A lot of machine learning practitioners don't have the statistical sophistication to appreciate this, and as ML becomes increasingly "democratized" I expect the situation will worsen.