What is stopping us from creating a "supervisory AI" that understands the difference between good content and bad content and having it perform the role of thes

What is stopping us from creating a "supervisory AI" that understands the difference between good content and bad content and having it perform the role of these humans? More broadly speaking training and overseeing a sufficiently advanced AI seems like it could only be done effectively by another AI. The human oversight could be temporarily redirected to validating the supervisory AI.