Machine Learning and the Progress of Science (Elements in the Philosophy of Science)

Machine Learning (ML) models nowadays solve prediction problems that had been unsolved for decades. Yet, why do real-world systems behave as predicted? How can we manipulate them to behave otherwise? The ability to answer such questions reflects understanding, which several philosophers take to be the measure of scientific progress. But ML models typically do not deliver answers. Can ML really foster scientific progress? On knowledge-, truth-, or justification-based approaches to progress, ML might advance science in different ways: On knowledge-based approaches, it may lead to progress through the discovery of novel phenomena; on truth-based ones, through the discovery of novel hypotheses; on justification-based ones, by delivering a new basis for justification. This Element enters new territory by investigating these ways in which ML may lead to scientific progress. It also constitutes an opinionated snapshot of the fast-growing philosophical literature on ML in science.