A recently published study from the Center for Security and Emerging Technology (CSET), a think-tank at Georgetown University, indicates commonly held perceptions about China’s supposed AI spending may be grossly off-base. Key takeaways suggest the PRC spends much less than the US on ‘military’ and ‘defense’ AI.
Here’s a hot take: All AI is military AI.
Let’s start with the study. You can read it here, but to briefly sum up, the authors write:
We assess with low to moderate confidence that China’s public investment in AI R&D was on the order of a few billion dollars in 2018. With higher confidence, we assess that China’s government is not investing tens of billions of dollars annually in AI R&D, as some have suggested.
The researchers at CSET had a near-impossible job set out before them. The public discourse on China’s spending is largely based on speculation. As you can imagine, the PRC doesn’t release information about its government spending.
CSET takes great pains – over dozens of paragraphs – to point out that there isn’t enough data available on China’s AI spending to come to any real conclusions. Here the researchers point out that these are estimates:
Given the pervasive uncertainties and assumptions in our analysis, we urge readers not to draw anything from these figures other than rough orders of magnitude.
However, they also point out that their study’s purpose was to research previous estimates and update with any new information, thus CSET’s guesses are far more educated:
We believe it is highly unlikely that China is investing tens of billions of dollars per year in AI R&D, as other sources suggest. Although our findings and assumptions are tentative, inferring tens of billions of dollars in annual R&D spending from publicly available data would require much more extreme assumptions.
So what can we take away from this knowledge? We thought China was spending tens of billions of dollars per a year on military AI, so we were scared that the PRC was going to become the supreme AI power and dominate the world by 2030. Now that we can better assume it isn’t wildly outspending the US on defense-oriented R&D, should we all sleep better at night?
No. What we can glean about China’s AI program from financial breadcrumbs and common sense won’t change the basic fact that there’s no such thing as non-military AI.
Does anyone actually believe that researchers like Andrew Ng and Ian Goodfellow developed deep learning techniques because they were driven by a lifelong passion to create an AI that could tell the difference between a cat and a dog? No, that’s preposterous. The big picture, whether it’s general AI or something else, has nothing to do with trivial problems.
The US calls everything even slightly related to the military “defense spending” because it’s pretty easy for a Republican-lead US government to get money for defense. The same tech that powers the CBP’s facial recognition systems also powers Amazon’s Ring cameras.
China can call it whatever it wants to – but the tech powering Alibaba’s fashion AI can be adapted to power missile targeting systems as easy as it can to predict whether your outfit is trendy or not.
Whether certain funds were earmarked for military, defense, education, or civilian R&D are irrellevent because artificial intelligence isn’t a product-based technology like hover tanks or railguns, it’s a science-based one like atomic bombs or electricity. Once we figure something out, the genie’s out of the bottle.
Is China spending less money on military AI than we previously assumed? It doesn’t matter. According to the CSET study, China is probably outspending the US on overall AI research (despite its GDP being roughly 40 percent lower). That seems like the more important takeaway.