HHS Is Using AI Effectively? Someone’s Hallucinating.

From day one, Health and Human Services Secretary Robert F. Kennedy Jr. has made “radical transparency” a top priority, while promising to dramatically reform public health with the help of artificial intelligence.

Today, the only thing seemingly transparent is that he’s failing to live up to his promise.

After setting an “aggressive timeline” for incorporating the latest AI technology, the U.S. Food and Drug Administration—the HHS agency responsible for food safety and the safety and effectiveness of drugs and medical devices—launched Elsa, its new AI assistant.

Tasked with the cumbersome job of reviewing clinical protocols, highlighting scientific reviews and streamlining other steps in the drug-approval process, Elsa was touted by top officials as a “dynamic force” that would usher in “the AI era at the FDA.”

What it ushered in instead was chaos and confusion.

After Elsa’s June launch, FDA employees were alarmed to find that it regularly made mistakes and routinely mischaracterized findings from clinical trials. Some claimed the assistant hallucinated and “was making stuff up” when summarizing documents.

FDA officials have asserted this isn’t true. An HHS official insisted the employees’ statements were “taken out of context” and came from “disgruntled former employees and sources who have never even used the current version of Elsa.”

Though glitches might be common when new technologies are introduced into an organization’s operations, technological failures have plagued those in charge of “making America healthy again” for months.

In May, weeks before Elsa’s debut, HHS released a report documenting concerns about vaccines, prescription drugs and food safety. The report was later criticized for citing some nonexistent studies and providing at least 21 dead links to others. Independent analysis concluded that the inaccuracies and other problems were probably due to the use of ChatGPT in preparing the report.

These events are worrisome, and more are likely to come since Kennedy and his team hope to use AI more extensively: to enhance telemedicine, create a data-sharing network enabling the Centers for Medicare and Medicaid Services to partner with tech giants, and phase out some animal testing, replacing it with AI for certain products seeking FDA approval.

The problem, however, isn’t AI. The problem is that HHS and its agencies appear to be using AI technology that isn’t ready, or even appropriate, for the task.

Though Kennedy has spent much of his career challenging and attacking the U.S. food and health care industries, he would be wiser to study how they are effectively using artificial intelligence to improve health outcomes and reduce costs.

Healthcare, an international peer-reviewed scientific journal, for example, reviewed 200 academic studies on the impact of AI on health care economics and found overwhelming evidence that, when used properly, AI significantly reduces costs, saving patients money on both diagnosis and treatment.

Consider obesity, arguably the greatest public health concern in the United States.

Although injectable weight-loss treatments such as Wegovy and Mounjaro have helped millions lose weight, they (like other pharmaceuticals) can be harmful if misused. Whereas Kennedy aims to reduce patient use of these drugs, private companies are channeling AI technology to help patients use them more effectively, by tailoring doses, adjusting their diets, and alerting them when the treatments are not working as expected.

Other AI-based apps provide exactly what Kennedy wants: a reliable way to lose weight without drug-related side effects. For example, the weight loss app Myndset utilizes AI to “retrain the brain” in ways that help patients lose weight naturally, and this is said to be five to seven times more effective than standard diet and exercise routines.

AI’s potential to improve health care is unprecedented. But it needs to be used properly by people who understand both its potential and its limitations.

Government efforts headed by Kennedy have so far been disappointing and have left little reason to expect improvement. Hoping otherwise is a hallucination that even AI might not be able to fix.