Opinion | We Spent a Year Investigating What the Chinese Army Is Buying. Here’s What We Learned.

1

Last week, the U.S. Department of Defense released its annual report on Chinese military power, mentioning “artificial intelligence” 20 separate times. The report echoed longstanding concerns that the Chinese People’s Liberation Army is investing heavily in “intelligentized warfare” — a strategy based on making weapons systems and military operations more networked and autonomous — and that artificial intelligence may be “changing the future of warfare faster than expected.”

The so-called arms race for AI has come to define debates about the competition between the United States and China. The idea that the two nations are racing to dominate in AI — and, in particular, that China is surging ahead in this race — has garnered high-profile supporters as well as skeptics. But while much discussion, including the DoD report, has focused on China’s longer-term grand plans to become an AI superpower, it has been less clear what the country is doing in the short term to make those ambitions a reality.

Over the past year, I was part of a team of researchers at the Center for Security and Emerging Technology that sifted through 350 Chinese military equipment contracts related specifically to AI. The sample we analyzed is part of a larger, publicly available dataset of 66,000 procurement records published between April and November 2020.

Our just-published report, based entirely on open-source information, shows how China is already using artificial intelligence in its military strategy, as well as how it plans to acquire future AI capabilities. We found that the Chinese military is “intelligentizing” warfare by purchasing AI systems for all manner of applications, including autonomous vehicles, intelligence analysis, decision support, electronic warfare and cyber operations. At the same time, we found reason to be skeptical of the most ominous predictions about China’s efforts to fully automate warfare through “doomsday”-like weapons.

Perhaps most importantly for U.S. policymakers, our investigation into the PLA’s buying habits shows how Chinese progress in military AI is being driven, in part, by access to American technology and capital. Our report highlights the critical role U.S. companies play in supplying China with data, software and funding. This points to serious shortcomings in the U.S. export control system, which wasn’t built to screen the high volume of technology transfer and capital flows into China, and which struggles to distinguish between military and civilian purchasers. Even as the United States attempts to decouple supply chains from China when it comes to American goods, it also needs to consider new strategies to prevent American know-how from inadvertently powering China’s technological advancements.

Procurement contracts show what the Chinese military wants to buy, what it’s actually buying and from where. Some of the contracts we reviewed are mundane, documenting purchases of office supplies like toilet paper and staplers. More salient are the records showing that China is working to make its military more connected and autonomous in the immediate future. We found several instances in which PLA units were already buying commercial, off-the-shelf autonomous drones and AI-enabled surveillance software — in other words, ready-made AI tools that can be used right away.

Intelligent and autonomous vehicles, for example, accounted for more than a third of the contracts in our dataset. These include swarms of fixed-wing drones and rotorcraft, which could be used to saturate U.S. or Taiwanese air defenses in a conflict over the island. Several PLA units seemed to be purchasing unmanned vehicles through “the Drone Network” — an online marketplace similar to eBay, but for military equipment.

Notably, based on the public contracts in our dataset, China doesn’t appear focused on lethal autonomous weapons or automated nuclear launch, as some have suggested. In fact, of the seven areas canvassed in our report, command and control systems and automated targeting software were purchased the least often — and most of these systems seem designed to support human decision-making processes, not to replace them outright. Like the United States, China’s most promising AI applications so far seem to be for back-office tasks like intelligence analysis and predictive maintenance. Overall, while the U.S. military’s annual budget still eclipses China’s by $500 billion, we estimate that the countries spend roughly the same amount on military AI, likely more than $1.6 billion.

Our research also highlights that U.S. companies are inadvertently powering Chinese military advances in AI. The overwhelming majority of advanced computer chips at the heart of China’s military AI systems are designed by U.S. firms like Intel, NVIDIA and Xilinx, and manufactured in Taiwan. We found that suppliers actually depicted NVIDIA-branded processors in photos of their products, providing clear evidence of the role U.S. technology plays in powering China’s advances. One company, which won a contract to supply chips for the PLA Strategic Support Force, even bought the domain "nvidiagpu.com."

Some Chinese suppliers make an entire business out of sourcing foreign data or components and reselling them to sanctioned Chinese defense companies and army units. Others partner with U.S. tech firms on AI research or buy ship-tracking data from U.S. satellite companies. Notably, some companies that supply the Chinese military with AI-based battle management and cybersecurity software are financed by U.S.-based venture capital companies.

Although China is working diligently to make its military supply chain more independent of the U.S. and free itself from reliance on Taiwan for semiconductors, industry leaders say it will likely be years before Beijing can achieve its dream of self-sufficiency.

The interdependence between the United States and China has been well-documented, but our report provides one of the clearest illustrations yet of how loopholes and shortfalls in the export control system are enabling the outflow of U.S. technical know-how to the PLA. Among the 273 AI equipment suppliers identified in the study, less than 8 percent are found on the Commerce Department’s Entity List of individuals and organizations subject to U.S. government licensing requirements. The overwhelming majority of PLA suppliers are also not blacklisted by either the Treasury Department or the Defense Department.

To be fair, the U.S. export control system was never intended to deny resources to the world’s second-largest defense-industrial base. It was stood up to put selective pressure on rogue states and companies involved in nuclear weapons proliferation or human rights violations — with the promise of an off-ramp if they comply with international norms and regulations. This ad hoc approach to technology transfer is simply not built to handle America’s “number one pacing challenge,” a titanic economy deeply intertwined with the U.S. market.

As it stands, adding an entity to one of the U.S. government’s many blacklists requires months of planning, and any proposed addition must pass a rigorous interagency process. The misspelling or omission of any part of a company’s name can render the entire listing ineffective — assuming the designee does not just change it. Due diligence and accountability are important, but these inefficiencies are a problem.

Another problem for the U.S. export screening system is that the lines between military and civilian entities in China are becoming less clear. Many firms sell to the PLA but also to civilian companies. This is part of Chinese President Xi Jinping’s emphasis on “military-civil fusion,” or integrating the civilian and military technology industries. As part of military-civil fusion, private companies that contribute to China’s military modernization receive extensive research subsidies and tax breaks from the state. Indeed, our research indicates that private companies are the PLA’s main AI equipment suppliers, in contrast to the longstanding narrative that China’s defense industry is dominated by bloated, state-owned enterprises. More than 60 percent of the 273 contractors in our study are private companies.

The growing role of China’s private sector poses a major challenge for U.S. regulators. Within the Commerce Department’s Bureau of Industry and Security, just seven export control officers are tasked with reviewing every application to sell equipment to individuals and organizations on its Entity List. BIS already processes more than 30,000 export license applications each year, and adding thousands or tens of thousands more Chinese companies to its watch list would expand its responsibility by orders of magnitude. As two analysts recently pointed out in these pages, with a stagnant budget and fixed billets, the Commerce Department already struggles to keep up.

While it is still vital that U.S. regulators keep identifying and blacklisting specific entities, the sheer size of the challenge requires more adaptable approaches. One solution could be to screen transactions that deal with broad categories of technology, rather than investigate individual companies or people. But the U.S. government hasn’t even established those categories: Regulators have yet to release the list of “emerging and foundational technologies” that Congress mandated in the Export Control Reform Act of 2018. And to empower U.S. agencies to screen more transactions, Congress must boost funding.

As the old saying goes, “Show me your budget, and I’ll tell you your strategy.” With the help of publicly available documents, we’ve shed light not only on how much China is spending on military AI, but how it’s already preparing for the future of warfare. We hope that our work will help policymakers turn an often abstract debate about China’s pursuit of technological primacy into a real blueprint on how to address it.

View original post