A new tool lets medical students practice on AI-powered patients

“Hello, doctor. I’ve been having some chest discomfort.”

It’s the opening line in the robotic voice of “Mr. Perry,” an AI patient in McGraw Hill’s Clinical Reasoning tool. Dr. Scott Stern, editor in chief of the platform, follows up with some usual questions: “When did this start? What medications are you taking? Do you smoke?”

The education platform’s new offering lets medical students practice in this manner, whittling down a differential diagnosis through simulated patient interaction. The GenAI involved is restrained; human content writers authored all the actual dialogue, and if the student ever strays off topic, the AI patient directs them back to the health dilemma at hand. (“I’m not much of a baseball fan, doctor. Can we focus on my chest discomfort?”)

“The only part of the entire platform that’s AI-powered is the ability of the interface to understand my question. All the patient’s responses, all of that is written out,” Stern said.

Students can collect patient history and order virtual tests in the course of the appointment. At the end, the entire interaction will be evaluated based on how an expert would have handled it. The platform also includes modules to help students reason through problems, and illness scripts for different diseases.

Real-world experience: Stern, a professor of medicine at the University of Chicago, said the idea is to give students hands-on experience without making a diagnostic error in the real world.

“We spend a lot of time teaching about diseases, but not a lot teaching about how when someone comes in with a complaint, you systematically attack that complaint,” Stern said. “One of the things that brought it to light was in 2015, the Institute of Medicine came out with a national report and estimated that there were 12 million diagnostic errors a year in the US alone, sometimes with terrible consequences.”

McGraw Hill’s tool is not the only patient simulator out there today; interactive learning scenarios like these are seen by some as an area where GenAI technology might shine. Stern said the data based on real-world scenarios and the focus on students learning the reasoning behind actions set Clinical Reasoning apart.

“I want to make sure that the students understand, at the end, not just that they can recognize this case, but they can recognize how to evaluate the problem,” Stern said. “It’s very evidence-based; it’s not just somebody writing off the top of their head.”

AI with care: Clinical Reasoning’s cautious embrace of AI fits with a wider company mandate. McGraw Hill Chief Data Science and AI Officer Dylan Arena told us earlier this year that in an education field already upended by AI, the 137-year-old company wants to tread carefully in integrating the technology.

Stern said that’s true when it comes to education and medicine, where AI models might make mistakes beyond outright hallucinations by misunderstanding nuances of the studies they’ve ingested in training.

“AI is a double-edged sword right now. There’s no question that it can help us build some tools and it can get us information quickly. There’s also no question it transmits inaccurate information,” Stern said. “I don’t think we’re there yet to allow free range of the AI—we’ll see where it goes. But I think there’s a lot of potential for creating certainty, but error. You know, ‘I was told by AI [that] it’s OK,’ but it’s wrong.”