Do artists have a legal claim over datasets for AI usage?

I am not a legal expert, but the law seem to be clear transformative and for learning purposes are covered under the law for US and EU:

EU:
>The Computer Programs Directive stipulates a mandatory exception in article 5(1) that acts of permanent or temporary reproduction and acts of translation, adaption and other alteration of a computer program shall not require authorisation by the rights holder ‘where they are necessary for the use of the computer program by the lawful acquirer in accordance with its intended purpose, including for error correction’.
https://www.lexology.com/library/detail.aspx?g=1013cebf-9e8e-41a0-bac0-595f0e04133b

US:
>Courts look at how the party claiming fair use is using the copyrighted work, and are more likely to find that nonprofit educational and noncommercial uses are fair.
>Additionally, “transformative” uses are more likely to be considered fair. Transformative uses are those that add something new, with a further purpose or different character, and do not substitute for the original use of the work.
https://www.copyright.gov/fair-use/

Isn't AI is fully covered under US and EU transformative and fair use laws?

ChatGPT Wizard Shirt $21.68

Beware Cat Shirt $21.68

ChatGPT Wizard Shirt $21.68

  1. 1 year ago
    Anonymous

    Maybe i worded the title incorrectly, i mean if artists have a valid claim of copyright violation to stable diffusion training.

    US and EU law seem to heavily imply they have no case.

    • 1 year ago
      Anonymous

      That's because they don't.
      They can't achieve any of their goals without shooting their foot, ankle, and anything below the knee completely off their bodies.

  2. 1 year ago
    Anonymous

    we will reach a point when the AI itself owns the copyright

    • 1 year ago
      Anonymous

      The AI doesn't already own the copyright?

    • 1 year ago
      Anonymous

      The AI doesn't already own the copyright?

      Only people can own copyrights (corporations are people too)

      • 1 year ago
        Anonymous

        >(corporations are people too)
        this is a good thing, it makes liberals seethe so hard.

        • 1 year ago
          Anonymous

          liberals love corporate personhood, so long as the corporation is liberal too
          >private companies can do what they want

  3. 1 year ago
    Anonymous

    What the law currently says is not relevant in the face of a new technology. Legislatures exist precisely to adapt the letter of the law to conform to it's spirit.

    • 1 year ago
      Anonymous

      >What the law currently says is not relevant
      It absolutely is in the US. Ex post facto law is explicitly forbidden by the Constitution.

  4. 1 year ago
    Anonymous

    The law this, the law that. Who cares?
    The law is whatever law enforcement does. Law enforcement obeys political pressure. Politicians obey financial incentives. Guess who provides those?
    When strict copyright enforcement is profitable, violators will be ritually executed (see Disney). When copyright violations are profitable, everything is fair use (see Microsoft).
    Always has been, etc.

  5. 1 year ago
    Anonymous

    I'm no lawyer but I can't imagine artists could have claim over AI dataset.
    The actual artworks aren't in the dataset, it's not just that you can't get at them, they don't exist.
    It's kind of like if you went to a museum and wrote down a description of the piece you saw, would the artist have a case against you publishing it or using this text as you see fit?
    Of course not.
    Even if that person did something like release a for-profit book containing these text descriptions of artworks they'd have no claim on it.
    Hell, it's not even transformative as I understand it, it's kind of like painting a picture inspired by a song.
    Now if those descriptions were very precise and detailed someone could get that book and use these descriptions to create his own artwork based on the descriptions in that book.
    Yes, his work would be based on something that's based on these artworks, but it's not "copying".
    AI does exactly that, except instead of a description per artwork it's matrices and vectors describing a "typical" tag, and most importantly it does it very fast and based off of a LOT of data.
    So to some "artists" it feels like cheating that a machine does what they spent years learning how to do effortlessly.

    I say "artists" because the only ones who are butthurt by SD are morons who think SD makes a collage out of real pictures and those who make their living making generic garbage and could be made obsolete with present day SD.
    Anyone with half a brain can see SD is very impressive but it won't make anyone but the most expendable of creators.
    Its biggest fault is precision, SD can do things very well if you don't care about minute details that add a lot of depth.
    Hell, even furry artists aren't in danger (yet), Furries pay big money to have precisely their OC drawn.
    The only ones that should rightfully be freaked out are those that draw simplistic, generic things.

    • 1 year ago
      Anonymous

      >The actual artworks aren't in the dataset, it's not just that you can't get at them, they don't exist.
      Wait what?
      The dataset (LAION) does not have actual images there?
      Then from where does the model learn the image-text matching?
      I was almost sure LAION has images (that's why they say they will honor any request to take down as long as it is valid).

      Actually, LAION themselves say so:
      >Overall, the dataset contains 2.32 billion images with English text, 2.26 billion with text from other languages, and 1.27 billion whose text language could not be determined unambiguously.

      • 1 year ago
        Anonymous

        >The dataset (LAION) does not have actual images there?
        While training the images are there, but when you deploy a trained solution the model doesn't have any pictures in it.
        It only needs the images to create a "typical" representation of a tag based on all the examples related to that tag it could fine.
        The thing you actually use to make new pictures doesn't have any images in it, just matrices describing what a typical image of that tag looks like.
        If the images can be found online and viewed by anyone I can't imagine there being a case against someone clicking "save as"
        If they were distributing these pictures as part of the finished model you might argue they're distributing their work unlawfully, but as it is the guy-writing-a-description-of-a-painting analogy stands.

      • 1 year ago
        Anonymous

        a ledditor has it explained really simply

      • 1 year ago
        Anonymous

        >Wait what?
        >The dataset (LAION) does not have actual images there?

        LAION publishes the image URL/caption pairs. They scrape the URLs from the internet at large, download the images themselves to generate the captions, but only distribute the text pairs.
        They don't keep/show/distribute the images, the dataset just links to them.
        You're expected to download the images yourself (as they did) to do your own processing on them (IE, train the AI using the images with the captions that LAION provide).
        Apparently the total size of images is about 240TB.

  6. 1 year ago
    Anonymous

    how can i make the ** text appear in a darker color

  7. 1 year ago
    Anonymous

    No artist is created in a vacuum. A lawsuit could not go ahead, as it would bring into question all art and artists.

  8. 1 year ago
    Anonymous

    >Do artists have a legal claim over datasets for AI usage?
    Short answer, no.
    Long answer, no, you can abuse legal loopholes like "for research, for academic purposes, education" etc.
    It is also a question of how your AI database came to be, how were these materials procured, were they from faceberg or instagoy, etc? take the L.
    Under what server, country are these works made and posted? what due diligence did the artist do for his artworks (as in what papers and docs does he have for his works)
    its a lot of "what ifs" to expand upon, but generally no, its hard to prove anything, not worth thinking about

  9. 1 year ago
    Anonymous

    no, and if theu were given claim over it then by extension you would have claim over youtube's and google's algorithims, for example, because your browser usage was monitored and used to train them, without your prior consent. in fact, this is still done without your consent.

  10. 1 year ago
    Anonymous

    It doesn't matter. If the law prohibits it, the law will change or will never be enforced. The will of the people trots toward self destruction and will do what it takes to pave the way for an AI overlord.

  11. 1 year ago
    Anonymous

    Im unironically a lawyer, and nope they dont

    • 1 year ago
      Anonymous

      I'm an ironic lawyer and I concur

  12. 1 year ago
    Anonymous

    There could be some claim from AI models memorizing and outputting images from the training dataset. But, there's no way random artists are going to win a case versus AI researchers, when most of them are backed by corporations like Microsoft, Google, or Nvidia

  13. 1 year ago
    Anonymous

    >copyright cases in the future will be settled based on the train/val loss diagram

  14. 1 year ago
    Anonymous

    AI art is a compelling reason to doubt that not only the artists, but the state has a legitimate claim to 'intellectual property'

  15. 1 year ago
    Anonymous

    Art is its own reward, and imitation is the most sincerest form of flattery

    just embrace that knowledge is owned by the universe, or adapt and begin focusing on not creating pieces but mediums to train your own AI's to finish it off.

  16. 1 year ago
    Anonymous

    I suspect once AI begins inching too close to the sun--er, music-- we'll have an abrupt crackdown on all AI generated content.

Leave a Reply to Anonymous Cancel reply

Your email address will not be published. Required fields are marked *