Artificial intelligence in medical diagnostics: A critical narrative review of risks, responsibility, and the epistemological limits of large language models

The rapid integration of artificial intelligence (AI) into diagnostic medicine represents both a technological advance and an epistemological disruption. This narrative critical review examines how large language models and related AI systems influence diagnostic reasoning, clinical accountability, and medical ethics. While AI has demonstrated remarkable potential in pattern recognition and data synthesis, its correlation-based logic lacks causal understanding, interpretability, and moral agency. The review identifies key risk domains including automation bias, dataset bias, model opacity, and liability asymmetries between developers and physicians. It argues that AI's integration challenges traditional boundaries of professional responsibility and informed consent, raising concerns over epistemic validity and patient trust. Current governance frameworks, adapted from static medical device regulation, remain ill-suited for continuously learning systems. The research shows that safeguarding the integrity of medical reasoning in the age of AI requires a renewed commitment to epistemic humility, distributed accountability, and the preservation of human judgment within computationally mediated care.