News
Gastvortrag "Negotiating the Proper Roles for Machine Learning in Science"
Dr. Donal Khosrowi von der Leibniz Universität Hannover wird als Teil der "Lecture Series on AI in Science and Society" am 13.06.2025 von 14:15–15:45 Uhr in H27 (GW II) einen Vortrag zum Thema "Negotiating the Proper Roles for Machine Learning in Science" halten.
Abstract: Machine learning (ML) systems play increasingly important roles in scientific discovery, e.g., to discover novel protein structures or materials, recover new physical quantities and equations, or reconstruct broken historical artefacts. In the wake of these developments, significant conceptual disruptions (Löhr 2023) take place: central concepts we use to understand and structure scientific pursuits come under pressure, leaving unclear how to understand the role of ML systems and their outputs in scientific enterprises. This talk focuses on disruptions affecting two key concepts: ‘evidence’ and ‘researcher’. The concept of ‘evidence’ is disrupted by recent attempts to use generative AI systems in the historical sciences to restore partially destroyed manuscripts or artefacts. Such use-cases raise uncertainty around whether to interpret their outputs as mere hypotheses in want of independent support, or as full-blown evidence that already entitles researchers to make new knowledge claims. Second, the concept of ‘researcher’ is disrupted by recent efforts to build ML systems that predict upcoming scientific discoveries and suggest hypotheses and experiments. Such systems go beyond automating execution-level tasks, such as efficiently searching a space for stable chemical compounds, and instead automate high-level agenda-setting roles, steering what to investigate and how. I argue that these novel roles envisioned for ML systems press us to consider what it means to be a researcher, whether emerging ML systems possess abilities central to this role, and what are good divisions of labor between human researchers and ML systems. Together, both disruptions illustrate the need for a larger research program that examines and provides responses to these ongoing disruptions.