A brief review of some artificial intelligence methods in nephrology.

This brief, focused review considers two of the more commonly used artificial intelligence (AI) methods encountered in nephrology publications: machine vision based on convolutional neural networks (CNNs) and chatbots, such as ChatGPT, based on large language models. It is intended to offer a mostly non-technical, intuitive understanding of these methods, including their uses and limitations. CNNs have been used for some time to segment and classify important features of digitized kidney biopsy images. In addition to the identification of pathologic primitives, CNN approaches may be used to predict so-called sub-visual features of biopsies, such as kidney survival rates. Large language models are newer players in the medical AI field. Although seemingly easy to use as natural language tools, most currently available chatbots have been characterized by inconsistent performance, hallucinations, and even a higher CO2-footprint than CNN-based models.