Indeed. And even if we were to minimize bias, machine learning has another fatal flaw: it still gives us systems where have no way of knowing what it is the system actually learned, and if we discover an error, no way to fix that exact problem, we can only retrain the system and hope it’s better.
