The authors analyze key problems of using artificial intelligence (AI) in scientific research. It considers such current challenges as insufficient interpretability of algorithms (the «black box effect»), dependence on data quality, the risk of false correlations, and ethical issues of authorship. Particular attention is paid to the problem of reproducibility of results obtained using AI, which calls into question their reliability in the academic environment. The work uses the methods of a systematic literature review, comparative case analysis, and conceptual modeling, which allowed us to identify the main trends and contradictions in this area. The study is interdisciplinary in nature, combining approaches from computer science, science studies, and the philosophy of science. As solutions, the authors propose the development of hybrid models that combine AI with traditional scientific methods, the introduction of explainable artificial intelligence standards, and the creation of a regulatory framework for verifying results. It is emphasized that AI should not replace, but rather complement classical research practices, enhancing the analytical capabilities of scientists. The work is of interest to researchers working at the intersection of artificial intelligence and fundamental science, as well as to specialists involved in the methodology of scientific knowledge in the digital age. The research materials can be used in the development of curricula on data science and scientific ethics.