Discussions around the value of AI in science lack clarity about what ‘AI’ actually entails—from basic neural networks to advanced foundation models—and how these tools are applied in the scientific process. This project will systematically map the use of various AI approaches (machine learning, deep learning, foundation models) across scientific disciplines and research workflows. The team will begin by mining scientific literature for citations of key AI papers. Using large language models and natural language processing, they will assess the degree of AI engagement in citing papers—ranging from simple references for context, to active use in research methods, to modification of the AI models themselves. The study will also explore how adoption varies across disciplines based on model attributes such as size and openness. The focus will remain on AI’s role within the research process itself, rather than on administrative applications. Ultimately, the project will shed light on the evolving role of AI in science and bring greater precision to how we talk about AI’s place in scientific research.