Doesn't (over-applied) machine learning just lead towards a variation of exactly the same problem? Instead of "this model fits the curve and sounds attractive

Doesn't (over-applied) machine learning just lead towards a variation of exactly the same problem? Instead of "this model fits the curve and sounds attractive and sciencey" it becomes "this model fits the curve and was done by computer!" Not really the "empiricism coming to economics" that the author claims, but I guess it could support better outcomes. For a while.