Why Chinese AI DeepSeek Won’t (and Can’t) Work

(Daily Signal)—The world is abuzz with the news about DeepSeek, the new open source AI platform that changed the paradigm regarding how much time, money, and processing power is required to develop advanced artificial intelligence systems.

Its origin in Communist China only adds to its mystery—and concern for those of us in the West.

Big Chinese companies, including automaker Great Wall Motor and the country’s three largest telecom companies, have recently announced they’re getting behind DeepSeek and integrating it into their operations.

If the global artificial intelligence race hadn’t already begun, it’s in full swing now.

There’s just one problem with DeepSeek: It can’t work.

We’ve all heard the term GIGO, which stands for “garbage in, garbage out.” It’s a truism not only about the data we put into computers, but the food we eat and the entertainment we consume. Given what we know about how AI works, a new term is apropos: BIBO, or “bias in, bias out.”

Just for fun, I asked American-made Chat GPT to confess whether or not it’s biased. Here’s what it said:

As an AI language model, I strive to be unbiased and provide information and assistance based on the data I have been trained on.

However, it’s important to note that my responses are generated based on patterns and information present in the training data, which may include biases inherent in the data itself.

While efforts have been made to mitigate bias during the training process, I may still inadvertently exhibit some biases or reflect the biases of the data I was trained on.

It’s always a good idea to critically evaluate the information and consult multiple sources to obtain a well-rounded perspective.

It’s comforting to hear such a confession, and it underscores how the age of AI will put an increasing premium on truth. The corollary to Richard Weaver’s post-World War II formulation that ideas have consequences is that bad ideas have victims, which means the incentives for effective AI applications will be to root out as much error as possible.

Using artificial intelligence, health care diagnoses and treatments will be better and more quickly administered, but only if they use accurate information. If not, they will harm people. With AI, transportation systems will be faster and safer, but only if they are informed by good data. If not, they’ll cause accidents. Manufacturing systems that leverage AI will be more efficient and productive, but only if they’re guided by factual inputs. If not, they won’t work.

Artificial intelligence is accelerating the speed of learning, as did the internet before it and the printing press before that. With AI, the old saying that a lie can make it halfway around the world while the truth is still lacing up its shoes might become an anachronism.

In the realm of public policy, the extent to which AI can more quickly expose bad ideas and confirm good ideas, the more humanity will flourish.

Politicians and pundits who misrepresent the facts may get busted faster and with better evidence. Economic data collected more quickly and in ever-more-detailed forms will better expose misinformed tax policy. Global intelligence gathered faster and more accurately will lower the uncertainty that often leads to destructive foreign policy decisions. Air traffic data, richer and more sophisticated, will help prevent tragic accidents in our skies.

There’s nary an arena of public policy or private enterprise in which AI can’t be of benefit. As long as it’s all based on truth, that is.

In DeepSeek’s case, incomplete or incorrect information isn’t a bug, it’s a feature. Totalitarian China has long demonstrated how technology can be used to increasingly track, deceive, and oppress its people, and its leaders aren’t interested in DeepSeek telling the whole truth to its users (to say nothing of its usefulness as yet another spy tool).

Early adopters of the platform have reported inquiring about topics like the 1989 massacre in Tiananmen Square or the current plight of the Uyghurs in China’s Xinjiang province, only to be politely stonewalled and stymied. Even left-leaning Politico recently published this headline: “We asked DeepSeek about geopolitics. It gave us Beijing talking points.”

DeepSeek can’t win for the simple reason that it can’t tell the truth, at least about politics. And policy. And persecution. And, say, the origin of certain viruses. The other existing AI applications are far from perfect, but at least their creators are working on improving them, knowing that when it comes to broad, long-term adoption, trust is the coin of the realm.

If we can’t trust what AI says (or ultimately does) in one area, we can’t trust it in any.

The other day someone asked me if the world would be a safer place in the age of artificial intelligence. Since all lies eventually get exposed, my answer was, ultimately, yes. But given DeepSeek’s example, that may happen only after the world first becomes more dangerous.

We may have to wait for it, like communism itself, to (finally and forever) end up on the ash heap of history.