Generative Artificial Intelligence in Clinical Medicine: Potential and Limitations

Generative artificial intelligence (GenAI) is rapidly penetrating various aspects of clinical medicine—from medical documentation generation to clinical decision support, patient communication, and medical education. However, the rate of technological progress significantly outpaces the accumulation of clinical evidence. This paper systematizes the main paradigms for applying GenAI in clinical medicine and analyzes its inherent limitations, including the uncertainty caused by random sampling, the problem of hallucinations and traceability of evidence, and the risk of "cognitive debt" leading to the degradation of clinical skills. Based on this, the need to shift from a model-centric paradigm to a "data-model co-evolution" paradigm is discussed and the logic for constructing a risk-stratified management system is outlined. The key thesis is that the clinical value of GenAI is determined not by benchmark results, but by its verifiability, monitorability, and adjustability in real clinical workflows.