Artificial intelligence knowledge, perceptions, and educational needs among undergraduate medical students in Oujda, Morocco: a cross-sectional study

¹Mohammed First University, Faculty of Medicine and Pharmacy, Oujda, Morocco, ²Mohammed V University, Faculty of Medicine and Pharmacy of Rabat, Rabat, Morocco, ³Hassan II University, Faculty of Medicine and Pharmacy of Casablanca, Casablanca, Morocco

^&Corresponding author
Imane Chenfouh, Mohammed First University, Faculty of Medicine and Pharmacy, Oujda, Morocco

Introduction    

Artificial intelligence (AI) is reshaping health care by supporting clinical decision-making, diagnostic workflows, biomedical research, patient monitoring, and health-system management. Its potential benefits include faster data processing, improved recognition of patterns in imaging and clinical records, support for drug development, and reduction of avoidable errors. However, these benefits depend on appropriate human oversight, data quality, algorithmic transparency, and ethical governance [1-3]. Medical students are future users, interpreters, and communicators of AI-enabled tools. Their readiness is therefore not limited to technical knowledge, it also includes understanding the clinical limits of AI, bias, accountability, patient autonomy, confidentiality, and the preservation of empathy in the physician-patient relationship. International studies have reported that students often recognize AI as useful but have limited formal training and variable confidence in applying it responsibly [4-6]. In Morocco, where national discussions about AI governance in health care are emerging, empirical data on students readiness for AI-enabled health care remain limited [7]. A recent Moroccan study explored perceptions among medical students in Agadir, but additional evidence from other regions and larger samples is needed to inform curriculum discussions in a context-sensitive way [8]. Oujda, located in eastern Morocco, provides a relevant setting because it represents a public medical school outside the largest national academic centers, within a nationally structured medical curriculum [9]. This study aimed to assess self-reported knowledge, perceptions, and educational needs regarding AI in medicine among undergraduate medical students at the Faculty of Medicine and Pharmacy of Oujda, Morocco.

Methods     

Study design and setting: a descriptive cross-sectional study was conducted at the Faculty of Medicine and Pharmacy of Oujda, Morocco, during the 2024-2025 academic year. Data were collected between February and June 2025. Reporting was guided by the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist.

Participants and sampling: the target population comprised all undergraduate medical students aged 18 years or older and enrolled at the Faculty of Medicine and Pharmacy of Oujda during the 2024-2025 academic year. At the time of recruitment, 2,740 students were enrolled and therefore potentially eligible. The Moroccan medical curriculum spans seven years, including preclinical years, years with combined theoretical teaching and partial clinical rotations, and final years dedicated to clinical clerkships [9]. For analysis, students were grouped as basic years (first to fifth years) and clinical years (sixth and seventh years). The survey used a non-probability online census invitation strategy. Invitations were distributed through official WhatsApp groups managed by student representatives, because these groups were the most practical way to reach students across all academic years.

Sample size: the minimum required sample size was calculated using Epi Info version 7.2 and the formula described by Charan and Biswas:

[10]. We used Z₁-α = 1.96 for a 95% confidence level, p = 0.50, and d = 0.05. After accounting for an anticipated 15% non-response rate, the minimum required sample size was 397 students. The final sample of 1,242 students exceeded this requirement.

Questionnaire development and measures

The questionnaire was adapted from previously published surveys on medical students views of AI in medicine, particularly the instrument used by McLennan et al. and related work on AI exposure in undergraduate medical education [11,12]. The instrument was not treated as a new psychometric scale. Items were selected to describe self-reported perceptions and educational needs rather than to generate a single latent construct score. The questionnaire was translated from English to French and back-translated to check semantic equivalence. The draft was reviewed by the research team for content relevance and face validity in the Moroccan medical education context, then pilot-tested with 20 students to assess clarity and completion. Minor linguistic changes were made after piloting, and pilot responses were excluded from the final analysis. The final questionnaire included four sections: demographic and academic characteristics, perceived AI knowledge and usefulness in medicine, perceived advantages and disadvantages of AI in medicine, and perceptions of AI education. Items used a five-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree) (Annex 1).

Operational definitions

In this study, "perceived digital competence" referred to agreement with the item "Overall, I have good digital skills and competences." "Self-reported high technological literacy" referred to a yes response to the demographic item asking whether students had good experience with technology or a high degree of technological literacy. "AI familiarity" referred to agreement with the item "I feel well informed about the role of AI in medicine." These measures were self-reported and should not be interpreted as objective assessments of digital literacy or AI knowledge.

Data collection, cleaning, and missing data: data were collected using Google Forms. Participation was voluntary, and electronic informed consent was required before accessing the survey. To prevent missing data, all questionnaire items were configured as mandatory in Google Forms, therefore, a questionnaire could not be submitted unless every required item had been completed. The exported dataset was then screened according to a predefined cleaning protocol. The final analytical dataset included 1,242 complete responses, with no item-level missing data for any variable included in the analysis.

Statistical analysis

data were analyzed using IBM SPSS Statistics version 26.0. Quantitative variables were summarized using means and standard deviations (SD) or medians and interquartile ranges (IQR), as appropriate. Categorical variables were summarized using frequencies and percentages. Normality was assessed using the Shapiro-Wilk test. Because Likert-scale responses were ordinal and not normally distributed, Mann-Whitney U and Kruskal-Wallis tests were used for unadjusted bivariate comparisons. Statistical significance was set at P < .05, and exact P values are reported except when P < .001. The analysis was exploratory and descriptive. We therefore report unadjusted associations rather than independent determinants or predictors. No multivariable regression model was fitted.

Ethical considerations

This study was an anonymous, minimal-risk educational survey. Under local regulations, formal ethics committee approval was not required because no intervention was performed, no clinical or sensitive personal data were collected, and no identifiable information such as names, student identification numbers, IP addresses, telephone numbers, or email addresses was stored [13]. Participation was voluntary, electronic informed consent was obtained, and participants could stop completing the survey at any time before submission. Data were analyzed in aggregate and stored securely by the research team in accordance with data protection principles and the Declaration of Helsinki.

Results     

Participants and response rate

Of 2,740 eligible undergraduate medical students, 1,242 completed the questionnaire, corresponding to a response rate of 45.3%. Respondents had a mean age of 22.21 years (SD 2.37). Most respondents were female (805; 64.8%) and were enrolled in the basic years of study (724; 58.3%). A minority reported a background in mathematics, statistics, or computer science (275; 22.1%), self-reported high technological literacy (293; 23.6%), or having a close family member with an AI-related degree (187; 15.1%) (Table 1).

Perceived AI knowledge and usefulness in medicine

Students broadly perceived AI as relevant to medicine, with 1,130 respondents (91.0%) agreeing that AI has useful medical applications. However, AI-specific familiarity was limited: 218 students (17.5%) felt well informed about AI in medicine, despite 495 (39.9%) reporting good digital skills and competences. The most frequently endorsed areas of usefulness were drug research and development (1,084; 87.3%) and support for diagnosis (828; 66.7%). Support was lower for treatment decision support (661; 53.2%), personalized medicine (560; 45.1%), and direct assistance during treatment (516; 41.6%). Most students (908; 73.1%) disagreed that patients would independently manage their care through AI-based health applications (Figure 1).

Perceived advantages and disadvantages of AI in medicine

The most frequently recognized advantages were the ability to analyze large amounts of clinically relevant data (1,065; 85.7%) and continuous operation without fatigue (981; 79.0%). Students also identified potential for reducing medical errors (795; 64.0%), giving physicians more time for patient discussions and clinical examinations (802; 64.6%), improving cost efficiency (742; 59.7%), and supporting more accurate treatment decisions (522; 42.0%) (Figure 2). Students also reported substantial concerns. The most frequently endorsed disadvantage was AI inability to develop empathy and consider patients emotional well-being (970; 78.1%), followed by uncertainty about liability in case of error (967; 77.8%), development by programmers with limited medical experience (884; 71.1%), limited usefulness in unforeseen situations (858; 69.1%), lack of flexibility for individual patients (821; 66.1%), potential undermining of physician autonomy (739; 59.5%), amplification of bias and discrimination (676; 54.4%), and potential undermining of patient autonomy (547; 44.0%) (Figure 3).

Perceptions of AI education

Most respondents reported limited formal exposure to AI in medicine. A total of 771 students (62.1%) disagreed that they had received AI training through research or professional experiences, and 615 (49.5%) disagreed that their learning opportunities about AI had been adequate. Nevertheless, 593 (47.8%) reported independently educating themselves about AI. Interest in future training was high: 1,022 respondents (82.3%) considered it important to study AI in medicine more thoroughly, and 1,083 (87.2%) wished to learn more if given the opportunity. Nearly half (615; 49.5%) did not perceive limited programming or mathematics knowledge as a barrier to understanding AI in medicine (Figure 4).

Bivariate associations

Unadjusted bivariate analyses showed that students with a background in mathematics, statistics, or computer science reported higher scores for perceived AI usefulness in medicine (P = .027) and digital skills (P < .001). Self-reported high technological literacy was also associated with higher perceived AI usefulness (P < .001) and digital skills (P < .001). Gender, academic level, and having a close family member with an AI-related degree were not significantly associated with the three AI knowledge and digital competence items examined in Table 2. These findings should be interpreted as unadjusted associations, not independent predictors.

Discussion     

This study assessed self-reported AI knowledge, perceptions, and educational needs among undergraduate medical students at the Faculty of Medicine and Pharmacy of Oujda. The findings show a clear mismatch between perceived relevance and preparedness: most students considered AI useful for medicine, but only a minority felt well informed about AI-specific medical applications. Respondents recognised potential benefits in clinical data analysis, drug research and development, diagnostic support, continuous availability, and error reduction, while expressing concerns about empathy, liability, bias, flexibility, autonomy, and the need for professional oversight. The strong interest in further training suggests an educational gap rather than resistance to AI. Because this was a single-centre study using non-probability online recruitment, the findings should be interpreted as local exploratory evidence rather than nationally representative estimates.

The finding that students value AI while reporting limited AI-specific familiarity is consistent with emerging Moroccan and international evidence. Chakri et al. provided early Moroccan data showing that medical students recognised AI relevance while having variable knowledge and exposure [8]. McLennan et al. similarly reported that German medical students perceived AI as important but had limited AI-specific familiarity and raised concerns about professional responsibility and oversight [11]. Studies from Canada and Indonesia also described limited curricular exposure despite student interest in AI learning [12,14]. Research from Saudi Arabia, Pakistan, Nepal, Turkey, and a multinational Arab sample reported broadly positive attitudes toward AI alongside recurrent gaps in formal instruction and ethical understanding [15-20]. These comparisons require caution because the studies differed in sampling frames, recruitment modes, instruments, educational stages, and local health-system maturity. Nevertheless, the recurrent pattern across settings supports the relevance of AI literacy as an emerging priority in medical education.

The distinction between general digital competence and AI-specific readiness is important. In this study, more students reported good digital skills than felt well informed about AI in medicine. This suggests that everyday digital confidence does not necessarily translate into the ability to interpret AI outputs, recognise the limits of training data, assess algorithmic bias, or communicate uncertainty to patients. Similar gaps were reported by McLennan et al. and in the Canadian study by Pucchio et al. where students and educators identified the need for structured exposure to AI rather than reliance on informal self-learning [11,12]. The Indonesian and Turkish needs-assessment studies also emphasise that AI education should be deliberately planned, because student enthusiasm alone does not ensure competence [14,20].

Students' perceptions of AI usefulness were strongest in areas involving data-intensive, physician-mediated tasks. High endorsement of clinical data analysis, drug research and development, diagnostic support, and continuous availability is consistent with the role of AI as a tool for pattern recognition, information synthesis, and workflow support [1,11,17]. Conversely, support was lower when AI was positioned closer to direct treatment decisions or patient self-management, and most students rejected independent patient management through AI-based applications. This gradient is important: students did not appear to oppose AI in medicine, but they favoured a model in which AI augments rather than replaces professional judgment. Similar caution has been reported among postgraduate doctors in the United Kingdom, who recognised the potential of AI but emphasised supervision, clinical accountability, and preservation of human decision-making [16].

The ethical concerns reported by students are consistent with broader debates on trustworthy AI in health care. The most frequently endorsed disadvantage was AI's inability to develop empathy or consider patients' emotional well-being, followed by uncertainty about liability in the event of error. These findings align with studies showing that learners and trainees often view empathy, contextual reasoning, and responsibility as domains that should remain anchored in human clinical practice [11,16,17]. Concerns about algorithmic bias are particularly relevant because biased data and poorly validated models can reproduce or amplify inequities. The documented example of racial bias in a commercial health care risk-prediction algorithm illustrates that bias can occur even when race is not explicitly used as an input variable [21,22]. For Moroccan medical education, this means AI teaching should include not only the capabilities of AI systems but also their failure modes, including data quality, representativeness, local validation, explainability, accountability, and human oversight [1,6].

The high level of interest in AI education has practical curriculum implications, but these implications should remain proportional to the study design. The findings support stepwise integration of AI literacy into existing teaching at the Faculty of Medicine and Pharmacy of Oujda, rather than immediate claims for national curriculum reform. A locally feasible curriculum could begin with introductory concepts in AI and machine learning, followed by applied sessions on clinical use cases, critical appraisal of AI studies, patient communication, data privacy, bias, and medico-legal responsibility. This approach is consistent with educational literature emphasising that medical students do not need to become engineers, but they should be able to critically evaluate AI-assisted recommendations, understand when human oversight is required, and collaborate with data scientists and health informatics specialists [6,12,20].

The Moroccan context strengthens the relevance of this issue while also requiring careful interpretation. Medical education in Morocco follows a nationally structured curriculum, but schools may differ in local resources, digital infrastructure, faculty expertise, and opportunities for interdisciplinary teaching [21]. Evidence from Oujda therefore adds to emerging Moroccan data. Future multicenter studies across different public and private medical schools are needed.

This study has several strengths and limitations. Strengths include the large sample size, participation from students across academic stages, transparent reporting of the response rate, mandatory questionnaire completion resulting in no item-level missing data, and a STROBE-guided reporting structure. The study also addresses a timely educational question in an underrepresented Moroccan setting. Limitations include its single-centre design, non-probability WhatsApp-based recruitment, potential selection of more motivated or digitally engaged students, a 45.3% response rate that leaves possible participation bias, and reliance on self-reported perceptions rather than objective AI knowledge or performance measures. The adapted questionnaire was translated, reviewed, and pilot-tested, but it was not formally psychometrically validated in Moroccan medical students, and no Cronbach's alpha was reported. Finally, analyses were descriptive and bivariate, without adjustment for confounders; therefore, associations should not be interpreted as independent determinants. These limitations mean the findings should be read as exploratory evidence of perceived educational needs in one medical school, not as nationally representative or causal conclusions.

Conclusion     

In this single-center cross-sectional survey, undergraduate medical students in Oujda generally perceived AI as useful for medicine but reported limited AI-specific familiarity and insufficient learning opportunities. Their concerns about empathy, liability, bias, flexibility, and professional oversight indicate that AI education should address ethical and clinical judgment issues, not only technical content. These findings support locally grounded curriculum discussions at the Faculty of Medicine and Pharmacy of Oujda, while further multicenter studies using probability-based sampling, validated instruments, objective knowledge measures, and adjusted analyses are needed before drawing national conclusions.

Competing interests     

The authors declare no competing interests.

Authors' contributions     

Conception and study design, data collection, guarantor of the study: Imane Chenfouh. Methodology: Imane Chenfouh, Mohammad Abusheikha, Oumaima Benhadi and Imane Elomri. Data analysis and interpretation: Imane Chenfouh, Mohammad Abusheikha and Othmane El Fahd. Manuscript drafting: Imane Chenfouh and Oumaima Benhadi. Manuscript revision: Mohammad Abusheikha, Othmane El Fahd, Imane Elomri and Narjiss Aji. Supervision: Narjiss Aji. All authors read and approved the final version of the manuscript.

Tables and figures     

Table 1: demographic and academic characteristics of undergraduate medical student respondents (N = 1,242)
Table 2: unadjusted bivariate associations between participant characteristics and selected perceived AI knowledge/usefulness items
Figure 1: artificial intelligence knowledge and perceived usefulness in medicine
Figure 2: advantages of AI in medicine
Figure 3: disadvantages of AI in medicine
Figure 4: perceptions about the future of AI education in medicine

Annexes     

Annex 1: english version of the questionnaire

References