It's a request for knowledgeable identifiers to check the observations and correct as many as possible. It suggests that the training of the computer vision has been skewed by incorrect identifications by users, and possibly by its own suggestions being wrong, but accepted by users.