Harnessing the Generative Power of AI to Move Closer to Personalized Medical Education

The traditional one-size-fits-all paradigm in medical education fails to address the diverse learning needs of students. Educational psychologist Benjamin Bloom estimated gains in student performance of up to 2 standard deviations ("2-sigma") with one-on-one tutoring. Providing one-on-one tutoring to every learner, as Bloom describes, is often infeasible in large cohorts. Artificial intelligence (AI) offers a promising solution, with the potential to tailor educational experiences to individual learners.

At the University of Cincinnati College of Medicine, from March to September 2023, the authors developed and piloted 2-Sigma, an AI-based simulation platform that incorporates adaptive clinical scenarios for second-year medical students. Informed by Bloom's 2-sigma tutoring theory, adult learning principles, and clinical reasoning development frameworks, the 2-Sigma platform leverages chain-of-thought and few-shot prompting to simulate diverse virtual patient encounters. Students completed 8 cases during their clinical skills course, receiving immediate AI-generated feedback after diagnosing or managing the case.

Across 176 second-year medical students, 1,603 unique sessions were recorded. Use of the 2-Sigma platform varied across cases, with notable discrepancies in students' diagnostic accuracy, particularly in cases like viral myocarditis and hypersensitivity pneumonitis. Students frequently asked more questions in scenarios where their diagnoses were incorrect.

The 2-Sigma platform has the potential to address Bloom's 2-sigma problem by providing scalable one-on-one style interactions and feedback. This platform allows educators to collect and analyze data from the case transcripts that would otherwise be inaccessible or complicated to obtain, providing new opportunities to assess students' clinical reasoning behaviors and to tailor feedback. Planned refinements include expanded analysis of learners' behaviors, ensuring AI accuracy, and improved personalization of feedback. Future work will systematically evaluate how AI-generated feedback influences clinical reasoning and learners' attitudes.