I think this is a fantastic analogy between AI tools and the art and new area of expertise brought on by cameras. Exploring the latent space with prompts, fine tuning new models, all will be elements of human creativity as the tools enable greater works.
I respectfully disagree. With AI, the quality of output depends mostly on the dataset of original art, not the program. In photography, the subject/landscape/etc. is just that, a moment in time, not the actual artistic output of thousands of other humans that would receive no credit, copyright, or compensation.
There are a LOT of digital artists. I've always assumed that mostly they simply paint with software, but perhaps they also cut and paste within a given canvas and do other things. If the copyright office starts screwing them, I'd expect a backlash.
If I was an AI artist I'd also be taking the position that my creation process was none of their business, and I'd consider it my right to conceal it from them and to sue them.
What are the precedents for included noncopyrightable things, like Campbell's Soup cans (or more detailed prior work directly included) in your multimillion dollar painting?
The problem I have with your approach is that AI is basically cherry-picking material from other artists, and just because we are unable to know exactly who a tiny detail is from, it still infringes on the rights of an existing creator.
It doesn't matter if you're a painter, a photographer, a 3D artist, you're still being shunned by every community if you steal other people's material without attribution or citation.
Currently a lot of AI systems are trained on unlicensed copyrighted works. Whether that's legal is a separate question that's currently being sorted out by the courts. But if the courts decide that training on unlicensed work is infringing, then Ai companies will negotiate licensing deals and create new generative AI tools that are 100 percent based on licensed content. At which point we'll still have to decide whether the resulting works can be copyrighted.
The Dall E 3 announcement looks like it's already headed that way, even though copyright law probably doesn't require it. Their approach makes sense. The AI just needs to understand what images look like. One trained on only (digitizations of) public domain paintings and photos would be very similar to the existing AIs, but without the ability to specify modern image attributes like lense flare and telephoto lens. The hardest thing about creating it would be creating the list of which images to exclude, which the Dall E 3 post suggests they're already starting to do.
But...doesn't something similar happen with text, with words when we write? Authors are creating a new whole out of parts that they may have read elsewhere - turns of phrase, etc. Musicians often do the same thing. But their work is copyrighted. The AI is doing something similar: creating a novel whole from parts often so distant from their original context as to make the original works unrecognizable in the final result.
I'm not sure AI works are so novel, certainly many have been shown to be near replicas of pieces in the dataset. Regardless, how AI programs work vs how humans learn seems entirely different. AI needs millions of images to create something half decent, whereas a human can hear just a song or two, or look at just a piece or two of art and make connections, inferences that can inform their creative process. Also, humans often pay for the music they listen to, or the book they read, or the painting they look at in a museum. AI offers no such compensation.
"AI needs millions of images to create something half decent, whereas a human can hear just a song or two, or look at just a piece or two of art and make connections, inferences that can inform their creative process"
Humans learn over the course of their entire lives, starting from the day they are born into the world. We don't just wake up one morning and create something new and different. What we do is no different than what computers do, other than we need a longer period of time to intergrade experiences.
"Also, humans often pay for the music they listen to, or the book they read, or the painting they look at in a museum"
There is no shortage of free art, music and books. No one HAS to pay for something to be able to integrate it into their personal knowledge base.
"What we do is no different than what computers do" imho is a clearly false and lazy argument. Extrapolation is not interpolation!! Why do we fail to appreciate the incredible nature of human, and generally biological cognition? Folks smarter than I have argued that ML has characteristics that are in many ways the *opposite* of what general intelligence demands: https://medium.com/@petervoss/why-machine-learning-wont-cut-it-f523dd2b20e3
My argument about creative output and compensation is that AI 'inspiration' is restricted, unlike humans. But those data inputs are often stolen copies of somebody's life work. Every creative/holder of copyright imho should be allowed to give consent (opt in, not opt out), receive credit, and compensation for their contribution to these models.
It's crystal clear that you are tunnel focused on the point that you wish to push forward and no amount of other information is going to change your POV.
Again, there is no difference between humans assimilating info from a variety of sources, including books they borrowed from free sources such as the library or their friends or Wikipedia or the Internet Archives and a computer scanning the same material, also without paying anyone for it.
Humans are just another living organism on this planet. We are not special. We were not created by a god.
In 50 years or so, machines and people may well meld together forming the next step in evolution, as a biomachine.
You must know something that leading minds in neuroscience and ML don’t. What is your expertise? What evidence do you have that despite having fundamentally different architectures, operations, compute, and outputs, biological brains and LLMs ‘learn’ the same?
You're making a compelling argument for granting a copyright to stable diffusion or midjourney itself, inasmuch as that makes any sense. If anything though, it's an argument against the prompt engineer receiving copyright, since they themselves are merely making a request of the entity creating the original work. It would be like immediately granting copyright for an image to anyone that makes a request to an artist, instead of the artist themselves. Artists will frequently sell or give away copyright for fan requested material, but it's absolutely not a given
Ah, but your argument is consistent with his. The Midjourney software isn't a legal entity here. Instead, people are legal entities, and they can create entities for shared ownership called corporations, in this case called Midjourney Inc.
Midjourney Inc. has chosen to make clear that when buyers use their tools the buyers own the product of the work they do with the tool. (This is probably natural. I guess artists wouldn't use it otherwise)
Yes, this seems like a much better description. Stochastic gradient descent (which is how AI "learning" works) doesn't cherry pick it's inputs, nor does it even copy them.
That's not really a good way to describe how these things work.
Or, alternatively, it doesn't sufficiently distinguish them from the way humans work.
As someone who has actually worked on engineering these things I've not come across a great way to analogolize their workings that doesn't fail as soon as you start poking at the analogy.
A vast flood of AI generated material would then be under copyright protection. Everything will be owned and it will be almost impossible for a human being to produce anything that cannot be challenged as an infringing work. (See what's happening in music?) I was hoping that AI would bring an end to copyright altogether. Copyright has evolved to inhibit creativity, not promote it as hoped by the framers of the Constitution.
Exactly. The author describes the "creative process" of selecting from 654 (x4?) AI-generated images before reaching the image that's the subject of the copyright application.
But are those hundreds of images also protected by copyright? And if not, by what standard?
Copyright trolls and pranksters can very quickly "create" an almost infinite universe of "creative works" that can be quickly compared against future works. All they need is time and computing power. A year from now, almost every new work would be infringing on *something* that an AI generated and filed away.
Naw, independent creation is a defense to copyright infringement. If you accidentally created an image that looked like someone else's AI-generated art, they wouldn't have grounds for a lawsuit unless they could demonstrate there was deliberately copying.
It would seem hard to argue then that AI outputs are independent creations. Even if AI works don't create an infringement crisis, might they flood the market and starve the already starving artists making original works? Maybe that's where we end up, AI programs that dominate the market and eventually cannibilize themselves, leading to more and more creative stagnation and relegating original creation to hobbyists who can afford the time. Sounds great /s
Is the independent creation defense premised on the difficulty of quality photography? To me, it seems likely. If person A publishes a photo they consider worth selling, this photo is usually of such quality that it would be time-consuming for person B to take a very similar one (knowingly or unknowingly).
Now imagine that person B has a good image-to-image model, plus good luck. Person B is mere seconds away from an image that closely resembles person A's earlier image (created by a text-to-image model), or person A's photo.
With no one (presumably) able to prove what person B did, it seems like they can rely on the independent creation defense. But this is far from what the defense is meant to protect, right? What do you think?
Copyright as such should be banned. It is largely idiotic and detrimental to creativity and detrimental to a much more share based world. Instead of an economy based world.
Hopefully, the advent of easily created computer generate books, music , art, etc. will bring this about because there will be no easy way to determine if an AI participated in creating something.
With this rule, the AI-generated image is in the public domain. But the prompter can modify it, creating a copyrighted derivative work, and never release the AI-generated image. If that's a valid workaround, then this isn't a very good rule.
I think it is highly disingenuous to suggest that AI-works are 'generated by a machine' when the quality of output depends mostly on the dataset of original art, not the program. In photography, the subject/landscape/etc. is just that, a moment in time, not the actual artistic output of thousands, maybe millions. In other words, copyright should consider 'what artist(s) are most responsible for the output?.' It's hard for me to attribute anything more than a fraction of AI output to the prompter, because the AI is so fundamentally a remix of other people's work. If anything, the copyright should be shared.
Tim, several of the comments call for more minimalist approaches to copyright in general. In the spirit of steelmanning, can you see a workable copyright minimal approach? Maybe the copyright office's approach is a decent minimalist approach?
One thought here is that I'm not sure registration is the right stage of the legal process to make judgments like this. If I take a photo of the Eiffel Tower and submit it to the Copyright Office for Registration, the Copyright Office will accept the registration. But my copyright in the photo doesn't allow me to sue other people who take their own picture of the Eiffel Tower, even if the photo looks almost identical due to similar angle and lighting conditions.
So maybe the law should be that certain kinds of AI-generated material is not copyrightable, but rather than forcing every artist to argue with the copyright office about this, we wait until there's an accusation of infringement, at which point the courts could hear evidence on whether the elements that got copy reflected the creative choices of the human artist or were generated autonomously by the computer.
It's possible that there would not actually be that many difficult cases because commercially valuable works with AI material in them tend to be more elaborate and involve more human creativity. This still seems like it would be a mess but I think it's less of a mess than expecting the Copyright Office to render a judgment about each and every piece of digital art that gets registered.
"If artists want their work to stand up in court, they may need to start carefully documenting their creative process so they can prove that their work was made without AI."
Honestly, this does not seem particularly burdensome, given how generous copyright is. Exclusive reproduction rights for a piece of art for life of the artist (which will hopefully continue to extend over the coming century) plus 70 years is a truly expansive property right. Asking digital artists to record their creative process by capturing their screen at the cost of pennies of electricity and drive space seems reasonable, given the expected reward.
Are there circumstances where we wouldn’t want to grant copyright protections? For example, an algorithm that generates endless images and posts them to social media automatically. Should they (each individual piece) get IP protections?
Yes, there does need to be an element of human creativity, so a fully automated system (for example, software that generates millions of images based on randomly-generated prompts) probably wouldn't be eligible for copyright. Of course those images likely wouldn't be worth much.
The problem here, as I see it, is that these AI programs have scraped billions of copyrighted works, which they then chew up and spit out according to the artist’s prompts. No permission was sought or granted for the use of the scraped images.
The logical next argument is that that is exactly what a human does. A human perceives innumerable images during their life, chews them up and then uses that collective knowledge to form their own original material. However, that is an incomplete argument, because humans also must develop refined motor-skills to generate their works, and they must navigate several perceptual layers in order for their work to be created – making the finished work truly unique to them.
As for the argument that landscape photographs are nothing more than pointing a mechanical device at a scene and clicking the shutter, that is the sort of argument made by someone who is unaware of photographic framing and positioning, let alone choices in timing to allow for different natural lighting effects – all aspects unique to the photographer’s skill and body of work.
I agree with your conclusion that this is not a simple problem and that it must be addressed. Using some degree of so called AI during the creation of original works must be allowed for copyright purposes, but where that boundary lies will be very difficult to determine.
If I tell someone they should write a book and suggest 600 ideas that are then incorporated into the text, I don't get to copywriter it because I didn't write it.
I mean ghost writing is a thing. The ghostwriter gets the copyright by default but typically there's a contract where the copyright transfers to the nominal author. Often the author will take the ghostwritten book and make changes to it. I think this is a pretty good analogy for what an AI artist does. He "hires" the AI system to produce some rough drafts, then picks the one he likes best and perhaps asks for more versions. With a human we might say that the two people are co-authors of the work. If one of the "co-authors" is a computer program then I think it's reasonable to say the other "co-author" should get the copyright.
My favorite example like this is The Art of the Deal, Trump's old biography. Trump's name is the only one on the cover, but if I recall it's written entirely by it's ghostwriter and Trump owns the work due to how they contracted it, via work-for-hire or whatever means.
I agree that lack of copyright of AI is inconsistent with how photography is treated. Legally this is indeed a quagmire and would encourage lying. But I wonder what would be the social effects if that holds? Perhaps this ruling is a way to preserve the value of work of human artists who don't use AI? You would still need to pay them I you need your image to be copywritten.
There's a separate question about whether it's copyright infringement to train a model based on unlicensed copyrighted works. The courts will decide on that in the next few years. I think it makes sense to deal with that question directly rather than using copyright eligibility as a backdoor way to get the result someone might want.
Last I checked, this was misinfo, Japan is moving to regulate AI. Copyrighted materials can be used for research and education only. Monetizing, publishing or infringing on copyright holders is still not allowed.
No I completely agree that there is substantial skill involved in a landscape photographer's work. It's just that the skill is in the use of the camera and the selection of images, rather than in directly controlling what's in front of the camera. Which I think makes it similar to what an AI artist does: there's substantial skill in knowing how to adjust a prompt to produce a perfect picture.
In many ways I think that we’re on the same page here, but I will argue the point of framing when it comes to photography, which takes a great deal of skill to do right. And while the counter-argument might be that prompting is like framing in that it arguably takes some skill, the truth is that the prompter can enter the same prompts multiple times and get completely different results with each entry – in other words, the prompts are only vague suggestions, not real creative input.
An interesting angle to look at, though, is the distinction of work-for-hire and who owns a photograph's copyright in that case; the photographer, or the employer, and how that could relate to AI generated images being regarded as created by a virtual ‘employee’.
You can configure generative AI systems to use a fixed seed so that a given prompt always produces the same image. I don't know see why that should make a difference in whether or not we view prompting as a creative activity.
Interesting - but how many prompters would actually configure the system to use a fixed seed, when not doing so would produce a wider and more interesting range of results?
I think this is a fantastic analogy between AI tools and the art and new area of expertise brought on by cameras. Exploring the latent space with prompts, fine tuning new models, all will be elements of human creativity as the tools enable greater works.
I respectfully disagree. With AI, the quality of output depends mostly on the dataset of original art, not the program. In photography, the subject/landscape/etc. is just that, a moment in time, not the actual artistic output of thousands of other humans that would receive no credit, copyright, or compensation.
Great points.
There are a LOT of digital artists. I've always assumed that mostly they simply paint with software, but perhaps they also cut and paste within a given canvas and do other things. If the copyright office starts screwing them, I'd expect a backlash.
If I was an AI artist I'd also be taking the position that my creation process was none of their business, and I'd consider it my right to conceal it from them and to sue them.
What are the precedents for included noncopyrightable things, like Campbell's Soup cans (or more detailed prior work directly included) in your multimillion dollar painting?
The problem I have with your approach is that AI is basically cherry-picking material from other artists, and just because we are unable to know exactly who a tiny detail is from, it still infringes on the rights of an existing creator.
It doesn't matter if you're a painter, a photographer, a 3D artist, you're still being shunned by every community if you steal other people's material without attribution or citation.
Currently a lot of AI systems are trained on unlicensed copyrighted works. Whether that's legal is a separate question that's currently being sorted out by the courts. But if the courts decide that training on unlicensed work is infringing, then Ai companies will negotiate licensing deals and create new generative AI tools that are 100 percent based on licensed content. At which point we'll still have to decide whether the resulting works can be copyrighted.
The Dall E 3 announcement looks like it's already headed that way, even though copyright law probably doesn't require it. Their approach makes sense. The AI just needs to understand what images look like. One trained on only (digitizations of) public domain paintings and photos would be very similar to the existing AIs, but without the ability to specify modern image attributes like lense flare and telephoto lens. The hardest thing about creating it would be creating the list of which images to exclude, which the Dall E 3 post suggests they're already starting to do.
But...doesn't something similar happen with text, with words when we write? Authors are creating a new whole out of parts that they may have read elsewhere - turns of phrase, etc. Musicians often do the same thing. But their work is copyrighted. The AI is doing something similar: creating a novel whole from parts often so distant from their original context as to make the original works unrecognizable in the final result.
I'm not sure AI works are so novel, certainly many have been shown to be near replicas of pieces in the dataset. Regardless, how AI programs work vs how humans learn seems entirely different. AI needs millions of images to create something half decent, whereas a human can hear just a song or two, or look at just a piece or two of art and make connections, inferences that can inform their creative process. Also, humans often pay for the music they listen to, or the book they read, or the painting they look at in a museum. AI offers no such compensation.
"AI needs millions of images to create something half decent, whereas a human can hear just a song or two, or look at just a piece or two of art and make connections, inferences that can inform their creative process"
Humans learn over the course of their entire lives, starting from the day they are born into the world. We don't just wake up one morning and create something new and different. What we do is no different than what computers do, other than we need a longer period of time to intergrade experiences.
"Also, humans often pay for the music they listen to, or the book they read, or the painting they look at in a museum"
There is no shortage of free art, music and books. No one HAS to pay for something to be able to integrate it into their personal knowledge base.
"What we do is no different than what computers do" imho is a clearly false and lazy argument. Extrapolation is not interpolation!! Why do we fail to appreciate the incredible nature of human, and generally biological cognition? Folks smarter than I have argued that ML has characteristics that are in many ways the *opposite* of what general intelligence demands: https://medium.com/@petervoss/why-machine-learning-wont-cut-it-f523dd2b20e3
My argument about creative output and compensation is that AI 'inspiration' is restricted, unlike humans. But those data inputs are often stolen copies of somebody's life work. Every creative/holder of copyright imho should be allowed to give consent (opt in, not opt out), receive credit, and compensation for their contribution to these models.
It's crystal clear that you are tunnel focused on the point that you wish to push forward and no amount of other information is going to change your POV.
Again, there is no difference between humans assimilating info from a variety of sources, including books they borrowed from free sources such as the library or their friends or Wikipedia or the Internet Archives and a computer scanning the same material, also without paying anyone for it.
Humans are just another living organism on this planet. We are not special. We were not created by a god.
In 50 years or so, machines and people may well meld together forming the next step in evolution, as a biomachine.
You must know something that leading minds in neuroscience and ML don’t. What is your expertise? What evidence do you have that despite having fundamentally different architectures, operations, compute, and outputs, biological brains and LLMs ‘learn’ the same?
You're making a compelling argument for granting a copyright to stable diffusion or midjourney itself, inasmuch as that makes any sense. If anything though, it's an argument against the prompt engineer receiving copyright, since they themselves are merely making a request of the entity creating the original work. It would be like immediately granting copyright for an image to anyone that makes a request to an artist, instead of the artist themselves. Artists will frequently sell or give away copyright for fan requested material, but it's absolutely not a given
Ah, but your argument is consistent with his. The Midjourney software isn't a legal entity here. Instead, people are legal entities, and they can create entities for shared ownership called corporations, in this case called Midjourney Inc.
Midjourney Inc. has chosen to make clear that when buyers use their tools the buyers own the product of the work they do with the tool. (This is probably natural. I guess artists wouldn't use it otherwise)
A director curates performances from actors, and is given credit for their curation.
Couldn't it be an argument to grant copyright to the dataset contributors, since the programs can do literally nothing without them?
Yes, this seems like a much better description. Stochastic gradient descent (which is how AI "learning" works) doesn't cherry pick it's inputs, nor does it even copy them.
That's not really a good way to describe how these things work.
Or, alternatively, it doesn't sufficiently distinguish them from the way humans work.
As someone who has actually worked on engineering these things I've not come across a great way to analogolize their workings that doesn't fail as soon as you start poking at the analogy.
A vast flood of AI generated material would then be under copyright protection. Everything will be owned and it will be almost impossible for a human being to produce anything that cannot be challenged as an infringing work. (See what's happening in music?) I was hoping that AI would bring an end to copyright altogether. Copyright has evolved to inhibit creativity, not promote it as hoped by the framers of the Constitution.
Exactly. The author describes the "creative process" of selecting from 654 (x4?) AI-generated images before reaching the image that's the subject of the copyright application.
But are those hundreds of images also protected by copyright? And if not, by what standard?
Copyright trolls and pranksters can very quickly "create" an almost infinite universe of "creative works" that can be quickly compared against future works. All they need is time and computing power. A year from now, almost every new work would be infringing on *something* that an AI generated and filed away.
Naw, independent creation is a defense to copyright infringement. If you accidentally created an image that looked like someone else's AI-generated art, they wouldn't have grounds for a lawsuit unless they could demonstrate there was deliberately copying.
It would seem hard to argue then that AI outputs are independent creations. Even if AI works don't create an infringement crisis, might they flood the market and starve the already starving artists making original works? Maybe that's where we end up, AI programs that dominate the market and eventually cannibilize themselves, leading to more and more creative stagnation and relegating original creation to hobbyists who can afford the time. Sounds great /s
I think you may be underestimating the impact of even frivolous claims and suits.
Is the independent creation defense premised on the difficulty of quality photography? To me, it seems likely. If person A publishes a photo they consider worth selling, this photo is usually of such quality that it would be time-consuming for person B to take a very similar one (knowingly or unknowingly).
Now imagine that person B has a good image-to-image model, plus good luck. Person B is mere seconds away from an image that closely resembles person A's earlier image (created by a text-to-image model), or person A's photo.
With no one (presumably) able to prove what person B did, it seems like they can rely on the independent creation defense. But this is far from what the defense is meant to protect, right? What do you think?
Erm, and i don't want to mention the NFT artwork flood, buuuuuuut... :)
Copyright as such should be banned. It is largely idiotic and detrimental to creativity and detrimental to a much more share based world. Instead of an economy based world.
Hopefully, the advent of easily created computer generate books, music , art, etc. will bring this about because there will be no easy way to determine if an AI participated in creating something.
With this rule, the AI-generated image is in the public domain. But the prompter can modify it, creating a copyrighted derivative work, and never release the AI-generated image. If that's a valid workaround, then this isn't a very good rule.
I think it is highly disingenuous to suggest that AI-works are 'generated by a machine' when the quality of output depends mostly on the dataset of original art, not the program. In photography, the subject/landscape/etc. is just that, a moment in time, not the actual artistic output of thousands, maybe millions. In other words, copyright should consider 'what artist(s) are most responsible for the output?.' It's hard for me to attribute anything more than a fraction of AI output to the prompter, because the AI is so fundamentally a remix of other people's work. If anything, the copyright should be shared.
Tim, several of the comments call for more minimalist approaches to copyright in general. In the spirit of steelmanning, can you see a workable copyright minimal approach? Maybe the copyright office's approach is a decent minimalist approach?
One thought here is that I'm not sure registration is the right stage of the legal process to make judgments like this. If I take a photo of the Eiffel Tower and submit it to the Copyright Office for Registration, the Copyright Office will accept the registration. But my copyright in the photo doesn't allow me to sue other people who take their own picture of the Eiffel Tower, even if the photo looks almost identical due to similar angle and lighting conditions.
So maybe the law should be that certain kinds of AI-generated material is not copyrightable, but rather than forcing every artist to argue with the copyright office about this, we wait until there's an accusation of infringement, at which point the courts could hear evidence on whether the elements that got copy reflected the creative choices of the human artist or were generated autonomously by the computer.
It's possible that there would not actually be that many difficult cases because commercially valuable works with AI material in them tend to be more elaborate and involve more human creativity. This still seems like it would be a mess but I think it's less of a mess than expecting the Copyright Office to render a judgment about each and every piece of digital art that gets registered.
"If artists want their work to stand up in court, they may need to start carefully documenting their creative process so they can prove that their work was made without AI."
Honestly, this does not seem particularly burdensome, given how generous copyright is. Exclusive reproduction rights for a piece of art for life of the artist (which will hopefully continue to extend over the coming century) plus 70 years is a truly expansive property right. Asking digital artists to record their creative process by capturing their screen at the cost of pennies of electricity and drive space seems reasonable, given the expected reward.
Are there circumstances where we wouldn’t want to grant copyright protections? For example, an algorithm that generates endless images and posts them to social media automatically. Should they (each individual piece) get IP protections?
Yes, there does need to be an element of human creativity, so a fully automated system (for example, software that generates millions of images based on randomly-generated prompts) probably wouldn't be eligible for copyright. Of course those images likely wouldn't be worth much.
This just shows that the concept of copyright is a kludge at best.
I've got some bad news for you about the law in general. It's kludges all the way down
True! However, I'd argue that the amount of kludge varies.
The problem here, as I see it, is that these AI programs have scraped billions of copyrighted works, which they then chew up and spit out according to the artist’s prompts. No permission was sought or granted for the use of the scraped images.
The logical next argument is that that is exactly what a human does. A human perceives innumerable images during their life, chews them up and then uses that collective knowledge to form their own original material. However, that is an incomplete argument, because humans also must develop refined motor-skills to generate their works, and they must navigate several perceptual layers in order for their work to be created – making the finished work truly unique to them.
As for the argument that landscape photographs are nothing more than pointing a mechanical device at a scene and clicking the shutter, that is the sort of argument made by someone who is unaware of photographic framing and positioning, let alone choices in timing to allow for different natural lighting effects – all aspects unique to the photographer’s skill and body of work.
I agree with your conclusion that this is not a simple problem and that it must be addressed. Using some degree of so called AI during the creation of original works must be allowed for copyright purposes, but where that boundary lies will be very difficult to determine.
If I tell someone they should write a book and suggest 600 ideas that are then incorporated into the text, I don't get to copywriter it because I didn't write it.
I mean ghost writing is a thing. The ghostwriter gets the copyright by default but typically there's a contract where the copyright transfers to the nominal author. Often the author will take the ghostwritten book and make changes to it. I think this is a pretty good analogy for what an AI artist does. He "hires" the AI system to produce some rough drafts, then picks the one he likes best and perhaps asks for more versions. With a human we might say that the two people are co-authors of the work. If one of the "co-authors" is a computer program then I think it's reasonable to say the other "co-author" should get the copyright.
My favorite example like this is The Art of the Deal, Trump's old biography. Trump's name is the only one on the cover, but if I recall it's written entirely by it's ghostwriter and Trump owns the work due to how they contracted it, via work-for-hire or whatever means.
I agree that lack of copyright of AI is inconsistent with how photography is treated. Legally this is indeed a quagmire and would encourage lying. But I wonder what would be the social effects if that holds? Perhaps this ruling is a way to preserve the value of work of human artists who don't use AI? You would still need to pay them I you need your image to be copywritten.
There's a separate question about whether it's copyright infringement to train a model based on unlicensed copyrighted works. The courts will decide on that in the next few years. I think it makes sense to deal with that question directly rather than using copyright eligibility as a backdoor way to get the result someone might want.
Note: Japan has already declared that using datasets for training AI models doesn't violate copyright law
Last I checked, this was misinfo, Japan is moving to regulate AI. Copyrighted materials can be used for research and education only. Monetizing, publishing or infringing on copyright holders is still not allowed.
No I completely agree that there is substantial skill involved in a landscape photographer's work. It's just that the skill is in the use of the camera and the selection of images, rather than in directly controlling what's in front of the camera. Which I think makes it similar to what an AI artist does: there's substantial skill in knowing how to adjust a prompt to produce a perfect picture.
In many ways I think that we’re on the same page here, but I will argue the point of framing when it comes to photography, which takes a great deal of skill to do right. And while the counter-argument might be that prompting is like framing in that it arguably takes some skill, the truth is that the prompter can enter the same prompts multiple times and get completely different results with each entry – in other words, the prompts are only vague suggestions, not real creative input.
An interesting angle to look at, though, is the distinction of work-for-hire and who owns a photograph's copyright in that case; the photographer, or the employer, and how that could relate to AI generated images being regarded as created by a virtual ‘employee’.
You can configure generative AI systems to use a fixed seed so that a given prompt always produces the same image. I don't know see why that should make a difference in whether or not we view prompting as a creative activity.
Fixed prompt or not, I think it highlights that other people's work are mostly responsible for the output, not the prompter.
Interesting - but how many prompters would actually configure the system to use a fixed seed, when not doing so would produce a wider and more interesting range of results?
The photo might be copyrighted but someone else could make their own copy of the AI art without infringing your copyright.
Only if you release the AI art or the prompt! Otherwise they'd have to independently discover it.
I'm sure someone could program a computer to input a piece of "art" and reverse engineer the prompt that created it.