Thread by @RyanFedasiuk

⚡️ We know the Chinese military is betting big on AI. But how exactly will it use the technology?

To find out, my team @CSETGeorgetown spent a year analyzing 66,000 of the People’s Liberation Army’s purchase records.

Here is what we found (1/17): cset.georgetown.edu/publication/ha…

2/ Like any large organization, the Chinese military buys a lot of stuff: Toilet paper, staplers, ball bearings...and weapons systems, like combat UAVs, submarines, and fighter jets.

A shocking amount of that info is public—if you know where to look.

3/ So Jennifer Melot, Ben Murphy, and I compiled every procurement record published by PLA units and state-owned defense companies from April-November 2020.

Our dataset includes 21,000 Chinese military equipment contracts, 350 of which were for AI-related systems and software.

4/ These contracts offer an incredibly detailed, open-source view of China's efforts to build an intelligentized (智能化) military force.

They amount to tens of millions of dollars in spending, mostly for autonomous vehicles, intelligence analysis, logistics, and info warfare.

5/ Specifically, we find the PLA is buying AI systems designed to identify undersea vehicles, wargame Taiwan operations, track U.S. Navy ships, and deploy electronic countermeasures, among other tasks.

“Harnessed Lightning” profiles each of these applications in depth.

6/ We can also get a sense of just how important AI is to the PLA.

Based on public contracts, PLA spending on AI-related tech is likely in the low billions of USD—about on par with 🇺🇸 military AI spending, based on this earlier report from @RitaKonaev.

cset.georgetown.edu/publication/u-…

7/ Moreover, we find that China’s military-civil fusion (军民融合) development strategy is paying real dividends.

Of the 273 AI equipment suppliers in our dataset, 60% are private companies. The overwhelming majority are quite small, established only in the last 10 yrs.

8/ And as @Kate_OKeeffe reported Monday for @WSJ, we find that < 10% of the PLA’s AI equipment suppliers face U.S. export controls. Almost none are subject to financial sanctions.

wsj.com/amp/articles/h…

9/ In fact, some companies that supply the Chinese military with AI systems are funded by U.S. venture capital companies like @sequoia and @GoldmanSachs.

For example, look to Eversec, whose clients include the PLA Strategic Support Force and Ministry of Public Security.

10/ Needless to say, the PLA’s rapid adoption of AI has big implications for the U.S. military and its partners in the Indo-Pacific, and our report details some of those challenges.

But we expect three points of tension to define 🇨🇳 military AI in the 2020s and beyond:

11/ FIRST, while it’s clear Chinese military leaders plan to jam, blind, and hack U.S. C4ISR systems in a conflict, it is not clear how the PLA would defend *its own* AI systems and "cloudified" sensor and communication networks.

➡️ Please fund more counter-AI research.

12/ SECOND, China’s “intelligentization” strategy is entirely predicated on access to AI chips designed by U.S. companies and manufactured in Taiwan and South Korea—whose continued supply is not guaranteed.

➡️ Expect more from us on this topic soon.

13/ FINALLY, there is the longer-term question of whether China will pursue lethal autonomous weapons.

Although Beijing decries AI weapon *deployment,* we do find that some Chinese research centers are actively developing AI-based target recognition & fire control systems.

14/ I hope this thread has persuaded you to read our report, which offers an unprecedented view of 🇨🇳’s adoption of military AI.

With 25 pages of appendices and endnotes, it has taken buckets of blood, sweat, and tears to create.

15/ As you read about the PLA’s meteoric AI gains, however, remember this: There are clear vulnerabilities in both the technology and Beijing’s plan to wield it for military use.

With an appropriate balance of tech promotion and protection, 🇺🇸 can prepare to meet the challenge.

16/ This report wouldn't have been possible without the help of a small army—28 people are listed in acknowledgments—but I am especially grateful to Twitterless coauthors Jennifer Melot and Ben Murphy; human and AI (🤖) research assistants Devon Colmer and "Elicit";

17/ mentors @RitaKonaev, @Andrew_Imbrie, @AnnaPuglisi838, and Igor Mikolic-Torreira; many talented folks on CSET's external affairs team, and reviewers @megan_lamberth, Ritwik Gupta, and Scott Harold.

Please direct media inquiries to: adrienne.thompson@georgetown (dot) edu

Likely of interest to mil/tech analysts @EBKania, @paul_scharre, @MartijnRasser, @kennethpayne01, @chrismeserole, @WriteArthur, @M__Verbruggen, @mchorowitz, @SamBendett, @RikeFranke, @MichaelJKanaan, @jessbrandt, @mattsheehan88, @klonkitchen, @Defense_AIO, @PartnershipAI

and China analysts @DaveShullman, @jmulvenon, @fravel, @nktpnd, @SheenaGreitens, @RollandNadege, @RushDoshi, @NinaPalmerCAP, @BrianTHart, @NathanMBM, @AndrewSErickson, @PLMattis, @Stokes2049, @He_Shumei, @mike_mazza, @casi_research, @ChinaMaritime

(@DefenseBaron @navybook: @DefenseOne readers might be interested in this—a draft op/ed is in your submissions inbox!)