Received: September 3, 2026
Accepted for publication: September 22, 2026
Published online: September 30, 2026
UDC: 378.147:004.8:616.2
DOI: 10.26212/2227-1937.2026.87.69.003
GENERATIVE ARTIFICIAL INTELLIGENCE IN TUBERCULOSIS AND RESPIRATORY MEDICINE EDUCATION: EFFECTS ON CLINICAL REASONING, RISKS, AND EDUCATIONAL STRATEGIES (LITERATURE REVIEW)
Izmailova S.Kh.¹, Kassenov B.Zh.¹, Zhumakazhi M.B.¹
¹ Asfendiyarov Kazakh National Medical University, Almaty, Kazakhstan
Introduction. Generative artificial intelligence (GenAI) is rapidly being integrated into medical education; however, its impact on independent clinical reasoning remains uncertain. This issue is particularly important in training in tuberculosis and respiratory medicine, where diagnostic and therapeutic decisions require the integration of clinical, microbiological, molecular genetic, functional, and imaging data, and where erroneous conclusions may have serious consequences.
Objective. To synthesize current evidence on the impact of GenAI on the development of clinical reasoning and to identify conditions for its safe educational use in the training of specialists in tuberculosis and respiratory medicine.
Materials and methods. A structured narrative review of the literature was conducted. PubMed, Scopus, and Web of Science were searched primarily for publications from 2020 to 2026; earlier seminal studies were also considered. Systematic reviews, meta-analyses, randomized and quasi-experimental studies, professional recommendations, and domain-specific studies on the use of large language models in tuberculosis and respiratory medicine were analyzed. Educational effects, risks of cognitive offloading, and errors in high-stakes clinical tasks were assessed separately.
Results. Meta-analytic evidence suggests possible short-term improvements in knowledge, selected practical skills, and learning satisfaction with GenAI use; however, the certainty of evidence remains heterogeneous. Individual studies demonstrate improvements in clinical and critical reasoning when GenAI is used in a structured and educator-supervised manner, whereas uncontrolled reliance on models is associated with risks of cognitive offloading, automation bias, and reduced independent analytical work. In tuberculosis and respiratory medicine, GenAI may be used for clinical case analysis, generation of differential diagnoses, and training in the interpretation of functional studies; however, mandatory verification is required when dealing with Xpert MTB/RIF, LF-LAM, drug resistance, HRCT, and antituberculosis treatment regimens.
Discussion. Evidence-based use of GenAI in tuberculosis and respiratory medicine education should be viewed not as a replacement for clinical reasoning, but as a supervised pedagogical scaffold. A three-step model is proposed: independent clinical hypothesis → comparison with the GenAI output → mandatory external verification against patient data, WHO recommendations, and national clinical protocols.
Conclusion. GenAI can enhance training in tuberculosis and respiratory medicine only when an initial independent clinical decision is preserved and the model output is systematically verified. A hybrid format combining AI-enabled and AI-free tasks, trust calibration training, and verification of clinical decisions against current guidelines appears to be the most appropriate approach.
Keywords: generative artificial intelligence; large language models; clinical reasoning; medical education; tuberculosis; respiratory medicine; pulmonology; drug resistance; artificial intelligence.
