> 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.<p>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).<p>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.<p>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?