The research analyzes the strengths and weaknesses of scientific texts produced by various artificial intelligence (AI) programs available on the Internet. The research aims to determine the extent to which scientists and university faculty can currently rely on AI achievements. By applying fundamental general scientific methods of analysis, the authors conclude that neural networks are useful in scientific discourse for writing formulaic texts, which broadly include annotations, reviews, and news reports on research discoveries (for popularizing university activities on websites and in the media), for selecting illustrative material for presentations, and in educational activities as a tool for students’ independent work in planning future scientific texts. However, neural networks cannot currently significantly aid in scientific work because they lack critical thinking and the ability to conduct innovative scientific activities. Additionally, the most significant weaknesses of AI include the absence of emotional intelligence and the pedagogical skills necessary for training new scientific personnel.