ИИ-детектив в Минздраве США: Роберт Кеннеди-младший ищет побочки там, где их не видели

Imagine you give the most powerful digital microscope to a person who is already convinced that microbes of a very specific type are to blame for all the world's troubles. He will surely find them, even if it's just dust on a lens or a glass defect. This is roughly what's happening now in the corridors of the US Department of Health and Human Services (HHS).

The department, now inextricably associated with the controversial figure of Robert F. Kennedy Jr., has decided to arm itself with artificial intelligence to find links between vaccination and various diseases.

The idea of data analysis itself is a good and even necessary matter. In medicine, there is always room to search for extremely rare side effects that may have escaped researchers' attention during standard clinical trials. However, the devil, as always, lies in the details—and in who exactly holds the remote control of this technology.

Kennedy Jr. has spent decades building his public career on harsh skepticism toward vaccines. Now at his disposal is a tool capable of turning scattered and often unconfirmed complaints into ostensibly scientifically grounded hypotheses.

The problem is that modern AI is neither an impartial judge nor a bearer of absolute truth. It is a mirror of the data loaded into it and of the prompts dictated to it. If you feed a neural network a massive VAERS database containing any complaints people have after vaccinations—from mild headaches to random injuries—and persistently ask it to find patterns, it will find them.

Neural networks are virtuosos at hallucinating correlations, especially if asked very insistently. Scientists around the world fear that the new tool will become a conveyor belt for producing sensational headlines about proven harm, completely ignoring the basic medical principle: after does not mean because of.

This case poses a crucial ethical question to the entire AI industry. We are used to discussing absurd chatbot errors, when they confuse celebrities' birth dates or invent non-existent books. But when an algorithm begins generating medical hypotheses capable of influencing state policy and the health of millions of people, the stakes skyrocket. Using machine learning to confirm one's own bias (so-called confirmation bias) is perhaps the most dangerous scenario for the application of technology in government. Instead of using algorithms to objectively verify drug safety, we risk obtaining a perfect tool for legitimizing conspiracy theories.

The situation is further complicated by the fact that the general public tends to trust conclusions made by computers more than the words of politicians. If the US Department of Health and Human Services begins releasing reports generated by neural networks, the scientific community will find it extremely difficult to counter these arguments. After all, arguing with numbers and graphs backed by mysterious and powerful intelligence is psychologically harder than arguing with a public figure.

We are entering an era of alternative data, where AI could become not a doctor's assistant, but a chief weapon in an information war against evidence-based medicine. And we will be dealing with the consequences of this experiment for a very long time to come.

Key: Will the scientific community be able to create a system of checks and balances, or are we headed for an era when medical standards will be dictated by algorithms with a predetermined direction of search?