Artificial intelligence (AI) is transforming medical education, enhancing knowledge acquisition, teaching, assessment, and curriculum delivery. While AI offers the potential to democratise access and improve inclusivity, little is known about how equity is addressed in AI-enabled medical education. This scoping review maps current evidence, identifies gaps, and provides insights for equitable implementation.
A scoping review was conducted following the Joanna Briggs Institute methodology. Peer-reviewed literature was identified through MEDLINE, Evidence-Based Medicine Reviews (Health Technology Assessment), Health and Psychosocial Instruments and Global Health. Screening, data extraction, and thematic analysis were performed by multiple reviewers. Quantitative data were summarised descriptively, and qualitative data were synthesised to identify narrative themes.
Of 256 records identified, 80 full-text articles were reviewed, and 35 studies were included. Most were published between 2022 and 2025 and originated predominantly from high-income countries. AI applications focused on large language models (48%), with curriculum design (49%) being the most common area of focus. Equity themes were: (1) equitable access (demographic, geographic, and distributed learning disparities); (2) bias (algorithmic, cultural, and socioeconomic); and (3) AI literacy (limited curricular exposure and preparedness).
Previous literature reviews have examined AI and its ethical use in medical education, but not through an equity lens. This scoping review highlights that medical education is at a pivotal juncture, shaped by converging imperatives: inclusivity and the rapid evolution of AI. To expand access and standardisation, equitable implementation requires AI literacy, contextually relevant tools, and active bias mitigation. Future research should prioritise inclusive, contextually appropriate, and accessible AI interventions across medical education.
1. Introduction
Artificial intelligence (AI) is rapidly transforming how both healthcare and medical education are delivered (Ahsan, 2025; Duan et al., 2025; Greengrass, 2024). Over the past decade, AI has evolved from early adoption to become an embedded, accepted, and integral component of medical education (Gordon et al., 2024). Currently, AI is used across many areas, including student selection, knowledge acquisition, teaching modalities, assessment, curriculum design and delivery, and administrative processes (Duan et al., 2025; Gordon et al., 2024). AI is not only a transformative tool for teaching and learning, but a core competency that future medical practitioners need to master (Greengrass, 2024).
Several broad scoping and narrative literature reviews have examined how AI is being applied and its ethical use in medical education (Ahsan, 2025; Gordon et al., 2024; Rincón et al., 2025). However, few have focused on issues surrounding equitable access. Gordon et al. (2024) emphasised the importance of collaboration between educators, policymakers, and AI developers to ensure “AI tools are available to all without bias” (Gordon et al., 2024). Despite this, their review did not explicitly evaluate included studies through an equity lens (Gordon et al., 2024). Similarly, Ahsan et al. (2025) and Rincón et al. (2025) did not focus on equity but raised concerns that AI favours high-resource settings, exacerbating existing disparities in medicine and widening the gap for underrepresented groups (Ahsan, 2025; Rincón et al., 2025). Key issues included limited technological infrastructure, financial constraints, biased datasets, poor connectivity in rural areas and limited support for educators and students in effective use (Ahsan, 2025; Akefe et al., 2025; Rincón et al., 2025). However, if purposefully implemented, with attention to geographical needs, resource availability and cultural context, AI has the capacity to enhance and democratise accessibility and equity in medical education (Ahsan, 2025; Bray et al., 2025; Rincón et al., 2025).
Equity is the state in which everyone has a fair and just opportunity (Organisation for Economic Co-operation and Development, 2012). Efforts to create a more equitable medical profession have led to strategies aimed at widening access and participation for diverse, underrepresented, and disadvantaged groups to study medicine (Fuller et al., 2025; George et al., 2025). These strategies vary and are shaped by unique cultural, geographic, economic and political contexts (George et al., 2025). Educating a more diverse cohort of medical professionals will help to ensure the workforce is representative of the communities they serve, which is anticipated to address workforce shortages and enhance health outcomes (Fuller et al., 2025; George et al., 2025). Widening access and participation are two separate yet interconnected concepts; widening access refers to reducing barriers to admission to medical school (opening the door), while widening participation ensures that the medical education is delivered in an inclusive teaching environment, which is accessible and appropriate for diverse and underrepresented populations (Akefe et al., 2025; Case, 2025; George et al., 2025).
AI offers the potential to widen access and participation in medicine by enhancing learning opportunities and making education more inclusive across diverse contexts (Ahsan, 2025; Akefe et al., 2025). Yet, despite its promise, there remains a limited literature synthesis addressing equitable access within AI-enabled medical education. This scoping review seeks to map existing approaches, therefore, identify gaps, and generate insights that can guide equitable implementation across a range of educational, cultural, and geographic settings.
2. Methods
This paper followed the Joanna Briggs Institute scoping review methodology, which is grounded in the foundational work of Askey and O'Malley (2005) and subsequently refined by Levac et al. (2010) (Arksey and O'Malley, 2005; Levac et al., 2010; M. Peters et al., 2024). The methodology comprised a five-step process (1) identifying the research question, (2) identifying relevant studies, (3) study selection, (4) charting the data and (5) collating, summarising and reporting the results (Arksey and O'Malley, 2005; Levac et al., 2010; M. Peters et al., 2024).
2.1 Identifying the research question
The scoping review aimed to capture a breadth of literature; hence, the broad question “what is known regarding equitable access to AI and medical education” was used to develop the search strategy.
The eligibility/inclusion criteria were defined using the PCC (Population, Concept, Context) framework, as recommended for scoping reviews (Aromataris and Munn, 2019).
Population: Medical undergraduate or medical postgraduate degree (pre-professional registration).
Concept: Equitable access to AI in medical education.
Context: Any setting related to tertiary or higher education institutions globally, providing training to become a medical practitioner (pre-professional registration).
2.2 Identifying the relevant studies
A comprehensive search strategy was developed to identify relevant peer-reviewed literature. The final search was conducted on August 27, 2025, across four major electronic databases: MEDLINE, Evidence-Based Medicine Reviews–Health Technology Assessment, Health and Psychosocial Instruments and Global Health. These databases were selected for their broad coverage of medical education literature.
The search strategy included a combination of controlled vocabulary (e.g. MeSH terms) and free-text terms. Keywords included, but were not limited to:
“Equity”
“Access”
“Medical Education”
“Artificial Intelligence”
Boolean operators (AND, OR) were used to combine search terms, and filters were applied to limit results to English-language publications. No date restrictions were applied to ensure the inclusivity of older foundational studies, as well as recent developments.
The inclusion and exclusion criteria used for study selection are summarised in Table 1.
The full search strategy, including specific database queries, is available in the Supplementary Material ( Appendix).
2.3 Study selection
All search results were exported into a reference management software (EndNote) (Clarivate, 2023), duplicates were removed and thereafter exported to Covidence (Veritas Health Innovation, 2021). Title and abstract screening were conducted independently by two reviewers (JB and NW) using the predefined eligibility criteria (Table 1). If consensus could not be reached at this stage, the full text was reviewed.
For studies deemed potentially relevant, the full texts were retrieved and assessed for inclusion (JB, NW, CC, CM, JM, EM). Discrepancies during the full-text screening stage were resolved through discussion with a third reviewer (JB, JM, CM), who provided an independent opinion to ensure consistency and reduce bias.
The Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) flow diagram illustrates the study selection process, including the number of records identified, screened, included, and excluded, along with the reasons for exclusion (Tricco et al., 2018) (Figure 1).
2.4 Data charting and extraction
A standardised data charting form was developed and piloted across a sample of five studies to ensure clarity and relevance. The charting process was iterative and flexible, allowing for the refinement of data items as familiarity with the literature deepened. The authors collaboratively reviewed and agreed upon the final data extraction table.
The following data were extracted from each study: Bibliographic information, study design and methodology, geographic location and institutional context, study population characteristics, type of AI, and context of AI and equity in medical education.
All data were independently extracted by one reviewer (JB) and cross-verified for accuracy by other reviewers (JM, NW). Any inconsistencies were resolved through consensus or consultation with a third reviewer.
2.5 Collating, summarising, and reporting the results
Quantitative and qualitative analyses were undertaken to collate and map the data. Descriptive statistics were used to collate the quantitative data. Qualitative thematic analysis was employed to develop relevant themes to address the objective of the scoping review. Notable patterns, facilitators, barriers, and research gaps were highlighted to inform future research directions and policy implications. Given the scoping nature of this review, no formal quality appraisal or risk of bias assessment was conducted (Peters et al., 2024).
3. Results
The search results yielded 256 studies for full-text screening (August 2025), with one duplicate removed (Figure 1). After abstract and title screening, 80 studies remained for full-text review. Following further evaluation against the inclusion and exclusion criteria, 45 studies were excluded. The primary reason for exclusion was the absence of a focus on equity (n = 23). Other reasons included ineligible population group (n = 7), ineligible study design (n = 5), ineligible population setting (n = 3), outcomes did not align with the focus (n = 3), was not original research (n = 3), and intervention was outside the inclusion criteria (n = 1) (Figure 1).
3.1 Study characteristics
Table 2 presents a quantitative description of the included studies. Of the studies reviewed, 97% (n = 34) were published between 2022 and 2025, with 51% (n = 18) in 2025. The studies originated from 16 countries, with the largest proportion (40%, n = 14) from the United States of America (USA). India contributed the second-highest number of studies (14%, n = 5). Notably, 66% of the countries were represented by a single study (Figure 2).
A variety of study methodologies were employed, with 29% (n = 10) case-based or educational intervention studies. The next most common methodology was comparative performance/observational studies (23%, n = 8), followed by commentary/perspective pieces (20%, n = 7). Most studies (54%, n = 19) involved participant-based research, with “medical students only” being the most studied group (37%, n = 13).
The types of AI investigated were categorised into eight categories. As individual studies sometimes referenced more than one type of AI, a total of 48 instances of AI types were identified. Large Language Models (LLM), particularly iterations of ChatGPT, were the most common category (48%, n = 23) followed by Machine Learning (15%, n = 7).
The areas of medical education 21’ studies most frequently addressed were curriculum-related topics (49%, n = 17), followed by general or non-specific themes (23%, n = 8), and distributed learning (14%, n = 5) (Table 2). Notably, only one study included an equity-related term (socio-economic disparities) in the study keywords (Monzon and Hays, 2025), and one included an equity-related term in the article title (disparities) (Mishra et al., 2025) (Table 3).
3.2 Narrative analysis
Three overarching equity-related themes were identified across the included studies: (1) equitable access, (2) bias and (3) AI literacy and education, with many studies addressing multiple themes (Table 3, Figure 3). Within equitable access, sub-themes related to demographics (e.g. socioeconomic status, gender, culturally and linguistically diverse (CALD)) and geography (e.g. low resource settings, urban–rural disparities). Within bias, sub-themes centred on algorithmic bias (e.g. cultural appropriateness, under-representation of minority groups). AI literacy and education were underpinned by curriculum development. Across all themes, the literature highlighted facilitators and barriers, along with recommendations to support more equitable integration of AI into medical education.
3.3 Equitable access
Access through an equity lens was described as both a facilitator and a barrier to democratising medical education globally, with concepts of demographic, geographic, and distributed learning often overlapping.
3.3.1 Demographic access
Demographic access included gender, CALD and socio-economic measures such as income. Studies that focused on gender reported contrasting results (Ahmed et al., 2025; Jebreen et al., 2024; Mashburn et al., 2025; Sunmboye et al., 2025). Two studies conducted in low- and middle-income countries (LMICs) reported gender-related disparities, with female students demonstrating lower levels of knowledge and access to LLMs and reporting less favourable perceptions of AI compared to their male counterparts (Ahmed et al., 2025; Jebreen et al., 2024). In contrast, a German study found females had higher self-rated confidence after using an LLM compared to males and employed different problem-solving techniques, e.g. more fact checking of the LLM (Mashburn et al., 2025). This latter finding was reinforced by a United Kingdom study that found female students exhibited a higher level of concern regarding inaccuracies in AI-generated content, and therefore were more likely to engage in behaviours aimed at mitigating these risks (Sunmboye et al., 2025).
Access to AI for CALD populations was framed in two ways: by its ability to bridge existing educational divides, but also as a source of concern that it could exacerbate existing access inequities. One study stated English literacy as a major barrier, noting that English is the de facto standard academic language globally (Biri et al., 2023). However, another study found that Generative AI facilitated more equitable access by successfully translating medical education resources into users' first language (Shimizu et al., 2023). Despite this, other studies raised equity concerns due to LLMs lower accuracy on non-English examinations, reflecting their English-dominant training data and limited handling of linguistic nuances (Cherif et al., 2024; Meyer et al., 2024). AI has the potential to reduce medical education disparities for CALD populations, provided that issues of language accessibility and appropriateness are addressed and collaborative contextual adaptation of AI tools occurs (Jagannath et al., 2025; Spallek et al., 2023). For example, Spallek et al. (2023) highlight that when LLM educational resources are specifically developed for and co-designed with specific population groups, such as Aboriginal and Torres Strait Islander students in Australia, they have a higher acceptability and ability to meet their needs (Spallek et al., 2023).
Costs surrounding AI were another conflicting concept (Ahmed et al., 2025; Divito et al., 2024; Rueff et al., 2025; Scherr et al., 2023). Medical students from low-socioeconomic backgrounds were found to be less likely to use LLMs (Ahmed et al., 2025), with one study addressing this issue by providing funding so all students had equitable access to a subscription version of an LLM (Rueff et al., 2025). It was noted that while some versions of LLM were cost-free, therefore more accessible, there were subscription versions of LLMs, which provide higher usage limits, and more sophisticated functionality. This may extend the divide amongst students based on their socio-economic background (Divito et al., 2024; Scherr et al., 2023).
3.3.2 Geographic access
Studies highlighted the presence of a medical education divide both between and within countries, with AI seen as a facilitator and barrier to the global standardisation of medical education (Biri et al., 2023; Divito et al., 2024; Jagannath et al., 2025; Ma et al., 2023; Mishra et al., 2025; Mondal, 2025; Noel, 2024; Pu et al., 2025; Theros et al., 2025). Studies used terms such as developing countries (Biri et al., 2023; Pu et al., 2025), underdeveloped regions (Ma et al., 2023), low-resource settings (Divito et al., 2024; Jagannath et al., 2025; Mishra et al., 2025; Mondal et al., 2024; Noel, 2024; Theros et al., 2025) and rural areas (Ahmed et al., 2025; Bray et al., 2025) when describing geographic settings of interest. Collaboration and sharing of AI platforms between institutions with formal partnerships or open-access globally were deemed to be an important facilitator to bridging this divide (Jagannath et al., 2025; Theros et al., 2025; Yli-Hallila et al., 2025). For example, one study described a virtual microscopy platform that was developed collaboratively and shared between institutions, providing an accessible standardised resource that had functionality for context-specific adaptation (Yli-Hallila et al., 2025).
While the accessibility of AI tools was generally seen as positive, due to their low cost, availability of internet access, and the abundance of available information, this was not universal across geographic regions (Biri et al., 2023; Divito et al., 2024; Jebreen et al., 2024; Kıyak and Kononowicz, 2024; Ma et al., 2023; Mishra et al., 2025; Mondal et al., 2024; Noel, 2024; Pu et al., 2025; Theros et al., 2025; Yli-Hallila et al., 2025). For example, Jebreen et al. (2024) surveyed Palestinian medical students' perceptions of AI in the curriculum, finding strong support for its inclusion but noting major barriers to its usage, due to limited access to technology, the internet, and electricity (Jebreen et al., 2024).
Distributed learning was categorised by hub and spoke models, where AI-enabled centralised teaching was accessed by remote learners (Ahmed et al., 2025; Bray et al., 2025; Caudell et al., 2003; Gin et al., 2025; Mondal et al., 2024; Mukadam et al., 2025; Scherr et al., 2023; Spear et al., 2022). This model was particularly relevant in rural areas (Ahmed et al., 2025; Bray et al., 2025; Gin et al., 2025; Mondal et al., 2024), in geographically distributed partner institutions (Caudell et al., 2003; Yli-Hallila et al., 2025), and for locations without a simulation centre (Scherr et al., 2023). It also addressed unequal access to a clinical setting during the COVID-19 pandemic (Spear et al., 2022).
A common application of distributed learning was the use of virtual patient simulations (Ahmed et al., 2025; Bray et al., 2025; Caudell et al., 2003; Mukadam et al., 2025). These were described as an effective supplement to traditional educational methods when access to standardised patients was limited. This facilitated learner engagement with a diverse, yet contextually appropriate patient population and enabled them to practice clinical scenarios (Ahmed et al., 2025; Mukadam et al., 2025).
3.4 Biases and mitigation strategies
Concerns relating to AI perpetuating existing biases in medical education were commonly associated with LLMs exhibiting algorithmic bias (Abouammoh et al., 2025; Bray et al., 2025; Cherif et al., 2024; Gin et al., 2025; Kıyak and Kononowicz, 2024; Laverde et al., 2025; Ma et al., 2023; Mashburn et al., 2025; Meyer et al., 2024; Monzon and Hays, 2025; Robleto et al., 2024; Sauder et al., 2024; Scherr et al., 2023; Sharma et al., 2025; Smirnova et al., 2025; Spallek et al., 2023; Theros et al., 2025; Xu et al., 2024). The type of algorithmic bias was not always transparently described (Monzon and Hays, 2025; Sauder et al., 2024; Sharma et al., 2025; Theros et al., 2025; Xu et al., 2024). However, when specified, biases included cultural (Abouammoh et al., 2025; Laverde et al., 2025; Meyer et al., 2024; Robleto et al., 2024; Spallek et al., 2023), geographic (Bray et al., 2025; Cherif et al., 2024), gender (Laverde et al., 2025; Spallek et al., 2023), social (Laverde et al., 2025; Ma et al., 2023; Mashburn et al., 2025; Robleto et al., 2024; Scherr et al., 2023) and age and disability (Robleto et al., 2024).
In AI-enhanced medical education, biases may emerge in contexts such as using LLMs to generate multiple-choice questions (Kıyak and Kononowicz, 2024), develop virtual patient chatbots (Laverde et al., 2025) and student self-directed learning (Meyer et al., 2024; Spallek et al., 2023). In studies that surveyed medical students about their perceptions of AI, concerns about the existence of biases were common, although detailed explanations were rarely provided (Abouammoh et al., 2025; Cherif et al., 2024; Mashburn et al., 2025; Robleto et al., 2024; Sharma et al., 2025; Xu et al., 2024).
Studies included recommendations on strategies to mitigate biases; these included raising students and educators awareness of the existence of algorithm bias (Mashburn et al., 2025; Sauder et al., 2024; Scherr et al., 2023; Sharma et al., 2025; Xu et al., 2024), collaboration between stakeholders and end users in model development (Gin et al., 2025; Monzon and Hays, 2025; Spallek et al., 2023), the implementation of monitoring and critical evaluation frameworks by universities (Gin et al., 2025; Robleto et al., 2024; Smirnova et al., 2025; Theros et al., 2025) and the development of context-specific Generative AI (Bray et al., 2025; Cherif et al., 2024; Kıyak and Kononowicz, 2024; Spallek et al., 2023). Addressing the latter strategy, Cherif et al. (2024) noted that the prevalence of Western-centric internet may not be representative of slight variations in clinical presentations and disease epidemiology in certain contexts, like African and Asian populations (Cherif et al., 2024).
3.4.1 AI literacy and education
Several studies recommended the critical need to improve AI literacy via education for both students and faculty (Ahmed et al., 2025; Gin et al., 2025; Jagannath et al., 2025; Sunmboye et al., 2025; Theros et al., 2025; Xu et al., 2024). Limited AI literacy was compounded by minimal exposure to AI within existing medical curricula (Jagannath et al., 2025; Ma et al., 2023; Theros et al., 2025). For example, a multi-institutional survey of medical students found that none of the participating universities had a formal AI education program (Jebreen et al., 2024). Another study found regional disparities, with students in lower resource settings less likely to have undertaken AI-related coursework, prompting authors to recommend a national initiative in AI curriculum design (Ma et al., 2023). Prior exposure to AI education appeared to have a protective effect against AI shortcomings, with these students more likely to critically review and verify AI-generated information (Xu et al., 2024).
Recommendations for integrating AI into curricula emphasised the importance of context-specific approaches that reflect population-specific AI usage (Abouammoh et al., 2025; Bray et al., 2025; Mashburn et al., 2025; Mondal, 2025). In low-resource settings, it was noted that, additional resources and infrastructure are required to support this (Ahmed et al., 2025; Jebreen et al., 2024). The breadth of AI literacy recommended to address equity concerns ranges from awareness of how LLMs generate information to the need to develop and teach skills such as computer coding (Ahmed et al., 2025; Gin et al., 2025).
In addition to curriculum design, studies also recommended that university AI policies include equity guidelines (Monzon and Hays, 2025; Rueff et al., 2025). To keep pace with the rapid evolution of AI, Rueff et al. (2025) presented a case study of a step-wise process of integrating national AI policy guidelines into the curriculum (Rueff et al., 2025). Faculty AI literacy was identified as critical, with recommendations including faculty supervision of AI models (Robleto et al., 2024) and faculty-led teaching on appropriate AI use (Sauder et al., 2024). Shimuzu et al. (2023) proposed a curriculum framework involving collaboration between students and faculty, highlighting the need to learn alongside Generative AI. They suggest that this would require a shift in cognitive domains from “know and understand” to “apply and analyse” (Shimizu et al., 2023).
4. Discussion
The scoping review identified three overarching and intersecting equity themes across the literature: equitable access, bias, and AI literacy and education. Collectively, these findings illustrate a paradox: while AI offers unprecedented potential to democratise access, addressing long-standing barriers that have historically limited inclusivity and equity within medical education, it simultaneously risks reinforcing structural inequities embedded within educational and technological systems. This duality underscores the need for AI literacy and education that is contextual and equity-centred (Varsik and Vosberg, 2024). This pattern reflects the inverse equity hypothesis, whereby newly introduced innovations are initially adopted by wealthier or better-resourced groups, who typically have the least need, before eventually diffusing to more disadvantaged populations (Victora et al., 2000).
Consistent with this pattern, AI was commonly described as a mechanism to enhance access to learning and information across geographic boundaries, yet its benefits were often contingent on resource availability and contextualisation (Ahmed et al., 2025; Biri et al., 2023; Bray et al., 2025; Caudell et al., 2003; Divito et al., 2024; Gin et al., 2025; Jagannath et al., 2025; Jebreen et al., 2024; Kıyak and Kononowicz, 2024; Ma et al., 2023; Mishra et al., 2025; Mondal et al., 2024; Mukadam et al., 2025; Noel, 2024; Pu et al., 2025; Scherr et al., 2023; Spear et al., 2022; Theros et al., 2025; Yli-Hallila et al., 2025). This tension is particularly relevant for LMICs, where pre-existing inequities have been exacerbated by the delayed adoption and more limited integration of AI-enabled medical education compared to high-income countries (Ahsan, 2025; Naseer et al., 2025). In these contexts, limitations in digital infrastructure and local contextualisation are compounded by institutional readiness challenges, alongside the dominance of Western-centric internet resources and English as the de facto language of medicine (Ahsan, 2025; Jagannath et al., 2025). For example, an asynchronous AI course for medical students developed in the USA, with the explicit intention of improving access for learners from LMICs, was instead found to disadvantage non-English speakers, highlighting the potential of an unintended linguistic bias (Jagannath et al., 2025).
Taken together, these factors illustrate how well-intentioned digital education initiatives may inadvertently reproduce or exacerbate existing inequities. Addressing these challenges in LMICs requires not only alternative access models such as partnerships, targeted grants, and sustainable funding arrangements but also deliberate design choices that embed cultural competency and accessibility features from the outset (Jagannath et al., 2025; Kıyak and Kononowicz, 2024). In particular, collaborations between well-resourced and under-resourced institutions offer a pathway to co-develop AI tools that are locally adaptable, affordable, and contextually appropriate for LMIC settings (Kıyak and Kononowicz, 2024; Yli-Hallila et al., 2025). The predominance of studies from high-income countries in this scoping review further highlights the need for expanded research in LMIC contexts to support evidence-informed and contextually appropriate practice.
Moreover, despite the rapid growth of AI in medical education, equity-focused research remains peripheral within the literature. The scoping review confirms that medical education is positioned at a pivotal juncture, shaped by converging imperatives: the pursuit of greater inclusivity and the rapid evolution of AI (Akefe et al., 2025). The momentum of AI in this space is evident in publication trends, with over half (51%) of the included studies published in 2025. Notably, only one study incorporated an equity-related keyword (Monzon and Hays, 2025), and another referenced an equity term in the title (Mishra et al., 2025). Many studies instead examined AI integration more broadly, often referring to “underrepresented” students collectively or listing multiple groups without a focus on any specific equity population (Ahmed et al., 2025; Gin et al., 2025; Meyer et al., 2024). This lack of specificity limits the ability to generate targeted, actionable insights for particular equity populations.
Globally, there is no universal list of equity groups in medical education, as classifications vary by country and context (Jones et al., 2019; Lv et al., 2025). Equity groups of focus in the scoping review included females, CALD populations, students from low socio-economic backgrounds, rural background students, and First Nations students (Ahmed et al., 2025; Biri et al., 2023; Cherif et al., 2024; Jagannath et al., 2025; Jebreen et al., 2024; Kıyak and Kononowicz, 2024; Mashburn et al., 2025; Meyer et al., 2024; Mondal et al., 2024; Rueff et al., 2025; Scherr et al., 2023; Shimizu et al., 2023; Spallek et al., 2023; Sunmboye et al., 2025). Only one paper briefly mentioned students with a disability as an equity group (Robleto et al., 2024), and another explicitly excluded participants unable to use the provided AI hardware due to physical disability (Mukadam et al., 2025). This omission is notable given the potential of AI to support learning for students with disabilities, provided tools are designed for inclusivity (Varsik and Vosberg, 2024; Zhao et al., 2025). Another equity group notably absent from the reviewed literature were LGBTQIA + students, despite their recognition as a priority equity group in national medical education standards (Australian Medical Council, 2023). As with other underrepresented cohorts, LGBTQIA + learners face structural barriers within medical education (Wong et al., 2024). Moreover, heteronormative assumptions are deeply embedded in the language and data used to train AI models, risking the perpetuation of bias and misrepresentation (Wong et al., 2024).
Concerns regarding algorithmic bias were common across studies, with particular emphasis on how LLMs reflect and reinforce the assumptions embedded within their training data. Most AI platforms are trained before public release and do not update continuously, therefore any biases embedded in the training data will persist unless addressed by retraining or fine-tuning (Bhargava and Singhal, 2024; Qiu et al., 2025). At a fundamental level of AI literacy, students and faculty require education on the principles and mechanisms underpinning the development of LLMs (Ahmed et al., 2025; Sauder et al., 2024; Sunmboye et al., 2025).
AI literacy in medical education has been described as minimal, with both students and faculty requiring further support (Jebreen et al., 2024; Theros et al., 2025). Without adequate literacy, there is a risk of overreliance on AI tools and a lack of critical evaluation (Bearman and Ajjawi, 2025; Theros et al., 2025). Given the rapid pace of AI integration, medical schools need to prioritise embedding AI literacy into their curricula to ensure informed and equitable use. A proactive approach is warranted, as accreditation bodies internationally are increasingly recognising digital competence as an essential component of medical education and are likely to formalise such expectations in future standards (Australian Medical Council and Australian Digital Health Agency, 2021; Sapci and Sapci, 2020; Schubert et al., 2025).
Beyond literacy and education, the co-creation of contextually appropriate AI tools with students, faculty and priority equity groups is essential (Kıyak and Kononowicz, 2024; Yli-Hallila et al., 2025). Effective partnerships should encompass data governance, tool development, and ongoing monitoring to ensure inclusion and prevent inequitable outcomes (Jones et al., 2019; Spallek et al., 2023). Further research is needed to obtain contextually translatable evidence on how AI-enabled equitable tools can be co-developed and implemented effectively.
While much of the literature on geographic inequity is focused on LMICs, geographic disparities also persist within high-income countries (Bray et al., 2025; Caudell et al., 2003; Mondal et al., 2024). In this context, AI-enabled distributed learning has been shown to expand access for rural students, enabling them to remain in their communities without compromising educational quality (Bray et al., 2025; Caudell et al., 2003; Mondal et al., 2024; Mukadam et al., 2025). For instance, Bray et al. (2025) proposed that AI-powered tools, including virtual patient simulations, personalised learning platforms, clinical decision support systems, and automated performance evaluations, could transform rural medical education by democratising access to resources, standardising learning across regions, and providing tailored learning pathways (Bray et al., 2025).
These innovations are significant given the maldistribution of the medical workforce, which affects countries regardless of income level (World Health Organization, 2021). Rural communities consistently have experienced fewer doctors-to-population ratios compared to metropolitan areas (Fuller et al., 2025; Medical Deans Australia and New Zealand, 2024; World Health Organization, 2021). One key strategy to address this is to provide end-to-end medical training for students from rural backgrounds within their own communities (Fuller et al., 2025). Until recently, structural barriers, such as centralised pre-clinical education in metropolitan centres, have hindered this approach (Fuller et al., 2025; McGrail et al., 2023). In this context, AI-enabled distributed learning offers a potential mechanism to overcome these barriers, supporting more equitable, place-based training models without compromising educational quality.
5. Limitations
As scoping reviews map and collate literature rather than critically appraise study quality, the included studies may have varied in research rigour (Peters et al., 2015). Furthermore, as included studies encompassed a broad range of methodologies, AI modalities, contexts and population groups, the ability to synthesise findings and draw generalisable conclusions may have been compromised. Restricting the search to English-language peer-reviewed literature may have excluded relevant studies in other languages, particularly from non-English speaking LMICs, potentially underrepresenting global perspectives and indicative of the high percentage of studies from high-income countries.
The scoping review relied on pre-defined search terms such as “equity” and “access,” but not all studies that address disparities or inclusivity explicitly used these terms. Consequently, some relevant studies may have been missed, and “access” alone does not necessarily indicate an equity focus. This limitation highlights the challenge of capturing the full breadth of literature on equity in AI-enabled medical education, particularly when terminology is inconsistent or context-specific.
Additionally, the rapid evolution of AI technologies and their integration into medical education means that some included studies may already be outdated, limiting the currency of the review findings. The COVID-19 pandemic also accelerated the adoption of digital and AI-enabled learning, and studies conducted before or during the pandemic may not fully capture post-pandemic educational practices and equity challenges. These factors should be considered when interpreting the generalisability of the findings.
6. Conclusion
AI presents unprecedented opportunities to widen access and participation in medicine for underrepresented groups and standardise medical education globally. Realising these benefits requires AI literacy, contextually relevant tools, and deliberate strategies to mitigate bias. Medical schools should prioritise embedding AI literacy and education into the curriculum for students and faculty in anticipation of international accreditation standards increasingly mandating AI competence. Future research should focus on developing and evaluating AI interventions that are inclusive, contextually appropriate, and widely accessible.Table A1
We acknowledge the contributions of the Harnessing AI and Technology for Equitable and Ethical Medical Education project team, including John Turchini, Curtis Lee, and Lachlan McOmish, for their contributions to the development and implementation of the project, which provided the foundation for this work.
