Who Uses AI in Medical Training? New Study Reveals a Sociodemographic Divide

Artificial intelligence is sweeping through hospitals, laboratories, and lecture halls at a pace that has left many educators scrambling to keep up. Yet while headlines celebrate AI’s potential to transform medicine, far less attention has been paid to a deceptively simple question: which students are actually engaging with these tools, and which are being left behind? A new cross-sectional study from Ecuador offers one of the most detailed answers yet for Latin America, and its findings reveal a portrait of AI adoption in health sciences education that is anything but uniform.

Researchers at Universidad Técnica del Norte in Ibarra surveyed 668 undergraduate students drawn from four health sciences programs at the public university. Their goal was to map how sociodemographic factors—sex, age, academic program, and ethnic self-identification—shape the way future health professionals perceive, use, and understand artificial intelligence. The results, published in BMC Medical Education, suggest that engagement with AI is not a single undivided attitude but a multidimensional phenomenon, with different student groups excelling, lagging, or diverging across distinct dimensions of AI interaction.

The methodological architecture of the study is worth examining, because it reflects the growing sophistication of educational measurement research. The team administered a 31-item Likert-type questionnaire, in which respondents rated their agreement with statements on graduated scales. Before analyzing group differences, the researchers subjected the instrument to exploratory factor analysis, a statistical technique that identifies latent constructs underlying patterns of responses. The analysis distilled the questionnaire to a final 30-item structure organized around three coherent dimensions: Critical Perceptions of AI in Health Sciences Education, Practical Use of AI Tools, and AI Knowledge and Readiness. Each dimension demonstrated high internal consistency, with Cronbach’s alpha values ranging from 0.905 to 0.911—figures that indicate the items within each scale measure the same underlying construct with remarkable reliability.

With the measurement framework established, the researchers deployed a battery of statistical approaches. Because Likert-scale data are ordinal rather than strictly continuous, they applied non-parametric tests alongside adjusted general linear models, which allowed them to estimate the independent effect of each sociodemographic variable while controlling for the others. They also employed Multiple Correspondence Analysis, a technique that visualizes associations among categorical variables in a low-dimensional space, revealing clusters and patterns that traditional hypothesis tests can obscure. This triangulation of methods strengthens confidence that the reported differences are not artifacts of a single analytical choice.

The headline finding concerns sex. After adjustment, sex was significantly associated with two of the three dimensions: Critical Perceptions (p = 0.013) and Practical Use (p = 0.043). Female students scored higher on critical perceptions—meaning they reported more nuanced, evaluative attitudes toward AI’s role in health sciences education—while male students reported higher practical use of AI tools. In other words, women in the sample were more likely to think carefully and skeptically about AI, while men were more likely to roll up their sleeves and experiment with it. The researchers are careful not to overinterpret this pattern, but it echoes broader literature on gendered technology adoption, in which early hands-on engagement with emerging technologies has often skewed male even as attitudes and confidence diverge in more complicated ways.

Academic program emerged as an even stronger predictor. The program a student was enrolled in was significantly associated with both Practical Use of AI Tools (p < 0.001) and AI Knowledge and Readiness (p = 0.003). This is perhaps unsurprising—students training in fields with heavier computational or diagnostic components may encounter AI differently than those focused on, say, nursing or public health—but its magnitude underscores a structural point: AI literacy in health professions education is currently being shaped by curricular silos rather than by a shared institutional strategy. A student’s exposure to and comfort with these tools depends substantially on which faculty corridor they happen to study in.

Ethnic self-identification also mattered, showing a significant association with AI Knowledge and Readiness (p = 0.029). In a country as ethnically diverse as Ecuador, where Indigenous and Afro-Ecuadorian communities have historically faced unequal access to educational and technological resources, this finding carries particular weight. It suggests that the digital divide surrounding AI may intersect with longer-standing patterns of social inequality, meaning that efforts to democratize AI education cannot be culturally or demographically blind. Notably, age was not independently associated with any of the three dimensions, defying the common stereotype that older students are uniformly slower to embrace new technologies.

One of the most scientifically honest aspects of the study is its treatment of effect sizes. The researchers report that the significant adjusted effects were small, with partial eta-squared values ranging from 0.006 to 0.028. In plain terms, while sex, program, and ethnicity are statistically reliable predictors of AI engagement differences, they explain only a modest fraction of the overall variation among students. This is an important corrective to sensational readings of the data: the story is not that women cannot use AI or that certain programs produce technophobes. Rather, sociodemographic factors subtly tilt the landscape of engagement, and even small tilts can compound across hundreds of thousands of students entering health workforces worldwide.

The authors argue that their findings support a specific prescription: structured and equitable AI education woven across all health sciences curricula, rather than left to chance or concentrated in select programs. Such curricula, they contend, should simultaneously strengthen practical skills, factual knowledge, critical appraisal capacity, and responsible use—a four-pillar approach that mirrors the study’s own dimensional structure. If AI competence is multidimensional, then educational interventions must be too, addressing not only how to operate a tool but when to trust it, when to question it, and how to deploy it ethically in patient care.

The broader stakes extend well beyond a single Ecuadorian university. Health systems across Latin America and the Global South are preparing to integrate AI-assisted diagnostics, triage algorithms, and predictive analytics, often without a clear picture of how equitably the incoming workforce is prepared for them. Studies like this one provide the granular, locally grounded evidence needed to design interventions that reach the students who most need them. As generative AI tools proliferate faster than curricula can adapt, the question is shifting from whether health professionals will use AI to whether they will use it well—and, crucially, whether that competence will be distributed fairly across sex, ethnicity, and discipline. The Ecuador data suggest that without deliberate institutional action, the answer to the second question may be no.

Subject of Research: Sociodemographic differences in health sciences students' engagement with artificial intelligence in Ecuador

Article Title: Sociodemographic differences in health sciences students’ engagement with artificial intelligence: a cross-sectional study

Article References: Sociodemographic differences in health sciences students’ engagement with artificial intelligence: a cross-sectional study. (n.d.). https://doi.org/10.1186/s12909-026-10393-6

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10393-6

Keywords: artificial intelligence, health sciences education, medical education, sociodemographic factors, AI literacy, higher education, Ecuador, university students, technology use, health equity, Sociodemographic, differences