Machine Learning and the Future of Scientific Explanation

On 17 and 18 February 2020, Rafaela Hillerbrand and Paul Grünke (both Karlsruhe Institute of Technology) from the research project “The Impact of Computer Simulations and Machine Learning on the Epistemic Status of LHC Data” organized the workshop “Machine Learning: Prediction Without Explanation?” at Karlsruhe Institute of Technology (KIT). The project is part of the interdisciplinary, DFG/FWF-funded research unit “The Epistemology of the LHC”; a unique collaboration between philosophers, physicists, historians and sociologists that was recently renewed for three more years.1 The workshop’s purpose was to bring together philosophers of science and scholars from various fields who study and employ Machine Learning (ML) techniques, in order to create an interdisciplinary setting for discussing the changing face of science in the light of ML’s constantly growing use. Because ML is, from a certain vantage point, nothing but (statistical) optimization executed by digital computers, one may speculate that its increased use exemplifies a paradigmatic turn away from science’s traditional aim of explanation, and towards mere pattern recognition and prediction. Moreover, it is also an open question how to explain ML’s exceeding utility, as witnessed by various benchmark studies in recent years. Given this difficult epistemological status of ML, one may ponder the societal implications of its use, as well as its historical and systematic positioning as a scientific method. Accordingly, the talks’ contents will be discussed as organized into (i) practitioners’ perspectives, (ii) explanations from ML, (iii) explanations of ML, (iv) societal implications, and (v) global and historical perspectives.