Inductive Risk, Understanding, and Opaque Machine Learning Models

Abstract Under what conditions does machine learning (ML) model opacity inhibit the possibility of explaining and understanding phenomena? In this article, I argue that nonepistemic values give shape to ML problem even if we keep researcher interests fixed. Treating models as an instance doing model-based science explain understand phenomena reveals there is (i) external problem, where presence inductive risk imposes higher standards on externally validating models, (ii) internal greater demands a level transparency regarding inferences makes.