> If you want to use incorrect data to get correct answers, you need a human.

More often, that just gets answers that are incorrect in ways other than (sometimes, in addition to, othertimes, instead of) those explicable by the incorrect data.