It’s no secret that doctors across the world are using AI tools to do stuff from taking notes to looking up questions. But what about the next generation of doctors still in training? According to two medical professionals on the ground, the situation is looking pretty alarming.

With AI being used by med students, “the danger is not just deskilling but never-skilling,” write Simar Bajaj, a journalist and med student at the Stanford University School of Medicine, and Joseph Sakran, a trauma surgeon and executive vice-chair of surgery at Johns Hopkins Medicine, in an essay for The Guardian.

“Although a doctor who has forgotten how to reason is recoverable,” they added, “one who never learned how may not be.”

Bajaj and Sakran point to two alarming recent findings.

One is internal data showing that around two thirds of doctors in the US are using OpenEvidence, an AI chatbot specifically designed for clinicians. The technology is already prevalent and deeply woven into how medical professionals carry out their jobs.

The next is a new study published in the journal Nature Medicine that evaluated the reliability of these medical-specific large language models. Shockingly, the clinical AI tools performed worse  at answering medical queries than general purpose chatbots like ChatGPT and Claude did, the work found. According to the authors, it performed about on par to Google’s AI Overviews, which are notorious for being wonky and inaccurate.

Even if AI tools may be handy for research — and putting aside the major questions over their reliability — they’re no substitute for relying on your brain, as fallible as it might be, because when you’re learning, “the struggle is the point.”

“For example, trainees once asked to build a list of potential diagnoses might struggle and offer an incomplete set, learning what they missed, sometimes painfully,” Bajaj and Sakran wrote. “Now, trainees can simply ask OpenEvidence and get a nearly perfect answer, complete with possibilities they might have never considered and none of the embarrassment of having overlooked them.”

Where AI is different from medical advances in the past is that it’s “not just expanding what doctors can see,” they wrote, “but inserting itself into the cognitive machinery that training is meant to build.” 

A bevy of research on the “cognitive offloading” that AI encourages, in which users naturally begin to outsource a lot of their thinking onto the tools, backs this diagnosis. Some studies have suggested AI impairs critical thinking, and leads to lower brain activity during cognitive tasks.

It’s a conundrum gripping the medical world. Earlier this year, doctors from top medical schools proposed a stricter framework on how medical students should use AI tools.

“The process of deliberately crafting the note forces us to use our brains to really wrestle with what’s happening,” Jaideep Talwalkar, associate dean of educational technology and innovation at Yale School of Medicine, told Stat News. “There’s an importance in doing that with great repetition.”

Bajaj and Sakran have a few suggestions for how to deal with this. One is requiring trainees to periodically work through cases without any AI help, similar to how the Federal Aviation Administration advises aviators to periodically disengage autopilot and fly by hand. Of course, another powerful deterrent could be shame. A recent John Hopkins student found that doctors who used AI are viewed with scorn and skepticism by their peers.

More on AI: AI Model Invents Completely New Viruses Unknown to Nature