Automating Pursuitworthiness: Four Concerns About The Proper Roles for Machine Learning in Scientific Discovery
Year: 2025
Type: preprint
Abstract: Machine learning (ML) systems play increasingly important roles in scientific discovery. Recent efforts seek to build ML systems that predict upcoming discoveries and who is likely to make them, identify emerging research trends, and suggest novel concepts, questions, hypotheses and experiments to investigators. These predictive discovery and recommender systems (PDRS) hence aim to augment and automate key activities that are central to the roles currently played by human researchers. This paper argues, first, that PDRS raise novel conceptual and methodological disruptions, creating uncertainty around whether PDRS can and should play such roles. Second, the paper draws out four major questions, and associated concerns, about the roles PDRS are envisioned to play. These issues have not received attention in the literature thus far, leaving unclear what the proper roles of PDRS in science could be and how these roles should be carved out through appropriate designs and divisions of labor. To address these issues, the paper explores concerns about PDRS’ potential impacts, their likely limitations, and how PDRS fit into our broader views of how science should function.
Author: Khosrowi, Donal
Institution: Leibniz University Hannover
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7110469074
