A half-serious thought I had is to allow AI to be trained on copyrighted materials but require all output to be public domain.
It doesn’t solve attribution, and you could get a sufficiently advanced ai to “launder” copyright but at least it would prevent those corporations from leveraging public works into copyrighted projects
That's genuinely the best solution I've heard for this issue so far.
It'd stop buisnesses from replacing artists with models wholesale, but it would allow people to keep using them, and allow the companies making the models to make money selling access to the models.
It doesn’t solve attribution, and you could get a sufficiently advanced ai to “launder” copyright but at least it would prevent those corporations from leveraging public works into copyrighted projects