The AI gap isn't about enthusiasm
Outside the US and Europe there is near-universal appetite for AI combined with very limited access.
This Week in MedEd
7 open funding calls, 31 upcoming events, 37 live jobs (18 new, 7 closing this week), including:
💰 Advanced Digital Skills for AI Uptake in Health (up to €3.9M from the European Commission’s Digital Europe Programme, deadline 1 Oct 2026)
🎓 AMEE 2026 (Vienna, Austria, 22-26 Aug 2026)
💼 Assistant Professor in Fair and Responsible AI (Imperial College London, closes 19 Oct)
The full database, with deadlines, links and eligibility, free for 30 days:
Usually when we read medical education research the papers are single-site and in the general context of reliable electricity, institutional licences, and an assumption that every student owns a laptop. This is obviously not representative of medical education everywhere; the majority of the world’s doctors are trained in settings where none of those things can be taken for granted. This week I have selected papers that describe the state of AI in Medical Education in these contexts, with a focus on infrastructure and readiness.
Key points
Fewer than 10% of AI in Medical Education publications come from Latin America, Africa or Eastern Europe, so most of the evidence base is biased towards western contexts.
Among 143 physician educators at the West African College of Physicians, 98% thought AI could improve medical education, but only 35% had access to AI tools and only 23% had received any training.
415 students at the University of Aden in Yemen scored a mean of 79.82 out of 110 on a validated AI readiness scale, with ethics scoring highest. Appetite and self-assessed readiness are not the bottleneck.
AI amplifies whatever educational practice already exists rather than creating new capacity, so implementation failure is organisational, not technological.
The recurring practical recommendation is unglamorous. Faculty training, mobile-first delivery, open-access tooling, and locally adapted content..
Adefolalu completed a narrative review of AI in medical education across low- and middle-income countries with sub-Saharan Africa as the primary case1. They identified 2803 records and included 37 in the final synthesis. Of these 26% came from African countries and 38% came from high-income countries with LMIC applicability.
Analysis showed that fewer than 10% of these publications originated from Latin America, Africa or Eastern Europe. Additionally, a 2025 international survey of over 4500 students across 192 heath professions faculties found that more than 75% had received no formal AI education.
The barriers to AI use in medical education in these contexts are infrastructural rather than pedagogical:
intermittent power
poor connectivity
low electronic medical record adoption starving local models of data
algorithmic bias from training data drawn from WEIRD populations, with worked clinical examples in dermatology, pulse oximetry and cardiovascular risk stratification.
His suggestions for the future are therefore very practical, including mobile-first delivery given smartphone penetration, South-South partnerships rather than default North-South ones, open educational resources to control cost, and ethics teaching grounded in local values,
Abbas et al narrow the same question specifically to UG medical education They identified 210 records from PubMed and Google Scholar published between Jan 2020 and Mar 2025 and sifted them down to 43 articles2. The main finding was that the AI functions as “an amplifier of existing educational practices” rather than as an independent driver of improvement, which means a school with weak curriculum governance gets amplified weak curriculum governance.
The LMIC-specific barriers they name overlap heavily with Adefolalu’s but there is a specific focus on model bias. Models trained on high-income datasets miss local epidemiology, culture and language.. They provide recommendations organised as five pillars, with faculty development and AI literacy as priorities. Their point that faculty need training in critical appraisal of AI outputs rather than in tool operation reminds me a lot of the Feigerlova paper we discused last week3.
They also argue for hybrid assessment, pairing AI evaluation of structured cognitive tasks with human oversight of empathy, communication and professionalism. All of this is very reasonable and aligns with the literature elsewhere, but this is a narrative review with no formal risk-of-bias assessment, only two databases, and almost no quantitative synthesis.
Mamven et al surveyed 143 physician trainers recruited at the West African College of Physicians annual meeting4. Three quarters reported familiarity with AI concepts, but only 9.8% described themselves as ‘very familiar’, and only about 36% were aware of AI tools specifically for medical education. 19% reported ‘frequent use’, mostly for research, and 55% used AI in their role as educators.
They reported a significant gap between belief and provision. 98% thought AI could improve medical education and 90% supported integrating it, yet only 35% had access to AI tools and only 23% had received training.
The named barriers were lack of training (63.4%), tools being unavailable (51.7%) and poor internet connectivity (51%). Frequent AI use was independently associated with holding a higher degree (OR 3.7, 95% CI 1.28 -10.96) and with having attended AI training (OR 2.79, 95% CI 1.01 - 7.72). Nearly everyone (99%) wanted AI as a supplement rather than a replacement.
Al-Madhagi et al ran a similar study, but this time focussed on students rather than educators5. They surveyed 415 undergraduate health professions students at the University of Aden in Yemen using the validated Medical Artificial Intelligence Readiness Scale (MAIRS-MS) . The mean readiness score was 79.82 out of 110, which is a reasonably favourable result and comparable to what has been reported from far better-resourced settings, with the ethics and ability domains scoring highest.
80.2% had prior awareness of AI technologies and 44.8% had used them in a medical context. Male students scored significantly higher on the cognition domain than female students (27.73 versus 26.07, p=0.005) and readiness showed no significant association with age, GPA, academic year or income level.
I know we wouldn’t normally place a great deal of weight on a single-site study, but I think these findings are very interesting. They show that students in one of the most infrastructure-constrained health systems in the world arrive with roughly the same self-assessed readiness and ethical awareness as their peers elsewhere.
So, what’s the common thread that runs through these papers? We know that in high-income settings the conversation is largely about persuading sceptical faculty and restraining over-enthusiastic students. In the settings these papers describe, the enthusiasm is already near-universal and the ethical awareness is already high. What is missing is access and infrastructure . It doesn’t seem to be the case that some places are behind on AI attitudes and need to catch up.
Adefolalu AO. Integrating Artificial Intelligence into Medical Education in LMICs: A Narrative Review. Advances in Medical Education and Practice. 2026. https://doi.org/10.2147/AMEP.S613617
Abbas U, Khalid MU, Tanveer M, Arshad F, Musawwir UA, Hasan SM, Hussain N. Opportunities and challenges of integrating artificial intelligence into undergraduate medical education in low- and middle-income countries. Discover Artificial Intelligence. 2026. https://doi.org/10.1007/s44163-026-01650-0
Feigerlova E. How to Teach Generative Artificial Intelligence in Undergraduate Medical Education. The Clinical Teacher. 2026. https://doi.org/10.1111/tct.70420
Mamven M, Oleribe OO, Offiong U, Acquah G, Ogunrinde OO, Haruna A, Ansa V, Uzochukwu BSC, Taylor-Robinson SD. Artificial intelligence awareness, usage, and readiness among fellows of the West African College of Physicians. BMC Medical Education. 2026. https://doi.org/10.1186/s12909-026-10063-7
Al-Madhagi OA, Al-Shadadi HH, Al-Hubaishi AK, Dobian SA, Alkhalifi AK, Hashem LM. Artificial Intelligence (AI) readiness among medical students in Yemen: a cross-sectional study using the MAIRS-MS Scale. BMC Medical Education. 2026. https://doi.org/10.1186/s12909-026-10106-z