How is AI changing the methods and practices of research in different scientific fields? Do artificial neural networks and other machine learning approaches allow for entirely new insights or is their function more about supporting the ordinary operations of the sciences, for example by automating workflows? How can the epistemological potentials, constraints, and ambivalences of the methodical use of AI be explored in any meaningful way from a comparative perspective? My talk will present the work of a research group funded by the VW Foundation since 2019 that analyzes these questions by combining the perspectives of three disciplines: history of science, media studies, and computer science. I would like to show, first, how our project is developing a practical tool that will make it possible to use AI for researching AI. Furthermore, I will discuss how a media-ethnographic approach can contribute to closely examine the transformations of the sciences through AI on a micro-perspective level.