Demonetizing generative AI: an approach to copyright in training and inference

Copyrights have been at the kernel of the disputes around ethical authorship practices around visual generative artificial intelligence. We can tackle these disputes by separating them between the input, process and output of models. Their input consists of millions of images downloaded from the Internet without any concern for the copyright status, what has been the cause for strong reaction from artists who saw their work be integrated into the neural network without any consent. The output, on the other hand - the assets generated by inference processes - stand on a gray area of ownership. The legal framework of several jurisdictions states that they can have no claim to copyright, as they were not created by humans. Other jurisdictions, like the United Kingdom, place the ownership in the hands of the humans who coordinated the generative process, a stance that also excludes from artists in the training dataset from titular rights. This paper makes the claim that, while there is no easy solution to this situation, the obliteration of copyright of generated pieces would beget a scenario that would enrich global culture while demonetizing the exploitation of cognitive labour of artists included in the training set. The grounds for this limitation should lie on the fact that they are trained on humanity's cultural commons, and not on the fact that they are engendered by machines. To make this assertion, we hark back to the Remix theories of the early 2000's, specifically as formulated by law scholar Lawrence Lessig, as a framework that will allow us to see generative media as a successful application of fair reapropriation of earlier works. This stance is, of course, not welcome by neither big tech players (who could see a devaluation of their generative products), nor by independent artists (who could feel under an even greater fear of exploitation), making it admittedly unrealistic. Nevertheless, humankind in general would benefit greatly from open access to the general cultural commons, allowing generative AI to become a tool for remixing and recreating culture in levels never achieved before.