21 ms·

How is this different to Googling “robot cop” or “video game plumber” and being served copyrighted material? Is it because Google will link to the image source

How is this different to Googling “robot cop” or “video game plumber” and being served copyrighted material? Is it because Google will link to the image source? Or does the infringement begin when I use the image for gain, or claim it as my own? Perhaps it is because Google was allowed to crawl the page with the original image, so presenting them with a link is fine?

geraldwhen 3y ago
Looking at a copyrighted image posted by an author is not infringement. Printing that image onto a shirt and selling it is infringement. That’s what OpenAI is doing.
- golol 3y ago
  But OpenAI is not selling the rights to any images, or are they? When I pay for Dall-E, does the contract give me any rights for a work? If not then there is no issue.
  - AlienRobot 3y ago
    Copyright is the right to copy things. You don't even need to sell it. This is why Wikipedia images are mostly Copyleft images. Google gets a pass because nobody is suing Google. When people try to sue Google, Google simply stops indexing them and then they start begging Google to infringe their copyright again.
    - golol 3y ago
      This interpretation of copyright only made sense while the transfer and storage of information was tied to physical objects. That time is long and we dont consider it infringement to remember a media or reproduce it at home. Furthermore, we are now entering an era where the production of information is also being untied from physical objects, so it'll only get worse for copyright. I made a post to diacuss this stuff as I find it interesting right now and want to hear more opinions.
      - AlienRobot 3y ago
        I completely disagree. Tech exceptionalism makes no sense. We should be making technology to ensure people have their rights protected, not to come up with technobabble excuses to pretend such rights don't exist. Just because people having been posting memes and reposting pictures and comics with cropped credits and pirating stuff that doesn't mean any of this is legal. Legality isn't about what you can technically do thanks to how the computer works, or how HTTP works, or how the laws of physics work. Legality is just about what is law and what is not. Redistributing copyrighted works without license has always been illegal. People don't get sued for it all the time because it isn't worth the hassle and most small time copyright holders simply lack the resources to pursuit action against random Internet strangers across the Internet. That doesn't mean they don't have a copyright, they merely chose to not exercise it. And that's not a W for technology. That's literally just more abuse than a person can cope with. It's an L for society. That's like if you started getting so much spam in your e-mail that you gave up marking them as spam. That doesn't make them not spam. For example, if I wrote something in my blog and someone made a scrapper that reposted it entirely in their website full of stolen posts, I could take legal action against them. For a blog post. For something I wrote on the Internet. That's my right. But imagine how much time I'd have to spend to do this. It would be easier to check if Google has a way to tell someone stole my content and just get them delisted from Google than going through legal channels.
        golol 3y ago
        But I'm not talking about legality, I'm talking about what we should make the law to be. Just imagine memory implants become commonplace, shouldn't they be allowed to store copyrighted media you have consumed? If not how do you separate between your natural memory and the artificial one? How is it going to work?
  - noitpmeder 3y ago
    OpenAI is selling a service. In the terms of this service they explicitly reassign rights of the output to the user. So implicitly they believe they own the rights and are legally able to reassign them to you, a user of their service. In my view they do not own those rights originally and thus are unable to resign them.
pointlessone 3y ago
Google directs you to the original work. It doesn’t present you a derivative work based on the original. That is, original author, presumably, benefits from distribution. AI, on the other hand, slurps multiple original works, chews them up and gives you something average but close enough, and not any specific work in particular.
- ls612 3y ago
  Google shows snippets of copyrighted work all of the time, and it certainly ingests the entire copyrighted work when googlebot views the page to index it. The only real issue here is that NYT figured out a way to get bingbot to look up an entire article from the internet and repeat it which may not be kosher. But if search engines can ingest the entire content of copyrighted works (subject to robots.txt) then I don't see why AI training should be different on that front. Of course, the real reason it is different is that it impacts different interest groups than search engines, and the rule of law is a sham. Creatives will do anything to ensure they don't get disrupted and can continue extracting rent from society, and have learned a lot of tools of rhetoric from their fancy colleges to put to use in that effort, compared to the industrial workers who got disrupted by automation a generation ago.
dkjaudyeqooe 3y ago
Search engines are ruled fair use because they use the copyrighted material in a limited way, they provide a public good and they benefit the copyright holder. Generative AI is more or less the opposite of that. It ingests the whole work, generates output that substitutes for the used work and profits the user of copyrighted work to the detriment of the copyright holder. Throw in the fact that it is purley a mechanical transformation of the copyrighted work and generative AI is on shaky ground.
- fallingknife 3y ago
  But transformative use is an exception to copyright. And I think it's going to be pretty hard to argue that the matrix of parameters inside an LLM is not sufficiently transformative from the input image.
  - FridgeSeal 3y ago
    If I run a thesaurus over a plagiarised text, it would be a long bow to draw to say that’s “transformative”. I feel like this “oh but it’s transformative” argument is becoming rapidly load-bearing in the context of LLM arguments and I don’t really see nearly enough justification for it.
  - regularfry 3y ago
    Legally, "transformative" means semantically, not pixel-level. It's hard to argue that all matrix transformations done by the LLM would be transformative in that sense.

geraldwhen 3y ago

Looking at a copyrighted image posted by an author is not infringement. Printing that image onto a shirt and selling it is infringement. That’s what OpenAI is doing.

golol 3y ago
But OpenAI is not selling the rights to any images, or are they? When I pay for Dall-E, does the contract give me any rights for a work? If not then there is no issue.
- AlienRobot 3y ago
  Copyright is the right to copy things. You don't even need to sell it. This is why Wikipedia images are mostly Copyleft images. Google gets a pass because nobody is suing Google. When people try to sue Google, Google simply stops indexing them and then they start begging Google to infringe their copyright again.
  - golol 3y ago
    This interpretation of copyright only made sense while the transfer and storage of information was tied to physical objects. That time is long and we dont consider it infringement to remember a media or reproduce it at home. Furthermore, we are now entering an era where the production of information is also being untied from physical objects, so it'll only get worse for copyright. I made a post to diacuss this stuff as I find it interesting right now and want to hear more opinions.
    - AlienRobot 3y ago
      I completely disagree. Tech exceptionalism makes no sense. We should be making technology to ensure people have their rights protected, not to come up with technobabble excuses to pretend such rights don't exist. Just because people having been posting memes and reposting pictures and comics with cropped credits and pirating stuff that doesn't mean any of this is legal. Legality isn't about what you can technically do thanks to how the computer works, or how HTTP works, or how the laws of physics work. Legality is just about what is law and what is not. Redistributing copyrighted works without license has always been illegal. People don't get sued for it all the time because it isn't worth the hassle and most small time copyright holders simply lack the resources to pursuit action against random Internet strangers across the Internet. That doesn't mean they don't have a copyright, they merely chose to not exercise it. And that's not a W for technology. That's literally just more abuse than a person can cope with. It's an L for society. That's like if you started getting so much spam in your e-mail that you gave up marking them as spam. That doesn't make them not spam. For example, if I wrote something in my blog and someone made a scrapper that reposted it entirely in their website full of stolen posts, I could take legal action against them. For a blog post. For something I wrote on the Internet. That's my right. But imagine how much time I'd have to spend to do this. It would be easier to check if Google has a way to tell someone stole my content and just get them delisted from Google than going through legal channels.
      - golol 3y ago
        But I'm not talking about legality, I'm talking about what we should make the law to be. Just imagine memory implants become commonplace, shouldn't they be allowed to store copyrighted media you have consumed? If not how do you separate between your natural memory and the artificial one? How is it going to work?
- noitpmeder 3y ago
  OpenAI is selling a service. In the terms of this service they explicitly reassign rights of the output to the user. So implicitly they believe they own the rights and are legally able to reassign them to you, a user of their service. In my view they do not own those rights originally and thus are unable to resign them.

pointlessone 3y ago

Google directs you to the original work. It doesn’t present you a derivative work based on the original. That is, original author, presumably, benefits from distribution. AI, on the other hand, slurps multiple original works, chews them up and gives you something average but close enough, and not any specific work in particular.

ls612 3y ago
Google shows snippets of copyrighted work all of the time, and it certainly ingests the entire copyrighted work when googlebot views the page to index it. The only real issue here is that NYT figured out a way to get bingbot to look up an entire article from the internet and repeat it which may not be kosher. But if search engines can ingest the entire content of copyrighted works (subject to robots.txt) then I don't see why AI training should be different on that front. Of course, the real reason it is different is that it impacts different interest groups than search engines, and the rule of law is a sham. Creatives will do anything to ensure they don't get disrupted and can continue extracting rent from society, and have learned a lot of tools of rhetoric from their fancy colleges to put to use in that effort, compared to the industrial workers who got disrupted by automation a generation ago.

dkjaudyeqooe 3y ago

Search engines are ruled fair use because they use the copyrighted material in a limited way, they provide a public good and they benefit the copyright holder. Generative AI is more or less the opposite of that. It ingests the whole work, generates output that substitutes for the used work and profits the user of copyrighted work to the detriment of the copyright holder. Throw in the fact that it is purley a mechanical transformation of the copyrighted work and generative AI is on shaky ground.

fallingknife 3y ago
But transformative use is an exception to copyright. And I think it's going to be pretty hard to argue that the matrix of parameters inside an LLM is not sufficiently transformative from the input image.
- FridgeSeal 3y ago
  If I run a thesaurus over a plagiarised text, it would be a long bow to draw to say that’s “transformative”. I feel like this “oh but it’s transformative” argument is becoming rapidly load-bearing in the context of LLM arguments and I don’t really see nearly enough justification for it.
- regularfry 3y ago
  Legally, "transformative" means semantically, not pixel-level. It's hard to argue that all matrix transformations done by the LLM would be transformative in that sense.