Automatic rejection for AI-generated submissions

Hello,

Every day I come across more AI generated content to review. Besides the usual signs (strange amount of limbs or fingers, jumbled-up writing, inconsistent texture) there is also a very obvious tell: the submission photograph and the supporting photograph are often the same image. Sometimes one of the images is slightly cropped. This is the result of AI struggling to generate two consistent images of the same POI.

I believe that having an automatic rejection for submissions with the same image (or a cropped version of it) would drastically cut down on the amount of AI slop reviewers have to check. It is also incredibly easy and intuitive for the people submitting true POIs to take multiple images, so the amount of false flags would be minimal.

You’d end up with a ton of false negatives and frustration on the part of first time submitters who legitimately use a crop of one photo for both images. Which isn’t to say it’s a bad idea, but it would cause issues.

I do think ML should be able to do SynthID checks, and as AI detection improves, it should be incorporated into ML if possible. Tools like Neuramancer would be a great addition (though I’m not sure if in the budget?)