Guides

The Complicated Line between Art and Artificial

By now, you've probably noticed the influx of "AI generated" art. If you've been following me for a while, you might know I use one of these generators to do "style transfers" for some of my own art. But we've grown from an era where DeepDream put creepy eyes on everything, to a point where typing a few words into a text box can create effective, expressive art. So what does that mean?

I won't lie, this topic is complicated, and it took a wall of text to break it down. It you're not here for that, this is the summary: Generators aren't theft, but they will replace a lot of jobs. That's bad, but it's bad for capitalistic reasons. Artistically, the world is about to get more creative, and our culture will grow, which is good. The part of this that hurts is mostly the economic context it is arriving into, and we'll do better focusing on changing that context than pretending we can stop these tools. On a personal level, I think you should use any tool that enables you to be more creatively expressive, and you should not listen to anyone who shames you for exploring an artistic interest that doesn't hurt anyone. If a generator lets you grow as an artist while wasting less paint and canvas, that's a win in a lot of ways.

If you're interested in a deeper dive into why I feel this way, keep reading.

I've been thinking about this stuff a lot, the art communities I'm in have been watching this tech arrive since about the time I started doing digital art. The conversation is moving into the mainstream, so I think it's worth providing a perspective that has had a little more time to bake.

First, let's make sure we're on the same page. While it's pretty popular to call this stuff "AI generated", that's a dramatic label. Deep Dream, Style Transfer, GPT and DALL-E are all neural network technology that benefits from Deep Learning, so people like to call it "AI". If we're being more technical, these are Generative Adversarial Networks.

The Generator part of the algorithm generates an image (often starting with random noise) and the Adversarial part judges how much the image looks like the goal with a rating. The generator tweaks some parameters and makes a new image, keeping tracks of which tweaks succeed in producing a better rating. The process is repeated hundreds or thousands of times to get to an image that often looks pretty coherent. The ratings are based on trends detected in millions of training photos with contextual labels.

The algorithm has seen a bunch of examples of things it knows are trees, so it can guess at whether some new pixels look like a tree. This isn't an "artificial intelligence" hanging out in cyberspace dreaming up paintings for us. It's just a lot of cleverly organized math. The difference isn't just brain food for sci-fi nerds, it had a pretty big impact on the idea of authorship.

If a generator trains on a database that includes, say, a hundred different works by Leonard Da Vinci, and then you ask it to draw something in a renaissance style, is it plagiarizing Da Vinci? Who really "created" the image, the person who typed the prompt, the programmer who wrote the code, the code itself? Or is it like sampling a track in a song, the parts that are sampled belong to the artist who created them (or whoever they've sold those rights to)? Even if we're OK with artists stealing from long dead masters, what about when we ask generators to emulate living artists?

But a generator isn't clipping pictures out of a magazine and pasting them together. If you watch the generator iterate, it resolves an image from a haze of random noise until it looks like something recognizable. Perceptually, it is closer to cloud watching than collage.

Recently, I've been trying something kind of new, making subjective illustrations of actual things, instead of just pretty abstract shapes. This has historically been a weak spot of mine, I don't visualize things in my head and my hands are not great a pen or brush work. But using some digital tools and working with vector shapes, I've been comfortable enough to give it a shot. When my partner requested a frog design, I couldn't simply imagine a frog and draw it. I knew things about a frog, it has round, bulbous eyes, it has a wide mouth, it has nose holes and a long tongue. But those are properties, not visual ideas. So I looked for some help.

I searched for cartoon frog faces, and looked at a couple dozen examples. I paid attention to how the eyes sat on the head, the way the curve of the lips defined the shape of the face, and the proportions of the nose holes. Different images had different answers, but looking at a bunch of them helped me understand the essential "frog" appearance they all shared. Then I took my blank page, and lined up shapes in ways that fit the patterns I had picked up on. Soon I had a rough outline of a frog face, and from there I was able to refine it into something charming, cute, and recognizably froggy.

Is that so different from what the generators do? When I did it in my head, I didn't have to worry about plagiarism and copyright. I'm free to look at public art, I'm free to analyze it's patterns, and I'm free to try to recreate them, as long as I'm not using them directly in the process or pretending to be someone I'm not. This is how most of us learn art, by observing the art around us and resynthesizing it with our own creations.

So the argument that these generators are stealing from artists doesn't really resonate to me. Especially when I consider the sources of the training data. While many models are constructed using public domain or academically licenses image archives, others are simply trained off anything they can find online. I am sympathetic to the argument that a creator should have control over how their work is consumed, and that many would prefer not to let a computer algorithm try to emulate them. But I'm wary when we're talking about art that is posted to the public internet by the creator.

We could have a whole long chat about the dangers of our modern ideas of privacy, and how profit motives often make it hard to tell the difference between exploitation and convenience. We'd probably agree on a lot of it. But that doesn't change the fact that when I upload my fractal art to Facebook so some of you will see it, I'm explicitly granting access to that art to a public audience. It sucks that surrendering that control is such a normalized part of being a modern artist, but it is still a thing I'm choosing to do.

And I don't mean that in a nit-picky "gotcha" way. When I'm sharing my art somewhere publicly for free, what is my goal? Maybe to attract interest that turns into sales or support, sure. But I'm also trying to make my mark. When you think of fractal art, I hope my art is some of what comes to mind. Beyond the money to keep making art, I genuinely want to make an artistic impression on you. When someone tells me my art has inspired them to try to make fractals like I do, I don't feel like they are stealing anything by wandering through my gallery. I feel honored!

To that end, I think one of the ways we will adapt to a future with generated art where one of the GOALS as an artist is to become notable enough to be included in a training database. To associate a style with your name so strongly that it biases any generator that invokes it.  To contribute to the collective pool of shared creativity we've been cultivating. Or to curate our own databases of our art and license our styles.

OK, well hopeful creative utopias are fun to dream about, but we can't just ignore the fact that a whole lot of artists are about to be efficiently replaced by automation. Let me be clear, I'm a pretty big fan of artists getting paid to make art, and I know firsthand how hard it is to secure any stability doing it. Watching your industry dry up is scary, and no amount of rationalizing it will keep it from hurting some good people along the way. For what it's worth, I've seen some pretty impressive results come out of prompts as simple as "fractal art", so I'm on the chopping block here too.

But which work am I losing? It's not the fans that commission me. Even with generated art widely accessible, I can build and maintain my audience by marketing my artistic persona. A patron, in the classic sense, is someone who buys my art because they want to support me as an artist. As long as I continue to present and interesting and charming artistic persona around new people, I will attract patrons.

What I won't be getting as much of is people using my art as a service. Sometimes I'm approached by an indie musician who needs an album cover. Self publishing on social media with no budget, they need a trippy cover for as cheap as possible. I don't disparage these customers, it's nice to make a few quick bucks when I have the time. But these aren't the customers I'm seeking. I won't make a successful career out of selling album covers for $25 a pop, even if I got popular doing it. I'm seeking the customers who can pay my full commission price, because that's the rate that I've calculated could make this art thing work for me.

Pretty soon, all of those musicians will have a free website where they can type "trippy fractal album art" in and get something they can use. This means less work for me. It also means the quality of the album covers will trend upwards. In the end, a world where musicians feel like their album art expresses them sounds better than one where they have to pay me for help.

Those customers are the least interested in supporting me. Maybe they are happy to support an indie artist, sure, and I appreciate their business. But once they have the art they need, I'm not a part of it anymore. If I'm lucky, they'll mention where the art came from, or maybe come back to buy more. But these aren't the customers that turn around and buy a large print because they like my art so much.

Sadly, a lot of art industries use the service model as an excuse to exploit artistic talent, and a lot of artists impose the same exploitations on themselves to compete. Drawing commissions of people's pets to help them express their love is rewarding and a way to grow as a pet portrait artist. Drawing 10 pet portraits a day so they can be cheap enough to sell at a sustainable profit is not so great for artistic development. It burns a lot of artists out.

Generated art is going to replace a lot of abusive art industry jobs. Not every job that will be lost is an abusive one, but most of the abusive ones will no longer have a need to abuse human artists. Yes, this means capitalists are reaping the benefits and the working class is left empty handed, I'm not denying that this progress casts long shadows. But it also means that what it means to be a working artist does is going to change.

Who still hires an artist when they have free generated art available? People who want a consistent or particular style, people who want to hire creative insight, patrons of particular artists, and anyone who wants to be sure the art in their project is genuinely expressive. There will still be artists who meet those needs. And in the future, artists who want to be professional will aim to do that, instead of pushing to make 10 pet portraits in a day.

It's fair to point out that there will be a new type of artist, the kind that specializes in coaxing quality out of a generator. After all, someone is still operating the generators to get the art, even if they taking place of dozens or hundreds of other artists. But I think it's pretty clear that we'll be losing more art jobs than we'll be making, so that's not all that much solace. I will say I've been very impressed by artists who have incorporated generator output into a more traditional digital art process. That approach is even how I made my own most popular design. Those tools allow me to expand my creative range, and I've grown artistically in the process of learning to use them.

That's the trade off that I don't think gets considered as much in these conversations. Yes, we will have fewer professional artists doing art as a service. We will also have a lot more artists. The market will be more saturated, but ask any modern creator, they'll tell you it already is supersaturated. You already have to do so much more to be a successful artist, you have to sell the artistic persona I talked about earlier. That approach still works in a market with generated art, so the path to real success isn't really changing with the generators around. And in the meantime, everyone with access to the tech can make art.

Of course, access to this kind of tech is a whole huge issue, with a lot of complicated factors. This article is already sprawling enough, but I need to at least mention that this tech relies on a lot of modern privilege, and the people we're failing to include in our digital world are getting left behind at an increasing rate. I may use linguistic shorthand in this article, but it's not fair to say we're all going to benefit from this tech, because some of us aren't privileged to access tech.

So, we get a lot more artists and a lot less of them get paid for it. That's bad news, right? Well, if you measure the success of an artistic practice through a capitalistic lens, yeah. Skilled labor is losing value, and people who practice it gain less for trying. But I can't ignore the fact that there's a lot more to artistic pursuits than making money. About the time I was ready to start selling my art in person, a global pandemic convinced us to stay inside for a year. That was a year where I made almost no money off my art. It was also a year where I grew a tremendous amount as an artist. Knowing that it would be a while before I could show off, I stopped worrying about what would sell, and started focusing on what was artistically compelling. By the time the markets came back, I had a much more developed artistic style.

Creating environments where artists have time to put into artistic pursuits is a matter of economics and policy. A lot of industries are looking at accelerating automation with the same fears of being replaced, not just artists. Writers, designers, programmers, managers, customer service, medical diagnostics, musicians, advertisers, journalists.. Generated content is impacting an increasingly large swath of labor. So at a fundamental level, we as people need to come up with a new plan that makes room for people to live even where there aren't enough jobs for all of us. This is not a fight for artists alone, this is a fight for humans.

But that fight isn't going to be quick. Transitioning to a automated society will take time and trouble. And during that time, whether we like it or not, this creative technology will continue to make art more accessible to more people. So another shift we will need to make is to encourage the value of art for non-commercial purposes. We can't all be popular artists, and maybe none of us should get all of the attention. It's probably better if we learn to create within smaller creative communities, where we can engage with each other, watch each other grow, and encourage each other with our explorations.

When we don't have to compete over who can sell out art, we can indulge in much more interesting, personal, and passionate art. We can progress our artistic culture so much more when we have more access to creative self-expression. If we can teach each other to value this cultural development as much as we've valued economic development up to this point, maybe we can stop producing leaders that prioritize profits and power. At the very least, we can stop sacrificing new artists at the alter of the popularity algorithms.

Generators are going to be awesome. They're going to be terrible. They're going to lift us up, and they are going to hurt us. They are tools, and we will develop better understandings of when and where to use them as we practice them more. Just like the internet, or the printed page, or the paintbrush. A favorite writer of mine offered some advice: "When you imagine a new tool, imagine it in your enemy's hand, with the sharpest part against your neck." Do that. Understand that a tool is also a weapon. But don't forget, you get one too, and maybe you can find something more interesting to do with yours than hurt people. When you do, we're all going to want to see it.

The Complicated Line between Art and Artificial

By now, you've probably noticed the influx of "AI generated" art. If you've been following me for a while, you might know I use one of these generators to do "style transfers" for some of my own art. But we've grown from an era where DeepDream put creepy eyes on everything, to a point where typing a few words into a text box can create effective, expressive art. So what does that mean?

I won't lie, this topic is complicated, and it took a wall of text to break it down. It you're not here for that, this is the summary: Generators aren't theft, but they will replace a lot of jobs. That's bad, but it's bad for capitalistic reasons. Artistically, the world is about to get more creative, and our culture will grow, which is good. The part of this that hurts is mostly the economic context it is arriving into, and we'll do better focusing on changing that context than pretending we can stop these tools. On a personal level, I think you should use any tool that enables you to be more creatively expressive, and you should not listen to anyone who shames you for exploring an artistic interest that doesn't hurt anyone. If a generator lets you grow as an artist while wasting less paint and canvas, that's a win in a lot of ways.

If you're interested in a deeper dive into why I feel this way, keep reading.

I've been thinking about this stuff a lot, the art communities I'm in have been watching this tech arrive since about the time I started doing digital art. The conversation is moving into the mainstream, so I think it's worth providing a perspective that has had a little more time to bake.

First, let's make sure we're on the same page. While it's pretty popular to call this stuff "AI generated", that's a dramatic label. Deep Dream, Style Transfer, GPT and DALL-E are all neural network technology that benefits from Deep Learning, so people like to call it "AI". If we're being more technical, these are Generative Adversarial Networks.

The Generator part of the algorithm generates an image (often starting with random noise) and the Adversarial part judges how much the image looks like the goal with a rating. The generator tweaks some parameters and makes a new image, keeping tracks of which tweaks succeed in producing a better rating. The process is repeated hundreds or thousands of times to get to an image that often looks pretty coherent. The ratings are based on trends detected in millions of training photos with contextual labels.

The algorithm has seen a bunch of examples of things it knows are trees, so it can guess at whether some new pixels look like a tree. This isn't an "artificial intelligence" hanging out in cyberspace dreaming up paintings for us. It's just a lot of cleverly organized math. The difference isn't just brain food for sci-fi nerds, it had a pretty big impact on the idea of authorship.

If a generator trains on a database that includes, say, a hundred different works by Leonard Da Vinci, and then you ask it to draw something in a renaissance style, is it plagiarizing Da Vinci? Who really "created" the image, the person who typed the prompt, the programmer who wrote the code, the code itself? Or is it like sampling a track in a song, the parts that are sampled belong to the artist who created them (or whoever they've sold those rights to)? Even if we're OK with artists stealing from long dead masters, what about when we ask generators to emulate living artists?

But a generator isn't clipping pictures out of a magazine and pasting them together. If you watch the generator iterate, it resolves an image from a haze of random noise until it looks like something recognizable. Perceptually, it is closer to cloud watching than collage.

Recently, I've been trying something kind of new, making subjective illustrations of actual things, instead of just pretty abstract shapes. This has historically been a weak spot of mine, I don't visualize things in my head and my hands are not great a pen or brush work. But using some digital tools and working with vector shapes, I've been comfortable enough to give it a shot. When my partner requested a frog design, I couldn't simply imagine a frog and draw it. I knew things about a frog, it has round, bulbous eyes, it has a wide mouth, it has nose holes and a long tongue. But those are properties, not visual ideas. So I looked for some help.

I searched for cartoon frog faces, and looked at a couple dozen examples. I paid attention to how the eyes sat on the head, the way the curve of the lips defined the shape of the face, and the proportions of the nose holes. Different images had different answers, but looking at a bunch of them helped me understand the essential "frog" appearance they all shared. Then I took my blank page, and lined up shapes in ways that fit the patterns I had picked up on. Soon I had a rough outline of a frog face, and from there I was able to refine it into something charming, cute, and recognizably froggy.

Is that so different from what the generators do? When I did it in my head, I didn't have to worry about plagiarism and copyright. I'm free to look at public art, I'm free to analyze it's patterns, and I'm free to try to recreate them, as long as I'm not using them directly in the process or pretending to be someone I'm not. This is how most of us learn art, by observing the art around us and resynthesizing it with our own creations.

So the argument that these generators are stealing from artists doesn't really resonate to me. Especially when I consider the sources of the training data. While many models are constructed using public domain or academically licenses image archives, others are simply trained off anything they can find online. I am sympathetic to the argument that a creator should have control over how their work is consumed, and that many would prefer not to let a computer algorithm try to emulate them. But I'm wary when we're talking about art that is posted to the public internet by the creator.

We could have a whole long chat about the dangers of our modern ideas of privacy, and how profit motives often make it hard to tell the difference between exploitation and convenience. We'd probably agree on a lot of it. But that doesn't change the fact that when I upload my fractal art to Facebook so some of you will see it, I'm explicitly granting access to that art to a public audience. It sucks that surrendering that control is such a normalized part of being a modern artist, but it is still a thing I'm choosing to do.

And I don't mean that in a nit-picky "gotcha" way. When I'm sharing my art somewhere publicly for free, what is my goal? Maybe to attract interest that turns into sales or support, sure. But I'm also trying to make my mark. When you think of fractal art, I hope my art is some of what comes to mind. Beyond the money to keep making art, I genuinely want to make an artistic impression on you. When someone tells me my art has inspired them to try to make fractals like I do, I don't feel like they are stealing anything by wandering through my gallery. I feel honored!

To that end, I think one of the ways we will adapt to a future with generated art where one of the GOALS as an artist is to become notable enough to be included in a training database. To associate a style with your name so strongly that it biases any generator that invokes it.  To contribute to the collective pool of shared creativity we've been cultivating. Or to curate our own databases of our art and license our styles.

OK, well hopeful creative utopias are fun to dream about, but we can't just ignore the fact that a whole lot of artists are about to be efficiently replaced by automation. Let me be clear, I'm a pretty big fan of artists getting paid to make art, and I know firsthand how hard it is to secure any stability doing it. Watching your industry dry up is scary, and no amount of rationalizing it will keep it from hurting some good people along the way. For what it's worth, I've seen some pretty impressive results come out of prompts as simple as "fractal art", so I'm on the chopping block here too.

But which work am I losing? It's not the fans that commission me. Even with generated art widely accessible, I can build and maintain my audience by marketing my artistic persona. A patron, in the classic sense, is someone who buys my art because they want to support me as an artist. As long as I continue to present and interesting and charming artistic persona around new people, I will attract patrons.

What I won't be getting as much of is people using my art as a service. Sometimes I'm approached by an indie musician who needs an album cover. Self publishing on social media with no budget, they need a trippy cover for as cheap as possible. I don't disparage these customers, it's nice to make a few quick bucks when I have the time. But these aren't the customers I'm seeking. I won't make a successful career out of selling album covers for $25 a pop, even if I got popular doing it. I'm seeking the customers who can pay my full commission price, because that's the rate that I've calculated could make this art thing work for me.

Pretty soon, all of those musicians will have a free website where they can type "trippy fractal album art" in and get something they can use. This means less work for me. It also means the quality of the album covers will trend upwards. In the end, a world where musicians feel like their album art expresses them sounds better than one where they have to pay me for help.

Those customers are the least interested in supporting me. Maybe they are happy to support an indie artist, sure, and I appreciate their business. But once they have the art they need, I'm not a part of it anymore. If I'm lucky, they'll mention where the art came from, or maybe come back to buy more. But these aren't the customers that turn around and buy a large print because they like my art so much.

Sadly, a lot of art industries use the service model as an excuse to exploit artistic talent, and a lot of artists impose the same exploitations on themselves to compete. Drawing commissions of people's pets to help them express their love is rewarding and a way to grow as a pet portrait artist. Drawing 10 pet portraits a day so they can be cheap enough to sell at a sustainable profit is not so great for artistic development. It burns a lot of artists out.

Generated art is going to replace a lot of abusive art industry jobs. Not every job that will be lost is an abusive one, but most of the abusive ones will no longer have a need to abuse human artists. Yes, this means capitalists are reaping the benefits and the working class is left empty handed, I'm not denying that this progress casts long shadows. But it also means that what it means to be a working artist does is going to change.

Who still hires an artist when they have free generated art available? People who want a consistent or particular style, people who want to hire creative insight, patrons of particular artists, and anyone who wants to be sure the art in their project is genuinely expressive. There will still be artists who meet those needs. And in the future, artists who want to be professional will aim to do that, instead of pushing to make 10 pet portraits in a day.

It's fair to point out that there will be a new type of artist, the kind that specializes in coaxing quality out of a generator. After all, someone is still operating the generators to get the art, even if they taking place of dozens or hundreds of other artists. But I think it's pretty clear that we'll be losing more art jobs than we'll be making, so that's not all that much solace. I will say I've been very impressed by artists who have incorporated generator output into a more traditional digital art process. That approach is even how I made my own most popular design. Those tools allow me to expand my creative range, and I've grown artistically in the process of learning to use them.

That's the trade off that I don't think gets considered as much in these conversations. Yes, we will have fewer professional artists doing art as a service. We will also have a lot more artists. The market will be more saturated, but ask any modern creator, they'll tell you it already is supersaturated. You already have to do so much more to be a successful artist, you have to sell the artistic persona I talked about earlier. That approach still works in a market with generated art, so the path to real success isn't really changing with the generators around. And in the meantime, everyone with access to the tech can make art.

Of course, access to this kind of tech is a whole huge issue, with a lot of complicated factors. This article is already sprawling enough, but I need to at least mention that this tech relies on a lot of modern privilege, and the people we're failing to include in our digital world are getting left behind at an increasing rate. I may use linguistic shorthand in this article, but it's not fair to say we're all going to benefit from this tech, because some of us aren't privileged to access tech.

So, we get a lot more artists and a lot less of them get paid for it. That's bad news, right? Well, if you measure the success of an artistic practice through a capitalistic lens, yeah. Skilled labor is losing value, and people who practice it gain less for trying. But I can't ignore the fact that there's a lot more to artistic pursuits than making money. About the time I was ready to start selling my art in person, a global pandemic convinced us to stay inside for a year. That was a year where I made almost no money off my art. It was also a year where I grew a tremendous amount as an artist. Knowing that it would be a while before I could show off, I stopped worrying about what would sell, and started focusing on what was artistically compelling. By the time the markets came back, I had a much more developed artistic style.

Creating environments where artists have time to put into artistic pursuits is a matter of economics and policy. A lot of industries are looking at accelerating automation with the same fears of being replaced, not just artists. Writers, designers, programmers, managers, customer service, medical diagnostics, musicians, advertisers, journalists.. Generated content is impacting an increasingly large swath of labor. So at a fundamental level, we as people need to come up with a new plan that makes room for people to live even where there aren't enough jobs for all of us. This is not a fight for artists alone, this is a fight for humans.

But that fight isn't going to be quick. Transitioning to a automated society will take time and trouble. And during that time, whether we like it or not, this creative technology will continue to make art more accessible to more people. So another shift we will need to make is to encourage the value of art for non-commercial purposes. We can't all be popular artists, and maybe none of us should get all of the attention. It's probably better if we learn to create within smaller creative communities, where we can engage with each other, watch each other grow, and encourage each other with our explorations.

When we don't have to compete over who can sell out art, we can indulge in much more interesting, personal, and passionate art. We can progress our artistic culture so much more when we have more access to creative self-expression. If we can teach each other to value this cultural development as much as we've valued economic development up to this point, maybe we can stop producing leaders that prioritize profits and power. At the very least, we can stop sacrificing new artists at the alter of the popularity algorithms.

Generators are going to be awesome. They're going to be terrible. They're going to lift us up, and they are going to hurt us. They are tools, and we will develop better understandings of when and where to use them as we practice them more. Just like the internet, or the printed page, or the paintbrush. A favorite writer of mine offered some advice: "When you imagine a new tool, imagine it in your enemy's hand, with the sharpest part against your neck." Do that. Understand that a tool is also a weapon. But don't forget, you get one too, and maybe you can find something more interesting to do with yours than hurt people. When you do, we're all going to want to see it.

By now, you've probably noticed the influx of "AI generated" art. If you've been following me for a while, you might know I use one of these generators to do "style transfers" for some of my own art. But we've grown from an era where DeepDream put creepy eyes on everything, to a point where typing a few words into a text box can create effective, expressive art. So what does that mean?

I won't lie, this topic is complicated, and it took a wall of text to break it down. It you're not here for that, this is the summary: Generators aren't theft, but they will replace a lot of jobs. That's bad, but it's bad for capitalistic reasons. Artistically, the world is about to get more creative, and our culture will grow, which is good. The part of this that hurts is mostly the economic context it is arriving into, and we'll do better focusing on changing that context than pretending we can stop these tools. On a personal level, I think you should use any tool that enables you to be more creatively expressive, and you should not listen to anyone who shames you for exploring an artistic interest that doesn't hurt anyone. If a generator lets you grow as an artist while wasting less paint and canvas, that's a win in a lot of ways.

If you're interested in a deeper dive into why I feel this way, keep reading.

I've been thinking about this stuff a lot, the art communities I'm in have been watching this tech arrive since about the time I started doing digital art. The conversation is moving into the mainstream, so I think it's worth providing a perspective that has had a little more time to bake.

First, let's make sure we're on the same page. While it's pretty popular to call this stuff "AI generated", that's a dramatic label. Deep Dream, Style Transfer, GPT and DALL-E are all neural network technology that benefits from Deep Learning, so people like to call it "AI". If we're being more technical, these are Generative Adversarial Networks.

The Generator part of the algorithm generates an image (often starting with random noise) and the Adversarial part judges how much the image looks like the goal with a rating. The generator tweaks some parameters and makes a new image, keeping tracks of which tweaks succeed in producing a better rating. The process is repeated hundreds or thousands of times to get to an image that often looks pretty coherent. The ratings are based on trends detected in millions of training photos with contextual labels.

The algorithm has seen a bunch of examples of things it knows are trees, so it can guess at whether some new pixels look like a tree. This isn't an "artificial intelligence" hanging out in cyberspace dreaming up paintings for us. It's just a lot of cleverly organized math. The difference isn't just brain food for sci-fi nerds, it had a pretty big impact on the idea of authorship.

If a generator trains on a database that includes, say, a hundred different works by Leonard Da Vinci, and then you ask it to draw something in a renaissance style, is it plagiarizing Da Vinci? Who really "created" the image, the person who typed the prompt, the programmer who wrote the code, the code itself? Or is it like sampling a track in a song, the parts that are sampled belong to the artist who created them (or whoever they've sold those rights to)? Even if we're OK with artists stealing from long dead masters, what about when we ask generators to emulate living artists?

But a generator isn't clipping pictures out of a magazine and pasting them together. If you watch the generator iterate, it resolves an image from a haze of random noise until it looks like something recognizable. Perceptually, it is closer to cloud watching than collage.

Recently, I've been trying something kind of new, making subjective illustrations of actual things, instead of just pretty abstract shapes. This has historically been a weak spot of mine, I don't visualize things in my head and my hands are not great a pen or brush work. But using some digital tools and working with vector shapes, I've been comfortable enough to give it a shot. When my partner requested a frog design, I couldn't simply imagine a frog and draw it. I knew things about a frog, it has round, bulbous eyes, it has a wide mouth, it has nose holes and a long tongue. But those are properties, not visual ideas. So I looked for some help.

I searched for cartoon frog faces, and looked at a couple dozen examples. I paid attention to how the eyes sat on the head, the way the curve of the lips defined the shape of the face, and the proportions of the nose holes. Different images had different answers, but looking at a bunch of them helped me understand the essential "frog" appearance they all shared. Then I took my blank page, and lined up shapes in ways that fit the patterns I had picked up on. Soon I had a rough outline of a frog face, and from there I was able to refine it into something charming, cute, and recognizably froggy.

Is that so different from what the generators do? When I did it in my head, I didn't have to worry about plagiarism and copyright. I'm free to look at public art, I'm free to analyze it's patterns, and I'm free to try to recreate them, as long as I'm not using them directly in the process or pretending to be someone I'm not. This is how most of us learn art, by observing the art around us and resynthesizing it with our own creations.

So the argument that these generators are stealing from artists doesn't really resonate to me. Especially when I consider the sources of the training data. While many models are constructed using public domain or academically licenses image archives, others are simply trained off anything they can find online. I am sympathetic to the argument that a creator should have control over how their work is consumed, and that many would prefer not to let a computer algorithm try to emulate them. But I'm wary when we're talking about art that is posted to the public internet by the creator.

We could have a whole long chat about the dangers of our modern ideas of privacy, and how profit motives often make it hard to tell the difference between exploitation and convenience. We'd probably agree on a lot of it. But that doesn't change the fact that when I upload my fractal art to Facebook so some of you will see it, I'm explicitly granting access to that art to a public audience. It sucks that surrendering that control is such a normalized part of being a modern artist, but it is still a thing I'm choosing to do.

And I don't mean that in a nit-picky "gotcha" way. When I'm sharing my art somewhere publicly for free, what is my goal? Maybe to attract interest that turns into sales or support, sure. But I'm also trying to make my mark. When you think of fractal art, I hope my art is some of what comes to mind. Beyond the money to keep making art, I genuinely want to make an artistic impression on you. When someone tells me my art has inspired them to try to make fractals like I do, I don't feel like they are stealing anything by wandering through my gallery. I feel honored!

To that end, I think one of the ways we will adapt to a future with generated art where one of the GOALS as an artist is to become notable enough to be included in a training database. To associate a style with your name so strongly that it biases any generator that invokes it.  To contribute to the collective pool of shared creativity we've been cultivating. Or to curate our own databases of our art and license our styles.

OK, well hopeful creative utopias are fun to dream about, but we can't just ignore the fact that a whole lot of artists are about to be efficiently replaced by automation. Let me be clear, I'm a pretty big fan of artists getting paid to make art, and I know firsthand how hard it is to secure any stability doing it. Watching your industry dry up is scary, and no amount of rationalizing it will keep it from hurting some good people along the way. For what it's worth, I've seen some pretty impressive results come out of prompts as simple as "fractal art", so I'm on the chopping block here too.

But which work am I losing? It's not the fans that commission me. Even with generated art widely accessible, I can build and maintain my audience by marketing my artistic persona. A patron, in the classic sense, is someone who buys my art because they want to support me as an artist. As long as I continue to present and interesting and charming artistic persona around new people, I will attract patrons.

What I won't be getting as much of is people using my art as a service. Sometimes I'm approached by an indie musician who needs an album cover. Self publishing on social media with no budget, they need a trippy cover for as cheap as possible. I don't disparage these customers, it's nice to make a few quick bucks when I have the time. But these aren't the customers I'm seeking. I won't make a successful career out of selling album covers for $25 a pop, even if I got popular doing it. I'm seeking the customers who can pay my full commission price, because that's the rate that I've calculated could make this art thing work for me.

Pretty soon, all of those musicians will have a free website where they can type "trippy fractal album art" in and get something they can use. This means less work for me. It also means the quality of the album covers will trend upwards. In the end, a world where musicians feel like their album art expresses them sounds better than one where they have to pay me for help.

Those customers are the least interested in supporting me. Maybe they are happy to support an indie artist, sure, and I appreciate their business. But once they have the art they need, I'm not a part of it anymore. If I'm lucky, they'll mention where the art came from, or maybe come back to buy more. But these aren't the customers that turn around and buy a large print because they like my art so much.

Sadly, a lot of art industries use the service model as an excuse to exploit artistic talent, and a lot of artists impose the same exploitations on themselves to compete. Drawing commissions of people's pets to help them express their love is rewarding and a way to grow as a pet portrait artist. Drawing 10 pet portraits a day so they can be cheap enough to sell at a sustainable profit is not so great for artistic development. It burns a lot of artists out.

Generated art is going to replace a lot of abusive art industry jobs. Not every job that will be lost is an abusive one, but most of the abusive ones will no longer have a need to abuse human artists. Yes, this means capitalists are reaping the benefits and the working class is left empty handed, I'm not denying that this progress casts long shadows. But it also means that what it means to be a working artist does is going to change.

Who still hires an artist when they have free generated art available? People who want a consistent or particular style, people who want to hire creative insight, patrons of particular artists, and anyone who wants to be sure the art in their project is genuinely expressive. There will still be artists who meet those needs. And in the future, artists who want to be professional will aim to do that, instead of pushing to make 10 pet portraits in a day.

It's fair to point out that there will be a new type of artist, the kind that specializes in coaxing quality out of a generator. After all, someone is still operating the generators to get the art, even if they taking place of dozens or hundreds of other artists. But I think it's pretty clear that we'll be losing more art jobs than we'll be making, so that's not all that much solace. I will say I've been very impressed by artists who have incorporated generator output into a more traditional digital art process. That approach is even how I made my own most popular design. Those tools allow me to expand my creative range, and I've grown artistically in the process of learning to use them.

That's the trade off that I don't think gets considered as much in these conversations. Yes, we will have fewer professional artists doing art as a service. We will also have a lot more artists. The market will be more saturated, but ask any modern creator, they'll tell you it already is supersaturated. You already have to do so much more to be a successful artist, you have to sell the artistic persona I talked about earlier. That approach still works in a market with generated art, so the path to real success isn't really changing with the generators around. And in the meantime, everyone with access to the tech can make art.

Of course, access to this kind of tech is a whole huge issue, with a lot of complicated factors. This article is already sprawling enough, but I need to at least mention that this tech relies on a lot of modern privilege, and the people we're failing to include in our digital world are getting left behind at an increasing rate. I may use linguistic shorthand in this article, but it's not fair to say we're all going to benefit from this tech, because some of us aren't privileged to access tech.

So, we get a lot more artists and a lot less of them get paid for it. That's bad news, right? Well, if you measure the success of an artistic practice through a capitalistic lens, yeah. Skilled labor is losing value, and people who practice it gain less for trying. But I can't ignore the fact that there's a lot more to artistic pursuits than making money. About the time I was ready to start selling my art in person, a global pandemic convinced us to stay inside for a year. That was a year where I made almost no money off my art. It was also a year where I grew a tremendous amount as an artist. Knowing that it would be a while before I could show off, I stopped worrying about what would sell, and started focusing on what was artistically compelling. By the time the markets came back, I had a much more developed artistic style.

Creating environments where artists have time to put into artistic pursuits is a matter of economics and policy. A lot of industries are looking at accelerating automation with the same fears of being replaced, not just artists. Writers, designers, programmers, managers, customer service, medical diagnostics, musicians, advertisers, journalists.. Generated content is impacting an increasingly large swath of labor. So at a fundamental level, we as people need to come up with a new plan that makes room for people to live even where there aren't enough jobs for all of us. This is not a fight for artists alone, this is a fight for humans.

But that fight isn't going to be quick. Transitioning to a automated society will take time and trouble. And during that time, whether we like it or not, this creative technology will continue to make art more accessible to more people. So another shift we will need to make is to encourage the value of art for non-commercial purposes. We can't all be popular artists, and maybe none of us should get all of the attention. It's probably better if we learn to create within smaller creative communities, where we can engage with each other, watch each other grow, and encourage each other with our explorations.

When we don't have to compete over who can sell out art, we can indulge in much more interesting, personal, and passionate art. We can progress our artistic culture so much more when we have more access to creative self-expression. If we can teach each other to value this cultural development as much as we've valued economic development up to this point, maybe we can stop producing leaders that prioritize profits and power. At the very least, we can stop sacrificing new artists at the alter of the popularity algorithms.

Generators are going to be awesome. They're going to be terrible. They're going to lift us up, and they are going to hurt us. They are tools, and we will develop better understandings of when and where to use them as we practice them more. Just like the internet, or the printed page, or the paintbrush. A favorite writer of mine offered some advice: "When you imagine a new tool, imagine it in your enemy's hand, with the sharpest part against your neck." Do that. Understand that a tool is also a weapon. But don't forget, you get one too, and maybe you can find something more interesting to do with yours than hurt people. When you do, we're all going to want to see it.