Ai learns scientific discovery steps from human research changes
Learning Scientific Exploration from Human Research Decision Trajectories
Machine LearningComputation and Language
Summary
Scientific research involves many small decisions and experiments, but published papers usually show only the final results, missing the step-by-step process. The authors created ResearchTrails, a dataset that uses version control histories from coding projects as a record of how research evolves over time. They show that these research paths reveal useful information about decision making during science, beyond just the final outcomes. Training AI with this data can help it better understand and predict research decisions, potentially improving how AI assists scientific discovery.
What this means in practice
- •For machine learning engineers: Use ResearchTrails to train models that better predict intermediate research decisions, improving AI guidance in scientific experiments.
- •For ai tool developers: Incorporate ResearchTrails data to build AI that mimics human scientific exploration steps, enabling smarter research assistance tools.
Authors
Xuchen Gong, Shane Gu, Haokun Liu, Dixi Yao, Chenhao Tan, Tian Li
Abstract
A key challenge in building AI systems for scientific research is enabling $\textit{scientific exploration}$: the systematic process of investigating unknown phenomena or ideas to gain new knowledge through sequences of research decisions and actions. Yet this process is largely missing from existing scientific corpora; for example, research papers primarily record final outcomes rather than the trajectories that produced them. In this work, we introduce $\textbf{ResearchTrails}$, a dataset of $\textbf{human research trajectories constructed from Git repositories}$, where $\textbf{commit histories}$ serve as proxies for research exploration. We develop an automated and scalable pipeline that extracts structured research trajectories from repository commits, capturing successive changes to methods, experiments, and ablations. We characterize the resulting dataset and show that these trajectories contain meaningful signals about intermediate research decisions beyond what final papers reveal. We further demonstrate utilities of ResearchTrails in multiple use cases, including retrieving human research experience as external skills at test time and training models on research trajectories to improve generalization to new research decisions. Our results suggest a path toward AI systems that learn not only from the products of science, but from the evolving process of discovery itself.