Feasibility of an AI-Simulated Patient Encounter Delivered in Group and Individual Learning Formats for First-Year Medical Students: A Two-Cohort Pilot Study
This pilot study demonstrates that an AI-simulated patient encounter is feasible and acceptable for first-year medical students in both group and individual formats, yielding exploratory positive shifts in engagement and psychological safety while highlighting the need for larger randomized trials to confirm these findings.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
In the early years of medical school, students must learn to talk to patients, ask the right questions, and piece together a diagnosis from a conversation. This skill is traditionally practiced with "standardized patients," actors trained to play a specific role so every student gets the same experience. However, these human actors are expensive and limited in number. Recently, a new tool has emerged: artificial intelligence. Specifically, large language models—computer systems capable of generating human-like text and speech—can now act as virtual patients. These digital characters can listen, respond, and maintain a consistent medical story, offering students a way to practice clinical conversations without needing a human actor present. The question for educators is not just whether these machines can talk, but how best to use them in a classroom. Should students practice alone, or should they gather in a group to discuss the case with a teacher?
Researchers at the Southern Illinois University School of Medicine set out to test these two different approaches with first-year medical students. They created a single, realistic medical scenario involving a patient with gastrointestinal issues and programmed an artificial intelligence to play that patient. They then divided the students into two groups. One group of seven students sat together in a classroom, where they took turns interviewing the AI patient projected on a large screen, guided by a faculty member. The other group of twelve students went to private spaces and completed the exact same interview on their own devices, without a teacher present and without other students watching. Before and after their sessions, all students filled out surveys to measure how engaged they felt and how safe they felt making mistakes or asking questions. The researchers also developed a way to automatically analyze the transcripts of these conversations to see what the students actually said.
The study found that both formats worked well and were acceptable to the students. The students in both the group setting and the individual setting reported feeling more engaged and more psychologically safe after the experience. This sense of safety is crucial in medical training; it means a student feels comfortable admitting they do not know an answer or trying a new line of questioning without fear of embarrassment. While the individual group showed a particularly noticeable increase in this feeling of safety, the study was too small to say with certainty that one format was definitively better than the other. The researchers noted that the numbers were not large enough to prove a difference between the two methods, so the results should be viewed as a promising sign rather than a final verdict.
One of the most interesting findings came from asking the students what they preferred. After trying the solo version, the majority of those students said they would prefer a hybrid model: doing the AI interview alone first, and then joining a group discussion with a teacher afterward. This suggests that students value the privacy to practice and make mistakes on their own, followed by the opportunity to refine their thinking with peers and an instructor. The study also tracked a slight drop in how prepared students felt to discuss the case afterward, but the researchers believe this might simply mean the students realized how much there was to learn, rather than a failure of the tool.
The researchers were careful to state that this was a pilot study, a preliminary test designed to see if the idea works and to refine their tools for a larger experiment later. They did not claim to have solved the problem of how to teach medical students, nor did they prove that AI is a perfect replacement for human actors. Instead, they demonstrated that it is possible to run these AI encounters in different ways and that students are willing to participate. They also introduced a new method for automatically reading and analyzing the conversation transcripts, which could help future researchers measure student performance more objectively. Ultimately, the work suggests that artificial intelligence can be a useful part of medical education, especially when used as a stepping stone to group learning, but more research with larger groups of students is needed to confirm these early observations.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.