> If I trained a model exclusively on Warhol art, and then had that model create new images in Warhol's style, I didn't do any of the creative work and probably don't get the copyright.
If I watch exclusively Warhol images for years and then paint something similar I get copyright.
There needs to be a gray are, because usually art is not done in a vacuum?
> If I watch exclusively Warhol images for years and then paint something similar I get copyright.
Not necessarily. If you copy one of Warhol's works but "change it a little" then that is a derivative work, and the copyright belongs to Warhol's estate. Depending on how close of a copy it is, you would have a tough time defending your claim to copyright in court. The advantage an offending artist has in court is that they can claim "inspiration" as long as they don't admit to copying.
For a computer model the difference maker is that the court can probably obtain records of a training set, so if the training set is exclusively Warhol works it is probably easy to get a court to side on "derivative" and assume the computer does not possess inspiration.
Courts have basically baked in "gray areas" in copyright cases. The historical copyright tests are all written as to sound like mathematical formulas but everything is kind of subjective.
You also cannot train only on Warhol imagery, unless he drew billions of pics. So this is hypothetical “if”. In reality you finetune an existing network based on a dataset much larger than Warhol’s.
If I watch exclusively Warhol images for years and then paint something similar I get copyright.
There needs to be a gray are, because usually art is not done in a vacuum?