Siliconvs. Stethoscopes: Why Healthcareis Resistingthe Ai Revolution

Abstract

The present study attempts to assess if modern artificial intelligence (AI), consisting of CNNs in medical imaging and LLMs in language processing, would replace the physicians in the near future and find the most significant barriers to this process. The narrative review of the scientific literature about AI performance and reproducibility in medical imaging, LLM's competence in assessing medical knowledge and communicating with patients, lack of generalization beyond the distribution, lack of physical examination and sensory input, and the existing regulatory and legal environment, especially the one in the European Union is performed. AI is highly accurate and reproducible in narrowly defined tasks of image interpretation, lesion measurement, triage, documentation assistance, and written communication. These abilities diminish the interobserver variation and improve the workflow efficiency. Physicians are here to stay, and that is for a reason. While the AI is impressively accurate in many tasks, it fails miserably the moment the case is out of the training set distribution, cannot do physical examination and surgery, and still invents some answers with full confidence. Apart from technical problems, the most significant issue here is the one of accountability: who will be sued once an algorithm makes a wrong decision? For a foreseeable future, the AI will remain a valuable assistant, but never a physician. It will be able to handle repetitive and well-defined tasks in the background, but the high-stakes part of medicine – physical examination, complex ethical decisions, bedside care, and overall legal responsibility – will remain a doctor's duty.