You would want your noticing that something is bad to, in some way, indicate what would be a better way to make the thing better. You want to know what in particular is bad and can be fixed, rather than the less informative "everything". If your classifier triggers on everything, it tells you less on average about any given thing.