What's Wrong With AI Art?
or: What we should remember about the Luddites
I realize that I’m late to the party here. Generative AI tools like ChatGPT and Midjourney have been out for a while now, and algorithmically generated artworks existed long before them. Others have already weighed in with with their predictions on whether they’ll save or destroy the world. I’ve heard some convincing arguments from both camps, but none of them really articulate the way I felt when I first saw AI-generated art scroll across my feed - or the way I still feel about it.
I don’t know if I can express that feeling well, or whether I’ll get to a clear, unambiguous position by the end of this post. Writing this, here, is as much an attempt to get my own thoughts in order as it is anything else. To that end, I’ll list out some of the major arguments against AI-generated art and work through them one at a time.
It Isn’t Good
Most art isn’t good. That’s never been an excuse not to make it.
It Isn’t Art
This is the most common argument I hear, and it’s also the most problematic. For starters, it relies on a clear definition of what art is.
Personally, I like Stephen King’s argument that art is a form a telepathy (he’s talking about writing specifically, but I’d argue that it applies to all art). The artist starts with an image, a story, a feeling, an experience, a perspective. Once they have it, they attempt to encapsulate it in their chosen medium. The spectator - usually a complete stranger, separated from the artist by time, distance, even a common language - experiences the art. In experiencing the artist’s work, the spectator closes the circuit, and the artist’s original idea appears in the spectator’s mind as if by magic.
This happened to me once in the National Portrait Gallery in Washington, D.C. In 2011, they opened an exhibit called Asian American Portraits of Encounter. I’d just spent a year living in South Korea, and was interested to see CYJO’s KYOPO project, a series of portraits where the subjects were asked to describe their experience of being ethnically Korean in a non-Korean culture.
But it was this video by Hye Yeon Nam that stopped me cold:
In the video, the artist spends four minutes eating a dozen cherry tomatoes off a plate using a ruler as a utensil. Each tomato takes about fifteen seconds of effort. At several points, she stops to shake the cramps out of her hand or massage her sore shoulder. It’s absolutely frustrating to watch, but I stood there and devoured every second of it while other people passed through the room.
Why?
Because I’d just spent a year living in a foreign culture. I knew exactly how she felt, the way a simple everyday task like eating could become awkward and uncomfortable for someone far from home. I remembered the novelty and the disbelief of ordering spaghetti at a restaurant and being handed a pair of smooth metal chopsticks. The bewilderment of standing in an aisle at a grocery store where you didn’t recognize a single product. The months of practice before I could order food over the phone. The artist had managed to convey that exact feeling in the most unlikely of media.
As far as I can tell, nothing about AI-generated art disqualifies it from this feat of telepathy. If your carefully worded prompt manages to connect your brain to mine, it counts, full stop.
In fact, I can easily think of an example where this works really well: the procedurally generated worlds in No Man’s Sky.
Given the impossible scale of the universe you’re exploring (there are, literally, 18,000,000,000,000,000,000 planets in the game), no team of human artists could ever handle the task of populating those worlds. The use of procedurally generated artistic assets is a prerequisite for telling the kind of story that the designers envisioned: a lonely journey through a vast, unknowable universe.
It’s Theft
This is still being decided in court. I’m not a legal expert, but to me, the most damning piece of evidence is the wording around fair use. Specifically, the “effect of the use upon the potential market for or value of the copyrighted work.”
The UK has similar wording in their fair dealing clause:
“If a use of a work acts as a substitute for it, causing the owner to lose revenue, then it is not likely to be fair.”
In August of 2022 - the month after Midjourney launched, and the same month Stable Diffusion went into open beta - Getty Image’s stock price peaked at $30.86 per share. On September 30th, it was $6.79 per share. That’s a clear loss of value, and it’s directly related to the way AI-generated art “acts as a substitute for it, causing the owner to lose revenue.” As it stands, Getty Images has a fair shot in court, but the artists who created the works will probably never see a dime.
Even if the complaints fail to meet the legal standards for copyright infringement, I’d argue that they meet the ethical standard of exploitation, which is a neat segue into the heart of the matter.
It Hurts Artists
One of the things I find interesting in this debate is the rampant misuse of the word Luddite.
People tend to talk about Luddites as stupid, stubborn anti-technologists, but this is a blatant mischaracterization. To correct that narrative, we need to understand the environment in which the movement was founded: early nineteenth century England.
Some highlights:
The outset of the Revolutionary War in 1775 meant that England could no longer transport its prisoners to America. To offset this, Parliament passed the Hulks Act in 1776, which allowed private contractors to run “convict hulks,” permanently moored ships where prisoners were kept in horrific conditions (on the Justitia alone, 25% of its inmates died within the first 19 months).
In 1787, England began sending its prisoners to Australia as a way to continue what they believed was a necessary evil. As Robert Hughes writes in The Fatal Shore, “the failure of language—the tyranny of moral generalization over social inspection—fed the ruling class’s belief that it was endangered from below.” Convict hulks and transportation continued until the Hulks Act was allowed to expire in 1857. The Luddites came onto the scene around 1811, about halfway between these two important dates (1775 and 1857).
Around the same time, English courts were sentencing roughly 10,000 people to debtors prisons every year (Charles Dickens’ own father was sentenced in 1824). Prisoners could generally expect a life of torture, malnourishment, disease, and hard labor for debts that were - in many cases - impossible to pay off.
Labor disputes in this era (like Luddism) existed against the important backdrop of mass incarceration, deep class divides, and the existence of unskilled labor as a primary form of punishment.
In 1803, England ended the Treaty of Amiens and declared war on Napoleon’s military dictatorship. The Napoleonic Wars would last until 1815 - another set of bookends for Luddism. During this time, industrial productivity was so important to the ruling class that it bordered on existential. Working conditions worsened, and threats to productivity were dealt with extremely harshly (death, or the aforementioned imprisonment and transportation).
Finally, on March 11, 1811, framework knitters in Nottingham gathered to protest an employer who had hired too many apprentices, which was illegal. When British troops dispersed them, they responded by smashing their own knitting frames. This began the violent period of the Luddism movement, which employed “collective bargaining by riot,” a term coined by historian Eric Hobsbawm. Their grievances, by and large, were the same grievances that laborers raise everywhere: falling wages, unemployment, worsening working conditions, and the inability to effect real political change.
As Adrian Randall writes in the foreword of Writings of the Luddites, the majority of them were not anti-technology, but skilled technologists who resisted “the notion that market forces rather than moral values should shape the fate of labor.”
They failed. Twenty years later, England added a new word to its labor lexicon: the sweatshop.
Fast forward to the present day. Many artists find work under work-for-hire contracts or as independent contractors. This exempts their employers from paying federally mandated benefits, and makes it illegal for them to unionize. This is an important distinction. If employers collude to set the price they pay to artists, that’s perfectly legal. If independent contractors (who aren’t covered by the National Labor Relations Act) agree to a minimum price for their labor, that’s price fixing - an imprisonable offense.
For those that sidestep the work-for-hire market and enter into licensing agreements, this puts them in a situation where they can’t sell their own product to an end consumer. As an example, a writer who follows the traditional publishing route will sign with one of the big five publishing houses. In this model, the reader pays the retailer, the retailer pays the publisher, and the publisher pays the agency who represents the writer. The writer gets what’s left.
If they don’t like that model, they have the option to buy into Amazon’s KDP monopoly - with the understanding that Amazon can change the royalty structure at any time for any reason (and probably will, if they ever finish crowding traditional publishers out of the market).
The same goes for musicians. Spotify doesn’t pay musicians directly, as explained in this video. It pays the rights holders, which typically means one the three record labels who control 70~80% of the recorded music industry. The labels divvy up the proceeds based on contract agreements that they write.
Set this against the backdrop of American politics, with its deep political divides, legislative deadlocks, stagnant wage growth, and pay-to-participate legal system. Luddism is an apt metaphor, but not for the reasons you think. Like the Luddites, most artists are skilled technologists who are already struggling against low wages, poor job security, flimsy legal protections, and an inability to lobby for meaningful improvements to their working lives.
How does generative AI improve that situation? Because I don’t see it.
In making it easier to mass produce art, existing artists have less negotiating power in a game that’s already rigged against them. The artists who reject the productivity-enhancing features of AI risk being outpaced by those who embrace it. The artists who do choose to make AI-generated art wade into murky legal territory where they might not be able to copyright their own works.
And if an artist manages to thread that needle and become wondrously productive through the use of generative AI, who cares? In a world where creative labor is routinely exploited and the benefits of added productivity always, always, always get diverted away from the worker, does anyone actually believe that artists will be the sole exception to the rule?
It Doesn’t Live Up To Its Promise
This, ultimately, is what annoys me most about AI-generated art.
With the release of generative AI, we were promised the “democratization of art,” which is a problematic term for two reasons.
First, art was already democratized. It’s never been inaccessible to anyone, and never required a minimum level of technical skill or mythical inborn talent. Of the millions of people making art right now, virtually all of us make it because we’re compelled to do so. The years of time and effort we’ve invested were never a barrier to entry. They were only-ever-always the natural result of practicing a craft that we considered worth our time and effort.
The only real barriers to entry are the significant legal and market forces that make it difficult to make a living as an artist - forces that generative AI does nothing to address.
Second, if AI is supposed to democratize art, why doesn’t it address any of the core issues facing its constituents?
Perhaps more importantly, what kind of democracy operates without the consent of the governed?