USE OF GENERATIVE LANGUAGE MODELS BY PHYSICIANS: BETWEEN PRODUCTIVITY GAINS AND RISKS OF ERROR AND DEPENDENCE IN CLINICAL PRACTICE

<jats:p>Generative language models, such as ChatGPT and similar tools, have been rapidly incorporated into the practice of healthcare professionals, who have begun using them for tasks such as document preparation, information research, and support for clinical reasoning. This article aims to analyze the use of generative language models by physicians, examining the tension between productivity gains and risks of error and dependency. A literature review of a qualitative nature was adopted, based on scientific literature on the application of these models to medicine. The results show that physicians adopted these tools rapidly, especially for documentation generation and differential diagnosis suggestions, obtaining productivity gains. However, it was found that language models present relevant risks, including the production of false information, or hallucinations, whose incidence in clinical tasks can be high, and the risk of dependency and disqualification of professional judgment. It is concluded that the use of generative language models in medicine can bring significant benefits, provided that it is responsible, transparent, and supervised, preserving the ethical and legal responsibility of the physician for clinical decisions and avoiding both error and uncritical dependence on technology.</jats:p>