![]() ![]() But a bit of thinking tells us that some generative AIs are detrimental to the most cherished aspect of human culture, like music, prose or graphical arts. Our culture and ways of life evolved through millennia, and the majority of its bits and pieces were not thought out for dealing with generative AIs. > I think at this point we have to ask ourselves if we want these models or not, and who will have them. 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. 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. I think that's really bad for the arts and for the economy, and eventually for the AI companies themselves. One could view these models as massive alienation machines, designed to separate data from the people who created it. GPT-4 would suck a lot of they hadn't hoovered up a significant chunk of the internet. 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. I 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. 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? 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. Acuff-Rose Music).įor 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. 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. So I don’t see how using copyrighted works to train generative AI models of this nature can be considered fair use.įair use's factors each weigh for or against a finding of fair use, as opposed to needing to strictly satisfy all four. Today’s generative AI models can clearly be used to create works that compete with the copyrighted works they are trained on. > 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”. It’s a complex topic and saying “humans work this way, computers are fine to as well” doesn’t IMO capture the nuance of the debate. I’m certainly sympathetic to the idea that it might, though. Having said all that, I don’t have enough skin in the game to decide whether that builds an argument for regulating input vs output. If you consider some finite amount of capacity in the world to gainfully support all artists, I’m only drawing from that for a finite period of time.Ĭontrast that with AI, where, theoretically, in 50 years a model will still be just as able to create outputs as it is today. Sooner or later I retire/die and someone else must become the cultural torchbearer. If I learn how to write music per the above, I’ve only got a set period of time wherein I’m collecting from the talents given to me by the shoulders upon which I’m standing. Human capacity is rate-limited, and that’s an important distinction.Īnother point: humans exist for a finite time. ![]() That is to say, if I spend years learning how to write music and being influenced by a suite of prior artists, I can’t then package myself up and arbitrarily clone my capability such that an infinite number of people can ask me to spit out new music drawing inspiration from those prior artists. Humans can’t automate and industrialise the process in the same way that AI can. ![]()
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |