The article says “maybe yes” for what I understood. At least a guy is making a lawsuit, we’ll see:
> The alleged wrongdoing comes down to what Butterick summarized to me as “the three ‘C’s”: The artists had not consented to have their copyrighted artwork included in the laion database; they were not compensated for their involvement, even as companies including Midjourney charged for the use of their tools; and their influence was not credited when A.I. images were produced using their work
Trying to pretend LAION has any culpability is how you know this is bunk.
It's a list of links to publically available content. The metadata is the alt-text of the linked images.
No one has the ability to demand they not be linked to, because otherwise the simple solution is "stop hosting" or "add authentication" and the ramifications of any other conclusion break the internet (EU right-to-forget would apply, but that compels action after a request, it doesn't imply linking is illegal).
Why? This is no different than software in general using third-party assets without permission. We, as in both society and the legal world, know how to deal with license violators and IP thieves.
The only question is whether we can stop confounding the discussion from spamming "AI" every-fucking-where every-fucking-time.
I believed that corporate theft from citizens was a valuable investment. Why all the commotion? Copyright is obsolete for the average person. They will optimize algorithms for free, after having fed them with user-generated, tag-labeled content for years. Now inside your Windows Desktop. Soon in Apple pods.
We the people shouldn't be wondering if this is violating the law.
We the people should be deciding if this should be allowed or not. The law is ours to write, and we should make our decision what we want it to say and then work to make sure it is implemented as we want.
This is from America, where national laws like copyright are very hard to change, due to lobbyists and both political parties thinking the other one literally works for the Devil. Your point about people deciding their own laws is great but it’s just not how our system works. Instead most law nowadays is made by judges interpreting anything controversial.
Ok, but, that system - "how things are" - is exploitative in the extreme to both people and the planet. It's unsustainable, eldritch, profoundly unjust and corrupt.
Using "how things are" (fucked) as a justification to prevent discussion of "how things could be" (amazing) is fucked. I'm sick to the teeth of seeing it.
You’re both right. It’s true the practical aspects many it hard to change laws, but I still appreciate OP’s starting from first principles.
It’s easy to get caught up in debates over how to interpret current laws and conflate what’s legal with what’s (a) ethical (b) conducive to progress. However, we shouldn’t lose sight of the fact that these laws are merely social contracts and subject to change with enough impetus and savvy.
The truth is that nobody cares. Your "We the people" Braveheart speech is just cognitive dissonance with the rest of the world if you look at the reactions irl and on the web normal people, or better said people not in the tech bubble and overwhelmed artists, are rather excited.
Whatever position you think that commenter holds is entirely created within your own mind. The comment makes just as much sense regardless of which "side" you think they're on.
That incorrectly presumes that you[1] can make any law you want, have it enforced, and not piss off the vast majority of people or damage the fabric of free society.
You can't make any law you want, and keep a free society. You can't tell people they can't train neural nets with publicly available data. You can't tell people they can't run something like stable diffusion without getting output approved by some approved entity.
This is a line in the sand. If lawmakers are coerced into passing such laws, it will consolidate almost all future power (computing, language, speech, images, and video will all be heavily dependent on potentially copyrighted AI output) in the hands of government or corporate authority.
[1] for all the people complaining about this:
a) That's the collective "you", not the parent poster.
b) The majority of people are not involved in Congress's decision to pass or not pass legislation. What moves the needle is the representative's own intuition (often flawed), or relatively small groups of people who have been motivated to contact their legislators, or corporate lobbying.
Arguing if a new tech is legal is like saying the Earth is flat.
Arguing that we can create and alter the laws and should therefore be discussing if we want this tech to be legal, that's like saying the Earth is a sphere.
Both are wrong, but treating them as equivalent is even more wrong than both of them.
> You can't make any law you want, and keep a free society. You can't tell people they can't train neural nets with publicly available data. You can't tell people they can't run something like stable diffusion without getting output approved by some approved entity.
Depends what you count as "a free society".
There are things which I'm not allowed to take photographs of or record videos of, such as the screen of a cinema while it's playing the latest superhero film.
There are communications I am not allowed to make, such as the sequence of bits required to recreate even an approximation of the new song from that famous pop singer everyone likes.
There is data I'm not allowed to be given or store, like a random stranger's medical records.
You're allowed to be an anarchist and say those things should be legal, but most people don't really care very much about any of that — closest we get is people campaigning to "keep memes legal".
Hmm.
I'm no lawyer, but there's probably one here: what even is the legal status of companies selling the service of searchable archives of screenshots with text on them and short video clips? Is that "fair use" or "probably copyright infringement but nobody cares" or "they all need to be licensed, there was a court case"? (I assume it varies by jurisdiction, an answer in any will do).
> There are things which I'm not allowed to take photographs of ...
But photography isn't illegal per se. You don't need a license to have and legally use a camera.
Taking some pictures may constitute a crime, but it won't be simply having the camera and clicking the shutter button.
> There is data I'm not allowed to be given or store, like a random stranger's medical records.
Not really. If I give you my medical PII you've committed no crime. Even if you store it you only may be liable.
> There are communications I am not allowed to make, such as the sequence of bits in copyrighted material]
If you intend to interact with the media it encodes then yes, but if you're a dev and I send you a test vector you can't be expected to know what those bits encode.
> > You can't make any law you want, and keep a free society. You can't tell people they can't train neural nets on public data
You can't tell people they can't add public numbers. Training an AI is all math and in theory could be done on paper.
Having laws that would proactively make training an AI on public data illegal would make our society not free.
Public data wouldn't mean anything, because the public wouldn't be able to read it. Training an AI is so vague that making notes about a text could qualify. If the courts tried to enforce such a law non-selectively it would criminalize everyone.
> You're allowed to be an anarchist and say those things should be legal
Anarchists don't believe more things should be legal, they think there should be less or no laws. It's quite different.
Laws are supposed to recognize and punish damages, not to broadly prohibit new technologies that sound scary when they were invented.
That’s the problem. We, the people, are not in control. We, the people, have given up control to bureaucrats who play god with permission. We, the people, make society function and our society will continue to function without the bureaucrats.
We, the people, should be demanding direct democracy.
It is the same as humans learning from art to imitate one or multiple other artists' styles. The differences are in scale, and flexibility in mixing and matching different styles (a very primitive, not-very-accurate model or description would be a vector of [0,1] values, one for each artist[1]), and which vector an AI art program happens to use as a default style for any prompt that doesn't specify something akin to "in the style of <x>"
[1] I'm not claiming these neural nets are learning models of each artist and linearly combining them, but if you squint hard enough it approaches that.
Everyone knows that the term machine learning doesn't actually have anything to do with human learning, right? It's just a set of words that were catchier than gradient descent or curve fitting.
Asserting legal rights based on a naming quirk seems pretty unfounded.
Let's try that statement again: 'It is the same as executing gradient descent over multiple artists works' That sounds a heck of a lot closer to running a Photoshop filter than learning anything now doesn't it.
I fail to see how that last sentence makes it sound like a photoshop filter. Do you mean if you can express it as an algorithm it cannot be compared to human learning?
What if we can reduce human learning to making predictions based on training data? Then it's not human learning anymore but more like a photoshop filter?
"Everyone knows that the term machine learning doesn't actually have anything to do with human learning, right?"
Do you have sources for that? They're not equal but I see a lot of parallels.
When someone successfully reduces human learning to predictions on training data, I will celebrate the Nobel prizes they are awarded for medicine, philosophy, and possibly physics and economics.
Until then maybe we shouldn't just assert that it's true.
Until then, we shouldn't just assert that it's false either.
GPT-3 is coming quite close to being able to write novel convincing research papers. With GPT-4 and GPT-5 on the horizon I don't see why language models couldn't be significantly involved in winning a Nobel prize in the future. At least I wouldn't categorically exclude that possibility.
Deep learning is definitely inspired by biological learning and also shows surprising parallels in terms of for example how language rules are learned by children.
"Machine learning doesn't have anything to do with human learning" is just not true. There are many parallels but sure they are not equal.
Theory of Mind May Have Spontaneously Emerged in Large Language Models
Theory of mind (ToM), or the ability to impute unobservable mental states to others, is central to human social interactions, communication, empathy, self-consciousness, and morality. We administer classic false-belief tasks, widely used to test ToM in humans, to several language models, without any examples or pre-training. Our results show that models published before 2022 show virtually no ability to solve ToM tasks. Yet, the January 2022 version of GPT-3 (davinci-002) solved 70% of ToM tasks, a performance comparable with that of seven-year-old children. Moreover, its November 2022 version (davinci-003), solved 93% of ToM tasks, a performance comparable with that of nine-year-old children. These findings suggest that ToM-like ability (thus far considered to be uniquely human) may have spontaneously emerged as a byproduct of language models' improving language skills.
Absolutely. Gradient descent is a fantastic technique for a lot of problems. This and many other algorithms can be performed by both humans and computers and give useful results.
Notice neither your or my comments connect any of this to learning.
After numerous times making chicken broth, something will change about your brain/body and we will say 'you learned how to season broth.' We have basically no idea what this process is, or whether it relies on gradient descent or anything else.
Machine learning is not human learning in any meaningful or legal way because laws are for humans and crucially machine is not human.
It will be the case as long as that machine lacks free will, human rights and all that important stuff. If it gets to a level where a machine is granted human rights this could be reassessed. Until then, whoever operates the machine is to whom laws apply, and in this case they violate copyright by capitalizing on others’ creative work and intellectual property through creating derivative works.
The large interests in question are guys like OpenAI/Microsoft, and it is in their interest that we continue to see AI as some human-like enigma and don’t think about the ethics of what they do. If you ask actual small artists, most do not in fact share this enthusiasm.
This is much less of a problem than people make it out to be.
For a lot of people, art is a way to connect to other human minds. If you know there is no human mind directly involved in the piece, you might not value the art the same way. No amount of code will change that.
Having said that, I think the current iteration of AI will be a fantastic tool for artists once it matures. Even if you're making some one-off disposable piece of visual/audio design for business purposes, you will want a professional to look it over.
You’ve summed up my issue with AI art. If no human was involved in creating the piece then as far as I’m concerned, it’s not art. It’s something but art it is not - and I’m not someone who is in to art at all, I just want to know that an actual person was trying to express an idea/opinion when making a piece.
Once I know a human didn’t make something my brain just looses interest. Telling if something was human or AI made us the trick.
To be clear, I’m only applying this thinking to art - if something is drawn or calculated by AI that solved a problem (circuit design, route planning etc) then I’m all for it.
This might add value to art that machines can't do yet, like oil painting. A painter could film themselves creating the art. You could go and see them painting.
If you write an 'algorithm' that copies someone's art and distributes it - then you are infringing. It doesn't matter if you 'automated it' and the word 'art' doesn't really apply so much either.
On the flip side, an artstyle can't be copyrighted. So if a model creates something that is similar, but not the same, and different enough, is it really infringing?
> I don't see how any corporation should be able to consume someone's copyrighted work without their permission and make something of it.
I'm not quite sure where I stand yet on this issue. I wonder, though, if an individual artist studies another long-dead artist's work, practices that style of painting/drawing/whatever, and then produces something in that same vein that is clearly inspired by the artist...is there a copyright issue? My instinct would be to answer 'no' as this is what art and inspiration has always been about. No one had to pay to look at Da Vinci's work or study a Renoir - the art was publicly available.
Now, with OpenAI, the capacity to 'study' an artist and produce something in their style is available to all. If the work is publicly available, is a machine studying/learning from it any different to an art student in a museum? And if OpenAI generates an image in the style of Da Vinci, should it be paying someone for doing so?
Again, my instinct is 'no'. I can understand why an artist working today might see things very differently but unless the source material was locked up or otherwise not available for people to see, study and learn from, I am not yet convinced there is a big difference between thousands of humans studying and imitating a master artist and a machine studying and imitating a master artist.
The article points to two problems, one being corporations profiting from other peoples copyrighted works with just an algorithm. Thats awful for sure, but the actual utility of content generated that way is questionable at best (Also, because it's new, there is a bit of a fad for it). The other is more existential and pertains to AI replacing humans in the cultural process. I don't think it's reasonble to assume this will happen anytime soon, given the state of technology which is just nowhere near enough capable for that. By the time it would be, you'd have a fully artifical human mind.
I agree with you for capital-A Art, but it turns out that most of content produced doesn't need to meet that bar - book covers, adverts, filler paintings, and so on. Those are the ways that a lot of people make their living. Is it OK that they'll become outdated? Maybe yes for the production of capital-A-Art, but it sure will suck for a lot of people who enjoy making their living from creative production.
From what I've seen in the art world it is common practice to build on the work of others without attribution. Is this really all that different? That said, I think these tools still implicitly depend on people to pick out good images that resonate with people, as if people don't like the image it creates it won't get shared.
1) When artists 'steal' from others, they generally build something 'new'. AI can't really create anything 'new'. Case in point: Umberto Eco's The Name Of The Rose steals the plot and outline from crime stories, steals the library idea from Borges, and steals the murderer's motive from Eco's medieval manuscript (forgot which one). Yet the outcome is something completely new. Same goes for hip-hop music; sampling is at its core, but the final music that comes out is nothing like what it samples, the samples are just a part of something new.
2) when artists steal from each other, it's generally poor artists ripping off other poor artists. No money can flow. When a million/billion-dollar company rips off poor artists to make more money via generative AI it's a different story. Money could flow but it doesn't.
Too your first point, I am not sure what constitutes new or novel here. I've seen the AI do exactly the type of borrowing from different sources you describe and producing something that seems very new to me. Though I admit, often these are under human direction with some carefully chosen sentences. Is that enough to generate something really new?
To your second point, most of the work I've seen generated from these models were done for free. It could be argued that these tools add to an artists toolbox rather than take something away. I can see for example where one poor artist could create a computer game of the same quality that it takes a triple-AAA game company to do today. Is that good or bad?
Why would you say AI can't create something new? It should be abler to merge ideas from its training data and create something like "The Name Of The Rose".
ChatGPT is able to write lyrics about a topic of my or its choice in the style of an unrelated philosopher for example. It creates something new by mixing ideas from different modalities.
a_bonobo.... I think you're digging in the right vein here, but to perhaps to make the point clearer we could separate the discussion into two:
In an AI laden world
1) How are artists relevant
2) How are artists fairly compensated
I believe the answer to 1) is as you say, simply in the word 'new'. AI functionally remixes toward a mean. You can improve the impression of this with exquisite sources or more nuanced algorithms, but the result is always, literally, average.
I believe that an artist works out of an authentic story that is much more than the sum of their influences, and is uniquely their own. This story is the value proposition that is offered to an appreciant - it is an expression of identity that rises above the noise, and that can be joined. Mechanisms of payment have changed with the gyrations of technology, culture, and power since the concept of 'art' was birthed. But as long as there is a distinctive story in an artist's work, I believe there will always be a path to compensation.
Why is it some of the worlds greatest achievers say something like: "everything I've done was built on those who came before, I'm standing on the shoulders of giants", and others say: "I and I alone made the entirety of this"?
We're all standing on the shoulders of giants, dwarves, and everything in between. That's not really arguable. Even if you create something "entirely unique" (yeah right), that somehow doesn't draw on any work done before (ha!), you can only communicate it because people made language.
You can only share it because people fed you, clothed you, educated you. Countless squintillions of interactions and evolutions, in nature, in humanity, in culture and science and art and language, in the very cosmos, went into putting you in the position you were when you created the "completely new" thing.
However, one of the things that people do is go into bubbles - religious bubbles, ego bubbles, psychotic delusions, etc. It can be powerful to give yourself all the credit for your work. It can be motivating, comforting, and generally advantageous sometimes to think that God has your back, that your tribe is chosen above others.
Think of all the performers with diva complexes - those complexes serve a role in helping them. It's absurd and kind of sad from the outside, but it works for those people to believe that they're the greatest.
But the truth that they're all forgetting is that they had teachers, helpers, warriors, workers, people of all kinds supporting them to get to the position they're in. They all had ancestors who broke their backs to bring them into the world and protect them. They've all eaten thousands of pounds of food grown with exploited labour. They've all benefited from countless artists and creators enriching the local and global culture.
So, I can see why someone might believe, or even choose to believe that they "made the entirety" of a thing... But it's not the truth - not even close. A wider and less egoistic perspective might not help them create or believe in themselves, but if we're talking about a broader truth than one person's viewpoint, it's utterly absurd, wholly delusional, to give yourself that much credit. I think it's a sign of sickness in our culture that any of this is news to people.
> “We’re not litigating image by image, we’re litigating the whole technique behind the system.”
Oh yes, "let's sue AI-anything", no time to sue one by one. Wildcard attack, not against any specific work, but against the algorithm itself. If they wanted that, they should have patented image diffusion.
And when they want to copyright their own styles, the same thing. They want a style to work like a trademark, to block any future work from using it. Another attempt to make copyright more powerful, making current holders have an advantage over newcomers.
Basically they would like copyright to get trademark and patent powers, to cover any future reimplementation or style variation. But AI isn't coming just for art, and diffusion models are useful for many other things. They are not the one and only voice that matters.
Don't think of this in terms of just art. In fact art is probably a confusing example since the creative process is a bit obscure.
Think of creating any immaterial object of value to you or others. A recipe book, a coding blog, a graphic novel, whatever.
Are you really going to go through the pains of creating that and putting it out there, if on the next day some AI startup is going to train its models on it and offer something seemingly different, but ultimately equivalent, commercially?
We're running into a huge problem for creators here, and it'll have to be solved with new laws around training models on original content some way or the other.
If we want to still have a creative industry going forward, that is.
Bill Gates says that ChatGPT will change the world, make Jobs more efficient. Weird that people don't feel the same about Stable Diffusion even though it will do the exact same thing
Bill is obviously reached, a statute of God. He is starting a Climate religion as we speak, he should hire Tom Cruise for a start. He knows a lot about this profitable startup idea.
To be fair I am also biased: I am responsible for creating an interface that uses Stable Diffusion https://capsizegames.itch.io/ai-runner, but the program is a result of my bias not the root cause.
I am old and biased to my core. My personal experience is that in reality the winner is the boring tech which people use because of real life needs and goals.
All of it related to their survival and well-being. Generating some pictures is entertaining and may be valuable for some, but when we talk about art, the process is more important than the results. Art is personal human expression. When my brushes are painting more than me, this is their art. Not mine.
All of this will give yet another tool in the corporate belt for reducing employment on a grand scale. And this is not the industrial revolution, which provided some alternative to the craftsman of the era.
This is the removal of human value in the production loop.
Edit: Did you not read the ART part of my response?
Boring is MS Excel, a real game changer.
what do you mean by winner, and what do you mean by "boring" tech?
advanced algorithms and the things you can make with them are pretty cool, and i'd argue the boring is subjective (ai could be considered boring).
"ai" is new. new markets and opportunities and areas to expand knowledge have opened up. i'm excited by it like i was excited by computers in the 80s and i'm dedicated to building applications that put it into the hands of as many people as possible.
The important thing here is not if jobs are getting more efficient or not, it's that everyone gets the same boost. All companies will get the same AI, and that makes people the differentiating factor. Let's stop with the theories of job loss by AI automation, competition will have humans+AI and beat your company that uses AI alone.
The early Google for example was also an AI. But it was open to everyone, nobody got exclusive Google access. So we all benefited equally and there were no overall job losses.
Soon after chess AI beat all humans, people found that human-AI-hybrid players were still better than pure AI.
That time has past, and now having humans in the team only make the AI worse.
Will that happen here? Possibly, but not certain, even with superhuman capacity:
I suspect, but am not certain, that art is a peacock tail for humans — a signal where the point is to be extravagant and expensive — which means that all automation of it, including pre-AI automation, results in that-which-is-now-cheap no longer counting as "art" in the minds of increasingly many humans.
Some of the famous renaissance still life paintings are recreated a million times a day as people use their phones to take photos of their food. The hard ones are art, the common ones are…?
‘Microsoft co-founder believes that his company’s own investment in the AI startup that created ChatGPT will change the world.’
It is open source AI models that are large in parameters, transparent and smaller in size that are the ones that will permanently change the world and will disrupt OpenAI’s closed AI SaaS business plans.
We don't want any creative human input, because it is expensive and gives opportunity to have a middle class. Sam is regularly at Davos. Ask the Davos futurists about their vision of tomorrow in which everything will be fine and dandy.
Did you see Sam's appearance at the Microsoft presentation? It is all about safety, he said. Safety.
> and offer something seemingly different, but ultimately equivalent, commercially?
> it'll have to be solved with new laws
Tbh, if an AI will fill out a need I had for cheaper than a human would, thn I'd see that as positive. Be it image asset creation, ghost writing or whatever. Restricting that with laws seems like it would only cut innovation because it takes away jobs. Kinda like forbidding robot assemblers in industry.
Tough, take that from a person on the consumer side of creative works.
If we replace a big chunk of our media with derivative works created by copyright laundering engines, we're destroying the incentives to be original and creative.
But we absolutely need originality and creativity!
Living in a world where everything looks sort of like SEO spam sounds like a dystopia to me.
Note, I don't think those technologies should or will be banned. There just have to be rules in place that allow creators to opt-in to having their work used as training data, with appropriate remuneration.
If that makes some AI business models unworkable, that's fine with me as long as we don't destroy our creative and inventive people's livelihoods.
Let's be honest most of that art is intended to make you buy things you don't really need. It's already a SEO spam all around. The internet is not aesthetically pleasing at all.
Real art will survive, don't worry. True creativity emerges naturally from human experience and is not dependent on a specific industry. Also now it will be more accessible.
I think baboon Freud agrees [id]!
If an artists work is used in training an AI model without a license, should it be legal to distribute the images generated by that AI model?
What about distributing the AI model itself?
Would you use an online service if the Terms and Conditions stated that "You agree that content uploaded to this service may be used in training AI models without compensation, rights or acknowledgement"?
The most interesting thing to me is how these systems are text-based. Before you could pick up a mouse/pencil and draw directly. Now we're describing art. Same thing will happen to code. This will increase attention to language, which tends to increase social cohesion, not to mention productivity gains. Maybe even one day advanced mathematical theories will be abstracted away from numbers.
The main problem seem to be the inclusion of the artist name in the metadata of the training images. This makes the style of individual, currently active artists exploitable in ways that affect the artists negatively.
One solution could be to allow the use of the images, but disallow the use of the artist name in the training set. From a privacy standpoint it seems reasonable to disagree to having your name and artistic characteristics connected and encoded into an AI model, even if it's legal to use the art itself as training data.
The Miyazaki quote really hit home: “I strongly feel that this is an insult to life itself.”
I think most AI art misses the point entirely. Art is really about the person / people that make the work and their response to the time and context in which they live. It's usually a struggle to bring something new into the world, but it does feel magical and exciting - it seems to come from a very different place than interpolating past influences. AI art takes away all the magic, and without the sense of a person I think it has no artistic or aesthetic value - it just adds visual noise to the world. I think it will, unfortunately, disincentivise people to make new work, as although there is a world of difference between AI art and good art, it can be tricky to tell the difference at first glance, and to make anything good takes years of dedication and effort.
Sure, it can come from a boring, unoriginal place without any investment and it could be bad art. Work like that is just a waste of time for everyone. People wouldn't listen to music that was created with the same indifference, it would sound awful. I agree it's not a very important question though, it's just a shame that we will have to look at the images AI generates.
95 comments
[ 3.0 ms ] story [ 152 ms ] thread> The alleged wrongdoing comes down to what Butterick summarized to me as “the three ‘C’s”: The artists had not consented to have their copyrighted artwork included in the laion database; they were not compensated for their involvement, even as companies including Midjourney charged for the use of their tools; and their influence was not credited when A.I. images were produced using their work
It's a list of links to publically available content. The metadata is the alt-text of the linked images.
No one has the ability to demand they not be linked to, because otherwise the simple solution is "stop hosting" or "add authentication" and the ramifications of any other conclusion break the internet (EU right-to-forget would apply, but that compels action after a request, it doesn't imply linking is illegal).
Same for this scraping algorithms, somebody has written them.
I don’t have a solution, it’s only a consideration.
Why? This is no different than software in general using third-party assets without permission. We, as in both society and the legal world, know how to deal with license violators and IP thieves.
The only question is whether we can stop confounding the discussion from spamming "AI" every-fucking-where every-fucking-time.
We the people should be deciding if this should be allowed or not. The law is ours to write, and we should make our decision what we want it to say and then work to make sure it is implemented as we want.
Using "how things are" (fucked) as a justification to prevent discussion of "how things could be" (amazing) is fucked. I'm sick to the teeth of seeing it.
It’s easy to get caught up in debates over how to interpret current laws and conflate what’s legal with what’s (a) ethical (b) conducive to progress. However, we shouldn’t lose sight of the fact that these laws are merely social contracts and subject to change with enough impetus and savvy.
You can't make any law you want, and keep a free society. You can't tell people they can't train neural nets with publicly available data. You can't tell people they can't run something like stable diffusion without getting output approved by some approved entity.
This is a line in the sand. If lawmakers are coerced into passing such laws, it will consolidate almost all future power (computing, language, speech, images, and video will all be heavily dependent on potentially copyrighted AI output) in the hands of government or corporate authority.
[1] for all the people complaining about this:
a) That's the collective "you", not the parent poster.
b) The majority of people are not involved in Congress's decision to pass or not pass legislation. What moves the needle is the representative's own intuition (often flawed), or relatively small groups of people who have been motivated to contact their legislators, or corporate lobbying.
In order to do so, we need the conversation and we need to educate inorder to have the conversation in the first place.
Law is evolving.
OPs "we" _is_ the vast majority of people. You are answering as if OP said "I".
Of course OPs implementation is complicated, but they are not presuming you can make any law you want.
Arguing if a new tech is legal is like saying the Earth is flat.
Arguing that we can create and alter the laws and should therefore be discussing if we want this tech to be legal, that's like saying the Earth is a sphere.
Both are wrong, but treating them as equivalent is even more wrong than both of them.
> You can't make any law you want, and keep a free society. You can't tell people they can't train neural nets with publicly available data. You can't tell people they can't run something like stable diffusion without getting output approved by some approved entity.
Depends what you count as "a free society".
There are things which I'm not allowed to take photographs of or record videos of, such as the screen of a cinema while it's playing the latest superhero film.
There are communications I am not allowed to make, such as the sequence of bits required to recreate even an approximation of the new song from that famous pop singer everyone likes.
There is data I'm not allowed to be given or store, like a random stranger's medical records.
You're allowed to be an anarchist and say those things should be legal, but most people don't really care very much about any of that — closest we get is people campaigning to "keep memes legal".
Hmm.
I'm no lawyer, but there's probably one here: what even is the legal status of companies selling the service of searchable archives of screenshots with text on them and short video clips? Is that "fair use" or "probably copyright infringement but nobody cares" or "they all need to be licensed, there was a court case"? (I assume it varies by jurisdiction, an answer in any will do).
But photography isn't illegal per se. You don't need a license to have and legally use a camera.
Taking some pictures may constitute a crime, but it won't be simply having the camera and clicking the shutter button.
> There is data I'm not allowed to be given or store, like a random stranger's medical records.
Not really. If I give you my medical PII you've committed no crime. Even if you store it you only may be liable.
> There are communications I am not allowed to make, such as the sequence of bits in copyrighted material]
If you intend to interact with the media it encodes then yes, but if you're a dev and I send you a test vector you can't be expected to know what those bits encode.
> > You can't make any law you want, and keep a free society. You can't tell people they can't train neural nets on public data
You can't tell people they can't add public numbers. Training an AI is all math and in theory could be done on paper.
Having laws that would proactively make training an AI on public data illegal would make our society not free.
Public data wouldn't mean anything, because the public wouldn't be able to read it. Training an AI is so vague that making notes about a text could qualify. If the courts tried to enforce such a law non-selectively it would criminalize everyone.
> You're allowed to be an anarchist and say those things should be legal
Anarchists don't believe more things should be legal, they think there should be less or no laws. It's quite different.
Laws are supposed to recognize and punish damages, not to broadly prohibit new technologies that sound scary when they were invented.
"We the people" means nothing. You are not in control.
We, the people, should be demanding direct democracy.
https://www.youtube.com/watch?v=5pIVVpoz5zk
It is the same as humans learning from art to imitate one or multiple other artists' styles. The differences are in scale, and flexibility in mixing and matching different styles (a very primitive, not-very-accurate model or description would be a vector of [0,1] values, one for each artist[1]), and which vector an AI art program happens to use as a default style for any prompt that doesn't specify something akin to "in the style of <x>"
[1] I'm not claiming these neural nets are learning models of each artist and linearly combining them, but if you squint hard enough it approaches that.
Small independent copyright interests trying to eke out a living from their art will also argue this.
In truth both sides are a mix of both individuals and monied interests but this rhetorical tactic seems to be irresistible.
Asserting legal rights based on a naming quirk seems pretty unfounded.
Let's try that statement again: 'It is the same as executing gradient descent over multiple artists works' That sounds a heck of a lot closer to running a Photoshop filter than learning anything now doesn't it.
What if we can reduce human learning to making predictions based on training data? Then it's not human learning anymore but more like a photoshop filter?
"Everyone knows that the term machine learning doesn't actually have anything to do with human learning, right?" Do you have sources for that? They're not equal but I see a lot of parallels.
Until then maybe we shouldn't just assert that it's true.
GPT-3 is coming quite close to being able to write novel convincing research papers. With GPT-4 and GPT-5 on the horizon I don't see why language models couldn't be significantly involved in winning a Nobel prize in the future. At least I wouldn't categorically exclude that possibility.
Unfortunately the answer is definitely no. People have heard the terms “learning” and “train” in this context and have anthropomorphized weak AI.
The level of ignorance on display is sad to witness.
"Machine learning doesn't have anything to do with human learning" is just not true. There are many parallels but sure they are not equal.
Theory of Mind May Have Spontaneously Emerged in Large Language Models
Theory of mind (ToM), or the ability to impute unobservable mental states to others, is central to human social interactions, communication, empathy, self-consciousness, and morality. We administer classic false-belief tasks, widely used to test ToM in humans, to several language models, without any examples or pre-training. Our results show that models published before 2022 show virtually no ability to solve ToM tasks. Yet, the January 2022 version of GPT-3 (davinci-002) solved 70% of ToM tasks, a performance comparable with that of seven-year-old children. Moreover, its November 2022 version (davinci-003), solved 93% of ToM tasks, a performance comparable with that of nine-year-old children. These findings suggest that ToM-like ability (thus far considered to be uniquely human) may have spontaneously emerged as a byproduct of language models' improving language skills.
Consider the act of adding salt to chicken broth for example but I can quite easily think of dozens of other things.
Notice neither your or my comments connect any of this to learning.
After numerous times making chicken broth, something will change about your brain/body and we will say 'you learned how to season broth.' We have basically no idea what this process is, or whether it relies on gradient descent or anything else.
It will be the case as long as that machine lacks free will, human rights and all that important stuff. If it gets to a level where a machine is granted human rights this could be reassessed. Until then, whoever operates the machine is to whom laws apply, and in this case they violate copyright by capitalizing on others’ creative work and intellectual property through creating derivative works.
The large interests in question are guys like OpenAI/Microsoft, and it is in their interest that we continue to see AI as some human-like enigma and don’t think about the ethics of what they do. If you ask actual small artists, most do not in fact share this enthusiasm.
For a lot of people, art is a way to connect to other human minds. If you know there is no human mind directly involved in the piece, you might not value the art the same way. No amount of code will change that.
Having said that, I think the current iteration of AI will be a fantastic tool for artists once it matures. Even if you're making some one-off disposable piece of visual/audio design for business purposes, you will want a professional to look it over.
Once I know a human didn’t make something my brain just looses interest. Telling if something was human or AI made us the trick.
To be clear, I’m only applying this thinking to art - if something is drawn or calculated by AI that solved a problem (circuit design, route planning etc) then I’m all for it.
It doesn't matter what 'a lot of people think art is' it matters what artists think, and what our rights are.
I don't see how any corporation should be able to consume someone's copyrighted work without their permission and make something of it.
AI will be fantastic for doing 'art' but it won't be good for 'artists' trying to make a living.
If OpenAI wants to use some artists material they can pay for it.
Otherwise, use the stuff that's not protected, make their own art, etc..
I'm not quite sure where I stand yet on this issue. I wonder, though, if an individual artist studies another long-dead artist's work, practices that style of painting/drawing/whatever, and then produces something in that same vein that is clearly inspired by the artist...is there a copyright issue? My instinct would be to answer 'no' as this is what art and inspiration has always been about. No one had to pay to look at Da Vinci's work or study a Renoir - the art was publicly available.
Now, with OpenAI, the capacity to 'study' an artist and produce something in their style is available to all. If the work is publicly available, is a machine studying/learning from it any different to an art student in a museum? And if OpenAI generates an image in the style of Da Vinci, should it be paying someone for doing so?
Again, my instinct is 'no'. I can understand why an artist working today might see things very differently but unless the source material was locked up or otherwise not available for people to see, study and learn from, I am not yet convinced there is a big difference between thousands of humans studying and imitating a master artist and a machine studying and imitating a master artist.
They definitely disagree with your point of view.
1) When artists 'steal' from others, they generally build something 'new'. AI can't really create anything 'new'. Case in point: Umberto Eco's The Name Of The Rose steals the plot and outline from crime stories, steals the library idea from Borges, and steals the murderer's motive from Eco's medieval manuscript (forgot which one). Yet the outcome is something completely new. Same goes for hip-hop music; sampling is at its core, but the final music that comes out is nothing like what it samples, the samples are just a part of something new.
2) when artists steal from each other, it's generally poor artists ripping off other poor artists. No money can flow. When a million/billion-dollar company rips off poor artists to make more money via generative AI it's a different story. Money could flow but it doesn't.
To your second point, most of the work I've seen generated from these models were done for free. It could be argued that these tools add to an artists toolbox rather than take something away. I can see for example where one poor artist could create a computer game of the same quality that it takes a triple-AAA game company to do today. Is that good or bad?
ChatGPT is able to write lyrics about a topic of my or its choice in the style of an unrelated philosopher for example. It creates something new by mixing ideas from different modalities.
In an AI laden world
1) How are artists relevant
2) How are artists fairly compensated
I believe the answer to 1) is as you say, simply in the word 'new'. AI functionally remixes toward a mean. You can improve the impression of this with exquisite sources or more nuanced algorithms, but the result is always, literally, average.
I believe that an artist works out of an authentic story that is much more than the sum of their influences, and is uniquely their own. This story is the value proposition that is offered to an appreciant - it is an expression of identity that rises above the noise, and that can be joined. Mechanisms of payment have changed with the gyrations of technology, culture, and power since the concept of 'art' was birthed. But as long as there is a distinctive story in an artist's work, I believe there will always be a path to compensation.
And which is closer to the truth?
You can only share it because people fed you, clothed you, educated you. Countless squintillions of interactions and evolutions, in nature, in humanity, in culture and science and art and language, in the very cosmos, went into putting you in the position you were when you created the "completely new" thing.
However, one of the things that people do is go into bubbles - religious bubbles, ego bubbles, psychotic delusions, etc. It can be powerful to give yourself all the credit for your work. It can be motivating, comforting, and generally advantageous sometimes to think that God has your back, that your tribe is chosen above others.
Think of all the performers with diva complexes - those complexes serve a role in helping them. It's absurd and kind of sad from the outside, but it works for those people to believe that they're the greatest.
But the truth that they're all forgetting is that they had teachers, helpers, warriors, workers, people of all kinds supporting them to get to the position they're in. They all had ancestors who broke their backs to bring them into the world and protect them. They've all eaten thousands of pounds of food grown with exploited labour. They've all benefited from countless artists and creators enriching the local and global culture.
So, I can see why someone might believe, or even choose to believe that they "made the entirety" of a thing... But it's not the truth - not even close. A wider and less egoistic perspective might not help them create or believe in themselves, but if we're talking about a broader truth than one person's viewpoint, it's utterly absurd, wholly delusional, to give yourself that much credit. I think it's a sign of sickness in our culture that any of this is news to people.
Oh yes, "let's sue AI-anything", no time to sue one by one. Wildcard attack, not against any specific work, but against the algorithm itself. If they wanted that, they should have patented image diffusion.
And when they want to copyright their own styles, the same thing. They want a style to work like a trademark, to block any future work from using it. Another attempt to make copyright more powerful, making current holders have an advantage over newcomers.
Basically they would like copyright to get trademark and patent powers, to cover any future reimplementation or style variation. But AI isn't coming just for art, and diffusion models are useful for many other things. They are not the one and only voice that matters.
Are you really going to go through the pains of creating that and putting it out there, if on the next day some AI startup is going to train its models on it and offer something seemingly different, but ultimately equivalent, commercially?
We're running into a huge problem for creators here, and it'll have to be solved with new laws around training models on original content some way or the other.
If we want to still have a creative industry going forward, that is.
i only quote him to say: high profile people are willing to make happy predictions about text generators but not art generators. biased and stupid.
I personally have no interest in generative art at all.
Just received my new handmade brushes from Escoda. Waiting for a spring to paint me some landscapes.
I get my inspiration just as a generative AI, from data set from countless artists before me.
The difference?
I am still human and love the process which is not driven by the profit margins or time to market optimization.
I am old and biased to my core. My personal experience is that in reality the winner is the boring tech which people use because of real life needs and goals.
All of it related to their survival and well-being. Generating some pictures is entertaining and may be valuable for some, but when we talk about art, the process is more important than the results. Art is personal human expression. When my brushes are painting more than me, this is their art. Not mine.
All of this will give yet another tool in the corporate belt for reducing employment on a grand scale. And this is not the industrial revolution, which provided some alternative to the craftsman of the era.
This is the removal of human value in the production loop.
Edit: Did you not read the ART part of my response? Boring is MS Excel, a real game changer.
advanced algorithms and the things you can make with them are pretty cool, and i'd argue the boring is subjective (ai could be considered boring).
"ai" is new. new markets and opportunities and areas to expand knowledge have opened up. i'm excited by it like i was excited by computers in the 80s and i'm dedicated to building applications that put it into the hands of as many people as possible.
The important thing here is not if jobs are getting more efficient or not, it's that everyone gets the same boost. All companies will get the same AI, and that makes people the differentiating factor. Let's stop with the theories of job loss by AI automation, competition will have humans+AI and beat your company that uses AI alone.
The early Google for example was also an AI. But it was open to everyone, nobody got exclusive Google access. So we all benefited equally and there were no overall job losses.
That time has past, and now having humans in the team only make the AI worse.
Will that happen here? Possibly, but not certain, even with superhuman capacity:
I suspect, but am not certain, that art is a peacock tail for humans — a signal where the point is to be extravagant and expensive — which means that all automation of it, including pre-AI automation, results in that-which-is-now-cheap no longer counting as "art" in the minds of increasingly many humans.
Some of the famous renaissance still life paintings are recreated a million times a day as people use their phones to take photos of their food. The hard ones are art, the common ones are…?
Making human-AI teams win sounds like a new challenge where AI could contribute.
‘Microsoft co-founder believes that his company’s own investment in the AI startup that created ChatGPT will change the world.’
It is open source AI models that are large in parameters, transparent and smaller in size that are the ones that will permanently change the world and will disrupt OpenAI’s closed AI SaaS business plans.
Did you see Sam's appearance at the Microsoft presentation? It is all about safety, he said. Safety.
> it'll have to be solved with new laws
Tbh, if an AI will fill out a need I had for cheaper than a human would, thn I'd see that as positive. Be it image asset creation, ghost writing or whatever. Restricting that with laws seems like it would only cut innovation because it takes away jobs. Kinda like forbidding robot assemblers in industry.
Tough, take that from a person on the consumer side of creative works.
But we absolutely need originality and creativity!
Living in a world where everything looks sort of like SEO spam sounds like a dystopia to me.
Note, I don't think those technologies should or will be banned. There just have to be rules in place that allow creators to opt-in to having their work used as training data, with appropriate remuneration.
If that makes some AI business models unworkable, that's fine with me as long as we don't destroy our creative and inventive people's livelihoods.
id: https://cdn.discordapp.com/attachments/1073730007437955135/1...
What about distributing the AI model itself?
Would you use an online service if the Terms and Conditions stated that "You agree that content uploaded to this service may be used in training AI models without compensation, rights or acknowledgement"?
One solution could be to allow the use of the images, but disallow the use of the artist name in the training set. From a privacy standpoint it seems reasonable to disagree to having your name and artistic characteristics connected and encoded into an AI model, even if it's legal to use the art itself as training data.
I think most AI art misses the point entirely. Art is really about the person / people that make the work and their response to the time and context in which they live. It's usually a struggle to bring something new into the world, but it does feel magical and exciting - it seems to come from a very different place than interpolating past influences. AI art takes away all the magic, and without the sense of a person I think it has no artistic or aesthetic value - it just adds visual noise to the world. I think it will, unfortunately, disincentivise people to make new work, as although there is a world of difference between AI art and good art, it can be tricky to tell the difference at first glance, and to make anything good takes years of dedication and effort.
https://sphericalbullshit.wordpress.com/2022/12/11/is-ai-art...