AI is Not Art: A History of Gatekeeping Creativity
- Dave "The Barber" Burdick
- 3 days ago
- 14 min read
There is a sentence being thrown around a lot right now when someone admits they used artificial intelligence to create something.
“You didn’t make that. The AI did.”
Sometimes it gets shortened to, “That's not art.”
I understand where some of that comes from. Generative AI can do things that would have required years of training not very long ago. Someone who cannot draw can generate an image. Someone who cannot play an instrument can make a song. Someone who struggles with grammar can produce clean prose. That is a massive change.
It turns out we have been deciding who is and is not a “real” artist for a very long time. The pattern is surprisingly consistent. A new technology gives people abilities they did not have before. Some people do incredible things with it. A whole lot of people make crap. The people who spent years mastering the old way see newcomers skipping skills they had to learn, and pretty soon somebody starts arguing that the missing labor was what made the work legitimate in the first place.
We Have Been Automating Creativity for a Very Long Time
The printing press is an early example. In the mid-fifteenth century, Johannes Gutenberg developed a movable-type printing system that transformed book production in Europe, making written material dramatically easier to reproduce and distribute. It also meant more bad books, bad information, and sloppy editions.
A generation later, Desiderius Erasmus, a Dutch Renaissance scholar and writer, became one of the new technology’s enthusiastic users. Printing helped spread his work across Europe, and he worked closely with printers and was demanding about the quality and accuracy of his editions. But even Erasmus complained about careless printers and “these swarms of new books,” arguing that the sheer quantity of poorly produced material could harm scholarship.
Then things got weird. In 1677, John Peter published Artificial Versifying, a system of rules and tables for constructing Latin verse. Peter claimed it was simple enough to be used even by “he that cannot Write or Read.” Our hypothetical knuckle-dragging mouth-breather only had to pick six numbers and follow them through the tables. One example, using 4, 6, 7, 1, 8 and 2, produces Tristia fata tibi producunt sidera prava, roughly, “Sad fates bring forth crooked stars for you.” Congratulations. Here’s your shitty poetry.
Then it got weirder. In 1845, John Clark took the idea a step further with Eureka, a machine he had spent more than a decade building to mechanically generate Latin verse. Instead of choosing numbers and following tables yourself, Eureka did the work, combining a small vocabulary into millions of possible Latin hexameter lines. (Alfred Gillett Trust, Cambridge University Press)
Martia castra foris praenarrant proelia multa
Pessima regna domi producunt vulnera mira
The camps of Mars abroad foretell many battles,
The worst kingdoms at home produce wondrous wounds.
Nearly two hundred years before ChatGPT, we had already figured out how to automate questionable poetry.
Around the same period, creatives gained another kind of assistance. In 1806 William Hyde Wollaston introduced the camera lucida in the early nineteenth century. It used optics to let an artist see the subject and drawing surface together, making placement and proportion easier to judge. I have not found evidence of some massive revolt when it appeared, but the cheating argument showed up much later. In 1999, artist David Hockney suggested that the remarkably precise portrait drawings of Jean-Auguste-Dominique Ingres may have been made with the help of a camera lucida. Hockney eventually argued that optical aids had been used by many Old Masters, and the idea caused a serious fight in the art world. Some critics treated the suggestion itself as an accusation that the artists had cheated, even though Hockney argued that using optics did not diminish their artistry.
Then the Machine Made the Picture
Photography came along shortly after and did not merely help an artist draw the image. The machine produced the image.
Photography created a much bigger problem for the old definition of artistic skill. A painter had to build an image by hand. A photographer could point a machine at something and let light and chemistry record it. If the machine was doing so much of the physical work, where exactly was the artist?
The answer turned out to be in the choices. What to photograph? Where to stand? What to include and leave out? When to press the shutter? How to use light, focus and exposure, and later how to process and present the image. Photography removed a huge amount of the handwork required to make an image, but it did not remove the need to decide what the picture should be.
In 1888, George Eastman introduced the Kodak camera, designed so an amateur could simply press the button and let Kodak handle the developing. This produced mountains of blurry children, crooked horizons, cut-off heads and photographs of absolutely nothing interesting. Serious photographers had reason to be annoyed. In 1909, Alfred Stieglitz warned amateurs, “Don’t believe you became an artist the instant you received a gift Kodak on Christmas morning.”
Giving somebody a camera did not automatically give them taste. But bad photographs did not prove photography was not art. They proved that making a tool easier to use does not automatically make the person using it good at what they are trying to do.
That feels extremely relevant now.
When the New Tool Threatens the Old Job
Gatekeeping is not always about artistic snobbery. Sometimes there is a paycheck underneath it.
In 1906, composer and bandleader John Philip Sousa wrote an article called The Menace of Mechanical Music, attacking phonographs, player pianos and other forms of recorded music. He worried that mechanical reproduction would replace human musicians and discourage people from learning instruments. His famous opening complaint described machines substituting for “human skill, intelligence, and soul.”
Sousa was not simply an old man yelling at a phonograph. He was worried about something real. If a machine can reproduce the music, what happens to the person who used to get paid to perform it?
Economists have a wonderfully bloodless phrase for that: technological unemployment.
The same tension showed up decades later with synthesizers. In 1982, when Barry Manilow toured Britain using synthesizers in place of the orchestra he had previously used, the Central London branch of the Musicians' Union responded by calling for an outright ban on synthesizers. It never became national union policy, and synthesizer players eventually became part of the profession.
New technology really can reduce the economic value of skills people spent years learning. That can hurt people, and pretending otherwise does nobody any favors. But technological unemployment does not mean the old craft necessarily disappears. People still paint with oils, shoot movies with film, play acoustic instruments, write with pens, and set type by hand. Sometimes the old way becomes a specialty and more valuable for it.
What changes is which skills are necessary to participate.
That is where economic disruption can become gatekeeping. A skill that once defined the profession becomes optional, and the people who mastered it understandably feel like newcomers skipped something important. But losing the economic value of a skill does not prove that people using the replacement technology are incapable of creativity.
How Much of the Thing Does the Artist Have to Make With Their Own Hands?
In 1917, Marcel Duchamp submitted a manufactured urinal to an exhibition, signed it “R. Mutt” and called it Fountain. He did not sculpt it or manufacture it. He selected it, changed its context, and presented it as art. The work was rejected, and people are still arguing about it more than a century later. An editorial defending Fountain at the time made the argument that whether the artist made the object with his own hands was not the point. He chose it and gave it a new meaning.
Andy Warhol pushed the artist’s hand even farther away from the finished object. He used photographic silk-screening, a commercial reproduction technique, and assistants helped produce work in his Factory. Warhol even said he wanted to be a machine. Yet nobody seriously argues today that Warhol's work is not art because a commercial process and other human hands were involved. (The Andy Warhol Museum)
His Flowers series was based on photographs taken by Patricia Caulfield, who took legal action against him and eventually settled. There was a legitimate dispute over the source material even while Warhol’s resulting work remained art. Those are two different questions, and we are going to run into that distinction again.
Film complicated authorship in another way. When the Museum of Modern Art created its Film Library in 1935, it faced the challenge of “articulating a pastime as art.” Today nobody finds it strange to call a director an artist even though a camera records the actors, an editor assembles the footage, and an entire crew contributes to the finished work. We learned to judge the movie as a whole instead.
Then artist Harold Cohen made the comparison to AI almost uncomfortably direct. Beginning in the late 1960s, Cohen developed AARON, a computer system that generated drawings according to rules he created. In 1983, the Tate Gallery exhibited work involving AARON, and people were already debating whether Cohen was the artist or whether the computer was.
Everybody Got the Tools
Computers accelerated the pattern. Word processors removed a lot of the mechanical labor of writing. Desktop publishing put professional-looking page layout tools onto personal computers, and the early Web let almost anybody publish to the world. Naturally, people immediately made ugly newsletters, unreadable pages, and websites covered in blinking text and bad clip art. But the tools were not the problem. Making creation easier increased the amount of bad work, while also giving more people the chance to learn how to make something good.
Photoshop arrived in 1990 and made sophisticated image manipulation dramatically more accessible. That created real concerns about deception, especially in journalism and documentary photography, where people expect an image to represent what actually happened. But the same tools also became ordinary parts of illustration, advertising, filmmaking and digital art.
In 2006, Reuters withdrew a photograph by freelance photographer Adnan Hajj after discovering that Photoshop had been used to add and darken smoke in an image of Beirut after an Israeli airstrike. Reuters cut ties with Hajj and eventually removed all 920 of his photographs from its database. Hajj denied deliberately trying to mislead anyone, but Reuters considered changing the content of a news photograph a serious violation of its standards. (The Guardian)
We see the same potential for deception with AI. A manipulated photograph, an AI-generated image, or any other creative tool can be used to mislead people. The ethical problem is not simply that the tool can create something convincing. The problem is what the person using it intends to make people believe.
Standards Are Not Gatekeeping
This may be the most important distinction here.
Saying something is bad is not gatekeeping. Some AI art is terrible. So are some photographs, novels, movies, websites and oil paintings made entirely by humans.
Making a creative process more accessible increases the number of people who can create something with it. It does not magically give those people taste, judgment, or an understanding of what works and why. When access spreads faster than those things develop, we get slop.
AI is doing the same thing on an enormous scale.
Standards, understanding, and taste still matter. Criticism matters. Bad art is bad art.
Gatekeeping begins when we stop judging the work and decide the work cannot possibly be legitimate because of the tool used to create it.
A particularly funny example happened in May 2026. Conceptual artist SHL0MS posted a cropped image of a real Claude Monet Water Lilies painting on X, labeled it “Made with AI,” and asked people to explain, in detail, why it was inferior to a real Monet.
Many people obliged. They criticized the composition, reflections, depth, colors and even the lack of emotion. Not everyone was fooled, but plenty were. The problem was that the image was an actual Monet painted around 1915.
The painting did not change. The only thing that changed was the tool people thought made it.
AI's Real Problems Are Still Real
I hear a lot of criticism about the environmental and social impact of data centers.
AI data centers use enormous amounts of electricity, and some cooling systems consume significant amounts of water. Those are real costs. If a new data center strains a local electrical grid, drives up rates or competes with a community for water, those concerns deserve to be taken seriously.
The good news is that these problems are solvable.
Water use depends heavily on how a data center is cooled. Traditional evaporative cooling can continually consume water, but better cooling methods are already available. Microsoft, for example, has begun adopting closed-loop chip cooling systems that recirculate coolant instead of continuously evaporating water. That does not make data centers environmentally harmless, but it does show that today's water consumption is not some permanent requirement of AI. Better systems already exist, and I am sure engineers will continue finding ways to improve them.
Electricity is the bigger challenge. Lawrence Berkeley National Laboratory estimates that all U.S. data centers, not just AI, could account for about 11.8 percent of American electricity use by 2030, with its modeled scenarios ranging from 9.5 to 15.3 percent. That is a serious amount of power.
But our electrical grid has needed improvements for quite some time. Manufacturing, electric vehicles, new homes, and other growing electrical loads are already creating pressure for more generation and transmission. The Department of Energy now includes hyperscale data centers alongside manufacturing and transportation electrification when discussing the need for more transmission infrastructure. The massive demand created by data centers makes those improvements more urgent, but it can also create the economic pressure to finally build more generation, transmission lines and substations.
The important question is who pays for it.
If billions of dollars of infrastructure are built specifically to serve a data center, ordinary ratepayers should not be left holding the bill if that project disappears or fails to deliver what it promised. FERC, the Federal Energy Regulatory Commission, is already dealing with how the costs created by enormous new electrical loads should be assigned. The companies creating that demand should pay their fair share for the infrastructure required to support it.
AI uses water. AI uses a lot of electricity. Those are legitimate concerns, and pretending otherwise does nobody any favors.
Better cooling methods are already available. Meeting the electrical demand will require more generation and better transmission infrastructure, improvements our grid has needed for quite some time anyway. Contracts and regulation can make sure the companies creating these enormous new demands pay their fair share of the cost.
How the Tool Was Built and What the Human Created Are Separate Questions
Authors, artists, photographers, musicians and other copyright holders have legitimate questions about whether companies have the right to use their work to train commercial models and whether compensation or licensing is required. Courts, Congress, licensing markets, and copyright law are still sorting through those questions. The U.S. Copyright Office has treated AI training and the copyrightability of AI-assisted outputs as separate issues in its ongoing AI reports.
If people are legally owed compensation, compensate them. If licensing is required, license the work. If a particular training practice violates copyright, change the practice.
But how the tool was built and what a person creates with that tool are still separate questions.
The Copyright Office has made a determination when looking at authorship. Simply providing prompts generally is not enough by itself to establish copyright in an AI-generated output, but human-created selection, arrangement, and modification can be protected. Using AI as part of a larger human creative process does not automatically make the human work unavailable for copywrite.
That makes sense to me, and I can see the difference in my own work. The first book cover here is one I used to get The First True Song to press in time for a contest. The contest was also my girlfriend’s clever excuse to finally make me finish the damn book. I needed a cover quickly, so I generated an image I liked, added the title and author information, and used it. I made some choices for that initial cover, but the AI made most of the visual decisions.
The final cover was a very different process. I developed the characters individually, generated and rejected multiple versions, and brought the pieces I wanted into Photopea. From there I combined AI-generated images with free images from Pixabay, arranged the characters into the scene, changed the lighting, moved objects, adjusted colors and shadows, and fixed AI artifacts like the classic six-fingered hand and weird eyes. I even dropped a little Easter egg for observant readers. Then I added the typography and kept working until the cover looked like the one I had in my head.
The difference between those two covers is not simply that I spent more time on the second one. In both cases AI did a tremendous amount of the work I could not have done by hand. The difference is how many of the decisions that shaped the finished image came from me. On the first cover, I mostly prompted and chose a result. On the final cover, I decided what the characters looked like, what stayed, what changed, where things belonged, how the scene was lit and when the image was finally finished.
That is why “AI art” is such an incomplete description. It tells us what tool was involved, but almost nothing about the relationship between the person and the finished work.
You can find my book The First True Song on Amazon. Its the first in a series of funny, fun, found family, fantasy fiction I have worked on for over four years.
Here We Are Again
Nearly all of the new tools have stayed but the older crafts stayed too. They changed place, changed value or became specialties, while the number of people who could create what was in their heads got a little bigger.
Now AI is doing the same thing.
There will be slop. There will be problems. There will also be people making things they could never have made before.
I know that last part because I've have one of those creative ideas niggling my brain for more than thirty years, one depicting a card game between Satan and godly beings. It used to have Allah at the table, but at some point I swapped him out for the Flying Spaghetti Monster because there is no visual reference for Allah. I have asked several artists over the years to make it. Sometimes they did not want to touch imagery like this. Sometimes we got to a rough sketch I paid for and I did not like where it was going. Between sketches that went nowhere and one person in the Navy who disappeared before finishing the job (if you are reading this can I please have the rough sketch that I made back), I probably spent about $500 trying to get this thing out of my head.
It still is not exactly what I want. But between ChatGPT and my limited skills in Photopea, I think I am finally going to have it. I would love to have it as a print on the wall.
Let's Stop Gatekeeping
We have spent centuries asking whether a new tool disqualifies a person's creativity.
Maybe we should spend a little less time policing the tools and a little more time celebrating when someone creates something.
Please be civil in your comments.