I wonder what the implications are for AI generated code. The implications for that could be massive, including all code generated by tools such as co-pilot.
I was wondering the same. If you can feed a bunch of code into an AI and then use the generated code that's basically like a money mixer. Do you get to decompile things then and then feed them with existing code into an AI? Can you reuse the leaked Microsoft code when you mix it with other code into a dataset? Can a third party do it and you claim ignorance when you use it?
Legal matters don't exist in a vacuum. Short answer is: yes you can do all of those things. But some of them might land you in hot water if done in the context of a contract. Copyright is not absolute, it's about what potential claims people might have.
A. I get a copy of ChatGPT's code (via some legitimate means) and train it exclusively on my written work. I argue that all output from the dsr_%_ChatGPT instance is derivative from my work. I assert that when I prompt it, the results are also my work. Further, when someone else prompts it, the results are infringing on my work.
B. Same as A, but I train it 50% on my written work and 50% on jacquesm's written work. I assert that when I prompt it, the results always infringe on jacquesm, and if jacquesm prompts it, the results always infringe on me.
It's far less general than the headline suggests, to the point of it being misleading:
The human behind the AI did insist that they had absolutely no part in creating the image, and the decision is taking that at face value. No human creator - no copyright. Since it's purely based on their input, it also doesn't look at questions like "was the AI created using input material that's copyrighted and might that have relevance"
It has zero applicability on "AI art"/code/text/... created by a human using an "AI" as a tool to do so.
But somehow this gets paraded around every few months as being some important decision...
That brings the question: in order for a work to be copyrighted, does a human need to put "some" effort into generating that work (prompt engineering, tweaking knobs, etc.) or does it suffice to click a "generate" button?
Let's say a photographer sets up his camera, aims it, composes it, focuses it, etc. But he decides the light isn't quite right yet, so he goes for lunch. While he is away, his assistant clicks the shutter on a whim, capturing more or less the exact image what the photographer intended to capture. The resulting photograph's copyright belongs to the assistant, not the photographer.
This is in theory, of course. In practice, this would be difficult to prove, since the picture is likely to be on a roll of many other pictures for which the photographer does own the copyright. The assistant would need some definitive evidence that they clicked the shutter on that exact frame if they wanted credit for their work.
And that might lead to an argument that if the human doesn't hold the copyright the AI does, but then you are back to square one because you need to be a legal entity to hold copyright.
Also EU makes this different. Just taking a generic image. As in one that many other people would take might not reach level of originality and thus not entail copyright.
Intellectual proterty is political. If you have a patent, I cannotncreate the se thing even if I came up with the idea independantly and neber heard of you.
You are using the power of the state to restrict my freedoms.
Not to be too pedantic but wouldn't that fall under the questioner's definition of fiddling with knobs?
I mean, if you have to point the camera, and set the fstop or set photo vs pano, or even just set the flash or what have you, aren't you kind of key to generating that image? (Even though you generated it with the press of the button.) I mean, if the human didn't point the camera and set the settings, it would be a different picture even if someone did push the button on a camera that was just sitting on a table say.
Or even more generally, does blindfolded painting while drunk, half by accident, for 3 seconds count? How do deliberate and skillful does it have to be?
Speculation: this will hinge on whether or not that input can be considered a copyrighted work by itself. A prompt may well be too short, and if someone else could come up with the identical prompt in a natural setting and generate the exact same output then I would lean towards 'no', if it takes who gives the prompt into account then the answer might be 'yes'.
Just a few questions before I decide what I think of this idea.
What exactly do you mean when you say "AI art"/code/text/... created by a human using an "AI" as a tool to do so.? I guess maybe a better question would be, how would the process you envision for generating AI art differ from the process that the human in this instance used?
It wouldn't, that's the point. The decision is about "if we accept as fact that there was no human element, then this doesn't get copyright". It explicitly does not consider the question if in fact there was a human involved and what that would mean.
I think I'm trying to describe the world we live in, and how the US copyright system works. A lot of creative work is just, 'Throwing shit at the wall and seeing what sticks', and essentially the business of 'curation' in art galleries is just picking what seems good out of a system generating massive amounts of random art.
But was the "human behind the AI" also not involved in priming/training the AI about what to generate? Did the AI really decide what to create (and when to create) of their own volition?
Chances are they were not involved in training the AI, and just provided instructions to a website like DALL-E. If I give you a paragraph of instruction to write a book, barring any other contracts we make, I would be surprised to find anyone arguing I could use that paragraph to claim copyright over your book. The threshold of creative input we require for such a claim is higher than "More than zero"
True, but we don't need to assume that without a prior claim, the standard for awarding copyright bottoms out at nothing. Copyright exists to serve a specific purpose, to encourage people to make substantial investments that they would not make otherwise if they could not profit off them. If the investment is trivial to non-existent, there is no reason for the government to protect it.
The decision is about "if we accept as fact that there was no human element, then this doesn't get copyright". Hence your questions are out of scope of the decision, which doesn't care if "no human element" is true or even possible, and that's why the decision means so little in practice, because once you consider those the situation changes.
The law moves slow but technology is moving fast. AI generated art is a curiousity now but will be the majority of all art produced (by at least some measures) in a few short years.
This makes a precedent that could dramatically shift the balance of power between corporate creators of commercial works and the general public.
Just a few says ago there was a thread where artists recreated Disney art with A.I. I simply cannot imagine a world, where this stance will stand. Sooner or later we will see a massive crackdown by the large copyright holders.
What if a marketing team generates 100 candidate characters with AI for the next feature film, selects the best one and then writes an incredibly detailed brief for a human artist to reverse engineer the character. Could you then copyright the human generated art?
A key litmus test: can an AI produce output that can substitute for a copyrighted work, to the extent that the commercial value of that copyrighted work is diminished? e.g. if a future hypothetical DALL-E-for-video can diffuse a mkv file containing something that plays very much like Star Wars, then surely possessing the model would be treated as possessing a copy of the movie.
The whole idea that the copyright holder needs to opt-out is seriously broken. Copyright is not 'opt-out'. Stable diffusion has set themselves up for trouble, this should have been opt-in.
While AI definitely will be used to help artists, nobody would be happy with somebody being the first to brute force run through tons of prompts on stable diffusion and submitting copyrights on all of those.. imagine if Getty did that..
Indeed it was, I missed it. But given the recent developments in AI driven art generation (and other creatives) I think it is getting to be more and more relevant.
But this does not mean that derivative works can't be copyrighted. So as long as there has been human input and it can be proved then it's copyright able.
Also, the courts still have to decide how to deal with the data that went into the AI. We are a long way from figuring out copyright law for AI's.
54 comments
[ 3.1 ms ] story [ 129 ms ] threadA. I get a copy of ChatGPT's code (via some legitimate means) and train it exclusively on my written work. I argue that all output from the dsr_%_ChatGPT instance is derivative from my work. I assert that when I prompt it, the results are also my work. Further, when someone else prompts it, the results are infringing on my work.
B. Same as A, but I train it 50% on my written work and 50% on jacquesm's written work. I assert that when I prompt it, the results always infringe on jacquesm, and if jacquesm prompts it, the results always infringe on me.
Does anyone disagree with either of the above?
The human behind the AI did insist that they had absolutely no part in creating the image, and the decision is taking that at face value. No human creator - no copyright. Since it's purely based on their input, it also doesn't look at questions like "was the AI created using input material that's copyrighted and might that have relevance"
It has zero applicability on "AI art"/code/text/... created by a human using an "AI" as a tool to do so.
But somehow this gets paraded around every few months as being some important decision...
"Generally, the author and initial copyright owner of a photograph is the person who 'shoots' or 'takes' the photo."
https://www.copyright.gov/engage/photographers/
Let's say a photographer sets up his camera, aims it, composes it, focuses it, etc. But he decides the light isn't quite right yet, so he goes for lunch. While he is away, his assistant clicks the shutter on a whim, capturing more or less the exact image what the photographer intended to capture. The resulting photograph's copyright belongs to the assistant, not the photographer.
This is in theory, of course. In practice, this would be difficult to prove, since the picture is likely to be on a roll of many other pictures for which the photographer does own the copyright. The assistant would need some definitive evidence that they clicked the shutter on that exact frame if they wanted credit for their work.
https://en.wikipedia.org/wiki/Monkey_selfie_copyright_disput...
And that might lead to an argument that if the human doesn't hold the copyright the AI does, but then you are back to square one because you need to be a legal entity to hold copyright.
You are using the power of the state to restrict my freedoms.
Copyright is slightly different, but still
I mean, if you have to point the camera, and set the fstop or set photo vs pano, or even just set the flash or what have you, aren't you kind of key to generating that image? (Even though you generated it with the press of the button.) I mean, if the human didn't point the camera and set the settings, it would be a different picture even if someone did push the button on a camera that was just sitting on a table say.
What exactly do you mean when you say "AI art"/code/text/... created by a human using an "AI" as a tool to do so.? I guess maybe a better question would be, how would the process you envision for generating AI art differ from the process that the human in this instance used?
I would expect if I used such software a number of times, and chose the one I liked the best, I could claim copyright over that image.
I wonder if there were any rulings over images of the mandlebrot set that would clarify it.
"Math class is tough! I love shopping!" - Barbie
The law moves slow but technology is moving fast. AI generated art is a curiousity now but will be the majority of all art produced (by at least some measures) in a few short years.
This makes a precedent that could dramatically shift the balance of power between corporate creators of commercial works and the general public.
[0]https://en.m.wikipedia.org/wiki/Monkey_selfie_copyright_disp...
US Copyright Office Rules A.I. Can't Be Owner of Copyright
This should not be controversial until we have general AIs that are people.
https://news.ycombinator.com/item?id=30702117
(328 points | 9 months ago | 166 comments)
The claimant here explicitly denied being involved in authoring the piece, and the copyright office took that at face value.
Anyone who actually wanted copyright would claim to be the author.
Also, the courts still have to decide how to deal with the data that went into the AI. We are a long way from figuring out copyright law for AI's.
That's the big one. Another commenter mentions that Stable Diffusion is 'opt-out', I think that was a huge mistake on their part.