I find some of the arguments compelling about how AI art plagiarizes the artists whose works the models are trained on.
But I'm also reminded of the controversy around sampling in hip hop, which gave marginalized people a way to create new art, changing the entire music world.
I think of Peter Gunz and Lord Tariq's hit Deja Vu. It sampled a Steely Dan song, and Donald Fagan and Walter Becker won 100% of the writing credit for Deja Vu in court. That doesn't seem right to me, but it's unsurprising the way the courts ruled against the Black dudes from the Bronx. https://en.m.wikipedia.org/wiki/Deja_Vu_(Uptown_Baby)
It's not exactly the same though. Samples in music can be identified, but the training source material in AI models might not be, let alone the contributions of specific pieces to a given generated output.
Cc @mykola
@acjay @mykola yeah this is a useful point.
I don't think we really have the language to talk about this stuff properly though, like there may be 'copyright' breaches in the process (maybe in a different way than most people think) but the thing that's annoying people is the way some fairly accepted things work if they're combined with automation
@lyndonhood @mykola Yeah, I'm faaar from a defender of the current IP regime. But attribution seems like table stakes for a fair system.
At the end of the day, we're basically talking about a new front in the struggle for how to conceptualize intellectual property, in which the works that a derivative work sources from become increasingly obscure to the author, let alone consumers.
