Evaluating the educational impact of an AI-powered totipotent interactive patient simulator (SimPatient)

AI-powered tools are increasingly used to address the growing demand for clinical teaching capacity in medical education. We present SimPatient, a novel AI-simulated patient built on a bespoke large language model (LLM) specifically trained for realistic clinical dialogue. Unlike conventional scripted or chatbot-based systems, TIPS offers fully interactive, multimodal simulations (text, voice, and video), allowing learners to practise consultation skills across diverse formats. Its design supports dynamic, context-sensitive patient behaviour and incorporates a wide range of demographic characteristics, addressing long-standing limitations in diversity and flexibility seen in human simulated patients and pre-scripted platforms. In contrast to commercial solutions, TIPS enables tailored content, alignment with local curricula, and enhanced control over pedagogical parameters. The use of a dedicated LLM and controlled training data also provides a research platform for investigating learning outcomes, bias, and student-AI interaction.

To evaluate educational impact, we conducted a mixed-methods study involving second-year medical students who used TIPS in repeated self-directed consultation exercises over one semester, alongside their existing communication skills curriculum. Outcomes were assessed using pre- and post-intervention questionnaires and structured performance tests.

Preliminary data from 44 participants (mean age 21; 59.1% female) indicate that 61% had prior experience with AI tools such as ChatGPT, using them primarily to summarise content, clarify complex topics, and organise study schedules. Around 68% reported use of AI in academic settings. These findings suggest a student cohort familiar with GenAI but holding varied expectations about its role in healthcare and learning. Post-intervention data, currently in analysis, will assess changes in consultation competence and shifts in learner attitudes following extended engagement with the platform.

These preliminary findings underscore the feasibility of integrating bespoke GenAI tools like SimPatient into undergraduate medical education. As post-intervention data become available, we aim to determine whether such tools can meaningfully enhance consultation skills and learner confidence. If effective, SimPatient may offer a scalable and curriculum-aligned complement to faculty-led teaching, particularly in settings where access to human simulated patients or diverse clinical scenarios is limited.