am finally going to take up one small part of the puzzle. There are a host of issues around AI: Is it going to kill all of us, or just some of us? Will it create mass unemployment? How do we ensure humans retain control? What, if any, safety guardrails are appropriate and feasible? What should we do about data centers? How will their demand for electric power and semiconductor chips affect the rest of the economy? And on and on. Those are important questions, and I leave them to the experts. Since this is a column about trade, I will discuss exporting AI technology: Should we allow it, and, if so, under what conditions or limitations?

Wading into that question reopens an old debate: Is it better to stringently protect the technology, both hardware and software, so it does not fall into the hands of our adversaries, primarily China? Or is it better to “flood the zone” and establish U.S. models and the chips that support them as the global standard? This is essentially an argument over whether a “safe” strategy or one that is high-risk and high-reward is more effective. This is not a new debate. It recurs with every new breakthrough technology development. The safe strategy tends to be favored by security hawks, particularly in the Pentagon, while the other finds a home in the business community and government agencies such as the Department of Commerce that are more responsive to corporate concerns. It would be nice if one of these had a clear record of success and was demonstrably more effective than the other. Sadly, in this case, both strategies have been employed, and both have had their successes and failures, which provides little guidance to government officials trying to decide what to do about AI.

The safe strategy requires extensive, detailed rules about which exports are permitted and which are not, competent licensing officials who understand the new and rapidly changing technology, and a strong enforcement program that can detect and constrain leakage. When it works, it enables the United States to maintain its technology leadership. Jet engine hot section technology is a good example of long-term success.

The risks include regulations that can be so complicated it is difficult for companies to understand what products apply, and, because of their complexity, inevitably leave loopholes that companies can exploit. The recent history of controls over semiconductor chips is a good example. Detailed regulations also impose a significant due diligence burden on companies to know their customers and make sure their exports are going to legitimate parties that will use them for their declared purpose. Large companies have sufficient resources to maintain strong compliance teams that do their best to make sure the rules are followed. That is much more difficult for smaller companies, particularly startups, which are dedicating most of their resources to product development and put compliance further down their list of priorities. While leakage is inevitable even in the best of circumstances, tight controls and effective enforcement can make it difficult for the adversary to scale.

The “flood the zone” strategy takes a different approach. It is based on the idea that the better way to maintain U.S technology leadership in a cutting-edge sector is to export and try to ensure that U.S. products become the global standard. Simply put, if everybody is buying American, then other countries will find it difficult to produce viable alternatives and will have little incentive to do so. Not always mentioned but also part of the strategy is that U.S. companies will make a lot of money, which they can then invest in next-generation development that will continue U.S. leadership. A classic example is Microsoft Windows, which the company successfully made a global product that is so widely used, it is difficult for others to compete. The security theory behind the strategy is that, if successful, it enables the United States to control the technology and its use and effectively isolate Chinese competition. The strategy does not mean no controls, but it permits more flexibility.

The main risk of this strategy is that rather than deterring China, it could instead encourage China to move faster to develop and market competing models while providing the tools to do it, because less restrained U.S. exports to other countries will inevitably fall into Chinese hands and make it easier for them to catch up. In other words, this strategy makes leakage more likely. In the case of AI, we have already seen this risk in the administration’s AI partnership deal with the United Arab Emirates, which has long had a reputation for being a place where exports find their way onwards to China or Russia.

My conclusion is that with respect to AI, where China is already actively in a contest for leadership, neither strategy is likely to be fully effective, and, while the United States should maintain controls, its best move is to run faster to stay ahead. Of course, that may end up killing all of us, in which case who won the leadership race won’t matter much.

William A. Reinsch is a senior adviser (non-resident) and Scholl Chair emeritus with the Economics Program and Scholl Chair at the Center for Strategic and International Studies in Washington, D.C. He can be reached at [email protected].