The question of copyright for AI-generated works is a complex legal and ethical challenge. Currently, most jurisdictions attribute copyright to human creators, raising debates about whether machines can hold rights. Addressing this issue involves reevaluating copyright laws to adapt to the evolving landscape of AI-generated content and finding a balance that respects both technological advancements and human involvement in the creative process. Legal frameworks and international discussions are essential to navigate this evolving aspect of intellectual property.

There are reasonable arguments on both sides. First argument is that the owner of the machine/AI system invested resources into creating and training the system, and directing it to generate the output. So they have some claim over works created using their system. However, the actual creative process is being done autonomously by the AI.

Notwithstanding, since the AI system is doing the creative work, perhaps it should hold some form of legal rights over works it generates. However, current copyright law applies to human creators, and there are open questions around legal personhood for AI systems. Additionally, practically it may be difficult to vest and enforce rights without a legal entity behind the AI to claim those rights.

Among the many proposals for alternative models are as follows:

Thus, having copyright vest in the dataset creators and trainers who helped teach the AI system. However, their individual contribution is hard to establish.

On the other hand, having works pass into the public domain by default to promote access, since no human directly created them. However, this removes commercial incentives for developing AI that can create works.

Overall there are good-faith arguments on multiple sides, and reasonable people can disagree on the ideal approach. There are open questions around whether current copyright laws need to adapt to account for the rise of creative AI systems. It's an emerging issue that may take some time for legal frameworks, philosophers, ethicists and technologists to fully work through.

First, any work produced by AI right now is a derivative work, that is all the creative process is being done by a permutation space. As of this point it has not been proven (as far as I know) that this permutation space provides enough distance between the original work and the derived work to be called creative (if anyone has proof of this please let me know. I am eager to verify such claims). It is interesting that the field that I work in treats so loosely the scientific standards that are needed to verify such claims.

As of this point I would say that neither the ML trainers nor the AI deserve credit for creating works besides derivative ones.