I might be referring to a different problem here, but often when ML gets something wrong, it gets it mind-bogglingly wrong.

Taking an extreme cartoon example, where a doctor might think a cancer is a cough, a computer might think the lamp is a patient.