I recently read ChinaTalk’s article, How Chinese AI Radicalizes. (though I preferred its previous title) I really enjoy reading their articles, but for this episode, I came away with a different perspective on why Chinese AI researchers don’t adore Aario. It’s a joint work with X.PIN’s CT Zhao, they write great articles covering Chinese tech and AI. Below are my thoughts.
China’s View of the US Is Realist, Not Conspiracy
The original article writes:
when discussing what Beijing casts as the ultimate motives of US actors. Beijing pushes the same grand-conspiracy narrative on many topics.
It reads more like applying an asymmetrical standard to the US and China. When talking about American AI leaders, it takes their thoughts seriously, carefully distinguishing among effective altruism, liberal internationalism, and national-security realism, and emphasizing that these people may sincerely believe their actions serve all of humanity. Yet when it comes to Chinese researchers drawing conclusions from US export controls, corporate policies, and national-security partnerships, the article reduces their views to conspiracy theories and the effects of the government “propaganda.”
As someone who once studied international relations theory in the US, I see the matter differently. The Chinese way of thinking about the forces underlying US behavior is basically realist. It would be an overstatement to call it a conspiracy.
In realist theory, great powers use technology, institutions, alliances, and, of course, rules to preserve their advantages, while the language of security inevitably serves a certain power structure. As the author points out, deeply rooted convictions and internal disagreements lie behind American doctrines. But this diversity of thought cannot obscure the fact that, when the US administrative state exercises its power, competition and containment remain its most fundamental instruments.
And, when the US government and American companies define AI as a strategic technology that will determine national power and the outcome of competition between political systems, I would say it is entirely natural and fair for Chinese researchers to interpret American AI-safety discourse through interstate competition.
Interpreting the views of AI company founders as the result of genuine ideas and political convictions, while reducing Chinese researchers’ views of their American counterparts to conspiracy theories or propaganda, underestimates the intellectual agency of Chinese researchers. It also carries a somewhat condescending undertone.
China’s AI community may be the segment of Chinese society with the closest ties to its American counterpart. Many of its researchers have studied in the US or worked for American AI companies. Even more have received American-style academic training, read frontier papers without difficulty, and collaborated with American colleagues.
The original article argues that Chinese media propaganda has led researchers to adopt views of the US that are even more radical than the Party’s. I would argue the opposite: that these researchers have reached their own conclusions precisely because they have observed the American technology industry firsthand and personally experienced changes in US policy in recent years.
Sincerity Does Not Change the Nature of a Policy
Of course, as the original article says, “no trillion-dollar business builds itself without leveraging the pragmatics of capitalism and quotidian stakeholder interests.” Nor do I wish to pass judgment on whether Dario Amodei’s ideological convictions are sincere. That is the business of the Spanish Inquisition. I would rather return to the policies he has publicly advocated.
Put simply, Dario believes that the US should:
Do everything possible to restrict China’s access to AI chips, models, and talent, thereby eliminating the threat posed by China’s development of uncontrollable AI models.
Communicate with official Chinese channels in order to establish a shared U.S.–China regulatory framework in which he would participate.
He may sincerely believe that he represents universal interests while promoting an exclusionary set of policies that plainly benefit his own country’s companies and national power.
But to Chinese researchers on the receiving end, these restrictions are insulting. Rather than asking them to distinguish between sincere and insincere motives, I would invoke the American saying Mike Wallace quoted in his 1990s interview with a former Chinese leader: “If it walks like a duck and quacks like a duck, it probably is a duck.”
From the Chinese perspective, Dario’s two positions contradict each other. He characterizes China as an authoritarian state and claims that it is developing uncontrollable AI. In doing so, he strikes one of the most sensitive nerves among the Chinese public: he affronts the country’s narrative of national rejuvenation.
We should recall that the legitimacy of the 新中国 (the New China), or I’d prefer, the People’s Republic. It was founded in part on overthrowing the “three great mountains,”三座大山, foremost among them imperialist interference in China’s affairs. When an American technology leader publicly advocates restricting China’s access to critical technologies while also criticizing its political system, Chinese audiences will naturally interpret his safety agenda as an extension of an old power relationship into a new technological domain.
Chinese society, meanwhile, places great value on using tech innovation to make the economy more efficient and thereby improve human well-being. In other words, Chinese people don’t need to be guided toward the conclusion that Dario’s position is intended to restrict their right to development. They will arrive at it quite naturally.
This introduces another problem. As I understand, criticism and negotiation can coexist. When there are shared interests, even adversaries that are deeply hostile toward each other can communicate. But if the party initiating dialogue unilaterally denies the other side’s political legitimacy while simultaneously demanding that it accept a regulatory agenda under its own leadership, the failure of negotiations is entirely foreseeable.
Consider the establishment of diplomatic relations between China and the US. As the Sino-Soviet divorce occurred, all three sides of the China-Soviet-US triangle had concrete realist reasons for cooperation. Nevertheless, nearly eight years passed between Henry Kissinger’s 1971 visit to China and the formal establishment of diplomatic relations in 1979. I seriously doubt that Dario can expect the Chinese to accept his agenda within a short period of time.
Of course, I do not blame them. In China, our exposure to rationalism comes from Descartes, Rousseau, Montesquieu, and the Enlightenment, not from a work of Harry Potter fan fiction. Given their different philosophical traditions, China and the United States will inevitably approach AI differently. The problem is that China has never been the one to initiate the confrontation.
I am not a believer in any religion, but allow me to borrow from Matthew 10:34:
“Think not that I am come to send peace on earth: I came not to send peace, but a sword.”
Cooperation Cannot Be Built on Imagined Goodwill
U.S.–China AI dialogue certainly requires some degree of goodwill. But it would be naive to expect cooperation to rest on believing each other’s moral self-descriptions. China will not believe that U.S. export controls are motivated by concern for humanity's safety, just as the U.S. does not believe that China’s commitment to open-source AI is entirely motivated by a desire to democratize technology. Yet none of this prevents the two sides from communicating on concrete issues.
If they truly want results from AI cooperation, they need to limit the domain. They need to acknowledge that the other has security concerns as well as clear competitive interests. Requiring one side to accept the other's ideological narrative won’t help at all. Instead, they should establish verifiable, reciprocal rules concerning specific risks, such as AI-enabled cyberattacks, military miscalculation, and loss-of-control incidents.
Rather than relying on goodwill, enforceable mechanisms are the foundation of cooperation. I believe so, but decision makers do not need to believe that the other side consists of “good people” to cooperate in specific areas. If anything, only after stripping away imagined goodwill and comforting illusions can the two sides clearly perceive each other’s actual capabilities, interests, and red lines—and build institutions capable of functioning even in the absence of mutual trust.
