warhol v. goldsmith and fair use

the great thing about this case is they have to clarify every time when they’re talking about campbell (famous copyright case concering parody) or campbell (soup). even so, this case is really about conde nast reproducing a warhol silkscreen of prince after his death, and the questions in this case are about as base as you can get when it comes to fair use: what is a transformative use, and how transformative do you really have to be?

fair use is wildly misunderstood by the public, mainly that it’s an affirmative defense which requires evidence to negate criminal liability — you can’t claim fair use at the time of creation. the fact that there is still judicial review into the intent of “purpose and use” and whether a transformative work can satisfy that condition of fair use by imbuing additional “meaning” should really help people understand that nothing about the way we are legally permissed to interact with culture and media is sacred. as more cases heat up about stable difussion and AI, which along with warhol are likely to be the most consequential concerning copyright for the next few decades, it’s easy to envision how these decisions might shake out to be broad or narrow for many different reasons:

• these technologies are inherently transformative, but in warhol it’s quite clear from the way justices responded to oral arguments that transformation itself does not override the other factors of satisfying fair use in section 107, and indeed there is an obvious appetite among justices to narrow transformative use that can be classified as fair use by bleeding the fourth factor (effect on the market for the copyrighted work) into the first “purpose and use” factor
• mechanically created works are already not protected in the US, meaning any AI generated work can be replicated, sold, and so on without impunity. most of the artists and publishers likely to pursue relief in court against AI are probably in the commercial market, which will definitely impact how their cases are perceived by justices
• with warhol, it’s quite clear that a substantive portion of the original copyrighted work was used (third factor of section 107), but it’s also possible to imagine that a similar reproduction could exist without using goldsmith’s particular photograph. can the same be said of current AI models, almost all of which are trained on datasets with copyrighted materials, and probably wouldn’t produce such fantastical specific results of certain artists or styles without them?
• copyright protecting smaller or independent artists in the US is largely the result of the expansion of copyright to protect enterprise: longer terms, licensing requirements, the DMCA, and much more. right now it is accepted practice to train data models on copyrighted material, but longstanding attitudes can quickly change as the result of one court’s interpretation, most recently obvious in how sampling became pretty much prohibitively expensive overnight in 1991. courts in my view are unlikely to enforce limitations on training data itself ¹, but there is a future I can envision that carves out tailored exceptions on AI, like generating on specific artist styles trained with hypernetworks or embeddings

copyright status quo currently favors large publishers, which I think is likely to emerge as a point in their favor in the courts as AI matures from research to commercial reality. courts are also increasingly used to play political football and in recent years are more willing to deliberately cause circuit splits to force review of what was once widely held standard and precedent². given the recent judicial turmoil, I wouldn’t be surprised to see more carve outs and exceptions to fair use³ that further upend the ways people have come to expect they’re able the interact with media. anyone that has a holistic or definitive answer to how these arguments will be settled is probably selling you a bridge

the oral arguments in this case are worth a listen, if only to expose yourself to how technical arguments about creativity can get: https://www.oyez.org/cases/2022/21-869