Whatever your opinion is on artificial intelligence, there’s a good chance that you’ve heard of the recent release of Deepseek AI. The system, released by Chinese hedge fund High-Flyer, recently made waves in the tech industry as a cheaper and more efficient large language model than many competing models, including OpenAI. LLMs operate by analyzing massive amounts of data and using it to predict sentence structures, usually based on topical prompts. For a while, AI development was comfortably a U.S.-dominated business with companies like OpenAI, Google and Meta being the top names in the field. Maybe it was the United States’ comfort in being number one that made it so shocking for a Chinese company to find success.
Developed amid a trade war between the U.S. and China, Deepseek had limited access to Nvidia computer chips. Partially because of this, Deepseek’s cost of development was much lower than its U.S. competitors, at $6 million contrasting the several billion paid by U.S. companies. Compared to OpenAI, it is also more effective at solving complex problems. Additionally, the software is open-weight. This means that while the training data itself is not public like an open source software would be, Deepseek allows free access to the parameters, or “weights,” it uses to function. This is a huge step for accessibility of AI research in particular, as many top AI systems are privatized.
Despite this being a huge in innovation in LLMs, you wouldn’t know it based on the media coverage from here in the U.S. In fact, it took me digging through several pages of fearmongering to find any substantial information on how Deepseek works to write this article. If you Google “Deepseek” and look through the news category, you’ll see pages upon pages of coverage on “concerns over security, censorship and dependence on China” and “DeepSeek – A Wake-Up Call For US Higher Education” — lines clearly written out of reigniting Cold War-esque posturing, or at least hoping to farm some clicks off the sentiment. An article by WIRED titled “Chinese Companies Rush to Put DeepSeek in Everything” features an eye-catching tagline on stock prices and nationalism, making it out to be anything but normal behavior, as if the U.S. doesn’t use our own AI for the exact same things.
Look, I get it. First, it was nuclear arms, then space, then the moon — now it’s AI. Beyond claims of national safety, legitimate or otherwise, technology and innovation have always been a point of pride and show of power on the international stage. We want to win, and with the amount of money and effort the U.S. is funneling into artificial intelligence, it’s frustrating for many to still not be consistently on top. It’s comical, though, to pretend that we’re somehow the only ones allowed to see success in technology. After all, China has one of the biggest economies in the world, a huge population with many experienced and intelligent developers and it is not currently being hostile to citizens pursuing careers in STEM. To put it bluntly, the way that Deepseek is discussed within tech news and many parts of the tech community is propaganda — fearmongering at best and flat-out xenophobia at worst. Very common technology features, like text prediction, are now being featured as some Chinese data-harvesting conspiracy just because it’s being done outside the Western world.
A similar phenomenon can be seen on TikTok; a few months ago, the U.S. banned TikTok after giving the company an ultimatum of handing control over to the United States or being banned from the country. This was preceded by months of bipartisan rhetoric around the safety of U.S. user data that no one could cite any substantial evidence for. Meanwhile, U.S.-based companies have been turning a blind eye to carelessness and the deliberate selling of user data for years. In fact, a whistleblower at Meta recently came forth with allegations of Meta helping the Chinese Communist Party build censorship software. Similarly, all of the accusations made against Deepseek on data privacy issues are equally matched by the repeated issues OpenAI has encountered in its poor handling of data privacy. If the basis of the hostility toward Deepseek is really “American data privacy,” then there are many issues to take care of domestically before we even worry about international companies. The cautionary rhetoric around Chinese tech companies only exists to deflect the public suspicion and outrage away from the government officials and U.S. CEOs who spout it.
The stereotype that Chinese people and products are untrustworthy and malicious to the United States began with the first immigrants arriving on the West Coast during the 1800s gold rush. Throughout the 19th century, American newspapers depicted these immigrants as morally corrupt, believing they were stealing jobs from Americans. Americans saw Chinese men as corrupting forces on white women, and Chinese women almost exclusively as prostitutes. This led to several laws that made it extremely difficult to both immigrate to the U.S. from China or live life as a Chinese American, in an effort to cleanse the “yellow peril,” as newspapers loved to say.
In the 1950s, the U.S. included China as an enemy in the Cold War after the Chinese Communist Revolution rose to prominence and garnered sympathy due to the atrocities and massive casualties the country had suffered from Japan in World War II. During this time, both sides pushed massive amounts of propaganda and the United States once again harped on the stereotype of Chinese people as suspicious, especially in the era of McCarthyism. It was only when Asian Americans formed coalitions with other communities of color in the Civil Rights Movement that the stereotype switched to “the model minority” in order to sow discourse by pitting minority groups against each other rather than the majority-white power system. In our modern era, the content of anti-Chinese messaging is almost identical — even the words are shockingly similar. Time and time again, the same propaganda is used and reused.
If we are going to commit to being a leader in AI, this bravado around Deepseek is counterproductive. There’s a lot to admire about how the program was made, from it being open-weight and research-focused, to the small team behind it, to the fact that they brought on experts and consultants in many fields beyond tech, such as history and sciences. If Deepseek has proven anything, it’s that an openness to new ideas and collaboration leads to success. Currently, the U.S. general public is woefully uneducated on AI, and if we’re to make sure future developers and researchers exist in the American AI field, we shouldn’t be misleading them. Frankly, the story of Deepseek is one that certain American companies could benefit learning from. We can be international leaders in the future of AI, but we need to actually be better rather than just pretending that we are.
Daily Arts Writer Lin Yang can be reached at yanglinj@umich.edu.