Why Regulating AI is a Losing Battle

This op-ed originally appeared in Standard Examiner on August 30, 2024.

If artificial intelligence could be regulated with export bans, we’d have nothing to worry about — but that’s wishful thinking.

The recent reports of China circumventing U.S. restrictions on AI technology through smuggling networks and front companies highlights a fundamental flaw in our ability to ban physical exports. This only underscores the flaws in Department of Commerce regulations, which ban the export of AI enhancing hardware, and the ENFORCE ACT, trying to clamp down on the export of open-source AI knowledge sharing the same way.

AI-powered software can do many things, such as improving the quality of low resolution photos.

While well-intentioned, these types of export bans are ultimately ineffective. As the saying goes, “if there’s a will, there’s a way” — and when it comes to AI software, even prominent tech CEOs such as Meta founder Mark Zuckerberg agree that “our adversaries are great at espionage, and stealing models that fit on a thumb drive is relatively easy.” This reality exposes the futility of trying to control AI through export restrictions.

The New York Times recently detailed how Chinese companies are skirting U.S. bans on AI microchips, through the use of smuggling operations and front companies which facilitate the flow of restricted technology into the country.

The U.S. government may have thought that these bans would slow China’s AI advancement, but the reality on the ground is vastly different. Vendors in markets like Shenzhen can still get their hands on Nvidia’s most advanced chips. Meanwhile, Huawei is busy creating AI chips of its own, further demonstrating how China is rapidly catching up in the AI race.

But hardware is just the tip of the iceberg. What about software?

As Zuckerberg pointed out during the recent Llama 3.1 release, espionage is easy for adversaries, particularly when the target can so easily be transferred to a thumb drive. AI models, unlike hardware, can be copied and distributed at the speed of light. A few lines of code can be smuggled out of a country and replicated anywhere in the world with ease. No amount of customs checks or export bans, even those enhanced by the ENFORCE Act, can stop that.

Moreover, these regulations unduly harm U.S. tech companies.

The compliance costs alone — navigating complex export controls, monitoring distribution chains and ensuring no product ends up in the wrong hands — can be astronomical. These export controls are creating a “death spiral” for U.S. firms, reducing their competitiveness and revenue, which in turn hinders their ability to invest in R&D and capital — allowing China to close the gap even faster. For companies like Nvidia, Intel and AMD, the loss of revenue due to these restrictions is substantial. Billions of dollars in potential sales are lost, not to mention the competitive disadvantage created when companies are forced to pull back from lucrative markets. The truth is this: Export controls, especially on AI hardware like semiconductors, are hurting U.S. companies more than they’re hindering China’s AI ambitions. The same goes for AI software. These restrictions are boxing in American firms, leaving them with fewer options and bigger risks.

If the U.S. is serious about staying ahead in the AI race, it needs to rethink its approach.

Hardware bans may slow down some aspects of AI development, but they’re not long-term, and they’re ineffective. China’s workaround tactics show that their will to obtain and advance AI is stronger than any regulation the U.S. can impose. And with AI models being so easily transferable, export regulations on software are about as useful as a screen door on a submarine.

The focus should instead shift to safeguarding intellectual property, enhancing cybersecurity measures and fostering innovation domestically to maintain a competitive edge. By doubling down on innovation and creating an environment where American AI thrives, the U.S. can mitigate the risks posed by international competitors without relying on easily bypassed regulations.

In the end, the battle over AI isn’t just about who has the best chips; it’s about who has the best minds, the most advanced algorithms and the strongest strategies for protecting them. The sooner we realize that, the better prepared we’ll be to lead in the AI era. If we don’t, we might find ourselves outpaced not just by smuggled microchips, but by the very technologies we sought to control.