Integration of Artificial Intelligence into Medical Education: A Mixed-Methods Study Identifying the Risks, Safeguards, and Optimal Approaches to Protect Clinical Skill Development across West African Countries

Background Artificial intelligence (AI) offers transformative potential for medical education through personalised learning, simulation, and decision support, especially in resource-limited settings like West Africa. However, there are concerns regarding over-reliance on AI, potentially undermining core clinical skills such as diagnostic reasoning, physical examination, and independent judgement. Empirical evidence on risks, safeguards, and optimal integration strategies in low-resource African contexts remains scarce. Methods This mixed-methods study targeted physicians affiliated with the West African College of Physicians. A structured online survey (n = 136) assessed AI literacy, perceived risks to clinical skill development, attitudes toward AI, and understanding of AI limitations using Likert scales and knowledge items. Purposive key informant interviews (n = 72) with physician educators, programme directors, clinicians, technical experts, and policymakers explored experiences, concerns, safeguards, and curriculum recommendations. Quantitative data were analysed descriptively on SPSS version 29.0; and qualitative data underwent thematic framework analysis in NVivo. Findings Survey respondents were predominantly Nigerian (84.6%), balanced by gender (51.5% female), with diverse clinical experience. AI literacy was low: 32.4% lacked basic familiarity, and 67.6% lacked confidence in assessing AI outputs. Major gaps included unawareness of algorithmic bias (41.9%) and population variability in AI performance (36.8%). Most (87.5%) rejected uncritical acceptance of AI recommendations. High concern existed about threats to autonomy (77.9%), weakened diagnostic skills (71.3%), and deskilling of younger clinicians (68.4%). However, 94.1% expressed willingness to use AI if proven effective and safe. Qualitative themes emphasised systemic training gaps, the need for “informed sceptics”, longitudinal curriculum integration, hands-on experience, and safeguards prioritising clinical reasoning over AI dependence. Conclusion West African clinicians show significant AI literacy deficits and apprehension about the erosion of clinical skills yet demonstrate conditional openness to AI. Integration requires foundational literacy, longitudinal embedding, practical training, ethical focus, and competency assessments that preserve independent judgement. Addressing faculty expertise and infrastructure barriers is essential to prepare competent, AI-literate physicians without compromising clinical proficiency.