Proposing a ban on AI-generated content

Status
Not open for further replies.
I'm not contradicting myself in the slightest, and this conversation still isn't about what is and isn't real art.
Yes it is; you are just trying to define art you don't like as not Real Art while also trying to shut down your opponents by declaring that the conversation isn't happening. With copious insults thrown in.

We would never say of someone who told a human artist to, say, "draw me a foxgirl cleric with a glaive" that they were making the art.
But we would say that of somebody who took nature photographs and decided which ones looked best, which is a lot closer to how AI art works.
 
You've certainly claimed it was no different. Haven't proven that quite yet though.

I believe I have. After all, we humans use other peoples art as reference, to learn from how to compose. We create an impression how it comes together, and then apply it to our ideas.

So far, counter-argument has been that there is some special feature of humans that converts "paint on canvas" to "art" that AI lacks, so AI just produces "paint on canvas".

No it wasn't. The basis of my argument was and is (as I've reminded you constantly throughout this conversation) that my objections to AI art are that the systems that generate it, at current, are unethical as they scrape artists' work without permission.

That you've refused to acknowledge this point makes me believe you have no counterargument against this, and that the 'real art' diversion is just that. A diversion.

I explained the differences because you keep bringing up comparisons to artists making pictures, but ironically enough the quote you just used proves my point above - that my argument is about the ethical concerns because it's a system that directly uses other artists' work in order to create an image. That was the purpose of the comparison, and tracing is the best comparison to reach for.

The fact that you think my complaint is about what 'real art' is means you don't understand my argument or position. Not that you disagree with it, but that you don't understand it to begin with.

And again, we go back to memes. Because what is the difference between AI "scraping" internet for publicly available images, and people taking this image:




And drawing these?



What is the difference? Because under the argument you present, these are "unethical" creations. It targets memes and I kinda doubt you are advocating that memes are unethical.
 
Last edited:
They use images as reference to figure out that when person ask"dog runnin on a green hill" what those elements are and how to position them.
So does rotoscoping, which is an animation technique which is based on… *drumroll* tracing images. It is used to figure out "how to position" elements in motion, sure, but it's specifically done through tracing.

As a rule of thumb, if your argument for why something shouldn't be considered plagiarism can be used to excuse actual plagiarism, then your argument doesn't work.

But we would say that of somebody who took nature photographs and decided which ones looked best, which is a lot closer to how AI art works.
But it's not, though. AI art isn't in any way comparable to photography, and even if you compare it to photo editing instead, which would be more on the mark, the person deciding what looks best would still be less of an artist and more of a commissioner.

And again, we go back to memes.
If your argument is "are AI generated meme images real memes", then the answer is "sure, why wouldn't they be". What art has to do with anything?
 
But it's not, though. AI art isn't in any way comparable to photography, and even if you compare it to photo editing instead, which would be more on the mark, the person deciding what looks best would still be less of an artist and more of a commissioner.
Why? They both involve applying aesthetic judgment to the output on nonhuman processes.

And I can't even figure out why you think it's like photo editing. If anything human-made art is closer to that than AI art, unless somebody edits it.

No, I for one would call that curating the actual photographer's work.
That's what such photographers do.
 
Yes it is; you are just trying to define art you don't like as not Real Art while also trying to shut down your opponents by declaring that the conversation isn't happening. With copious insults thrown in.

But we would say that of somebody who took nature photographs and decided which ones looked best, which is a lot closer to how AI art works.

I explained the differences between photography and AI art, that you didn't understand me is not my problem. I don't think this is really worth reiterating everything I've said verbatim at this point, so just go back and read my posts.

Also
This conversation still isn't about what is and isn't real art. I'm going to begin quoting this portion of this reply at you every time you reply to me and ignore the 'ethics' portion of the conversation.


Why? They both involve applying aesthetic judgment to the output on nonhuman processes.

And I can't even figure out why you think it's like photo editing. If anything human-made art is closer to that than AI art, unless somebody edits it.

That's what such photographers do.
I literally explained what photographers do.
They don't just take a billion pictures and then just decide what looks best. That you think that's what artistic photography is about shows you literally do not understand photography as an artform despite my best attempt at explaining it to you.

And not even like "oh you took a picture that's not real art" I mean you don't understand what photography actually is!

[The counter-argument has been that there is some special feature of humans that converts "paint on canvas" to "art" that AI lacks, so AI just produces "paint on canvas".]
[...]
[... Jojo Memes]
Counterargument to what? I'm still not here to talk about what is and isn't Real Art. I'm not interested in discussing what is 'real art', and you're not doing a good job of convincing me that it's relevant to this thread. It's kind-of irrelevant to, in my mind, most actual discussions of the ethics of AI art, and the ethical aspects of AI art are (again, to me) the best reasons to consider policies regarding them.

That you subscribe to an overly simplistic notion of what a person is and have some particularly funny concepts about the differences between a computer and a human... that's more for a corncob thread. Please don't pretend my argument has anything to do with 'the specialness of humans' - calling our minds as algorithms is inaccurate and... er... dehumanizing? Because 'the human brain is a computer' isn't supposed to be a literal statement of fact. At best you could argue about the processes of a human brain, but comparing, say, the mechanics of serotonin reuptake to SNMPD would be extremely bizarre.

You're taking some metaphors as literal and it's really kinda not great?
 
Last edited:
If your argument is "are AI generated meme images real memes", then the answer is "sure, why wouldn't they be". What art has to do with anything?

Argument is that if using, again, publicly available images is "unethical", then does not under this definition memes be unethical, considering how many of them often directly trace these images? Like, are those images I list "unethical" because they use the original?

Because if not, why is suddenly for AI researchers to use them?

So does rotoscoping, which is an animation technique which is based on… *drumroll* tracing images. It is used to figure out "how to position" elements in motion, sure, but it's specifically done through tracing.

As a rule of thumb, if your argument for why something shouldn't be considered plagiarism can be used to excuse actual plagiarism, then your argument doesn't work.

Man, can't wait for argument that rotorscoping is not art. Because that is what you are saying here. Also, again no: art models don't trace original images. They use impression of the original. Just like humans do. We don't have "original.png" in our mind telling us how exactly to copy original when we start working, we have impression of the original. Because that is how human brain stores data.

Counterargument to what? I'm still not here to talk about what is and isn't Real Art. I'm not interested in discussing what is 'real art', and you're not doing a good job of convincing me that it's relevant to this thread. It's kind-of irrelevant to, in my mind, most actual discussions of the ethics of AI art, as the ethical reasons of AI art are (again, to me) the best reasons to consider policies regarding them.

JoJo memes have nothing to do with art. You are confusing two separate responses. Again: are people who draw memes based on publicly available images being unethical for using someone else art as reference? And if not, what exactly is difference "humans reference image on internet" vs. "AI references image on internet" when we speak about ethics?
 
Last edited:
The people being "spiteful" are the anti-AI crowd calling their opponents thieves and in general showing contempt for them.
Maybe if you don't want people to scorn at you, you should stop claiming that being able to chose the one picture that please you the most in a set definitively puts you on the same level as people who actually bother to create things.
 
Last edited:
Maybe if you don't want people to scron at you, you should stop claiming that being able to chose the one picture that please you the most in a set definitively puts you on the same level as people who actually bother to create things.
I never said that in the first place, nor is what "level" somebody on a relevant issue. This isn't about status games.
 
Why? They both involve applying aesthetic judgment to the output on nonhuman processes.

And I can't even figure out why you think it's like photo editing. If anything human-made art is closer to that than AI art, unless somebody edits it.
Because it's not taking actual photos, it's making a collage of a subset of actual photos from its training data. As AI art advocates keep mentioning, it's not literally cut-and-paste. Due to the way neural networks work, it's not plausible to expect it to spit out an actual photo instead of something that looks like a photo. In that regard, at least, they really are similar to humans.

Also, "nonhuman" in "applying aesthetic judgment to the output on nonhuman processes" is doing hell of a lot of work in that sentence. My aesthetic judgment is that most of Van Gogh's paintings look bad, except for some naturemorts and a draft of a human skeleton that I saw on wikipedia, but taking one of those images and saying "I like this" wouldn't make me an artist, so why would doing the same with procedurally generated pictures be any different except by assuming humans be special by default?

Man, can't wait for argument that rotorscoping is not art. Because that is what you are saying here. Also, again no: art models don't trace original images. They use impression of the original. Just like humans do. We don't have "original.png" in our mind telling us how exactly to copy original when we start working, we have impression of the original. Because that is how human brain stores data.
You keep doing that thing where you take an argument about non-originality of AI generated art and pretend that it's about artistic value. I wonder why.

Also, you know what else is an "impression of the original image"? A jpeg-compressed image. Believe it or not, they aren't literally same! Lossy image compression is, by definition, lossy. That doesn't mean that converting a bunch of PNGs of Sonic the Hedgehog characters into JPEG before making a collage of their body parts is any more original than doing the same with PNGs themselves.
 
You keep doing that thing where you take an argument about non-originality of AI generated art and pretend that it's about artistic value. I wonder why.

Also, you know what else is an "impression of the original image"? A jpeg-compressed image. Believe it or not, they aren't literally same! Lossy image compression is, by definition, lossy. That doesn't mean that converting a bunch of PNGs of Sonic the Hedgehog characters into JPEG before making a collage of their body parts is any more original than doing the same with PNGs themselves.

You say I am being obtuse, yet here you try to argue that image, that still maintains same appearence, is same as having general impression.

Tell me, can you get a close enough copy of, say, Batman comic out of art generator that I can not tell which is the original and which is the generated copy?

Because with image formats, you can't tell which is original as long as there is no major loss of data.

What you keep trying to argue here is that MLs are just copy-pasting blurry images. That is not what they are doing.
 
Last edited:
Argument is that if using, again, publicly available images is "unethical", then does not under this definition memes be unethical, considering how many of them often directly trace these images? Like, are those images I list "unethical" because they use the original?

Publicly-availability is not a license to use them for your own purposes. Memes are shared as-is - and if you want to argue the ethics of sharing images from somewhere else, then cool but that doesn't really matter here. However, using artists' art by scraping and downloading and shoveling it into your computer in order to spit out an image using their work (and style) is different than reposting memes.

If I'm making a work that's a multi-layered reference to some multiple things, that's my creative output using an idea and my own labor. An AI art generator takes an idea, but it uses others' works to composite an image.

Your 'then meme redraws should be banned too' idea fixates on the unoriginality of the idea, not the labor being performed (and how it's being performed). The ethics issue is how that's being performed - chiefly, the AI is being fed others' works and that is being used to composite an image, and there is no consent in the use of those works for compositing.

Man, can't wait for argument that rotorscoping is not art. Because that is what you are saying here. Also, again no: art models don't trace original images. They use impression of the original. Just like humans do. We don't have "original.png" in our mind telling us how exactly to copy original when we start working, we have impression of the original. Because that is how human brain stores data.

Actually those images you linked aren't traced (at least, as far as I can tell, off of the original image). And I'll restate this so you understand: I used the argument that AI art is more like tracing as a point of differentiation (though I think 'compositing' is more accurate)... and compositing isn't the same as drawing something from reference, experience, or knowledge. Compositing is taking multiple elements and making a whole thing out of its parts.

And before you say 'mosaics', I'd argue that a mosaic made out of other artists' works without permission should also be regulated (at least credited as such, with individual artists' works).

(edit: Okay the Chad versus Virgin meme is traced, but the others aren't)

[JoJo memes have nothing to do with art.]

Minor digression - I have a legitimate serious argument that these JoJo redraws have some actual artistic merit, but that's neither here nor there. I do want to sorta make a corncob thread about it now though.
 
Last edited:
You say I am being obtuse, yet here you try to argue that image, that still maintains same appearence, is same as having general impression.

Tell me, can you get a close enough copy of, say, Batman comic out of art generator that I can not tell which is the original and which is the generated copy?
You argued that a bunch of Jojo memes some of which still maintain same appearence as the original manga page are the same as "having general impression" just a few posts ago, lmao. What are you even doing?

Because with image formats, you can't tell which is original as long as there is no major loss of data.
That's a very important if, and is also sometimes the very point. See also: Sweet Bro And Hella Jeff, which artistically incorporates deliberately low quality highly compressed elements of itself.

And on another end of the spectrum, it is, infact, possible to overtrain a network into reproducing an image indistinguishable from the original.
 
You argued that a bunch of Jojo memes some of which still maintain same appearence as the original manga page are the same as "having general impression" just a few posts ago, lmao. What are you even doing?

Because they do? Or are you arguing they are literally exact same images?

When, exactly, do you start treating work as "original" instead of "copy"?

And on another end of the spectrum, it is, infact, possible to overtrain a network into reproducing an image indistinguishable from the original.

Yeah, it is possible. It is also generally considered failure state for these models because the model gets stuck on specific things. You more or less have to intentionally overtrain the model on specific image to get it to just copy original.

But by the same token, it is possible for person to paint exact copy of the original image. People can train themselves to copy exact style of another artist. This, as far as I can tell, is not treated as good enough reason to ban/restrict traditional painting skills.
 
It's pretty nakedly clear to me that you don't care about anything beyond that this threatens a tool you like. You don't care about the method in which it's being used or the consequences of using others' art to feed it.

I don't think you have a leg to stand on in a conversation regarding its regulation, because your only valid arguments seem to fixate on its similarities to an artist's output, not that the commonly-used generators utilize training data gathered without consent.

I mean you tried to promote the idea that 'they were publicly accessible therefore fair game' but like that's a garbage reason along the lines of how I'm not allowed to use someone's pictures from Facebook to advertise my business.
 
Last edited:
Yeah, it is possible. It is also generally considered failure state for these models because the model gets stuck on specific things. You more or less have to intentionally overtrain the model on specific image to get it to just copy original.

But by the same token, it is possible for person to paint exact copy of the original image.
…which is also generally considered failure state. There are even several words for it, and some of them even cover cases when multiple original images are used without permission to create a new image.

If your point is that AI generated art should be treated the same, then good news! This is also what my point is! A collage of a thousand pictures made by giving computer a text prompt should be treated the same as a collage of a thousand pictures made by using Cut, Paste and Transform in your image editor of choice! Which is to say, it should be treated as original when using your own original art as source, as a plagiarism when trying to take credit for other artists' pictures without permission, and on case-by-case basis in edge cases.

And if most common models weren't trained by using pictures that were scraped without permission, we wouldn't be having this argument to begin with.
 
If your point is that AI generated art should be treated the same, then good news! This is also what my point is! A collage of a thousand pictures made by giving computer a text prompt should be treated the same as a collage of a thousand pictures made by using Cut, Paste and Transform in your image editor of choice! Which is to say, it should be treated as original when using your own original art as source, as a plagiarism when trying to take credit for other artists' pictures without permission, and on case-by-case basis in edge cases.

And again you resort to falsehood of "AI art is just cut and copy". It is not. At no point does the model actually recreate the original. In fact these models are incapable of copy-paste which you claim them to do.

Please stop trying to push to falsehood. It has been, multiple times now, explained how this is not correct. There is no original image in the model from which it cuts and pastes.

The way these generators work is by starting from noise, then using the weighs they have to slowly (from their perspective) to "clean up" image by figuring out what pixels should be next to which ones. At no point does it copy/cut something from its training data and just paste wholesale onto the image and apply effect.
 
Last edited:
Alert: Stay on topic
stay on topic
Alright I don't think what qualifies as true art is really the business of SV staff. In the interests of that, Avernus and Mandemon have been threadbanned.


While I'm otherwise happy to let the discussion run, I'll address this point here:

That said I don't really see the big issue with the idea of putting up a rule that it should be tagged even without punishment, just honors system. If we think tagging is preferable, then I don't think there's really anything lost by an honors system rule, compared to no rule; bar the time and effort to set the category and checkboxes up, anywhosie. A toothless rule is better than no rule.

A toothless rule is worse than no rule. It teaches users that we don't take the rules seriously and that they can be disregarded. "Don't give an order that won't be followed" is a close sibling to this — rules that not only aren't followed, but cannot be enforced are worse than useless, and are instead actively detrimental.

If you can't make a rule enforceable, you shouldn't make it at all.
 
A toothless rule is worse than no rule. It teaches users that we don't take the rules seriously and that they can be disregarded. "Don't give an order that won't be followed" is a close sibling to this — rules that not only aren't followed, but cannot be enforced are worse than useless, and are instead actively detrimental.

If you can't make a rule enforceable, you shouldn't make it at all.
This has some interesting corollaries. On one hand, a rule against AI-generated content would be effectively unenforceable — both because it's not always obvious when something is AI-generated from the first glance, and because humans have a habit of imitating imperfections of products of machines for artistic purposes. On another hand, the very nature of training data that goes into creating an AI model makes some of the forum's rules retroactively unenforceable — specifically, the rule about only posting stuff when you are sure you have the rights to do that and the right to give SV rights to do that.

Because, while Mandemon and others might sneer at their strawman of "cut-and-copy", it's still fairly obvious that AI models don't just have impressions of their training data. I still remember how Github Copilot was caught reproducing GPL-licensed code verbatim, for example. It is clear that AI content is full of plagiarism, but what isn't clear is how much of that plagiarism is accidental and how much is because obfuscation of plagiarism was the point to begin with.
 
As far as I can tell, the last few pages have basically just amounted to two thoroughly entrenched sides unproductively yelling 'I'm right' at each other. Neither of you's going to convince the other, so just let it go.
 
Never thought it was a discussion theme before on SV. So I want to ask some questions about it.

What are good and bad consequences of banning AI-generated content in this situation?
How could it be implemented hypothetically and within what limits?
 
[ALERT="Stay on topic"]
Alright I don't think what qualifies as true art is really the business of SV staff. In the interests of that, Avernus and Mandemon have been threadbanned.[/ALERT]

While I'm otherwise happy to let the discussion run, I'll address this point here:



A toothless rule is worse than no rule. It teaches users that we don't take the rules seriously and that they can be disregarded. "Don't give an order that won't be followed" is a close sibling to this — rules that not only aren't followed, but cannot be enforced are worse than useless, and are instead actively detrimental.

If you can't make a rule enforceable, you shouldn't make it at all.

I'm pretty sure I remember SV trying to have a rule about tagging source for art, do you remember where that went? Because this seems adjacent, at least in the case of AI images specifically.
 
As far as I can tell, the last few pages have basically just amounted to two thoroughly entrenched sides unproductively yelling 'I'm right' at each other. Neither of you's going to convince the other, so just let it go.
On the one hand, I agree, shouting matches aren't really productive. On the other hand, it's disheartening how this always starts with hate from the anti-AI side. I really wanna go to sleep one day and wake up when the hate train has exhausted its fuel, artists can go back to art corners, and AI enthusiasts can go back to AI appreciation corners without being harrassed.
 
Status
Not open for further replies.
Back
Top