Save the date: Tomorrow—Saturday November 14—at noon US Eastern Time I’ll host a zoom call about AI with paid NonZero subscribers. I’m eager to hear about your AI anxieties and enthusiasms and your ideas about how the newsletter and/or podcast and/or future NonZero projects might contribute to humankind’s successful navigation of the AI revolution. You’ll find the link to the Zoom call at the bottom of this newsletter.
To get you in the mood for the call, here’s a highlight reel featuring recent NonZero podcast guests talking about AI. Below the video you’ll find my quasi-defense of conspiracy theories about the fall of NVIDIA’s Jensen Huang.
“Man the deep state is doing a full coup of Jensen,” tweeted “Beff Jezos,” leader of the “effective accelerationism” movement, to his 300,000 followers as this week’s Jensen Huang scandal approached its climax. After the climax—with Huang now forced out of his role as CEO of NVIDIA—Prakash Narayanan, a tech entrepreneur and podcaster, tweeted that Huang’s ousting was “obviously a warning shot to other tech CEOs: toe the deep state line or we will get you.”
Narayanan is prone to starkly stated opinions that sometimes qualify as outliers, and Beff Jezos has said a number of things that strike me as flat-out crazy. But I do think there’s a version of the deep state conspiracy theory about the Huang scandal that is plausible, so long as you define “deep state” and “conspiracy” carefully. And, in any event, here’s one moral of the story that I think is clear: China hawks in the national security establishment are pushing their agenda fiercely and effectively, and that doesn’t bode well for the US-China cooperation that I think will be critical to human welfare as the AI revolution unfolds.
Two weeks ago the New York Times reported that Alice Huang, an executive at NVIDIA customer Megaspeed, was a “Chinese intelligence asset” and had been in a “close relationship” with Jensen Huang, according to the Times’s anonymous sources—“people briefed on the surveillance” of her.
“Intelligence asset” is a broad term. Trained spies who work for the CIA or China’s MSS (Ministry of State Security) are intelligence assets, but so are other kinds of people. If you’re an American entrepreneur and after you attend a Shanghai tech conference you have coffee with someone in America’s intelligence bureaucracy and they ask you what you learned there and then encourage you to share observations after future trips, you are an “intelligence asset.” But you’re not a “spy” in the common sense of the term—you’re not a trained agent or even an amateur who was sent on an intelligence-gathering mission by an intelligence agency.
Alice Huang (no relation to Jensen Huang except in the sense of being a paramour) wasn’t called a spy anywhere in the Times piece—just “an intelligence asset.” Nonetheless, the Times headline said, “NVIDIA Chief is Target of Chinese Spy Plot.” And thereafter, US media and social media were replete with references to Alice Huang as a “honeypot”—as if we knew she had been deployed by China’s MSS to lure Jensen Huang into bed.
A subsequent Wall Street Journal piece about Alice’s career, while not quite contradicting the “honeypot” scenario, painted a very different picture. She and her serial entrepreneur husband had worked their way to a point where they could start a cloud computing company in Singapore (Megaspeed) that put advanced American microchips in data centers located outside of China and rented the processing power to Chinese companies (which is legal, whereas Megaspeed’s alleged sale of some of the chips to Chinese companies violated US export restrictions). The couple landed a commitment from Alibaba to rent lots of processing power, and this commitment cemented their business relationship to Jensen Huang (whom Alice’s husband had long known from the video game business, where NVIDIA had gotten its start). Alice’s charms seem to have further cemented the relationship with Jensen—and apparently her husband knew about the Alice-Jensen affair and was OK with it.
So in the Journal’s telling, Alice Huang’s commercial ambitions, not China’s intelligence service, led her to Jensen Huang. “Nvidia’s AI chips were the latest play for a couple with a track record of riding hot trades and, on occasion, running afoul of regulators… US officials believe that China’s intelligence services recruited Alice Huang as an asset because of her access to the chip-designer chief.”
A subsequent CNN story cast doubt even on the part about her recruitment as an intelligence asset: “Multiple sources” had told CNN “that the intelligence collected on her to date does not prove that she was acting as a spy for Beijing. There is disagreement among US officials about the strength of the evidence…”
Now back to the New York Times story and its anonymous sources:
The “people briefed on surveillance” were presumably in the government. As a rule, people in the government leak things for a reason—especially things this explosive. Also as a rule, the likely reasons for the leak can be found in its most predictable consequences.
In this case one very predictable consequence is to increase the pressure on politicians to do things like (1) tighten the enforcement of the chip export controls that Megaspeed was violating if it indeed was selling NVIDIA chips to Chinese companies (which it very likely was); and (2) add new export controls—like banning the part of Megaspeed’s business model that was legal: putting the chips in those data centers in Malaysia and elsewhere and renting their use to companies inside China. Such a ban has long been advocated by, for example, Anthropic CEO Dario Amodei, who two years ago co-authored a Wall Street Journal op-ed advocating it with uber-China-hawk Matt Pottinger, who served in the first Trump administration.
In addition to giving momentum to these particular policies, the Times story put pressure on President Trump to be more generally hawkish toward China. The Times’s sources disclosed that “roughly a year ago, Mr. Trump was briefed about the intelligence findings… and chose not to do anything with the information.” Being cast as a panda hugger will raise the political cost of embracing Xi Jinping when Trump meets Xi at next week’s Asian-Pacific Economic Cooperation summit in China.
Trump’s failure to at least inform Jensen Huang of the suspicions about his paramour does sound negligent, and I can see why a government official with no particular China policy ax to grind might be so outraged as to air this information. But when the information is aired via multiple anonymous sources shortly before a rare Trump-Xi meeting, you have to wonder whether there isn’t another goal.
So was this a “conspiracy”? Well, if multiple government sources had a shared goal and coordinated their actions behind closed doors, then, by definition, yes. But even if that wasn’t the case—even if these sources acted independently but were motivated partly by their aligned world views—it could still be the “deep state” at work, depending on how you define the deep state.
One defensible characterization of the “deep state” is: the part of the government that serves the national security establishment consensus even when the nominal leader of the executive branch defies that consensus. That establishment is currently pretty China hawkish—and is very disapproving of Jensen Huang’s longstanding advocacy of loosening chip export controls, something Trump has done in a limited way and seems open to doing in a bigger way if China were to offer something in return on the economic or diplomatic fronts. But the national security establishment need no longer worry about Trump doing it in a bigger way next week when he meets Xi Jinping.
More broadly, the national security establishment has long been committed to a democracy-versus-autocracy framing of geopolitics that isn’t compatible with Trump’s value-neutral realpolitik.
At any rate, the fallout from this scandal decreases the chances of a near-term warming of US-China relations and increases the chances of a sustained cooling. (New chip export rules could well provoke China to tighten, for example, rare-earth-mineral export rules.) So this scandal isn’t good news for people like me, who think that without extensive US-China cooperation the planet is very unlikely to get through the AI revolution in good shape.
And it’s not good news for people (also like me) who worry that a breakneck race to artificial superintelligence could lead to war with China. Here’s how I describe that scenario in my book on AI, The God Test:
Various signs point to accelerating AI progress, among them the exponential improvement in AI performance, on various kinds of tasks, that I noted at the end of chapter 9. It’s not crazy to think that, if the curve of progress keeps getting steeper, at some point being only a few months ahead of a rival could make a world of difference. You don’t have to subscribe to a strict version of the superintelligence takeoff scenario to believe that a sudden and potentially explosive imbalance of technological power between superpowers is plausible.
What’s more, the very plausibility of such a thing—the fact that national leaders could take this prospect seriously—might itself prove explosive. If there are two AI superpowers that are on adversarial terms, and one of them fears that the other is nearing a stage that will bring utter dominance, drastic measures might seem in order. Like, for example, sabotaging the rival superpower’s AI program, maybe via cyberattack, or maybe via, as they say in national security circles, “kinetic action”—like bombing data centers being used to train AIs or factories being used to make advanced AI chips. And bombing has been known to start wars.
By the way, some definitions of the “deep state” are broad enough to include people outside of government—such as people in the private sector who are intertwined with the Pentagon and US intelligence agencies through contracts and less formal symbiotic relationships. By this definition, Dario Amodei—who emphatically buys a “democracy versus autocracy” framing of geopolitics—is a member of the Deep State. So I’ll close with this summary of Amodei’s proposed strategy toward China, also drawn from my book:
Which leads us to the one major part of Amodei’s “Machines of Loving Grace” paper that I haven’t talked about: a section called “Peace and Governance.”
Here Amodei ponders the prospect of the most powerful AIs being in the hands of an “authoritarian bloc”—whose most formidable member, needless to say, would be China. To preempt this development, he says, the US should lead a coalition of democracies that would restrict the authoritarians’ access to key technologies and then would “use AI to achieve robust military superiority,” after which these democracies could “parlay their AI superiority into a durable advantage.” This could “lead to an ‘eternal 1991’—a world where democracies have the upper hand.” Then, Amodei says, the democratic coalition could use this stick to get these repressive autocracies to “give up competing with democracies” unless they want to “fight a superior foe.” And, for added incentive, they could be offered the carrot of access to advanced AI as a reward for finally seeing the light.
Amodei acknowledged that “this will be very difficult to achieve.” Still, he wasn’t so daunted by the challenge that he mentioned the possibility of avoiding it—like, say, by the United States talking things over with China in hopes of working out agreements about AI development that would leave both countries feeling secure.
You can read the introductory chapter of my book, and excerpts from all other chapters, at thegodtest.net.
Banners and graphics by Clark McGillis.
And here’s that Nov 14 Zoom link: