4 ms·
> one of the factors affecting whether the act of copying is fair use, according to Congress, is “the effect of the use upon the potential market for or value o
by Ukv 3y ago
> one of the factors affecting whether the act of copying is fair use, according to Congress, is “the effect of the use upon the potential market for or value of the copyrighted work”. Today’s generative AI models can clearly be used to create works that compete with the copyrighted works they are trained on. So I don’t see how using copyrighted works to train generative AI models of this nature can be considered fair use.
Fair use's factors each weigh for or against a finding of fair use, as opposed to needing to strictly satisfy all four. In particular, what machine learning is likely heavily resting on is that "The more transformative the new work, the less will be the significance of other factors" (Campbell v. Acuff-Rose Music).
For instance, Google Translate was trained on translator's works, and may in part compete with the market for translations, but I'd claim is transformative by nature of adding something new (instant on-demand translation of novel text) and not merely superseding the static works it was trained on.
How to decide "the effect of the use upon the potential market for or value of the copyrighted work" is also a bit of a gray area and one of the questions the US Copyright Office were seeking comments on. Should it be about the impact models have on the market for that general class class of works? Or, the extent to which training on a specific work impacted the market for that specific work compared to if the model was not trained on that work?
- __loam 3y agoI think it's pretty easy to argue that generative ai has no value whatsoever without a large set of training data, which happens to include the works of creators without permission. You could even go further and say most of the value involved with these models is in the training set, and not any of the engineering or models themselves. GPT-4 would suck a lot of they hadn't hoovered up a significant chunk of the internet. One could view these models as massive alienation machines, designed to separate data from the people who created it. I think that's really bad for the arts and for the economy, and eventually for the AI companies themselves. It's going to be really hard to keep getting training data when nobody is posting anything for you to steal anymore because of the egregious thing you already pulled off. It may not be strictly illegal, but in the view of many, including the author of this tweet and myself, it's incredibly unethical to cut artists away from the value they were responsible for, and even more perverse to then use that work to put them out of a job while claiming it's for the good of humanity.