I did not cite a collage to argue that ai are collages.
And indeed they are not. Collages, unlike AI images, contain creative input beyond their source material, which qualifies them for copyright protection separate from their source material.
Using material for training an ai is currently also ruled as derivative.
Is that what you meant to say? Because it completely undermines your previous arguments.
Derivative works are still bound by the terms of the copyright of the source material. If I create a game that incorporates GPL-licensed code, I have created a derivative work, and I am bound to release my game source code under the terms of the GPL. Furthermore, under United States copyright law, copyright ownership in a derivative work attaches only if the derivative work is lawful, because of a license or other “authorization.”.
But AI images are not derivative works because they are not works in the first place. They are merely derivative. In the US, the output of an algorithm not copyrightable separately from the input.
Well, in my opinion, the morally wrong thing here is not the training of ai. We all stand on the shoulders of giants. But rather the replacement of human work by ai output.
Using procedural art instead of of hiring an artist is not immoral. Using photographs instead of paintings isn’t immoral. Respectfully using public domain art isn’t immoral. Making a text-based game with no art at all isn’t immoral. Nobody is obligated to be a patron of the arts. None of these are based on stealing.
Meanwhile the ai output of such models would still replace human work.
You have no idea how much work goes into training an AI before it starts producing useful output, do you? If each AI were trained only on commissioned artwork produced for that AI, the AI companies could employ every artist on Earth and the AIs still wouldn’t be as good as what we have now.