Could machine learning fuel another reproducibility crisis in science?

There's a bit of denial about the framing of the article, which is maybe telling.

The article is about a slightly different form of reproducibility crisis I think, one more specific to ML studies. But the title maybe isn't quite right.