copyright exists in the US under this clause in the constitution:
> “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.”
not sure how letting people copyright the output of a program that ate the work of a million other people would "promote the progress of science and useful arts"? or how there's an author to reward? or how there's writings or discoveries?
of course this also applies to the US endlessly extending copyright also - nothing Congress can do, no matter how well intentioned, is going to encourage Walt Disney himself to produce anything new of value, nor did keeping Happy Birthday copyrighted until 70 years after the last Hill sister died assist any of them in coming back to life to come up with a second hit.
the dumb optimistic part of my dreams that having matrix multiplications shitting out words and pictures will cause a reassessment of why we let IP laws get so fucked up to begin with, but of course, it won't - it'll just mean Disney and Warner et al will find some way to keep selling tv shows and movies and the humans who produced the art who get shoved into the matrices get shafted.
> not sure how letting people copyright the output of a program that ate the work of a million other people would "promote the progress of science and useful arts"?
I agree with you. It's a mess.
AI output should be copyrightable, but only if the provenance of its input is carefully audited. That is, the owner of its output owns or licenses all copyrights of its inputs. An artist that trains a model on only her own work should be able to copyright its output.
Midjourney is IP theft to the degree that it was trained on copyrighted work without permission.
> An artist that trains a model on only her own work should be able to copyright its output.
Why?
If she made a LORA, well she's using someone else's Stable Diffusion checkpoint, which was trained on things she wasn't permitted to use.
If she made a fine tune, same thing.
If she trains the UNets and the VAE from scratch, on just her own work, what did she use for conditioning? CLIP? Which was trained on LAION2b?
Did she also train her own text encoder? What is her conditioning approach?
All these models, why are we talking about them? If it weren't for training on the unpermissioned data, these models would be toiling away in obscurity.
If someone conducts a bunch of unethical medical research, and they discover some important biomechanics, do I get to repeat all of that stuff, "but ethically?" Am I untainted? If the police discover real evidence in a procedurally problematic way, for example by torturing a confession that turns out to be true, do they get a "do-over"?
> but only if the provenance of its input is carefully audited
All I am saying is, you haven't "carefully audited" the feel-good story "artist that trains a model on only her own work should be able to copyright its output."
No idea what you're taking about. Throwing arcane technical terms about does not advance your argument. Nor does the triumph or obscurity of the AI research elide the copyright interest of artists in how their own work is used.
I don't think you should be downvoted for having opinions that are based on a lack of information.
Everything I talked about is to illustrate that it is impracticable to resolve the issue of authorship by doing something like, "I got permission for the data."
The first step on the expanding brain meme is understanding all the stuff I wrote, it basically means "in order for the generative AI technology to work, you use a lot of other people's unpermissioned data in many, many ways, that are impractical to replace."
The second step in the expanding brain meme is to understand that your opinion, "if you use art you have permission for in a generative AI model, you're In The Ethical Right," is co-opted by Adobe. Because they happen to have the good fortune to run an image licensing bureau, so they can "get permission" for "the data" and you were convinced that's "all they need." When of course, they use tons of unpermissioned data, because that's how these models work, but you just don't know enough about the problem yet to understand that. So you're going around essentially astroturfing this feel good position that has no stronger technical basis than doing what CivitAI users do. It just has a really appealing starch to it.
You don't win an argument by gatekeeping and patronizing like that. There's information you don't have, apparently. This is not a technological issue alone, but also a social, legal and artistic issue.
It is very clear that artists whose art was used to train AI have copyright interest in that AI's output. The implementation details of the model are irrelevant to the interest. The technical difficulty of auditing its inputs' provenance is irrelevant to that interest.
No. The only things that's clear is that works from AI should be non-copyrightable and under at least a creative commons, and in my opinion a copyleft license akin to AGPL and gpl3 (I think disagreeing on this opinion is fine however).
Output from AI should be non-mercantile.
Why? Because most text/images used by AI company to create, test then pretrain/train their tools are creative common /unlicensed (I trained/fine-tuned an OCR in 2016), and thus, copyrighting the output of any model would be stealing from the general public.
> Because most text/images used by AI company to create, test then pretrain/train their tools are creative common /unlicensed (I trained/fine-tuned an OCR in 2016), and thus, copyrighting the output of any model would be stealing from the general public.
"Most"? If any of the training inputs are copyrighted and the AI company does not own or license the work, nor applies all licenses correctly, then the AI company should be liable for copyright infringement.
There's really no other way forward as a society, otherwise creatives' work can always just be stolen at will by laundering it through an AI model. There would be no incentive for the next HR Geiger or Moebius to work hard on a distinctive aesthetic when anyone with a computer could make art just like theirs.
If all of an AI input is public domain, I'm not sure I agree with you but I think your argument has some merit.
> There's really no other way forward as a society, otherwise creatives' work can always just be stolen at will by laundering it through an AI model. There would be no incentive for the next HR Geiger or Moebius to work hard on a distinctive aesthetic when anyone with a computer could make art just like theirs.
The games market has been dealing with this for a while and things work out. We get original, meaningful games that are easily pirated, and original, meaningful game mechanics copied by giant, powerful corporations. Nonetheless those people thrive. So I don't know. This is a kind of doomerism that isn't going to play out, it's reacting to a lack of creativity among monetization for illustrators and a lack of opportunities that have always existed, rather than anything specific about the mechanics of laundering this or AI models that.
The copyright infringement isdue is orthogonal to my point.
Let's say you have all the IP of a LORA. Nice. Now, the model you're using will output something similar to your copyrighted work, but this will be barely .001 percent of what the model used. Most (if not all) other images it used will be cc, unlicensed (most images on the internet) or copyrighted but unidentifiable.
In this case, even if the output resemble the copyrighted work used for the Lora, the rights should be owned by the public.
My opinion is that it should be copyleft (because you can't possibly attribute all the work used for the generated image, so the only fair use is copyleft use).
> Most (if not all) other images it used will be cc, unlicensed (most images on the internet) or copyrighted but unidentifiable
You keep saying "most". I get that you believe copyright infringement is orthogonal to your point, but it's really not.
If you can argue your point by saying "all, 100%" free and clear - public domain or otherwise permitted by the owners - then I can consider your point.
Otherwise, no, the model does not produce copyleft or public domain work. It's copyright infringement and IP theft.
"Unlicensed" does not mean it is not under copyright, by the way. CC usually means there are restrictions or requirements for use. Attribution, for instance.
But if the model were trained only on public domain work, I'm still not completely convinced. Perhaps. I do find Jason M. Allen's argument more compelling than Tamara Pester's
The alternative being what? You can't use images produced by the model?
I don't see a middle ground here, and I fail to imagine see any real-world instance where it's logical that ai-generated images fall under copyright. It's either public domain or no public use (which, tbh, I think is fair too)
> I don't see a middle ground here, and I fail to imagine see any real-world instance where it's logical that ai-generated images fall under copyright.
Here's a point of view on the difference between painters and everyone else: everyone else lives in the world and deals with its realities, but bonafide painters try to paint the world the way they wish it worked. Some even live the world the way they wish it worked, even when it doesn't make sense (like in terms of an objective reality), as long as they're alive they can do their thing and they can prove that something is possible, even if impracticable.
This isn't a bad thing.
If logical means, "the way orwin wants the world to work," then sure. You're doing the painter thing in the tech world. You said you want the copyright to be this way and the way things to work to be that way. If you want, you can write a license on a model that maybe you trained on all public domain data, and that the license says all the things it makes are also public domain, and maybe people go along and respect that license. This can all coexist with other stuff, it can be real.
It just doesn't mean it makes sense. Just because something happens doesn't mean it makes sense. Another way I've heard this put is, feelings that are more strongly felt are not more valid. That said, in my world, I know people try to pitch "give me a $25,000 grant to write a license," instead of engaging with the core creative part of the RFP, and they don't get the grant. There is little appetite for the painter's way of doing things in the tech world. That's a bad thing but it's also true.
Do we agree that in the _real_ world, it is impossible to determine which image have been used to generate the model (not talking about just the LORA part here).
So the solution i see is:
- we ban AI-generated image from public use because of IP: basically, the use of any image not in the public domain, even CC0, will cause that
- Most non-public-domain images on the internet you can use to train stuff are CC and CC0 (which is kinda public domain if i remember my wikipedia days). My point is: the majority of the images used to create the model (again, not the LORA part) are public domain or free to use as long as they are attributed. It's impossible to attribute according to what CC is demanding. The list would be impossible to find. So either we get back to the first sentence, and sue AI-generated images to oblivion, or the image are public domain/CC0.
Also, the legal point: the creator of the art isn't a human, so it isn't copyrightable. A curator do not have the copyright of the image/painting he found at a garage sale, even if he put a lot of effort to restore it.
Humans produce art in part by taking in what others have created before them and the output is some combination of many input factors, including influences by the works of others (whether we're talking painting or architecture). And today the influence by pre-existing art would be likely to be considerable in terms of factors, given the Internet, access to art and visual media broadly.
As AI advances radically over the next decade or two, it's going to end up being comically hypocritical to pretend humans are more deserving as originators.
> As AI advances radically over the next decade or two, it's going to end up being comically hypocritical to pretend humans are more deserving as originators.
What does that have to do with anything?
> “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.”
Matrix maths isn’t a human so can’t be an author or inventor so there’s no reason to grant any protection to any of the results of any multiplication .
It's odd to me that an image such of this is rejected unless labeled as generated by "AI", yet an image heavily edited in Photoshop or some other software with algorithms is not.
The point of copyright is to protect the rights of human creators. AI tools are not human creators.
If I go to an artist and say "make me a picture with X, Y, and Z features", they own the copyright unless they agree to transfer it to me. An AI tool never owns the copyright to its work because it's not a person, so you cannot acquire the copyright on the work since you didn't generate it.
It's an entirely different process from using Photoshop (&c), and I think conflating them requires either a significant lack of understanding of how AI-based generation tools work (the most common issue) or a deliberate desire to confuse the issue.
How is this different than attaching 10 paint cans of different color from strings and poking hold in them and letting them swing?
Could the creator of the ai producing code then hold copyright, or maybe it’s those who found the weights?
This ruling seems to suggest any method of automating works would not qualify for copyright.
If so. We are just one step closer to abandoning the notion of a copyright, which still is the only logical solution in a world where we can have and do anything we desire at the click of a button.
> How is this different than attaching 10 paint cans of different color from strings and poking hold in them and letting them swing?
The "creator" of the art in that case is also the creator of the system which generates the art. You didn't create Stable Diffusion, MidJourney or whatever else, it's just a tool you have access to.
I'm not sure if that's a difference that should matter, but it's clearly a notable difference.
If a developer writes their own Stable Diffusion variant from scratch, uses nothing more supportive than Numpy and writes all the supporting code themself, scrapes the web for data themself, trains image generating models on their own hardware, and then uses their own self created tool/system to generate AI, should that be copyrightable? In this case, everything short of Numpy would be the hand written code logic of that one developer (or small team). The data used to train the system could be argued is no different than a person viewing the world they live within... copyrightable?
Stepping into the thread, but as someone who agrees with the copyright office, in the case you laid out, I think there would be a better argument there actually for copyright, yes. Personally, I disagree with the last part (a tired argument about AI imho) but it feels like it would have a better argument.
I disagree with the copyright office. I think my creative output as an engineer is not necessarily less creative or original than an artist. We are making an arbitrary distinction here. Artists make devices, machines sometimes. Is the output of those machines not copyrightable? I bet it is.
So. The string lengths, colors, and hole position in the paint cans along with initial velocity and angle are what can qualify for copyright?
Are paint brushes just tools people have access to, does the manufacturer of the paint brush own the painting?
Prompts are no different than positions and angles the artist instructs the tool to motion. The only real difference is the ratio of perceived work to output.
Even a prompt could take years to derive. I am 40. And 30 years ago I could have not com up with the following, making the prompt below 40 years in the making.
“Field of dead dreams, with copyright monsters flying above looking for there next meal”.
Spell check fixed a few things, along with typo correction, does Apple know own a copyright of my work?
I doubt it, that would be like someone setting up the hypothetical "swinging paint cans" for public use then trying to claim copyright on the product. I think the slam dunk case would be you create your own system, train it on data you own, and then you own the output.
Again, this doesn't strike me as a useful standard for copyright, but I think it's the state of the law today. Obviously with new technology there's a need to revisit the law, and honestly even without the new tech copyright law was already a steaming mess.
I bet I can create a machine that uses those 10 cans of paint to do computations. That's all a computer is, really. Those 10 cans of paint swung by an artist are a creative endeavor, but my output as a software engineer with the 10 cans of paint is not? Seems arbitrary and unfair and probably shouldn't stand careful legal scrutiny.
In this case, the winning image wasn't a single prompt to image. The author generated a lot of images and invested lots of hours into it till he considered it ready. AI was more like a custom stock asset generator in this case.
We will need to seriously revise copyright laws in the age of AI because this new technology is simply not compatible with the assumptions the current ones are based on. (regardless how much lobby power are in them)
> The Board finds that the Work contains more than a de minimis amount of content generated by artificial intelligence (“AI”), and this content must therefore be disclaimed in an application for registration.
The human-authored portion of the work would be copyrightable if the author disclaimed the portions that were created by an LLM.
Does this mean that if I use an even more deterministic tool like, say, autocomplete to write software, that bit (say an auto-generated function definition), doesn't attract copyright?
Not that it doesn't make sense to say what they did, but it seems not quite compatible with how copyright already seems to work in practice. Which is, instead another demonstration that copyright in general is a pretty internally-inconsistent system when it comes to the details.
If the autocompleted part were copyrighted, why would you be able to use it freely? What claim to you have to authorship?
Did you, perhaps, write the stuff that the autocomplete used as context to give you a selection of choices to make about what to put there? Because that still sounds like a human making choices about things to me.
This is a vaguely recognizable metaphor for prompt engineering, but I think you're orders of magnitude off in estimating the ratio of human choice to computer extrapolation. Sure, a compiler or something maybe is a better metaphor, but even then the concept, the source material is generated by a human.
You could say that a prompt is source material and the generator is just a compiler analogy, but I think they made the only choice they could here. Otherwise you can use a generator to make literally a billion images, copyright them all (including aspects you did not make a choice to add, like details on people in the background there), and control art for the rest of time.
I'm not a copyright lawyer, but I think its pretty easy to argue that generated boiler plate constitutes a "de minimis" amount of content and so you would still own the copyright of all the code
> In this case, the winning image wasn't a single prompt to image. The author generated a lot of images and invested lots of hours into it till he considered it ready.
This is exactly the “sweat of the brow” argument for copyrightability that courts have consistently rejected in favor of a creativity requirement since long before AI tools existed, though, so its probably not the argument you want to make for copyrightability.
What's the threshold between the AI input and my input that moves it from copyrightable to not? We can automate a lot of those things an artist did on a photo, does that make it not copyrightable (that's a long word, let's use CR)).
We kind of already have automated image-processing today, where does it change from cr to not-cr. I can take creative photos and then edit them and I can own the the copyright. If I took my photo and used automated improvement tools (like in google photo), I think it's still copyrightable. Now I take my automated outdoor camera at my house and take pictures of things, then I put them through google photos in my automated image pipeline, I guess they are copyrightable?
Then I use those photos as input into a non-copyrightable-by-definition system like midjourney or whatever, now the output is not copyrightable. Seems pretty arbitrary. If an artist creates an artistic work (lets say it is a mirror of people), it's cr. If I as a software engineer create software, that software is cr. But the output of the cr is not cr?
Still, it's a hard case to make that any AI-generated imagery lacks sufficient human input into the creation process. Both Midjourney and Stable Diffusion offer a number of parameters allowing the user to control how the generation will go (CFG, samplers, etc). So IMO there is necessarily some human input required to make the process work well.
As someone who has been doing a lot of work in SD, I just wanted to strongly second this.
Stable Diffusion and the like are much closer to when photoshop first appeared than they are to than the completely automatic art generation machine people make them out to be.
I remember well both digital cameras and photoshop receiving initial criticism from "real" artists... until they were both widely adopted by these communities.
The critique of digital photography was that it's not really art because you can just take a bajillion pictures and then filter out the ones that don't look good. Photoshop users (especially web comic creators) got similar flack that it didn't take real artistic skill to make nice images in photoshop.
For anyone young enough to always have photoshop and digital photography will likely be surprised that there was such resistance and snobbery around using these tools. There was also the same fear that these tools would ruin art by making "real" artists obsolete.
> it's a hard case to make that any AI-generated imagery lacks sufficient human input
The test is not "input", but "authorship". If you commissioned some art from an artist, you would have input on what they make but the artist would be the author and copyright holder of their work.
Its hard for me to accept that the prompter "authored" the output of an LLM. The output of the LLM was entirely generated and authored by the LLM. It did so at your command, but that doesn't count as authorship.
Artists can copyright their work no matter what. If they create a computer program to scatter paint, they can copyright the output. If I as a programmer do creative work (ie programming), it should have the same support.
This was just a bad decision about copyright, the only reason to support this distinction is to protect art and artists.
> Artists can copyright their work no matter what. If they create a computer program to scatter paint, they can copyright the output.
Not if the computer program uses GenAI, at least without disclosing and disclaiming from copyright the GenAI contribution, under the current (absolutely ludicrous) Copyright Office interpretation.
> If I as a programmer do creative work (ie programming), it should have the same support.
It does: neither copyright law nor the copyright office defines classes of people (“artists” vs. “computer programmers”) with different effects on if they create artwork with the same type of automated tooling.
The copyright office may make a spurious distinction between GenAI and other tools, but it doesn't distinction between classes of creator the way you seem to think is an issue that needs corrected.
What If I personally train a network on my own art? The computer is mine, the software is mine, the reference art is mine, but I have no rights to anything produced?
I think the question in front of the courts then would be to what extent your creation of the network is important vs. the training data you used. Is it really feasible to train a robust network on whatever reference material you're personally able to generate?
It will definitely end up in the courts. This* is a list of artists that have sold the rights to their music catalog. If Sony, who spent $200m on Bob Dylan's master recordings, sets up servers, pays developers, and creates a network using only Dylan recordings they have the rights to, can they copyright newly generated songs? I bet we will find out soon.
I don’t quite understand what you are saying. In his scenario, what details would make it not copyrightable? If the network is paramount to the creation does that make it more or less copyrightable?
It seems like currently the chain of ownership is so complicated and spread out that its not feasible. But I dont see a good claim by another party in your case. Best one might be a library or hardware but thats getting ridiculous.
To rephrase my comment below - what if Sony, who owns Bob Dylan's master recordings, use their own hardware, software, & reference materials to generate new "Bob Dylan" music?
If I generate a base image with Midjourney, and then apply a mild gaussian blur in Photoshop to a 10x10 pixel space within that image, can I say I've created it? If not, why not?
Given this it would seem running any simple algo such as balance colors or adjust lighting would disqualify the output for copyright. Only change here was the algo used.
No, that isn’t true. If you do substantial original work and employ AI that is different than if the entire work is the output of AI. Nobody is saying using AI for any one part taints what is otherwise a substantially original work.
The substantial original work here is the prompt authorship. That’s the “paintbrush” in this work. And the “paint” is the AI.
It’s very easy to spend hours iterating on prompts for the perfect photo while using human judgment to figure out how to steer the image. That’s meaningful human involvement and should be able to be protected.
Does that mean an art gallery curators spending hours, days even over a thousand pictures that people took randomly finally finding the perfect one for a project he has should have the copyright to that picture? Considering it's not a professional, just a tourist that took it.
Sure. But where should that exist on the spectrum of copyrightability vs painting with a physical brush? I don’t think we can just handwave AI generated art away but isnt your contribution a lot less than painting?
There's a perception (which AI hype has fed) that “AI” systems like MJ and SD are categorically different than traditional or digital artist tools, and are something like brute animals with paint on their paws making images: neither humans whose work copyright protects nor tools wielded by the hand of humans, but forces of nature where the human role is effectively a matter of selecting one of the generated outputs from a metaphorical pile left in the cage.
This is a factor in shaping the legal treatment of work generated with the use of “AI” tools.
Ok - I'm going to pick on you since you responded with this first:
Here's my argument:
1. Copyright in some sense is probably a net win for society.
2. Copyright in it's current legal form in the US is - fucking unquestionably - abusive.
7 years with a 7 year renewal after review? Probably ok, feels long in today's digital age.
To the author's death? oof, that's a long time to protect works, but maybe I guess, in some specific fields of work.
To the author's death PLUS 70 fucking years? Get the fuck out of here. Literally - who the fuck other than giant corporations is POSSIBLY benefiting from this? This is a fucking insane level of protection, and is outrageously harmful to generating new content (Just take a peak at all the fucking remakes that constitute our new films, and the miserable content licensing restrictions that pop up everywhere and kill so many new things)
---
So my stance is firmly: Yep, they probably are taking some of your work. But also: Your work is also grounded in the work of others (again - fucking unquestionably). Further - the copyright system at large is SO fucking harmful that I don't really think it should exist anymore (side note: I'm really underplaying my feelings here, "nuke it from orbit" is probably still an understatement... "I'd like to see it suffer before it dies" is probably close).
So... if you want to talk about bulk exploitation - go fucking yell at the current media companies, because holy fuck are they the worst fucking companies in the world to make this claim. They are leeches.
If you want me to step to your side: CURB THE FUCKING ABUSE.
Otherwise... you are the shitty cab company and the LLM is uber: Are they ethical? Eh - Nope. Are they unquestionable better for most people? Yes.
Will I take a useful and unethical company over a stagnating and unethical company? Yes, yes I will. Would I prefer real governmental action to either of those? 100%.
I agree with most of what you are saying, but the key gap is that the victims here are not "unethical companies". They are 5+ digits of independent artists who had no idea this was going to happen to them.
If big art was a thing, your argument would be good, but Disney doesn't account for most of the input to these image algorithms.
> They are 5+ digits of independent artists who had no idea this was going to happen to them.
But these people are already making essentially zero income from copyright (it's mostly tours/merch), and have been (again, and again, and again) utterly fucked by the current publishing companies that are so interested (financially) in killing this AI work.
They are already victims of the current system. I don't see how introducing AI changes their position all that much: They still have low bargaining power, low financial returns (avg is under 6k/year) and mostly aren't doing it as a career.
Are a few of them exceptions? Absolutely.
Will there keep being successful exceptions with AI: I'd bet yes (with a lot of cash).
Will the way they work change? Probably. Some will enjoy the change, and some won't.
I just don't see any compelling reason in your comment to argue to keep the current cluster-fuck of abuse that is modern copyright.
Arguing that the currently abused might continue to be abused might be a valid reason to reject both systems, but doesn't pass the bar to stopping AI works in my opinion.
To argue an alternative:
If anything - many indie artists might be much more successful if they're allowed to play/remix/sample/use works that are currently sitting unused behind the outrageous copyright duration we have right now.
If I open Photoshop then via my inputs it indirectly generates pixels using various tools/brushes/automations, it is copyrightable.
If I open an LLM, enter a bespoke prompt, then the tool generates pixels using that prompt, it isn't copyrightable.
Things can be copyrightable even if there isn't a 1:1 between human input and copyright output. Where is the cut-off exactly? What about Photoshop's Content-Aware Fill, is that copyright or not?
What if I take LLM output then color correct it, is that now my copyright? Since it effectively has no authorship before I color shifted it.
This is the difference between legal thinking and CS. There is a spectrum between 100% human-generated (and copyright-eligible) and 100% machine-generated (and ineligible), and in between there are gray areas.
To us, that is a problem, because it implies ambiguity. Where is the cut-off?
In law, that is normal. No actions are the same; everything is on a spectrum from allowed to illegal. Judging the boundary is made on a case-by-case basis through legal argument, evaluated by a judge or jury.
So to the Copyright Office, it is sufficient to say that this particular way of AI generating art doesn't involve enough human creativity. But you won't get them to define enough in the abstract, for any general case, because that is essentially impossible, and can only be done by evaluating concrete cases.
I often find it interesting how when these types of issues come up, technologists often interpret it using "transitive rules" that are natural in math and engineering, but "the real world" doesn't follow this logical progression. That is, if you have "A implies B, and B implies C, and C implies D, then A implies D". The problem is that in "the real world" there are enough slight differences at each step that most people feel that A does not imply D.
Point being, there are lots of comments here along the lines of "If creating art with a tool can be copyrighted, how is art created with Photoshop copyrightable but art created with AI is not". The difference is that most people (and, most importantly, the US Copyright Office) believe that AI crosses the line between being "just a tool" to actually being the author of significant parts of the work. The fact that that line is gray is not a fundamental problem, and as you point out it occurs all the time in the legal realm.
> That is, if you have "A implies B, and B implies C, and C implies D, then A implies D". The problem is that in "the real world" there are enough slight differences at each step that most people feel that A does not imply D.
So you just got it wrong and A does not imply D. Obviously logic applies to the world outside math and computer science.
Normal, sure, but still a problem in the sense that they need to clarify it and try not to write ambiguous laws. The existence of the law, lawyers, and the whole judicial and lawmaking systems is proof that its difficult and complicated.
I just dont see the contrast with computer science here.
Sorry, not everything is evaluated on a "case-by-case" basis per say, there is a difference between a "Bright-Line" (clearly specified rule in court) and a "Balancing-Test" (what you describe).
I think it's actually pretty damn easy to define "enough" precisely. If the AI + human generated image cannot be easily recreated by AI, than it's enough.
So, for example, random image from civit.ai which is nothing more than a prompt, and negative prompts. This fails the test because I can run the image through CLIP and get (almost) the same exact prompts used to generate it
Now let's say I generate an image using a bunch of custom tools and extensions, such as regional color-loss (i.e. top right of the image must be blue and I will force the model to prefer blue outputs), and leveraging a lora model at 0.3 weight trained on my kids sketch-book drawing style. There's no way for any AI model that hasn't been configured exactly by my human hands to deduce anything approaching the prompt used to generate it (CLIP/Other LLMs doesn't know about extensions and loras).
Courts need a bright-line here, balancing tests lead to miscarriages of justice, which is what looks like is going to happen here a lot.
>I think it's actually pretty damn easy to define "enough" precisely. If the AI + human generated image cannot be easily recreated by AI, than it's enough.
That's a terrible test, because the capabilities of LLMs are constantly evolving. And the words "easily" and "recreate" just creates more ambiguity.
There are fairly few bright-line tests, because they're hard to write.
For instance, consider your proposed test: The copyrightability of works would change retroactively depending on the capabilities of AI. Something that can't be reproduced by AI today could be next year. Is that a desirable outcome?
> I think it's actually pretty damn easy to define "enough" precisely. If the AI + human generated image cannot be easily recreated by AI, than it's enough.
Would you be on-board with the inverse of this?
You cannot copyright what cannot be generated by a human.
This seems like a misunderstanding of the argument. It has nothing to do with inputs, it has to do with the creative process of making of content. It doesn't have to with the method but the creative input and making of content.
Example used above: I tell an artist to make me a picture of a dog. They created the image, and they own the copyright. They would have to give the copyright to me, I do not own it merely because I made the request.
No. You don't know who wrote the software that implemented that algorithm. In effect, me as the engineer who created photoshop was the creator of that. You needed me in this example to do that. You don't know if ai created that algorithm - it has been happening for decades that some software was created by other software - and I'm not just talking about optimizing compilers.
There was a creative algorithm, and 99.999% of the time there was not a specification that determines every single bit.
The work output of the programmers, the thousands of people involved in that pipeline to create the software - "that's different", say the artists.
You also don't know who created the paint or what their secret formula is in traditional art.
Drawing your art in Photoshop is clearly authorship. You are manipulating each pixel from scratch to be how you want it. There may be algorithms between your mouse clicks and the final product, but you as the author still have complete control of the whole art. You can tweak each pixel until its exactly how you want.
Asking an LLM for a picture of a duck is clearly not authorship. You did not create any pixels in that image, the LLM did. You just asked for it. Its no different than googling for a picture of a duck and claiming what you find because you engineered the google search query.
There is a lot of murkiness in between, where you take assets from an LLM and use them in your own creation. I think having to disclaim the assets which were not created by you is a very reasonable solution.
so if i tell an artist to do some revisions i own that work now? what exactly did the human do in this case? wrote some words in a certain order? so the prompt is copy writable then? if that's the case can i just copywrite certain prompts that are commonly used?
If a prompt is commonly used, then it wasn't created by you, and isn't copyrightable. It has to have enough original content so it could be copyrighted. I am sure a short poem is copyrightable.
But then again, fair use also allows citing most (if not all) of the short poem.
The difference is the underlying source material. With photoshop it comes from your brain, with AI tools like Midjourney it is derived from other peoples brains.
While we're at it, why not reject copyright on photographs? After all, it's just a machine capturing whatever's in front of it, all the human does is push a button. /s
What troubles me so much about copyright in the context of AI generated images are that we don't have any clear "Bright-Line" for what constitutes human authorship.
For example, let's say that I generate a bunch of images using Automatic1111 (the main Stable Diffusion webUI). This webUI has hundreds of extensions which give tons of additional functionality. If I generate an image using two controlnets, regional prompting (human deciding where particular prompt or part of prompt is applied in resulting image), and fancy prompt engineering (i.e. blending embeddings of multiple tags together, or deciding to change the prompt midway through the diffusion process), then I claim that I have put in enough human authorship that I think current courts should grant copyright in such a situation.
The courts have decided that a human simply gathering a bunch of AI images into a kids book was "enough work" for the kids book to be copyrightable. If I put at least that kind of effort into making an image (i.e. by writing what is ultimately a program, see comfyUI workflows for a visual example), than I think that image should be owned by the human who wrote the program to generate it.
Courts are dangerously close to deciding that any diffusion model being used as part of the images creative process would nullify the ability to copyright it.
> The courts have decided that a human simply gathering a bunch of AI images into a kids book was "enough work" for the kids book to be copyrightable.
Yes, the book is, the images aren't.
> Courts are dangerously close to deciding that any diffusion model being used as part of the images creative process would nullify the ability to copyright it.
I honestly think it should. Generative AI can't exist without public domain/unlicensed work. Even if your Lora is 100% copyrighted/personal work, most of the images that constitute the model are public property.
Sale with any stuff written by ChatGPT, or any code by copilot. All should be open-source or public domain,at least. To me it should even be copyleft,so the book (to come back to it) should be public domain too.
I'm somewhat curious about the legal status of publicly using / exhibiting and/or selling uncopyrightable media, such as AI generated imagery and video. If it is uncopyrightable, is it also a free for all - as in once publicly exhibited it is "free media" anyone can take and remix?
What about AI generated media that includes / incorporates copyrighted elements, but used the legal loophole of parody? As parody, anyone can include copyright protected elements of others in their original works under US copyright law because satire is protected speech.
If someone were to create a "satire AI" does that completely break copyright law?
I think this decision is incorrect, even after reading the decision itself. There is human intervention at every step.
Midjourney was made by humans.
Were I to construct a robot that wandered about randomly snapping hundreds of photos, then I select one of the photos, apply Photoshop filters and then win an award, it sounds like the review board would reject registration.
If I trained an AI model only on my own illustrations alone, it sounds like they would reject copyright registration.
In all of these cases, humans were involved at every step.
There’s going to be a lot of low effort takes about this.
In the context of the Copyright Office’s previous few decisions this makes sense.
Midjourney is toast if it does nothing to stop the by-design inertia of the quasi-caselaw way these things are decided.
The source of the most flaws in the Zarya of the Dawn registration rejection came from the Copyright Office copying and pasting deeply flawed technical explanations directly from Midjourney. It’s on Midjourney and no one else. They ought to be paying a great IP firm to submit rebuttals on behalf of losing users, but whatever.
“Yeah bro, but predict the future.” In this particular case, you're going to be able to register AI generated artwork - maybe not in the specific way that Midjourney does it, because the Copyright Office isn't going to pull their own pants down after issuing the Zarya of the Dawn decision, pulling their pants back up, in reaction to social media pulling their pants down. But ControlNets and T5-style grammatically substantive prompts will do.
The recent Andy Warhol decision and the way Google Books actually panned out means Stability and Runway are boned, and probably Midjourney too. The Author's Guild-like litigants will get a small payout, they will not prevail in a huge way, but they will prevail - they actually prevailed in Google Books too but people don't understand that, the news reported it as "Google won" but it actually lost in that it settled and didn't get to do any of the things it really wanted to do.
Meta will be forced to stop distributing its tainted LLaMA, express permission will be needed for all intents and purposes, but academics will be "fair use." I don't know if there's going to be some loophole to academize their work, it's just going to be hard to argue that compute resources and devops - the things Stability's, Runway's, Meta's teams really bring to the table compared to the academy - mean they ought to benefit from the same protections as a traditional education institution, in spite of being commercial entities.
It remains to be seen what will happen with Adobe's efforts. Nobody would be talking about latent diffusion models or pixel diffusers or whatever if it weren't for the fact that they were trained on content that wasn't expressly permitted. So it doesn't matter if Adobe got express permission for the training inputs of its models, so long as it's also using this code that maybe wasn't "trained" (in the ML sense) on unpermissioned content in this particular case but was definitely authored in the context of unpermissioned content in an equally essential way. It's the same reason you can't conduct unethical human research and launder it into something clean. It taints everything.
Anyway, I doubt Adobe trained their own text encoder unencumbered. It's acutely obvious with these community efforts at "training" stable diffusion "checkpoints" on "CC0" content. My dudes, you're using CLIP, which is trained on LAION 2b. It's just so painfully low information. Adobe will not prevail.
But Adobe surely will convince the low information bros that their way is bulletproof. In an intellectually honest way, nobody using diffusers or transformers with permissioned content will just get away with it.
Favorite-ing your comment to see if you end up being right.
My opinion is the tools are here, whether we like it or not, so AI as a thing will remain in some form, it's just that these efforts where people train on anyone else's content is where things fall apart and thus it will not look like how it looks today, where you have a company or a community effort that provides a NN trained on other people's data. Instead, the data generators and the AI trainers will have to be the same person or entity (company). If I for example trained an LLM on the code I've generated since I was 16 and made a nooberminGPT, it would be fine because I won't sue myself.
The only other way things could generally work is like how artists sign their copyrights to be owned by a company or publisher, basically people or entities will have to sell their content to be trained instead of it being trained on for free...And then we'll see old fashioned exploitation, basically people being suckered into contracts where they sign their life away, but it will be on paper as opposed to now where it's the wild west and you can take anyone's content without asking.
> If I for example trained an LLM on the code I've generated since I was 16 and made a nooberminGPT, it would be fine because I won't sue myself.
That's like saying you don't need IRB approval to inject yourself with your experimental research chemicals. That's true, but it's impracticable. It's as rare as monkeys taking pictures of themselves, which did produce an interesting but objectively bullshit copyright registration decision. It might as well be random at that scale.
nooberminGPT is impracticable because GPTs won't generalize on your tiny dataset, if that's all they're trained on. Making your own math is impracticable, your own alternative to transformers, because you need commercial-entity-scale resources to run that research program. Nobody could have proven that transformers work as well as they do without all the unpermissioned data, and we're talking about it only because it was proven to work. I mean transformers were just as powerful when they were first conceived as Python code, in an objective sense, but we're talking about it because of the data. And if you allow that something changed between that stuff entering the researcher's consciousness and actually reacting to the results of experiments on unpermissioned data, which is definitely true, transformers and diffusers are irreparably tainted.
> where it's the wild west and you can take anyone's content without asking.
Like long term, if you put something on the Internet, you are implicitly saying that people can do whatever the fuck they want with it. Video games, as an industry and creative pursuit, have evolved, in a Darwinian way, within that objective reality for a long time. And consequently the multiplayer backend servers are highly protected, the clients are free, there's a thriving indie scene of easily pirated games, there's a thriving first party market of expensive prestige single player titles, and the video game industry is huge and makes tons of money, and it has no more (or fewer) exploited creatives toiling in it than any other creative industry, except that they are compared to novelists, actors and directors, always toiling away in obscurity. All of this works.
That Hollywood and book publishers are struggling to innovate monetization at all isn't really the copyright office's problem; that these models can generate images and text, which directly threatens Hollywood and doesn't really move the needle on video games, is also not really the copyright office's problem. The AI stuff is a red herring for deeper problems.
I mean this is just going to promote "bad actors" that are not going to disclosured the use of AI with actual pieces that looks like created by a human.
Did you read the EULA of the LLM you're using closely enough?
I know that sounds shifty and underhanded, but the providers of these services have every right to slide a little nugget into their EULA that shares the rights to your prompts with them. Most even explicitly say they'll use prompts and other data to "improve the offering".
copyright exists in the US under this clause in the constitution:
> “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.”
not sure how letting people copyright the output of a program that ate the work of a million other people would "promote the progress of science and useful arts"? or how there's an author to reward? or how there's writings or discoveries?
of course this also applies to the US endlessly extending copyright also - nothing Congress can do, no matter how well intentioned, is going to encourage Walt Disney himself to produce anything new of value, nor did keeping Happy Birthday copyrighted until 70 years after the last Hill sister died assist any of them in coming back to life to come up with a second hit.
the dumb optimistic part of my dreams that having matrix multiplications shitting out words and pictures will cause a reassessment of why we let IP laws get so fucked up to begin with, but of course, it won't - it'll just mean Disney and Warner et al will find some way to keep selling tv shows and movies and the humans who produced the art who get shoved into the matrices get shafted.