6 ms·

It could be, if the human uses their skills to draw infringing outputs!

a_bonobo 3y ago
hence the entire Andy Warhol case! >Echoing a recommendation from Joe Biden’s administration, the supreme court focused on the specific use that allegedly infringed Goldsmith’s copyright – a license of Warhol’s work to Condé Nast – and said it was not transformative because it served the same commercial purpose as Goldsmith’s photo: to depict Prince in a magazine. https://www.theguardian.com/artanddesign/2023/may/18/andy-warhol-copyright-prince-paintings-lawsuit https://www.theguardian.com/artanddesign/2023/may/18/andy-wa...
brookst 3y ago
So why are we even arguing about this new form of copyright that restricts who can learn from a work? If an AI model produces an infringing work and someone distributes it, sue them like you would a human doing the same thing. This idea of training being a violation of copyright is insane.
- mattigames 3y ago
  No, is not insane at all, a computer can generate millions of derivate copies in a second, a human has limited time for learning and for making artwork, so it doesn't make sense to compare them, not in the slightest. If suddenly there was a race of humans who could read and retain entire books in a matter of seconds (and other similar feats) the implications would be almost the same, as it would break basic expectations of time and personal resources that our society relies on, and a lot of people would be rightfully worried about such people entering the workforce.
  - Solvency 3y ago
    Computers can do all sorts of things humans can't at inhuman scale and speed. Why is AI leveraging the same training data humans do inherently bad just because it can do it better and faster than humans?
    - mattigames 3y ago
      Yeah, I'm a programmer, I'm aware of all that by definition. The issue here it's getting to he point where is inevitable that it will replace a huge amount of job positions in a tiny frame of time in ways society is not prepared for, if you think the amount of jobs at risk is anywhere in a similar scale we have seen before you haven't been paying attention; of course the flaw here lies on society's itself and what it means to earn a living, and not really over ML researchers or Adobe or any other company creating those AIs.
  - zirgs 3y ago
    Adobe trained their own model only on public domain images and on images that they own. Do you think that they should not be allowed to do that?
    - mattigames 3y ago
      That's a different discussion and should have wide ramifications with way more nuance that "allowed" or "not allowed", if suddenly tomorrow Adobe or any other were able to create a model that is able to replace any existing job position how we should it be managed? The societal ramifications are complex, is not binary and should be discussed by politicians and intellectuals, not by corp marketing teams.
      - zirgs 3y ago
        They can already replace stock image sites. Why pay getty/shutterstock an obscene amount of money for a photo if you can simply generate something similar with a prompt for free? I checked how much some images cost on getty images - and their prices are ridiculous - 100-1000eur or more. No wonder they are trying to sue stability.ai now.
  - astrange 3y ago
    Speed isn't an existing factor in copyright law.
    - mattigames 3y ago
      That is not a factor may have something to do with the fact the law was created before even sci-fi authors could have predicted machine learning and the huge difference it makes for creating artwork. Generating synthetic copies of someone voice, or even combining voices (e.g. a CG voice that is a equal parts Kurt Cobain's voice and Laine Staley's voice) are also not currently a factor in copyright law, but it will be.
  - brookst 3y ago
    Wait, so do your same opinions about copyright for AI also apply to copyright for corporations? How is employing 1000 people to generate derivative works any different from using AI?
    - mattigames 3y ago
      Because they have to hire those 1000 people and all that entails? They will have to weight if it's worth it to pay that many people, and if in the future they ever want to generate even more images they have to hire them again, instead of just re-running a program which costs them near nothing. The more I think about it the less similar are both processes.

hence the entire Andy Warhol case! >Echoing a recommendation from Joe Biden’s administration, the supreme court focused on the specific use that allegedly infringed Goldsmith’s copyright – a license of Warhol’s work to Condé Nast – and said it was not transformative because it served the same commercial purpose as Goldsmith’s photo: to depict Prince in a magazine. https://www.theguardian.com/artanddesign/2023/may/18/andy-warhol-copyright-prince-paintings-lawsuit https://www.theguardian.com/artanddesign/2023/may/18/andy-wa...

brookst 3y ago

So why are we even arguing about this new form of copyright that restricts who can learn from a work? If an AI model produces an infringing work and someone distributes it, sue them like you would a human doing the same thing. This idea of training being a violation of copyright is insane.

mattigames 3y ago
No, is not insane at all, a computer can generate millions of derivate copies in a second, a human has limited time for learning and for making artwork, so it doesn't make sense to compare them, not in the slightest. If suddenly there was a race of humans who could read and retain entire books in a matter of seconds (and other similar feats) the implications would be almost the same, as it would break basic expectations of time and personal resources that our society relies on, and a lot of people would be rightfully worried about such people entering the workforce.
- Solvency 3y ago
  Computers can do all sorts of things humans can't at inhuman scale and speed. Why is AI leveraging the same training data humans do inherently bad just because it can do it better and faster than humans?
  - mattigames 3y ago
    Yeah, I'm a programmer, I'm aware of all that by definition. The issue here it's getting to he point where is inevitable that it will replace a huge amount of job positions in a tiny frame of time in ways society is not prepared for, if you think the amount of jobs at risk is anywhere in a similar scale we have seen before you haven't been paying attention; of course the flaw here lies on society's itself and what it means to earn a living, and not really over ML researchers or Adobe or any other company creating those AIs.
- zirgs 3y ago
  Adobe trained their own model only on public domain images and on images that they own. Do you think that they should not be allowed to do that?
  - mattigames 3y ago
    That's a different discussion and should have wide ramifications with way more nuance that "allowed" or "not allowed", if suddenly tomorrow Adobe or any other were able to create a model that is able to replace any existing job position how we should it be managed? The societal ramifications are complex, is not binary and should be discussed by politicians and intellectuals, not by corp marketing teams.
    - zirgs 3y ago
      They can already replace stock image sites. Why pay getty/shutterstock an obscene amount of money for a photo if you can simply generate something similar with a prompt for free? I checked how much some images cost on getty images - and their prices are ridiculous - 100-1000eur or more. No wonder they are trying to sue stability.ai now.
- astrange 3y ago
  Speed isn't an existing factor in copyright law.
  - mattigames 3y ago
    That is not a factor may have something to do with the fact the law was created before even sci-fi authors could have predicted machine learning and the huge difference it makes for creating artwork. Generating synthetic copies of someone voice, or even combining voices (e.g. a CG voice that is a equal parts Kurt Cobain's voice and Laine Staley's voice) are also not currently a factor in copyright law, but it will be.
- brookst 3y ago
  Wait, so do your same opinions about copyright for AI also apply to copyright for corporations? How is employing 1000 people to generate derivative works any different from using AI?
  - mattigames 3y ago
    Because they have to hire those 1000 people and all that entails? They will have to weight if it's worth it to pay that many people, and if in the future they ever want to generate even more images they have to hire them again, instead of just re-running a program which costs them near nothing. The more I think about it the less similar are both processes.