The debate over DeepSeek puts on clear display how political nationalism risks stifling technological progress. While some push for strict AI decoupling, experts warn that restricting collaboration could undermine innovation. We must find a way to balance national security with the open exchange of ideas that drives scientific advancement.
Within days of DeepSeek’s startling global debut on January 20, 2025, one of the most vociferous, and predictably strident U.S. voices against China, Missouri Senator Josh Hawley, introduced the Decoupling America’s Artificial Intelligence Capabilities from China Act (S. 321).
In short, Senator Hawley wants the U.S. to have nothing to do with AI from China, even DeepSeek, even though it’s open-sourced, not seeking profit and can be used independently on American servers without fear of data mining by potential adversaries.
In throwing down the gauntlet against anything and everything Chinese, Hawley is cutting his own country to spite another.
U.S. tech is a dominant force in the world that got where it is on open exchange, shared knowledge and the ongoing give-and-take across borders between engineers and coders, industry and individual users of innovative products.
Lawfare, a Brookings Institution-affiliated publication that focuses on U.S. national security issues, raised the alarm about the U.S. needlessly closing itself off from the world. Taken aback by Hawley’s proposed AI decoupling bill, scholars Ritwik Gupta and Andrew Reddie put it this way:
“The bill’s rigid structure fails to take into account the globalized nature of software, hardware, and talent supply chains that benefit U.S. firms aiming to develop AI technologies. Ultimately, the DAAICCA will disrupt the open-source development of new tools and methods and create significant enforcement challenges. Taken together, the limitations of the bill, as currently drafted, will stifle American AI progress.”
Less than a month after DeepSeek’s debut, its parent company High-Flyer released Janus Pro, another innovative AI product that is licensed by the Massachusetts Institute of Technology for commercial purposes without restrictions. Described as “a novel autoregressive framework that unifies multimodal understanding and generation,” Janus Pro handles image recognition and generation with text-to-image instruction-following capabilities.
Sam Altman, the tech bro hero of Open AI who stunned the world with the beta-release of Chat GPT two years ago, has gone from a cool, breezy insouciant tech wizard to a thin-skinned tech bro crying about sour grapes, invoking nationalism while alleging, in a litany of mostly spurious complaints against DeepSeek, intellectual theft and dishonesty.
To be fair, the uncanny timing of DeepSeek’s release was awkward for Altman, for he had just successfully convinced the U.S. government that Open AI was not just the best at AI in the world, but the most trustworthy, with great fealty to the nation state. This made it the poster boy of Donald Trump’s $500,000,000 gala Stargate plan to bolster American AI with American dollars.
Then along comes Liang Wenfeng, a mild-mannered engineering nerd who dazzles the world with inexpensive, innovative AI on a par with ChatGPT. After reinvesting the profits of his High-Flyer quant fund in Hangzhou and tinkering for two years, Liang rattled cyberspace with the one-two punch of creating a product just as good as ChatGPT at a fraction of the cost.
Although the details of DeepSeek’s rise from nowhere remain sketchy due to Liang’s avoidance of the press and more recent information controls that now include a ban on DeepSeek researchers going abroad, DeepSeek was released free to the world and has been examined in depth by coders. It is widely recognized as a good product, all the more remarkable because it was, by all indications, not incubated and nurtured by the state but a self-funded venture.
Intense, idealistic, innovative and adept, Liang looks a lot like the backyard inventor of American yore, the self-reliant genius who tinkers for the pleasure of it and keeps hitting ‘em out of the park.
AI may or may not pan out to be the panacea that digital evangelists claim it will soon be, but Liang Wenfeng’s story has a pleasingly familiar ring to it, one that admirers of quirky tinkerers like Thomas Edison, Samuel Morse, Henry Ford and Steve Jobs can appreciate.
DeepSeek prides itself on the power of teamwork in a non-hierarchical office culture coupled with whiz-kid engineering, love of math and eye-popping innovation.
There’s a palpable whiff of idealism in the effort to date, reflected in Liang Wenfeng’ insistence that “China must gradually transition from being a beneficiary to a contributor, rather than continuing to ride on the coattails of others.”
DeepSeek’s neat feat continues to raise eyebrows, and cries, suggesting it was only possible because of a massive, secretive, sanctions-violating acquisition of Nvidia chips and massive state support, but there is little proof of that to date. It is true that many of the R&D costs were borne by Liang Wenfeng’s High-Flyer Quant fund, and he never claimed otherwise. DeepSeek’s own website acknowledges the six million figure reflects solely the cost of creating the app, but even that is a huge breakthrough, made possible by inspired coding and clever engineering.
For Liang Wenfeng to build so good an AI property for so little money, a sum of money less than the cost of Sam Altman’s flashy $7,000,000 sports car, was a shot across the bow of the tech bros of Silicon Valley.
Furthermore, the beauty of the open source model is that it can be copied and adapted by any player, rich or small, anywhere. It is a boon for developers in the U.S. and around the world who don’t have the dough of the Silicon Valley titans.
There exists the option of utilizing DeepSeek AI, with all its prowess, clever algorithms and economy, free of Chinese monitoring and influence by hosting it on servers outside of China, a cooptation exploited by Perplexity and Microsoft with evident success.
In today’s highly-charged political atmosphere, the rhetoric of techno-nationalism and AI supremacy tends to cast the AI race as a zero-sum game akin to a new Cold War.
The unexpected prowess of DeepSeek, acknowledged by professionals on both sides of the Pacific to be as good or better than comparative models on the market, has lessons for both the U.S. and China.
The U.S. needs to keep the door open to truly useful and mutually beneficial exchange with China. Protectionist policies, where necessary, should carefully differentiate between things that are hurtful to the US economy and things that are not. Just as the indiscriminate banning of Chinese students is stupid, cruel and hurtful to U.S. academia, banning good ideas from China is a bad idea. The knee jerk reaction to slam all doors shut on the part of reactionary politicians such as Senator Hawley is ill-tempered and counterproductive.
China, for its part, should be more open to sharing knowledge, science and technology with the U.S. since it has gained immensely from the generosity of spirit in U.S. exchanges to date. The authorities would also be wise to observe the age-old wisdom of Aesop’s tale about the Goose and the Golden Egg, letting citizens do what they do well without unwarranted state meddling and interference.
AI certainly needs guardrails, but nationalism should remain a subsidiary concern. Patents serve a purpose in regulating the course of innovation and sovereign nations need to protect intellectual property within reason, but ultimately all advances in technology and science, to paraphrase the great Isaac Newton, stand on the shoulders of giants. Science is not bound to man-made notions of nation and state nor should it be. Knowledge is not an island and never has been.
Science and engineering are universal languages enriched by the sharing of ideas, technology, factual findings and cross-fertilization. The forward course of human knowledge is impeded not just by the firewalls of nationalism and silos of isolation but in the stubborn refusal to acknowledge the good ideas and sometimes brilliant contributions of others.
