Is it bad that this will be used to displace individual artists/creatives in the value chain of media production? Of course it is. But we shouldn't be responding to that by clinging harder to schemes that have outlived their usefulness, we should be developing new models for funding production.
Great! So given your articles are in the public domain on your website I can make millions out of it without given you a cent or direct credit and sources without paying you and can claim it all as my own then.
The hypothetical of lost revenue needs to be validated by the evidence of actual revenue being made in this way. Right now, I see lots of interest in paying for the _tool_, but almost none in paying human wages for the _output_ of generative models.
(I am not proposing that my above distinction is a legal test. Just pointing out that all these arguments would be more credible if actual ai generated works were being sold by AI companies)
There is no recognition of 'lost revenue' for the commercialization of public domain works.
My point was that we haven't yet seen the products of GenAI stuff really making money yet. People are paying for the tool, and people are paying for work that is being done using the tool, but nobody buys a book, image, movie, or similar from OpenAI or Google directly.
I have sympathy for the artists, but frankly this is progress and it can't be stopped. The economics are so lop-sided in favour of silicon that the law won't be able to hold it back without crippling society at large. Artists aren't the only ones affected and they may not even be the profession most impacted.
Kinda sad that artists having their income even more crippled is seen as progress.
On the other hand, I grew up in one of those places, the UK, and there were a lot of people moaning that the TV License (which funds the BBC) was an abominable stealth tax or words to that effect.
I'm now in Germany, where everyone has to pay the equivalent even if they don't own a TV or watch live over the internet (unlike the UK where not doing that means you don't need to pay); I've not heard anyone complain so far… but I don't know if that's because they genuinely don't, or if I'm peacefully oblivious by never having been suckered into reading a German-language comments section.
Sadly, I think most artists would bite my head off right this minute? So I guess I'm going to have to wait until the storm blows over.
Not sure about that. It's like saying if AIs are good at art, they'll be better at math.
Also, GPT-3.5 is already a better coder than a few humans whose mistakes I've had to fix. 3.5 is nowhere near the best, yet it's already eating at the bottom rungs despite being free.
This describes how ideas work - and ideas are rarely (if ever) IP-locked by a first originator - because there usually isn't one.
Everything by everyone, everywhere is built on the output of predecessors. Progress is a shared effort made up of minuscule increments or slight reorderings - which are typically done several times before they catch on.
IP exists to hinder this process by preventing 99.99999% of potential people from advancing ideas.
But if they claimed that they wrote the content then they would be defrauding their customers, since saying they wrote them would just be lying. You don't need intellectual property for fraud (as in "lying for material gain") to exist and be bad. At the very least it would be dishonest academically speaking and they should be criticized for it.
You're mischaracterizing me by suggesting that it's "economic suicide", as I said in my original comment in this thread:
> But we shouldn't be responding to that by clinging harder to schemes that have outlived their usefulness, we should be developing new models for funding production.
Sure it would be disruptive if we snapped our fingers and said "no more IP starting tomorrow", there should be a gradual phasing out of these unfair protections and effort put into sustainable pro-creative models.
A lot of people are very myopic about generative AI, thinking that large Hollywood studios are going to steal people's work and put everyone out of a job. Hollywood studios won't exist soon enough because anyone will be able to put together a movie that rivals existing expensive productions. In fact, generative AI is democratizing, as long as it's not gatekept by a few large corporations, which is exactly what trying to misapply copyright here would do.
Tens of millions of fans of a living celebrity would pay for it and the fans do not care about the original author as long as it is the celebrity's name that is selling it and can claim it as their own.
They don't need to give credit or sources to whatever their selling to generate millions.
Either way, they can make millions out of it, even if you are the original author and don't need to give you credit.
Also, those emails seem very likely to be ordered to be produced during discovery.
This thing could really go either way at this point but I feel like Stability has the upper hand.
Imagine training a model without any of the plaintiffs images, then using that side by side with the model that does. This could then be used to show the jury that those individual works are of no importance to the system if the images are of the same quality.
They will probably argue that the individual expressions of each work are not copied, rather the abstract ideas of two-dimensional representations present across any and all images.
Expect lots of side by side pictures as Exhibits from both side! Grandma and her fellows have to weigh in on this one!
This is a fun one!
If the images are REALLY of no importance, they wouldn't have been used anyway.
And yes, the nearly identical copies will definitely be presented to the jury.
The gist was if you overtrain a model, and try to recreate an exact image by prompting very similar things, and run it several thousand times, then you can recreate an image. If you seriously crank up the overtraining it's even easier. But normal use of the models do not just pump out recreations of training data.
If you put your "work" out in the world, anyone who views it, is automatically training their brains on it. Viewing is training.
Same thing that happened on the construction of the "corporation as a person", built on top of rulings made to protect African Americans.
We've achieved AGI, it's just yet a baby. Only a few years ago, if you wanted a model that could complete a task, you'd have to train a model specific to that task. GPT-4 can do a large variety of tasks without being specifically trained on them, i.e. it's suitable for a wide range of general tasks. Yeah, it sucks at some of them, but focusing on that is short-sighted. Look at the progress made from just a few years ago and where it's going, especially with multimodality and increased compute power.
This thing is a machine intelligence. My point is that a machine intelligence to be equivalent to humans in every law is curiously never mentioned, except when it is advantageous to the owners of said machine.
Or are we saying that we can't turn off GPT-4 at this point, because following your metaphor, we would be committing murder or abortion?
Again, you can't reason using metaphors and inferences and analogies. It's not helpful.
> except when it is advantageous to the owners of said machine.
That's just not true. You're ignoring vast amounts of discussion on the topic, particularly the discussions had when Blake Lemoine claimed that LaMDA was sentient. There's no reason to dismiss all that just because you're dead set on assuming everyone involved in AI is up to no good.
A perfectly reasonable view for humans, since you shouldn't be able to copyright a brain.
Not at all a reasonable view for a computer, until we also get to freely use all the copyrighted works ourselves. The problem here is that AI training is asymmetric: the people training an AI use works in violation of their licenses, but don't let their own works be used in the same way. For instance, Microsoft uses code on GitHub to train Copilot, but you still don't get to freely the source code of Windows or GitHub.
I am absolutely in favor of eliminating copyright and patent law. I am not in favor of keeping it around while letting AI become a laundering mechanism to get around it. AI training should not get to uniquely ignore copyright; copyright should cease to exist.
2. Despite the differences, the similarities are rhetorically convincing for many.
I recon there will be a lot of bad analogies for the next decade at least.
For the purpose of creating new art from experiencing previously created art, why should we treat them differently?
If AI training should be fair use has nothing to do about how it works (search engines fall under fair use regardless of if they think or not) but the effects it have on the original author.
So when people talk about humans and AIs "doing the same thing", AIs are not legal entities like humans are to begin with, the entire premise is moot.
That's what they're upset about.
If there were no use for 2D artists, then Stability Ai wouldn't be making an AI to replace them.
Key word here is: replace. 2D artists are not becoming obsolete - they're being replaced by a machine that was trained on their works without permission.
If you want to make an AI that does amazing paintings, and doesn't use human training data, more power to you. I can't compete with that. But if you use MY WORK to make a machine that's going to replace me, you do it under the cover of darkness and without permission - yeah i'll get pretty mad.
What happened to visual artists was more like Logitech announcing Logitech CoPilot and revealing they've extracted code from keylogging for the past 20 years.
This is kind of petulant then, respectfully. They got paid to produce that work, and they sold their services for that paycheck and the knowledge that they were putting their work into the world. The fruits of their labor were already being used to make Disney, etc. massively wealthy.
But this is even another step of indirection from that totally fine and reasonable situation: people (or machines) are learning from the work and producing their own. They might as well be angry at interns who learned their style to get junior jobs at Disney and who are now up and coming, replacing them.
A lot of the works were just ripped off Arstation and similar websites. It's likely a very large number was never "paid for".
A lot of that work was never paid for to begin with. It was just spec work by artists who a. might be too young b. might be in between jobs c. just doing passion work.
Again a lot of people at HN don't really seem to know the story very well, or how artists are compensated.
However most of the images that are being used for training data don't come from Disney. Most of the imagery used for Stable Diffusion and Midjourney was from portfolio websites and other sources, afaik. The Artstation website, which is a popular portfolio website, was one of the main targets, alongside many others. Other websites used include DeviantArt and Flickr. You can check using this website https://haveibeentrained.com
A lot of people placed their passion work, personal work, and unpaid work on Artstation, DeviantArt and Flickr. Some of those people are brilliant. Stability AI used their work as training data, without announcing it or asking for permission. The result is that a lot of more profilic and famous artists can see a lot of work that directly seems to be a reference to their own, copying their style.
A lot of the work on these websites is quite simply not commercial work. The only commercial application in fact, so far, is these image engines.
What we have here is someone hoovering up copyrighted and protected content, from all over the internet, and using it to generate income. They then are turning around and claiming that they do not need to pay for these protected works because... Well, it's always something hand-wavy like "machine learn like people" but it boils down to "I want to keep the money."
I'm still confused as to their business model.
Now someone wants to use one of the songs in an ad. We "got paid" to produce the work, but the people making the ad still have to pay us for that use of the song, because it's a different use than the one we got paid for.
I'm not sure what your comment about "Logitech CoPilot" is. GitHub CoPilot exists, was trained on code from the past 20 years, and by and large developers enjoy the additional automation, and are looking for ways to leverage this automation to be more productive. I would think artists and writers would adopt a similar approach. Experience tells me that fighting automation is a waste of time. Best to stay ahead of it.
Yes.
"Under the cover of darkness and without permission" implies quite a bit, but you're coming at it from the wrong angle, which this court ruling affirms. Try thinking of it as a new tool that will act as a force multiplier for your work.
This "but progress" argument is tiring. The Industrial Revolution was a complete failure if you measure quality of life in its first decades. It got turned around by people who you know, fought back, created unions, and demanded fair pay.
Good. That's the goal, and we can stop inventing stupid make-work for people when that happens. Jobs aren't the goal, making robots do all of the work is the goal. You're falling into the myopic trap of assuming that the current way of things won't change along with generative AI becoming commonplace.
My fear is that in between the world we have now and the world where robots do everything, we have a giant valley of madness where robots take all the jobs, all the profits, and a lot of people are left with nothing.
It would take an incredible amount of altruism to share your goal, and you don't become a billionaire by being altruistic.
But progress is part of the foundational purpose of copyright:
> 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.
I also take issue with the assumption that AI will "benefit everyone enormously, even if it won't be equally distributed"; I don't see any factual basis for this assumption. On the contrary, it seems much more likely that AI will be used to concentrate wealth even further. Given the high cost, I find it hard to believe it will ever by "equally distributed".
For as long as I can remember big corporations have been merciless in their preservation of "intellectual property". I didn't love it then and I don't love it now. OTOH, the idea that Microsoft can train their LLM on code I've written and then sell access to that LLM for money (sharing no dollars with me) strikes me as outright theft.
The scenario where AI training is locked down doesn't result in 1,000,000 individuals getting paid. (What would they get paid, and by whom?) It results in Disney, Adobe, etc.—massive companies with existing licenses to use content just about however they want—training their own models and locking everyone else out of the large AI model training game, until AI gets good enough to start generating human-quality creative work on its own (the same kind of progression as alphago/lee to alphago/zero), perhaps with the addition of a small set of purely copyright-free material.
Excluding all copyrighted material would be tying an AI model's metaphorical hands behind its back, since humans, although capable of producing great works through much iterative effort in isolation, all rely on having learned from some copyrighted work. Find an author who hasn't read plenty of recent books as well as older classics, or a musician (other than classical) who hasn't listened to plenty of modern music, or a director or editor who hasn't watched tons of movies and films. Recall Newton, "[I]f I have seen further, it is by standing on the shoulders of giants." Many of those "shoulders" are copyrighted.
My understanding is that people want to be compensated when their intellectual property is used as training data for a machine. That strikes me as an entirely reasonable expectation.
One person memorizing Harry Potter for their own amusement, even if they make money doing public appearances where they recite sections of the work verbatim for the amusement of the audience is not in any way similar to the process of training an LLM or of that LLM's output. The scale alone is so vastly different that it renders the comparison useless and misleading.
That's fine that they want to. The question is whether copyright law gives them that and that's very unlikely.
They pay for it.
They buy the books. They buy tickets to theatre. They buy entrance to the gallery.
The trick that's being done now is hey, we don't have to pay since it's not a person. (to the creator) But hey, it is just like a person when it learns! (legal system)
If AI models require human training data, then they should pay for it. Easy.
Books3 has separate legal concerns, but Google has a legally acquired corpus of tons of books, which they've mostly cleaned up from scans (probably far better than IA has), and have probably used to train Bard on. Their lawyers must be biting their nails waiting to see how these lawsuits turn out, though.
Until AGI arrives, or some other method of training LLMs from the ground up on sparse examples by incrementally building on structural knowledge of language.... training on ridiculous amounts of copyrighted content is required. Not because anyone wants to copy those works, but because training that way fills in for a lack of real-world experience that every child gets, which includes consuming and interacting with a bunch of copyrighted content that isn't tracked because it's not practical to do so.
You could train a LLM only on project gutenberg, and the LLM would churn out stilted English and the occasional iambic pentameter. That's great if you want works that seem like they were written over a century ago, but nearly useless otherwise.
>Until AGI arrives, or some other method of training LLMs from the ground up on sparse examples by incrementally building on structural knowledge of language.... training on ridiculous amounts of copyrighted content is required.
that's not my problem. Those AI model folk should just compensate the people they're using training data from, and they should ask for permission.
>You could train a LLM only on project gutenberg, and the LLM would churn out stilted English and the occasional iambic pentameter. That's great if you want works that seem like they were written over a century ago, but nearly useless otherwise.
Not my problem. Why are the problems of the wonderful AI developers suddenly human, global problems that we all have to find a way to fix?
If they want access to training data - they should pay for the privielige.
Physical library books are governed by the doctrine of first sale. That's why google has one of the largest (maybe excluding l-bg-n and IA) corpus of books on the internet. They might have the cleanest corpus of OCR'd book content of anyone, since IA uses commercial or open source OCR and that's it, while google for a long time used recaptcha to check OCR results.
For physical books, the cost per read of a library book is an order of magnitude smaller than the cost per read of privately purchased books. How can you tolerate the economic model of libraries when the net effect is a theft of maybe 80%-95% from the author and publisher? Libraries subsidize books that nobody wanted to read, but steal from authors and publishers whose books are read multiple times per physical copy.
Even libraries' onerous ebook licenses are not commercial retail ebook pricing. They're just close to it than publishers could ever manage with physical books, because there's no pesky right of first sale which allows libraries to lend physical books on their own terms to library patrons without getting permission or licensing from the publisher.
I would prefer to get away from OpenAI and Facebook and all the other people using potentially tainted sources like books3. The legal question there is not whether training was legal, but whether the acquisition of the training data was legal to begin with, and that's a straightforward copyright issue.
What is your position on Google using its corpus of books (legally acquired and possessed, as the content behind google books) to train a LLM? Do they need to acquire additional rights from copyright holders? Why, and under what legal theory?
First, there's this thing called a deposit library: https://en.wikipedia.org/wiki/Legal_deposit
Second: I, as a user of the service, who learns things, still pay nothing.
Should I be required to directly pay for the things I learned from, or is it sufficient that someone is? Because if the latter, then picking up a book from a normal (non-deposit) library, showing it to an OCR system, and having an AI learn from that, would involve just as much payment as I ever made to read a library book (with the possible exception of late return fines, I can't remember if I ever had any of those).
You: Yes, and you know how humans acquire works to learn from? They pay for it.
Them: False. Libraries exist. Borrowing books from neighborhood libraries or friends exists.
You: Libraries exist? Do you think books fly onto library shelves for free? As far as I know, someone bought them.
Them: then [pick] up a book from a normal ... library, [show] it to an OCR system, and [have] an AI learn from that
You: Everybody finds an edge case they can justify their entire worldview on.
However, it's not an edge case; it's a direct consequence of your argument. If humans learning from libraries rewards copyright holders because libraries paid for the books, then an AI training from libraries rewards copyright holders in exactly the same way since the libraries paid for the books.
They're saying, "You can't consume my book in that way."
This doesn't cover everything: I did, indeed, also buy books on HTML and JS, and my first C compiler, and a licence to REALbasic[1]. But that doesn't refute the fact that I did learn a lot for free.
> If AI models require human training data, then they should pay for it. Easy.
You can do that if you like, but that won't stop any of the economic issues that arise. The cost of running Stable Diffusion is so low that even if you had literal slaves, and you were spending only the UN extreme poverty threshold on keeping them alive and housed, the pro-rata cost of keeping them alive for long enough to type in the prompt dominates the total cost of making images.
Right now these models are still, despite their impressiveness, flawed: while an artist can use them to great effect, most of us will have our generations easily spotted by some flaw we have never trained ourselves to notice. If the models become good enough to fully replace all artists, the only way the profession called "artist" isn't going to go the same way as the profession called "computer" is if the arts are to humans as fancy tails are to peacocks: the effort being the point, extravagantly wasting effort to show you're fit enough to manage fine despite the penalty.
[0] UK rules at the time, thanks to my dad's early retirement and therefore "low income" status
[1] as it was so named at the time, Xojo now
This will happen in every industry and occupation, one by one.
Is this what you think is desirable?
The alternative is perversely simple: PAY for the right to use training data.
Post-scarcity economics, AKA "fully automated luxury communism" (not the book of the same name): https://en.wikipedia.org/wiki/Post-scarcity
The main Stable Diffusion models themselves are licensed under Creative ML OpenRAIL-M, and freely available third-party models exist; OpenAI gets stick for not making their models downloadable (at least GPT-x does, though nobody seems to care too much about them keeping the DALL•E 2/3 models private), other players are making and distributing other LLMs of varying quality.
That said, we do also have actual reasons to be concerned about this remaining anarchic (though a lot of people very loudly scoff at them), so who knows what the future will bring on that axis.
We've been here before several times: Silhouette painting, Photography, Airbrushing, Pianos, Synthesizers, Sampling, Photoshop, Ray Tracing, and many more. "It's not real art", "they're stealing from us", "we'll go hungry!" .
Some of these are already quite old. For others, I've actually been asked the question back when I was in school: "Are you really making music if your instrument has a microprocessor in it?". Um, yes, yes I claim I am making music thank you very much.
First people complain, then they adapt, and then they end up making awesome art with the new tools and/or instruments. Which isn't to say historically it was all rainbows and roses, but it was never the end of the world either. Seeing the newer generation of AI tools and how the tools end up getting integrated into regular workflows, it seems to be going the same direction.
To quote the song, I think it's "all just little bits of history repeating".
I'm sure you felt quite clever after writing this, but it isn't.
Even the Industrial Revolution, which you probably think increased the general quality of life, decreased it for a significant amount of time. https://web.archive.org/web/20230622001458/https://www.econo...
It's the usage of unlicensed images and text in their training data that I have an issue with. If you wrote a book or made a painting, GPT-4 should pay for the rights to read or look at it.
It doesn't really matter though. The models that exist today aren't going away, and future work will just be built on top of their output, tainting every model forevermore from your perspective. I suggest you adapt to the new way of things and enjoy the ride.
Buying one copy is certainly better than 0. But I doubt rights holders would be satisfied with that.
Copyright is archaic and holding humanity, and especially creatives, back.
I'd really like to see people drop the inequality argument. If you actually cared about that instead of virtue signaling, you'd push for a mandatory GPL-style license that forces models to be available to anybody that uses them. That would avoid trying to unsuccessfully put the genie back in the bottle, while also preventing a few companies from benefiting at the expense of everyone else. Just like OpenAI's original mission of making AI available for everyone, before they got dollar signs in their eyes.
Good.
Intellectual Property is a mistake. If AI brings about its end, I welcome it.
All because judges will be befuddled about what to do after hearing terms like “training data” and “compression”. We will, of course get the emails in 10-20 years showing that it’s all lies and that the CEOs of these companies knew exactly what they were doing.
If this continues, AI will be the great inequality machine in history. Take data from 1,000,000 individuals, train your AI to replace them, compensate no one.
You can do this in every area: driving, farming, cooking. Music. Just dispossess everyone of all their property by training an AI to copy all their work! What could be easier (and less morally right)…