4 ms·

This seems obvious, and the correct decision. Copyright requires an author, a purely mechanically produced work by definition does not have an author.

brucethemoose2 4y ago
Maybe for an absolute case like this, but theres a grey area where models take a mostly human made input image. What if, for instance, the foreground is hand drawn and distinct while the background is machine generated? Or vice versa?
- JohnFen 4y ago
  In past court cases that resembled those situations, I believe the courts have ruled that it depends on what the bulk of the work consisted of (to oversimplify). Worst case, I think, any ambiguity would be around how much is human work. I could imagine an extreme situation where you created a work modified by AI where the AI contribution was great enough to turn it into a new work. I'd expect in such a situation, a court might rule that your original work was copyrightable, but the derivative work is not. So there is an area where things are not clear-cut, but it appears to me to be edge cases. Not surprising with something new on the scene. I expect future court rulings will clarify things.
courseofaction 4y ago
If I draw a picture with a pencil, did the pencil draw the picture? My major concern is that this doesn't protect artists - it takes the rights away from the individual creator or user, even if they're using these AI tools, and moves towards the model being the creator. This is a step towards centralizing ownership of human creative output based on the tools used, rather than the creative work or output.
- lcnPylGDnU4H9OF 4y ago
  This is an interesting call-out because a pencil is a (simple) machine which is used to create drawings with human input. So this ruling suggests that a sufficiently complex machine, even with human input, is not being used to create drawings but is instead the thing which is creating the drawings. That "sufficiently complex" line is being drawn somewhere between Photoshop and Stable Diffusion.
  - courseofaction 4y ago
    I'm not sure the complexity is sufficient to explain the decision, I think it's more the 'type' of work. Many things have insanely complex manufacturing processes, but we never credit microchips to the factory. Spitballing, but I think we are conceptualizing creative work as somehow from other kinds of work, and that creative work loses its specialness (especially in the production of art and ideas) when augmented with tools. As someone who works in a creative industry, this is nonsense: all kinds of art are majority skill and work. In my opinion, it's the use of art that confers its value and 'specialness'. An image generated and shown to no one is not equal to an image generated and used as part of graphic novel.
    - lcnPylGDnU4H9OF 4y ago
      I considered how one might give a prompt (the same prompt) at two different times and get two different images which started me towards thinking about a similar result in the context of other image editing software (GIMP, Photoshop, probably other things I haven't heard of). > ... creative work loses its specialness (especially in the production of art and ideas) when augmented with tools. I think about image editing software which does "fuzzy" stuff. If I don't drag my mouse in exactly the same way twice, I will get different results, however slight, when editing images but the end result is always something "I created". I do tend to land on the side that these things use different skills, which opens up these arguments that "generative" is not "creative". I just see a lot of gray when I consider these things in detail. > it's the use of art This is a very succinct way to put it. Your last paragraph hits home.
  - pishpash 4y ago
    If you read the USCO administrative letter [1], it wasn't complexity that killed the claim, it was stochasticity. That's a much shakier ground to stand on and only serves to highlight the deciding officer's inability to grasp how creative input, partial information, and stochasticity do not conflict with each other. The decision is full of holes because of the officer's narrow conception of mechanical devices. Probably a boomer. [1] https://fingfx.thomsonreuters.com/gfx/legaldocs/klpygnkyrpg/AI%20COPYRIGHT%20decision.pdf https://fingfx.thomsonreuters.com/gfx/legaldocs/klpygnkyrpg/... Instead, as explained above, Midjourney begins the image generation process with a field of visual “noise,” which is refined based on tokens created from user prompts that relate to Midjourney’s training database. The information in the prompt may “influence” generated image, but prompt text does not dictate a specific result... Because of the significant distance between what a user may direct Midjourney to create and the visual material Midjourney actually produces, Midjourney users lack sufficient control over generated images to be treated as the “master mind” behind them.
cypress66 4y ago
There is nothing obvious here. If I draw the image using photoshop, including many of its ai tools, most would say it's copyrightable. If I take a photo and use an AI denoiser (your phone already does), most would say it's copyrightable. That denoiser could be stable diffusion (it can be used for "traditional denoising"). Is that copyrightable? If I infill part of my manual drawing with SD, is that copyrightable? What if I draw a sketch, and use control net to get an image of that? There is definitely no clear line that would make this decision "obvious".
- JohnFen 4y ago
  All of your examples are clearly copyrightable, because in all of them, you were the one who created the works (and I assume that you're a human). There's no legal ambiguity in any of your examples. The use of AI or any other sorts of tool to modify or improve your work doesn't alter that. This is clearly shown in legal precedents.
  - cypress66 4y ago
    So the control net example (which is image to image) is copyrightable, but the OP case (text to image) is not? Then the line to draw is that when there's a change in modality, it's not copyrightable? Most people are probably going to end up using something like controlnet anyway because of the extra control it gives you.
    - JohnFen 4y ago
      No, I don't think that it has to do with modality. Again, IANAL, but as I understand it, the legal line is how much of the art was created by a human. There is certainly a gray area here that will have to be hashed out, but I don't think it's as large as you think it is. The rulings are going to be about whether or not the creative work is the work of a human. I imagine that a case can be made that describing a work to a machine and having the machine engage in the creativity part means the creative work is done by the machine, not the human. But if the human did the creative work and the machine modified it, then the work is done by the human. Of course, we're all speculating here. I have no idea how the courts will rule on these things. I could even see the necessity of Congress making some clarifying laws. Also, understand that I am not expressing what I personally think should be the case. I'm just trying to work out what is, or is likely, to be the case.
hex4def6 4y ago
It doesn't seem that obvious to me. Cameras and computer software really blur the line. A photo is a "mechanically produced work," and yet we still afford them copyright. If I put a camera in the middle of a busy shopping mall, set to capture images every 10 seconds, and them come back later and choose the top 5 images, I'm reasonably sure those would be granted copyright protection (If you disagree, replace that example with a timelapse of the sky). I've played around with landscape photography, and I can confidently say that I've gotten some stunning photographs that were mainly luck and being in the right place at the right time. Conversely, I've spent multiple hours trying to wrangle Stable Diffusion into giving me the image I want through different prompts, generating VAEs / Hypernets, ControlNet, img2img, etc. If we were judging "creative effort", I'd argue the latter required significantly more on my part. Likewise, if I take Blender 3D, and start creating a cityscape I'm pretty sure that would be considered for copyright. But what if part of that process is creating a python script that randomly selects a building object to place in a city block grid? Is that algorithmic work grounds for losing copyright? What if instead of a random choice, it does it according to some process of my design (1 story buildings can't go next to skyscrapers, etc)? What if it's a ML algo that takes feedback from me based on which ones I accept / reject? What if it cycles through a set for each position, and lets me press Y/N for the choice? What setting up a bonsai in front of a white sheet and taking a photograph of it. Obviously, copyrightable. What about writing an algorithm that generates bonsai tree objects, and making a rendering of it? What about 3D printing it out, and then placing in where the real bonsai was, and taking a picture? I feel like I should have less claim to a copyright in the case of the bonsai photo (which I had zero input in growing or making) versus the 3D version (in which I created an algorithm to generate bonsai-like objects, which means it is "algorithmically created" and not worthy of a copyright). It seems from my perspective that these works should be afforded copyright. There was at least a modicum of creative effort that went into the prompt selection, generation, and image choice. Enough that it meets the minimum effort that taking a random snapshot of a city street would require, for instance, and especially if effort was taken in generating the Hypernet / embedding, posing through ControlNet, etc. And more importantly, enough that the author decided it was interesting enough to keep and publish (and that a potential third-party also considers valuable enough to yoink a copy of). You've "created" / "discovered" / "put to paper" something that did not exist before, and which clearly has value. I guess I'm having trouble understanding what the downsides of allowing these to be copyrighted are.
pishpash 4y ago
Nothing is purely mechanically produced. There is always some input. In this case, the images were produced from prompts and entropy. The author clearly should own the prompts. Nobody should own the entropy. So who owns the output, if not the author?
- saurik 4y ago
  If I ask you to make 100 images of "a monkey sitting at a computer realizing it is out of a job" and then I select the best one that best matched what I intended, I do not own the copyright: you do. I can instead compensate you something to agree to assign the copyright of the image (or even all of the images) to me, and structure it as a "work for hire", but the expression of my idea into an image is the copyrightable work there, not my idea itself; and, neither, does curation of your works convert any of your work into my work. Now, if I ask an AI to give me the same 100 images and merely select the best one, in some sense maybe the AI owns the copyright?... but we don't want to support that concept, as that AI doesn't have rights, so I guess no one owns the copyright. If you want copyright over its work, I guess you need to make sure it has some basic inhuman rights to begin with and then be prepared to pay it some fair compensation for its copyright.
  - pishpash 4y ago
    Suppose you wrote this AI yourself, and you trained it on data you own. Does it change anything? It makes no sense that because there is an AI in the middle, that you lose copyright over the output. What if you trained it on public domain data? I think, in the case of an AI without rights being the upstream of a piece of work (and no other licensing issues with the owner of the AI), the next proximate entity to the work owns the rights, so it would be the author/curator, whatever you want to call it. It doesn't matter how little they did. It's that they did something and nobody else is claiming rights. There is no point referring back to a world where only humans are interacting with each other. It's a dangerous precedent to say that AI output receives no copyright, because it is the same as what you're implying, that AI is not a tool, but an entity that siphoned off some of the rights, like maybe a child or a slave. If the USCO wants to declare that then they should do so, otherwise the case is wrongly decided and needs to go in front of a more intelligent judge.
mc32 4y ago
So what happens if the infinite monkey theorem is applied to images and AI, having ben trained on copyrighted and CC images eventually produces copies that pass for copyrighted originals...
- avereveard 4y ago
  You can think copyright as a flavor of bits, something that exists in it's own acid, along it's own axis. infinite monkeys can't produce copyrighted content (as you can't for example copyright all cbinations of 8 notes and claim infringement on all future music) but distribution of their work can infringe on someone else copyright. You can take inspiration from their work to produce an original, if you put enough effort in it. That doesn't absolve the original of infringements of characters or other intellectual properties.
  - mc32 4y ago
    Does this infer that someone not associated with an ART AI, take a published output of an AI then render it with their own tools and then claim copyright on their "handmade" artwork that was uncopyrightable as an AI output?
    - avereveard 4y ago
      https://www.jcms-journal.com/articles/10.5334/jcms.1021217/ https://www.jcms-journal.com/articles/10.5334/jcms.1021217/ the artwork need to be transformative, non transformative artwork don't generate new copyright claims.
tenebrisalietum 4y ago
A mechanically reproduced work is still covered and can be infringed upon. Why is what AI does not considered reproduction of existing works - it seems like automated derivative works.

brucethemoose2 4y ago

Maybe for an absolute case like this, but theres a grey area where models take a mostly human made input image. What if, for instance, the foreground is hand drawn and distinct while the background is machine generated? Or vice versa?

JohnFen 4y ago
In past court cases that resembled those situations, I believe the courts have ruled that it depends on what the bulk of the work consisted of (to oversimplify). Worst case, I think, any ambiguity would be around how much is human work. I could imagine an extreme situation where you created a work modified by AI where the AI contribution was great enough to turn it into a new work. I'd expect in such a situation, a court might rule that your original work was copyrightable, but the derivative work is not. So there is an area where things are not clear-cut, but it appears to me to be edge cases. Not surprising with something new on the scene. I expect future court rulings will clarify things.

courseofaction 4y ago

If I draw a picture with a pencil, did the pencil draw the picture? My major concern is that this doesn't protect artists - it takes the rights away from the individual creator or user, even if they're using these AI tools, and moves towards the model being the creator. This is a step towards centralizing ownership of human creative output based on the tools used, rather than the creative work or output.

lcnPylGDnU4H9OF 4y ago
This is an interesting call-out because a pencil is a (simple) machine which is used to create drawings with human input. So this ruling suggests that a sufficiently complex machine, even with human input, is not being used to create drawings but is instead the thing which is creating the drawings. That "sufficiently complex" line is being drawn somewhere between Photoshop and Stable Diffusion.
- courseofaction 4y ago
  I'm not sure the complexity is sufficient to explain the decision, I think it's more the 'type' of work. Many things have insanely complex manufacturing processes, but we never credit microchips to the factory. Spitballing, but I think we are conceptualizing creative work as somehow from other kinds of work, and that creative work loses its specialness (especially in the production of art and ideas) when augmented with tools. As someone who works in a creative industry, this is nonsense: all kinds of art are majority skill and work. In my opinion, it's the use of art that confers its value and 'specialness'. An image generated and shown to no one is not equal to an image generated and used as part of graphic novel.
  - lcnPylGDnU4H9OF 4y ago
    I considered how one might give a prompt (the same prompt) at two different times and get two different images which started me towards thinking about a similar result in the context of other image editing software (GIMP, Photoshop, probably other things I haven't heard of). > ... creative work loses its specialness (especially in the production of art and ideas) when augmented with tools. I think about image editing software which does "fuzzy" stuff. If I don't drag my mouse in exactly the same way twice, I will get different results, however slight, when editing images but the end result is always something "I created". I do tend to land on the side that these things use different skills, which opens up these arguments that "generative" is not "creative". I just see a lot of gray when I consider these things in detail. > it's the use of art This is a very succinct way to put it. Your last paragraph hits home.
- pishpash 4y ago
  If you read the USCO administrative letter [1], it wasn't complexity that killed the claim, it was stochasticity. That's a much shakier ground to stand on and only serves to highlight the deciding officer's inability to grasp how creative input, partial information, and stochasticity do not conflict with each other. The decision is full of holes because of the officer's narrow conception of mechanical devices. Probably a boomer. [1] https://fingfx.thomsonreuters.com/gfx/legaldocs/klpygnkyrpg/AI%20COPYRIGHT%20decision.pdf https://fingfx.thomsonreuters.com/gfx/legaldocs/klpygnkyrpg/... Instead, as explained above, Midjourney begins the image generation process with a field of visual “noise,” which is refined based on tokens created from user prompts that relate to Midjourney’s training database. The information in the prompt may “influence” generated image, but prompt text does not dictate a specific result... Because of the significant distance between what a user may direct Midjourney to create and the visual material Midjourney actually produces, Midjourney users lack sufficient control over generated images to be treated as the “master mind” behind them.

cypress66 4y ago

There is nothing obvious here. If I draw the image using photoshop, including many of its ai tools, most would say it's copyrightable. If I take a photo and use an AI denoiser (your phone already does), most would say it's copyrightable. That denoiser could be stable diffusion (it can be used for "traditional denoising"). Is that copyrightable? If I infill part of my manual drawing with SD, is that copyrightable? What if I draw a sketch, and use control net to get an image of that? There is definitely no clear line that would make this decision "obvious".

JohnFen 4y ago
All of your examples are clearly copyrightable, because in all of them, you were the one who created the works (and I assume that you're a human). There's no legal ambiguity in any of your examples. The use of AI or any other sorts of tool to modify or improve your work doesn't alter that. This is clearly shown in legal precedents.
- cypress66 4y ago
  So the control net example (which is image to image) is copyrightable, but the OP case (text to image) is not? Then the line to draw is that when there's a change in modality, it's not copyrightable? Most people are probably going to end up using something like controlnet anyway because of the extra control it gives you.
  - JohnFen 4y ago
    No, I don't think that it has to do with modality. Again, IANAL, but as I understand it, the legal line is how much of the art was created by a human. There is certainly a gray area here that will have to be hashed out, but I don't think it's as large as you think it is. The rulings are going to be about whether or not the creative work is the work of a human. I imagine that a case can be made that describing a work to a machine and having the machine engage in the creativity part means the creative work is done by the machine, not the human. But if the human did the creative work and the machine modified it, then the work is done by the human. Of course, we're all speculating here. I have no idea how the courts will rule on these things. I could even see the necessity of Congress making some clarifying laws. Also, understand that I am not expressing what I personally think should be the case. I'm just trying to work out what is, or is likely, to be the case.

hex4def6 4y ago

It doesn't seem that obvious to me. Cameras and computer software really blur the line. A photo is a "mechanically produced work," and yet we still afford them copyright. If I put a camera in the middle of a busy shopping mall, set to capture images every 10 seconds, and them come back later and choose the top 5 images, I'm reasonably sure those would be granted copyright protection (If you disagree, replace that example with a timelapse of the sky). I've played around with landscape photography, and I can confidently say that I've gotten some stunning photographs that were mainly luck and being in the right place at the right time. Conversely, I've spent multiple hours trying to wrangle Stable Diffusion into giving me the image I want through different prompts, generating VAEs / Hypernets, ControlNet, img2img, etc. If we were judging "creative effort", I'd argue the latter required significantly more on my part. Likewise, if I take Blender 3D, and start creating a cityscape I'm pretty sure that would be considered for copyright. But what if part of that process is creating a python script that randomly selects a building object to place in a city block grid? Is that algorithmic work grounds for losing copyright? What if instead of a random choice, it does it according to some process of my design (1 story buildings can't go next to skyscrapers, etc)? What if it's a ML algo that takes feedback from me based on which ones I accept / reject? What if it cycles through a set for each position, and lets me press Y/N for the choice? What setting up a bonsai in front of a white sheet and taking a photograph of it. Obviously, copyrightable. What about writing an algorithm that generates bonsai tree objects, and making a rendering of it? What about 3D printing it out, and then placing in where the real bonsai was, and taking a picture? I feel like I should have less claim to a copyright in the case of the bonsai photo (which I had zero input in growing or making) versus the 3D version (in which I created an algorithm to generate bonsai-like objects, which means it is "algorithmically created" and not worthy of a copyright). It seems from my perspective that these works should be afforded copyright. There was at least a modicum of creative effort that went into the prompt selection, generation, and image choice. Enough that it meets the minimum effort that taking a random snapshot of a city street would require, for instance, and especially if effort was taken in generating the Hypernet / embedding, posing through ControlNet, etc. And more importantly, enough that the author decided it was interesting enough to keep and publish (and that a potential third-party also considers valuable enough to yoink a copy of). You've "created" / "discovered" / "put to paper" something that did not exist before, and which clearly has value. I guess I'm having trouble understanding what the downsides of allowing these to be copyrighted are.

pishpash 4y ago

Nothing is purely mechanically produced. There is always some input. In this case, the images were produced from prompts and entropy. The author clearly should own the prompts. Nobody should own the entropy. So who owns the output, if not the author?

saurik 4y ago
If I ask you to make 100 images of "a monkey sitting at a computer realizing it is out of a job" and then I select the best one that best matched what I intended, I do not own the copyright: you do. I can instead compensate you something to agree to assign the copyright of the image (or even all of the images) to me, and structure it as a "work for hire", but the expression of my idea into an image is the copyrightable work there, not my idea itself; and, neither, does curation of your works convert any of your work into my work. Now, if I ask an AI to give me the same 100 images and merely select the best one, in some sense maybe the AI owns the copyright?... but we don't want to support that concept, as that AI doesn't have rights, so I guess no one owns the copyright. If you want copyright over its work, I guess you need to make sure it has some basic inhuman rights to begin with and then be prepared to pay it some fair compensation for its copyright.
- pishpash 4y ago
  Suppose you wrote this AI yourself, and you trained it on data you own. Does it change anything? It makes no sense that because there is an AI in the middle, that you lose copyright over the output. What if you trained it on public domain data? I think, in the case of an AI without rights being the upstream of a piece of work (and no other licensing issues with the owner of the AI), the next proximate entity to the work owns the rights, so it would be the author/curator, whatever you want to call it. It doesn't matter how little they did. It's that they did something and nobody else is claiming rights. There is no point referring back to a world where only humans are interacting with each other. It's a dangerous precedent to say that AI output receives no copyright, because it is the same as what you're implying, that AI is not a tool, but an entity that siphoned off some of the rights, like maybe a child or a slave. If the USCO wants to declare that then they should do so, otherwise the case is wrongly decided and needs to go in front of a more intelligent judge.

mc32 4y ago

So what happens if the infinite monkey theorem is applied to images and AI, having ben trained on copyrighted and CC images eventually produces copies that pass for copyrighted originals...

avereveard 4y ago
You can think copyright as a flavor of bits, something that exists in it's own acid, along it's own axis. infinite monkeys can't produce copyrighted content (as you can't for example copyright all cbinations of 8 notes and claim infringement on all future music) but distribution of their work can infringe on someone else copyright. You can take inspiration from their work to produce an original, if you put enough effort in it. That doesn't absolve the original of infringements of characters or other intellectual properties.
- mc32 4y ago
  Does this infer that someone not associated with an ART AI, take a published output of an AI then render it with their own tools and then claim copyright on their "handmade" artwork that was uncopyrightable as an AI output?
  - avereveard 4y ago
    https://www.jcms-journal.com/articles/10.5334/jcms.1021217/ https://www.jcms-journal.com/articles/10.5334/jcms.1021217/ the artwork need to be transformative, non transformative artwork don't generate new copyright claims.

tenebrisalietum 4y ago

A mechanically reproduced work is still covered and can be infringed upon. Why is what AI does not considered reproduction of existing works - it seems like automated derivative works.