Computational Assistants for Scientific Discovery

Objective

Societal challenges like climate change, healthcare, and food security demand rapid scientific and technological innovation. Recently, AI demonstrated its transformative potential for biology, chemistry, and materials science. However, while AI already excels in domain-specific tasks, its role as a general-purpose research assistant remains underexplored, even though this use case could fundamentally enhance scientific discovery as a whole. This proposal aims at developing human-centric AI assistants to support researchers by addressing cognitive challenges, fostering creativity, and enabling new research across disciplines. More concretely, we will develop: (A) Structured document models to analyze and relate content within and across documents; (B) Computational methods for predicting and understanding scientific impact; (C) Hierarchical models to map the evolution of scientific concepts over time; and (D) Ideation assistants to support researchers in generating novel and impactful ideas. Our human-centric approach will focus on trust, interpretability and user agency, ensuring that AI serves as an enabler rather than a replacement for human intellect. By deploying the models via our Scholar Inbox research platform, we will provide easy access to the developed tools and evaluate their long-term impact on scientific discovery across all fields of science.