Generative AI in Health Professions Education: Educational Equity Implications – A Scoping Review Protocol

Generative AI tools—including large language models like ChatGPT, Gemini, and Claude, and image generation models such as DALL-E and Midjourney—have rapidly permeated health professions education since late 2022 . These technologies offer transformative potential: personalized tutoring at scale, simulated patient interactions for clinical training, multilingual content delivery, and creation of accessible educational materials . For learners in resource-limited settings, Indigenous students, non-native language speakers, and those with disabilities, GenAI has been heralded as a potential educational equalizer . Recent scoping reviews have documented the breadth of GenAI applications in HPE. A review of LLM-generated cases in health professions education found that 23 studies across 23 clinical domains demonstrated the feasibility of GenAI for case generation, though methodological heterogeneity and limited higher-level outcome evaluation remain concerns . In nursing education, a PAGER scoping review of 107 studies identified GenAI implementation across four domains—assessment, clinical simulation, content development, and faculty/student support—revealing hybrid implementation models as most effective . However, these reviews also highlighted significant gaps: limited methodological rigor (29.0% of empirical studies), inconsistent quality control, and equity concerns . Geographic distribution shows North American (42.1%) and Asian (29.9%) dominance in GenAI research, with African regions nearly absent . A growing body of evidence challenges the optimistic narrative of GenAI as an automatic equalizer. Algorithmic biases in AI-generated content have been documented across multiple dimensions. Studies have identified demographic bias in AI-generated clinical imagery, with systematic overrepresentation of lighter skin tones and males, and near-complete absence of certain ethnic and age groups. In simulated career guidance contexts, AI models have exhibited racial and gender biases that may disadvantage underrepresented minority applicants. Moreover, research on ChatGPT's integration into healthcare education reveals stark geographical concentration, with over 70% of studies originating from North America and East Asia . This raises concerns that the benefits of AI-enhanced education are not being studied—or perhaps not being realized—in the regions where they are most needed. Additional concerns include potential over-reliance on AI leading to diminished critical thinking, automation bias, and ethical challenges regarding data privacy and algorithmic accountability . This apparent contradiction—GenAI as both "bridge" and "barrier" to educational equity—constitutes what we term the educational equity paradox. Understanding this paradox requires systematic synthesis of a rapidly expanding but fragmented empirical literature base.