DeepSeek is being portrayed as a wakeup call, but it seems American policy makers and tech companies are hitting the snooze button. That may not be the right metaphor. They are alarmed, very much so. But instead of waking up, getting out of bed and questioning what they have been doing, they seem to be pulling the covers over their head and doubling down on the multiple policy failures that DeepSeek exposed. They really want to stay in bed and keep dreaming of a tech nationalism that gives the US unilateral control over a global digital economy. They are not learning.
DeepSeek is a powerful generative AI application. A good exposition of how it responds to tests can be viewed here.
How could such an advanced machine learning application come from China? The US government, during both Trump and Biden, adopted a policy of trying to retard or disable China’s digital development by means of export controls, restrictions on incoming capital investment, and attempts to kick Chinese firms out of our market. DeepSeek’s story was preceded by the expulsion of Huawei, the attempt to control semiconductors, then by bans on TikTok and drone maker DJI, and others. Although the US claimed that the Chinese firms were national security threats, the real underlying motivation was a belief that we could retard or cripple China’s digital development by cutting its companies off from advanced chips and blocking them from US markets. We insisted that Chinese tech companies could not possibly be commercial businesses bringing productive capabilities into the US market, they had to be Trojan Horses for the Chinese state.
That approach did put a temporary dent in Huawei’s growth, but it has recovered and is doing fine. Our attempt to ban TikTok was also not successful, as it became clear to many Americans that the threat was minimal and did not outweigh the benefits and preferences of the American public.
DeepSeek confronts us with the clearest, most unambiguous failure of this policy. Here is an advanced capability in a cutting edge area. DeepSeek seems to have leveraged a tradeoff between processing before a question is asked and the amount of time and power needed while answering a question. “DeepSeek’s shortcuts help it train AI at a fraction of the cost of competing models,” according to Lennart Heim of RAND. And while the application does contain Chinese censorship – it will not answer questions about Tiananmen Square and parrots the Communist Party line about Taiwan – the Trojan Horse claims are ridiculous in this case because it’s open source, and anyone can modify those constraints. Maybe, just maybe, this shows that China is a dynamic economy capable of bringing good products and innovations to the global ecosystem, and we need to start asking some fundamental questions about our approach to this problem. Maybe isolating ourselves from China tech is not the right approach.
Granted, the combination of a political dictatorship and a robust market economy is unusual for the 20th and 21st century (it was common in the 17th and 18th centuries). Yes, China’s CCP is dangerous and authoritarian to the people who must live under it. But where did we get the idea that the U.S. is in a position to alter China’s political regime and control its technological development? Why can’t the US benefit from trade with it in the meantime, just as it trades with authoritarian governments like Saudi Arabia, Israel, Vietnam?
Does it make sense to believe that the U.S. is in a position to control the technical development of an economy that is almost as big as its own and has the potential to be bigger? Is the suppression of global economic and technological development entailed by that strategy in the global public interest, or even in the interests of American citizens? How, exactly, do Americans benefit from shrinking or retarding the economic development of 17% of the world’s population? Do American businesses really benefit from being shut off from one of the world’s biggest markets? Are we retarding innovation and progress not just in China, but in the global economy as well?
But we are not having that conversation. Instead, we see the government and our state-subordinated tech entrepreneurs doubling down on its failures.
Here is Dmitry Alperovitch, a cybersecurity technology innovator who has convinced himself that he is an expert on geopolitical strategy. Quoted in the Wall Street Journal, he says, “DeepSeek couldn’t have trained its model without Nvidia chips acquired on the black market and less powerful chips that the U.S. permitted for export to China until 2023.” So instead of questioning the effectiveness of export controls, he wants us to seal off the industry even more.
Here is OpenAI, which is reverting to the oldest lament in the trade book, namely that the Chinese could not possibly have innovated but must have copied OpenAI. It says it is investigating whether DeepSeek trained its new chatbot by repeatedly querying the U.S. company’s AI models. In the words of the WSJ,
[this] raises the specter that companies spending hundreds of millions of dollars to train state-of-the-art models may have trouble keeping rivals from copying their work.
And how will we prevent learning and copying? Once again, we seem to be drifting towards more walls, more barriers, or perhaps shutting fast followers out of the the US market. The idea that it might be economically healthy for new competitors to find cheaper more accessible ways to do things does not seem to be entering the minds of Washington policymakers.
The worst response to DeepSeek is the one most predictable in Washington – the companies need to look to Washington for subsidies or support. President Trump, we are told, has to “confront China’s AI prowess.” What does this mean? Here is a letter from Alexandr Wang to President Trump, providing guidance on “how to win the AI war.”
This way of thinking about the problem is wrong. It’s not a war and it’s not a race with a winner. It’s an ongoing process of market development and technological evolution. There is no finish line. Insofar as there are “winners,” it will be companies who can maintain a profitable business in the face of intense competition for a long period of time. The markets will be global, not national. The competitors are not “countries,” they are businesses. The beneficiaries are not nations but consumers and users. OpenAI is not “America;” DeepSeek is not “China.” Except for the specific firms and employees, it is a matter of indifference to American citizens whether the best generative AI model or service is developed by a company headquartered in Silicon Valley or in Beijing, especially if the model is open source and can be used and modified by anyone, anywhere.
The post DeepSeek Disruption appeared first on Internet Governance Project.
Source: Internet Governance Forum