I fail to see why this is a good argument against AI replacing humans, given that humans also produce errors.