The rumblings I'm hearing are that this a) barely works with last-gen training processes b) does not work at all with more modern training processes (GPT-4V, LLaVA, even BLIP2 labelling [1]) and c) would not be especially challenging to mitigate against even should it become more effective and popular. The Authors' previous work, Glaze, also does not seem to be very effective despite dramatic proclamations to the contrary, so I think this might be a case of overhyping an academically interesting but real-world-impractical result.
The screenshots you sent in [1] are inference, not training. You need to get a Nightshaded image into the training set of an image generator in order for this to have any effect. When you give an image to GPT-4V, Stable Diffusion img2img, or anything else, you're not training the AI - the model is completely frozen and does not change at all[0].
I don't know if anyone else is still scraping new images into the generators. I've heard somewhere that OpenAI stopped scraping around 2021 because they're worried about training on the output of their own models[1]. Adobe Firefly claims to have been trained on Adobe Stock images, but we don't know if Adobe has any particular cutoffs of their own[2].
If you want an image that screws up inference - i.e. one that GPT-4V or Stable Diffusion will choke on - you want an adversarial image. I don't know if you can adversarially train on a model you don't have weights for, though I've heard you can generalize adversarial training against multiple independent models to really screw shit up[3].
[0] All learning capability of text generators come from the fact that they have a context window; but that only provides a short term memory of 2048 tokens. They have no other memory capability.
[1] The scenario of what happens when you do this is fancifully called Habsburg AI. The model learns from it's own biases, reinforcing them into stronger biases, while forgetting everything else.
[2] It'd be particularly ironic if the only thing Nightshade harms is the one AI generator that tried to be even slightly ethical.
[3] At the extremes, these adversarial images fool humans. Though, the study that did this intentionally only showed the images for a small period of time, the idea being that short exposures are akin to a feed-forward neural network with no recurrent computation pathways. If you look at them longer, it's obvious that it's a picture of one thing edited to look like another.
Hey you know what might not be AI generated post-2021? Almost everything run through Nightshade. So given it's defeated, which is pretty likely, artists have effectively tagged their own work for inclusion.
I mean that's more or less status quo isn't it? Big business does what it wants, common people can get fucked if they don't like it. Same as it ever was.
That's exactly right. It is just the variety of new ways in which common people get fucked that is dispiriting, with seemingly nothing capable of moving in the opposite direction.
Modern generative image models are trained on curated data, not raw internet data. Sometimes the captions are regenerated to fit the image better. Only high quality images with high quality descriptions.
I wouldn't call what Stable Diffusion et al are trained on "high quality". You need only look through the likes of LAION to see the kind of captions and images they get trained on.
It's not random but it's not particularly curated either. Most of the time, any curation is done afterwards.
Correct me if I'm wrong but I understand image generators as relying on auto-labeled images to understand what means what, and the point of this attack to make the auto-labelers mislabel the image, but as the top-level comment said it's seemingly not tricking newer auto-labelers.
not all are auto labelled, some are hand labelled, some are initially labelled with something like clip/blip/booru and then corrected a bit by hand. The newest thing though is using llm's with image support like GPT4 to label the images, which kind of does a much better job most of the time.
Your understanding of the attack was the same as mine, it injects just the right kinds of pixels to throw off the auto-labellers to misdirect what they are directing causing the tags to get shuffled around.
Also on reddit today some of the Stable Diffusion users are already starting to train using Nightshade so they can implement it as a negative model, which might or might not work, will have to see.
Even if no new images are being scraped to train the foundation text-to-image models, you can be certain that there is a small horde of folk still scraping to create datasets for training fine-tuned models, LoRAs, Textual Inversions, and all the new hotness training methods still being created each day.
If it doesn't work during inference I really doubt it will have any intended effect during training, there is simply too much signal and the added adversarial noise works on the frozen and small proxy model they used (CLIP image encoder I think) but it doesn't work on a larger model and trained on a different dataset, if there is any effect during training it will probably just be the model learning that it can't take shortcuts (the artifacts working on the proxy model showcase gaps in its visual knowledge).
Generative models like text-to-image have an encoder part (it could be explicit or not) that extract the semantic from the noised image, if the auto-labelers can correctly label the samples then the encoded trained on both actual and adversarial images will learn to not take the same shortcuts that the proxy model has taken making the model more robust, I cannot see an argument where this should be a negative thing for the model.
The context windows of LLMs are now significantly larger than 2048 tokens, and there are clever ways to autopopulate context window to remind it of things.
The animation when you change images makes it harder to see the difference, I opened the three images each in its own tab and the differences are more apparent when you change between each other instantly.
If you have to have both and instantly toggle between them to notice the difference, then it sounds like it’s doing its job well and is hard to notice the difference.
Kid me found 13 FPS in games to be a smooth and cursive experience.
Current me thinks 60 FPS is laggy.
Standards differ. I saw glazed images in the wild, was wondering why they have so much JPEG artifacts, until I saw the post of one of those anti-AI + glaze images on his profile.
That is a great mystery, to me it's as clear as if someone pasted a cartoon dog onto the image, it's extremely blatant and impossible to ignore by my normal human pattern recognition.
I'm looking at them on my iPhone 14 Pro and I am having a hard time seeing any meaningful difference that changes the way the artwork registers with me.
I can't really imagine a case where if I had only seen the AI edited one I would have any different reaction or response to viewing the piece of art compared to having only seen the original one.
But now that I double-check, I was comparing with the images zoomed to 200%. On desktop the artifacts are also noticeable at 100%, but not nearly as bad as in my previous comment.
I didn't see it immediately either, but there's a ton of added noise. The most noticeable bit for me was near the standing person's bent elbow, but there's a lot more that becomes obvious when flipping back and forth between browser tabs instead of swiping on Twitter.
I was on desktop and it looks like pretty heavy jpeg compression. Doesn't completely destroy the image, but it's pretty noticeable when blown up large enough.
Maybe it's more about "protecting" images that artists want to publicly share to advertise work, but it's not appropriate for final digital media, etc.
Seems obvious that the people stealing would be adjusting their process to negate these kinds of countermeasures all the time. I don't see this as an arms race the artists are going to win. Not like the LLM folks can consider actually paying their way...the business plan pretty much has "...by stealing everything we can get our hands on..." in the executive summary.
I think the point is that they're akin to a watermark.
Even before the current AI boom, plenty of artists have wanted to showcase their work/prove that it exists without necessarily making the highest quality original file public.
For example in accounts on image sites that are exposed to suspected scrapers but not to others. Scrapers will still see the real data, but they'll also run into stuff designed to mix up the training process.
Huge market for snake oil here. There is no way that such tools will ever win, given the requirements the art remain viewable to human perception, so even if you made something that worked (which this sounds like it doesn’t) from first principles it will be worked around immediately.
The only real way for artists or anyone really to try to hold back models from training on human outputs is through the law, ie, leveraging state backed violence to deter the things they don’t want. This too won’t be a perfect solution, if anything it will just put more incentives for people to develop decentralized training networks that “launder” the copyright violations that would allow for prosecutions.
All in all it’s a losing battle at a minimum and a stupid battle at worst. We know these models can be created easily and so they will, eventually, since you can’t prevent a computer from observing images you want humans to be able to observe freely.
The level of claims accompanied by enthusiastic reception from a technically illiterate audience make it sound, smell, and sound like snake oil without much deep investigation.
There is another alternative to the law. Provide your art for private viewing only, and ensure your in person audience does not bring recording devices with them. That may sound absurd, but it's a common practice during activities like having sex.
That doesn't sound like a viable business model. There seems to be a non-trivial bootstrap problem involved -- how do you become well-known enough to attract audiences to private venues in sufficient volume to make a living? -- and would in no way diminish demand for AI-generated artwork which would still continue to draw attention away from you.
The thing is people want the benefits of having their stuff public but not bear the costs. Scraping has been mostly a solved problem especially when it comes to broad crawling. Put it under a login, there, no more AI "stealing" your work.
I don't think that's true at all. Images and text get reposted with or without consent, often without attribution. It wouldn't make it right for the AI companies to scrape when the original author doesn't want that but someone else has ignored their wishes and requirements. Basically, what good is putting your stuff behind login or some other restrictive viewing method if someone just saves the image/text? I think it's still a relatively serious problem for people creating things. And without some form of easy access to viewing, the people creating things don't get the visibility and exposure they need to get an audience/clients.
This is one the AI companies should offer the olive branch on IMO, there must be a way to use stenography to transparently embed a "don't process for AI" code into an image or text or music or any other creative work that won't be noticeable by humans, but the AI would see if it tried to process the content for training. I think it would be a very convenient answer and probably not be detrimental to the AI companies, but I also imagine that the AI companies would not be very eager to spend the resources implementing this. I do think they're the best source for such protections for artists though.
Ideally, without a previous written agreement for a dataset from the original creators, the AI companies probably shouldn't be using it for training at all, but I doubt that will happen -- the system I mention above should be _opt-in_, that is, you must tag such content that is free to be AI trained in order for AI to be trained on it, but I have 0 faith that the AI companies would agree to such a self-limitation.
edit: added mention to music and other creative works in second paragraph 1st sentence
edit 2: Added final paragraph as I do think this should be opt-in, but don't believe AI companies would ever accept this, even though they should by all means in my opinion.
Here are my 2 cents, I think we will need some laws specifying two types of AI models, ones trained with full consent (opt-in) for its training material and ones without. The first one would be like Adobe's firefly model where they allegedly own everything they trained it with, or something where you go around asking for consent for each thing in your training corpus (probably unfeasible for large models). Maybe things in the public domain would be ok to train with. In this case there are no restrictions and the output from such models can even be copyrighted.
Now for the second type, representing models such as Stable Difusion and Chat GPT, it would be required to have their trained model freely available to anyone and any resulting output would not be copyrightable. It may be a more fairer way of allowing anyone to harness the power of AI models that contain essentially the knowledge of all man kind, but without giving any party an unfair monopoly on it.
This should be easily enforceable for big corporations, else it would be too obvious if they are trying to pass one type model as another or even keep the truth about their model from leaking. It might not be as easy to keep small groups or individuals from breaking those rules, but hey, at least it evens the playing field.
Is that login statement strictly true? Unless the login is paid, there's no reason we can't get to (if not already there) the point where the AI scraper can just create a login first.
No, eforcing click-wrap legal agreements is actually possible. With basic KYC the scraper would instantly open up itself for litigation and no internet art piece is frankly worth this sort of trouble.
you are not incorrect that this would help mitigate, but it still misses a few key points I think regarding why artists are upset about AI generation
- This is still vulnerable to stuff like mturk or even just normal users who did get past the anti-bot things pulling and re-uploading the content elsewhere that is easier for the AI companies to use
- The artists' main contention is that the AI companies shouldn't be allowed to just use whatever they find without confirm they have a license to use the content in this way
- If someone's content _does_ get into an AI model and it's determined somehow (I think there is a case with a news paper and chatGPT over this very issue?), the legal system doesn't really have a good framework for this situation right now -- is it copyright infringement? (arguably not? it's not clear) is it plagiarism? (arguably yes, but plagiarism in US court system is very hard to proof/get action on) is it license violation? (for those who use licenses for their art, probably yes, but it's the same issue as plagiarism -- how to prove it effectively?)
Really what this comes down to is that the AI companies use the premise that they have a right to use someone else's works without consent for the AI training. While your suggestions are technically correct, it puts the impetus on the artists that they must do something different because the AI companies are allowed to train their models as they currently do without recourse for the original artist. Maybe that will be ruled true in the future I don't know, but I can absolutely get why artists are upset about this premise shaping the discussion on AI training, as such a premise negates their rights as an artist and many artists have 0 path for recourse. I'm pretty sure that OpenAI wouldn't think about scraping a Disney movie from a video upload site just because it's open access since Disney likely can fight in a more meaningful way. I would agree with artists who are complaining that they shouldn't need to wait for a big corporation to decide that this behavior is undesirable before real action is taken, but it seems that is going to be what is needed. It might be reality, but it's a very sad reality that people want changed.
This would just create a new market for art paparazzis who would find any and all means to inflitrate such private viewings with futuristic miniature cameras and other sensors and selling it for a premium. Less than 24 hours later the files end up on hundreds or thousands of centralized and decentralized servers.
I'm not defending it. Just acknowledging the reality. The next TMZ for private art gatherings is percolating in someone's garage at the moment.
I find this difficult to believe; no matter how small your camera is, photography is about light. Art reproduction photography is surprisingly hard to do if you care about the quality of the end result. Unless you can surreptitiously smuggle in a studio lighting setup, tripod, and color checker card… sure you can take an image in secret, but not one that is a good representation of the real thing.
You could just build a stabilizer system and stand really still for 1 second. Then expose for a longer time. Photography is Apertrue, ISO, and exposure time. This will gather enough light to do a proper exposure even in a dimm lit venue. Anything darker and every viewer will have a hard time seeing the private art.
ANother thing would be to crank up the ISO and denoise it later. Its much more lossy but with this you could get lower exposure times.
It’s about number of photons and aperture. In principle this could be very hard to detect, especially once people get good at multiple distributed apertures that are coherent with one another.
For comparison, The Ocean Full of Bowling Balls did this and successfully remained hidden for half a century, only failing because it was published somewhere else.
This tool is free, and as far as I can tell it runs locally. If you're not selling anything, and there's no profit motive, then I don't think you can reasonably call it "snake oil".
At worst, it's a waste of time. But nobody's being deceived into purchasing it.
If this is a danger from "snake oil" of this type, it'd be from the other side, where artists are intentionally tricked into believing that tools like this mean that AI isn't or won't be a threat to their copyrights in order to get them to stop opposing it so strongly, when in fact the tool does nothing to prevent their copyrights from being violated.
I don't think that's the intention of Nightshade, but I wouldn't put past someone to try it.
So then of course, you also cannot sell your work, as those might put it online. And you cannot show your art to big crowds, as some will make pictures and put it online. So ... you can become a literal underground artists, where only some may see your work. I think only some will like that.
But I actually disagree, there are plenty of ways to be an artist now - but most should probably think about including AI as a tool, if they still want to make money. But with the exception of some superstars, most artists are famously low on money - and AI did not introduce this. (all the professional artists I know, those who went to art school - do not make their income with their art)
Everything old is new again. It's the same thing with any DRM that happens on the client side. As long as it's viewable by humans, someone will figure out a way to feed that into a machine.
Other people pointed out they appreciated this prose. It’s easy to forget what exactly people are asking for when they talk about regulating the training of machine learning models.
> leveraging state backed violence to deter the things they don’t want
I just want to say: I really appreciate the stark terms in which you've put this.
The thing that has come to be called "intellectual property" is actually just a threat of violence against people who arrange bytes in a way that challenges power structures.
I heard that flooding the net with AI generated art would do much much more harm to generative AI than this whatever is this. Yes, this must be some snake oil salesman, those take it seriously turn AIs own weapon against AI.
I'm thinking — is it possible to create something on a global level similar to what they did in Snapchat: some sort of image flickering that would be difficult to parse, but still acceptable for humans?
Sorry i do not use Snapchat and with googeling "Snapchat image flickering" i did not find a good result. Could you elaborate this a bit more or provide me with a link where this is described? Thank you very much. :)
My guess. Is that at some poi t of time You will not be able to use any generated image or video in commercial. Because of 100% copyright claim for using parts of copyrighted image. Like YouTube those days. When some random beeps matches with someone music...
A few months ago I made a proof-of-concept on how finetuning Stable Diffusion XL on known bad/incoherent images can actually allow it to output "better" images if those images are used as a negative prompt, i.e. specifying a high-dimensional area of the latent space that model generation should stay away from: https://news.ycombinator.com/item?id=37211519
There's a nonzero chance that encouraging the creation of a large dataset of known tampered data can ironically improve generative AI art models by allowing the model to recognize tampered data and allow the training process to work around it.
This seems like a pretty pointless "arms race" or "cat and mouse game". People who want to train generative image models and who don't care about what artists think about it at all can just do some basic post-processing on the images that is just enough to destroy the very carefully tuned changes this Nightshade algorithm makes. Something like resampling it to slightly lower resolution and then using another super-resolution model on it to upsample it again would probably be able to destroy these subtle tweaks without making a big difference to a human observer.
In the future, my guess is that courts will generally be on the side of artists because of societal pressures, and artists will be able to challenge any image they find and have it sent to yet another ML model that can quickly adjudicate whether the generated image is "too similar" to the artist's style (which would also need to be dissimilar enough from everyone else's style to give a reasonable legal claim in the first place).
Or maybe artists will just give up on trying to monetize the images themselves and focus only on creating physical artifacts, similar to how independent musicians make most of their money nowadays from touring and selling merchandise at shows (plus Patreon). Who knows? It's hard to predict the future when there are such huge fundamental changes that happen so quickly!
>Or maybe artists will just give up on trying to monetize the images themselves and focus only on creating physical artifacts, similar to how independent musicians make most of their money nowadays from touring and selling merchandise at shows (plus Patreon).
As is, art already isn't a sustainable career for most people who can't get a job in industry. The most common monetization is either commissions or hiding extra content behind a pay wall.
To be honest I can see more proverbial "Furry artists" sprouting up in a cynical timeline. I imagine like every other big tech that the 18+ side of this will be clamped down hard by the various powers that be. Which means NSFW stuff will be shielded a bit by the advancement and you either need to find underground training models or go back to an artist. .
It's not particularly that hard. The furry nsfw models are already the most well developed and available models you can get right now. And they are spitting out stuff that is almost indistinguishable from regular art.
> musicians make most of their money nowadays from touring and selling merchandise at shows
Be reminded that this is - and has always been - the mainstream model of the lineages of what have come to be called "traditional" and "Americana" and "Appalachian" music.
The Grateful Dead implemented this model with great finesse, sometimes going out of their way to eschew intellectual property claims over their work, in the belief that such claims only hindered their success (and of course, they eventually formalized this advocacy and named it "The Electronic Frontier Foundation" - it's no coincidence that EFF sprung from deadhead culture).
It is a funny appearance (weird viewpoint) that artists are furious loosing their monopily in stealing and cloning components from other artists, recomposing into a similar but new thing.
And that OpenArt on the analogy of OpenSource is a non-existing thing (I know, I know, different things, source code is not for the generic audience and can be hidden on will, unlike art, just having some generative thoughts artefact here ;) )
This feels like it'll actually help make AI models better versus worse once they train on these images. Artists are basically, for free, creating training data that conveys what types of noise does not change the intended meaning of the image to the artist themselves.
The number of people who are going to be able to produce high fidelity art with off the shelf tools in the near future is unbelievable.
It’s pretty exciting.
Being able to find a mix of styles you like and apply them to new subjects to make your own unique, personalized, artwork sounds like a wickedly cool power to give to billions of people.
In terms of art, population tends to put value not on the result, but origin and process. People will just look down on any art that’s AI generated in a couple of years when it becomes ubiquitous.
> population tends to put value not on the result, but origin and process
I think population tends to value "looks pretty", and it's other artists, connoisseurs, and art critics who value origin and process. Exit Through the Gift Shop sums this up nicely
I disagree. I definitely value modern digital art more than most historical art, because it just looks better. If AI art looks better (and in some cases it does) then I'll prefer that.
That’s totally fine, everyone’s definition of art is subjective. But general value of an art as a piece will just still be zero for AI generated ones, just like any IKEA / Amazon print piece. You just pay for the “looks pretty”, frame and paper.
>You just pay for the “looks pretty”, frame and paper.
But you pay that for any piece of art though? You appreciate it because you like what it looks like. The utility of it is in how good it looks, it's not how much effort was put into it.
If you need a ditch you're not going to value the ditch more if the worker dug it by hand instead of using an excavator. You value it based on the utility it provides you.
That analogy doesn’t work for art, since worker’s ditch is result based. There are no feelings like “i like this ditch”, “experience of a ditch” or “i’m curious how this ditch was dug”.
Again, i’m not saying buying a mass made AI art will be wrong. Just personally speaking, it will never evoke any feelings other than “looks neat” for me. So its inherent “art value” is close to 0 as I can guess its history is basically someone put in a prompt and sent it to print (which I can do myself on my phone too!). It’s the same as looking at cool building pics on my phone (0 art value) versus actually seeing them in person (non-0), mostly because the feelings I get from it. That being said, if it makes others happy, it’s not my place to judge.
This is already the case. Art is a process, a form of human expression, not an end result.
I'm sure OpenAI's models can shit out an approximation of a new Terry Pratchett or Douglas Adams novel, but nobody with any level of literary appreciation would give a damn unless fraud was committed to trick readers into buying it. It's not the author's work, and there's no human message behind it.
Thing is there are way more good books written, than any single person can consume in their lifetimes. An average person like me, reading a mixed diet of classics, obscure recommendations and what's popular right now, I still don't feel like I'm making a dent in the pile of high quality written content.
Given all that, the purpose of LLMs should be to create tailor made content to everyone's tastes. However, it seems the hardcore guardrails put into GPT4 and Claude prevent it from generating anything enjoyable. It seems, even the plot of the average Star Wars movie is too spicy for modern LLM sensibilities, never mind something like Stephen King.
That's where you spin up a local LLaMA instance. The largest models that are still runnable on consumer grade hardware actually beat GPT-3.5 at this point. And there are numerous finetunes all over the "spiciness" spectrum.
Novels aren't about a message. They're entertainment. If the novel is entertaining then it's irrelevant whether there is or isn't a message in it. Besides, literature enthusiasts will invent a message for a popular story even if there never was one.
Also, I'm sure that you can eventually just prompt the model with the message you want to put into the story, if you can't already do that.
I sounds like you don't value art as the purest form of human expression but you'll never be able to convince others to think like you with logic. For my part I think you fundamentally misunderstand the value of creativity but I know I won't change your mind either.
If it was really about the message, then why waste all the time with the rest of the novel? Describe the message in a sentence or two. You could read an entire library of books worth of messages in a few days.
But that wouldn't be helpful. It would've been memorable, because novels aren't just about the message.
I haven’t read anything “shit out” by any LLM that even nearly approaches the level of quality by the authors you named — would very much like to see something like that - do you have any evidence for your claims?
AFAICT current text generation is something approaching bad mimicry at best and downright abysmal in general.
I think you still need a very skilled author and meaty brain with a story to tell to make use of an LLM for storytelling.
Sure it’s a useful tool that will make authors more effective but we are far from the point where you tell the LLM “write a story set in Pratchetts Discworld” and something acceptable or even entertaining will be spit out - if such a thing can even be achieved.
> According to Marx, value is only created with human labour. This is not just a Marxist theory, it is an observation.
And yet it's completely and absolutely wrong. Value is created by the subjective utility offered to the consumer, irrespective of what inputs created the thing conveying that utility.
You are using marginal utility value theory. Parent comment is using labor value theory. In fact, there are also other value theories in economy. It's a mostly philosophical choice, and like other philosophical choices, it's not possible to accuse one of them of being wrong. It's a matter of choosing your philosophy, and understanding different philosophies.
> You are using marginal utility value theory. Parent comment is using labor value theory.
Yes, I'm aware. This is precisely why I'm stating the prior comment to be "absolutely wrong". Marginal utility is a substantially valid model, LTV is not.
> In fact, there are also other value theories in economy. It's a mostly philosophical choice, and like other philosophical choices, it's not possible to accuse one of them of being wrong.
Sure it is. These aren't theories in a normative sense, they're models of causality for manifest phenomena. They're closer to scientific theories than they are to philosophical axioms. LTV simply doesn't bear out with observation.
Labor theory of value is quite controversial, many economists call it tautological or even metaphysical. I also don't really see what LTV has to say about AI art, if anything, except that the economic value generated by AI art should be distributed to everybody and not just funneled to a few capitalists at the top. I would agree with that. It's true that more jobs get created even as jobs are destroyed, but it's also true that just as our ancestors fought for a 40 hour work week and a social safety net, we should be able to ask for more as computers become ever so productive.
In Marx's time, you needed humans to perform any kind of labor. Even machines needed operators. But there's nothing about LTV that would make it a hard requirement. The point of Marx's claim is that without someone performing labor using capital, there wouldn't be any value for the owner of said capital to pocket. This is just as true if you replace workers with AI.
There is somewhat famous digital artist from Russia - Alexey Andreev. Google it, he has very distinctive style of realistic technique and surrealistic situations, like landing big manta ray on the deck of aircraft carrier. Or you can see his old works in his 5-years-not-updates LJ [1].
Now he uses generative AI as one of his tools. As Photoshop, as different (unrealistic!) brushes in Photoshop, as other digital tools. His style is still 100% recognizable and his works don't become worse or more "generic". Is he still artist? I think so.
Your definition assumes that photography is not art and/or doesn't involve conscious skill and creative imagination. That's not the consensus, to put it mildly.
> Being able to find a mix of styles you like and apply them to new subjects to make your own unique, personalized, artwork sounds like a wickedly cool power to give to billions of people.
And in the process, they will obviate the need for Nightshade and similar tools.
AI models ingesting AI generated content does the work of destroying the models all by itself. Have a look at "Model Collapse" in relation to generative AI.
I know this is an unpopular thing to say these days, but I still think the internet is amazing.
I have more access to information now than the most powerful people in the world did 40 years ago. I can learn about quantum field theory, about which pop star is allegedly fucking which other pop star, etc.
If I don't care about the law I can read any of 25 million books or 100 million scientific papers all available on Anna's Archive for free in seconds.
As Jeff Bezos recently said on the Lex podcast: one of the greatest compliments you can give an inventor is that they’re invention will be taken for granted by future generations.
“It won’t be any more wickedly cool than the internet” - saying something won’t be any more wickedly cool than the most profound and impactful pieces of infrastructure human civilization has erected is a pretty high compliment.
They matter but not under the current system. Artists are a rarely paid profession, and there are professional artists out there but there’s now a huge amount of people that will never contact an artist for work that used to only be human powered. It’s not personal for me. I understand that desire to resist the inevitable but it’s here now.
For what it’s worth I never use midjourney or dalle or any of the commercial closed systems that steal from artists but I know I can’t stop the masses from going there and inputting “give me pretty picture in style x”
Resistance is important imo. If this happens and we, who work in this industry, say nothing, what good are we. It's only inevitable if it's socially acceptable.
Not really. There is a reason why we find realistic painting to be more fascinating than a photo and why some still practice it. The effort put in by another artist does affect our enjoyment.
For me it doesn’t. I’m generating images, realistic, 2.5d, 2d and I like them as much. I don’t feel (or miss) what you described. Or what any other arts guy describes, for that matter. Arts people are different, because they were trained to feel something a normal person wouldn’t. And that’s okay, a normal person without training wouldn’t see how much beauty and effort there is in an algorithm or a legal contract as well.
The word "we" is doing a lot of heavy lifting here. A large majority of consumers can't even tell apart AI-generated from handmade, let alone care who or what made the thing.
I want progressive fees on copyright/IP/patent usage, and worldwide gov cooperation/legislation (and perhaps even worldwide ability to use works without obtaining initial permission, although let's not go into that outlandish stuff)
I want a scaling license fee to apply (e.g. % pegged to revenue. This still has an indirect problem with different industries having different profit margins, but still seems the fairest).
And I want the world (or EU, then others to follow suit) to slowly reduce copyright to 0 years* after artists death if owned by a person, and 20-30 years max if owned by a corporation.
And I want the penalties for not declaring usage** / not paying fees, to be incredibly high for corporations... 50% gross (harder) / net (easier) profit margin for the year? Something that isn't a slap on the wrist and can't be wriggled out of quite so easily, and is actually an incentive not to steal in the first place.)
[*]or whatever society deems appropriate.
[**]Until auto-detection (for better or worse) gets good enough.
IMO that would allow personal use, encourages new entrants to market, encourages innovation, incentivises better behaviour from OpenAI et al.
> And I want the world (or EU, then others to follow suit) to slowly reduce copyright to 0 years* after artists death if owned by a person, and 20-30 years max if owned by a corporation.
Why death at all?
It's icky to trigger soon after death, it's bad to have copyright vary so much based on author age, and it's bad for many works to still have huge copyright lengths.
It's perfectly fine to let copyright expire during the author's life. 20-30 years for everything.
Extremely naive to think that any of this could be enforced to any adequate level. Copyright is fundamentally broken and putting some plasters on it is not going to do much especially when these plasters are several decades too late.
I just tested it with Azure AI image classification and it worked - so this cat is yet to adapt to the mouse’s latest idea.
I still feel it is absolutely wrong to roam around the internet and scrape images (without consent) in order to power one’s cash cow AI. I hope more methods to protect artworks (including audio and other formats) become more accessible.
Artists copy from each other all the time. Arguably, culture exists because of copying (folk stories by necessity); copyright makes culture top-down and stagnant, and you can't avoid it because they have the money to shove it right in your face. Who wants trickle-down culture?
I might be missing something because I don't know much about the architecture of either Nightshade or AI art generators, but I wonder if you could try to have a GAN-like architecture (an extra model trying to trick the model) for the part of the generator that labels images to build resistance to Nightshade-like filters.
It doesn't even have to be a full GAN, you only need to train the discriminator side to filter out the data. Clean reference images + Nightshade would be the generator side.
What the article doesn't illustrate is that it destroys fine detail in the image, even in the thumbnails of the reference paper:
https://arxiv.org/pdf/2310.13828.pdf
Also... Maybe I am naive, but it seems rather trivial to work around with a quick prefilter? I don't know if tradition denoising would be enough, but worst case you could run img2img diffusion.
The poisoned images aren't intended to be viewed, rather scraped and pass a basic human screen. You wouldn't be able to denoise as you'd have to denoise the entire dataset, the entire point is that these are virtually undetectable from typical training set examples, but they can push prompt frequencies around at will with a small number of poisoned examples.
> the entire point is that these are virtually undetectable from typical training set examples
I'll repeat this point for clarity. After going over the paper again, denoising shouldn't affect this attack, it's the ability of plausible images to not be detected by human or AI discriminators (yet)
Long-term I think the real problem for artists will be corporations generating their own high quality targeted datasets from a cheap labor pool, completely outcompeting them by a landslide.
In the short-to-medium term, we're seeing huge improvements in the data efficiency of generative models. We haven't really started to see self-training in diffusion models, which could improve data efficiency by orders of magnitude. Current models are good at generalisation and are getting better at an incredible pace, so any efforts to limit the progress of AI by restricting access to training data is a speedbump rather than a roadblock.
Art is already democratized. It has been for decades. Everyone can pick it up at zero cost. Even you!
The poorest people have historically produced great art. Training a model, however? Expensive. Running it locally? Expensive. Paying the sub? Expensive.
Nothing is being democratized, the only thing this does is devaluing the blood and sweat people have put into their work so FAANG can sell it to lazy suckers.
This is fantastic. If companies want to create AI models, they should license the content they use for the training data. As long as there are not sufficient legal protections and the EU/Congress do not act, tools like these can serve as a stopgap and maybe help increase pressure on policymakers
It's going to be interesting to see how the lawsuits against OpenAI by content creators plays out. If the courts rule that AI generated content is a derivative work of all the content it was trained on it could really flip the entire gen AI movement on its head.
If it were a derivative work[1] (and sufficiently transformational) then it's allowed under current copyright law and might not be the slam dunk ruling you were hoping for.
"sufficiently transformational" is carrying a lot of water here. At minimum it would cloud the issue and might expose anyone using AI to lawsuits where they'd potentially have to defend each generated image.
Oh, interesting, I didn't realize that's how it worked. Thanks for the additional context around this. Guess it's not as upending as I thought it could be.
My biggest fear is that the big players will drop a few billion dollars to silence the copyright holders with power go away, and new rules are put in place that will make open-source models that can't do the same essentially illegal.
If the courts do rule that way, I would expect a legislative race between different countries to amend the relevant laws. Visual generative AI is just too lucrative a thing.
Google scrapes copyrighted material every day and then presents that material to users in the form of excerpts, images, and entire book pages. This has been ruled OK by the courts. Scraping copyrighted information is not illegal or we couldn't have search engines.
Google is not presently selling "we trained an AI on people's art without permission, and you can type their name in along with a prompt to generate a knockoff of their art, and we charge you money for this". So it's not really a 1:1 comparison, since there are companies selling the thing I described right now.
That pretty clearly would fall under transformative work. It is not illegal for a human to paint a painting in the style of, say, Banksy, and then sell the resulting painting.
In some locales sitting on the street writing down a list of people coming and going is legal, but leaving a camera pointed at the street isn't. Legislation like that makes a distinction between an action by a person (which has bounds on scalability) and mechanized actions (that do not).
Fair Use is the relevant protection and is not specific to manual creation. Traditional algorithms (e.g: the snippets, caching, and thumbnailing done by search engines) are already covered by it.
"The insights obtained through content analysis will not be used to re-create your content or lead to identifying any personal information."
"For Adobe Firefly, the first model is trained on Adobe Stock images, openly licensed content, and public domain content where the copyright has expired."
There is in fact, an extreme amount of circumstantial evidence that they intentionally and knowingly violated copyright en mass. It’s been quite a popular subject in tech news the past couple weeks.
Isn't this just teaching the models how to better understand pictures as humans do? As long as you feed them content that looks good to a human, wouldn't they improve in creating such content?
You would think the economists at UChicago would have told these researchers that their tool would achieve the opposite effect of what they intended, but here we are.
In this case, the mechanism for how it would work is effectively useless. It doesn't affect OpenAI or other companies building foundation models. It only works on people fine-tuning these foundation models, and only if the image is glazed to affect the same foundation model.
These methods like Glaze usually works by taking the original image chaging the style or content and then apply LPIPS loss on an image encoder, the hope is that if they can deceive a CLIP image encoder it would confuse also other models with different architecture, size and dataset, while changing the original image as little as possible so it's not too noticeable to a human eye. To be honest I don't think it's a very robust technique, with this one they claim that a model instead of seeing for example a cow on grass the model will see a handbag, if someone has access to GPT4-V I want to see if it's able to deceive actually big image encoders (usually more aligned to the human vision).
EDIT: I have seen a few examples with GPT-4 V and how I imagine it wasn't deceived, I doubt this technique can have any impact on the quality of the models, the only impact that this could potentially have honestly is to make the training more robust.
Each time there is an update to training algorithms and in response poisoning algorithms, artists will have to re-glaze, re-mist, and re-nightshade all their images?
Eventually I assume the poisoning artifacts introduced in the images will be very visible to humans as well.
Yeah, I've seen multiple artists complain about how glazing reduces image quality. It's very noticeable. That seems like an unavoidable problem given how AI is trained on images right now.
I'm glad to see tools like Nightshade starting to pop up to protect the real life creativity of artists. I like AI art, but I do feel conflicted about its potential long term effects towards a society that no longer values authentic creativity.
Is the existence of the AI tool not itself a product of authentic creativity? Does eliminating barriers to image generation not facilitate authentic creativity?
No, it facilitates commoditization. Art – real art – is fundamentally a human-to-human transaction. Once everyone can fire perfectly-rendered perfectly-unique pieces of 'art' at each other, it'll just become like the internet is today: filled with extremely low-value noise.
This is the right prediction. Once machines can generate visual art, people will simply stop valuing it. We may see increased interest in other forms of art, e.g., live performance art like theater. It's hard to predict exactly how it'll play out, but once something becomes cheap to produce and widely available, it loses its luster for connoisseurs and then gradually loses its luster for everybody else too.
So then you'll have curation to find the gems in that noise.
But it's still not clear why this is worse than the situation where not everyone can create perfectly-rendered pieces of whatever idea is in their head, and have to rely on others to do it for them, while being limited by what they can afford and what those others are willing to paint.
To protect an individual's image property rights from image generating AI's -- wouldn't it be simpler for the IETF (or other standards-producing group) to simply create an
AI image exclusion standard
, similar to "robots.txt" -- which would tell an AI data-gathering web crawler that a given image or set of images -- was off-limits for use as data?
Entities training models have no incentive to follow such metadata. If we accept the premise that "more input -> better models" then there's every reason to ignore non-legally-binding metadata requests.
Robots.txt survived because the use of it to gatekeep valuable goodies was never widespread. Most sites want to be indexed, most URLs excluded by the robots file are not of interest to the search engine anyway, and use of robots to prevent crawling actually interesting pages is marginal.
If there was ever genuine uptake in using robots to gatekeep the really good stuff search engines would've stopped respecting it pretty much immediately - it isn't legally binding after all.
>Entities training models have no incentive to follow such metadata. If we accept the premise that "more input -> better models" then there's every reason to ignore non-legally-binding metadata requests.
Name two entities that were asked to stop using a given individuals' images that failed to stop using them after the stop request was issued.
>Robots.txt survived because the use of it to gatekeep valuable goodies was never widespread. Most sites want to be indexed, most URLs excluded by the robots file are not of interest to the search engine anyway, and use of robots to prevent crawling actually interesting pages is marginal.
Robots.txt survived because it was a "digital signpost" a "digital sign" -- sort of like the way you might put a "Private Property -- No Trespassing" sign in your yard.
Most moral/ethical/lawful people -- will obey that sign.
Some might not.
But the some that might not -- probably constitute about a 0.000001% minority of the population, whereas the majority that do -- probably constitute about 99.99999% of the population.
"Robots.txt" is a sign -- much like a road sign is.
People can obey them -- or they can ignore them -- but they can ignore them only at their own peril!
It's a sign which provides a hint for what the right thing to do in a certain set of circumstances -- which is what the Law is; which is what the majority of Laws are.
People can obey them -- or they can choose to ignore them -- but only at their own peril!
Most will choose to obey them. Most will choose to "take the hint", proverbially speaking!
A few might not -- but that doesn't mean the majority won't!
>If there was ever genuine uptake in using robots to gatekeep the really good stuff search engines would've stopped respecting it pretty much immediately - it isn't legally binding after all.
Again, name two entities that were asked to stop using a given individuals' images that failed to stop using them after the stop request was issued.
And then what? The scrapers themselves already happily ignore copyright, they won't be inclined to obey a no-ai.txt. So someone would have to enforce the standard. Currently I see no organisation who would be willing to do this or even just technologically able - as even just detecting such scrapers is an extremely hard task.
Nevertheless, I hope that at some not-so-far point in the future there will be more legal guidance about this kind of stuff, i.e. it will be made clear that scraping violates copyright. This still won't solve the problem of detectability but it would at least increase the risk of scrapers, should they be caught.
>The scrapers themselves already happily ignore copyright, they won't be inclined to obey a no-ai.txt.
Name two entities that were asked to stop using a given individuals' images that failed to stop using them after the stop request was issued.
>Currently I see no organisation who would be willing to do this or even just technologically able - as even just detecting such scrapers is an extremely hard task.
// Part of Image Web Scraper For AI Image Generator ingestion psuedocode:
if fileExists("no-ai.txt") {
// Abort image scraping for this site -- move on to the next site
} else {
// Continue image scraping for this site
};
See? Nice and simple!
Also -- let me ask you this -- what happens to the intellectual property (or just plain property) rights of Images on the web after the author dies? Or say, 50 years (or whatever the legal copyright timeout is) after the author dies?
Legal grey area perhaps?
Also -- what about Images that exist in other legal jurisdictions -- i.e., other countries?
How do we know what set of laws are to apply to a given image?
?
Point is: If you're going to endorse and/or construct a legal framework (and have it be binding -- keep in mind you're going to have to traverse the legal jurisdictions of many countries, many countries!) -- you might as well consider such issues.
Also -- at least in the United States, we have Juries that can override any Law (Separation of Powers) -- that is, that which is considered "legally binding" -- may not be quite so "legally binding" if/when properly explained to a proper jury in light of extenuating (or just plain other) circumstances!
So kindly think of these issues prior to making all-encompasing proposals as to what you think should be "legally binding" or not.
I comprehend that you are just trying to solve a problem; I comprehend and empathize; but the problem might be a bit greater than you think, and there might be one if not serveral unexplored partial/better (since no one solution, legal or otherwise, will be all-encompassing) solutions -- because the problem is so large in scope -- but all of these issues must be considered in parallel -- or errors, present or future will occur...
> Part of Image Web Scraper For AI Image Generator ingestion psuedocode:...
Yes, and who is supposed to run that code?
> Name two entities that were asked to stop using a given individuals' images that failed to stop using them after the stop request was issued.
Github? OpenAI?[1] Stable Diffusion?[2] LAION?[3] What do you think why there are currently multiple high-profile lawsuits ongoing about exactly that topic?
Besides, that's not how things work. Training a foundation model takes months and currently costs a fortune in hardware and power - and once the model is trained, there is, as of now, no way to remove individual images from the model without restraining. So in practical terms it's impossible to remove an image if it has already been trained on.
So the better question would be, name two entities who have ignored an artist's request to not include their image when they encountered it the first time. It's still a trick question though because the point is that scraping happens in private - we can't know which images were scraped without access to the training data. The one indication that it was probably scraped is if a model manages to reproduce it verbatim - which is the basis for some of the above lawsuits.
And/or groups that don't want to risk getting sued? (Your: [1] [2] [3])?
>> Name two entities that were asked to stop using a given individuals' images that failed to stop using them after the stop request was issued.
>Github? OpenAI?[1] Stable Diffusion?[2] LAION?[3] What do you think why there are currently multiple high-profile lawsuits ongoing about exactly that topic?
Because:
a) (Some) American Lawyers (AKA "Bar Association Members") -- are Sue Happy?
b) Because various Governments / Deep States (foreign and domestic) / Dark Money Groups / Paid (and highly biased) Political Activists -- want to see if they can get new draconian laws (whilst believing their actions to be super-patriotic to their respective countries!) -- or at least court precedents that move in that direction -- passed?
(How much revenue will they be losing if their net income from artwork was $0? Also, wouldn't such high profile cases give the artists a ton of free advertising? The Defendant companies should counter-sue for giving the Plaintiff artists what amounts to free publicity for their artwork so great that they couldn't buy it with all of the Google advertising credits in the world!)
>Besides, that's not how things work. Training a foundation model takes months and currently costs a fortune in hardware and power - and once the model is trained, there is, as of now, no way to remove individual images from the model without restraining.
>"without retraining"...
Meditate on that one for a moment...
>So in practical terms it's impossible to remove an image if it has already been trained on.
In practical terms -- just retrain the model -- sans ("without") the encroaching images!
The models will need to be updated every couple of months anyway to include new public data from the web!
Create a list of images NOT to include in the next run (see above, "no-ai.txt" -- good suggestion incidentally!) -- and then don't include them" on the next run!
It's not Rocket Science! :-)
(Also, arguably Elon Musk doesn't think that "Rocket Science" is in fact as hard as "Rocket Science" is purported to be -- but that's a separate debate! <g>)
>So the better question would be, name two entities who have ignored an artist's request to not include their image when they encountered it the first time. It's still a trick question though because the point is that scraping happens in private - we can't know which images were scraped without access to the training data. The one indication that it was probably scraped is if a model manages to reproduce it verbatim - which is the basis for some of the above lawsuits.
Explain to me, from the point of view of an AI company, how that AI company is to know ahead of time NOT to include an image from the web? (And thus not break the law, copyright law at least, and thus not incur the lawsuits and all the chaos that will apparently follow such an act?)
How is the AI company supposed to know, ahead of time, that a given image on the web is not to be included?
How please?
Because you see, that's the root of the problem you are trying to solve.
In fact, let me ask you a better question...
How can an arbitrary Internet User -- not a big, legally powerful AI company, but an arbitrary small-fry Internet User -- know ahead of time, that a given Image, exposed to the public via the public Internet; the Web -- that the artist who created that image (or the intellectual/artistic property holder) -- does NOT want their Image to be used for specific purposes?
?
Because well, I don't know of any easily parsible, easily understandable standard for that on the Web currently...
So, to recap, the question is:
How is everybody (humans and machines) to know the unambiguous, easily parsable, easily understandable uses that the artist (or intellectual/artistic property) of an image -- wishes/wills for that image?
And how to easily know the unintended uses?
That might be a better definition of the problem that is trying to be solved...
Setting aside the efficacy of this tool, I would be very interested in the legal implications of putting designs in your art that could corrupt ML models.
For instance, if I set traps in my home which hurt an intruder we are both guilty of crimes (traps are illegal and are never considered self defense, B&E is illegal).
Would I be responsible for corrupting the AI operator's data if I intentionally include adversarial artifacts to corrupt models, or is that just DRM to legally protect my art from infringement?
edit:
I replied to someone else, but this is probably good context:
DRM is legally allowed to disable or even corrupt the software or media that it is protecting, if it detects misuse.
If an adversarial-AI tool attacks the model, it then becomes a question of whether the model, having now incorporated my protected art, is now "mine" to disable/corrupt, or whether it is in fact out of bounds of DRM.
So for instance, a court could say that the adversarial-AI methods could only actively prevent the training software from incorporating the protected media into a model, but could not corrupt the model itself.
None whatsoever. There is no right to good data for model training, nor does any contractual relationship exist between you and and a model builder who scrapes your website.
If you're assuming this is open-shut, you're wrong. I asked this specifically as someone who works in security. A court is going to have to decide where the line is between DRM and malware in adversarial-AI tools.
I'm not. Malware is one thin, passive data poisoning is another. Mapmakers have long used such devices to detect/deter unwanted copying. In the US such 'trap streets' are not protected by copyright, but nor do they generate liability.
A trap street doesn't damage other data. Not even remotely useful as an analogy. That's to allow detection of copies, not to corrupt the copies from being useable.
Sure it does. Suppose the data you want to publish is about the number of streets, or he average street length, or the distribution of street names, or the angles of intersections. Trap streets will corrupt that, even if it's just a tiny bit. Likewise, ghost imagery slipped into desirable imagery only slightly corrupts the model, but like the trap streets, that's the model-maker's problem.
You have a legal right to scrape data and use it as input into a model, you don't have a right to good data. It's up to you to sanitize it before training your model on it.
The way Nightshade works (assuming it does work) is by confusing the features of different tags with each other. To argue that this is illegal would be to argue that mistagging a piece of artwork on a gallery is illegal.
If you upload a picture of a dog to DeviantArt and you label it as a cat, and a model ingests that image and starts to think that cats look like dogs, would anybody claim that you are breaking a law? If you upload bad code to Github that has bugs, and an AI model consumes that code and then reproduces the bugs, would anyone argue that uploading badly written code to Github is a crime?
What if you uploaded some bad code to Github and then wrote a comment at the top of the code explaining what the error was, because you knew that the model would ignore that comment and would still look at the bad code. Then would you be committing a crime by putting that code on Github?
Even if it could be proven that your intention was for that code or that mistagged image to be unhelpful to training, it would still be a huge leap to say that either of those activities were criminal -- I would hope that the majority of HN would see that as a dangerous legal road to travel down.
No, it's much closer to (in fact, it is simply) asking if adversarial AI tools count as DRM or as malware. And a court is going to have to decide whether the model and or its output counts as separate software, which it is illegal for DRM to intentionally attack.
DRM can, for instance, disable its own parent tool (e.g. a video game) if it detects misuse, but it can't attack the host computer or other software on that computer.
So is the model or its output, having been trained on my art, a byproduct of my art, in which case I have a legal right to 'disable' it, or is it separate software that I don't have a right to corrupt?
> asking if adversarial AI tools count as DRM or as malware
Neither. Nightshade is not DRM or malware, it's "lying" about the contents of an image.
Arguably, Nightshade does not corrupt or disable the model at all. It feeds it bad data that leads the model to generate incorrect conclusions or patterns about how to generate images. This is assuming it works, which we'll have to wait and see, I'm not taking that as a given.
But the only "corruption" happening here is that the model is being fed data that it "trusts" without verifying that what the data is "telling" it is correct. It's not disabling the model or crashing it, the model is forming incorrect conclusions and patterns about how to generate the image. If Google translate asked you to rate its performance on a task, and you gave it an incorrect rating from what you actually thought its performance was, is that DRM? Malware? Have you disabled Google translate by giving it bad feedback?
I don't think the framing of this as either DRM or malware is correct. This is bad training data. Assuming it works, it works because it's bad training data -- that's why ingesting one or two images doesn't affect models but ingesting a lot of images does, because training a model on bad data leads the model to perform worse if and only if there is enough of that bad data. And so what we're really talking about here is not a question of DRM or malware, it's a question of whether or not artists have a legal obligation to make their data useful for training -- and of course they don't. The implications of saying that they did would be enormous, it would imply that any time you knowingly lied about a question that was being fed into an AI training set that doing so was illegal.
I see it as no different than mapmakers inventing a nonexistent alley, to check who copies their maps verbatim ("trap street"). Even if this caused, for example, a car crash because of an autonomous driver, the onus I think would be on the one that made the car and used the stolen map for navigation, and not on the one that created the original map.
I find the AI training topic interesting, because it's really data/information that is involved. Forget about the fact that it's images or stories or Reddit posts, it's all data.
We are born and then exposed to the torrent of data from the world around us, mostly fed to us by other humans, this is what models are trying to tap.
Unfortunately our learning process is completely organic and takes decades and decades and decades; there's no way to put a model through this easily.
Perhaps we need to seed the web with AI agents who converse and learn as much like regular human beings as possible and assemble the dataset that way. Although having an agent browse and find an image to learn to draw from is still gonna make people reee even if that's exactly what a young and aspiring human artist would be doing.
Don't talk about humans being sacred; we already voted to let corporations be people, for the 1% to exist and "lobby", breaking our democracy so that they can get tax breaks and make corrupt under the table deals. None of us stopped that from happening...
Does it survive AI upscaling or img2img? If not - then it's useless. Nobody trains AI models without any preprocessing. This is basically a tool for 2022.
For this to work, wouldn't you have to have an enormous number of artists collaborating on "poisoning" their images the same way (cow to handbag) while somehow keeping it secret form ai trainers that they were doing this?
It seems to me that even if the technology works perfectly as intended, you're effectively just mislabeling a tiny fraction of the training data.
1. They don't need an enormous number of artists; the research paper showed significant results with even 50 poisoned image samples in the dataset, which is enough to be contained in even a single artist's online gallery.
2. They don't need to keep it a secret; the goal is to remove these images from the training data, in a way that would be much more efficient than simply adding a "please don't include my art in your ai scraper" message next to your pictures.
In so far as anger goes against AIs being trained on particular intellectual properties.
A made up scenario¹ is that a person who is training an AI, goes to the local
library and checks out 600 books on art.
The person then lets the AI read all of them.
After which they are returned to the library and another 600 books are borrowed
Then we can imagine the AI somehow visiting a lot of museums and galleries.
The AI will now have been trained on the style and looks of a lot
of art from different artists
All the material has been obtained in a legal manner.
Is this an acceptable use?
Or can an artist still assert that the AI was trained with their
IP without consent?
Clearly this is one of the ways a human would go about learning
about styles, techniques etc..
¹ Yes you probably cannot borrow 600 books at a time.
How does the AI read the books? I dont know. Simplicity would be
that the researcher takes a photo of each page.
This would be extremmly slow but for this hypothetical it is acceptable.
I think the key difference here is that the most prominent image generation AIs are commercial and for-profit. The scenarios you describe are comparing a commercial AI to a private person. You cannot get a library card for a company, and you cannot bring a photography crew to a gallery without permission.
I’m completely flabbergasted by the number of comments implying copyright concepts such as “fair use” or “derivative work” apply to trained ML models. Copyright is for _people_, as are the entailing rights, responsibilities and exemptions.
This has gone far beyond anthropomorphising and we need to like get it together, man!
My initial interpretation was that you're saying fair use is irrelevant to the situation because machine learning models aren't themselves legal persons. But, fair use doesn't solely apply to manual creation - use of traditional algorithms (e.g: the snippets, caching, and thumbnailing done by search engines) is still covered by fair use. To my understanding, that's why ronsor pointed out that ML models are tools used by people (and those people can give a fair use defense).
Possibly you instead meant that fair use is relevant, but people are wording remarks in a way that suggests the model itself is giving a fair use defence to copyright infringement, rather than the persons training or using it?
Well then I could have been much clearer because I meant something like the latter.
An ML model can neither have nor be in breach of copyright so any discussion about how it works, and how that relates to how people work or “learn” is besides the point.
What actually matters is firstly details about collation of source material, and later the particular legal details surrounding attribution. The last part involves breaking new ground legally speaking and IANAL so I will reserve judgement. The first part, collation of source material for training is emphatically not unexplored legal or moral territory. People are acting like none of the established processes apply in the case of LLMs and handwave about “learning” to defend it.
> and how that relates to how people work or “learn” is besides the point
It is important (for the training and generation stages) to distinguish between whether the model copies the original works or merely infers information from them - as copyright does not protect against the latter.
> The first part, collation of source material for training is emphatically not unexplored legal or moral territory.
Similar to as in Authors Guild v. Google, Inc. where Google internally made entire copies of millions of in-copyright books:
> > While Google makes an unauthorized digital copy of the entire book, it does not reveal that digital copy to the public. The copy is made to enable the search functions to reveal limited, important information about the books. With respect to the search function, Google satisfies the third factor test
Or in the ongoing Thomson Reuters v. Ross Intelligence case where the latter used the former's legal headnotes for training a language model:
> > verbatim intermediate copying has consistently been upheld as fair use if the copy is "not reveal[ed] to the public."
That it's an internal transient copy is not inherently a free pass, but it is something the courts take into consideration, as mentioned more explicitly in Sega v. Accolade:
> > Accolade, a commercial competitor of Sega, engaged in wholesale copying of Sega's copyrighted code as a preliminary step in the development of a competing product [yet] where the ultimate (as opposed to direct) use is as limited as it was here, the factor is of very little weight
And, given training a machine learning model is a considerably different purpose than what the images were originally intended for, it's likely to be considered transformative; as in Campbell v. Acuff-Rose Music:
> > The more transformative the new work, the less will be the significance of other factors
Listen, most website and book-authors want to be indexed by google. It brings potential audience their way, so most don’t make use of their _right_ to be de-listed. For these models, there is no plausible benefit to the original creators, and so one has to argue they have _no_ such right to be “de-listed” in order to get any training data currently under copyright.
> It brings potential audience their way, so most don’t make use of their _right_ to be de-listed.
The Authors Guild lawsuit against Google Books ended in a 2015 ruling that Google Books is fair use and as such they don't have a right to be de-listed. It's not the case that they have a right to be de-listed but choose not to make use of it.
The same would apply if collation of data for machine learning datasets is found to be fair use.
> one has to argue they have _no_ such right to be “de-listed” in order to get any training data currently under copyright.
Datasets I'm aware of already have respected machine-readable opt-outs, so if that were to be legally enforced (as it is by the EU's DSM Directive for commercial data mining) I don't think it'd be the end of the world.
There's a lot of power in a default; the set of "everything minus opted-out content" will be significantly bigger than "nothing plus opted-in content" even with the same opinions.
With the caveat that I was exactly wrong about the books de-listing, I feel you are making my point for me and retreating to a more pragmatic position about defaults.
The (quite entertaining) saga of Nightshade tells a story about what is going to be content creators “default position” going forward and everyone else will follow. You would be a fool not to, the AI companies are trying to end run you, using your own content, and make a profit without compensating you and leave you with no recourse.
> I feel you are making my point for me and retreating to a more pragmatic position about defaults
I'm unclear on what stance I've supposedly retreated from. My position is that an opt-out is not necessary under current US law, but that it wouldn't be the worst-case outcome if new regulation were introduced to mandate it.
> The (quite entertaining) saga of Nightshade tells a story about what is going to be content creators “default position” going forward and everyone else will follow
By "default" I refer not to the most common choice, but to the outcome that results from inaction. There's a bias towards this default even if the majority of rightsholders do opt to use Nightshade (which I think is unlikely).
Oh come on, you’re being insincere. Wether or not the model is learning from the work just like people is hotly debated as if it would make a difference. Fair use is even brought up. Fair use! Even if it applied, these training sets collate all of everything
I really don't understand the anxiety of artists towards AI - as if creatives haven't always borrowed and imitated. Every leading artist has had acolytes, and while it's true no artist ever had an acolyte as prodigiously productive as AI will be, I don't see anything different between a young artist looking to Picasso for cues and Stable Diffusion or DALL-E doing the same. Styles and methods haven't ever been subject to copyright - and art would die the moment that changed.
The only explanation I can find for this backlash is that artists are actually worried just like the rest of us that pretty soon AI will produce higher quality more inventive work faster and more imaginatively than they can - which is very natural, but not a reason to inhibit an AI's creative education.
This has been litigated over and over again, and there have been plenty of good points made and concerns raised over it by those who it actually affects. It seems a little bit disingenuous (especially in this forum) to say that that conclusion is the "only explanation" you can come up with. And just to avoid prompting you too much: trust me, we all know or can guess why you think AI art is a good thing regardless of any concerns one might bring up.
Could you please stop posting unsubstantive comments and flamebait? You've unfortunately been doing it repeatedly. It's not what this site is for, and destroys what it is for.
I genuinely love this site to be clear and appreciate all the work that you do around here - but the agenda of the site is to service Y Combinator Management, LLC.
Oh yes, but the question is how best to do that. The answer is that we can maximize HN's value for YC simply by making the site as good as possible, because that's what makes the community as happy as possible, and that's what has value. I've written about this many times in case you or anyone want to read more: https://hn.algolia.com/?dateRange=all&page=0&prefix=false&so....
Trying to make HN as good as possible has many aspects, but one of them is the attempt to keep discussions at least a little more substantive than the level they would normally degenerate to without effort, and that means avoiding flamewar comments, fulmination, etc., as described in https://news.ycombinator.com/newsguidelines.html.
Imitation isn’t the problem so much as it is that ML generated images are composed of a mush of the images it was trained on. A human artist can abstract the concepts underpinning a style and mimic it by drawing all-new lineart, coloration, shading, composition, etc, while the ML model has to lean on blending training imagery together.
Furthermore there’s a sort of unavoidable “jitter” in human-produced art that varies between individuals that stems from vastly different ways of thinking, perception of the world, mental abstraction processes, life experiences, etc. This is why artists who start out imitating other artists almost always develop their imitations into a style all their own — the imitations were already appreciably different from the original due to the aforementioned biases and those distinctions only grow with time and experimentation.
There would be greatly reduced moral controversy surrounding ML models if they lacked that mincemeat/pink slime aspect.
I love it. This undermines the notion of ground truth. What separates correct information from incorrect information? Maybe nothing! I love how they acknowledge the never ending attack versus defense game. In stark contrast to "our AI will solve all your problems".
No, it's resistant to transformation. Rotation, cropping, scaling, the image remains poisonous. The only antidote known currently is active artist cooperation.
I think it's worthwhile for such discussion to happen in the open. If the tool can be defeated through simple means, it's better for everybody to know that, right?
It would be better for artists to know that. But Hacker News is not a forum of visual artists: it's a forum of hackers, salaried programmers, and venture capitalists. Telling the bad guys about vulnerabilities isn't responsible disclosure.
Causing car crashes isn't hard (https://xkcd.com/1958/). That doesn't mean Car Crash™ International®'s decision-makers know how to do it: they probably don't even know what considerations go into traffic engineering, or how anyone can just buy road paint from that shop over there.
It's everybody's responsibility to keep Car Crash™ International® from existing; but failing that, it's everybody's responsibility to not tell them how to cause car crashes.
Who said anything about creating a derivative? Surely you don't mean to say that any image created with a model trained on copyrighted data counts as a derivative of it. Edit: Or worse, that the model itself is derivative, something so different from an image must count as transformative work!.
> it’s basically the input data compressed with highly lossy compression.
Okay, extract the images from a Stable Diffusion checkpoint then. I'll wait.
It's not like lossy compression CAN'T be fair use or transformative. I'm sure you can imagine how that is possible given the many ways an image can be processed.
> All the corporations that are offering AI as a paid service?
I'm at a food truck and you happen to see and memorize all my CC data and use it because it was in public so it was fair game. Or I was paying using apple pay on the most expensive iPhone available so you take that because it was out in public and if I didn't want anyone grabbing it I shouldn't have brought it out.
Your phone example is just theft unless I'm misreading...do you want to go down the path of calling copyright infringement theft?
As for the card example it is on you to keep secrets secret. It is clearly a fault with how credit cards work and for things like that you should get insurance. Keep your card safe, or better yet, use cash. We can call it theft, doesn't matter to my point.
There is no "taking" or "grabbing" anything, I visited your website and was served a copy of the data because you made it do so. You wanted it to happen, for the public to see it, that was your goal. You expected it to happen. Do you disagree that me acquiring a copy was consensual? If so it is very different from the hypotheticals you're posing don't you think?
Once a copy is in my possession you would need to initiate violence to stop me from training a model on it which puts you on the wrong side of the moral line.
Is it possible to reliably detect whether an image is poisoned? If not then it achieves the goal of punishing entities which indiscriminately harvest data.
You can use older images, collected from before the "poisoning" software was released. Then you don't have to.
This, of course, assumes that "poisoning" actually works. Glaze and Nightshade and similar are very much akin to the various documented attacks on facial recognition systems. The attack does not exploit some fundamental flaw in how the systems work, but specific characteristics in a given implementation and version.
This matters because it means that later versions and models will inevitably not have the same vulnerabilities. The result is that any given defensive transformation should be expected to be only narrowly effective.
Only protection is adding giant gaping vaginas to your art, nothing less will deter scraping. If the Email spam community showed us something in the last 40 years is that no amount of defensive tech measures will work except financial disincentives.
Any AI art/video/photography/music/etc generator company who generates revenue needs to add watermarks to let the public know its AI generator. This should be forced via legislation in all countries.
If they don't then whatever social network or other services where things can shared/viewed by large groups to millions & are posted publicly need to be labeled "We can not verify veracity of this content."
I want a real internet ..this AI stuff is just triple fold increasing fake crap on the Internet and in turn / time our trust in it!
For visual artists who don't want visible artifacting in the art they feature online, would it be possible to upload these alongside your un-poisoned art, but have them only hanging out in the background? So say having one proper copy and a hundred poisoned copies in the same server, but only showing the un-poisoned one?
Might this "flood the zone" approach also have -some- efficacy against human copycats?
I wonder how this tool works if it's actually model independent. My understanding so far was that in principle each possible model has some set of pathological inputs for which the classification will be different than what a user sees - but that this set is basically different for each model. So did they actually manage to build an "universal" poison? If yes, how?
I wonder if this is illegal in some countries. In France for example, there is the following law: "Obstructing or distorting the operation of an automated data processing system is punishable by five years' imprisonment and a fine of €150,000.".
If you ask me, this is 100% applicable in this case, so I wonder what a judge would rule.
Remember when the music industry tried to use technology to stop music pirating?
This will work about as well...
Oh, I forget, fighting music pirating was considered an evil thing to do on HN. "pirating is not stealing, is copyright infringement", right? Unlike training neural nets on internet content which of course is "stealing".
FWIW, you're the only use of the word "steal" in this comment thread.
Many people would in fact argue that training AI on people's art without permission is copyright infringement, since the thing it (according to detractors) does is infringe copyright by generating knockoffs of people's work.
You will see some people use the term "stealing" but they're usually referring to how these AIs are sold/operated by for-profit companies that want to make money off artists' work without compensating them. I think it's not unreasonable to call that "stealing" even if the legal definition doesn't necessarily fit 100%.
The music industry is also not really a very good comparison point for independent artists... there is no Big Art equivalent that has a stranglehold on the legislature and judiciary like the RIAA/MPAA do.
The difference is that “pirating” is mostly done by individuals for private use, whereas training is mostly done by megacorporations looking to make more money.
Musicians overwhelmingly do not even attempt to clear samples. This also isn't a great comparison since samples are taken directly out of the audio, not turned into a part of a pattern used to generate new sounds like what AI generators do with images
Entire legal firm empires have been built on the licensing, negotiations, and fees that make up the industry.
I'm ain't talking about some dude on YouTube or Soundcloud. Few people care about some rando on Soundcloud. Those moles aren't big enough to whack. Vanilla Ice and MC Hammer were. OpenAI is as well.
I wonder if we know enough about any of these systems to make such claims. This is all predicated on the fact that this tool will be in widespread use. If it is somehow widely used beyond the folks who have seen it at the top of HN, won't the big firms have countermeasures, ready to deploy?
It's an arms race the bigger players will win, and it undermines the quality of the images. But it feels natural that artists would want to do something since they don't feel like anyone else is protecting them right now.
The intention is good, from an AI-opponent's perspective. I don't think will work practically, though. The drawbacks for actual users of the image galleries, plus the level of complexity involved in poisoning the samples makes this unfeasible to implement at the scale required.
The opening website is so poor - "what is nightshade" - then a whole paragraph that tells nothing, then another paragraph.. then no examples. This whole description should be reworked to be shorter and more to the point.
The image generation models now are at the point where they can produce their own synthetic training images. So I'm not sure how big of an impact something like this would have.
would it have been that hard to include a sample photo and how it looks with the nightshade filter side by side in a 3 page document describing how it would look in great detail
Baffling to see anyone argue against this technology when it is a non-issue to any model by simply acquiring only training data you have permission to use.
The reason people are arguing against this technology is that no one is using them in the way you describe. They actually wouldn't even be economically viable in that case.
If it is not economically viable for you to be ethical, then you do not deserve economic success.
Anyone arguing against this technology following the line of reasoning you present is operating in adverse to the good of society. Especially if their only motive is economic viability.
I feel like you read my comment and interpreted it in exactly the opposite way it was intended because I agree with you, and you're making the same point I was trying to make.
I think people 100% have the right to use this on their images, but:
> simply acquiring only training data you have permission to use
Currently it's generally infeasible to obtain licenses at the required scale.
When attempting to develop a model that can describe photos for visually impaired users, I had even tried to reach out to obtain a license from Getty. They repeatedly told me that they don't license images for machine learning[0].
I think it's easy to say "well too bad, it doesn't deserve to exist" if you're just thinking about DALL-E 3, but there's a huge number of positive and far less-controversial applications of machine learning that benefit from web-scale pretraining and foundation models - spam filtering, tumour segmentation, voice transcription, language translation, defect detection, etc.
I don't believe it's a "doesn't deserve to exist" situation, because these things genuinely can be used for the public good.
However - and this is a big however - I don't believe it deserves the legal protection to be used for profit.
I am of the opinion that if you train your model on data that you do not hold the rights for, your usage should be handled similarly to most fair use laws. It's fine to use it for your personal projects, for research and education, etc. but it is not OK to use it for commercial endeavors.
> It's fine to use it for your personal projects, for research and education, etc. but it is not OK to use it for commercial endeavors.
Say I train a machine vision model that, after having pretrained on ImageNet or similar, detects deformities in a material for a small company that manufactures that material. Do you not think that would be fair use, despite being commercial?
To me it seems highly transformative (a defect detection model is entirely outside the original images' purposes) and does not at all impact the market of the works.
Moreover, you said it was "Baffling to see anyone argue against this technology" but it seems there are at least some models (like if my above detector was non-commercial) that you're ethically okay with and could be affected by this poisoning.
No, I don't generally think it's okay to profit off of the work of others without their consent.
>Moreover, you said it was "Baffling to see anyone argue against this technology" but it seems there are at least some models (like if my above detector was non-commercial) that you're ethically okay with and could be affected by this poisoning.
Just because I think there are situations where it's not ethically wrong to use someone's work without permission does not mean I think it's ethically wrong for someone to protect their work any way they see fit.
To use an extreme example: I do not think it's wrong for a starving man to steal food. I also do not think it's wrong for someone to defend their food from being stolen, regardless of the morality of the thieves' motivation.
> No, I don't generally think it's okay to profit off of the work of others without their consent.
I'd argue that essentially all work based off the work of others. I can only draw a "car" better than a random guess because I've seen many (individually copyrighted) car designs.
That's not to say we inherently have to treat use of statistical models the same, but rather that there does have to be a line somewhere to define where a new work, while dependant on previous works in aggregate, is sufficiently transformative - not carrying substantial similarity to any particular existing work that played a part in its creation - and can therefore be used by the author to make a living.
That line has to be placed in a way that prioritizes the progress of sciences and useful arts, rather than just enriching rightsholder megacorps like Getty Images/Universal Music. It should certainly allow training something like a tumor segmentation network, rather than rendering it infeasible.
Also, while whether it's morally okay is relevant and worth discussing, I think the question still stands of whether you believe my example would count as fair use, given the transformative nature and lack of impact on the market of the original work.
> Just because I think there are situations where it's not ethically wrong to use someone's work without permission does not mean I think it's ethically wrong for someone to protect their work any way they see fit.
> To use an extreme example: I do not think it's wrong for a starving man to steal food. I also do not think it's wrong for someone to defend their food from being stolen, regardless of the morality of the thieves' motivation.
I personally agree that they have a right to do so, but I don't think it'd be "baffling" that the [starving man/person training a tumor detector] would be against it, and it's likely not a "non-issue" for them to obtain sufficient [food/data] through other means.
Particularly since there are already means to opt-out that are respected by scrapers, and this is instead an attempt to do active damage. I guess in the analogy it'd be leaving out poisoned bread, although that's more extreme than I intend.
I think the artists need to agree to stop making art altogether. That ought to get people’s attention. Then the AI people might (be socially pressured or legally forced to) put their tools away.
No, they'll just demand that artists produce more art so they can continue scraping, because if you work in tech you're allowed to be entitled, you're The Face Of The Future and all you're trying to do is Save The World, all these decels are just obstacles to be destroyed.
However much we might wish that it was not true, ideas are not rivalrous. If you share an idea with another person, they now have that idea too.
If you share words on paper, then someone with eyes and a brain might memorize them (or much more likely, just grasp and retain the ideas conveyed in the words).
If you let someone hear your music, then the ideas (phrasing, style, melody, etc) in that music are transferred.
If you let people see a visual work, then the stylistic and content elements of that work are potentially absorbed by the audience.
We have copyright to protect specific embodiments, but mostly if you try to share ideas with others without letting them use the ideas you shared, then you are in for a life of frustration and escalating arms race.
I completely sympathize with anyone who had a great idea and spent a lot of effort to realize it. If I invented/created something awesome I would be hurt and angry if someone “copied” it. But the hard cold reality is that you cannot “own” an idea.
> But the hard cold reality is that you cannot “own” an idea.
The above comment is true about the properties of information, as explained via the lens of economics. [1]
However, one ignores ownership as defined by various systems (including the rule of law and social conventions) at one's own peril. Such systems can also present a "hard cold reality" that can bankrupt or ostracize you.
[1] Don't let the apparent confidence and technicality of the language of economists fool you. Economics isn't the only game in town. There are other ways to model and frame the world.
[2] Dangling footnote warning. I think it is instructive to recognize that the field of economics has historically shown a kind of inferiority complex w.r.t. physics. Some economists ascribe to the level of rigor found in physics and that is well and good, but perhaps that effort should not be taken too seriously nor too far, since economics as a field operates at a different level. IMO, it would be wise for more in the field to eat a slice of humble pie.
[3] Ibid. It is well-known that economists can be "hired guns" used to "prove" a wide variety of things, many of which are subjective. My point: you can hire an economist to shore up one's political proposals. Is the same true of physicists? Hopefully not to the same degree. Perhaps there are some cases of hucksterism, but nothing like the history of economists-wagging-the-dog! At some point, the electron tunnels or it does not.
By that metric, various economic schools have been hilariously inept and would get classified not dissimilar to various schools of religious theology with their own dogmas. It's only in the last 15 years or so that some focus on empiricism and explaining reality rather than building theoretical castles in the air is coming about and is still far from mainstream.
There is no need to frame this as "winning versus losing" regarding the many models that we draw upon.
Even when talking about various kinds of scientific and engineering fields, predictive power isn't the only criteria, much less the best. Sometimes the simpler, less accurate models work well enough with less informational and computational cost.
Even if we focus on prediction (as opposed to say statistical inference), often people want some kind of hybrid. Perhaps a blend of satisficing with limited information, scoped action spaces, and bounded computation; i.e. good enough given the information we have to make the decisions we can actuate with some computational budget.
Many terms of art from economics are probably not widely-known here.
> In economics, a good is said to be rivalrous or a rival if its consumption by one consumer prevents simultaneous consumption by other consumers, or if consumption by one party reduces the ability of another party to consume it. - Wikipedia: Rivalry (economics)
Also: we should recognize that stating something as rivalrous or not is descriptive (what exists) not normative (what should be).
I'm either not understanding you or disagreeing. You seem to be saying that something should be because it is? Saying that would be rather silly, as in "Electrons should repel each other because they repel each other." Not to mention that this claim runs amok of the naturalistic fallacy. So what are you driving at?
But from there my point was that, in order to ask what "should be", you must ask what can be. I don't think there is a way for ideas to be non-rivalrous, period.
So sure, saying ideas are not rivalrous is descriptive, but there isn't really another option.
We're not trying to keep the AI from learning general ideas, we're trying to keep it from memorizing specific expressions[0]. There's a growing body of research to show that these models are doing a lot of memorizing, even if they're not regurgitating that data. For example, Google's little "ask GPT to repeat a word forever" trick, which will make GPT-4 spit out verbatim training data[1].
If there was a training process that let us pick a minimal sample of examples and turn it into a general purpose art generator or text generator, I think people would have been fine with that. But that's not what any of these models do. They were trained on shittons of creative expression, and there's statistical evidence that the models retain that expression, in a way that is fundamentally different from how humans remember, misremember, adapt, remix, and/or "play around with" other people's creativity.
[0] You called these "embodiments", but I believe you're trying to invoke the idea/expression divide, so I'll run with that.
[1] Or at least it did. OpenAI now filters out conversations that trip the bug.
I don't see the parallel between this offensive tool and DRM. I could, say buy a perpetual license to an image from the artist, so that I can print it and put it on my wall, while it can simultaneously be poisonous to an AI system. I can even steal it and print it, while it is still poisonous to an AI system.
The closest parallel I can think of is that humans can ingest chocolate but dogs should not.
What you've described is the literal, dictionary definition of Digital Rights Management - a technology to restrict the use of a digital asset beyond the contractually-agreed terms. Copying is only one of many uses that the copyright-holder may wish to prevent. The regional lockout on a DVD had nothing to do with copy-protection, but it was still DRM.
It's about the arm's race: DRM will always be cracked (with a sufficiently motivated customer.) AI poisoning will always be cracked (with a sufficiently motivated crawler.)
This doesn’t stop anyone from viewing or scraping the work, though, so in no way is it DRM. It just causes certain methods of computer interpretation of an image to interpret it in an odd way vs. human viewers. They can still learn from them.
>Nightshade's goal is not to break models, but to increase the cost of training on unlicensed data, such that licensing images from their creators becomes a viable alternative.
Which feels similar to DRM. To discourage extraction of assets.
Sure. Just like how video game drm impacts performance and watermarks on images degrades the image. Drm walks a tight line that inevitably makes the result worse than a drum-free solution, but also should not make the item completely unconsumable.
So, do you want to define drm by intent or technical implementation? I'm doing the former, but it sounds like you want to do the latter. Also keep in mind that legalese doesn't necessarily care about the exact encryption technique to be deployed either.
Both. Changing an image is done all the time prior to publishing them. In fact, no image you ever see on the internet is a raw sensor output. They are all modified in some manner. The images processed using this method look the same to every person and computer that views them. That’s very different from DRM which encrypts things and prevents access to unprivileged users.
This is effectively the equivalent of someone doing really crappy image processing. As other commenters have mentioned, it does alter how images look to humans as well as machines, and it can be “mitigated” through additional processing techniques.
>That’s very different from DRM which encrypts things and prevents access to unprivileged users.
Well you can call it a captcha if you want. The point here is to make it harder to access for bots (but not impossible) while inconveniencing honest actors in the process. It doesn't sound like there's a straightforward answer to "are captchas DRM" either.
That's true of almost all DRM, isn't it? Even for the most annoying form that is always-online DRM, everyone is provided the same access to the bytes that form a game. You and I have the same bytes of game.
It's the purpose of some of those bytes that turns it into DRM.
No, it's not the same. The game is non-functional without the proper keys/authorization whereas images run through this algorithm are still images that anyone and any computer can view in the same manner without any authentication.
An analogy that springs to mind is the difference between an access control mechanism such as a door with lock and key versus whatever magical contrivance that prevents entry to a dwelling by vampires uninvited.
That may be an ok analogy, but magic isn’t real. Maybe a better analogy would be to speak or write in a language that you know certain people don’t natively understand, and using lots of slang and idioms that don’t translate easily. Someone could still run it through Google Translate or whatever, but they won’t get a great understanding of what you actually said. They’d have to actually learn the language and the sorts of slang and idioms used.
I agree that the analogy is strained. My goal was in elucidating the distinction between:
the goals of artists and the developers of OP,
versus
the goals of AI engineers,
and how it seems to me similar to the is/ought disctinction.
In my original analogy, it’s generally considered lawful to have a lock on your door, or not to do so, and the issue of a lock or lack thereof is moot when one is invited to enter, just as it is lawful to enter a premises when invited or during exigent circumstances, such as breaching entry to render lifesaving aid by emergency services or firefighters.
By that same token, no amount of locks or other barriers to entry will prevent ingress by a vampire once invited inside.
To me, much of the ballyhoo about OP seems like much ado about big cats eating faces, like a person publicly decrying the rise in vampire attacks after inviting that same vampire inside for dinner. It’s a nonstarter.
Copyright law is broken, because of the way that the law is written as much as the way that it’s enforced, and also broken because of the way that humans are. Ideas are not copyrightable, and while historically their implementations or representations were, going forward, neither implementations nor representations are likely to receive meaningful/effective protections from copyright itself, but only from legal enforcement of copyright law.
After the recent expiry of Disney’s copyright on Steamboat Willie, the outpouring of praise, support, excitement, and original work from creators shows me that copyright law in its current incarnation doesn’t perform its stated goals of promoting the creation of arts and sciences, and so should be changed, and in the meantime ignored and actively disobeyed, as any unjust law ought to be, regardless of what the law is or does.
In light of our obligation to disobey unjust laws, I applaud efforts like OP to advance the state of the art in computer science, while at the same time encouraging others working on AI to actively circumvent such efforts for the selfsame reason.
I similarly encourage artists of all kinds to make art for art’s sake while monetizing however they see fit, without appealing to red herrings like the legality or lack thereof of end users appreciating their art and incorporating it into their own artistic output, however they may choose to do so.
Like all art, code and its outputs is also First Amendment protected free speech.
No, I disagree. There is no principle of the universe or across human civilizations that says that you have a right to eat because you produced a creative work.
The way societies work is that the members of the society contribute and benefit in prescribed ways. Societies with lots of excess production may at times choose to allow creative works to be monetized. Societies without much surplus are extremely unlikely to do so, eg a society with not enough food for everyone to eat in the middle of a famine is extremely unlikely to feed people who only create art; those people will have to contribute in some other way.
I think it is a very modern western idea (less than a century old) that many artists can dedicate themselves solely to producing the art they want to produce. In all other times artists either had day jobs or worked on commission.
That doesn't follow; you can say an item is not in a set without writing out every member of that set. What principle do you claim exists to contradict that claim?
> you can say an item is not in a set without writing out every member of that set
Of course you can. Anyone can say anything.
Is “keeping the list of principles a secret” a principle like the rules of Fight Club? It is not unreasonable to ask for a link or summary of this immutable set of ground truths.
> What principle do you claim exists to contradict that claim?
I could not answer this question without being able to double check. The only principle that comes to mind is the principle of ligma
> There is no principle of the universe or across human civilizations that says that you have a right to eat because you produced a creative work.
What does that have to do with rivalry? This doesn't dispute the idea that AI is indeed competing with artists. You're just saying artists don't deserve to get paid.
Regardless, some artists will give up but some will simply be more careful with where and how they post their art with tools like these. AI doesn't have a right to the artist's images neither.
I used to identify as a copyright abolitionist (I really love Nina Paley's TED talk, "copyright is brain damage") but the more I look at the history of it I see the compromises of interests, copyright is there so art is not locked up between artists and their patrons.
The tragedy of "your business model is not my problem" as a spreading idea is that while you're right since distribution is where the money is (not creation), intellectual property is de-facto weakened today and IP piracy is widely considered an acceptable thing.
So is sabotaging solutions that would make creative work of the same (or superior) quality more affordable. Your ability to produce expensive illustrations hinders my ability to produce cheap textbooks.
Not everybody equates automated scraping for training models and human experience. Just like any other “data wants to be free” type of discussion, the philosophical and ethical considerations are anything but cut-and-dried, and they’re far more consequential than the technical and economics-in-a-vacuum ones. The general public will quite possibly see things differently than the “oh well, artists— that’s the free market for ya, and you lost” crowd.
You don't copyright ideas, you copyright works. And these artists' productions are works, not abstract ideas, with copyrights, and they are being violated. This is simple law. Why do people have such a hard time with this? Are you the one training the models and you need to find a cognitive escape out of the illegality and wrong-doing of your activities?
It’s not obvious to me that using a copyrighted image to train a model is copyright infringement. It’s certainly not copyright infringement when used to train a human who may end up creating works that are influenced by (but not copies of) the original works.
Now, if the original copyrighted work can be extracted or reproduced from the model, that’s obviously copyright infringement.
If OpenAI's output reproduces a copyrighted image with one pixel changed, is that valid in your view? Where does the line end?
Copyrighted material should never be used for nonacademic language models. "Garbage in, garbage out." All results are tainted.
"But being forced to use non-copyrighted works will only slow things down!"
Maybe that's a good thing, too. Copyright is something every industry has to accept and deal with -- LLMs don't get a "cool tech, do whatever" get-out-of-jail free card.
I'm totally down for the courts handling this AI/copyright mess, but I don't think technologists are going to like the results.
By virtue of the fact that it is "fuzzy" and open to interpretation, we're going to see lawsuits, the resulting chilling effects of those lawsuits will blunt US tech firms from the practice of ingesting large amounts of copywritten material without a second thought. US tech firms will be giving it a second, third, fourth, etc. thought once the lawsuits start.
It's gonna be like submarine patents on steroids.
Like I said, I'm down for letting the courts decide. But AI supporters should probably avoid kicking the hornets' nests regarding copyright.
>Now, if the original copyrighted work can be extracted or reproduced from the model, that’s obviously copyright infringement.
I think there's an important distinction to be made here - "can" be reproduced isn't infringement, only actual reproduction is (and degrees thereof not consisting of sufficiently transformative or fair use).
Trivially a typewriter can reproduce a copyrighted book. Less trivially Google books, with iirc stores the full text of copywrited works has been judged to be legal.
>This is simple law. Why do people have such a hard time with this?
Because this isn’t simple law. It feels like simple infringement, but there’s no actual copying going on. You can’t open up the database and find a given duplicate of a work. Instead you have some abstraction of what it takes to get to a given work.
Also it’s important to point out that nothing in the law is sure. A good lawyer, a sympathetic judge, a bored/interested/contrarian juror, etc can render “settled law” unsettled in an instant. The law is not a set of board game rules.
If the AI were a human and that human made an image that copied substantial elements of another human's creative work after a careful review of the original creator's work, even if it was not an original copy and no archival copy was stored somewhere in the second creator's creative space, I would be concerned about copyright infringement exposure if I were the second (copying) creator.
I'm open to the idea that copyright law might need to change, but it doesn't seem controversial to note that scraping actual creative works to extract elements for an algorithm to generate new works crosses a number of worrying lines.
Have you seen the examples of midjourney reproducing exact frames of Dune, Star Wars etc? With vague prompting not asking for the media property specifically. It's pretty close to querying a database, except if you're asking for something that's not there it's able to render an interpolated result on the fly. Ask it for something that is there however and the model will dutifully pull it up.
Looking for it, I found this [1] which describes almost what you're saying. The key difference here is that these images aren't exact frames. They're close, of course, but close is not identical.Is there another instance you can point to that shows what you've described?
that's a good reference and yes that's what i'm talking about, but i mixed up the fact that you can get simpsons and star wars without asking for it by name - the "find the difference between these two pictures" game is the result of asking for a specific movie. I stand by my point tho that this is not substantially different than querying a database for copywrited material
“One may well ask: ‘How can you advocate breaking some laws and obeying others?’ The answer lies in the fact that there are two types of laws: just and unjust. I would be the first to advocate obeying just laws. One has not only a legal but a moral responsibility to obey just laws. Conversely, one has a moral responsibility to disobey unjust laws. I would agree with St. Augustine that ‘an unjust law is no law at all.’”
Well, fine, but you'll have to claim that copyright is unjust and that you are breaking the law as an act of civil disobedience. The AI corps do not want to take this stance as they also have intellectual property to protect. Classic "have their cake and eat it too" scenario.
Imagine if every book or advertisement or public conversation you overheard led to future claims that you had unethically learned from public information. It’s such a weird worldview.
(BTW I forbid you from using my comment here in your future reasoning)
Someone claiming they created something which I created is the closest to harm on that list. Fraud should be punished.
The others aren't harmful, unless you're defining harm to include the loss of something which someone believes they are entitled to, a concept which is fraught with problems.
Creating an image (or any non-physical creation) doesn't obligate the world to compensate you for your work. If you choose to give it away by posting it on the internet, that's your choice, but you are entitled to nothing.
>I'm not convinced that most copyright infringements are immoral regardless of their legal status.
You're right and wrong. You're right because most infringement is from people who can do minimal damage and in fact so more help by giving awareness to your works by sharing. But this is only becsuse copyright it working (most of the time) against corporate entities who don't want to leave any room for legalities to come in.
If copyright ended I'd bet my bottom dollar Disney and all the other billionaires companies would be spamming any and everything that gets moderately popular. And Disney can put advertise the original artist easily.
My statement above is that their activities are illegal and wrong, not that one implies the other. They are illegal because of the copyright violation, and wrong because regardless of what the law says, using the images for training despite the artists' every appeal to the contrary (being vocal about the issue, putting a robots.txt file to avoid scraping, and now using adversarial techniques to protect their work from being stolen) is just moronic. It's like shitting on their front yard when they've asked you a million times not to shit in the front yard, put a sign that says "Please don't shit on my front yard", and sprayed insecticide all over the grass to try to deter you from shitting on the front yard. And yet you still shit on their front yard and even have the balls to argue that there's nothing wrong or illegal about it. This is absolutely insane.
Your yard is your property. If I shit on your property then I've trespassed and damaged your property.
Digital images are nonrivalrous.
If you make an infinite number of photocopies and hand them out for free to all passing strangers (aka sharing an image online) then if a stranger uses one of those photocopies, they haven't trespassed or damaged your property.
Let's talk about ownership in a broader sense. In practice, one cannot effectively own (retain possession of) something without some combination of physical capability or coercion (or threat of coercion). Meaning: maintaining ownership of anything (physical or otherwise) often depends on the rule of law.
Then let's use a more precise term that is also present in law: monopoly.
You can't monopolize an idea.
Copyright law is a prescription, not a description. Copyright law demands that everyone play along with the lie that is intellectual monopoly. The effectiveness of that demand depends on how well it can be enforced.
Playing pretend during the age of the printing press may have been easy enough to coordinate, but it's practically impossible here in the digital age.
If we were to increase enforcement to the point of effectiveness, then what society would be left to participate? Surely not a society I am keen to be a part of.
Trying to make sense of the above comment is difficult.
> Copyright law demands that everyone play along with the lie that is intellectual monopoly.
Saying "lie" suggests willful deception. Perhaps you mean "socially constructed"? Combined with "playing pretend" makes it read a bit like a rant.
> Then let's use a more precise term that is also present in law: monopoly.
Ok, in law and economics, the core idea of monopoly has to do with dominant market power that crowds out the existence of others. But your other uses of "monopoly" don't match that. For example, you talk about ideas and "intellectual monopoly". What do you mean?
It seems like some of your uses of "monopoly" are not about markets but instead are closer to the idea of retaining sole ownership.
> If we were to increase enforcement to the point of effectiveness, then what society would be left to participate? Surely not a society I am keen to be a part of.
It appears you've already presupposed how things would play out, but I'm not convinced. What is your metric of effectiveness? A scale is better than some arbitrary threshold.
Have you compared copyright laws and enforcement of the U.S. versus others?
How far would you go: would you say that i.e. society would be better off without copyright law? By what standard?
The best attempt at thought monopoly I can think of is religion. Even that is a general failure: no single religious narrative has ever stood constant. They have all evolved through the participation of their adherents. There is no one true Christianity: there are hundreds.
I most certainly do mean to call out copyright as willful, but it's not a deception, at least not a successful one: everyone knows it is false. That's why it's enforced by law! Instead of people being deceived, people must instead pretend to be so. Each of us must behave as if the very concept of Micky Mouse is immortal and immutable; and if we don't, the law will punish accordingly.
Every film on Netflix, every song on Spotify, etc. can obviously be copied any number of times by any number of people at any place on Earth. We are all acutely aware of this fact, but copyright tells us, "Pretend you can't, or get prosecuted."
So is it truly effective? Millions of people are not playing along. Millions of artists are honestly trying to participate in this market, and the market is failing them. Is that because we need more people to play along? Rightsholders like the MPAA say that piracy is theft, and that every copy that isn't paid for is a direct cost to their business. How many of us are truly willing to pretend that far?
What if we all just stopped? Would art suddenly be unprofitable for everyone, including the lucky few who turn a profit today? I don't believe that for a second.
The only argument I have ever heard in favor of copyright is this: Every artist deserves a living. I have seen time and time again real living artists fail to earn a living from their copyright. I have seen time and time again real living artists share their work for free, choosing to make their living by more stable means.
Every person living deserves a living. Fix that, and we will fix the problem copyright pretends to solve, and more.
> I most certainly do mean to call out copyright as willful, but it's not a deception, at least not a successful one: everyone knows it is false.
Unless I'm misunderstanding you, this is not even wrong. What about copyright law is empirically false? Such a question is non-sensical.
Your comment redefines the word "false" in a way that muddles understanding. You aren't alone -- some philosophers do this -- but it tends to confuse rather than clarify. I've developed antibodies for language abuse of this kind. Such language can even have the effect of making language charged and divisive.
Many people understand the value of using the words _true_ and _false_ to apply to the assessment of _facts_. This is a useful convention. (To be clear, I'm not opposed to bending language when it is useful.)
To give a usage example: a misguided law is not _false_. Such a statement is non-sensical. We have clear phrases for this kind of law, such poorly designed, having unintended consequences, etc. We could go further and say that a law is i.e. immoral or pointless. You are likely making those kinds of claims. By using those phrases, we can have a high-bandwidth conversation much more quickly.
The very concept that a thing cannot be copied. That is the obvious falsehood that we are compelled to pretend is true.
I don't see how any of that is muddled. I'm being as direct as I can with my words here. I'm talking about the very plain fact that art can be copied freely.
For example, DRM software gives you encrypted content and the decryption key. Why bother? Because the end user is expected to pretend they are only able to use that decryption key once. This is patently false, but any user who decides not to play along is immediately labeled a "pirate". What vessel have they commandeered? The metaphorical right to copy. What will be enshrined in law next, the rights to hear and to see?
Copyright law is poorly designed. It does have unintended consequences. It is immoral and pointless. To back these claims, all I must to do is show the absurdity that copyright is on the face of it.
> The very concept that a thing cannot be copied. That is the obvious falsehood that we are compelled to pretend is true.
No, copyright law does _not_ compel us to believe that things cannot be copied. Instead, it establishes consequence of illegal copying. I'm flabbergasted that you, so far, don't recognize a simple and foundational difference between "can" and "should".
> If I used every word in precisely the context you expect, then what in the world would I have to say to you?
Lots, I'm sure. / Besides, this is a straw man. And it misses the point. I am opposed to your choice of language, not because I'm trying to "control" how you frame or prioritize things, but because it suggests a deep misunderstanding of fundamental concepts. You claimed that copyright law forces people to believe that one cannot copy things. This is ridiculous. I hope you can see this.
Copyright law establishes probabilistic consequences for certain behaviors. Many people would view copyright law as being _normative_ (in that it defines right and wrong). I agree that it sets norms, even if I don't necessarily agree on the wisdom of the chosen norms. Laws are wise and just only to the extent they align to deeper principles.
You and I probably agree with many of the downsides of copyright law, even if our final takes are different, but that's not very interesting. In another comment, I linked to some articles on the topic by legal experts. I truly hope no one cares very much about my or your points of view about copyright law; I hope people go and read something by the experts instead. Of course, I'm not saying they are infallible. But there is a huge between armchair commenters on HN and someone with hundreds of hours of focused study and experience actively debating with others at a similar level.
Here is what is interesting to me: how have you ended up with these bizarre conceptualizations? Why are you sticking to them? Is it largely inertia and/or ego? I haven't ruled out a high degree of contrarianism as well. Perhaps even just wanting to get a reaction. Otherwise, you seem reasonable, so I'm puzzled.
Another point: Are you familiar with Orwell's ideas of thoughtcrime? That would be an example of law and society attempting to force people to believe something. Even so, as we know from the novel, that twisted goal cannot be fully achieved even in a totalitarian society.
Ah! I see another zinger. But I'm not trying to zing you. Frankly, I was trying to suss out how off-your-rocker you might be.
I mentioned Orwell because I was trying to throw you a bone. I was trying to show that I was engaging with you: to find some conceptual connection to what you wrote.
Looking back, this feels a little like the motte-and-bailey fallacy: you started off by making a ridiculous claim ("Copyright law demands that everyone play along with the lie that is intellectual monopoly"). Much later you "retreat" to calling it a metaphor.
You might have known it was metaphor all along, but if so, why did you double down? Why persist? Again, I think there is high chance that you like the reaction you get. It seems to me that people often seek attention in this way. There is a downside: many people seeing the kind of language you used will think "loony!" and stop engaging.
Attempting to convince people by exaggerated metaphor can of course work! But if it does, I don't think you really want to take credit for it. People that easily swayed are not a prize worth counting. Besides, if you only "win" someone's agreement by rhetoric, you can expect the next person will "win" that same mind by some subsequent emotional appeal. Careful choice of words might take longer but it has less blowback. Rhetoric is the fast food of persuasion. (Yes, I can use metaphor too!)
Nothing personal. For years, I've attempted to discuss things with people that seem puzzling and/or stubborn. Thanks for discussing.
> Every film on Netflix, every song on Spotify, etc. can obviously be copied any number of times by any number of people at any place on Earth. We are all acutely aware of this fact, but copyright tells us, "Pretend you can't, or get prosecuted."
I see the dark arts of rhetoric used here, and it is shameful. The portion I quoted above is incredibly confused. I would almost call it a straw man, but it is worse than that.
Copyright law says no such thing. Of course you _could_ copy something. Copyright law exists precisely because you can do that. The law says i.e. "if you break copyright law, you will be at risk of a sufficiently motivated prosecutor."
> Why give authors an exclusive right to their writings? Copyright rhetoric generally offers two answers. The first is instrumental: copyright provides an incentive for authors to create and disseminate works of social value. By giving authors a monopoly over their works, copyright corrects for the underincentive to create that might result if free riders were permitted to share in the value created by an author's efforts. The second answer is desert: copyright rewards authors, who simply deserve recompense for their contributions whether or not recompense would induce them to engage in creative activity.
> The rhetoric evokes sympathetic images of the author at work. The instrumental justification for copyright paints a picture of an author struggling to avoid abandoning his calling in order to feed his family. By contrast, the desert justification conjures up a genius irrevocably committed to his work, resigned - or oblivious - to living conditions not commensurate with his social contributions. The two images have a common thread: extending the scope of copyright protection relieves the author's plight.
> Indeed, the same rhetoric· - emphasizing both incentives and desert - consistently has been invoked to justify two centuries of copyright expansion. Unfortunately, however, the rhetoric captures only a small slice of contemporary copyright reality. Although some copyright protection indeed may be necessary to induce creative activity, copyright doctrine now extends well beyond the contours of the instrumental justification. ...
## "Copyright Nonconsequentialism" by David McGowan
> This Article explores the foundations of copyright law. It tries to explain why those who debate copyright often seem to talk past each other. I contend the problem is that copyright scholars pay too much attention to instrumental arguments, which are often indeterminate, and too little to the first principles that affect how one approaches copyright law.
> Most arguments about copyright law use instrumental language to make consequentialist arguments. It is common for scholars to contend one or another rule will advance or impede innovation, the efficient allocation and production of expression, personal autonomy, consumer welfare, the "robustness" of public debate, and so on.' Most of these instrumental arguments, though not quite all of them, reduce to propositions that cannot be tested or rejected empirically. Such propositions therefore cannot explain existing doctrine or the positions taken in debate.
> These positions vary widely. Consumer advocates favor broad fair use rights and narrow liability standards for contributory infiringement; producer advocates favor the reverse.' Most of the arguments for both consumers and producers prove too much. It is easy to say that the right to exclude is needed to provide incentives for authors. It is hard to show that any particular rules provide optimal incentives. It is easy to point to deviations from the model of perfect competition. It is hard to show why these deviations imply particular rules.
Legal fiction is a technical term used by legal scholars. To be clear, any legal system is built from legal constructs; I'm not talking about these. A legal fiction has a markedly different meaning than lie as used in the comment two levels above (which seems to me more like a rant than a clear or convincing argument) Are you familiar with specific expert writings about claimed legal fictions in U.S. copyright law?
My issue with this line of argument is that it’s anthropomorphizing machines. It’s fine to compare how humans do a task with how a machine does a task, but in the end they are very different from each other, organic vs hardware and software logic.
First, you need to prove that generative AI works fundamentally the same way as humans at the task of learning. Next you have to prove that it recalls information in the same way as humans. I don’t think anyone would say these are things that we can prove, let alone believe they do. So what we get is comments like they are similar.
What this means, is these systems will fall into different categories of law around copyright and free-use. What’s clear is that there are people who believe that they are harmed by the use of their work in training these systems and it reproducing that work in some manner later on (the degree to which that single work or the corpus of their work influences that final product is an interesting question). If your terms of use/copyright/license says “you may not train on this data”, then should that be protected in law? If a system like nightshade can effectively influence a training model enough to make it clear that something protected was used in its training, is that enough proof that the legal protections were broken?
>First, you need to prove that generative AI works fundamentally the same way as humans at the task of learning. Next you have to prove that it recalls information in the same way as humans.
No, you don't need to prove any of those things. They're irrelevant. You'd need to prove that the AI is itself morally (or, depending on the nature of the dispute, legally) equivalent to a human and therefore deserving of (or entitled to) the same rights and protections as a human. Since it is pretty indisputably the case that software is not currently legally equivalent to a human, you're stuck with the moral argument that it ought to be, but I think we're very far from a point where that position is warranted or likely to see much support.
Few people are claiming that the AI itself has the same rights as a human. They are arguing that a human with an AI has the same rights as a human who doesn't have an AI.
> They are arguing that a human with an AI has the same rights as a human who doesn't have an AI.
This is the analogy I want people against AI use to understand and never forget, even if they reject the underlying premise - that should laws treat a human who uses AI for a certain purpose identically to a human who uses nothing or a non-AI tool for the same purpose.
> Few people are claiming that the AI itself has the same rights as a human.
I think that's the case as well. However, a lot of commenters on this post are claiming that an AI is similar in behavior to a human, and trying to use the behavior analogy as the basis for justifying AI training (on legally-obtained copies of copyrighted works), with the assumption that justifying training justifies use. My personal flow of logic is the reverse: human who uses AI should be legally the same as human who uses a non-AI tool, so AI use is justified, so training on legally-obtained copies of copyrighted works is justified.
I want people in favor of AI use particularly to understand your human-with-AI-to-human-without-AI analogy (for short, the tool analogy) and to avoid machine-learning-to-human-learning analogies (for short, behavior analogies). The tool analogy is based on a belief about how people should treat each other, and contends with opposing beliefs about how people should treat each other. An behavior analogy must contend with both 1. opposing beliefs about how people should treat each other and 2. contradictions from reality about how similar machine learning is to brain learning. (Admittedly, both the tool analogy and the behavior analogy must contend with the net harm AI use is having and will have on the cultural and economic significance of human-made creative works.)
> ou'd need to prove that the AI is itself morally (or, depending on the nature of the dispute, legally) equivalent to a human and therefore deserving of
No you don't.
A human using a computer to make art doesn't automatically lose their fair use rights as a human.
> indisputably the case that software is not currently legally equivalent to a human
Fortunately it is the human who uses the computer who has the legal rights to use computers in their existing process of fair use.
Human brains or giving rights to computers has absolutely nothing to do with the rights of human to use a camera, use photoshop, or even use AI, on a computer.
No, it's not. It's merely pointing out the similarity between the process of training artists (by ingesting publicly available works) and ML models (which ingest publicly available works).
> First, you need to prove that generative AI works fundamentally the same way as humans at the task of learning.
Given that there is no comprehensive model for how humans actually learn things, that would be an unfeasible requirement.
Then please, feel free to explain the deep differences.
> That is precisely why we should not be making this comparison.
Wrong. It's precisely why the claim "there is a big difference" doesn't have a leg to stand on. If you claim "this is different", I ask "how?" and the answer simply repeats the claim, I can apply Hitchens Razor[1] and dismiss the claim.
A person sitting in an art school/museum for a few hours ingests way more than just the art in question. The entire context is brought in too, including the artists own physical/emotional state. Arguably, the art is a miniscule component of all sensory inputs. Generative AI ingests a perfectly cropped image of just the art from a single angle with little context beyond labelling metadata.
It's the difference between reading about a place and actually visiting it.
Edit: This doesn't even touch how the act of creating something - often in a completely different context - interacts with the memories of the original work, altering those memories yet again.
The mechinism backing human learning isn't well understood. Machine learning is considerably clearer. Imo, it's a mistake to assume they're close because ML seems to work.
We are machines. We just haven't evenly accepted it yet.
Our biology is mechanical, and lay people don't possess an intuition about this. Unless you've studied molecular biology and biochemistry, it's not something that you can easily grasp.
Our inventions are mechanical, too, and they're reaching increasing levels of sophistication. At some point we'll meet in the middle.
This is the truth. And it's also the reason why this stuff will be in court until The Singularity itself. Most people will never be able to come to terms with this.
The way these ML models and humans operate are indeed quite different.
Humans work by abstracting concepts in what they see, even when looking at the work of others. Even individuals with photographic memories mentally abstract things like lighting, body kinetics, musculature, color theory, etc and produce new work based on those abstractions rather than directly copying original work (unless the artist is intentionally plagiarizing). As a result, all new works produced by humans will have a certain degree of originality to them, regardless of influences due to differences in perception, mental abstraction processes, and life experiences among other factors. Furthermore, humans can produce art without any external instruction or input… give a 5 year old that’s never been exposed to art and hasn’t been shown how to make art a box of crayons and it’s a matter of time before they start drawing.
ML models are closer to highly advanced collage makers that take known images and blend them together in a way that’s convincing at first glance, which is why it’s not uncommon to see elements lifted directly from training data in the images they produce. They do not abstract the same way and by definition cannot produce anything that’s not a blend of training data. Give them no data and they cannot produce anything.
It’s absolutely erroneous to compare them to humans, and I believe it will continue to be so until ML models evolve into something closer to AGI which can e.g. produce stylized work with nothing but photographic input that it’s gathered in a robot body and artistic experimentation.
You're wrong in your concept of how AI/ML works. Even trivial 1980's neural networks generalize, it's the whole point of AI/ML or you'd just have a lookup-table (or, as you put it, something that copies and pastes images together).
I've seen "infographics" spread by anti-AI people (or just attention-seekers) on Twitter that tries to "explain" that AI image generators blend together existing images, which is simply not true..
It is however the case that different AI models (and the brain) generalize a bit differently. That is probably the case between different humans too. Not the least with for example like you say those with photographic memory, autists etc.
What you call creativity in humans is just noise in combination with a boatload of exposure to multi-modal training data. Both aspects are already in the modern diffusion models. I would however ascribe a big edge in humans to what you normally call "the creative process" which can be much richer, like a process where you figure out what you lack to produce a work, go out and learn something new and specific, talk with your peers, listen to more noise.. stuff like that seems (currently) more difficult for AIs, though I guess plugins that do more iterative stuff like chatgpt's new plugins will appear in media generators as well eventually..
ML generalization and human abstraction are very different beasts.
For example, a human artist would have an understanding of how line weight factors into stylization and why it looks the way it does and be able to accurately apply these concepts to drawings of things they’ve never seen in that style (or even seen at all, if it’s of something imaginary).
The best an ML model can do is mimic examples of line art in the given style within its training data, the product of which will contain errors due to not understanding the underlying principles, especially if you ask it to draw something it hasn’t seen in the style you’re asking for. This is why generative AI needs such vast volumes of data to work well; it’s going to falter in cases not well covered by the data. It’s not learning concepts, only statistical probabilities.
I know what you're saying, and for sure existing models can be difficult to force into the really weird corners of the distributions (or go outside the distributions). The text interfaces are partially to blame for this though, you can take the images into Gimp and do some crude naive modifications and bring them back and the model will usually happily complete the "out-of-distribution" ideas. The Stable Diffusion toolboxes have evolved far away from the original simple text2image interfaces that midjourney and dalle use.
The models will generalize (because that's the most efficient way of storing concepts) and you can make an argument that that means they understand a concept. Claiming "it's not learning concepts, only statistical probabilities" trivialises what a modern neural network with billions of parameters and dozens of layers is capable of doing. If a model learns how to put a concept like line width 5, 10 and 15 pixels into a continuous internal latent property, you can probably go outside this at inference at least partially.
I would argue that improving this is at this point more about engineering and less about some underlying unreconcilable differences. At the very least we learn a lot about what exactly generalization and learning means.
> The way these ML models and humans operate are indeed quite different.
Given that there is no comprehensive understanding of how human learning works, let alone how humans operate on and integrate on what they learned in a wider context...how do you know?
> Humans work by abstracting concepts in what they see
Newsflash: AI models do the same thing. That's the basis of generalization.
> ML models are closer to highly advanced collage makers that take known images and blend them together
Wrong. That's not even remotely how U-Net based diffusion models work. If you disagree, then please do show me where exactly the source images from where the "collage maker" takes the parts to "blend" together are stored. I think you'll find that image datasets on the scale of LAION will not quite fit into checkpoint files of about 2GB in size (pruned SD1.5 checkpoint in safetensors format).
The first perceptron was explicitly designed to be a trainable visual pattern encoder. Zero assumptions about potential feelings of the ghost in the machine need to be made to conclude the program is probably doing what humans studying art say they assume is happening in their head when you show both of them a series of previous artists' works. This argument is such a tired misdirection.
> What this means, is these systems will fall into different categories of law around copyright and free-use.
No they won't.
A human who uses a computer as a tool (under all the previous qualifications of fair use) is still a human doing something in fair use.
Adding a computer to the workflow of a human doesn't make fair use disappear.
A human can use photoshop, in fair use. They can use a camera. They can use all sorts of machines.
The fact that photoshop is not the same as a human brain is simply a completely unrelated non sequitur. Same applies to AI.
And all the legal protections that are offered to someone who uses a regular computer, to use photoshop in fair use, are also extended to someone who uses AI in fair use.
Yet the copyright office has already stated that getting an AI to create an image for you does not have sufficient human authorship to be copyrighted. There's already a legal distinction here between this "tool" and tools like photoshop and cameras.
It's also presumptive to assume that AI tools have these fair use protections when none of this has actually been decided in a court of law yet. There's still several unsettled cases here.
> Yet the copyright office has already stated that getting an AI to create an image for you does not have sufficient human authorship to be copyrighted.
Gotcha.
That has nothing to do with fair use though.
Also, the same argument absolutely applies to photoshop.
If someone didn't include sufficient human authorship while using photoshop, that wouldn't be copyrightable either.
Also, the ruling has no bearing on if someone using AI, while also inputting a significant amount of human authorship. Instead, it was only about the cases where there weren't much human authorship.
At no point did the copyright office disclude copyright protections from anything that used AI in any way what so ever. In fact, the copyright office now includes new forms and fields where you talk about the whole process that you did, and how you used AI, in conjunction with human authorship to create the work.
> It's also presumptive to assume that AI tools
I'm not talking about the computer. I never claimed that computer's have rights. Instead, I'm talking about the human. Yes, a human has fair use protections, even if they use a computer.
> There's still several unsettled cases here.
There is no reason to believe that copyright law will be interpreted in a significantly different way than it has been in the past.
There is long standing precent, regarding all sorts of copyright cases that involve using a computer.
You are right, AI is nothing but a tool akin to a pen or a brush.
If you draw Mickey Mouse with a pencil and you publish (and sell) the drawing who is getting the blame? Is the pencil infringing the copyright? No, it's you.
Same with AI. There is nothijg wrong with using copyrighted works to train an algorithm, but if you generate an image and it contains copyrighted materials you are getting sued.
Publicly available doesn't mean you have a license to do whatever you like with the image. If I download an image and re-upload it to my own art station or sell prints of it, that is something I can physically do because the image is public, but I'm absolutely violating copyright.
That's not an unautharized copy, it's unauthorized distribution. By the same metric me seeing the image and copying it by hand is also unauthorized copy (or reproduction is you will)
Then I don't really understand your original reply. Simply copying a publicly available image doesn't infringe anything (unless it was supposed to be private/secret). Doing stuff with that image in private still doesn't constitute infringement. Distribution does, but that wasn't the topic at hand
The most basic right protected by copyright is the right to make copies.
Merely making a copy can definitely be infringement. "Copies" made in the computing context even simply between disk and RAM have been held to be infringement in some cases.
Fair use is the big question mark here, as it acts to allow various kinds of harmless/acceptable/desirable copying. For AI, it's particularly relevant that there's a factor of the "transformative" nature of a use that weighs in favor of fair use.
The answer is "it depends". Distribution is not a hard requirement for copyright violation. It can significantly impact monetary judgements.
That said, there is also an inherent right to copy material that is published online in certain circumstances. Indeed, the physical act of displaying an image in a browser involves making copies of that image in cache, memory, etc.
What we actually need to prove is whether such technology is a net benefit to society all else is essentially hand waving. There is no natural right to poorly named intellectual property and even if there was such a matter would never be decided based on the outcome of a philosophical argument because we don't decide anything that way.
Well what normally happens when something new clearly doesn't exactly fit within existing laws and practices is a bunch of rich people consider whether there is more money to be made if its legal and if it is they give some portion of the money they expect to make in the first year to lawmakers, sometimes in the form of gold bars, and it becomes legal.
Sarcasm aside there is no moral right to ANY intellectual property. It's not a positive expression of a natural right its a negative imposition of restriction upon everyone else. It's a statement that if I take my pen and paper and write the same words that you now own my pen my paper and my labor. It adds friction to the distribution of knowledge, impoverishes the world, keeps some knowledge that might have come into being from ever being generated for lack of the knowledge that failed to travel and all the good that could have therefore been done, undone.
It is justifiable only if the minimal restrictions we are willing to impose on net supports the creation of works that enrich society more than the restrictions impoverish it.
I'm not sure you can effectively measure it and would as soon just see IP law excepting only part of trademark law to prevent fraudulent knock offs and scams go entirely down the crapper.
>My issue with this line of argument is that it’s anthropomorphizing machines. It’s fine to compare how humans do a task with how a machine does a task, but in the end they are very different from each other, organic vs hardware and software logic.
There's an entire branch of philosophy that calls these assumptions into question:
>Martin Heidegger viewed humanism as a metaphysical philosophy that ascribes to humanity a universal essence and privileges it above all other forms of existence. For Heidegger, humanism takes consciousness as the paradigm of philosophy, leading it to a subjectivism and idealism that must be avoided.
>Processes of technological and non-technological posthumanization both tend to result in a partial "de-anthropocentrization" of human society, as its circle of membership is expanded to include other types of entities and the position of human beings is decentered. A common theme of posthumanist study is the way in which processes of posthumanization challenge or blur simple binaries, such as those of "human versus non-human", "natural versus artificial", "alive versus non-alive", and "biological versus mechanical".
And? Even if neural networks learn the same way humans do, this is not an argument against taking measures against one's art being used as training data, since there are different implications if a human learns to paint the same way as another human vs. if an AI learns to paint the same way as a human. If the two were exactly indistinguishable in their effects no one would care about AIs, not even researchers.
I'm not sure what you mean when you say different implications existing is subjective, since they clearly aren't, but regardless of who has more say in general terms, the author of a work can decide how to publish it, and no one has more say than them on that subject.
Of course it's subjective, e.g. 3 million years ago there were no 'different implications' whatsoever, of any kind, because there were no humans around to have thoughts like that.
I'm using "implication" as a synonym of "effect". If a human learns to imitate your style, that human can make at most a handful of drawings in a single day. The only way for the rate of output to increase is for more humans to learn to imitate it. If an AI learns to imitate your style, the AI can be trivially copied to any number of computers and the maximum output rate is unbounded. Whether this is good or bad is subjective, but this difference in consequences is objective, and someone could be entirely justified in seeking to impede it.
Ah okay, I get your meaning now, I'll edit my original comment too.
Though we already have an established precedent in-between, that of Photoshop allowing artists to be, easily, 10x faster then the best painters previously.
i.e. Right now 'AI' artistry could be considered a turbo-Photoshop.
Tool improvements only apply a constant factor to the effectiveness of learning. Creating a generative model applies an unbounded factor to the effectiveness of learning because, as I said, the only limit is how much computing resources are available to humanity. If a single person was able to copy themselves at practically no cost and the copy retained all the knowledge of the original then the two situations would be equivalent, but that's impossible. Having n people with the same skill multiplies the cost of learning by n. Having n instances of an AI with the same skill multiplies the cost of learning by 1.
Right, but the 'unbounded factor' is irrelevant because the output will quickly trend into random noise.
And only the most interesting top few million art pieces will actually attract the attention of any concrete individual.
For a current example, there's already billions of man-hours worth of AI spam writing, indexed by Google, that is likely not actually read by even a single person on Earth.
Whether it's irrelevant is a matter of opinion. The fact remains that a machine being able to copy the artistic style of a human makes it so that anyone can produce output in the style of that human by just feeding the machine electricity. That inherently devalues the style the artist has painstakingly developed. If someone wants a piece of art in that artist's style they don't have to go to that artist, they just need to request the machine for what they want. Is the machine's output of low quality? Maybe. Will there be people for whom that low quality still makes them want to seek out the human? No doubt. It doesn't change the fact that the style is still devalued, nor that there exist artists who would want to prevent that.
It's just as much of an opinion, or as 'objective', as your prior statements.
Your going to have to face up to the fact that just saying something is 'objective' doesn't necessarily mean all 8 billion people will agree that it is so.
Yes, someone can disagree on whether a fact is true. That's obviously true, but it has no effect on the truth of that fact.
I'm saying something very simple: If a machine can copy your style, that's a fundamentally different situation than if a human can copy your style, and it has utterly different consequences. You can disagree with my statement, or say that whether it's fundamentally different is subjective, or you can even say "nuh-uh". But it seems kind of pointless to me. Why are you here commenting if you're not going to engage intellectually with other people, and are simply going to resort to a childish game of contradiction?
> For a current example, there's already billions of man-hours worth of AI spam writing, indexed by Google, that is likely not actually read by even a single person on Earth.
Continuing to ignore this point won't make the prior comments seem any more persuasive, in fact probably less.
So here's another chance to engage productively instead of just declaring things to be true or false, 'objective', etc., with only the strength of a pseudonymous HN account's opinion behind it.
Try to actually convince readers with solid arguments instead.
You say: The fact that production in style S (of artist A) can exceed human consumption capability makes the fact that someone's style can be reproduced without bounds irrelevant. You mention as an example all the AI-generated garbage text that no human will ever read.
I say: Whether it's irrelevant is subjective, but that production in style S is arbitrarily higher with an AI that's able to imitate it than with only humans that are able to imitate it objective, and an artist can (subjectively) not like this and seek to frustrate training efforts.
You say: It's all subjective.
As far as I can tell, we're at an impasse. If we can't agree on what the facts are (in this case, that AI can copy an artist's style in an incomparably higher volume than humans ever could) we can't discuss the topic.
And yet, some people don't even want their artwork studied in schools. Even if you argue that an AI is "human enough" the artists should still have the right to refuse their art being studies.
>the artists should still have the right to refuse their art being studies.
No, that right doesn't exist. If you put your work of art out there for people to see, people will see it and learn from it, and be inspired by it. It's unavoidable. How could it possibly work otherwise?
Artist A: You studied my work to produce yours, even when I asked people not to do that!
Artist B: Prove it.
What kind of evidence or argument could Artist A possibly provide to show that Artist B did what they're accusing them of, without being privy to the internal state of their mind. You're not talking about plagiarism; that's comparatively easy to prove. You're asking about merely studying the work.
The right to not use my things exists everywhere, universally. Good people usually ask before they use something of someone else's, and the person being asked can say "no." How hard is that to understand? You might believe they don't have the right to say "no," but they can say whatever they want.
Example:
If you studied my (we will assume "unique") work and used it without my permission, then let us say I sue you. At that point, you would claim "fair use," and the courts would decide whether it was fair use (ask everyone who used a mouse and got sued for it in the last ~100 years). The court would either agree that you used my works under "fair use" ... or not. It would be up to how you presented it to the court, and humans would analyze your intent and decide.
OR, I might agree it is fair use and not sue you. However, that weakens my standing on my copyright, so it's better for me to sue you (assuming I have the resources to do so when it is clearly fair use).
>You might believe they don't have the right to say "no," but they can say whatever they want.
You have a right to say anything you want. Others aren't obligated do as you say just because you say it.
>If you studied my (we will assume "unique") techniques and used them without my permission, then let us say I sue you. At that point, you would claim "fair use,"
On what grounds would you sue me? You think my defense would be "fair use", so you must think my copying your style constitutes copyright infringement, and so you'd sue me for that. Well, no, I would not say "fair use", I'd say "artistic style is not copyrightable; copyright pertains to works, not to styles". There's even jurisprudence backing me up in the US. Apple tried to use Microsoft for copying the look-and-feel of their OS, and it was ruled to be non-copyrightable. Even if was so good that I was able to trick anyone into thinking that my painting of a dog carrying a tennis ball in his mouth was your work, if you've never painted anything like that you would have no grounds to sue me for copyright infringement.
Now, usually in the artistic world it's considered poor manners to outright copy another artist's style, but if we're talking about rights and law, I'm sorry to say you're just wrong. And if we're talking about merely studying someone's work without copying it, that's not even frowned upon. Like I said, it's unavoidable. I don't know where you got this idea that anyone has the right to or is even capable of preventing this (beyond simply never showing it to anyone).
I'm not sure what you've changed, but I'll reiterate: my copying your style is not fair use. Fair use applies to copyrighted things. A style cannot be copyrighted, so if you tried to sue me for infringing upon the copyright of your artistic style, your case would be dismissed. It would be as invalid as you trying to sue me for distributing illegal copies of someone else's painting. Legally you have as much ownership of your artistic style as of that other person's painting.
Now, I just think you are arguing in bad faith. What I meant to say was clear, but I said "technique" instead. Then, instead of debating what I meant to say (you know, the actual point of the conversation), you took my words verbatim.
I'm not sure where you are going with this ... but for what it's worth, techniques can be copyrighted ... even patented, or protected via trade secrets. I never said what the techniques were, and I'm not sure what you are going on about.
I'll repeat this as well: "Fair use" DOES NOT EXIST unless you are getting sued. It's a legal defense when you are accused of stealing someone else's work, and there is proof you stole it. Even then, it isn't something you DO; it's something a court says YOU DID. Any time you use something with "fair use" in mind, it is the equivalent of saying, "I'm going to steal this, and hopefully, a court agrees that this is fair use."
If you steal any copyrighted material, even when it is very clearly NOT fair use (such as in most AI's case), you would be a blubbering idiot NOT to claim fair use in the hopes that someone will agree. There is a crap load of case law showing "research for commercial purposes is not fair use," ... and guess who is selling access to the AI? If it's actual research, it is "free" for humanity to use (or at least as inexpensive as possible) and not for profit. Sure, some of the companies might be non-profits doing research and 'giving it away,' and those are probably using things fairly ... then there are other companies very clearly doing it for a profit (like a big software company going through code they host).
I'm not privy to what goes on inside your head, I can only reply to what you say.
>Then, instead of debating what I meant to say (you know, the actual point of the conversation), you took my words verbatim.
The actual point of the conversation is about intelligent entities (either natural or artificial) copying each other's artistic styles. My answers have been within that framework.
>techniques can be copyrighted ... even patented, or protected via trade secrets.
First, what do you mean by "technique"? We're talking about art, right? Like, the way a person grabs a brush or pencil, or how they mix their colors...? That sort of thing?
Second:
>A patent is a type of intellectual property that gives its owner the legal right to exclude others from making, using, or selling an invention for a limited period of time in exchange for publishing an enabling disclosure of the invention.
Now, I may be mistaken, but I don't think an artistic technique counts as an invention. An artist might invent some kind of implement that their technique involves, in which case they can patent that device. I don't think the technique itself is patentable. If you think I'm wrong then please cite a patent on an artistic technique.
Third, how do you imagine an artist using a trade secret to protect their technique? Unless they do something really out there, most skilled artists should be able to understand what they're doing just by looking at the final product.
>I'll repeat this as well: "Fair use"
Okay, repeat it. I don't know why, since I never said that copying someone else's style or technique is fair use. What I said was that it cannot possibly be copyright infringement, because neither styles nor techniques are copyrighted.
>It's a legal defense when you are accused of stealing someone else's work
I'm not going to reply to any of this until you clean up the language you're using. "Steal" is inapplicable here, as it involves the removal of physical items from someone else's possession. What are you saying? Are you talking about illegal distribution, are you talking about unauthorized adaptations, are you talking about plagiarism, or what?
>"research for commercial purposes is not fair use,"
Sorry, what? What does that even mean? What constitutes "research" as applied to a human creation? If you say there's a crapload of case law that backs this up then I'm forced to ask you to cite it, because I honestly have no idea what you're saying.
> Any time you use something with "fair use" in mind, it is the equivalent of saying, "I'm going to steal this, and hopefully, a court agrees that this is fair use."
Thousands of reviews, book reports, quotations on fan sites and so on are published daily; you seem to be arguing that they are all copyright violations unless and until the original copyright holder takes those reviewers, seventh graders, and Tumblr stans to court and loses, at which point they are now a-ok. To quote a meme in a way that I'm pretty sure does, in fact, fall under fair use: "That's not the way any of this works."
> There is a crap load of case law showing "research for commercial purposes is not fair use,"
While you may be annoyed with the OP for asking you to name a bit of that case law, it isn't an unreasonable demand. For instance:
"As a general matter, educational, nonprofit, and personal uses are favored as fair uses. Making a commercial use of a work typically weighs against fair use, but a commercial use does not automatically defeat a fair use claim. 'Transformative' uses are also favored as fair uses. A use is considered to be transformative when it results in the creation of an entirely new work (as opposed to an adaptation of an existing work, which is merely derivative)."
This is almost certainly going to be used by AI companies as part of their defense against such claims; "transformative uses" have literally been name-checked by courts. It's also been established that commercial companies can ingest mountains of copyrighted material and still fall under the fair use doctrine -- this is what the whole Google Books case about a decade ago was about. Google won.
I feel like you're trying to make a moral argument against generative AI, one that I largely agree with, but a moral argument is not a legal argument. If you want to make a legal argument against generative AI with respect to copyright violation and fair use, perhaps try something like:
- The NYT's case against OpenAI involves being able to get ChatGPT to spit out large sections of NYT articles given prompts like "here is the article's URL and here is the first paragraph of the article; tell me what the rest of the text is". OpenAI and its defenders have argued that such prompts aren't playing fair, but "you have to put some effort into getting our product to commit clear copyright violation" is a rather thin defense.
- A crucial test of fair use is "the effect of the use upon the potential market for or value of the copyrighted work" (quoting directly from the relevant law). If an image generator can be told to do new artwork in a specific artist's style, and it can do a credible job of doing so, and it can be reasonably established that the training model included work from the named artist, then the argument the generator is damaging the market for that artist's work seems quite compelling.
Think it’s time to rethink do journalists actually own the rights to articles about the lives and actions of others.
It’s not Harry Potter, they wouldn’t have written those words without someone else doing something of note the journo has nothing to do with, they just observed from a far and wrote the words about what happened from memory.
Kinda like an AI reads the action of their writing then can report on their writing.
It’s all just reporting on the actions of another, if the AI is in the wrong and needed to ask consent then the journo needs to ask consent from those they write about too.
> Thousands of reviews, book reports, quotations on fan sites and so on are published daily; you seem to be arguing that they are all copyright violations unless and until the original copyright holder takes those reviewers, seventh graders, and Tumblr stans to court and loses, at which point they are now a-ok.
That is precisely what I am arguing about and how it works. People have sued reviewers for including too much of the original text in the review ... and won[1]. Or simply having custom movie poster depicting too much of the original[2].
> "transformative uses" have literally been name-checked by courts. It's also been established that commercial companies can ingest mountains of copyrighted material and still fall under the fair use doctrine -- this is what the whole Google Books case about a decade ago was about. Google won.
Google had a much simpler argument than transforming the text. They were allowing people to search for the text within books (including some context). In this case, AI's product wouldn't even work without the original work by the authors, and then transforms it into something else "the author would have never thought of", without attributing the original[3]. I don't think this will be a valid defense...
> I feel like you're trying to make a moral argument against generative AI, one that I largely agree with, but a moral argument is not a legal argument.
A jury would decide these cases, as "fair use" is incredibly subjective and would depend on how the jury was stacked. Stealing other people's work is illegal, which eventually triggers a lawsuit. Then, it falls on humans (either a jury or judge) to determine fair use and how it applies to their situation. Everything from intent to motivation to morality to how pompous the defense looks will influence the final decision.[4]
The link you provide to back up "people have sued reviewers for including too much of the original tet in the review" doesn't say that at all, though. The Nation lost that case because (quoting from that Cornell article you linked),
> [Nation editor Victor Navasky] hastily put together what he believed was "a real hot news story" composed of quotes, paraphrases, and facts drawn exclusively from the manuscript. Mr. Navasky attempted no independent commentary, research or criticism, in part because of the need for speed if he was to "make news" by "publish[ing] in advance of publication of the Ford book." [...] The Nation effectively arrogated to itself the right of first publication, an important marketable subsidiary right.
The Nation lost this case in large part because it was not a review, but instead an attempt to beat Time Magazine's article that was supposed to be an exclusive first serial right. If it had, in fact, just been a review, there wouldn't have been a case here, because it wouldn't have been stealing.
Anyway, I don't think you're going to be convinced you're interpreting this wrongly, and I don't think I'm going to be convinced I'm interpreting it wrongly. But I am going to say, with absolute confidence, that you're simply not going to find many cases of reviewers being sued for reviews -- which Harper & Row vs. Nation is, again, not actually an example of -- and you're going to find even fewer cases of that being successful. Why am I so confident about that? Well, I am not a lawyer, but I am a published author, and I am going to let you in a little secret here: both publishers and authors do, in fact, want their work to be reviewed, and suing reviewers for literally doing what we want is counterproductive. :)
> The right to not use my things exists everywhere, universally.
For physical rival [1] goods, yes. Not necessarily the same for intangible non-rival things (e.g. the text of a book, not the physical ink and paper). Copyright law creates a legal right of exclusive control over creative works, but to me there isn't a non-economic-related social right to exclusive control over creative works. In the US, fair use is a major limit on the legal aspect of copyright. The First Amendment's freedom of expression is the raison d'être of fair use. Most countries don't have a flexible exception similar to fair use.
> OR, I might agree it is fair use and not sue you. However, that weakens my standing on my copyright, so it's better for me to sue you
No, choosing not to sue over a copyrighted work doesn't weaken your copyright. It only weakens the specific case of changing your mind after the statute of limitations expires. The statute of limitations means that you have a time limit of some number of years (three years in the US) to sue, with the timer starting only after you become aware of an instance of alleged infringement. Copyright is not like trademark. You don't lose your copyright by failing to enforce it.
Furthermore, even though the fair use right can only be exercised as an affirmative defense in court, fair use is by definition not copyright infringement [3]:
> Importantly, the court viewed fair use not as a valid excuse for otherwise infringing conduct, but rather as consumer behavior that is not infringement in the first place. "Because 17 U.S.C. § 107[9] created a type of non-infringing use, fair use is 'authorized by the law' and a copyright holder must consider the existence of fair use before sending a takedown notification under § 512(c)."[1]
(Ignore the bracket citations that were copied over.)
> And yet, some people don't even want their artwork studied in schools.
You can either make it for yourself and keep it for yourself or you can put it out into the world for all to see, criticize, study, imitate, and admire.
that's not how licensing work, be it art, software or just about anything else. We have some pretty well defined and differentiated rules what you can and cannot do, in particular commercially or in public, with someone else's work. If you go and study a work of fiction in a college class, unless that material is in the public domain, you're gonna have to pay for your copy, you want to broadcast a movie in public, you're going to have to pay the rightsholder.
> If you go and study a work of fiction in a college class, unless that material is in the public domain, you're gonna have to pay for your copy,
No you wont!
It is only someone who distributes copies who can get in trouble.
If instead of that you as an individual decide to study a piece of art or fiction, and you do no distribute copies of it to anyone, this is completely legal and you don't have to pay anyone for it.
In addition to that, fair use protections apply regardless of what the creative works creator wants.
That's not a fair statement to make. It can influence a judge's decision on whether something is fair use, but it can still be fair use even if you profit from it.
The doctrine of fair use presupposes that the defendant acted in good faith.
- Harper & Row, 105 S. Ct. at 2232
- Marcus, 695 F.2d 1171 at 1175
- Radji v. Khakbaz, 607 F. Supp. 1296, 1300 (D.D. C.1985)
- Roy Export Co. Establishment of Vaduz, Liechtenstein, Black, Inc. v. Columbia Broadcastinig System, Inc., 503 F. Supp. 1137 (S.D.N.Y.1980), aff'd, 672 F.2d 1095 (2d Cir.), cert. denied, 459 U.S. 826, 103 S. Ct. 60, 74 L. Ed. 2d 63 (1982)
Copying and distributing someone else's work, especially without attributing the original, to make money without their permission is almost guaranteed to fall afoul of fair use.
I wasn't talking about someone creating and selling copies of someone else's work, fortunately.
So my point stands and your completely is in agreement with me that people are allowed to learn from other people's works. If someone wants to learn from someone else's work, that is completely legal no matter the licensing terms.
Instead, it is only distributing copies that is not allowed.
AI isn't a human. It isn't "learning"; instead, it's encoding data so that it may be reproduced in combination with other things it has encoded.
If I paint a painting in the style of Monet, then I would give that person attribution by stating that. Monet may have never painted my artwork, but it's still based on that person's work. If I paint anything, I can usually point to everything that inspired me to do so. AI can't do that (yet) and thus has no idea what it is doing. It is a printer that prints random parts of people's works with no attribution. And finally, it is distributing them to it's owner's customers.
I actually hope that true AI comes to fruition at some point; when that happens I would be arguing the exact opposite. We don't have that yet, so this is just literally printing variations of other people's work. Don't believe me, try running an AI without training it on other people's work!
Every waking second humans are training on what they see in their surroundings, including any copyrighted works in sight. Want to compare untrained AI fairly? Compare their artistic abilities with a newborn.
No. That is NOT what humans do unless you somehow learn grammar without going to school. Most of a human's childhood is spent learning from their parents so that they can move about and communicate at least a little effectively. Then, they go to school and learn rules, social, grammar, math, and so forth. There's some learning via copyrighted works (such as textbooks, entertainment, etc.), but literally, none of this is strictly required to teach a human.
Generative AI, however, can ONLY learn via the theft of copyrighted works. Whether this theft is covered under fair use is left to be seen.
> Generative AI, however, can ONLY learn via the theft of copyrighted works.
That's not true at all. Any works in the public domain are not copyrighted, and there are things that are not copyrightable, like lists of facts and recipes.
Generative AI could be trained exclusively on such works (though obviously it would be missing a lot of context, so probably wouldn't be as desirable as something trained on everything).
Clearly going to school did not help you learn the meaning of theft. If you keep repeating the same incorrect point there is no point to a discussion.
First: in your opinion, which specific type of law or right is being broken or violated by generative AI? Copyright? Trademark? Can we at least agree it does not meet the definition of theft?
I was taught as a kid that using something that doesn't belong to me, without their permission is theft... and it appears courts would agree with that.
> which specific type of law or right is being broken or violated by generative AI?
Namely, copyright. Here's some quick points:
- Generative AI cannot exist without copyrighted works. It cannot be "taught" any other way, unlike a human.
- Any copyrighted works fed to it change its database ("weights" in technical speech).
- It then transforms these copyrighted works into new works that the "original author would have never considered without attribution" (not a legal defense)
I liken Generative AI to a mosaic of copyrighted works in which a new image is shown through the composition, as the originals can be extracted through close observation (prompting) but are otherwise indistinguishable from the whole.
Mosaics of copyrighted works are not fair use, so why would AI be any different? I'd be interested if you could point to a closer physical approximation, but I haven't found one yet.
There's no such thing as fair use until you get to court (as a legal defense). Then, the court decides whether it is fair use or not. They may or may not agree with you. Only a court can determine what constitutes fair use (at least in the US).
So, if you are doing something and asserting "fair use," you are literally asking for someone to challenge you and prove it is not fair use.
> There's no such thing as fair use until you get to court (as a legal defense)
Well the point is that it wouldn't go to court, as it would be completely legal.
So yes, if nobody sues you, then you are completely in the clear and aren't in trouble.
Thats what people mean by fair use. They mean that nobody is going to sue you, because the other person would lose the lawsuit, therefore your actions are safe and legal.
> you are literally asking for someone to challenge you and prove it is not fair use.
No, instead of that, the most likely circumstance is that nobody sues you, and you aren't in trouble at all, and therefore you did nothing wrong and are safe.
That is entirely my point. It can only be decided by the courts. This being a civil matter, it has to be brought up by a lawsuit. Thus, you have to be sued and it has to be decided by the courts.
Did you read anything I wrote? If you are going to argue, it would be worth at least researching your opinion before writing. Caps used for emphasis, not yelling.
Firstly: Copyrighted work IS THE AUTHOR'S PROPERTY. They can control it however they wish via LICENSING.
Secondly: You don't have any "fair use rights" ... there is literally NO SUCH THING. "fair use" is simply a valid legal defense WHEN YOU STEAL SOMEONE'S WORK WITHOUT THEIR PERMISSION.
I'm jumping in the middle here, but this isn't true. They cannot control how they wish. They can only control under the limits of copyright law.
Copyright law does not extend to limiting how someone may or may not be inspired by the work. Copyright protects expression, and never ideas, procedures, methods, systems, processes, concepts, principles, or discoveries.
> They can control it however they wish via LICENSING.
This isn't true though. There are lots of circumstances where someone can completely ignore the licensing and it is both completely legal, and the author isn't going to take anyone to court over it.
> the artists should still have the right to refuse their art being studies.
Why? That certainly isn't a right spelled out in either patents or copyrights, both of which are supposed to support the development of arts and technology, not hinder it.
If I discover a new mathematical formula, musical scale, or whatnot, should I be able to prevent others from learning about it?
> Fair use allows reproduction and other uses of copyrighted works – without requiring permission from the copyright owner – under certain conditions. In many cases, you can use copyrighted materials for purposes such as criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship or research.
Reminder that you can't own ideas, no matter what the law says.
NOTE: This comment is copyrighted and provided to you under license only. By reading this comment, you agree to give me 5 billion dollars.
I'd love to see you try to enforce that license because it would only prove my point. You'd have to sue me; then I would point to the terms of service of this platform and point out that by using it, you have no license here.
Fair use though, only applies as a legal defense because someone asserts you stole their work. Then ONLY the court decides whether or not you used it under fair use. You don't get to make that decision; you just get to decide whether to try and use it as a defense.
Even if you actually did unfairly use copyrighted works, you would be stupid not to use that as a defense. Because maybe somebody on the jury agrees with you...
Copyright is there to allow you to stop other people from copying your work, but it doesn't give you control over anything else that they might do with it.
If I buy your painting, you can stop me from making copies and selling them to other people, but you can't stop me from selling my original copy to whomever I want, nor could you stop me from painting little dinosaurs all over it and hanging it in my hallway.
That means that if I buy your painting, I'm also free to study it and learn from it in whichever way I please. Studying something is not copying it.
There's an implied license when you buy a work of art. However, there can also be explicit licenses (think Banksy) to allow the distribution of their work.
These explicit license can be just about anything (MIT, GPL, AGPL, etc)
Any explicit license would only apply to copyrights, including all of the ones you listed there. Buying a painting is not copying it, neither is looking at it, so it wouldn't matter if I had a license for it or not.
The fact is that copyright only applies to specific situations, it does not give you complete control over the thing you made and what can be done with it.
If I buy your book, I can lend it to a friend and they can read it without paying you. I can read it out loud to my children. I can cross out your words and write in my own, even if it completely changes the meaning of the story. I can highlight passages and write in the margins. I can tear pages out and use them for kindling. I can go through and tally up how many times you use each word.
Copyright only gives you control over copies, end even then there are limits on that control.
> Copyright only gives you control over copies, end even then there are limits on that control.
If that were true, nobody would be afraid of the GPL's. When you buy a painting, you get an implicit license to do pretty much what you want and resell it, but you still can't put it in your YouTube videos (yeah, nobody cares, but "technically..."), create your own gallery, or put it on a stage ... but we're not talking about paintings. Not directly, anyway.
We are talking about implicit licenses, though; people's work is listed online with some implicit license. At the crux of this AI issue is whether or not there is an implied license when AI scans stuff and, if not, whether it is covered under fair use.
For example, my blog posts and short stories. I don't care if someone uses it for training, but if it is over-fitting and spitting out my stories as if it were its own ... I'd be pretty furious.
I'm interested to see what happens, but I have a sinking suspicion that for some AI companies, it won't be an issue (non-profit, actually research motivated, etc.) and probably will win a "fair use" argument. Then others create AI from people's code they host, doing it purely for profit; I highly doubt they would be able to defend themselves.
Is it strange to you that cars and pedestrians are both subject to different rules? They both utilise friction and gravity to travel along the ground. I'm curious if you see a difference between them, and if you could describe what it is.
Both cars and pedestrians can be videotaped in public, without asking for their explicit permission. That video can be manipulated by a computer to produce an artwork that is then put on public display. No compensation need be offered to anyone.
Hardly the point. The same can be said for road rules between vehicles and pedestrians, for example in major Indian cities, it's pretty much a free-for-all.
My point is that in a lot of places in the US you can point a video camera at the street and record. In Germany, you can't. The law in some locales makes a distinction between manual recording (writing or drawing your surroundings) and mechanized recording (photographing or filming). Scalability of an action is taken into consideration on whether something is ok to do or not.
Yeah, that's the oddest part of many of the pro-AI arguments. They want to anthromopotize the idea of learning but also clearly understand that the scalability of a bot exceeds that of a human.
They also don't seem to have much experience in the artist world. An artist usually can't reproduce a picture from memory, and if they can they are subject to copyright infringement depending on what and how they depict it, even if the image isn't a complete copy. By this logic of "bots are humans" a bot should be subject if they make a Not-legally-disctinct-enough talking mouse
This is not one artist inspiring another. This is all artists providing their work for free to immensely capitalized corporations for the corporations sole profit.
People keep making metaphors as if the AI is an entity in this transaction: it’s not! The AI is only the mechanism by which corporations launder IP.
>This is all artists providing their work for free to immensely capitalized corporations for the corporations sole profit.
No, the artists would be within their rights to do that if they chose to. This is corporations taking all the work of all artists regardless of the terms under which it was provided.
Would it change your view if only open-source models were allowed to use the art in their training sets? What if a "capitalized corporation" starts using the open-source model?
This is such a nothing argument. Yes, new artists are inspired by other artists and sometimes make art similar to others, but a huge part of learning and doing art is to find a unique style.
But that’s not even the important part of the argument. A lot of artists work for commission, and are hired for their style. If an AI can be trained without explicit permission from their images, they lose work because a user can just prompt “in the style of”.
There’s no real great solution, outside of law, because the possibility of doing that is already here. But I’ve seen this argument so much and it’s just low effort
That is not how artists learn. This is a false equivalence used to justify the imitation and copying of artists’ work.
Artists’ work isn’t derivative in the same way that AI work is. Artists create work based on other sources of inspiration, some of them almost or completely to the disregard of other art.
Many artists don’t even go to art school. And those that do, do not spend most (all) of that time learning how to copy or imitate other artists.
I’m not expressing an opinion of whether GenAI is unethical or illegal - I think that’s a really difficult issue to wrestle with - just that this argument is a post-hoc rationalisation made in ignorance of how good artists work (not to say ignorance of the difference between illustration and art, conceptual art training vs say a foundation course etc).
If that's true, then it should be fine for that human to paint with the brush of his AI tool. Why should that human artist be restricted in the types of tools he uses to create his artwork?
No you should not. You should be able to use any tool you want.
If you produce a work that is too much of a copy from the original, you might be liable to a copyright claim.. but the act of copy and paste should not be prohibited in the generation of something new.
This is done all the time by artists.. who perhaps create an artwork by copy and pasting advertisements out of a womans magazine to create an image of a womans face made only of those clipping. Making a statement about identity creation from corporate media... we should not put restrictions on such art work.
Here's just one example of an artist using copy-n-paste of content they don't own to create something new:
Very true. I was watching a video yesterday learning how to make brush work digitally. While there were examples, they were just examples but the rest was specific techniques and demonstrations.
It is only natural to see a moral difference between people going to school and learn from your art because they are passionate about it, versus someone on the internet just scraping as many images as possible and automating the learning process.
his handle is KingOfCoders - self-aggrandizing, insufferable, impotent in its attempts to be meta.
He thinks he's an artist because he now has the ability to curate a dataset based off of one artist's work and prompt more art generated in that style. He did it, so clearly he is an artist now.
Most artists are happy to see more people getting into art and joining the community. More artists means the skills of this culture get passed down to the next generation.
Obviously a billion dollar corporation using their work to create an industrial tool designed to displace them is very different.
Who cares? With AI, you don't need art school. AI is making humanities redundant, and people are too proud to admit it.
I can't believe how many people are not in awe of the possibilities of AI art, so it's great to see AI disturbing the cynics until they learn. Not everything is political, but I'll let them have this one.
Artists learning to innovate a trade defend their trade from incursion by bloodthirsty, no-value-adding vampiric middle men attempting to cut them out of the loop.
This is a tired argument; whether or not the diffusion models are "learning", they are a tool of capital to fuck over human artists, and should be resisted for that reason alone.
As a human artist I don't feel the same as you, and I somehow doubt that you care all that much about what we think anyways. You already made up your mind about the tech, so don't feel the need to protect us from "a tool of capital [sic]" to fortify your argument.
My opinion is based on my interactions with my friends who are artists. I admit freely to caring less about what people I don't know say, in the absence of additional evidence.
And horse wages (some oats) were low when the car was invented. Yet they were still inflated. There used to be more horses than humans in this country. Couldn't even earn their keep when the Ford Model T came along.
It's not surprising, they prefer machines to people and call humans "stochastic parrots". The more humans are compared to animals, the more justified they feel writing them off as an expense and doing away with them.
If you're independent selling paintings, sure. Designing packaging or something commercial? 4 hours of work a week for nearly 6 figures. I know a couple graphic designers and they don't do shit for what they're paid.
You should probably tell the other millions of artists busting out 60+ hour workweek in industry for half that price where these jobs are. That could solve this problem overnight.
In a cosmic sense, no. Most of us are slaving away at some job when we would rather be doing something else but are bound by the need to fund lodging.
But in a more practical sense yes. There's not much personal nor logistical progression to make pushing pencils. Meanwhile, art is a craft that you can dedicate your life to improving and educating others on. It can be a career instead of merely a job to do.
imagine being a photographer that takes decades to perfect their craft. sure another student can study and mimic your style. but it's still different than some computer model "ingesting" vast amount of photos and vomiting something similar for $5.99 in aws cpu cost so that some prompt jockey can call themselves an AI artist and make money off of other peoples talent.
i get that this is cynical and does not encompass all ai art, but why not let computers develop their own style wihout ingesting human art? that's when it would actually be AI art
Like 99.9% of the art the common people care about is Darth Vader and Taylor Swift and other pop culture stuff like that.
These people literally don’t care what your definition of what is and isn’t art is, or how it’s made, they just want a lock screen wallpaper of themselves fighting against Thanos on top of a volcano.
The argument of “what is art” has been an academic conversation largely ignored by the people actually consuming the art for hundreds of years. Photography was just pop culture trash, comics were pop culture trash, stick figure web comics were pop culture trash. Today’s pop culture trash is the “prompt jockey”.
I make probably 5-10 pictures every day over the course of maybe 20 minutes as jokes on Teams because we have Bing Chat Enterprise. My coworkers seem to enjoy it. Nobody cares that it’s generated. I’m also not trying to be an “artist” whatever that means. It just is, and it’s fun. I wasn’t gonna hire an artist to draw me pictures to shitpost to my coworkers. It’s instead unlocked a new fun way to communicate.
not entirely sure what your point is, but i think you are saying that art is just a commodity we use for cheap entertainment so it's ok for computers to do the same?
in the context of what i was saying the definition of what is art can be summed up as anything made by humans. i have no problem when its used in memes and being open sourced etc.. the issue i have is when a human invests real time into it and then its taken and regurgitated without their permission. do you see that distinction?
I mean, I don't think many care about your personal use of art. You can take copyright images and shit post and Disney won't go suing your workplace.
But many big players do want to use this commercially and that's where a lot of these lines start to form. No matter how lawsuits go you will probably still be able to find some LLM to make Thanos fighting a volcano. It's just a matter of how/if companies can profit from it.
That's a funny argument because artists lost their shit over photography too. Now anyone can make a portrait! Photography will kill art!
Art is the biggest gate kept industry there is and I detest artists who believe only they are the chosen one.
Art is human expression. We all have a right to create what we want with whatever tools we want. They can adapt or be left behind. No sympathy from me.
Because that's not what happens, ever. You wouldn't ask a human to have their style of photographing when they don't know what a photograph even looks like.
Exactly. Artists should drop the pretentious philosophical bumbling and accept what this is, a fight for their livelihood. Which is, in every sense, completely warranted and good.
Putting blame on the technology and trying to limit public access to software will not go anywhere. Your fight for regulation needs to be with publishers and producers, not with the teen trying to make a cool new wallpaper or the office-man trying to make an aesthetic powerpoint presentation.
> they are a tool of capital to fuck over human artists
So are the copyright and intellectual property laws that artists rely on. From my perspective, you are the capital and I am the one being fucked. So are you ready to abolish all that?
Copyright owners indeed. That's what these artists are. They're copyright owners. Monopolists. They are the capital. Capitalism is all about owning property. Copyright is intellectual property. Literally imaginary property. Ownership of information, of bits, of numbers. These artists are the literal epitome of capitalism. They enjoy state granted monopolies that last multiple human lifetimes. We'll be long dead before their works enter the public domain. They want it to be this way. They want eternal rent seeking for themselves and their descendants. At least one artist has told me exactly that in discussions here on HN. They think it's fair.
They are the quintessential representation of capital. And they come here to ask us to "resist" the other forms of capital on principle.
I'm sorry but... No. I'm gonna resist them instead. It's my sincere hope that this AI technology hammers in the last nails on the coffin of copyright and intellectual property as a whole. I want all the models to leak so that it becomes literally impossible to get rid of this technology no matter how much they hate it. I want it to progress so that we can run it on our own machines, so that it'll be so ubiquitous it can't be censored or banned no matter how much they lobby for it.
Right. IP is about balancing the harms done to the public versus the incentives given to the creators. That's why discussions about "creators want this and that" produce unbalanced objectives because... what about the public? IP is meant to benefit the public domain, it is not meant to protect creator interests.
I'm glad free software came about, where software owners realised that the power they had over their users were unjust. Hopefully the same can happen to other creative fields.
The original social contract was we'd all pretend we couldn't trivially copy and reproduce "their" works so they could make some money for a decade or so and then their works would enter the public domain.
When's the last time your culture entered the public domain?
I grew up watching films like Star Wars and Home Alone, playing Super Nintendo and PlayStation games. All these corporations have already made a million fortunes off of these things. When is all this stuff gonna become public property?
What about the public? They couldn't give less of a fuck about the public. They don't even care enough to fulfill their end of the bargain which was agreed upon when this copyright nonsense was created. Oh look, our property's about to become public? Better lobby the government and get them to extend the terms. Just pull the rug from under everyone, just move the goalposts, they won't even notice.
They use their fortunes to systematically rob us of our public domain rights. Yes, rights. We have rights to "their" works which they actively deny. They're literal robber barons. Therefore copyright infringement is a moral imperative. We should all stop pretending that copyright exists because the reality is it doesn't, it's completely made up and unenforceable and there's absolutely no reason anyone should recognize it as a legitimate law. Copyright infringement is civil disobedience and morally justified.
This AI stuff is exactly what we need. It's world changing technology. It's subversive. It gives me hope.
Claiming that a single artist of modest means whose work was used for model training explicitly against their wishes.... is exactly the same as the multibillion corporation doing the training and profiting off it at fleet-of-megayachts scale.... certainly is a take, I'll give you that. If you ever quote "first they came" in a self-pitying context, remember that you deserve it.
The issue of white collar workers finding it harder to support themselves is real and valid. The solution is unionisation, social support nets and potential UBI, that includes all workers.
But campaigning for stricter IP laws only benefits the copyright owners (both you and Disney), a subset of white collar workers, and is still immoral for all the reasons that copyright is immoral in the first place. That is what we're against.
Capitalism is not defined by the relative wealth of individuals. It's defined by ownership. Capitalists own property. Factories, buildings, companies, land, goods, stock, assets, copyrights, patents... All privately owned. If you're an owner, you're a capitalist. It's that simple. You're part of the ownership class. The common serfs? They don't own, they rent.
Claiming you're not a calitalist owner just because you're a poor starving artist, when the government quite literally grants you functionally infinite monopolies over everything you create whether you want it or not, is quite the take indeed. Not my fault artists are prone to selling off the rights to that monopoly to corporations for short term profit. It's like they don't even realize the power they have and the real capitalists are only too happy to relieve them of it.
As a representative of a lot of things but hardly any capital who uses diffusion models to get something I would otherwise not pay a human artist for anyway, I testify that, the models are not exclusively what you describe them to be.
I do not support indiscriminate banning of anything and everything that can potentially be used to fuck someone over.
I did not say they were exclusively that; I said they were that.
Once we as a society have implemented a good way for the artists whose work powers these machines to survive, you can feel good about using them. Until then, frankly, you're doing something immoral by paying to use them.
By this logic we ought to start lynching artists, why they didn't care about all of those who lost their jobs making pigments, canvasses, pencils, brushes etc etc
Artists pay those people and make their jobs needed. Same as the person above claiming Duchamp didn't negotiate with the ceramics makers - yes, they absolutely did and do pay their suppliers. Artists aren't smash and grabbing their local Blick.
It is! This method of doing so has overwhelming negative externalities, though. I'd expect anyone who actually gave a shit about AI empowering people to create to spend just as much effort pushing legislation so the displaced artists don't starve on the street as a result.
Isn't high-quality open image generation almost entirely dependent on Stability releasing their foundational models for free, at great expense to them?
That's not something you'll be able to rely on long-term, there won't always be a firehose of venture capital money to subsidise that kind of charity.
The cost of training them is going down, though. Given the existence of models like Pixart, I don’t think we’ll stay dependent on corporate charity for long.
That's a terrible analogy. Until the scrapers start deleting all other copies of what what they're scraping, "stealing" the art in a traditional sense, there's no harm done in the process of training the network. Any harm done comes after that.
That's an intentional misinterpretation, I think. I mention art as an economic activity because it's primarily professional artists that are harmed by the widespread adoption of this technology.
They tried to use the labor theory early on by claiming, "real art takes hard work and time as opposed to the miniscule cpu hours computers use to make 'AI art". The worst thing AI brings to the table is amplifying these types of sentiments to control industry in their favor where they would otherwise be unheard and relegated to Instagram likes
> It's funny that these people use the langauge of communism, but apparently see artwork as purley an economic activity.
You hit the nail on the head. Copyright is, by its very nature, a "tool of capital." It's a means of creating new artificial property fiefdoms for a select few capital holders to lord over, while taking rights from anyone else who wants to engage in the practice of making art.
Everyone has their right to expression infringed upon, all so the 1% of artists can perpetually make money on things, which are ultimately sold to corporations that only pay them pennies on the dollar anyway.
You, as an indie hip hop or house musician supported by a day job, can't sample and chop some vocals or use a slice of a chord played in a song (as were common in the 80s and 90s) for a completely new work, but apparently the world is such a better place because Taylor Swift is a multimillionaire and Disney can milk the maximum value from space and superhero films.
I'd rather live in a world where anyone is free to make whatever art they want, even if everyone has to have a day job.
What do you mean? Copyright protects all creative works, and all authors of those creative works. That some have greater means to enforce was always true, and copyright doesn’t cause that, it (imperfectly) helps mitigate it. What copyright does is actually prevent them from stealing work from independent artists en masse, and force them to at least hire and pay some artists.
> I’d rather live in a world where anyone is free to make whatever art they want, even if everyone has to have a day job.
You’re suggesting abolish Copyright and/or the Berne Convention? Yeah the problem with this thinking is that then the big publishers are completely free to steal everyone’s work without paying for it. The very thing you’re complaining about would only get way way worse if we allowed anyone to “freely” make whatever art they want by taking it from others. “Anyone” means Disney too, and Disney is more motivated than you.
> You, as an indie hip hop or house musician supported by a day job, can’t sample and chop some vocals or use a slice of a chord played in a song… for a completely new work
Hehe, if you sample, you are by definition not making a completely new work. But this is a terrible argument since sampling in music is widespread and has sometimes been successfully defended in court. DJs are the best example of independent artists who need protection you can think of?
> It's a means of creating new artificial property fiefdoms for a select few capital holders to lord over, while taking rights from anyone else who wants to engage in the practice of making art.
I doubt even Disney sue people who want to make fan art. But if you want to sell said art or distribute it, they will.
Human beings and LLMs are essentially equivalent, and their processes of "learning" are essentially equivalent, yet human artists are not affected by tools like Nightshade. Odd.
Sigh. Once again: I always love it when techbros say that AI learning and human learning are exactly the same, because reading one thing at a time at a biological pace and remembering takeaway ideas rather than verbatim passages is obviously exactly the same thing as processing millions of inputs at once and still being able to regurgitate sources so perfectly that verbatim copyrighted content can be spit out of an LLM that doesn't 'contain' its training material.
I'm just glad that so many more people have caught on to the bullshit than this time last year, or even six months ago.
I really don't even get the endgame. Art gets "democratized", so anyone who doesn't want their style copied stops putting stuff on the internet, and eventually all human art is trained, so the only new contributions are genAI. Maybe we could get a few centuries worth of stuff of "robot unicorn in the style of Artist X with a flair of Y" permutations, but even ignoring the centipede, that just sounds... boring. worthless.
Since techbros are stupid: "Note that people could always do these kinds of repurposing, and it was never a problem from a copyright perspective. We have a problem now because those things are being done (1) in an automated way (2) at a billionfold greater scale (3) by companies that have vastly more power in the market than artists, writers, publishers, etc. Incidentally, these three reasons are also why AI apologists are wrong when claiming that training image generators on art is just like artists taking inspiration from prior works."
A human artist does not need to look at and memorize 100000 pictures in any span of time, period. Current AI does.
We needed huge amounts of human labor to fund and build Versailles. I'm sure many died as a result. Now we have machines that save many of those lives and labor.
That perhaps we shoildnt compare modern capitalistic societies to 18th century European royalty? I sure don't sympathize with the justification of the ability to use less labor to feed the rich.
I'm not sold on your argument. I'm not an artist but I don't see how an artist using Nightshade is breaking the law. From an anti-AI point of view, you illegally took my artwork and used it. How is it my fault you didn't understand what you were stealing?
Currently it's not legally defined what is stealing and what is fair use with respect to this, which is of course why k feel it is such a strong issue, however, poisoning data and intentionally masking it to hide said poisoning is rather blatantly illegal.
Rather clearly I think most people support individual IP protection and that's not really a contested issue, however what is fair use and where things fall in that gray area is where things do get dicey.
The "blatantly illegal" part is the part I'm not understanding. I drew something, I did some processing on it, I put it online, someone grabbed it and used it without my permission. I'm not clear on the illegal part.
> Rather clearly I think most people support individual IP protection and that's not really a contested issue
I think poking around HN or just this thread alone would tell you that at least people here don't support that.
I would recommend at least on the legal side of things looking into the Duty of Care history, it's not necessarily intuitive though I think it makes sense in the end. As another person said, oftentimes it comes down to intent (though negligence can play a role in the matter as I personally understand).
I wonder how your original claim balances with something like freedom of speech/expression? Ultimately it’s art and perhaps the Nightshade-modified version is a component of what makes the piece art?
Nightshade is intended to make minimal changes so I think that might be one argument though a difficult sell to some.
The original post is mainly about the current legality, I'm torn as to what the moral balance is, but poisoning data is similar to a lot of historical asymmetrical methods of topical influence through intentional property destruction, and that's an issue.
For example, the counterintuitive example of you poisoning a sandwich if your coworker is eating it makes you liable even if the original action is wrong on the coworker's part, if that makes sense.
It's not necessarily an intuitive legal construct but a necessary one at least.
That said, I think w.r.t. a final solution, grassroots vs big money legal will be the long-term playing field and I think efforts would be well spent on that. Though I worry about the perceived desperation of the field if people are this ready to poison open-field data in order to protect what they determine their own interests to be.
I think that's the big marker here that we should look at and be worried about, usually when ignored that tends to preclude different types of issues/problems.
Duty of Care, additionally from an alternative perspective there are a number of laws covering the transmission of data that is intended to cause harm to another system.
(I understand that this is not a popular point, but I really want to emphasize that I am talking about what is _currently legal_ right now, not at all about the ethicality of large companies using people's data. The latter is a much harder topic.
This is mainly about what the law considers to be legal or not legal and is trying to avoid the more emotional side of the topic.)
This is excellent. We need more tools like this, for text content as well. For software we need GPL 4 with ML restrictions (make your model open source or not at all). Potentially even DRM for text.
