UK judge delivers a 'damp squib' in Getty AI training case, no clear precedent set

London's High Court has dismissed the major portions of Getty Images' lawsuit against generative AI firm Stability AI for training its image-generation model on copyrighted images, which some legal experts say could weaken intellectual property laws. However, others saw daylight for trademark and copyright protection in the …

[steelpillow]

In the news: Dog bites Man! UK Judge makes sense! (Only kidding, m'Lud)

> by the end of that process the Model itself does not store any of those Copyright Works.

Right. It just reproduces their subject matter and/or style, which is perfectly legal. Like, when I was at college I drew cartoons in the style of the Peanuts characters we all knew well. Had I traced them off a published strip, I could have been sued; were an AI to copy pixel for pixel, it too would be in trouble.

And don't reproduce the feckin' trademark! So I took care to sign myself Schplatz.

Some sense at last emerging from the AI shitfest.

Now, how the hell do you get an AI to understand what a trademark is and recognise one when it sees it? IMHO this problem might just focus minds rather usefully.

[Richard 12]

Smith's ruling says that transcoding any Hollywood blockbuster movie is not infringement, because at the end of the process the bytes on disk and pixels on screen are different.

Glad that's cleared up, I'm off to The Pirate Bay.

[Yet Another Anonymous coward]

But that's the same ruling that claims you have "created" whatever illegal content by transferring the bytes to screen

[Anonymous Coward]

Yeah, her ruling contradicts itself by stating "generative models [...] don't store actual copies of the content they're trained on" and also "responsibility for the presence of [water]marks lies with the model provider". If a watermark is reproduced verbatim in the model's output, then obviously the model has stored an actual copy of the content it was trained on.

Newman (of Addleshaw Goddard) is also half-right and half-wrong imho. Wrong for claiming that "extracting value from protected works is an act reserved to the copyright owner" which would prohibit all forms of education from copyrighted books for example. But entirely right in pointing the major problematic issue that "In practice, models trained on infringing data outside of the UK can be imported into the UK without legal repercussions", like cocaine(?), and Shein's child-like sex dolls currently rocking France(?).

Overall then, I'd say Smith was 1/3 right and 2/3 wrong on this judgment.

[Bebu sa Ware]

If a watermark is reproduced verbatim ...

If a watermark is reproduced verbatim in the model's output, then obviously the model has stored an actual copy of the content it was trained on.

I am not sure this inference is entirely sound.

I might say that the model has implicitly encoded an image that can later produce an explicit image that at some resolution appears identical to the watermark at the same resolution but at higher resolutions diverge. The question then might be at what level of fidelity does the copyright breach occur — would a child's crayon rendering be sufficient ?

Is there some Platonic "ideal" of the watermark against which copies might be compared ?

If implicit versus explicit copies are treated equally then we might consider a box of oil paints as containing an implicit copy of The Mona Lisa ? Curiously I recall some jurisdictions placing a levy on blank audiocassette tapes on the basis that were certain to eventually contain copyright music. ;)

Personally I would be overjoyed if the whole AI crew were taken behind barn... but I am more concerned that the reaction to AI/LLM training will restrict fair use and other acceptable use by natural persons, of copyright material.

[Anonymous Coward]

Re: If a watermark is reproduced verbatim ...

Yeah, I haven't seen how well the actual watermarks were reproduced in this particular case but based my perspective on priors such as NYT where "The Times elicited examples of memorization by prompting GPT-4 with the first few words or sentences of Times articles". The lawsuit's exhibit J (linked under "PDF" here) has great examples of verbatim reproduction (starting on page 238). Granted this is text-based though ...

Overall, it seems to me the issue of "what the meaning of the word 'is' is" may have already left the barn on this one ... Model makers have to find ways to ensure their outputs don't plagiarize existing works (maybe source citations could help?) beyond fair use, and/or develop strategies to compensate rights holders, imho.

[Eric 9001]

As far as I can tell, it has never been illegal to download a movie via bittorrent in the UK provided the client is configured to be a leech that doesn't seed - after all, you wouldn't be distributing the movie (although an extortive letter demanding x dollars, or that you'll be sued may still be received).

But I guess there's now some new precedent that you can share generally useful public information with your neighbors by seeding the movie, as long as the movie is re-encoded (as the bytes on disk and pixels are indeed different after after re-encoding H.264 to H.265 or better VP9).

[VoiceOfTruth]

Hmmm

I wonder how this plays out with copying software.

>> But Getty couldn't prove that any of that training had taken place in the UK, forcing it to drop its more general claim of copyright infringement

So if Joe in the UK so-called pirates some software, and is met with a claim of copyright infringement, if the claimant can't prove it was actually copied in the UK, what then?

[Yet Another Anonymous coward]

Re: Hmmm

I made a new AI model, it's "trained" on my movie collection and has a very simple "inference" step which uses the builtin "cp" command

As a bonus it uses much less resources than traditional AI models

[beast666]

Re: Hmmm

"As a bonus it uses much less resources than traditional AI models"

The original 'cp' command might but the new Rust version probably uses more...

[Eric 9001]

Re: Hmmm

If Joe in the UK uses a boat as part of an operation where he steals a storage medium with software on it (unless the act includes a boat, it is not piracy), generally there would no grounds for a claim of copyright infringement (as he didn't copy the storage medium - he stole it) - but there would be grounds for a claim of theft of the storage medium and of piracy - which has severe charges available on the books.

As far as I can tell, it has always been the case that if Joe in the UK downloads an unauthorized copy some software without distributing it, there aren't any grounds for a claim of copyright infringement, but a company will still be able to send a threatening letter demanding thousands of dollars.

[ecofeco]

Stored or not stored?

That has nothing to do with anything.

It was obviously used. By a for profit company.

Goddamn we live in idiot world.

[Anonymous Coward]

Re: Stored or not stored?

Does UK copyright prohibit use? Or copying? By a for-profit company?

[doublelayer]

Re: Stored or not stored?

Without permission, yes it does. Try pirating some software from a litigious company and find out. There are terms on copyrighted works which apply to your use or copying of it. If you're a for-profit company, non-profit company, or individual, you're equally culpable and can be punished for doing so. Evidently, if you are an AI company, laws don't apply to you. I wonder if any other ones do? I think I need to found an AI company.

[Long John Silver]

Re: Stored or not stored?

Disobedience to anachronistic (pre-digital era) copyright has reached a scale which begins to render copyright enforcement a fools' errand.

[doublelayer]

Re: Stored or not stored?

Even if your claims were correct, which they aren't, that crimes aren't being punished does not make any case about whether they should be crimes. You can try to make a case that the law should be changed*, but that it's easy to get away with isn't any argument for that point of view. Otherwise, I could say that, because I could probably get away with robbing you, it's totally justified for me to do so.

* I wouldn't suggest you bother though. I've seen your case. It's a very wordy version of "But I want it".

[Eric 9001]

Re: Stored or not stored?

It is an error to equate unauthorized copies to theft - as making a copy of digital data doesn't change the original by one bit, while robbing someone takes something physical from them.

It shouldn't be a crime to share generally useful public information with your neighbors just to be nice.

In the case of proprietary programs, maybe it wouldn't be a bad thing to make distributing them a crime - as the distribution of proprietary software does harm those who receive the software.

[doublelayer]

Re: Stored or not stored?

"It is an error to equate unauthorized copies to theft - as making a copy of digital data doesn't change the original by one bit, while robbing someone takes something physical from them."

1. I didn't. The robbery mentioned in the post referred to physical robbery and was an analogy pointing out the fact that the ease with which a crime can be committed is no argument at all for whether it should continue to be a crime.

2. We have lots of things equated with theft which work in exactly the same form. For an example, wage theft is a crime where I do work for someone who agreed to pay me but criminally withholds owed money. On that basis, they did not take anything I had and keep it away from me. Yet it is still theft because I am owed money and they are deliberately not giving it to me. As law stands, someone owning the copyright has the right to place terms on the distribution and use of their work, and you commit a crime if you violate those unless you and your use comply with explicitly written exceptions. Most such crimes do take the form of refusing to pay them money which they have the right to require if you obtain the covered work. Others, for example violating an open source license with specific terms, are less financial but are still criminal and can result in financial penalties.

As long as you continue to use flawed logic like this to either claim that these aren't crimes or to distract from the fact that they are crimes by quibbling with language you understand as well as anyone else, you are and will be wrong. You can argue that they should not be crimes and thus that we should consider changing laws. We would likely agree on some of that. For example, the specific exceptions where you can use copyrighted work for free without permission, I don't think there are enough of those. I also think copyrights last too long. You have not made any such argument, however.

[Eric 9001]

Re: Stored or not stored?

Always relying on the law to determine right or wrong is trite logic that fails hard - as there's a lot of laws that are immoral - but magically those laws are moral because the law says so.

>For an example, wage theft is a crime where I do work for someone who agreed to pay me but criminally withholds owed money.

In that particular case, the work has been done and money has been earned, but the contractually agreed percentage isn't given to the person who assisted with making that money.

Such act would be immoral whether it is criminal or not.

>As law stands, someone owning the copyright has the right to place terms on the distribution and use of their work

If you actually read the copyright laws - those state that copyright is of a limited time and eventually expires - while something that is owned never expires (too bad both the term lengths and the legal texts have been corrupted immorally).

It's often the case that the copyright holder is not the author of the work and is it moral that some business who didn't author the work has the ability to restrict the work (especially so against the author(s) wants)?

>you commit a crime if you violate those unless you and your use comply with explicitly written exceptions.

If you actually read the UK law; https://www.legislation.gov.uk/ukpga/1988/48/section/107 receiving unauthorized copies for private use is not an offense.

Distribution is only considered a offense if it is part of a business, or is for financial gain, or it is imagined that the copyright holder would make a financial gain if the distribution had not occurred (such claims are ludicrous).

>Others, for example violating an open source license with specific terms, are less financial but are still criminal and can result in financial penalties.

In the UK, violating an "open source" license would likely only be regarded as an offense if done commercially (reviewing section 107 - there is no offense for deemed losses that are not financial).

The typical case of copyright infringement of a strong free software license is to take the free software and to make it proprietary (usually commercially, but sometimes noncommercially) - which is immoral no matter what the law say (immorally it seems that individuals making software proprietary noncommercially in the UK would get away with such disgraceful act).

>As long as you continue to use flawed logic like this to either claim that these aren't crimes

Receiving an unauthorized copy is not an offense in the UK, nor is distribution that isn't imagined to cause a financial loss.

>You can argue that they should not be crimes and thus that we should consider changing laws.

Noncommercial sharing of generally useful public information should not be an offense not matter how, or how often it is done - after all, computers are the ultimate copying machine and never using them to copy is serious brain damage

In the UK, that could be achieved via removing remove the parts about imagined losses and also the parts forbidding breaking digital handcuffs from the copyright acts - but that will never happen.

[doublelayer]

Re: Stored or not stored?

"Always relying on the law to determine right or wrong is trite logic that fails hard - as there's a lot of laws that are immoral - but magically those laws are moral because the law says so."

Read the comments again until you understand that I did not say that. There's a difference between legal and moral, and my point is that, no matter whether you think something is moral, if the law says it's a crime, you have to change the law to make it not one. If you keep using logic that you think it's moral thus it's allowed, you're going to get very surprised when you lose in court. The rest of us will either try to get the law changed or will work within it.

The arguments so far that it is moral have been...well they haven't really been arguments. You just say it. The closest things to arguments we got were that it's easy, so you might as well make it not an offense, which is a really stupid attempt. Your new one is about whether certain things count as an offense in the UK which is mostly unimportant because you have clearly stated that commercial violation of copyright is, and the people the article is talking about are very commercial indeed. This thread has clearly outlived its usefulness, but I suggest thinking about your arguments next time you want to convince people; I'm much closer to your view than you recognize, but you've used and defended a lot of clearly bad arguments which don't respond to any of my points but appear to have been picked out of a bag of cliches.

[Yet Another Anonymous coward]

Re: Stored or not stored?

>It was obviously used. By a for profit company.

So if an employee at a for profit company read a textbook and used ideas they "learned' from it ?

It's going to make film studies programs a bit tricky. We can only employee you if you can prove you learned nothing from any of the copyrighted works you studied. I suppose English Lit is OK if you only study Chaucer and Shakespeare ( but not any modern translations)

[doublelayer]

Re: Stored or not stored?

This rubbish again? If your employee needs to learn from some materials and they steal copies, then yes, they can be punished for doing so and if you told them to do it, so can you. The AI companies did not obtain licenses to the stuff they trained on because they were not willing to pay for the things that, if we used them, we would have to pay for or commit a crime.

[Long John Silver]

Re: Stored or not stored?

First, you have to catch them. Second, catching them is only worth the bother when infringers have deep pockets. Third, 'law' is not worthy of respect when it obstructs cultural progression.

[doublelayer]

Re: Stored or not stored?

Well yes, in order to have consequences for your crime, you have to be caught first. Not really a useful point here, is it?

And in my opinion, copyright supports cultural progression and violating it obstructs it by making it much harder for those who progress culture by making creative works. You want to use those works, meaning you recognize that they have value, but you want to find a way where you get them without having any benefit accrue to the person who did all the work. If I had total control over copyright law, I would weaken it, but it would remain because none of your ill-considered replacements have or will work.

[BasicReality]

We grant far too many rights to copyright holders. Be nice to see those laws weakened a bit. Book publishers, the music/movie industries, these places have abused customers for years, it would be nice to see courts stop ruling in their favor so often.

[Long John Silver]

Abuse is inevitable when supply and price are not determined by market economics. In most other contexts, monopolies are abhorred.

[that one in the corner]

Of course, Getty has always been whiter-than-white

https://www.forbes.com/sites/legalentertainment/2016/08/03/pay-up-getty-sends-trolling-letter-to-photographer-highsmith-demanding-money-for-her-own-photos/

https://www.theregister.com/2015/09/09/i2600i_girds_loins_to_fight_off_copyright_troll/

[Snowy]

Re: Of course, Getty has always been whiter-than-white

Just what I was thinking about, as I was sure they had "found" other people's work and adopted it as their own.

[Long John Silver]

Amusing, but irrelevant

The major 'AI' producers and the copyright-dependent rentiers are gearing up for a battle of the Titans. Neither party in the dispute has attractive qualities. The former are seeking to garner de facto monopoly power by virtue of their sheer financial clout, coupled with persuading enterprise that it must tap into huge 'AI' models which are feasible to run only in vast power-hungry centres. The latter, a complacent bunch with an enormous sense of 'entitlement', are desperate to maintain a status quo based upon the now anachronistic, and risible notion, that the expression of ideas and of artefacts existing in the digital realm, can be equated to traditional property rights such as pertain to land and physical artefacts.

The 'AI' vendors may have partial success pushing their poorly conceived and massive models, wherein reside the shadows of indiscriminately chosen 'training data' coupled with a pretence that the 'AIs' have emergent properties qualitatively differing from ordinary databases. Perhaps, these vendors can be tied up in legal knots and forced to pay rent. However, the truly useful 'AI' models shall be of modest size, capable of accommodation by simple computers, and 'trained' for very specific purposes; these will be created by small groups and independently of the leviathans. When information outlets like Anna's Archive integrate their data stores with 'AI's trained on coherent subsets to act as helpful librarians, a huge swathe of formerly jealously guarded cultural materials will become permanently part of 'the commons'. Rentierism shall be replaced by a more defensible economic milieu. Many of the bespoke 'AIs' will immediately be freely accessible for anybody wishing to host them. Other similar 'AIs', when not requiring huge data-processing centres, will be 'escaped' into the wild on the principle of digits refusing to be corralled.

With ubiquitous 'local use' 'AIs', rentiers, regardless of 'law', will be powerless to curb so-called 'infringing use' by individuals. That will parallel the present day inability of rentiers of film, music, literature, academic materials, and so forth, to influence 'consumer' behaviour.

Anyway, let the giants slug it out. Getty Images is unattractive even by the mainstream standards of the generality of rentiers; it makes no pretence of involvement in the creative process of photography.

[hh121]

Nonsense

So crawling copyright content and making weighted models that can generate something extremely similar or even apparently identical to those sources....how is that different to making a low quality mp3 of a cd? or a compressed rip of a DVD? Or a dodgy jpeg of a work of art?

They will all collapse under the weight of their own irrelevance when the bubble bursts (most likely) or they keep training themselves on their own output till all that comes out is (even more) garbage, and Altman finally buggers off (unfortunately much richer).

[Eric 9001]

Re: Nonsense

Every creative work automatically falls under copyright whether its author wants it to or not - thus it's well known that it must be assumed that all images downloaded from the internet are under copyright, unless such images were marked as a scan of an ancient image or have been validly placed into the public domain on the page the image was downloaded from.

Exact copies do happen, but in the general case it's a bit like slamming parts of a bunch of CD's together and re-encoding into mp3 - the only difference is that buzzwords are plastered everywhere and it's magic as it's "AI".

[hh121]

Re: Nonsense

I'd agree, except for me it looked more like 'take all of artist/copyright holder (x)'s material, smoosh it together and create something (y) in the style of (x)'. Where 'x' is your music/art/movie/literature/news outlet of choice. Springsteen, Spielberg, Stephen King...doesn't matter.

But it will be the BMG/Universal/News Corps etc that will be able to either fight it (temporary) or get *their* payday (also temporary).

The rest of you, good luck.