JMIR Publications

ABSTRACT

Artificial intelligence (AI) is rapidly becoming embedded in medical education through generative AI, intelligent tutoring systems, virtual patients, automated assessment, adaptive learning, and clinical decision-support applications. However, the conditions under which these technologies are being adopted differ profoundly between high-income countries and low- and middle-income countries (LMICs). In many LMIC settings, AI is entering educational systems already affected by unequal internet access, equipment shortages, limited digital infrastructure, restricted institutional budgets, insufficient faculty training, and fragmented governance. Consequently, the principal challenge is no longer whether AI should enter medical education, but how it can be incorporated without reproducing or widening pre-existing educational inequalities. In this Perspective, we argue that AI adoption in medical education in LMICs is constrained by three interconnected divides: an access divide, related to connectivity, hardware, affordability, and infrastructure; a capability divide, involving faculty preparedness, institutional governance, and AI literacy; and an evidence divide, characterized by the limited number of locally generated, longitudinal, and implementation-oriented studies. We propose a resource-aware pathway for adoption based on six principles: start with the educational problem rather than the technology; design for the minimum viable infrastructure; build faculty capability before institutional scaling; preserve human oversight and local validation; incorporate governance and equity from the outset; and evaluate educational value, cost, sustainability, and unintended consequences before expansion. Experiences from clinical AI implementation in resource-constrained settings provide additional transferable lessons regarding realistic hardware, workflow integration, local validation, and human-in-the-loop safeguards. AI could expand access to high-quality medical education in LMICs, but only if implementation is context-aware, pedagogically justified, locally governed, and evaluated against outcomes that matter.