At Weill Cornell Medical College, first-year student Kellen Vu doesn’t just learn how to diagnose illness—he’s also practicing how to make patients feel heard. But instead of a live actor in a simulation room, Vu is talking to an artificial intelligence-powered patient named MedSimAI.
AI is now stepping into the training ground of America’s future physicians—not to replace clinical education, but to transform how it’s delivered.
Traditionally, med students sharpen their communication and diagnostic skills by interacting with trained actors during Objective Structured Clinical Examinations (OSCEs). These graded encounters simulate real-world interactions, but they’re expensive, time-consuming, and limited in number. Feedback often takes days—or weeks.
Early attempts to replace actors with chatbots or virtual reality struggled with realism and accessibility. Chatbots sounded robotic. VR systems were costly and clunky. Neither scaled well.
But now, generative AI may have cracked the code.
Developed through a collaboration between Cornell Bowers College of Computing and Weill Cornell Medicine—along with partners at Yale and UCSF—MedSimAI uses state-of-the-art large language models to simulate lifelike patient conversations.
It offers:
- Text and voice modes for natural interaction
- Scripted case scenarios for consistency in training
- Real-time feedback that mimics OSCE grading
- Evaluation of empathy, accuracy, and reasoning via a second AI model
Students aren’t just learning what questions to ask—they’re seeing how their tone and word choice land. And that, educators say, is critical.
“Part of the physician’s job is being able to communicate in a way that helps patients be comfortable,” said Dr. MacKenzi Nicole Preston, a clinical pediatrician helping to integrate MedSimAI into the curriculum.
But it’s more than bedside manner. It’s also about asking the right questions and synthesizing information. With MedSimAI, students get to practice both in a low-pressure, repeatable environment—what educators call deliberate practice.
This year, first-year students used it to interview a virtual rheumatic heart disease patient. Second-years applied it during pediatric rotations, speaking with an AI parent about a child’s symptoms.
“It’s very natural to use,” said Vu. “It’s voice-based, so that lets the conversation flow smoothly. I think it’s important for practicing your bedside manner, because tone and phrasing matter a lot in real life with patients.”
The platform also shows possible diagnoses based on symptoms mentioned—and flags ones the student forgot to ask about. That kind of immediate feedback is priceless for learners, and nearly impossible to provide at scale with traditional methods.
The development team, led by René Kizilcec at Cornell, is already expanding MedSimAI to include:
- Doctor-to-doctor consult simulations
- Custom cases for different specialties
- Multi-institution integration with Yale and UCSF
Eventually, the platform could serve nursing students, veterinarians, or clinicians in rural or underserved areas who have fewer live training options.
“Nothing replaces practice with human patients,” Kizilcec acknowledged. But MedSimAI isn’t trying to replace human connection—it’s enhancing it.
By giving students more opportunities to mess up, reflect, and improve without risking patient safety, AI becomes a mirror—not a mask.
And in a time when healthcare is increasingly high-tech, tools like MedSimAI remind us: great care starts with a conversation.
