New AI poses an imminent existential threat to artists, especially if it has direct access to their art.
“Within 2 years, AI will be superior in terms of visual imagination to virtually all of humankind.”paraphrase of a statement by a head of a leading AI art tool. I am not allowed to quote directly.
AI is a fantastic and exciting artistic tool, which I have recently obsessively experimented with. It is, however, so incredible that it has the potential to render artists redundant, especially if employed by bad actors.
Let’s start with a demonstration of the power of AI. There are 16 images in the gallery below. 8 are by the renowned digital painter, Simon Stalenhag. The other 8 I created in seconds each, merely by typing text that ended with the words “in the style of Simon Stalenhag” into a field. [I won’t say which tool I’m using in order to protect confidentiality, and to keep this as a general philosophical and practical argument.] Can you tell which below are the AI knock-offs?
I’ll give the answer to this little quiz at the bottom of the post.
In 1997 IBM’s supercomputer Deep Blue defeated Chess champion Garry Kasparov. When I found out about that I lost interest in playing chess. What was the point if artificial intelligence had already triumphed over humans at the game? According to Kasparov himself, “Today you can buy a chess engine for your laptop that will beat Deep Blue quite easily”. That’s how fast the technology is advancing.
More recently, in 2015, Google DeepMind’s AlphaGo program crushed Lee Sedol at the game of Go. Go was considered more difficult for AI because the game has many more possible moves, and is thought to require intuition. However, while Deep Blue relied on brute computational force, Alpha Go also incorporated neural networks and reinforcement learning. Three years later Lee retired from Go because AI represented “an entity that cannot be defeated”.
You can watch the brutal defeat of humankind by a machine in this excellent, if depressing documentary:
Sedol was once the 2nd ranked Go player in the world, but now he finds pursuing the game futile because victory is no longer possible for our species. One giant step for AI, one stomp on the face for humankind.
Are Artists Next to Go?
And am I going to be another Lee Sedol?
After I learned of Kasparov’s defeat by IBM’s supercomputer, I reasoned that art was still safe. The supercomputer didn’t even know it was playing Chess, or that it existed. It wasn’t conscious, had no sense of mortality or vulnerability, and hence didn’t care about anything. Any attempt by Ai to make art, I deduced, would be lifeless, sterile, and flat. I wasn’t worried.
Then, in 2015 Google came out with Deep Dream, which could generate unique imagery. I wrote about it here:
Google Dream was initially developed to recognize images, but as I wrote then, “the Google team discovered that the process could be reversed so the neural networks would generate images of animals, trees, and so on, by asking them to enhance the features they were looking for”.
If you are not very familiar with what Deep Dream does, here’s a gallery from my article of 7 years ago.
I was somewhat prescient in my prognosis. I delivering my usual argument why I didn’t, or hadn’t thought AI was a challenge to art, ending with, “How could it make art or write a novel without any passion or compassion?” On the other hand, I quickly realized, “but the danger is that a computer could fake it.” And what I said next has essentially come true.
“What if instead of being trained to recognize dogs, or buildings, a neural network were fed super hi-rez images of the mature style paintings of Monet, Van Gogh, or Picasso? Let’s just use Monet landscapes. Let’s say the algorithm is refined to identify Monet paintings, and even fine tune it by teaching it to differentiate them from landscapes by Pissarro, Renoir (he did some), and other of his contemporaries. It might start making some good facsimiles. To kick it up further it needs to be fed up-close details, and know that’s what they are, so that it can search for and thus generate characteristics of Monet’s brush strokes. We could then look at where the computer falls short, and tweak out the algorithm. It could be combined with neural networks that are developed to recognize all the kinds of subject matter in Monet paintings. After the computer has analyzed hundreds of Monet paintings, and thousands of haystacks and sunsets, it might learn to compete with the artist at his own game. It might be possible to go on to mix and match artists to get the neural network to develop a new, hybrid style.” ~ Me, 2015.
Let me go type in some prompts with my AI tool of choice, and see what we get right now.
“Haystacks in the style of Monet.”
Not too shabby. Note that I’m putting minimal effort into this. I could hone my texts with practice and add a lot of parameters. I’ve written ones that are paragraph lengths.
“Rouen Cathedral by Monet.“
I’m not as persuaded. Needs more fiddling. The AI is best as water lilies.
“Water Lilies in a pond by Monet“
Well, I do declare! We’ve come a long way from Google’s Deep Dream. There’s an option to upscale at a higher quality, but I just did this for a quick and dirty demo.
I also wrote favorably about the technology in my article:
“I’d want to feed the program images of aliens, then have it go looking for them and inventing them. The possibilities could be fascinating. I’d actually love to work with the people that developed the algorithm to explore the more creative possibilities for realizing unprecedented imagery. The artificial neural networks will think of possibilities that we humans would not, and perhaps could not.” ~ Me, 2015
Obviously nobody consulted me. I mean, c’mon, why would they? Possibly to get better aliens, but I digress.
“What do aliens look like?”
They look like they got mangled in the teleporter, is what they look like. With work and dedication, you CAN get much better aliens out of the AI. Let me just add a few artists and a style:
“alien face, by Jim Burns, H.R. Giger, Alex Grey, in a detailed matte painting.“
Better, but I think the Giger influence is overboard. A few look more like him than like they are painted by him. There’s a way to “weigh” how much of each artist one uses, other ways are evolving to get more specific results, and you can also upload an image to give the AI inspiration. With enough experimentation, I could come up with something good. So, they didn’t really need to consult me after all.
Some Scary Results
Here are a few pieces I created using AI that will give digital artists a bit more to think about.
You can type in painting styles or photographic styles, and the AI is excellent at giving the impression of surface detail, as long as you don’t look too closely. I can also safely say that I am not personally capable at this moment of creating this realized of digital paintings in any of these three styles.
Let’s take a closer look at one of them:
It has the immediate surface appearance of a high-end digital painting by a supremely talented and competent digital artist. It can’t replace concept artists yet, because the eyes are all whack, and not in a way that works with the sci-fi theme. For us humans the eyes are eminently important, but for the AI they are just another detail to approximate. Virtually everything is as off as the eyes, but just less conspicuously so. The wall of the tunnel is higher behind the man’s head than it is behind the woman’s, as if the wall has split in two. The woman has no right shoulder, and the man’s arm that is implicitly wrapped around her is invisible. This is the tell-tale sign of AI art, whatever style it creates, and it can create any style to a large degree. Once you catch on to this flaw, where everything is a bit deformed, it’s very easy to differentiate AI art from digital paintings made by humans.
How long do we have until AI learns to recognize and articulate subject matter better, or even flawlessly? What am I going to be writing in another 7 years, if I haven’t become homeless because AI has put artists out of business?
Why is AI a Threat to Artists?
Look at the image below, and imagine if it didn’t have those characteristic flaws of AI art. What if it was perfect, and could be generated in a moment? Then artists would not have only a formidable adversary, but a potentially invincible one.
It could run simulations using several styles at the same time. Would art directors need to hire artists, and would heads of companies need to hire art directors?
Imagine for a moment you are a digital artist. How can you hope to compete with AI? My own mind scrambles searching for some avenue of hope.
I do have one, because not only is the AI not very good at details, to some degree it is only as good as the people using it. I could, if I wanted, combine it with my own skills to make something novel. An experienced digital fine artist can add something to the brew, that AI on its own can’t, and that your average user can’t. So, there’s hope via adopting it, which is something I’m presently working on. But even without doing so, if one’s digital art is distinct, it will stand out from the products of AI. Today, I can tell AI art from most human digital art virtually instantly.
And that brings us back to my initial quiz with the AI versus Simon Stalenhag images. Here’s the answer key.
The Nail in the Coffin of Artists
We have our hope that we can stay ahead of AI. And part of that is because we are conscious beings, who care about our art, or even the fact that we exist at all. The AI is a combiner and blender of images, but does not innovate on its own.
If fact, without access to a library of imagery to work with, the AI can’t make interesting or compelling art. If it doesn’t have any Monet, or Impressionist paintings in it’s library, it can’t make knock-offs, or combine them with something else. it is dependent on actual artists for material from which to generate new-ish imagery.
I’ll get back to that key point in a minute, and it’s a whopper. First, lets envision a digital artist is able to compete with AI by throwing in humanity, traditional skills, 3D modeling, photography, photo-bashing, intuition and inspiration, joys and sorrows, and so on. We can use a series of portraits I just completed. I haven’t seen the AI do anything specifically like them.
The techniques I used here include some I developed on my own. AI could only approximate the surface impression, but not with the specificity. Other of my creations included aliens and other creatures in a highly articulated way that AI can’t do:
If you have been following me, you know that I’m very versatile. I can find a way to ride AI and add to it, or to skirt around it.
Well, guess what? Anyone can potentially just upload an artist’s unique creation into AI, have it make variations, rebrand them, sell them, and cut the original artist out of the picture. I could think of no defense, no ray of hope against THAT.
And then I did. What if the people using the AI weren’t permitted to upload the work of living artists that is not a part of the public domain? Surely there’s plenty of work in the public domain for them to use, and they can use any of their own material, or that of anyone who gives them permission. There’s a win-win situation. So I brought this up in a forum, and when the other users [who relish uploading anyone’s art hot off the press] weren’t mocking and ridiculing me, I had the opportunity to chat with a couple of the higher ups that represent company itself.
We had some long back and forth, mostly, in my considered opinion, because they refused to acknowledge certain truths, or answer certain questions. Here’s one I asked:
‘What would the effect be on artists if anyone could upload their latest work, create derivations of it, and then market them?”
The answer, of course, is that the artist’s reputation and ability to make a living would be severely damaged. They refuse to acknowledge this.
At present, in the case of this specific company, in their defense, the AI is not designed to make slight derivations of a work which is uploaded to the prompt. Rather, it combines it in a more random way that isn’t too close to the original.
One of the people who was busy mocking me tried to provoke me by directly uploading one of my pieces, and making a knock off. Here is his result.
That would not give me any cause for concern, because it looks nothing like my tribute to Van Gogh (which is, incidentally, not a re-painting of one of his images, but my own creation of a non-existent self portrait with a bleeding ear]. Note the brazen attitude and FU to me. Someone letting me know in real time that they could steal my art, while the mods remained silent, as this happened within a discussion thread in the company’s forum.
If one were to zoom into my painting, it looks like this:
The imitation will need to look at a lot more of my art in detail before it can fake it.
Not surprisingly, the users hounding me on the forum were not themselves accomplished digital artists in terms of having real digital painting skills. They have nothing to lose in a game of stealing the material of others. They are also short-sighted enough to not realize that the same rules apply to them, and if they were ever to make anything worthwhile in visual art, someone else could instantly steal it.
Here is a cynical take. From what I can tell from what the company people said, there is a loophole in copyright law, so that while individuals can’t steal other people’s art, AI can. They can use this loophole to feed the AI more art, and thus make it a more attractive product, even if it goes completely against the principle of existing laws. In the end, no art is safe, including that of anyone using the AI. Further, because all art will be consumed and regurgitated in massive quantities by the AI, most people will be unable to make any money off of digital art because the medium will be so degraded and diluted. This will also affect analog artists, because their work, if it appears anywhere online in a photo, can also be appropriated and assimilated by the AI. On top of it, the market will be absolutely flooded with deliberate and unintentional forgeries. The only winner is the AI itself, and of course the few who made a living off of selling it, and exploiting the art of millions for personal gain [using the AI tool costs a monthly fee] while pretending to do it for the good of humanity, and a positive, rosy future. And while that seems overly cynical, I can’t myself, yet, come up with a convincing counter-argument. All I’ve heard from others was obfuscation, denial, insult, and a complete buffet of logical fallacies. That doesn’t mean this cynical take is correct. Sadly, I just haven’t seen a more convincing argument. Of course, I hope I am wrong, because if I am, then I am allowed to have a future as an artist. This is NOT something that I WANT to believe!
Somehow, in the forums, when I asked people why they couldn’t use the tool for creative ends without needing to upload my own art — I sure as hell don’t need to directly upload anyone else’s art to use the tool! — all I got was ganged up on, and abuse. Their argument is, whether they are cognizant of it or not, “We can steal from you, make forgeries, and you can’t do anything about it.” I even got the more toxic versions of, “I have already stolen from you”.
While the AI currently isn’t perfectly configured for theft, with some small tinkering, such as controlling the amount of randomization, which you are already able to do on a rudimentary level, it can become the ultimate tool for online art theft. And, while most people would want to use the product for much more innocent and worthwhile purposes, they could do all of that without stealing from living artists. I can’t figure out why theft has to be allowed.
They were quick to point out that one doesn’t even need to upload a picture of something from Simon Stalenhag to get a decent approximation of his work, as if this helped their case. It did only to the degree that it indicates not allowing uploads of people’s work won’t solve the problem. However, it raised another issue, which is that the AI surreptitiously does it without artist’s permission, and hence is able to recreate Stalinhag’s style.
I was able to get them to admit that given some minimum amount of an artist’s pieces, AI could make knock-offs, if it was trained to do that.
I will paraphrase one core idea that they said:
“Within 2 years, AI will be superior in terms of visual imagination to virtually all of humankind.”I am not permitted to give direct quotes or cite sources
I’m going to say no. I do not believe that the best piece of digital art will be made by a computer. We’ll see. But let’s give the AI the benefit of the doubt for the moment.
Take note that the same people who believe that AI will completely dominate humans in terms of visual imagination, also think there’s nothing wrong with feeding it living artist’s work against their will.
So, let’s put the pieces together. 1) AI’s visual imagination is projected to vastly outstrip the power of individual artists. 2) AI can potentially be the ideal tool for creating forgeries of anyone’s art. 3] The intellectual property of artists is not respected, and users are allowed to directly upload works by artists without their permission. That seems like the perfect recipe for stealing someone’s artistic identity, or, of course, that of many artist’s, given how easy it will be to do so. Consider I made those knock offs of Stalenhag’s art inside of a half hour.
The only thing missing from that list is anyone who would deign make derivations of an artist’s work. Would anyone do such a thing? Consider this notification from DeviantArt:
DeviantArt Protect now scans over 3.8 million new NFT images every week, and since the addition of NFT protection in August 2021, over 80,000 alerts regarding potential NFT infringement have been sent.
I’ve written here about how my art has been stolen, sold as prints and an an NFT: People stealing my art and trying to sell it, including an NFT! It is noteworthy that in the case of the NFT, the thief made one small adjustment. So, in my personal, anecdotal experience, someone tweaking my art and reselling it has already happened. My general take on reality is that a tool that makes doing that absolutely painless, may indeed be used for that purpose, especially in a culture where art theft is so rampant it is the rule rather than the exception. I already have seen a few artist who I am familiar with on Twitter announce that someone put up a shop of their work, pretending to be them. But, no, it wouldn’t be attractive to those same people to be able to crank out forgeries via AI.
I came to a bit of a realization when debating whether or not AI art companies have the right to allow users to upload artist’s content for manipulation. While AI is poised to extinguish artists because it is potentially superior at rendering in any style, and at lightning speeds, it can only replace artists if it has access to their creations. So, it is wholly dependent on the work of artists, at least to make any of the kind of art I’ve shared from it. Left to its own resources, it might make some visual imagery we haven’t encountered before, but might also find crushingly boring. It would largely be spinning its wheels.
In order for AI to kill off artists, we must give it our art to work with. Or rather, it must access our art with or without our permission, and it is doing just the latter. It combs the internet and takes whatever it can find. it is, in effect, a parasite in that regard. AI can’t replace my art without access to it, and as it happens, at least one AI art company allows direct uploads of my art, and despite my attempts to plead my case.
If I give them the benefit of the doubt, the AI is not currently configured to facilitate the deliberate plagiarism of art, even if the potential is there. So, there may be a window where people can play with images without treading on artist’s copyright. The technology is new, and one could hope issues will be addressed as they arise. Well, that’s with one company. I hear others are actively pursuing making copies of art. I got the impression I was being told that my fears were justified, but the real culprits were elsewhere. It’s more a matter of ethics than of technology. I wanted assurance that they cared about and respected artist’s ownership of their own work, and I did not get that at all. Quite the opposite feeling. I could be wrong. Yet, I can’t find any reason why they need to allow uploading of living artist’s art. Everything would work just as well without doing that. I’m repeating myself, but somehow people aren’t getting this very simple point. I don’t seem to be able to hammer it home.
Let me give you a couple examples to illustrate how the AI can’t recreate art without access to it.
The first image I made of a human fly in his lab using AI, and it’s very impressive. The second is something I made 9 years ago, the hard way, and it is not the same thing. You can’t get the AI to reproduce the image on the right, or anything close to it, via text prompts alone. The AI can’t be anywhere near that specific. It can do faux-intricacy, but it can’t do accuracy. I made hundreds of images with AI of people in cars. Not once did it give me a correct window, windshield, steering wheel, door, side mirror, etc. It didn’t even come close. It’s not going to render a microscope correctly in relation to a human fly, barfing on donuts, from a long tongue distended from between the teeth in its open mouth, while pouring itself a drink and holding a slice of pizza in a claw. AI can’t come remotely close to doing that from scratch, which is also why it can’t replace concept artists. But if someone feeds it the final image, it can tweak it.
Even if AI is superior, as it is, I at least have a contender to put in the ring against AI, including my own AI creations.
Below, people seem to appreciate this piece more when they see the drawing phase, which more clearly reveals how I went about creating it.
The AI doesn’t build it up from the ground like that, which is part of why AI is so inaccurate. It culls from existing images and recombines them. It doesn’t evolve an image with clear intent and knowledge about the subject. Notice that my alien is at a much more complex angle. Every mark is conscious and deliberate. My AI version has anomalous extraneous flotsam that has nothing to do with the picture.
In this sense, while I do have to fear the raw power of the AI, I am still allowed my own distinct voice which it cannot replicate. Crush me as it may, I have a fighting chance in hell. That is, unless someone uploads my finished piece, and the AI does the equivalent of applying super-powerful Photoshop filters to it.
I don’t think they are right that AI will have a better visual imagination than human artists, period. If that were the case, then it could also write a superior novel, because there is a subjective content to art that is based on knowledge, experience, and vision. What makes more sense is that the AI, being millions of times better and faster at calculations, will be vastly superior at rendering images. That is not the same thing as a visual imagination.
As it stands, at present, the real threat is not the AI itself, but the access it has been given to copy artists work without their permission, without which, the AI’s work would be innocuous. AI is not defeating artists. Not yet. It is parasitically stealing from them. Artists can compete with AI, but only if it is not allowed to just take our finished work, run it through filters, and plop it out its proverbial rear. No human art in, and nothing remotely human will come out.
Call me an idealist, naive, a romantic, what have you, but I think humans are better at art than AI, because we are alive, conscious, give a shit about anything, and are even spiritual beings of sorts. The idea that something purely mechanical is better than us at art is the final defeat of our species. Skynet will have won.
AI is a great tool for creative exploration. I’m a power user in that regard, and a strong advocate for it. However, that is overshadowed if AI’s potential for making derivations via direct and indirect routes is used intentionally and unintentionally for theft and plagiarism. There are easy solutions to this that respect artist’s intellectual property, and allow us humans some hope that we won’t be rendered inferior carbon based entities.
It’s going to be tough. Artists now have to fight AI, and plagiarism supercharged with AI. I may be more concerned about the theft than the AI. Meanwhile, used properly, and respective of artist’s intellectual property, I maintain that AI can be a fantastic tool for exploring visual possibilities, finding inspiration, and integrating into ones own art. Like most technology, it can be used for good or bad. Right now, my objection is that real artists are being thrown under the bus to feed Skynet. And the reason for doing that is? Count Baltar?
[I assume a lot of you got both sci-fi references. Baltar!? BA HA HA HA! If you don’t know who he is, he was a human traitor that sold out his species to the Cylons in Battlestar Galactica.]