Other case mentioned was with face recognition, imagine a coop trusts the AI that tells him that you are a dangerous criminal, you could get killed if you are unfortunate because the cop could assume the computer is right and you are very dangerous and never thinks about the false positives