The impact of computer simulations on the epistemic status of LHC data

• Applicants: Professorin Dr. Rafaela Hillerbrand; Professor Dr. Gregor Schiemann; Professor Dr. Christian Zeitnitz
• Subject Area: Theoretical Philosophy
• Term: from 2016 to 2024
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 234743567

We investigate the involvement of computer-based methods in the ATLAS experiment and how they impact the epistemic status of experimental results. Phase 1 focused on the use of computer simulations (CSs) at ATLAS, analysed the epistemological status of CSbased knowledge in an experimental context. It investigated the structure of simulation models at ATLAS, including their relations to theoretical and phenomenological models from particle physics and empirical knowledge. Our research contributes to contemporary discussions on computer simulations within the philosophy of science, and hence has relevance beyond the philosophy of physics. We argued, for example, that, while CSs in particle physics are epistemically inferior to laboratory experimentation, they are often de facto superior to theoretical arguments. Another result of our research in Phase 1 is that the CSs used at ATLAS are epistemically not more opaque than other epistemic practices. Further work (in progress) aims to show how CSs are dispensable in particular ATLAS experiments.In Phase 2, we plan to analyse, on the basis of case studies, how the involvement of CSs might impact the LHC’s discovery potential and its ability to find (subtle) signatures of new physics; we will also include machine learning as a new topic. While used in particle physics and other areas of science for a long time, machine learning has not yet been widely addressed in the philosophy of science. It provides another class of computerbased methods, built around learning algorithms. While CSs are largely model-driven, machine learning is more data-driven. While the simple models underlying CSs may be viewed as reducing the complexity of theories in particle physics, machine learning reduces the complexity of data by finding patterns, at the expense of an apparent increase in opacity. Advances in deep learning (DL), a subclass of machine learning, have recently opened up new possibilities for utilisation in particle physics. As epistemic practices, both CS and DL come with various epistemic risks, associated, e.g., with modelling choices in the former and concept formation in the latter. We aim to provide a detailed account of these risks and how they are managed in specific case studies from ATLAS research.

• DFG Programme: Research Units
• Subproject of: FOR 2063: The Epistemology of the Large Hadron Collider