Summary

Core Thesis: Art as a Social Behavior

The central argument of this talk is that art is fundamentally a social behavior—a communicative act performed by humans, for humans, to affect social relationships, bond, and share culture. Consequently, computers and AI cannot be considered "artists." All computer-generated or AI-generated art is ultimately human-made art, with the computer serving as a tool. True automation of art is impossible because the human origin, context, and intent behind an artwork are intrinsic to its value.

Historical Parallels in Art Technology

The speaker draws extensively on art history to demonstrate that while new technologies inevitably disrupt artistic labor and methods, they do not replace the human artist.

• Oil Paint vs. Fresco: When oil paint emerged, masters of the older fresco style (like Michelangelo) dismissed it as amateurish. Yet, it allowed for unprecedented realism and fundamentally changed painting.

• Photography vs. Painting: The invention of photography in the 19th century caused panic among traditional painters. Paul Delaroche famously declared it "the end of art." Photography decimated the livelihood of portrait painters by offering a faster, cheaper alternative. However, it ultimately liberated painting from the burden of pure realism, directly ushering in the Modern Art movement (as championed by Van Gogh and Whistler) and eventually earning recognition as a distinct art form.

• Recorded Music vs. Live Performance: At the dawn of the 20th century, musical recording was attacked by figures like John Philip Sousa, who feared it would destroy the soul of music and turn people into automatons. While it did reduce communal music-making and displace performance musicians in theaters ("talkies"), it also democratized music appreciation and birthed entirely new genres, such as musique concrète and hip-hop.

The Evolution of Computer Art

Computer-assisted art has a 60-year history, and the debate over machine autonomy is not new.

• Conceptual Foundations: Sol LeWitt's idea that "the idea is a machine that makes the art" paved the way for algorithmic art.

• Early Generative Art: Artists like Harold Cohen spent decades writing complex algorithms (like AARON) to generate paintings. Despite the autonomous execution of the code, Cohen realized the machine lacked a "modifiable worldview"—it possessed no independent intent.

• Computer Animation: Early digital animators feared computers would steal their jobs. Instead, 3D animation became a highly labor-intensive, human-driven artistic performance. Pixar’s early mantra, "Art challenges technology, technology inspires art," reflects this synergy.

The Danger of AI Hype and Anthropomorphism

The speaker warns against the language of "Artificial Intelligence," which invites false comparisons to science-fiction characters like those in Star Wars or Star Trek. Modern AI (like DALL-E or ChatGPT) is essentially a complex, high-dimensional curve-fitting procedure, not a conscious entity.

• Historical Hype: In 1958, the Perceptron—a basic linear classifier—was hyped by the press as the embryo of a conscious machine.

• The Illusion of Agency: Humans naturally anthropomorphize machines (as seen with the 1970s ELIZA chatbot). Attributing artistic agency to AI ignores that its output is the result of human-written code, human-curated training data, and human-inputted prompts.

• The "Button Press" Argument: Critics argue AI text-to-image isn't art because it's just pressing a button. The speaker counters that photography is also just pressing a button; the artistic merit lies in meaning, expression, and context, not the physical labor of rendering.

Future Predictions and Ethical Implications

Looking forward, the speaker predicts that simple text-to-image generation is a superficial fad. The true impact of AI will be found in deeply integrating data-driven techniques into professional artistic pipelines, offering unprecedented control and birthing entirely new styles.

• Labor Disruptions: Just as streaming triggered the 2023 Hollywood writers' strike, AI will cause painful short-term labor disruptions, fundamentally redefining what it means to be an artist.

• The Necessity of Guardrails: The widespread impact of these tools will necessitate new ethical frameworks, guardrails, and copyright considerations to protect human creators.

• The Enduring Value of Human Art: Ultimately, pure automation is uninteresting to audiences. Because art is a social behavior, we will always seek out the human connection behind the work. Good art will remain difficult to make because it requires a unique, meaningful human perspective.

Transcript

Introduction: Can Computers Create Art?

In this talk, I'm going to discuss the question of whether computers can be considered artists, including so-called AI. To do so, I'll describe many other times in history when technology changed the way that we make art and the way that we understand art. And I think that looking at this history will provide a lot of useful lessons for the challenges that we face today. I'm going to explain why I believe that art is really a social behavior, one that we do for and with other humans. This talk originally began as a paper that I published way back in 2018, and a lot has changed since then, but a lot of the lessons of history remain the same.

Before I begin, a bit about myself. I studied art and computing in college, and over the past 30 years, I've done research in computer graphics algorithms inspired by art. I also still like to spend lots of time drawing. My art experience has informed a lot of my research over the years. So, for example, here's an installation that we made as part of my PhD thesis. It's a canvas that continually paints a picture of you as you move around in front of it. And we showed it at some art exhibitions in New York in 2001.

The Rise of AI-Generated Art

The past decade or so has really felt like a whirlwind of activity and attention for AI-based art. I date this to 2015 with the introduction of DeepDream, which was a kind of fun, interesting technique for a while. Neural style transfer was presented in 2015 as well, and this was a fun toy a lot of people liked to play with. Both of these methods were shown in art exhibitions, such as this one that was shown in San Francisco in 2016. Another notable moment was the auction of a GAN-generated artwork for half a million euros, and this was even signed with the GAN loss function.

Now, academic researchers have gotten involved in making statements about AI artwork. Here's a technique from a paper called "Creative Adversarial Networks," and it has a very interesting approach in the paper to defining new visual styles. But the thing I want to focus on is in the abstract where they say, "We propose a new system for generating art. The system generates art by looking at art and learning about style and becomes creative." So the authors are making very strong statements about the role of the computer in the system.

This was picked up by the news media. Headlines said, "Artificially intelligent painters invent new styles of art," "The artist that can create its own painting style," and "Critics even prefer some of its work to human efforts." And again, this is back in 2017. Here's a video of one of the preeminent art critics in the world, Jerry Saltz, talking about that algorithm. And the thing to notice here is the level of agency that he gives to this piece of computer code: "Initial thoughts: incredibly dull, generic, boring. The programmers are not freeing up the program. I want the robot to tap into its inner robot. Be free." So it's a simple GAN model, but he's essentially presented it as though it's an independent artistic entity.

More recently, in 2022, DALL-E was made available to the public. You could just type in a bit of text and images would be generated, and suddenly this created a lot of excitement, energy, anger, and controversy. And this is kind of the world we live in now.

Framing the Question: Human vs. Machine

In this talk, I want to focus on the question of whether computers can create art, whether we can think of computers as artists. Often, people treat this as just a matter of technological capability—that if the pictures are good enough that come out of the machine, then that makes the machine an artist. On the other hand, other people I've talked to say, "Of course computers can't be artists. Art requires intent or expression." There's a sense that art is fundamentally a human activity, like having a soul. I agree with a lot of these intuitions, but they're not very scientific, and it would be nice to make them a little more concrete—like, what do they actually mean? What are they saying?

The main points I'm going to make here are that people make art, sometimes using computers, and so all of the art that we make with computers is human-made art. This is in part because art is a social behavior. It's a thing that we do with and for other people. However, new technologies transform art, the way we make art, and the way we appreciate it. History provides many useful lessons for what happens when these technologies come along and the way people respond to new artistic technologies.

Now, these are very, in some cases, controversial topics. There are a lot of concerns and strong emotions here, often for very good reasons. There are lots of legitimate concerns around copyright, but I'm not going to discuss that here so much. I'm not making policy recommendations. I'm really here recommending ways to talk about art and technology that avoid a lot of the pitfalls. There are a lot of ways to have a gut response to these new technologies that are really quite misleading, and I think it's worthwhile being a lot more careful about how we think of the role of each of these new technologies.

Historical Lessons: Oil Paint and Photography

As you know, one of my main themes here is that technology has transformed art many times throughout history. One early example is the development of oil paint. Oil paint, as compared to the earlier technology of fresco, has a much greater tonal range and can depict basically more colors. It's also much easier and more practical than fresco. In response, Michelangelo, who was more a master of the older fresco style, said the new stuff is for amateurs. Within the Western tradition, over the following centuries, artists got better and better at using oil paint to create highly realistic and dazzling depictions of reality. I think it's really hard for us right now to appreciate how special it would have been to see pictures like this. Today, we are surrounded by photographs and realistic imagery online, in print, and on our cameras. It's hard for us to appreciate just how special it would have been to see what could only have been done by a painter 200 years ago. So the role of the painter and the identity of the artist were very highly tied up in their ability to create these kinds of pictures, which only they could make.

At the same time, photography was initially invented in the early 1820s. The first known existing photograph was initially made by technology tinkerers playing around with chemistry and optics. Here's a picture that Daguerre took out his window. Because these were long exposures, around 10 minutes, most people walking by would be blurred out. But you can see there's a fellow getting his shoe shined in the lower left there. And so this is the first known photograph of a person.

This technique really became widely available when Daguerre publicly described his invention. The French government bought him out so that he wouldn't patent it, because they believed that this technology was widely useful. Immediately with these first demonstrations of the technique, traditional artists said, in the words of the great Paul Delaroche, "This is the end of art." Or J.M.W. Turner, who said, "I am glad I have had my day." Because here is a machine that does what artists do: it makes realistic pictures.

So what actually happened? Well, one of the roles of painting was portraiture. Just as today we like to have pictures of ourselves, our family, friends, and ancestors. In 1838, if you were very, very wealthy, you could hire a painter to paint a picture of you. If you were not so wealthy, you might have a silhouette picture made—not a great likeness. Once photography came around, portrait studios emerged where people could get their portraits made through photography. And even though you had to sit with your head in a brace and sit very still for 10 or 30 minutes, it became very, very popular. As a result, we have all these wonderful photos by people like the colorful photographer Nadar of figures from the 19th century, such as this picture of the photographer Mathew Brady.

Within several decades, painters, at least one painter, said, "Photography has harmed painting considerably and has killed portraiture, especially once the livelihood of the artist." This is because portrait painters were unable to find work in the way they had before because photography was faster and cheaper, and so painters either had to switch to photography or were unable to find work as portraitists.

On the higher end, there was a discussion of whether photography is art. On one hand, there were the tinkerers and people exploring the technology, making pictures and seeing what happened. One thing that often happens with a new technology is that people try to mimic the existing artistic styles, both to develop the technology artistically and also to justify it. With photography, that movement was called Pictorialism. Here's a classical style tableau created through a multiple exposure composite within Pictorialism that had the same kind of horizontal tableau as a lot of classical painting. It took many, many decades before photography emerged into its own. It eventually had its own style, its own language, and was, by the beginning of the 20th century, accepted by major museums and galleries as a separate art form.

Now, throughout this, there were of course the haters, people like the poet Charles Baudelaire, who said, "If photography is allowed to stand in for art it will corrupt it completely. Thanks to the stupidity of the multitude." So this technology is going to replace art and make it stupid because people are stupid.

Ultimately, painters began to see this as a challenge. Whistler wrote, "The imitator is a poor kind of creature. If the man who paints only the tree, or flower, or other surface he sees before him were an artist, the king of artists would be the photographer. It is for the artist to do something beyond this." And so he made these atmospheric gauzy paintings that were quite different from what photography was capable of at the time. Vincent van Gogh, in his pivotal year of 1888, wrote to his brother that accurate drawing is not the thing to aim at, because a reflection of reality would not be a picture at all, no more than a photograph. So now this is a 180. He is saying that actually making realistic pictures is not what artists do, because that's just photography. And this whole story ushers in the Modern Art movement of the early 20th century, where making realistic pictures is no longer viewed as one of the major goals of painting.

Conceptual and Early Algorithmic Art

An important step in the development of Modern Art is the notion of conceptual art, especially this famous work by Duchamp, which, if you're not familiar with it, is a urinal that Duchamp turned upside down, signed, called it a fountain, and submitted to an art show. This is considered one of the most famous and influential works of 20th-century art.

So in summary, it looked like photography automated art because it does what artists do: it makes realistic pictures, and many artists feared it and condemned it for those reasons. What actually happened is it created a new art form—photography—which is considered distinct from painting. It invigorated the old art form of painting; I would argue that modern art emerged in part because of photography. Many jobs were affected; many portraitists were replaced with photo studios, and so they had to retrain or find other work, or they lost their jobs. Furthermore, image creation was made much more easily available to hobbyists. Nowadays, we're all carrying phones around in our pockets with cameras attached to them, making it very easy for all of us to make pictures all the time in ways that we couldn't have done 200 years ago.

There are many trends here. They're very complicated. And I argue that in some form, many of these trends repeat with each new art technology to varying degrees. There are many common features that different technologies share when they change art. Some people have responded to this history by saying I'm making it sound like there's nothing to worry about with the new changes, and one could ask, "Are the new changes different?" Of course, every new technology is different. There's a lot now that people didn't have to worry about 200 years ago. My only point here is that what's different is not the things people usually think they are. I think studying these trends of history is a way to avoid naive gut reactions and cognitive biases.

For example, people have complained that the new AI tools can't be art because it's just "pressing a button." But if that's true, then photography is not art as well, because photography is also just pressing a button. Whatever concerns we have about the new technology, they have to be a little bit more thoughtful than that. If you can apply that same concern to photography or other things I'll talk about, then that may be a problem for that criticism.

As part of this discussion, I may need to tell you a little bit about what art is, or the kind of art I'm talking about here. I use the term art very broadly. I talk about visual art, photography, conceptual art, and I would say children's drawing and amateur drawing is also art. Movies, music, video games, theater, and many other things. I take a very broad notion of what art is, and I don't see the value in slicing it up more finely than that. Even though most of my examples in this talk are primarily visual art, some people have inferred what my definition of art is. The thing is, there is no single definition of art. I think it's not really possible to come up with a short, simple definition of art. If you want to know more about that, I recommend the book The Art Question by Nigel Warburton. A lot of people's intuitions about the definition of art don't really generalize well or follow through if you think through the implications. But I still think it's useful and worthwhile to talk about art without trying to come up with a strict definition.

Let's come back to conceptual art. Sol LeWitt is another important figure in conceptual art. These are examples of paintings that he made. In these cases, he didn't actually make them by hand himself, though. He wrote down sets of instructions, and then other people would actually execute the paintings. In his writings about conceptual art, he argued that it's really the idea behind the work, the definition of how the paintings are made, that is the real work of the artist. He wrote, "The idea becomes a machine that makes the art."

This very naturally leads into early computer art. In the 1960s, as soon as people could make pictures with computers, they started making art with it. These are three examples of generative art where people wrote code that made artworks. The one on the left is a picture in the style of a particular Mondrian painting. Throughout the 20th century since then, there's an enormous, amazing variety of computer-based art of all different kinds, which I clearly don't have time to summarize in this talk because it's a vast history. I'll just mention three of my favorites: the evolutionarily generated screen savers of Scott Draves, which autonomously evolve over time in response to people's upvotes or downvotes; Jason Salavon's visualizations, including this picture which is an average of many different wedding photos over time; and Sofia Crespo's interesting GAN-generated artificial botany.

The Myth of Algorithmic Autonomy

One of the most important AI artists in the 20th century is Harold Cohen. He began as a contemporary fine artist from one of the major art schools, the Slade School of Art in London. He started out writing rules for himself to follow when painting by hand on paper, and then he would simply follow those rules and see what happened. In the early 1970s, he discovered computer programming in FORTRAN. From that point on, he started writing those rules in code rather than by hand. He spent the rest of his artistic career programming algorithms that made paintings. These are paintings collected by major galleries and institutions. There was a recent retrospective of his work at the Whitney a few years ago.

In the more popular realm of computer animation, Alvy Ray Smith tells a story about how, before he co-founded Pixar with Ed Catmull, they would go down to Disney and try to convince them to adopt computer animation tools. They said that back then, the animators were afraid of the computer. They thought it was going to take their jobs away. They spent a lot of time telling them that the computer is just a tool; it doesn't do the creativity. Indeed, if you've ever worked with computer animation tools, you know how incredibly labor-intensive they are, and how much talent, skill, and artistic ability they require. Computer animation is very much an artistic performance in much the same way that previous hand-drawn animation had been. This is why Pixar's early days were really driven by the mantra: "Art challenges technology, technology inspires art." Nowadays, if you watch the credits of any animated movie, you see an enormous number of artists were employed in order to create the animation. This is also true for so many of our live-action and VFX films that involve computer animation. There's an entertaining series of videos online about how "no CGI is just hidden CGI," which I recommend watching if you're interested.

In my own work, I began my studies with my first paper, which was a technique for taking a photograph and making a painting from it. This was published at SIGGRAPH 98. This original paper was based on just a series of rules and instructions that I wrote that used the material source image to decide where to place brush strokes. Anyone can read the paper and understand the reasoning and decision-making process involved in this algorithm. This is the source of the interactive installation that I showed earlier. Now, through this process, I found it difficult to define different rules for making paintings, and so I came up with the idea of doing it from examples. This is a paper called "Image Analogies" that we published in 2001. In this example, it's using elements of the texture in the picture on the top and applying them to the photo on the left. Even though it's learning from examples, you can again read the paper and understand how the decision-making process works and how the pastel illustration is being made.

I've summarized this long history of computer-generated art. Throughout this whole history, people are saying, "The robot can paint, but is it art?" Go back 40 years: "Computer art, is it really art?" For the past 60 years, people have been making art with computers and asking the same questions over and over again. Each time someone sees that the computer made a picture, they ask if the computer is an artist. The answer has always been the same for 60 years: the computer is a tool for people to make art. All computer art is really human-made art.

Now, in many of my discussions over the years, I've seen a lot of people make the explicit statement that being an artist is a matter of making good pictures. If you can make good pictures, that makes you an artist, and it's the same for a machine. I think this is a thought process a lot of good people go through in various ways. I want to look at how Harold Cohen went through that process:

"Ten years after that, I would have said, 'Look, the program is doing this on its own.' Another ten years on and I would have said, 'The fact that the program is doing this on its own is the central issue.' Here it was producing complex images of a high quality and I could have had it go on forever without rewriting a single line of code. How much more autonomous than that can one get?"

So he's saying, "I have a computer algorithm that makes pictures. They're being sold, framed, and displayed in major art galleries and museums. Doesn't that make it an artist?"

He continues:

"Well, of course, that's exactly the point. It's virtually impossible to imagine a human being in a similar position. The human artist is modified in the act of making art. For the program to have been similarly self-modifying would have required not merely that it be capable of assessing its own output, but that it had its own modifiable worldview to provide a basis for any meaningful assessment."

It's a thought process a lot of us have gone through. I made a computer algorithm that generates images. Maybe that makes it an artist. And then you keep using it over and over again, and you realize it actually feels like it's missing something. Harold Cohen thought it was missing a modifiable worldview.

Other people have made other statements for what computers would have to have to make them autonomous artists. Yet whatever statement you make, there is some existing code or algorithm that actually does that. For example, people often define creativity in terms of making things that are aesthetically valuable, surprising, novel, and perhaps unpredictable. Here's a video of the Mandelbrot set, which was developed in the 1980s. You can watch it for a very long time; it's visually dazzling, really fun to watch, and unpredictable. By that definition of creativity, it's quite creative, and yet it's 10 lines of code. Chaos theory tells us it is unpredictable. If those were enough definitions to make an artist, then this would be an artist. Same for the code that I wrote as part of my thesis. You can't tell exactly what it's going to do. And so, by a lot of definitions, it's an artist.

When people say, "If AI makes good pictures, that makes it an artist," I would say we have over 150 years of history of machines that can make art by that definition. We have photography, smart phone cameras, generative art, and all of the artwork that Harold Cohen's machines made. If it were the case that that was enough to make an artist, then we would already be calling these things artists. Yet instead, it's the case that computers are yet another technology that people use to make art, even when the code is running autonomously. In short, everything that we call computer-generated art is really human-made art.

People say, "Okay, well what about text-to-image? Surely that makes it an artist." I say it's the same thing: you type in text, you produce pictures. These things on some level draw better than I would, and certainly faster, and people ask the same question: whether it's art. I say, of course it's still art. It's art made by a person using an algorithm. The question we should really be asking about this is: Is it expressive? Is it meaningful? Is it ethical, beautiful, skillful, culturally significant? These are the axes on which we should discuss these things. Whenever we get into discussions about whether it's art or not, it ends up just being a huge waste of time, or we're talking past each other. It could be art, but it's bad art. It's just not meaningful. It's sloppy, or not expressive. That doesn't make it not art, it just makes it bad art, or unethical art.

Moreover, I really want to argue the point that it matters how a work of art was made. I've talked to a lot of people, at least within computing, who seem to believe that the only thing that matters is the visual effect that a set of pixels has on you. It doesn't matter where they came from. I want you to look at this picture here, and I'll tell you a few different versions of where it came from, and just see if you feel differently about it with these different stories. It could be that I painted this by hand with real oil paint. Or maybe I just typed in a text prompt and this was generated in ChatGPT. Does that change how you feel?

Let me tell you where this actually came from. I volunteer at an animal shelter where I spend a lot of time with shelter dogs. When you do that, you form relationships with them; you feel connected, emotionally attached to them. This is one dog that I spent a lot of time with. I took the photo on the left when I was hanging out with him, petting him, and he was looking me in the eye. It was really a moment. The picture on the right I made later by painting by hand digitally on my iPad, using the photo on the left as a reference. I would argue that this story gives you a different relationship with the picture than you had before. It means I have a very different relationship to the picture than other people do. And other people who knew this dog and know me have a different relationship to it than people who don't. These differences are very important. Where the picture came from and how it was made is very important. How much you know about that, or what your assumptions are, really affects the way you approach the artwork.

The Illusion of AI and the "Hype Cycle"

Now, all this happens within the context of AI hype. This is really what I see as the most dangerous part of all these new trends. AI hype is not new. With the development of the Perceptron in 1958, the New York Times said on the front page that the Navy revealed the embryo of an electronic computer that it expects will be able to walk, talk, see, write, reproduce itself, and be conscious of its existence. This was just a perceptron. This is equivalent to adding up a set of rows in an Excel spreadsheet and comparing it with zero. It's a simple linear classifier, and this was claimed to be the foundation of consciousness.

Over the years, we have these headlines about artificially intelligent painters and computers magnifying this hype beyond all proportion. I think part of the problem is just the phrase "AI" and "artificial intelligence." If it was up to me, we would ban this phrase. Part of the problem is that the phrase AI, to most of the public, signifies artificial intelligences that are much like people, because we have in our popular culture artificial intelligence like the friendly droids from Star Wars or the psychopathic Terminator. When the term artificial intelligence is used, it refers to these kinds of things. Yet our science fiction is not really about how our algorithms actually work. They're really about people, and these fictional AIs are really kind of like modified people.

What we're calling AI is really primarily a set of complicated data-fitting procedures. Instead of fitting a line to a set of data points, it's about fitting extremely high-dimensional, complicated functions to very, very large datasets with massive amounts of resources. But in the end, it's still just fitting curves to data. I do not consider that human-like intelligence on many axes. We as humans have a very strong propensity to infer intelligence and agency in things that we don't really understand. For example, the ELIZA chatbot was developed in the 1970s as a parody of talk therapists. It's a really simple set of rules that asks you questions and reflects back based on things you said. People had emotionally intense relationships with it until they found out how it was made, and then they felt a little bit let down.

When you say, "My AI algorithm is an artist," that tells people it's like Data in Star Trek—that it is a human-like entity. I think that's actually an irresponsible statement. Again, I argue that all of what we call AI-generated art is computer-generated art, which means it's human-made art. Humans using computers to make art, often through these so-called AI algorithms. With database algorithms, the authorship is diffuse; different people are involved in different parts of the process that led to the final work. But again, in the end, it is ultimately a human-driven process.

Art as a Distinctly Human Behavior

So, that's a discussion of where we are today with computer-generated art. But things could change. Maybe in the future, we will consider computers to be artists. I want to ask whether that's possible. Essentially, another way of asking this is: why haven't we accepted computers as artists already? What would it take for us to agree that the computer would be an artist? In order to do this, I want to think about what it means to be an artist. What does it take?

I think these are really social behaviors. What do I mean by social behavior? A social behavior is something that we do, at least in part, to affect our social relationships—like conversation, gifts, having meals together, and fashion. These are all things that have their own benefits. We wear clothes to protect our bodies from the environment, but we choose which clothes we wear often based on how we want to be seen and what we want to communicate. But fashion is really a social behavior.

I claim that art is also a social behavior. We share our art with other people. We go to see other people's art, often together. We like to talk about art with other people, or talk about movies, books, and music. We teach it to the next generation to communicate our values and our culture. We buy it and display it publicly to communicate our taste, values, or wealth. These are things that have happened since before written history. There's an argument that art is really a product of our evolution. I really like the book The Art Instinct by Dennis Dutton. The claim is that art is a product of our evolutionary history—something that we do for gifts, sharing, status, mating, and so on. Ultimately, all of these different functions of art throughout human history are about our social relationships. There's another paper making the case specifically for music, arguing our ability to make and care about music is a product of using it for social bonding.

When I've given this talk, people have said, "Well, I make art just for myself, so doesn't that disprove your point?" I think that's great. I find making art to be a valuable, personally fulfilling thing. But a lot of our social behaviors have their own individual benefits. Language is for communication, yet people might talk to themselves, sing to themselves, or take notes for themselves. That doesn't mean language isn't fundamentally for communication.

In short, art exists for us to help affect our social relationships. As a result, we care about art made by people. Only people make art, and computers are not people. I think this is a really important point. This is why we have not accepted computers as artists in the past, and why I don't think we're going to anytime in the near future.

To illustrate these points, let's look at alternative examples. There are many natural processes that produce things like landscapes, flowers, and trees that have a lot of the properties we associate with art. They're beautiful, emotionally impactful, and create meaningful experiences, and yet we don't consider the ocean to be an artist. Conversely, we do have social relationships with some animals. I think we're very open to the idea that if a dog painted a picture and really seemed to care about the painting itself, we'd be open to the idea that animals are artists. It's just that we haven't found animals that really seem to care about a painting as an aesthetic artifact, rather than just a fun activity.

If that's true, then we have an answer for when we might agree that a computer could be an artist. If we developed human-level personhood—AI that we truly see as people just like us—then we would consider them artists. But this is science fiction. The algorithms we have right now are not people. They are text generators or image generators fitted from data that follow understandable code. I don't see how we can see them as artists. There are various kinds of AI designed to be social and conversational—Siri, Alexa, Cortana, ChatGPT—and we are hearing stories of people having intense emotional or romantic relationships with chatbots. Personally, I find that really scary, and I don't think we should consider them as people. But that is the situation in which we would accept them as artists.

Another thing I've heard, especially among computer science audiences, is that we are all just computers, or the brain is just a computer, or that the whole universe is a computer, and therefore that computer could be an artist. Theoretically, in a science fiction world, maybe these are true statements, but they are not true statements about today's computers. I'm not talking about 1,000 years from now. I'm talking about the computers we can actually build right now. When I say computer, I mean a laptop, CPU, GPU.

Here's a picture of when I adopted my dog. That's a person, and there's a dog. If it were the case that people are just computers, then we would be saying that computers and people are morally, ethically, and socially equivalent. I see no moral or ethical problem with wiping the memory in my computer, throwing it away, erasing it, or recycling it. Doing that with another person or with a dog would be illegal and morally unacceptable. Anyone who says people are just computers needs to think about the difference between a person and a computer, because I think there's a very sharp divide here. It's okay to wipe computers, and not okay to kill people.

In short, computers are not people, and people make art. All of our computer-generated art is really human-made art.

The Consequences of Technology: Labor and Culture

As I mentioned, when I've given this talk, people have said, "Well, it sounds like you're saying there's nothing to worry about and everything's okay." That's not what I'm saying. I am saying that in the long run, we are always going to care about human artists. So in the long run, art is safe from complete automation. But in the short term, there can be lots of problems, concerns, and disruption. Moreover, these things aren't simple. It's not just "computer AI good or bad." It's much more complicated. I really like the writing of the historian of science Melvin Kranzberg. Kranzberg's laws state that technology is neither good nor bad; nor is it neutral. It can be a very powerful force for good and for bad, and it's worth understanding the different kinds of consequences it can have.

I want to give a final example with a different artistic technology: musical recording. This was invented at the end of the 19th century by Edison. Before music recording, the only way you could hear music was to be in the room with the person performing it. With musical recording, that was no longer the case. You could buy recordings and listen to them later. It became very popular, of course. Within a decade or so, John Philip Sousa, the famous composer, wrote this wonderful essay called "The Menace of Mechanical Music," in which he wrote that he foresaw a marked deterioration in American musical taste due to recording technology. It's easy to laugh at the ridiculous things he said—he made scary predictions about how recorded music was going to turn children into automatons and remove soul from music. But he also made points that are relevant to our discussions today. This article was really part of a successful campaign to add copyright protection for composers.

But I want to focus on the communal aspects. Before recorded music, people would get together, families and social groups, and make music together. There's strong evidence that the function of music is for social bonding. I've heard that making music together is a really powerful social bonding activity. It's something that is relatively rare in our modern lives because we can just buy recordings instead of learning to play instruments.

There was another campaign against recorded music with the invention of "talkies"—the ability to have soundtracks in movies rather than having performance musicians in the theater. Performance musicians were worried, claiming recorded music has no soul. Indeed, a lot of performance musicians lost work as a result of this technology. On the other hand, we got new media and new styles, such as tape loops used by pop artists like The Beatles and Pink Floyd in seminal works. Moreover, the development of recorded music was absolutely key to the development of hip-hop, where someone can create entirely new kinds of music by sampling elements of existing recordings using just two turntables and a fader.

The development of recorded music, in addition to creating new styles, created a lot of labor disruptions. Such as the Hollywood writers' strike in 2023, which was essentially the result of streaming technologies. Again, a different way to record and distribute media allowed studios to exploit loopholes and change the terms of how writers were paid. If you look through the history of writers' strikes, many were essentially responses to technological shifts that changed the way writers were paid. Or, as a friend of mine who was active in the latest strike puts it, "They use each new technology as a new way to screw us over."

In short, musical recording and streaming made musical appreciation much more convenient. We can all listen to music all the time. Revolutionary new kinds of music came out of it: musique concrète, tape loops, and hip-hop. On the other hand, most of us don't make music socially anymore. Moreover, artists really struggle to get fair pay; music streaming online is especially bad for artists right now. And yet, it's very hard to imagine giving up music streaming.

Conclusion and Future Outlook

With all this background, we can make a few predictions about what the new tools will do for art in the future. It's hard to make specific stylistic predictions. Here's Les Paul, the inventor of the electric solid body guitar in 1947. He played swing tunes. I don't see how he or anyone else could have predicted how his invention was going to change music and popular culture. You just can't predict what a new technology is going to do to styles.

Yet, with all the trends I've talked about, we can still predict gross trends. Future AI tools are not going to look like simple text-to-image. So much of the debate has been around typing in a text prompt and getting an image. That is a fad. It's often very boring, superficial art. The interesting action is where data-driven techniques occur deeply within existing artistic pipelines—things that give artists lots of control to make things that are truly unique in ways that are very different from how they worked in the past. Artists are going to find great ways to use these tools that are much more powerful than just a text prompt, and new styles and techniques will emerge.

This also means that non-experts will have new ways to create and communicate. However, it's going to cause many painful disruptions for artists and a redefinition of what an artist is. That's going to be really painful and change a lot of how artistic production works. It creates difficult ethical challenges: Who benefits and who is harmed? What guardrails or adjustments need to be put in place to account for these changes?

I believe that good art will always be hard to make. There will always be discerning audiences who care about the human behind the work. Pure automation is not interesting. We care about things that are unique and special, and that requires more than just typing a prompt. We're always going to care about artists. AI is not a human-like person and therefore not an artist. All so-called AI-generated art is really human-made art. This technology is going to create significant benefits and also harm. It's going to change how we make, benefit from, and understand art. The question is: how do we develop tools, guardrails, and safeties so that we, as much as possible, benefit our own humanity through these new technologies?