JMIR Medical Education (10/20/23) Vol. 9 Preiksaitis, Carl; Rose, Christian
A scoping review of available literature explored the potential opportunities and challenges of generative artificial intelligence (AI) in medical education to identify key areas for investigation and expansion. The reviewers primarily concentrated on literature published from 2022 onward following the broad uptake of generative transformer models like ChatGPT. The initial search yielded 2,761 articles that were ultimately narrowed down to 41 for final analysis. Twenty-one were opinion pieces, while 20 consisted of original research, of which 16 covered the performance of generative AI in standardized evaluations within medical education. Some studies showed impressive performance on standardized tests of medical knowledge overall, although just a handful explored the technology's ramifications in terms of self-directed learning or exam preparation. Many articles highlighted the AI models' potential to adjust to individual learners' needs to provide personalized learning strategies and materials in addition to individualized feedback. Several studies considered applying generative AI to augment emergency medicine physicians' communication skills, while two discussed AI image generation for case-based learning in plastic surgery and reflective sessions in a medical humanities curriculum. Some studies focused on generative AI as an assistive writing or research tool, especially for non-native English speakers. Also cited was the technology's potential to aid literature reviews and summarizations, although authors warned of models fabricating references and data. Many authors also cautioned against the possibility of generative AI models undermining academic integrity and potential misuse, including circumvention of traditional learning exercises designed to cultivate skills. A notable pitfall is the models' tendency to produce and promulgate misinformation, but also worrisome is the technology's potential to decrease learners' critical thinking and problem-solving capacities. "We propose the following three main areas of investigation relevant to learners, educators and both: (1) improving learners' AI literacy, (2) considering implications for assessment and (3) exploring human-AI interaction," the reviewers wrote.