A new research paper laid out ways in which AI developers should try and avoid showing LLMs have been trained on copyrighted material.

OpenAI now tries to hide that ChatGPT was trained on copyrighted books, including J.K. Rowling’s Harry Potter series::A new research paper laid out ways in which AI developers should try and avoid showing LLMs have been trained on copyrighted material.

  • TwilightVulpine@lemmy.worldEnglish
    101·
    1 year ago

    But this falls exactly under what I just said. To say that using Machine Learning to imitate an artist without permission is fine, because humans are allowed to learn to each other, is making the mistake of assigning personhood to the system, that it ought to have the same rights that human beings do. There is a distinction between the rights of humans as opposed to tools, so to say that an AI can’t be trained on someone’s works to replicate their style doesn’t need to apply to people.

    Even if you support that reasoning, that still doesn’t help the writers and artists whose job is threatened by AI models based on their work. That it isn’t an exact reproduction doesn’t change that it relied on using their works to begin with, and it doesn’t change that it serves as a way to undercut them, providing a cheaper replacement for their work. Copyright law as it was, wasn’t envisioned for a world where machine learning exists. It doesn’t really solve the problem to say that technically it’s not supposed to cover ideas and styles. The creators will be struggling just the same.

    Either the law will need to emphasize the value of human autorship first, or we will need to go through drastic socioeconomic changes to ensure that these creators will be able to keep creating despite losing market to AI. Otherwise, to simply say that AI gets to do this and change nothing else, will cause enormous damage to all sort of creative careers and wider culture. Even AI will become more limited with less fresh new creators to learn elements from.

    • Even_Adder@lemmy.dbzer0.comEnglish
      64·
      1 year ago

      The system doesn’t get personhood, it is your tool, and as said in the article:

      Fair use protects reverse engineering, indexing for search engines, and other forms of analysis that create new knowledge about works or bodies of works. Here, the fact that the model is used to create new works weighs in favor of fair use as does the fact that the model consists of original analysis of the training images in comparison with one another.

      It is your right, not the system’s you’re upholding.

      • TwilightVulpine@lemmy.worldEnglish
        62·
        1 year ago

        There is a difference between “analyzing” and derivating. The authorship of AI-created works is also not the user’s, it takes more than a prompt for that, and that seems to be the conclusion courts are leaning towards.

        Still, even if that turns out to be technically correct, it still doesn’t help the creators getting undercut who might be driven out of their careers by AI.

          • TwilightVulpine@lemmy.worldEnglish
            61·
            1 year ago

            They do specify that the human’s involvement needs to be more extensive than prompting for a certain image or text. The output itself is not copyrightable. If we are speaking about the process of “analysis” that the ML model does, then the user does not get the rights over it.

            This discussion is becoming increasingly overly specific and getting away from my point. My sole concern in all this is what happens to the artists who’ll have to compete with AI?

            • Even_Adder@lemmy.dbzer0.comEnglish
              31·
              1 year ago

              They do specify that the human’s involvement needs to be more extensive than prompting for a certain image or text. The output itself is not copyrightable. If we are speaking about the process of “analysis” that the ML model does, then the user does not get the rights over it.

              It says :

              In other cases, however, a work containing AI-generated material will also contain sufficient human authorship to support a copyright claim. For example, a human may select or arrange AI-generated material in a sufficiently creative way that “the resulting work as a whole constitutes an original work of authorship.” Or an artist may modify material originally generated by AI technology to such a degree that the modifications meet the standard for copyright protection.

              And you do get rights to your own original analysis of data. That isn’t even in question.

              This discussion is becoming increasingly overly specific and getting away from my point. My sole concern in all this is what happens to the artists who’ll have to compete with AI?

              I guess all I have to say here is that generative models are a free and open source tool anyone can use. It took us 100,000 years to get from cave drawings to Leonard Da Vinci. This is just another step, like Camera Obscura.

              • TwilightVulpine@lemmy.worldEnglish
                41·
                1 year ago

                When you call the output itself “analysis”, that’s not what they say.

                In February 2023, the Office concluded that a graphic novel comprised of human-authored text combined with images generated by the AI service Midjourney constituted a copyrightable work, but that the individual images themselves could not be protected by copyright

                This is in your own link. Simply prompting Midjourney doesn’t get the user copyright.

                I guess all I have to say here is that generative models are a free and open source tool anyone can use. It took us 100,000 years to get from cave drawings to Leonard Da Vinci. This is just another step, like Camera Obscura.

                That is not something many of those people whose work is being used to enable it even want to use. Not to mention, if AI art were to be the “next evolution” in media, which it isn’t since it output the same medium, there wouldn’t be a need for as many AI prompters as there are artists right now. This glosses over the issue entirely.

                • Even_Adder@lemmy.dbzer0.comEnglish
                  1·
                  1 year ago

                  This is in your own link. Simply prompting Midjourney doesn’t get the user copyright.

                  There’s more to generative models than just prompting. In that specific case, the images were generated with just a prompt because Midjourney doesn’t have the tools to let you do anything else.

                  That is not something many of those people whose work is being used to enable it even want to use.

                  It might not be for everyone, but there are already plenty of artists leveraging these new techniques.

                  Not to mention, if AI art were to be the “next evolution” in media, which it isn’t since it output the same medium, there wouldn’t be a need for as many AI prompters as there are artists right now. This glosses over the issue entirely.

                  That might be the case, but people often say that most art is only appeals to a few people, so just like Source Film Maker allowed more animations to be made, I expect the same kind of widening of scope for projects solo artists can make. You can already see this happening already. Not everyone has the time or motivation to do something as simple as make their own sandwich, I don’t think they’re going to want to sit down and hammer out a picture or whatever themselves.