Hacker News Reader

Comment by ar-nelson | Hacker News Reader

I've seen a lot of confidence on HN and other tech communities that a court would never rule that training an AI on copyrighted images is infringement, but I'm not so sure. To be clear, I hope that training AI on copyrighted images remains legal, because it would cripple the field of AI text and image generation if it wasn't!

But think about these similar hypotheticals:

1. I take a copyrighted Getty stock image (that I don't own, maybe even watermarked), blur it with a strong Gaussian blur filter until it's unrecognizable, and use it as the background of an otherwise original digital painting.

2. I take a small GPL project on GitHub, manually translate it from C to Python (so that the resulting code does not contain a single line of code identical to the original), then redistribute the translated project under a GPL-incompatible license without acknowledging the original.

Are these infringements?

In both of these cases, a copyrighted original work is transformed and incorporated into a different work in such a way that the original could not be reconstructed. But, intuitively, both cases feel like infringement. I don't know how a court would rule, but there's at least some chance these would be infringements, and they're conceptually not too different from distilling an image into an AI model and generating something new based on it.

View on HN

It’s not about reconstruction, it’s about the notion of a “derivative work”. Translating a work would absolutely be derivative (consider the case of translating a literary work between languages: this is a classic example of a derivative work). Blurring a work but incorporating it would nonetheless still be derivative, I think.

The challenge with these models is that they’ve clearly been trained on (exposed to) copyrighted material, and can also demonstrably reproduce elements of copyrighted works on demand. If they were humans, a court could deem the outputs copyright infringement, perhaps invoking the subconscious copying doctrine (https://www.americanbar.org/groups/intellectual_property_law...). Similarly, if a person uses a model to generate an infringing work, I suspect that person could be held liable for copyright infringement. Intention to infringe is not necessary in order to prove copyright infringement.

The harder question is whether the models themselves constitute copyright infringement. Maybe there’s a Google Books-esque defense here? Hard to tell if it would work.

> If they were humans, a court could deem the outputs copyright infringement

I'm not sure I understand how this is self-evident. The closest equivalent I can see would be a human who looks at many pieces of art to understand:

- What is art and what is just scribbles or splatter?

- What is good and what isn't?

- What different styles are possible?

Then the human goes and creates their own piece.

It turns out, the legal solution is to evaluate each piece individually rather than the process. And, within that, the court has settled on "if it looks like a duck and it quacks like a duck..." which is where the subconscious copying presumably comes in.

I don't know where courts will go. The new challenge is AI can generate "potentially infringing" work at a much higher rate than humans, but that's really about it. I'd be surprised if it gets treated materially different than human-created works.

> The new challenge is AI can generate "potentially infringing" work at a much higher rate than humans, but that's really about it

The other challenges are: (i) the model isn't a human that can defend themself by explaining their creative process, it's a literal mathematical transformation of the inputs including the copyrighted work. (And I'm not sure "actually the human brain is just computation" defences offered by lawyers are ever likely to prevail in court, because if they do that opens much bigger cans of worms in virtually every legal field...) (ii) the representatives of OpenAI Inc who do have to explain themselves are going to have to talk about their approach to licenses for use of the material (which in this case appears to have been to disregard them altogether). That could be a serious issue for them even if the court agrees with the general principle that diffusion models or GANs are not plagiarism.

And possibly also (iii) the AI has ridiculous failure modes like implementing the Getty watermark which makes the model look far more closely derived from its source data than it actually is

It's worth pointing out that the problem, in this scenario, is for the creator (ie, the human running the algorithm). They will need to determine whether a piece might violate copyright before using it or selling it. That seems like a very hard problem, and could be the justification for more [new] blanket rules on the AI process.

Proving artwork you created is free from all copyright issues is similarly impossible, but in practice isn’t an issue. So, I don’t see any AI specific justification being relevant.

How common is it for an artist to accidentally generate a work that resembles an existing work?

I can't speak for digital art because I'm not an artist but I can say it's extremely common for original music to (often accidentally/subconsciously) include melodies or pieces of melodies from other music.

That's more like two paintings sharing the same color scheme or using the same brand+color of paint but still being unique. The performance of those melodies and structures is what makes a song unique and creative.

Adam Neely the YouTuber has produced several videos about recent legal cases where musicians sue because of some superficial similarities. In most cases the higher courts in the US recognize that copying is part of the normal creative process and is not infringement unless it is a blatant rip-off.

This is exactly my thinking. If the court finds somebody guilty of infringing on a human-made piece of digital art the response is to punish the human, not to ban or impose limits on photoshop.

At risk of stretching the analogy, you don’t charge the gun with murder…

Except, of course, the human has very little control over what the AI outputs in TXT2TXT scenarios, at least in terms of whether the output would match the definition of copyright infringement of someone else's work. IMG2TXT is kinda different -- I think you could make a much stronger case for derivative work there. So you have a tool that can randomly create massive liability for you, and you can't know if its done so or not until someone sues you.

The humans are the cause of it happening in the first place. Maybe the gun analogy is not such a stretch: if you pull the trigger, you own the consequences.

Reasonable fair use principles could distinguish personal and R&D use from commercial use.

I think it's more common with music. Some musician goes to a foreign country and heard an obscure local song. 20 years later the musician has forgotten completely about the song and the trip. One day a catchy melody appears in the head of the musician out of the blue, and the musician complete the song and add a lyric. The song get famous, and later reach the foreign country, and everyone acuse the musician of plagiarism.

Reminds me of a recent scandal involving Adele, where she is accused of plagiarizing a Brazilian composer: https://english.elpais.com/usa/2021-10-19/toninho-geraes-vs-...

>I'm not sure I understand how this is self-evident. The closest equivalent I can see would be a human who looks at many pieces of art

...and then gets told "Hey, go and paint me a copy of that Andy Warhol piece from memory".

The model might not violate the copyright, but its output is derivative work if the copyrighted works are included in the training set.

That would be over-fitting which is certainly a failure mode of ML. But it's still a failure mode, not an inherent property.

Note that a derivative work doesn't instantly make it 'fair use' for the purposes of copyright. You typically still need 'adaptation' permission from the copyright holder to made a derivative work of it, so you can't make 'Breaking Bad: The Musical' by recreating major scenes in a play format, at least not without substantially changing it[0].

For the purpose of fair use, copyright.gov has an informative section titled "About Fair Use" which details what sort of modifications and usage of a copyrighted work would be legal without any permission from the copyright holder https://www.copyright.gov/fair-use/#:~:text=a)(3).-,About%20...

0: https://en.wikipedia.org/wiki/Say_My_Name!_(Musical)

It’s the opposite of what you’re suggesting in the first sentence. Something being a derivative work is a fairly clear sign that it’s not fair use.

It's kind of a combination of these two positions: something being a derivative work means that it requires an excuse under copyright law, whether that be a license, or a fair use defense (which may be fact-specific), or some other justification. Without such a justification, the derivative work is treated as a copyright infringement.

This is a result of 17 USC §106(2), which says that one of the things that "the owner of a copyright [...] has the exclusive rights to do and to authorize" is "to prepare derivative works based upon the copyrighted work", unless an exception (including fair use) applies.

https://www.law.cornell.edu/uscode/text/17/106

> If they were humans, a court could deem the outputs copyright infringement

Why? I am a human, I can learn the composition of stock photos and use the idea in my work. There is no copyright infringement, or there is simply no way to prove it.

The connection needs to be really strong to consider something as derivative work.

> consider the case of translating a literary work between languages: this is a classic example of a derivative work

This is a classic example of "you need a license on the original material to sell it" (unless it is public domain)

You need the rights also if you are translating from a pre existing translation in a different language (i.e. you're translating a Japanese book from the French authorized translation).

Translating an opera is not automatically derivative work and does not fall under the umbrella of fair use.

source: my sister is a professional translator

EDIT: technically you'd need to acquire the rights even if you translate it by yourself, for fun, and show it to someone else.

> The challenge with these models is that they’ve clearly been trained on (exposed to) copyrighted material, and can also demonstrably reproduce elements of copyrighted works on demand. If they were humans, a court could deem the outputs copyright infringement, perhaps invoking the subconscious copying doctrine (https://www.americanbar.org/groups/intellectual_property_law...).

Every single human has been exposed to copyrighted material, and probably can reproduce fragments of copyrighted material on demand. Nobody ever writes a book or paints a picture without reading a lot of books and looking at a lot of paintings first. For a "subconscious copying" suit to apply, you need to demonstrate "probative similarity" - that is, similarity to copyrighted material that is unlikely to be coincidental.

In other words - it's not clear to me that the situation with AI is any different than with a human, or that it presents new legal challenges. If it looks new, it is new.

> Blurring a work but incorporating it would nonetheless still be derivative, I think.

you have to show that the derivative work is a substantial part of the new work.

If your background is a small portion of the new image, and the blue makes it difficult to see that it was the original, and that any other blurred image would've done the same job, then i would argue that the final new painting does not constitute a derivative work.

A similar argument could be made for AI models. The model consists of billions of training images. None of these images are individually substantial in the final output, despite that on some part of the output, you can trace a derivative work. For example, if an author used 1 million books, and copied every nth word from each book and merged it, to produce a final book (which happens to produce a coherent book), i would argue that the author did not infringe copyright on any of the original 1million source books.

>> Translating a work would absolutely be derivative (consider the case of translating a literary work between languages: this is a classic example of a derivative work)

True for literature, which can be copyrighted. Not true for logic. Code that performs the same function but is written independently can't, by definition, violate copyright. It could violate a patent, but it's much harder to patent code.

However, in the AI case, the models themselves do nothing even approaching such translation. They fully incorporate what are essentially just compressed representations of existing works into their weights.

I'd guess if any single copyright holder could show that their work could be substantially retrieved from the model with the right query string, they'd have a reasonable claim on royalties from the entire model, or else the whole model would have to be thrown out.

Yes, I expect that if you ask the model for "Getty images photo of [famous person] doing [thing Getty Images has only one photo of that person doing]" you might well get the original photo out.

Should be easy to try. My guess is that it most likely won't.

You won't get the exact image, because the model doesn't perfectly memorize all inputs, but you'll likely get something so close that everyone would consider it a derivative work.

This is very similar to getting Copilot to spit out Carnack's fast inverse square root code with the right prompt.

There's some interesting thought experiments around this:

1. You do the same thing but don't use "getty" in the prompt.

2. You do the same thing with "getty" in the prompt on a model NOT trained with Getty images

3. You ask a human photographer to do the same thing and you show him a Getty image

4. You ask a human photographer to do the same thing but without showing him a Getty image (but he's presumably seen many in the past).

If 1 and 2 produce images that are also similar to a Getty image, where do we stand? I imagine it's likely that a trained model can learn "getty-like" without any actual Getty images to make 2 happen.

And currently IP law treats humans and AIs the same (in the sense that infringment rules don't distinguish between them) so would you consider 3 and 4 to be similar to 1 and 2?

For 1, there's probably a case where a picture that exactly matches the prompt is very famous, so maybe the right prompt would manage to effectively select an image licensed by Getty. The GPLed fast inverse square root routine was an example of this kind of thing: one good match and the model finds it. Finding such a case might or might not be possible.

For 2, you can't find what isn't there, so something random would come out.

For 3 or 4, I suppose you could pay paparazzi to stalk the celebrity and try to produce a similar original shot, and this might be impossible depending on the prompt (for example, photo of [person] at [event] on [date]"), but if it's possible it has absolutely no resemblance to 1 or 2. The photographer would produce an original work, unless your #3 contractor tries to remove the watermark and pass off the Getty image as their own, which would be idiotic.

There is no particular "Getty Images" style, other than their quality requirements, they are a huge company that acquires and licenses a ton of pro photography. There's no such thing as "getty-like".

So, no, only option 1 might possibly produce a problematic derivative work.

> For 2, you can't find what isn't there, so something random would come out.

It's not a search engine. It can synthesise things that it hasn't seen to some degree (assuming it knows the elements that make up the request.

The tricky bit would to teach it "Gettyness" without showing it Getty images but I think that's entirely possible. Getty images aren't astonishing examples of unprecedented originality. So it just needs to know a) the celebrity b) the action and c) what people mean when they ask for something that looks like a Getty image.

EDIT - I answered in a rush and realise you made a similar point to me about "Gettyness" - which makes it even harder for me to understand why you think it would be a violation.

How many different ways can George Clooney eat a burrito in Times Square?

Suppose that there's a really iconic pic of George Clooney eating a burrito in Times Square, with a "Getty Images" watermark on it, in the training set. It's perfectly framed, has a somewhat comic expression, and a bit of the contents spilled on his shirt. It seems quite possible that the model could produce an image that isn't identical to this but so much so that people assume it's a copy, maybe produced by an artist based on the picture. Worse, it will have most of the "Getty Images" watermark right on the image.

If that doesn't convince you that there might be an issue I'll just stop here.

Both hypotheticals are likely infringement. The first example may be considered de minimus, but the courts hate using those words, so they might just argue that you didn't blur it enough to be unrecognizable or that it could be unblurred.

However, the thing that makes AI training different is that:

1. In the US, it was ruled that scraping an entire corpus of books for the purpose of providing a search index of them is fair use (see Authors Guild v. Google). The logic in that suit would be quite similar to a defense of ML training.

2. In the EU, ML training on copyrighted material is explicitly legal as per the latest EU copyright directive.

Note that neither of these apply to the use of works generated by an AI. If I get GitHub Copilot to regurgitate GPL code, I haven't magically laundered copyrighted source code. I've just copied the GPL code - I had access to it through the AI and the thing I put out is substantially similar to the original. This is likely the reason why Getty Images is worried about AI-generated art, because we don't have adequate controls against training data regurgitation and people might be using it as a way to (insufficiently) launder copyright.

Searching books is different than generating books and selling them.

> To be clear, I hope that training AI on copyrighted images remains legal, because it would cripple the field of AI text and image generation if it wasn't!

To be clear, there's no law banning training an AI. There are laws for what you can do with other people's stuff.

In short, maybe the AI field would indeed be crippled if they no longer freely take input from others without asking permission and/or offering compensation. And maybe that's far, far from a bad thing.

That's true, but AI models trained on copyrighted images already exist and can't just be removed from the internet, and their output will often be indistinguishable from that of "clean" models. What I fear is a kind of legal hazard that would make even the possibility that AI had been used anywhere in a work radioactive.

Imagine another hypothetical: I create a derivative work by running img2img on another artist's painting without their permission. Whether the AI model in question contains copyrighted content or not, this is probably infringement.

Now suppose that, instead, I create an original work, without using img2img on someone else's art. But, as part of my process, I use AI inpainting, with a clean AI model, so that the work has telltale signs of AI generation in it.

And then suppose an artist I've never heard of notices that my painting is superficially similar to theirs--not enough to be infringement on its own, even with a subconscious infringement argument. But they sue me, claiming that my image was an img2img AI-generated derivative of theirs, and the AI artifacts in the image are proof.

With enough scaremongering about AI infringement, it might be possible for a plaintiff to win a frivolous lawsuit like this. After all, courts are unlikely to understand the technology well enough to make fine distinctions, and there's no way for me to prove the provenance of my image! If it becomes common knowledge that AI models can easily launder copyrighted images, and assumed that this is the primary reason people use AI, then the existence of any AI artifacts in a work could become grounds for a copyright lawsuit.

It’s completely ridiculous to believe that copyright claims are unenforceable because you ran it through an ML transformation engine.

I hope Getty sues and wins. Train your datasets on your own data! This is mass IP theft.

Be careful not to watch any copyrighted films through your kranke engine. The copyright owner could ask for all copies to be removed from wet storage, fair damages. I wouldn’t wish that on ainybody.

https://en.m.wikipedia.org/wiki/Monkey_selfie_copyright_disp... Slater claims the copyright (the monkey and the Canon EOS 5D DSLR definitely don’t have the copyright).

Nonsense. Observations about certain characteristics of a copyrighted work are not covered under that work's copyright. If I take a copyrighted book and produce a table of word frequencies in that book, no serious person would claim that the author's copyright domain extends to my table.

Everyone in HN keeps pretending like ML transformation = human inspiration. This is really funny - we don’t have AGI but we have an AGI-like capability to avoid copyright. Seems to be the only place where human rights and AI rights are matched is where it most benefits AI research. How interesting.

A for profit computer program =! A human being.

Can you articulate a meaningful (and more importantly legally provable) difference? Both human brains and these sorts of AI programs are intractable black boxes. Maybe you think computer programs lack some sort of divine spark, but even if we accept that it seems to me that it's not a given that humans apply their divine spark every time they create something either.

That’s a bizarre definition.

So because they’re both black boxes we suddenly treat them both the same legally?

I don’t have to prove that an ML transform is equivalent to a human being. Divine spark isn’t necessary to protect you from copyright - we have laws that determine what you need to do and those laws have allowed IP to exist as a profitable area for a century.

Here’s a simple question - if I were to take an image from an artist on artstation and announce my for money Warhammer tournament with it, without paying him, I’d be violating his IP rights.

But if I build an ML engine I apparently can take 100 of his images, produce similar images and charge money for it. Magic!

Explain to me - how did the artist suddenly lose his IP rights exactly ?

I submit it is up to ML researchers to prove this isn’t the biggest IP land grab in history. Not for me to prove that somehow humans are different. We know humans are different and we’ve codified in law just how much new IP needs to be different in order to avoid copyright. That’s good enough for humans.

But somehow ML enthusiasts want to say that if I feed an ML all the Disney movies and I get it to make it derivative work, it’s somehow protected? I can’t wait to see all the Mickey Mouse movies made by AÍ. Guess what - that’s never going to happen. If you think Disney or any other IP empire will let that go, I don’t know what to tell you. It’s absolute madness to say that anything that goes into an ML transformation is uncopyrightable. Or somehow I have no rights over my artwork because I don’t have the ability to sue OpenAI to oblivion like Disney does? Because that’s literally what we’re saying - those artists from artstation and Getty whose images were used were used exactly because they believed that they wouldn’t be able to sue - thanks to distinctions like your own!

Disney won’t let that stand, they would sue and push for annihilation, but hey the 100,000 artists on artstation? F** them, they can’t sue us. Let’s use their shit.

That’s basically where we’re at.

Let’s work out an example. So this AI can be feed all of Lucien Freud’s works. Then it can produce Lucien Freud-like artwork. And in this process, where all that is missing is Lucien Freud’s own signature, it could potentially make a virtual replica of one of his most famous paintings! Thus I would have a copy of Lucien Freud! But completely copyright free.

Are you for real? This cannot, absolutely cannot be allowed to happen. It is the biggest data theft in history, larger than Facebook or anything, to basically say that any human data when run through an ML transformation engine is no longer property of the human being who produced it but if the ML engineer. This is absolute madness if you consider the second order and third order effects.

Essentially all human data that can be automated through ML would be automated to the gain of only the ML engineers involved. The intellectual property of millions of human beings producing the data would be nothing but compost. This will lead to an unsustainable situation, where no one but ML engineers will be able to make any profit in the world. Human beings must be compensated for their data, or we’re headed straight into a dystopia.

>But if I build an ML engine I apparently can take 100 of his images, produce similar images and charge money for it. Magic!

You can pay a human artist to produce "similar" images too, and as long as they aren't too similar it's fine.

You seem to be conflating - repeatedly - exact copies of a work with merely copying a style. Family Guy had an episode where Brian and Stewie visit a Disney universe, and everything was animated in the style of a classic Disney film. A bunch of humans did indeed watch a bunch of Disney films, and make a new animation based on what they saw. Did Disney - a notoriously litigious company - sue? Of course not. Copying a style is not infringement! I don't see why it's any different if an AI program does that.

> A bunch of humans did indeed watch a bunch of Disney films, and make a new animation based on what they saw. Did Disney - a notoriously litigious company - sue? Of course not.

Why would Disney sue themselves? Disney owns family guy.

In a world where Disney did not own it, Disney would probably sue if they used a cartoon mouse. "Style" may not be covered by copyright, but characters likely are. A human animator would know enough not to make a parody cartoon mouse look too Micky-like, but an AI wouldn't know to avoid that.

Parody is a well known copyright exemption.

I just don’t see how you can go - this is just like a human being, let me make a million dollars out of it.

It wouldn't extend to your table, but if you used that table to generate a book of your own, it's very possible a serious person would claim the author's copyright extends to the generated book.

there is only one way frequency tables would return original work, all other versions would be gibberish, its like when using Huffman tables for compression where table itself is not equal to original data

Is the copyright still not applicable if your encoding table has rules to reconstruct text from? No serious person would argue that copyright doesn't extend to the encoded version of the book and prevent me from profiting off it.

I believe the same applies to AI generation of text/images. Just because you're encoding the data in statistical models doesn't mean that it's not encoded.

> I've seen a lot of confidence on HN and other tech communities that a court would never rule that training an AI on copyrighted images is infringement, but I'm not so sure. To be clear, I hope that training AI on copyrighted images remains legal, because it would cripple the field of AI text and image generation if it wasn't!

Regardless of the copyright of the training data which really is unresolved, the copyright-ability of output of AI is questionable at best. There's no way to monetize the generated images for a stock art that isn't at risk of a court ruling pulling the rug from under it.

AI-produced art is still human-made, as a person does the job of engineering a prompt and selecting from the generated images. The copyrightability of such work is unlikely to ever seriously be in question.

This is not as obvious as you may think. This is closely related to the "monkey selfies" copyright claim issues. The court didn't seem to agree with the photographer who said "I own the copyright of those pictures because I configured the camera by myself and put it there, the monkey only pushed the button."

The court didn't rule on the monkey selfies copyright claim. The copyright owner just ran out of money to pay for a lawyer and gave up fighting against Wikimedia.

That's not so obviously clear cut. Can the model produce identical output from the same simple prompt?

But that's exactly what happens, AI isn't randomness, it's a set of predefined calculations. The randomness is in the seed/starting point. For e.g Stable Diffusion it is given/user input, resulting in perfect reproducibility.

I should have been clearer - that was rhetorical to point out that if you and I use the same prompt and pick the same resultant image, it's harder to claim either of us have copyright. This is a gray area. The tools themselves could introduce some stochastic aspect so that outputs are never identical, also.

None of this leads to an obviously clear cut legal position wrt copyright.

Your (and the other person's) set pf inputs drive a mathematical function that derives the same output.

If we were all honest, we'd give up the idea of copyright altogether where ML is concerned. You can't get much closer to "It's just math, man" than what it is currently.

I think we are saying the same thing, roughly.

Of course, "it's just math, man" isn't precisely a legal argument, either.

Vector art is just math, yet you can get copyright on it.

It might be hard to argue that a prompt is original enough to be covered by copyright. Shorter, simpler prompts might be ruled unoriginal.

This is what I would like to believe… but needs to be proven before I stake my business on it.

> because it would cripple the field of AI text and image generation

I don't disagree with this statement. But arguably, it's becoming clear that these fields exist, on an economic level, as a means for already powerful corporations and technocrats to gain ownership over and repurpose the labor of previous generations for profitable automation (not simply to create C-3PO or something).

Not unlike all those other non-digital areas of the economy (e.g. railroads were built on the blood, sweat, and tears of previous generations and they are now owned by a small few, likely unrelated to the descendants of the laborers who built them).

For some ML-related commentary on this subject, see e.g. https://nathanieltravis.com/2022/08/01/ai-research-the-corpo...

I've seen a lot of confidence on HN and other tech communities that a court would never rule that training an AI on copyrighted images is infringement, but I'm not so sure.

Indeed and your examples are intended to point to gray areas. But a much more problematic (for the user) example is: some Dall-E-like program spits out seemingly original images but 0.1% are visibly near duplicates of copyrighted images and these form the basis of a lawsuit that costs someone a lot of money. Copyright in general tends to use the concept provenance - knowing the sequence of authors and processes that went into the creation of the object [1] and naturally AI makes this impossible.

The AI training sequence either muddies the waters hopeless or creates a situation where the trainer is liable to everyone who created the data. And I don't think the question will answered just once. The thing to consider is that anyone can just create an "AI" that just spits out stock images (which is obviously a copyright violation) and so the court would have to look at detailed involved and neither the court nor the AI creator would want that at all.

ianal... [1] https://serc.carleton.edu/serc/cms/prov_reuse.html

> To be clear, I hope that training AI on copyrighted images remains legal, because it would cripple the field of AI text and image generation if it wasn't!

Temporarily, yes. However it wouldn’t be the worst thing if they were forced to work on sample-efficiency. And I maintain that if they want a large dataset of images they can collaborate with Twitter and Instagram to add an image license option to the image uploader, and provide a license option that explicitly allows this kind of use.

AI needs sample efficiency research and if they actually had to get consent for their images there are a lot of artists who wouldn’t feel like their work was being ripped off.

It’s probably better if this kind of use is broadly permitted, but I don’t think it’s the disaster some think it would be if they were forced to get consent. There’s already millions of openly licensed images in several collections online (Creative Commons, Wikimedia Commons) and if they forced people to provide licenses on sites like twitter we could actually have open datasets for this instead of these kind of grey areas of privately scraped datasets.

There's a literal arms race behind the scenes in AI right now.

I think it's very unlikely that corporate IP claims that could substantially hold back progress in domestic AI development will end up being successful.

I don't know if there's a clear answer to (1), but with respect to (2) I believe there is precedent that the copyright owner would have a strong case if they had reason to believe the python library author had seen their work and it played a roll in the transcription to python. You can't look at a GPL work and literally transcribe it to license launder. Companies have tried. You can implement a similar idea in a clean room, though.

Honestly, as much as I am rooting for AI "art" (still not sure about that term here) I can see how Getty would easily have a claim in court if the AI was indeed trained on some of their images AND they can prove it somehow. If that's not derivative then I don't know what is. Maybe a special niche could be carved out for people who are only researching and experimenting and not really "selling" or profiting from the resulting images. It would seem if they're right that maybe they could bury a watermark in their images that identifies it as Getty (or just whomever) and that it's copyrighted by them and they don't give permission to use it for training AI. Maybe I just don't know enough about how the algorithms work though shrug

Unfortunately for the Getty et al, "feels like" is worth exactly $0.

Any court that understands the technology at even a lay level, has no path to find infringement applying existing precedent.

NB "style" is not protected.

I do understand the technology at at least a lay level (unless your Scotsman is true by your conclusion), and it seems to me that the idea that it's "style" and not "permuting the input data" is one that seems to be a postulate, not a fact.

For case 2, translating a novel from, say, English to Japanese, still requires permission of the holder of the copyright to the English version, even though the resulting novel "does not contain a single line ... identical to the original".

It might technically be infringement, but the proof is in the pudding. It may be very hard to prove a specific image (or set of millions of images) were used in training.

Reply on news.ycombinator.com