During Georgie Nahass’ first year of medical school at the University of Illinois Chicago, many of his fellow students expressed interest in the computer science research he was conducting on the side.

When Nahass decided to host a series of lectures to teach his peers some basic elements of programming, he found an enthusiastic audience.

“Lots of people came,” Nahass said. “There were like a hundred people that showed up at the first one.”

This made Nahass aware of an unmet need among medical students: While over 80% of physicians currently use AI in their work, medical schools still provide little education on these technologies.

In a 2024 global survey on the attitudes of medical, dental, and veterinary students in 48 countries, 67.6% of respondents reported positive attitudes toward AI in healthcare, but less than 25% had actually received education on it, and 75.3% reported little to no knowledge of AI.

“I haven’t necessarily had trainings on [AI],” said Kailyn Geter, Howard University medical student and president of the Student National Medical Association, though she sees the attending physicians around her using it.

Similarly, Jeff Kim, an MD/PhD student at the University of Chicago Illinois, says his AI training has been limited to single slides slapped onto the ends of lectures or informal advice from practitioners he worked alongside.

Fear of Falling Behind

With new AI tools constantly being released, some students fear they won’t keep up.

“I think my worry is that the hospital might not look the same today as it will 20 years from now,” said Kim. “If I don’t adapt to the changing environment in the hospital, I might be pushed out.”

Geter says she feels comfortable using the AI tools that exist now but doesn’t know what to expect later on. “We don’t know how extensive [AI] is going to get,” she said.

Even as some medical schools are starting to teach on AI, they vary widely in the material they cover.

For example, a 2025 Association of American Medical Colleges (AAMC) survey of medical and DO schools in the US and Canada found that while 77% of schools said they offered coursework on AI, only 39% provided training on data analysis and evaluation — which students and researchers agree may be more valuable than teaching on AI’s practical applications.

“There will always be a better tool within the next 2 years,” said Felix Busch, MD, based at the Technical University of Munich in Munich, Germany, and author of the 2024 global survey. Busch believes schools should educate students on how AI models are trained and where their data comes from rather than on individual AI tools.

Some Schools Don’t Address Larger AI Concerns

Meanwhile, about 70% of schools reported teaching on social, ethical, and legal aspects of medical AI.

Tina Nguyen, PhD, a bioethicist at The University of Texas Medical Branch in Galveston, Texas, said ethics training can counteract automation bias, the implicit assumption that AI’s output can be trusted without verification from a human.

Because students may be more comfortable with some digital technologies than older clinicians, Nguyen says, they may disproportionately fall victim to this bias.

“If they use the outputs and it’s completely wrong,” Nguyen said, “that blame is not going to be on the AI. It’s going to be on the student.”

Kim has noticed many of his peers relying on ChatGPT to make decisions — sometimes even during real patient consultations.

“They’ll talk to patients, and then if they forgot what the diagnosis might be, they would just say, ‘I’m going to the restroom,’” Kim said. “And they would sneak out and use these large language models [LLMs] to help them identify the diagnosis.”

He worries that this reliance on AI is preventing his peers from developing their own clinical reasoning skills.

If they “are sitting on their laptop typing,” he said, “and ChatGPT gives an answer in like five seconds, [they’re] not really practicing and training that skill.”

Gigi Magan, MD, MPH, a family physician and assistant clinical professor at UC Riverside School of Medicine, says some schools are responding by prohibiting LLM use among students, but she sees this as the wrong approach.

“Even if it’s not formally part of the curriculum, I do think that students are probably still going to use [LLMs],” she said, which means teaching students to use LLMs responsibly could be more valuable.

Magan also believes students should learn about how uncritically applying AI outputs can exacerbate healthcare inequities.

In addition, students should learn to ensure patient confidentiality and consent when using AI. Without this training, Magan says, students could inadvertently enter private patient information, which comes with consequences.

“There have been some cases I’m aware of where clinicians have put in patient identifiable information into these tools and it’s come back to bite them,” said Sarup Saroha, a medical student at University College London in London, England. Some healthcare providers are even facing lawsuits from patients who say their privacy was invaded when AI tools were used to record and transmit their information.

Several official bodies, including the AAMC and the World Medical Association, have recently called for AI training in medical education, but institutions face barriers to adopting it. For example, some institutions lack faculty with AI expertise or access to the necessary digital tools.

Little Consensus on Medical Education About Clinical AI

There are still no standardized frameworks to guide AI education in medical schools, and medical school curriculum is already jam-packed with material.

But the frameworks are necessary because “[AI] is a massive field of research — just as large as medicine.” Students can’t be expected to become data science experts, he says, making it important to determine which information is most essential to cover.

Nahass compares this to the training students receive on other medical technologies.

“[Students] don’t need to know what screw goes into what panel of the MRI machine,” he said. “But if you’re going to use an MRI machine, you probably should have some intuition as to what’s happening in the MRI machine.”

What makes MRI education work, he suggests, is that there’s agreement about the level of understanding students should develop before using these machines. “With AI, though, those agreed upon tenets haven’t really been established,” he said.

Optional AI courses like Nahass’ lecture series are one way around this, giving interested students a way to learn about AI without interrupting the core curriculum.

Following the success of these lectures, Nahass says, he and his college roommate, University of Washington medical student Hugh Alessi, founded Code Grand Rounds, a startup that provides curricula on AI and other data science topics to medical students and residency programs. For individual students, the platform is completely free.

“We didn’t want to make it prohibitive for people like us to get on and learn,” said Nahass. “I think it’s important to know what tools you use when you’re engaging with people and you’re making decisions that impact lives.”

Experts quoted in this story reported no disclosures.