The End of Art: An Argument Against Image AIs

[Danioover9000]

@Leo Gura

1 minute ago, Leo Gura said:

Oops! Yeah, maybe, although aren't you guilty of this kind of logic too?

You are sneaky.

   Well, actually... I am guilty! ? I posted Steven Zapata's video on @Space A.I. is ruining art thread, so @Scholar must have got inspired to make a more concrete thread to build a more convincing case against A.I. use. So, I copied that video for that thread, then @Scholar copied that video to make his thread, and Steven Zapata must have copied those images he's drawing from his memories and imagination. Yeah, the issue of copying and copyright is a deep and convoluted one.

[something_else]

3 minutes ago, Danioover9000 said:

Because human artists are in the business of communicating divine and spiritual truths, through their drawing. So, of course it is threatening when a robot is trying to take that purpose and take the process away from artists. Arguable, the most enjoyable of artwork is the art working process and the result as well. It's just bad and lazy if robots and A.I. do all human work for us, especially creative processes.

If humans are contributing that much extra divine and spiritual truths through their art, then you have no reason to be scared of AI art.

The spiritual and divine truths could be expressed through AI assisted art just as much as they could through hand drawn.

Most of those artists are not doing it for the money anyway, they're doing it because they enjoy it. And they'll do it whether AI art exists or not. Maybe they'll even use AI to assist them.

[Danioover9000]

@Scholar

15 hours ago, Scholar said:

Did you watch the video? What AI does and what humans do when creating and getting inspired by art is fundamentally not the same.

The AI's basically take an image and relate certain keywords to the shapes found within that image. When you generate a new image, it will create shapes based on on a seed and other parameters. It will generate a series of images, and at each point of the image it will ask the network of relationships, in relationship to the prompt, what kind of shape is most likely to exist next to the shapes currently existing on the image.

It does not understand what it is drawing, it does not understand composition, anatomy, object relations and so forth. It has no interprative lense, it is not inspired, it is not studying art. It is literally taking the shapes within an image, extracting them into a network of relationships, and using that exact set of information in the process of image generation. Again, all that is happening is that it is looking at the shapes it has created (most likely given the networks of relationships related to certain keywords) and then iterating and saying "What is the most likely shape that I have extracted from the sets of images that would be placed next to all the shapes currently existing on this canvas?".

That's fundamentally different from how humans create or learn from other artists. And in the music industry this is very clear. Because of the lack of complexity, the AI's would commit copy right infringements if they were to train on copy righted music. That's simply because music is less complex, and rhythmical relationship are far easier to note and analyze. Meaning that, with music AI's, it would become very apparent that all the AI does is say "If this were music made by artist X, what is the most likely rhythm that music would have? What is the most likely series of notes that would exist next to this set of notes?". Because individual parts of copy righted music can be copyrighted, unlike with art, you would very quickly infringe on the rights of those artists. Because all the AI does is use those very information it has found in the initial training set, and use that to generate new information that will be the most "most likely result" based on the prompts given.

 

For this process it is not inspired by the initial training set, it is literally using the network of relationships of shapes found within the images it was trained on. So no, it's not what is happening with AI art whatsoever. And making it seem like that is just flawed thinking.

 

Take the example of recently deceased Kim Jung Gi. Here is his artwork:

 

 

 

Here is an image generated by an AI trained on his art:

 

If you look closer at the image, you will notice it completely lacks coherence. That's because it is not actually creating a scene, with people in in a busy market place. It is creating shapes based on training data that are most likely to exist next to other shapes. It doesn't know what a head is, what a line is, what a person is or what a market is. All it does is put down shapes based on a prompt, then put down more shapes that it feels most likely would exist at that point given the information that is related to the initial prompt.

 

 

Fundamentally, the AI cannot create novel information. If you give the AI images of all objects in the universe, other than art, it will never be capable of ending up at where Kim Jung Gi is here. It will never be able to draw a stick figure. Why? Because it does not know what a stick figure is, it does not know what a human is, it does not know what any of the objects are. All it knows is visual relationships correlated to certain keywords, and even that it doesn'T truly know, like a calculator doesn't know what math or numbers are.

This is 40.000 year old art. The AI could never create this if it did not have these shapes in it's training data. It it had photos of all animals, all objects, all it could create is photorealistic images. It would never venture into stylization, because those shapes do not exist in it's training set.

 

There can be no abstraction without understanding. What this means is that the AI cannot ever generate new art that is not a combination of previously existing art. This is not true for humans, because humans do abstract, and humans do have understanding, and because the process of art creation is fundamentally different in humans than in AI's. While a calculator might imitate and be very proficient at math, it is not doing what the human is doing when he is calculating something.

 

If AI was to reign, and somehow develope enough coherence to actually create genuinely coherent art, then it would halt the evolution of art, for the reasons I have mentioned above. If AI art existed before art styles were invented, no art styles would have ever existed. Why would it? People would have never created the art, and put in thousands of hours to create abstractions and idealizations and so forth. What AI art will do, is flood the internet with art which only has human spirit within it because it is imitating the art of people who have human spirit in them.

If you look at the art above, you can see the human spirit not because someone has learned from another artist, but because it was a human who created it. A human who knew what he was creating, a human who could abstract, understand, and then engage in the creation of genuinely novel information.

This is not just because the AI's are not yet ready yet, but because they are not individuated consciousness. In other words, the reasons why the AI cannot create new information is because it's metaphysical limitations. But for that to be discovered scientifically, it might take hundreds if not thousands of years. For as long as that is not the case, we will stumble in the dark, and be imitating forces on nature and deluding ourselves we have surpassed the wisdom of hundreds of millions of years of evolution.

 

Another point would be very easy to make, even if I was wrong about everything above. The difference between an artist getting inspired and creating art is that the artist does not nullify the value of the artist he is studying. I can study Iain Mccaig for my entire life. One, I will never be truly able to create art like Iain Mccaig, I will always have individual biases within my art that I will never get rid of, because I am human. I will never be able to replace the effort Iain Mccaig puts into his art, meaning even if I studied decades to copy Iain Mccaigs style perfectly, I would still have to put in the effort to create the art, therefore not nullifying his life work and his value as an artist. If I was a God, and I could look at all artists and copy their styles effortlessly, then create trillions of images that nobody can compete with, rendering them all homeless, that would be unethical. The only reason why we allow other artists to get "inspired", by each other is because of human limitations, and because the way you study art is fundamentally different from the way AI's do it. You would, of course, know this if you were an artist.

Ever artist that studies the greats of the past and present, cannot help but add his own voice to that art. He is not just taking these images and saying "Oh look, I have these shapes now, what is the most likely shapes next to the shape if this was a drawing by Iain Mccaig?". The fact that people are framing these things as comparable is mindblowing to me, and it shows me that schools have failed in educated people about art, but that is not very surprising to me considering the level of art education I experience myself in school.

An artist will understand the process of other artist, sure they will get inspired, maybe they will try to understand the way certain shapes are expressed, but humans are humans. We cannot help but inject our own voice into what we create. Nobody can become a second Iain Mccaig, because nobody is Iain Mccaig. To be Iain Mccaig, you have to have the genetics and the life path of Iain Mccaig. That's what makes his art what it is. If he had different dispositions, his style would look differently. And with his dispositions he added new, novel information to the world of art. He is not just a mixture of previous art styles. His art contains an imprint of his own unique consciousness, his unique sense of appeal, his unique way of understanding the world, his unique way of learning how to draw and how to try to create the illusion of life, probably even the unique anatomy of his body. All of these are factors, limitations, which create new art that humans can relate to.

Without people like him, these AI's could do nothing but photorealism, because that's all that would exist. Photos of objects.

   First off, really good post.

   So, technically, is it more a danger of misusing A.I. then?

[Scholar]

5 hours ago, Danioover9000 said:

@Scholar

   First off, really good post.

   So, technically, is it more a danger of misusing A.I. then?

Thank you, seems like you are the only one who took the time to read it.

 

Yes, the problem is misusage and complete lack of ethics because of how excited everyone is for this new technology. It's so funny, but this really shows you that humanity has not become more wise with technology whatsoever. This kind of attitude will be our doom sooner or later, if we don't wise up.

[something_else]

54 minutes ago, Scholar said:

Thank you, seems like you are the only one who took the time to read it.

I'm assuming this is directed at me. I read it, but you grossly oversimplified how AI art models work and that was what I felt like I wanted to respond to the most.

Your main point was that AI art can't generate anything new, it can only compose existing styles. But like 99%+ of human art ever created is just composing various existing ideas/styles/objects as well. And most of that doesn't require the artist to cite every experience they had in their life that lead them to produce a certain piece of art. That was my initial point.

'Adding your own twist' to a piece of art is really just you combining multiple experiences you have had in your life into a piece of artwork which is exactly what an AI art model is doing too. Your mind is just exponentially better at this right now so it feels like something different to you.

 

[Danioover9000]

@something_else

1 hour ago, something_else said:

I'm assuming this is directed at me. I read it, but you grossly oversimplified how AI art models work and that was what I felt like I wanted to respond to the most.

Your main point was that AI art can't generate anything new, it can only compose existing styles. But like 99%+ of human art ever created is just composing various existing ideas/styles/objects as well. And most of that doesn't require the artist to cite every experience they had in their life that lead them to produce a certain piece of art. That was my initial point.

'Adding your own twist' to a piece of art is really just you combining multiple experiences you have had in your life into a piece of artwork which is exactly what an AI art model is doing too. Your mind is just exponentially better at this right now so it feels like something different to you.

 

   That post was directed at me than you, but to quickly address your point about his oversimplification of A.I, his first, I think, opening post in the first page, @Scholar did say he doesn't have all the technical inner workings of artificial intelligence beforehand, so really assuming that he should have all the inner workings required to make a critique against use of the A.I. drawing programs that online businesses provide, is too reductionistic. At best, it's a safer assumption that the majority of users here, are not computer scientists or IT job related workers unless specified beforehand. That doesn't mean he's automatically not allowed to have a critique on A.I. And also, more importantly than the technical details, is the holistic understanding of how the use of A.I. technologies are impacting others in other contexts besides the one we're specifically discussing.

   Some cases, those A.I. companies also withhold some details of their A.I. programs and how they acquire and learn from their sources, so we also can't have all the technical inner workings at all, and we also can't trace every point of source as well.

    

[Danioover9000]

@something_else  A simplified example of making claims with little technical knowledge of the inner workings needed, is in a case of a murder or sexual assault, I don't need to enter into the rapists or murderer's mind completely, and completely know all his/her memories and life experiences. I just need to gather enough information from the victims, the perpetrators, the crime scenes and the situation to build my case and make educated and as accurate as possible assumptions and go forward with that, because if I don't I get stuck in only gathering information forever and the perpetrators would still be out there instead of gathering just enough and go forward with making decisions and taking action. Most of life it's rare to gain 100% knowledge and be all knowing at the get go.

[something_else]

@Danioover9000 In this case the technical understanding is important to a degree. His critique was largely based around the idea that the AI is just copying shapes from source images which is pretty much just wrong and very much an oversimplification.

[Scholar]

 

[Scholar]

On 24.10.2022 at 11:26 PM, something_else said:

@Danioover9000 In this case the technical understanding is important to a degree. His critique was largely based around the idea that the AI is just copying shapes from source images which is pretty much just wrong and very much an oversimplification.

 

That's not what I said, but here is a nice simplification that should help you understand what's going on:

 

Read the entire thread.

[Leo Gura]

Once the AI models become robust enough, they will be indistinguishable from human learning. And thus any claims that AI are stealing or copying will equally apply to all artists, which of course all artists are in denial about.

You are not the originator of all your best ideas. God is. And God works by interconnecting with its own self, cross-pollinating itself in an incestuous manner. But the human ego wants to take credit and possession of this creative process.

As an artist, if you introspect carefully, you will be able to notice the source material from which your "original" work is "copied" through cross-pollination as you do your art. Which is exactly what the AIs are doing. It's just harder to track you because it doesn't serve you to track where your art ideas come from.

Edited November 15, 2022 by Leo Gura

[something_else]

3 hours ago, Scholar said:

That's not what I said

I mean, that sounds like pretty much what you said here:

Quote

It is literally taking the shapes within an image, extracting them into a network of relationships, and using that exact set of information in the process of image generation. Again, all that is happening is that it is looking at the shapes it has created (most likely given the networks of relationships related to certain keywords) and then iterating and saying "What is the most likely shape that I have extracted from the sets of images that would be placed next to all the shapes currently existing on this canvas?".

Quote

but here is a nice simplification that should help you understand what's going on:

His argument is based on latent spaces, so meaningful mappings from high dimension to low dimension. Which is a reasonable argument. But your brain also has a latent space. It's just absurdly, unimaginably gigantically complex, and always changing, so we call it something different. AI art models already have absurdly large latent spaces, even if they're not anywhere near the complexity of something like our brains yet.

His arguments kind of make sense for smaller latent spaces in like one, two or maybe even a few hundred dimensions. But once you have a latent space with billions or even trillions of dimensions, his argument loses meaning because the model is so flexible and multi-dimensional. And the complexity of AI art models is only going to get greater and greater. I know he claims that 5 billion parameters is not enough to be sufficiently flexible but if you look at a lot of AI art you can see that he is kind of just wrong on that point. Much of it is extremely unique and not just pulling from a few data points, unless you prompt it with extremely specific things.

When you have something as complex as an AI art model, it is pulling from so many different sources when generating an image that it can create stuff that is for all intents and purposes new. it just doesn't have the capability to create something quite as new as a human because a human is exponentially more multi-dimensional. But the fundamental process both the AI and the human go through is getting very similar as time goes on.

And my original argument was simply that humans and AI drawing on past art/experience to generate new things is quite a comparable process, yet we don't impose especially strict copyright requirements on human artist's sources of experience and inspiration when they produce new art.

He also makes the point that AI art models don't change once they're trained and suggests that this is important distinction between AI and humans, which is true. But there are many AI models that continue to learn once trained, for example reinforcement learning based models. Or even just periodically training fixed models on new data. Although I'm not aware of reinforcement learning really being used for AI art yet, it likely will at some point.

He makes good points about models overfitting and essentially storing near perfect representations of training data. That's absolutely an issue that developers of AI art models will need to be very careful with.

But even a well trained model that isn't overfitted will be capable of spitting out nearly identical images from training data with the correct input, in quite a similar way to how a skilled artist could probably approximately paint many famous paintings from memory if asked.

His second argument is kind of silly. It's very reminds me of a factory worker complaining about how automation is taking over his job. Well yea, technological innovation and automation will remove the need for certain jobs. But it will also open up many more opportunities at higher levels of creativity and abstraction too if you take advantage of it instead of sticking your heels in the mud.

[Adam M]

13 minutes ago, Leo Gura said:

Once the AI models become robust enough, they will be indistinguishable from human learning. And thus any claims that AI are stealing or copying will equally apply to all artists, which of course all artists are in denial about.

You are not the originator of all your best ideas. God is. And God works by interconnecting with its own self, cross-pollinating itself in an incestuous manner. But the human ego wants to take credit and possession of this creative process.

As an artist, if you introspect carefully, you will be able to notice the source material from which your "original" work is "copied" through cross-pollination as you do your art. Which is exactly what the AIs are doing. It's just harder to track you because it doesn't serve you to track where your art ideas come from.

I love this. A big confirmation for me. Well said

[AtheisticNonduality]

31 minutes ago, Leo Gura said:

Once the AI models become robust enough, they will be indistinguishable from human learning. And thus any claims that AI are stealing or copying will equally apply to all artists, which of course all artists are in denial about.

You are not the originator of all your best ideas. God is. And God works by interconnecting with its own self, cross-pollinating itself in an incestuous manner. But the human ego wants to take credit and possession of this creative process.

As an artist, if you introspect carefully, you will be able to notice the source material from which your "original" work is "copied" through cross-pollination as you do your art. Which is exactly what the AIs are doing. It's just harder to track you because it doesn't serve you to track where your art ideas come from.

Humans have been stealing for hundreds of thousands of years. These machines were built very recently, and they haven't stolen anything. They've only been fed / gifted all they have.

[Leo Gura]

1 hour ago, KH2 said:

That's sounds so demoralizing

It only sounds demoralizing to those who have no vision.

You could be using these tools to make much better and more art. Which great artists are already starting to do.

It's silly to get demoralized over evolution.

Edited November 15, 2022 by Leo Gura

[Tyler Robinson]

You could have pretty much made the same arguments against early photography in 1900s that you make against AI. 

Artists who made portraits would have been out of business with the advent of photography yet they didn't. 

 

[Extreme Z7]

2 hours ago, Leo Gura said:

Once the AI models become robust enough, they will be indistinguishable from human learning. And thus any claims that AI are stealing or copying will equally apply to all artists, which of course all artists are in denial about.

Not me. I've had this same idea for a couple years now. Although, I was mostly thinking in relation as to whether AI will ever become smarter than humans and enslave humanity. I always thought that when AI reaches a level of intelligence equivalent to that of a human that it's basically similar to just another human. I don't think there's anything unique to AI that allows them to easily become smarter than all humans.

(Or I don't know, maybe someone can create a Psychedelics.exe someday )

So just fucking compete I say. Adapt. Be Creative.

I've already used Dall-E Mini in the past to create references for art I had trouble finding with Google Images.
I've envisioned using AI Art in the future to hasten the relatively-boring process of making a background so I can just focus on the character art.
There's so much more possibility for its use in art and also beyond it.

[Juan]

Art = Consciousness = Infinite ??‍♂️

Great example of good use:

 

[John Paul]

The end of art *as you have known and defined it as god*

[Danioover9000]

   There seems now to be an increase in copyright strikes from those tech companies to artists that use the A.I. programs, and some vice versa. HERE WE GO!