A Chinese security officer wears a protective mask at the end of the closing session of the National People’s Congress at the Great Hall of the People on May 28, 2020 in Beijing, China. (Photo by Kevin Frayer/Getty Images)

While there is an ongoing, lively debate about who’s ahead in the artificial intelligence race, it’s clear that both the US and China, in addition to militaries the world over, see the critical advantage the technology could provide in the event of a conflict. In the op-ed below, Georgetown researcher Ryan Fedasiuk explains what he and his colleagues discovered about China’s AI push through public records, and how it might reveal a couple strategic weaknesses.

For the first time on record, earlier this year an artificial intelligence system reportedly beat one of the People’s Liberation Army’s (PLA) top fighter pilots in a simulated dogfight. Chinese state media hailed the achievement as a watershed moment in the country’s military modernization. But almost as significant as the event itself was the fact that it came just months after the US military had achieved the same milestone.

For years, experts have written of China’s plan to wield AI for battlefield advantage, but cited US advantages in hardware and workforce development as enduring sources of US strength.

As tensions mount between the United States and China, and some experts warn of an impending crisis over Taiwan, US policymakers and defense planners should understand the kinds of AI systems already available to the Chinese military and take steps to defend the United States’ edge.

In a new report for Georgetown University’s Center for Security and Emerging Technology (CSET), my coauthors and I sifted through 21,000 equipment contracts published by PLA units and state-owned defense companies in 2020. Nearly 350 records in our dataset were related to AI systems and equipment, providing an extraordinarily detailed, entirely open-source view of China’s efforts to build an “intelligentized” force.

AI is the foundation of the PLA’s mission to become a “world-class” military capable of rivaling the United States. First unveiled in 2017, Beijing’s New Generation AI Development Plan established China’s goal to become “the world leader” in AI by 2030 — and it is clear that this objective extends to military affairs.

Indeed, many of the AI projects identified in our study are explicitly focused on degrading and countering systems at the heart of the US military’s Joint Warfighting Concept, using techniques like adaptive radar jamming and vulnerability fuzzing. Research papers and textbooks from China’s defense universities even discuss using machine learning systems to counter specific US drone swarm projects like Locust and Gremlins.

What’s more, the PLA is backing up its ambitious AI development goals with significant investment. Despite the several-hundred-billion-dollar difference in the topline budgets of the US and Chinese militaries, we estimate that both countries are investing about the same amount in AI for military use — in the low billions of dollars each year. Between April and December 2020, more than one in 20 public contracts awarded by the PLA’s main service branches were related to AI or “intelligent” equipment. These projects include all manner of autonomous vehicles, surveillance systems, training simulators, and battlefield decision support software.

In particular, the PLA is investing in AI capabilities meant to jam, blind, and hack the C4ISR systems that bind US assets together. Dozens of the Chinese military contracts in our study are for AI systems used in cognitive electronic warfare. Throughout 2020, PLA units and state-backed research institutions also awarded contracts for “microwave reconnaissance jamming drones” and “electromagnetic weapon” payloads that can be attached to swarms of small unmanned aerial vehicles (UAVs) and flown into enemy airspace. One textbook assigned to Chinese electronic warfare operators likens the US military’s battlefield information networks to the human nervous system: “Once the ‘tendons and veins’ or ‘blood vessels’ are cut off, people will be paralyzed or even killed.”

Moreover, the PLA Navy hopes that AI will compensate for its longstanding disadvantages in undersea warfare. Since 2015, Chinese research institutes have initiated hundreds of autonomous undersea vehicle research programs, and have made steady progress in foundational technologies like battery life and deep-sea communications. In a potential crisis over Taiwan, AI-based systems could extend the PLA Navy’s undersea reconnaissance operations well beyond the first island chain, and cheap, autonomous platforms could prove useful for mine-laying or anti-submarine warfare.

Despite the PLA’s significant progress in adopting AI-enabled systems, however, there are at least two clear vulnerabilities in its blueprint to build an “intelligentized” force.

First, while Chinese military leaders plan to exploit weaknesses in US sensor and communication networks, it is not clear how they plan to build resilient, cloud-based networks of their own. PLA officers often write that the US military is susceptible to information manipulation and data poisoning, even calling data integrity “the Achilles’ heel” of the US joint all-domain command and control strategy.

But in a potential conflict, the PLA itself would also likely struggle to ensure the integrity of data used to train its own AI systems — to say nothing of the inherent fragility of AI-based computer vision and object recognition systems. None of the 350 unclassified Chinese military contracts in our study focus on building resilient networks or secure datasets.

To mitigate the threat posed by Chinese military AI systems, US defense planners should boost investment in counter-autonomy and adversarial AI research that exploits Chinese system vulnerabilities, at the same time they shore up the robustness of US AI systems.

Second, China’s “intelligentization” strategy is entirely predicated on access to AI chips designed by US companies and manufactured in Taiwan and South Korea. The supply of these high-end microelectronics, however, is far from guaranteed. In fact, the United States and its allies have already adopted several measures to starve Chinese military companies of the chips required to train advanced machine learning models.

But to effectively slow Chinese military progress on AI, US policymakers should continue to scale up investment in the organizations meant to regulate technology outflow, like the Department of Commerce’s Office of Export Enforcement; and crack down on third-party intermediaries who supply the Chinese military and defense industry with US-made equipment.

While US leaders should not dismiss the Chinese military’s progress in AI, there are clear vulnerabilities in its blueprint for “intelligentized” warfare. By striking an appropriate balance between promoting innovation at home and preventing the leakage of US technology abroad, the United States can keep its edge in military AI.

Ryan Fedasiuk (@RyanFedasiuk) is a Research Analyst at Georgetown University’s Center for Security and Emerging Technology (@CSETGeorgetown), focused on military applications of AI, and China’s efforts to acquire foreign technology.