Jump to content

Should AI-generated images be copyrightable?


Horn of Gabriel

Recommended Posts

Really interesting technology at play here, with tools like DALL-E and Midjourney.  Some of the images they spit out are incredible, and some are trash - much like human artists.  But aside from the interesting tech and the threat (?) of AI-artistry to human-artists, the article raises another big implication: are AI-generated images copyrightable?  Currently it looks like the US Copyright Office says no:

Quote

A spokesperson for the US Copyright Office told Insider that works generated only by artificial intelligence lacked the human authorship necessary to support a copyright claim.

But what about the human input to the AI?  These tools could be considered like a paintbrush - they don't do anything on their own, it takes a human hand to give the right prompt to create the image.  And while the prompt could be really basic like: "Elon Musk holding a Twitter bird," some are also very specific and detailed, like: "Close-up of [pick your favorite person] in Hellblade: Senua's Sacrifice, emerging from black mud, long hair with dreads, war blue paint, paint fading, angry expression, dirty face, finely detailed eyes, moody, viking clothes, epic scene, epic composition, Photography, Cinematic Lighting, Volumetric Lighting, ethereal light, intricate details, extremely detailed volumetric rays"

spacer.png

Some other examples in a variety of styles:

spacer.png

spacer.png

spacer.png

spacer.png

spacer.png

spacer.png

If you want to go down the rabbit hole: https://www.reddit.com/r/midjourney/

Edited by Horn of Gabriel
Link to comment
Share on other sites

  • Horn of Gabriel changed the title to Should AI-generated images be copyrightable?

This is incredible technology, but Dall-E 2 and neural nets and all of that are themselves the copyrightable artifact (read: THE MODEL!) - those weights, biases, neural layers, etc etc are what takes human insight and creation. The art is simply the output from using a tool that someone (or many other people) made - could you copyright pressing "go" on a tormach CNC?

The results are incredible, but are not the works of a person or entity that can otherwise originate ownership. 

  • Hook 'Em 1
  • Like 1
Link to comment
Share on other sites

The key provision here is Section 102 of the Copyright Act:

(a) Copyright protection subsists, in accordance with this title, in original works of authorship fixed in any tangible medium of expression, now known or later developed, from which they can be perceived, reproduced, or otherwise communicated, either directly or with the aid of a machine or device. Works of authorship include the following categories:

(1) literary works;

(2) musical works, including any accompanying words;

(3) dramatic works, including any accompanying music;

(4) pantomimes and choreographic works;

(5) pictorial, graphic, and sculptural works;

(6) motion pictures and other audiovisual works;

(7) sound recordings; and

(8) architectural works.

(b) In no case does copyright protection for an original work of authorship extend to any idea, procedure, process, system, method of operation, concept, principle, or discovery, regardless of the form in which it is described, explained, illustrated, or embodied in such work.

Authorship rather heavily implies and has been consistently interpreted in other contexts to refer to natural persons and their works or output.  It was recently held that a selfie taken by an orangutan is not a work of authorship because it lacks human authorship, despite meeting criteria of originality and expression.  Note also the use of "machine or device" in the statute.

AI "engines," of whatever type, are themselves not expressions, either, but rather the "idea" type of thing excluded from copyright by subsecion (b).  Sub (b) is recognized as both a delineator between what is eligible for copyright and what is eligible for patent (which has more rigorous requirements for protection and grants a much shorter monopoly) and a "comment" on what is not expression, the heart of copyright.

Also, it's never been expressly articulated in the case law, but there is a definite tendency in copyright to extend greater protection to "art for art's sake" rather than more "commercial" art.  Part of that is that commercial things often are dictated by something other than the artist's choices.  So, even if a non-person were properly considered an author, I think the nature of AI-generated works might entitle them to slim or no protection, because such works are dictated by the "strictures" of the AI engine more than any "natural" expression.

Edited by TwiceHorn
  • Hook 'Em 1
Link to comment
Share on other sites

I've messed around with Midjourney and some of the photos it pushes out are incredible. I see it as being a huge killer to the stock photo industry (believe it or not, it exists). I won't have to pay out licensing fees for marketing photos when I can get them to do it for free and no one can stop me.

Link to comment
Share on other sites

2 minutes ago, TwiceHorn said:

None of them will do Trump smearing shit in his hair.

I am dissapoint.

They won't do any specific public person, what for fear of deepfakery. The folks who built this dataset and model have also published reams on the measures they've taken to protect against abuse and creation of disinformation 

  • Hook 'Em 1
Link to comment
Share on other sites

On 10/24/2022 at 12:25 PM, HRSchenker said:

I've messed around with Midjourney and some of the photos it pushes out are incredible. I see it as being a huge killer to the stock photo industry (believe it or not, it exists). I won't have to pay out licensing fees for marketing photos when I can get them to do it for free and no one can stop me.

But then you'll miss the drama like when (I think) the Austin Chronicle in the 90s had some ad pushing some expensive widget, along with the face of a young woman who was smugly beyond everyday concerns.

The ad started a mini shit-storm in the letters to the editor-- the woman embodied everything wrong with what was happening around town, and she should just go to some other place if she wanted to be so hoity-toity.

She sort of did. Turns out the woman was a model for a photographer in Paris who had probably been thinking about lunch, but who managed to enrage a few Chronicle readers.

Link to comment
Share on other sites

Does the AI produce the same image for the same prompt? Does it do different things based on whether the input is typed or spoken? If the human can manipulate the AI's performance in ways that others found hard to replicate, that'd be interesting but not meaningful-- people copy other artists all the time.

I'm not sure it even matters. Take a hypothetical-- suppose I'm funding my baller lifestyle by throwing paint into a fan, which splatters it all over an 11 x 14" canvas (I chose a small canvas because it's easy to ship, and I'm big in Estonia.)

What's more, I do it blindfolded. Just set up the canvas, aim at the fan, splat.

Pretty sure I possess copyright for the results.

So why wouldn't I have copyright if I went through AI to simulate fan-splattered art? All I'd have control over would be rights to reproduce the images produced by those techniques. Would the AI creators be my creative partners?

  • Hook 'Em 3
Link to comment
Share on other sites

26 minutes ago, RDCanecutter said:

Does the AI produce the same image for the same prompt? Does it do different things based on whether the input is typed or spoken? If the human can manipulate the AI's performance in ways that others found hard to replicate, that'd be interesting but not meaningful-- people copy other artists all the time.

I'm not sure it even matters. Take a hypothetical-- suppose I'm funding my baller lifestyle by throwing paint into a fan, which splatters it all over an 11 x 14" canvas (I chose a small canvas because it's easy to ship, and I'm big in Estonia.)

What's more, I do it blindfolded. Just set up the canvas, aim at the fan, splat.

Pretty sure I possess copyright for the results.

So why wouldn't I have copyright if I went through AI to simulate fan-splattered art? All I'd have control over would be rights to reproduce the images produced by those techniques. Would the AI creators be my creative partners?

Good questions, especially the first.

Currently, there are "test cases" out there where they don't bother to name a human author on a copyright application, or a human inventor on a patent application.  They're just naming AI (actually, I don't think this has been tried at all with a copyright application, the Copyright Office is taking statutory cues from the Patent Office, where it has been attempted and failed) https://www.science.org/content/blog-post/ai-inventors-not-so-fast

A question that I don't believe has been answered is, what if a human applied for a copyright or patent for an work/invention made with AI assistance (or almost entirely by AI save the input), naming joint authors/inventors with at least one being human.

And, you're dead right about "machine aided" art:  there would be almost no infringement by "copying," because the similarity would be the result of chance, not the human input or action (in your fan-paint hypothetical, and if AI doesn't produce the same result from the same input).  Only the reproduction right (literal copies) and derivative works (modifications of the original work if not literal copies) would be implicated.  And, while I think you are right that a human applying for copyright on "machine aided" art would receive a registration and at least some protection, that protection would be narrow, for the reason above (limited human creativity at work).

Also, for right now, the question I think is purely statutory:  did Congress intend to permit non-human inventors and authors to secure copyright and patents?  Congress could change that tomorrow.  There are different policy arguments in favor of either, and the Constitution might come into play:  "[Congress shall have the power] To promote the Progress of Science and useful Arts, by securing for limited Times to Authors and Inventors the exclusive Right to their respective Writings and Discoveries;"

Edited by TwiceHorn
  • Drool 1
Link to comment
Share on other sites

1 hour ago, RDCanecutter said:

Does the AI produce the same image for the same prompt? Does it do different things based on whether the input is typed or spoken? If the human can manipulate the AI's performance in ways that others found hard to replicate, that'd be interesting but not meaningful-- people copy other artists all the time.

Dall-E is a generative neural net that takes a natural language prompt (spoken or written doesn't matter) and parses it with natural language processing (NLP) to understand the subject and style it is meant to produce. From there, it goes through the black box of layers of neurons and weights and biases (read: the "AI") to produce an image of the subject along with any other modifiers.

Generative NN's like Dall-E usually don't do the same thing twice for the same reason that you usually can't do the same thing exactly the same twice - the input conditions are never exactly the same. But it's going to be pretty dang similar every time. Either way, the output isn't really the thing of value here that you would want to keep protected - it's the model itself and it's bespoke neural net architecture, calibration of it's neurons and their interconnections, and the controlling logic to orchestrate turning language into imagery. 

 

Although much of the copyright complaints I've heard is in relation to Dall-E looking at other works while training it's NN, and copy holders arguing that Dall-E is simply copying the work, which is also inaccurate. The NN has nothing in it that resembles the original image, just the concept of what the image is. 

It's a little bit heady, but there's an emerging field of neural graphics in which rather than rendering pixels into a scene based on static reference data in a file, you train a neural net to "remember" it for you. It doesn't really fit to the classical copyright definition or model. Check out my thread in nerdz if you would like to know more

Interestingly, this conversation is going around the AI/ML community at my work from an academic and technical perspective, and nobody there really has any emerging consensus aside from the lawyers will probably get it wrong 

Edited by Captainant
  • Drool 1
Link to comment
Share on other sites

3 minutes ago, Captainant said:

Generative NN's usually don't do the same thing twice for the same reason that you usually can't do the same thing exactly the same twice - the input conditions are never exactly the same. But it's going to be pretty dang similar every time. Either way, the output isn't really the thing of value here that you would want to keep protected - it's the model itself and it's bespoke neural net architecture, calibration of it's neurons and their interconnections, and the controlling logic to orchestrate turning language into imagery. 

If it is true that the output is not the concern, but the architecture of the NN, the latter is not the subject of copyright, anyway.  Perhaps the literal code, but nothing else.

Link to comment
Share on other sites

17 minutes ago, TwiceHorn said:

If it is true that the output is not the concern, but the architecture of the NN, the latter is not the subject of copyright, anyway.  Perhaps the literal code, but nothing else.

Bingo. Even someone knowing your NN architecture, all the layers and interconnects, doesn't really get them anything. They also need weights (how important a neuron is in the next layer), and biases (how much a neuron reacts to it's connections) to actually define the behavior of the NN. Finding the right weights and biases is what model training is, and where you'll spend the big bucks in your data science and ML practice

Edited by Captainant
Link to comment
Share on other sites

  • 8 months later...

So, the first copyright suits against AI proprietors are hitting.

https://llmlitigation.com/

Recall that the Copyright Act protects against "copying" that takes the form of reproducing (verbatim) portions of copyrighted works and preparing derivative works (non verbatim reproductions).

So, the first objectionable act (of infringement) is training the AI engines on copyrighted works.  That seemingly inevitably produces a copy or reproduction of the work in the process of "feeding it" to the machine.  The second would be anytime the engine spits out reproductions of copyright works.  The latter seemingly occurs without human intervention, except that of the prompter and that of the trainer.  Both or either could be liable, although the trainer is probably more liable as it is volitionally copying the work, while the prompter may not intend anything.

As you see from the link above, the targets are coding/programming, images, and text/books.

Link to comment
Share on other sites

2 minutes ago, TwiceHorn said:

So, the first objectionable act (of infringement) is training the AI engines on copyrighted works.  That seemingly inevitably produces a copy or reproduction of the work in the process of "feeding it" to the machine.  The second would be anytime the engine spits out reproductions of copyright works.  The latter seemingly occurs without human intervention, except that of the prompter and that of the trainer.  Both or either could be liable, although the trainer is probably more liable as it is volitionally copying the work, while the prompter may not intend anything.

Is my phone loading a song into local memory while streaming it a violation of copyright? I would argue that at a mechanical and technical level, there is no copying happing of those works. That's simply now how extracting embeddings from a document or image works. What you call "feeding it" to the machine is handwaving away quite a significant amount of mathematically provable data science. 

Link to comment
Share on other sites

22 minutes ago, Captainant said:

Is my phone loading a song into local memory while streaming it a violation of copyright? I would argue that at a mechanical and technical level, there is no copying happing of those works. That's simply now how extracting embeddings from a document or image works. What you call "feeding it" to the machine is handwaving away quite a significant amount of mathematically provable data science. 

Legally, loading code into RAM is making a copy.  Whether it is an unlawful copy depends on whether the person loading it into RAM has license to do so.  

So, if you have a license to load a song into local memory, as by legitimate streaming service, it's not unlawful.  If it's a bootleg mp3, then it is.

Do you think using a photocopier to reproduce images isn't copyright infringement because the tech is cool and has nothing to do with painting or photography?  Or that digital recordings of music aren't infringements because it's not how the music was made in the first place (in some instances).

The technology is utterly irrelevant so long as the intention and result is to make a human-readable reproduction of a copyright work.

Is the purpose of training the AI engine so that the AI engine can regurgitate identifiable pieces of the copyrighted work at some point?  Does the AI engine accomplish that?  Did the trainer of the AI engine have the right to use the source material for that purpose?

Edited by TwiceHorn
Link to comment
Share on other sites

22 minutes ago, TwiceHorn said:

Is the purpose of training the AI engine so that the AI engine can regurgitate identifiable pieces of the copyrighted work at some point? Does the AI engine accomplish that? 

Specifically, no. The entire point of training an ML model, in this case a supervised model with labeled data, is to be able to capture the essence of meaning of a dataset or work to make predictions or variations on that theme.  Depending on the type of work and the possibility space of it's artifacts, it may be either impossibly remote of a chance to replicate an initial work -  or just a matter of time due to number of possible permutations (and perhaps an overly permissive original copyright). But it's virtually impossible to train an ML model to be a replica

27 minutes ago, TwiceHorn said:

Did the trainer of the AI engine have the right to use the source material for that purpose?

I do think there's some "there" there on data rights for training the model, but again depending on the nature of the work, is not simply viewing and internalizing a work an example of fair use? DJs are allowed to sample other people's works as fair use, and that's broadly analogous to what generative ML models are mechanically actually doing 

  • Hook 'Em 1
Link to comment
Share on other sites

3 minutes ago, Captainant said:

Specifically, no. The entire point of training an ML model, in this case a supervised model with labeled data, is to be able to capture the essence of meaning of a dataset or work to make predictions or variations on that theme.  Depending on the type of work and the possibility space of it's artifacts, it may be either impossibly remote of a chance to replicate an initial work -  or just a matter of time due to number of possible permutations (and perhaps an overly permissive original copyright). But it's virtually impossible to train an ML model to be a replica

I do think there's some "there" there on data rights for training the model, but again depending on the nature of the work, is not simply viewing and internalizing a work an example of fair use? DJs are allowed to sample other people's works as fair use, and that's broadly analogous to what generative ML models are mechanically actually doing 

Well, bear in mind that mere phrases or sentence (fragments) of, say, a literary work, if copied, can constitute infringement.  Same with song lyrics, musical notes, pieces of images. So, even though the purpose of the AI engine is not to strictly duplicate the original work, barfing out identifiable chunks of it may constitute infringement.

It could potentially be excused fair use, as well, but that's going to depend on the output of the AI and the nature of the use of the copyright work.  I don't think you can predictively call it fair use unless the courts want to go inventing another category of fair use.

What will likely happen is that a new "type" of license develops for use of copyright material to train AI engines.  It would probably be cheaper than a straight-up duplication license because only fragments of the works would be used and in a sort of random and unpredictable fashion.  But the existence of the license will still be premised on the notion that infringement occurs in the process of training and operating AI engines.  The trick is going to be that there is so much material out there with very diverse ownership, so it would seem that an "artists rights" organization like ASCAP/BMI is going to be necessary.

  • Hook 'Em 2
Link to comment
Share on other sites

Would it be analogous to a musician B being influenced by musician A and producing similar sounding music albeit slightly different?

I realize one could argue that musician B had to purchase either a ticket to see musician A, buy their CD, download their song, etc, therefore any influence obtained was obtained legally and musician A was rightfully compensated. However, say musician C bought the CD of musician A and let musician B borrow it? Musician B didn’t compensate musician A for the influence, but yet produced similar sounding music

What if someone had thousands of digital images of an artist. They were friends with an AI developer. The developer used the AI to influence it, and the AI produced similar artwork 

 

Link to comment
Share on other sites

2 hours ago, Neonmoon said:

Would it be analogous to a musician B being influenced by musician A and producing similar sounding music albeit slightly different?

That can still be copyright infringement, that is, copying can be done "from memory." 

The unlawful act in copyright infringement is "copying."  Because we don't always have proof of literal copying (but sometimes you do), copying is proven by inference:  did the infringer have "access" to the copyright work (exposure, knowledge, familiarity) and is the alleged copy "substantially similar," which means similar enough given the circumstances to infer that it was copied.

As a factual matter, copying "from memory" (as opposed to rote or "mechanical" duplication) tends to be less literal and may not be found by a judge or jury to be "substantially similar."  In these music copyright trials, that's usually the issue and the musicologists come on and explain how the copied "jingle" isn't actually that unique and, although the alleged copier was familiar with the copyright work, the similarity isn't such that you can infer copying from it.  And sometimes they find copying anyway.

2 hours ago, Neonmoon said:

I realize one could argue that musician B had to purchase either a ticket to see musician A, buy their CD, download their song, etc, therefore any influence obtained was obtained legally and musician A was rightfully compensated. However, say musician C bought the CD of musician A and let musician B borrow it? Musician B didn’t compensate musician A for the influence, but yet produced similar sounding music

Whether the copier had lawful possession of a copyright work isn't dispositive of the infringement issue.  For example, you can't play a CD in public (or even a radio station), especially in a for-profit venue, because the license you have lets you listen for personal private use, but not for public peformance.  Nor can you lawfully make more than one "archival" copy of your CD (or other digital formats).

Similarly, in your hypothetical above, even if you own the CD, you don't have the right to sample it (reproduce) and resell and/or publicly perform it. Which is why we get the sampling cases.

So, unless the AI trainer has a license to duplicate the work or otherwise use it to input to an AI engine, that would be infringement.

2 hours ago, Neonmoon said:

What if someone had thousands of digital images of an artist. They were friends with an AI developer. The developer used the AI to influence it, and the AI produced similar artwork 

 

Well, as explained (sort of) above, I think inputting the images means duplicating and putting all of the data that represents the image into the engine.  That's an act of copyright infringement in and of itself, even if the engine is "abstracting" that data and not storing it directly.  Unless you own a license to do it, of course.

Then, at some later date, the AI engine spits out a similar image.  Then you have the factual questions, is it substantially similar and is it fair use that determine infringement liability. Input on the left, outputs on the right.

warhol-4_custom-56581aa1ad743e3c83ecf7ad3e91734c33d436a3-s1100-c50.jpg

 

Edited by TwiceHorn
  • Hook 'Em 1
Link to comment
Share on other sites

Also, as hinted at above, copyright protects an expression of an idea, but cannot protect the idea itself, which is closer to the domain of patents.  That means broad plot ideas are not copyrightable, while particular plays, books, or stories are. A portrait of Prince is an idea.  The particular portrait photo by Lynn Goldsmith is an expression of that idea. The distinction between idea and expression can get hairy though, especially in works of visual art.

I think what Captainant was driving at is that AI engines strive to abstract ideas from the material that they are trained with.  That is, after all, the essence of intelligence:  to deal with information on the abstract, idea level, independent of expression, while computers heretofore were extremely literal and dependent on very exact expression for input, output, and intermediate processing.  If an AI engine is fully successful at abstracting ideas from input expressions, then nothing going on "inside" the AI engine should constitute infringement and seemingly neither should its output.  Whether AI engines really operate on that level is probably unknown, maybe unknowable.

Whether the engine avoids output of literal chunks of input altogether has a lot to do with the sophistication of the engine.  I know in textual things they use quotations, so the literal is in there somewhere and liable to come spilling out.  And it's a reasonable bet that quotation is fair use.  But I don't think you can extend that to any output of literally similar or identical material.

Edited by TwiceHorn
Link to comment
Share on other sites

11 hours ago, TwiceHorn said:

I think what Captainant was driving at is that AI engines strive to abstract ideas from the material that they are trained with.  That is, after all, the essence of intelligence:  to deal with information on the abstract, idea level, independent of expression, while computers heretofore were extremely literal and dependent on very exact expression for input, output, and intermediate processing.  If an AI engine is fully successful at abstracting ideas from input expressions, then nothing going on "inside" the AI engine should constitute infringement and seemingly neither should its output.  Whether AI engines really operate on that level is probably unknown, maybe unknowable.

It's extremely knowable, and aspects like feature importance and provability is a major portion of dataset and model analysis. When you take a large set of training data and train N number of models against it, you're going to have N number of unique models - even if you used the exact same dataset, hyperparameters, training epochs, etc etc etc. It's a stochastic process by design, it's the best mechanism we have to create a diverse set of neural nets that give us the best chance to find the best solution, not just one of several good solutions.

11 hours ago, TwiceHorn said:

Whether the engine avoids output of literal chunks of input altogether has a lot to do with the sophistication of the engine.  I know in textual things they use quotations, so the literal is in there somewhere and liable to come spilling out.  And it's a reasonable bet that quotation is fair use.  But I don't think you can extend that to any output of literally similar or identical material.

I think you're conflating natural language processing (NLP) generative pretrained transformer (GPT) models with machine learning writ large. When an NLP GPT model is attempting to generatively reply to a prompt, it's simply stringing together tokens (read: words) in whatever is the most normative sentence to its training data. It literally is not just copy/pasting, but building the reply word by word. And in some cases, there really can only be but a narrow set of responses to a prompt - which is probably where you would cry foul in this case. But it's not like that original copyrighted document is stored in a database somewhere inside the neural network - it's been encoded into an extremely complex set of weighted and interconnected neurons. In the case of chatGPT, they use approximately 50,000,000,000 different parameters (read: neurons) to encapsulate their model. It's not storing the raw data itself, but rather the insights it learned from that data so that it can model possibilities not encapsulated in the data itself.

  • Hook 'Em 1
Link to comment
Share on other sites

12 hours ago, TwiceHorn said:

As a factual matter, copying "from memory" (as opposed to rote or "mechanical" duplication) tends to be less literal and may not be found by a judge or jury to be "substantially similar."  In these music copyright trials, that's usually the issue and the musicologists come on and explain how the copied "jingle" isn't actually that unique and, although the alleged copier was familiar with the copyright work, the similarity isn't such that you can infer copying from it.  And sometimes they find copying anyway.

Is there a difference between "copying" and "influencing" legally? In the Prince example above, it's clearly a copy, and just because you add some different colors and techniques you can clearly see the original work. The same can be said for sampling, even if you alter it a little, you can clearly hear the original work. 

Isn't that materially different from an artist being influenced by another artist? And similarly if an AI was fed a 1000 works of impressionist artwork, and creates something "unique" in that style, would that be subject to copyright laws?

 

Link to comment
Share on other sites

7 hours ago, nineliveslost said:

Here AI text to art with the Simpsons. I say CR infringement. 

Bizarre AI Simpsons characters spark serious debate | Creative Bloq

 

AI Simpsons charactersXrEKL9wiPrimGE5oDvrkbY-970-80.jpg.webp

 

I dunno.

My understanding is that my copyright gives me the right to make copies of my own work to sell, and I can rent out that right if I feel like it, and no body else can make copies of my work to sell without my permission.*

In this case, is this guy even successfully copying? The only reason I might think they were Simpsons characters is because the creator said these were Simpsons characters. Otherwise I'd think they were random clip art. At this stage it looks more like the Simpsons inspired it rather than he copied some art. Emma Watson might have a case for them using her face for Lisa.

Now if he slaps together a Simpsons-esque script to go with them, sure.

 

 

 

*except for the entire country of China, which does what it wants.

[EDIT: OK I prolly just made up that part about selling, because obviously some gazillionaire can't print up 100,000 of somebody else's best-selling novel and give them away for free without permission. I was going to get a Law degree last month, but it rained too much and I got tired.]

Edited by RDCanecutter
Link to comment
Share on other sites

  • 1 month later...

Well, here's the first judicial decision that the Copyright Act requires human authors.

https://storage.courtlistener.com/recap/gov.uscourts.dcd.243956/gov.uscourts.dcd.243956.24.0_2.pdf

This, at page 13, implies a different result had Thaler characterized authorship differently in his copyright application:

This case, however, is not nearly so complex. While plaintiff attempts to transform the issue presented here, by asserting new facts that he “provided instructions and directed his AI to create the Work,” that “the AI is entirely controlled by [him],” and that “the AI only operates at [his] direction,” Pl.’s Mem. at 36–37—implying that he played a controlling role in generating the work—these statements directly contradict the administrative record. Judicial review of a final agency action under the APA is limited to the administrative record, because “t is blackletter administrative law that in an [APA] case, a reviewing court should have before it neither more nor less information than did the agency when it made its decision.”

  • Hook 'Em 1
Link to comment
Share on other sites

On 7/12/2023 at 7:41 PM, RDCanecutter said:

I dunno.

My understanding is that my copyright gives me the right to make copies of my own work to sell, and I can rent out that right if I feel like it, and no body else can make copies of my work to sell without my permission.*

In this case, is this guy even successfully copying? The only reason I might think they were Simpsons characters is because the creator said these were Simpsons characters. Otherwise I'd think they were random clip art. At this stage it looks more like the Simpsons inspired it rather than he copied some art. Emma Watson might have a case for them using her face for Lisa.

Now if he slaps together a Simpsons-esque script to go with them, sure.

 

 

 

*except for the entire country of China, which does what it wants.

[EDIT: OK I prolly just made up that part about selling, because obviously some gazillionaire can't print up 100,000 of somebody else's best-selling novel and give them away for free without permission. I was going to get a Law degree last month, but it rained too much and I got tired.]

You're right close.  The exclusive rights protected by copyright are:

Subject to sections 107 through 122, the owner of copyright under this title has the exclusive rights to do and to authorize any of the following:

(1)

to reproduce the copyrighted work in copies or phonorecords;

(2)

to prepare derivative works based upon the copyrighted work;

(3)

to distribute copies or phonorecords of the copyrighted work to the public by sale or other transfer of ownership, or by rental, lease, or lending;

(4)

in the case of literary, musical, dramatic, and choreographic works, pantomimes, and motion pictures and other audiovisual works, to perform the copyrighted work publicly;

(5)

in the case of literary, musical, dramatic, and choreographic works, pantomimes, and pictorial, graphic, or sculptural works, including the individual images of a motion picture or other audiovisual work, to display the copyrighted work publicly; and

(6)

in the case of sound recordings, to perform the copyrighted work publicly by means of a digital audio transmission.

So, the ersatz Simpsons people would presumably violate 1, 2, 3, and 5.  Maybe more 2 than 1 because they aren't complete copies, but only derived from the original cartoons.

Underlying all of these exclusive rights is the assumption that what is reproduced, derived, distributed, or performed/displayed is the copyright work, or something derived from the copyright work, or a copy.

So, the act of infringement is to do one or more of those things with something copied from the copyright work.  We rarely have direct evidence of copying, so we use a substitute inference:  Did the alleged infringer have access or exposure to, or knowledge of, the copyright work and is the infringing work substantially similar to the copyright work.  If those things are proven, then we have copying and we have infringement.

The Simpsons infringer no doubt had access and seems to acknowledge same, but the substantial similarity prong may be missing.  Of course we also have an admission of an intention to imitate the Simpson's characters, which may be the same as or nearly the same as an admission of copying.

Can really bad copies of copyright work constitute infringement?  Even when the infringer admits the intention to copy or derive?

Also note that even if a non-human cannot be an author, that non-human can likely infringe a copyright, or at least the "instructors" of that non-human can.  And, aside from the act of "creation" of the copies, there's probably also another act of infringement, in this case, distribution and/or public display that is undoubtedly human-controlled.

Edited by TwiceHorn
  • Hook 'Em 2
  • Drool 1
Link to comment
Share on other sites

16 hours ago, TwiceHorn said:

Well, here's the first judicial decision that the Copyright Act requires human authors.

https://storage.courtlistener.com/recap/gov.uscourts.dcd.243956/gov.uscourts.dcd.243956.24.0_2.pdf

This, at page 13, implies a different result had Thaler characterized authorship differently in his copyright application:

This case, however, is not nearly so complex. While plaintiff attempts to transform the issue presented here, by asserting new facts that he “provided instructions and directed his AI to create the Work,” that “the AI is entirely controlled by [him],” and that “the AI only operates at [his] direction,” Pl.’s Mem. at 36–37—implying that he played a controlling role in generating the work—these statements directly contradict the administrative record. Judicial review of a final agency action under the APA is limited to the administrative record, because “t is blackletter administrative law that in an [APA] case, a reviewing court should have before it neither more nor less information than did the agency when it made its decision.”

Does this mean I can use that image however I wish and Thaler can't do shit about it?

  • Like 1
Link to comment
Share on other sites

  • 3 weeks later...

Here's an interesting one.  This piece, generated using Midjourney, won a prize at the Colorado State Fair.

An AI-generated image that won a prize at the Colorado State Fair

The author/artist, Jason Allen, filed for a copyright, naming himself (not Midjourney) as the author.

So, problem one overcome.

But, the Office inquired as to the author contributions, relatively, of Allen and AI and required Allen to disclaim any copyright in those portions of the work Midjourney created.  **As an aside, people file for copyright registrations on works including "preexisting material" that was not created by the author.  It is a pretty standard practice to require the copyright claimant to identify and disclaim such material.**

He refused the disclaimer, the Copyright Office refused the registration.  https://arstechnica.com/information-technology/2023/09/us-rejects-ai-copyright-for-famous-state-fair-winning-midjourney-art/

The actual decision, here:  https://fingfx.thomsonreuters.com/gfx/legaldocs/byprrqkqxpe/AI COPYRIGHT REGISTRATION decision.pdf

So, it seems we have a conundrum here.  You can't name AI as the author, you must include a human.  But you will also have to identify and disclaim what the AI created, which seems to be an insurmountable task, even if you are inclined to do it.

Also notable, the Copyright Office has put out Guidance on how it will handle registration applications for AI-generated works. https://copyright.gov/ai/ai_policy_guidance.pdf

Edited by TwiceHorn
Link to comment
Share on other sites

3 minutes ago, TwiceHorn said:

 

So, it seems we have a conundrum here.  You can't name AI as the author, you must include a human.  But you will also have to identify and disclaim what the AI created, which seems to be an insurmountable task, even if you are inclined to do it.

 

I’m trying to follow along, but explain this to me like I don’t know what the fuck you’re saying. 

  • Like 1
Link to comment
Share on other sites

Just now, UT_OB1 said:

I’m trying to follow along, but explain this to me like I don’t know what the fuck you’re saying. 

You try to copyright an image you prompted AI to generate.  You can't name AI as the author.  You can try to name yourself as an author in conjunction with AI.

You will then have to identify the nature of your authorship, as well as what was created by the AI.  The stuff created by AI will not be copyrighted. 

You can say that you prompted the AI 624 times, as did Allen, but that actually doesn't show up in the image.  The image is 100% AI generated. (Allen did do some stuff like clean it up in photoshop or something, and that would count, but not text prompts).

No copyright for you.

Catch-22.

Link to comment
Share on other sites

6 minutes ago, TwiceHorn said:

You try to copyright an image you prompted AI to generate.  You can't name AI as the author.  You can try to name yourself as an author in conjunction with AI.

You will then have to identify the nature of your authorship, as well as what was created by the AI.  The stuff created by AI will not be copyrighted. 

You can say that you prompted the AI 624 times, as did Allen, but that actually doesn't show up in the image.  The image is 100% AI generated. (Allen did do some stuff like clean it up in photoshop or something, and that would count, but not text prompts).

No copyright for you.

Catch-22.

From a technical perspective, the only point of authorship in that whole thing is the prompt given to the generative model, and maybe the other parameters and seed condition if they were specified. Given the same seed conditions and prompt, a generative model will produce a consistent and repeatable output that IS proveably deterministic. 

But then again, there's not much value to owning a copyright to a specific set of inputs, as any minute change can produce unique results with zero effort. Just punch in a new seed value and away you go on a new ideation path within the generative model. 

Link to comment
Share on other sites

5 minutes ago, Captainant said:

From a technical perspective, the only point of authorship in that whole thing is the prompt given to the generative model, and maybe the other parameters and seed condition if they were specified. Given the same seed conditions and prompt, a generative model will produce a consistent and repeatable output that IS proveably deterministic. 

But then again, there's not much value to owning a copyright to a specific set of inputs, as any minute change can produce unique results with zero effort. Just punch in a new seed value and away you go on a new ideation path within the generative model. 

The other thing is that the individual inputs, at least, are probably in the nature of non-copyrightable ideas, or damn close to it.  "Make the circle in the middle bigger."  I suppose, though, in aggregate, they will amount to the sort of "choices" artistes make in paintings, drawings, literary and musical works all the time.

But what the owner really wants is to be able to prevent the reproduction and distribution of the image and that seems like it's going to be a no-go.  He or she can prevent someone from using the same prompts to create the same image, but if they just photo/copy it, they're free to do so.

Edited by TwiceHorn
Link to comment
Share on other sites

56 minutes ago, TwiceHorn said:

Here's an interesting one.  This piece, generated using Midjourney, won a prize at the Colorado State Fair.

An AI-generated image that won a prize at the Colorado State Fair

The author/artist, Jason Allen, filed for a copyright, naming himself (not Midjourney) as the author.

So, problem one overcome.

But, the Office inquired as to the author contributions, relatively, of Allen and AI and required Allen to disclaim any copyright in those portions of the work Midjourney created.  **As an aside, people file for copyright registrations on works including "preexisting material" that was not created by the author.  It is a pretty standard practice to require the copyright claimant to identify and disclaim such material.**

He refused the disclaimer, the Copyright Office refused the registration.  https://arstechnica.com/information-technology/2023/09/us-rejects-ai-copyright-for-famous-state-fair-winning-midjourney-art/

The actual decision, here:  https://fingfx.thomsonreuters.com/gfx/legaldocs/byprrqkqxpe/AI COPYRIGHT REGISTRATION decision.pdf

So, it seems we have a conundrum here.  You can't name AI as the author, you must include a human.  But you will also have to identify and disclaim what the AI created, which seems to be an insurmountable task, even if you are inclined to do it.

Also notable, the Copyright Office has put out Guidance on how it will handle registration applications for AI-generated works. https://copyright.gov/ai/ai_policy_guidance.pdf


 

fuck him

  • Hook 'Em 2
  • Drool 1
Link to comment
Share on other sites

10 hours ago, TwiceHorn said:

You try to copyright an image you prompted AI to generate.  You can't name AI as the author.  You can try to name yourself as an author in conjunction with AI.

You will then have to identify the nature of your authorship, as well as what was created by the AIThe stuff created by AI will not be copyrighted. 

You can say that you prompted the AI 624 times, as did Allen, but that actually doesn't show up in the image.  The image is 100% AI generated. (Allen did do some stuff like clean it up in photoshop or something, and that would count, but not text prompts).

No copyright for you.

Catch-22.

Thanks!  That cleared it up. 

Link to comment
Share on other sites

If AI art is allowed to be copyrighted, then there is nothing to stop someone from telling AI to generate a trillion(s) images for a variety of subjects. Then you tell the same AI to continually monitor the internet for anyone that is publishing anything similar, if not identical, to your copyrighted images. Then start the legal process to demand small dollars amounts. AI can probably handle the emails, legal filings, etc.  Sit back and watch your bank account get larger.

Link to comment
Share on other sites

  • 1 month later...

So, on one of the first lawsuits by creators for infringement by AI engines for using copyright material as input, the first decision was largely in favor of the AI outfits.

https://news.artnet.com/art-world/federal-judge-sides-with-ai-companies-in-artists-copyright-dispute-2387654

Seems like something of a technicality because a lot of the works allegedly copied weren't registered prior to filing suit.  Copyright is weird in the US in that a claim to a copyright must be registered before you can file suit for infringement of the copyright.  Hardly any other country has a registration scheme.

Link to comment
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.



×
×
  • Create New...