Using Generative AI to Simulate Patient History-Taking in a Problem-Based Learning Tutorial: A Mixed-Methods Study

Abstract Background Medical educators who implement problem-based learning (PBL) strive to balance realism and feasibility when simulating patient cases, aiming to stimulate collaborative group discussion, engage students’ clinical reasoning, motivate self-directed learning, and promote the development of actionable scientific understanding. Recent advances in generative artificial intelligence (AI) offer exciting new potential for patient simulation in PBL Method This study used a between-groups, mixed-methods approach to (1) form a comprehensive picture of Year 2 medical student interactions with a generative AI-simulated patient in a PBL tutorial, as compared to interactions with multimedia patient case materials; and (2) triangulate on the impact these interactions had on learning. Two groups of students (N = 13) gathered patient history information from a generative AI-enabled, 3D-animated avatar (AI condition). Two other student groups (N = 13) gathered patient history information from a multimedia database using keyword searching (Electronic PBL Module ePBLM condition). We used descriptive observation to explore student interactions with both forms of the simulated patient, and we quantitatively compared students’ perceptions of their learning experience and recall of patient history information across conditions. Results Students in the AI condition rated their present, AI-augmented PBL learning experience—particularly its clinical accuracy and teamwork aspects—significantly higher than they rated their previous PBL learning experiences using ePBLMs. Recall of patient history information did not differ between conditions. Descriptive observation indicated that the AI avatar presented case content accurately, with an appropriate amount of information provided in response to students’ questions. Students were highly engaged as a group in taking a history from the avatar. However, although students used language suggestive of anthropomorphizing of the AI (e.g., gender pronouns), they appeared to orient to it as an augmented “question bank” for gathering patient history information, using a questioning strategy akin to querying an ePBLM. Discussion Optimizing AI implementation to stimulate clinical reasoning and patient communication skills in PBL could include (1) starting early, perhaps in Year 1, before an alternative interactional framework can take hold; (2) orienting students to the AI to help them understand its capabilities; and (3) encouraging “play” with or “discovery learning” of the AI’s capabilities.