In a post-scarcity solarpunk future, I could imagine some reasonable uses, but that’s not the world we’re living in yet.
AI art has already poisoned the creative environment. I commissioned an artist for my latest solarpunk novel, and they used AI without telling me. I had to scrap that illustration. Then the next person I tried to hire claimed they could do the work without AI but in fact they could not.
All that is to say, fuck generative AI and fuck capitalism!
Receiving stolen property is still a crime. You can’t hire an independent contractor to draw you Disney characters and use the IP to make money. That’s still illegal.
But that’s not what these generative AIs do. They use actual content for training, but all generations are unique… Just like actual art
If you go to college for art you are actively required to use specific licensed learning materials to learn from. They don’t just go get random training material off the web and go “draw like this but make it your own”. The same principles apply. The AI has no filters. It has no way of determining what is copyright infringement and what isn’t. It can’t decide what is fair use and what isn’t.
But that’s specific to universities as institutions, not art as a concept. There are plenty of artists without formal education that got inspiration from the things we saw. We could have a discussion about how internet scrapers get their data, however that’s a different conversation. AI art isn’t stealing content, it’s using existing content (in albeit questionable ways) to generate new and unique content.
@atrielienz @SleezyDizasta my opinion is if I, as an artist, can look at publicly posted content and use that to inform my own unique work then why shouldn’t an AI be able to? If I try to sell a drawing of bugs bunny, then WB can sue me, but I can sell as many bugs bunny inspired rabbit drawings as I want. That should be the rule for an algorithm too.
Because you as the artist are going to change that to make a unique work within certain legal guidelines. The fact is, the laws have not caught up to regulate this and protect artists.
Additionally though you’re not thinking about this the right way. Your work as an artist is copyrighted. Meaning you own it and the right to license it to other entities. You as the artist did not license the use of your work to the company that used it for training data to give a result similar to your work when queried.
There are LLM’s that do only use licensed work that they have purchased a license for or the rights to. Getty images is a really good example. But ChatGPT did not license anything. So everything that comes out the other end of a query is tainted by the stolen data or art that went into it.
Look up why the actors guild striked and protested to protect their art and likenesses. And then tell me you don’t feel the same way. There’s multiple lawsuits going on right now with multiple of these LLM’s that have stolen data to use as training material.
A college can’t just take your work offline and use it in their curriculum. Neither should an LLM be allowed to do that.
@atrielienz what I’m saying is that if the artwork is viewable in public, I have given the public license to hold that information in their brain and use it to influence their own output.
If a member of the public makes too similar of a replica then I can sue. We do not regulate the intake of public information into human storage/retrieval systems (brains) so why should we do that for synthetic ones?
We should only regulate the output to not reproduce art or an actors likeness etc.
If you go to college for art you are actively required to use specific licensed learning materials to learn from. They don’t just go get random training material off the web and go “draw like this but make it your own”. The same principles apply. The AI has no filters. It has no way of determining what is copyright infringement and what isn’t. It can’t decide what is fair use and what isn’t.
@atrielienz the reason they have to use specific licensed material is because they are charging rhe art student and therefore must pay for the materials they provide to the student.
But as a student, you can look at any public art you want and allow it to inform your work as long as you don’t copy. So that’s another example of the same principle: you must pay to reproduce/distribute someone else’s art for money. So we come to the same point: no reproduction, but intake is allowed.
Two things. One. You agree that they are charging the student and therefore providing a service and thereby would need to use licensed material because they are charging for that material or its use. Why is that different that a generative AI firm providing a paid service using unlicensed training data? We’re not talking about generative AI firms as individuals. They’re businesses. Making money off a training set that was acquired through means that took the IP of other individuals and business without their knowledge and consent and used it to create something that they are selling as a service.
Two. There are a myriad of reasons why companies license materials and a lot of them don’t include the direct use, redistribution of, or copying of any of that material. There’s also a number of reasons schools license materials up to and including uniformity, consistency, and to put their spin on things so to speak. That’s why you might be able to find the same art course on offer just about any higher learning institution but the one at Julliard is not going to be the same as the one at the community college of Kenosha Wisconsin. The community college can’t just get a copy of the training materials used by Julliard and reproduce those exactly. What you’re saying is just a gross oversimplification of the real reasons, and I feel like it might be on purpose at this point.
@atrielienz let’s look at writing computer code. LLMs used public copyrighted code to get really good at writing code blocks. That’s like 85% of my job, but I don’t care that they are making me obsolete because that means I can now spend more time figuring out how to do better science.
Artists should do the same. Anything that could be adequately created by thinking of a good text prompt should be done in 10 s and spend the rest of the time on hard creative stuff 🤷♀️
This is a terrible one to one comparison. I can’t even begin to tear this apart it’s so bad. LLM’S aren’t even good at writing code which is like half the reason people have to go back and fix the code they generate.
Artists don’t do art because it’s work. They do art because they like to create things. You code because it’s work which is why you don’t care.
@atrielienz so you said it right here: “…can’t just get a copy of the training materials used by Julliard and reproduce those exactly.”
They can’t reproduce, but if Juliard posted their materials online for free, then the professor at the community college could look at those materials and use that to inform their own material selection.
You are muddling up a bunch of random side issues rather than addressing the principle issue: anyone at any company can view public information.
You seem to think for free means just take it and use it to generate revenue. That’s not what it means to have something be posted on the Internet. An artist’s online portfolio isn’t free. That’s not how that’s supposed to work and you know it.
If it were these LLM’S wouldn’t shy away from using music on the Internet to train their LLM’s. At least one of these firms has literally said they don’t do this specifically because they don’t want to get into trouble with any record labels. But sure. LLM’S can steal from Getty images and the NYT and be fine. That totally makes sense.