This paper on copyright in relation to AI models is an interesting read. It takes quite a different approach to the subject compared to what you usually encounter, particularly regarding how training data should be accessible to many more entities than just the largest labs. The paper also discusses why issues of bias, inclusion, and other factors need to be part of the pursuit of a sustainable solution to these questions. It suggests that focusing solely on the financial aspects of copyright probably won't lead us to where we want to be.
Visit the source: papers.ssrn.com