A new paper, co-authored by Cluster Principal Investigators Eva Blasco, Pavel A. Levkin, Yolita M. Eggeler, and Pascal Friederich and published in Nature Machine Intelligence, explores how artificial intelligence can help identify new research directions in materials science. The team analyzed more than 200,000 scientific abstracts using large language models (LLMs) to extract key concepts and build a large-scale “concept graph” of the field. They then trained machine learning models to predict novel combinations of scientific concepts that could lead to future discoveries. Interestingly, evaluations with domain experts revealed that a significant proportion of these AI-generated suggestions were found to be interesting and potentially valuable. This work highlights the potential of AI as a tool that can support researchers in navigating complex knowledge spaces and accelerating innovation.
© Marwitz et al., Nature Machine Intelligence (2026), CC BY 4.0