In a blog published on Wednesday, Facebook admitted its use of AI for detecting harmful content was 'not perfect.'