But presumably all the images used to train the AI had copyright, so shouldn't the the copyright in the AI's derivative work belong to them ?
Otherwise I could "train" an AI on a single work, end up with a training matrix just consisting of "1" and then "generate" as many of the new AI-derived, non-copyrighted not-at-all-a-copy of the work as I want
Biting the hand that feeds IT
