Could machine learning fuel a reproducibility crisis in science?