[jhanschoo]

> The concern with the 'AI' is that it's copyrighted content being recombined in ways that don't create a new copyright, so the output can't have a new copyright.

I believe the conversation was with respect to copyright ownership of the transformation applied to the training materials, and not the training inputs itself, (to the extent that the AI has been so poorly trained that its outputs aren't sufficiently transformative of its originals; A transformative work with a human author is copyrightable and does not infringe on the copyright of the original work).

On a liminal case, are song performances by songbirds who have heard it on the radio copyrightable?