I toured a house for sale today. A framed art print hung in the living room: a photograph of six horseshoes arranged in a grid. Climbing up the stairs, I was greeted by an abstract painting made of horizontal bands in soft pastel colors. Studying it closely, I imagined I was looking at a beach, but that may have been my imagination projecting its own meaning onto the canvas.
These pieces were placed there by the real estate company, chosen to evoke feelings of coziness while being politically inoffensive. They were purchased from a wholesale home decorator.
Their designs were created by human hands. I found myself wondering about the process of these artists. What techniques did they use? Did they have an artistic vision? Were they commissioned, or did they design the pieces independently and then license them? Did the artists see themselves reflected in these pieces or was it just a job? If there was an artistic intent, it’s now far enough removed from the artist that we can only speculate.
I suspect, although I can’t prove, that the main reason these pieces were made was to meet a demand created by offices and model homes. Generic, mass-produced home art is an entire industry. Its purpose is not to make any artistic statement, but to fill the silence of bare walls with the white noise of color. It’s filler.
“Filler” is a word with a negative connotation. In television, filler episodes pad out the season, meandering to avoid any plot or character developments that would disrupt the status quo. In cooking, it’s material to add bulk to food without adding any nutrients. It takes up space but has no real substance.
And yet, there is an entire industry devoted to the production of “filler” visual art. It is separate from the visual entertainment industry, although it employs the same artists, and draws upon similar talents.
This isn’t a put-down of the industry: I too, have bought art like this. At home, hanging above my office desk is a canvas print of a painting of a deer. I bought the print at IKEA, where it had been mass produced for the express purpose of filling out the walls of new homes. It has thousands of copies; I wouldn’t have been shocked to spy one of its many twins during the house tour.
And yet, I treasure my copy of the print. It reminds me of a friend, and a photograph he showed me once. That evocation of memory is the reason I bought it. The painting is meaningful to me, although it’s a meaning of my own creation. Like the abstract painting aforementioned, I projected my own interpretation onto the piece.
The “filler” industry is real, it’s alive, and it’s made a positive (if minor) impact in my life. And it will be the first to be severely disrupted by modern generative AI tooling.
So let’s talk about it. This is “The Filler Episode.”
Generative AI is, in a certain sense, the artistic equivalent of a “white noise machine.” Its results are incredibly lifelike, but without human guidance and interpretation, they produce little more than a perfectly generic simulacrum of human creation. They generate what is essentially “background” text and images, and invite us to project our own signal onto the noise.
(This is part of why it’s so dangerous to use AI outputs as an authoritative source of knowledge. They don’t just reflect the biases of the training data, they also reflect the biases of the user back at them.)
As I discussed in a previous post about generative AI, when a user projects their own value judgments onto the results and curates them, it becomes possible for that act of curation to produce an output that reflects the user’s artistic intent, and is in that sense, creative. But absent that, what they produce is filler. It takes up space on the canvas, it doesn’t look out of place, but it doesn’t say anything.
But I’m not denigrating filler text; I use it all the time! Oftentimes, when I’m writing a story, I have an outline of important scenes that need to happen and images that I want to describe. Then, I begin the task of writing the sinew that connects these scenes to a single narrative. That process of writing the connective tissue is the most laborious part of my writing process. It is also the least creative. Usually the process of writing this sinew allows me to identify opportunities for additional scenes, so both the story and my personal growth still benefit from writing it. But the writing of these scenes is still boring and mechanical. It’s filler.
It’s the exact kind of writing that generative AI has been shown to excel at. I’m tempted to see if I can get a language model to write those parts for me. Or at least have AI write the first draft which I can then go back and edit, replacing its generic prose with my own voice.
A better writer than me wouldn’t need filler. In fact, the better a writer you are, the less filler you need. A master writer doesn’t just make every word count, they make every word count multiple times. Every sentence is maximally efficient and is dripping with so much authorial intent that no one other than the author could possibly have written it.
For example, one of my favorite written sentences is the opening to Robert Charles Wilson’s Axis:
In the summer of his twelfth year—the summer the stars began to fall from the sky—the boy Isaac discovered that he could tell east from west with his eyes closed.
I love this sentence because of how much work it does. It establishes the main character and gives him not only a name and age, but also a disposition: he’s curious, observant, and experimental. It establishes the story’s place in the world’s chronology (“the summer the stars began to fall from the sky”), while simultaneously creating a sense of mystery and arousing the reader’s curiosity. And it documents the incident that kicks off the story’s plot: Isaac’s discovery of his unique ability, and the investigation of its underlying cause. It’s not a long sentence, but it pulls so much weight.
It’s a sentence that not only has a clear authorial throughline, but it’s a throughline that is consistent with the remainder of the work in which it appears. And there’s no filler: each word is in service of this intent. These are qualities that can’t be generated even by ChatGPT. (Or at least, making ChatGPT generate a sentence like this would require so much interactivity from the user, so much editing of prompts and curating of responses, that the output is more a reflection of the user than the model. In neither case did ChatGPT create such a densely meaningful sentence wholly from the training data.)
Now, this isn’t to say that a master of their craft will always produce art as “densely meaningful” as this. While an artist can create a work that is entirely devoid of “filler”, where each and every detail in the final work is chosen to have maximal creativity and authorial intent, there are multiple compelling reasons why they might choose not to. An artist working against a deadline or a budget may not consider it wise to put the maximum attention to detail into every aspect of the project. For projects beyond a certain scope, it’s simply not possible for a single creative vision to dictate every detail. In those cases, even a master artist might outsource some of the labor to an autonomous system that can follow their design goals and directions.
And this is where the current working relationships that power the art industry will be disrupted by generative AI tooling. Because highly skilled and organized artists working on large-scale projects already outsource their labor to “autonomous systems”. But until now, those systems were called “other people.”
The exact name for these people differs depending on the details. Some of them are apprentices helping to implement their teacher’s vision. Some of them are employed by studios, working to strict style guides. But they are united by their job description: to provide labor and artistic skill in exchange for pay and a potential path toward career advancement. It’s not a position that allows them to exercise creative control. It’s a job remarkably similar to the job of designing “filler” art for offices and model homes.
Consider Dale Chihuly. Chihuly is an artist known for his impressive, large-scale blown-glass sculptures, and he’s still active today. But Chihuly has not actually operated a glass blower since 1979, when he dislocated his shoulder. He now designs his sculptures and directs their construction, but the physical labor is outsourced to employees at his studio. He is still undeniably an artist: the fact that he does not physically realize his creations himself does not undermine his creative vision. This is true regardless of whether he blows the glass himself, hires other artists to do it, or relies on some yet-to-be-invented glass blowing automation tool. Such a tool would help Chihuly, but it would also create changes in the industry’s working relationships that would pose difficulties for the rising artists who would otherwise be employed by him.
This is the impact of every form of automation, and generative AI is no exception. It can substitute for human labor but not human creativity; but as it turns out, lots of artists are employed to provide the former but not the latter. I expect to see many of these jobs replaced with automation in the near future. Studios will still hire artist employees, but fewer of them, with the job of guiding generative tooling and overseeing / cleaning up the results. As I talked about last time, this job of overseeing an AI is still a job that requires artistic skill to do well. Artists aren’t going to be replaced by non-artists. But there will be fewer job positions for artists available. And worse, because these positions were used as traditional avenues of networking and advancement, those advancement opportunities will be diminished as well.
All hope is not lost: the eternally-online nature of our generation has created new, nontraditional means for artists to network, advertise, and be discovered. Artists today can find an audience on social media, and fund their craft through not just through direct sales, but through crowdfunding and taking commissions from fans. But traditional employment opportunities are a stable, reliable way to make an income from artistic skill while advancing your craft and career.Gaining visibility on online art platforms is risky and harrowing in comparison, to say the least.
And even though the highest quality AI generated art will be created by already skilled artists, the impact of the oncoming glut of low-effort, low-cost creations can’t be ignored. Because while these creations won’t compete on quality, they will compete on cost. And in any market where art is undervalued (which is to say, any market; just ask a freelance artist what their hourly rate is), artists will need to find a way to compete with free. Which can be done, and will be done… but not everyone will be able to do it, purely due to the fact that artists aren’t just competing with low-cost alternatives, but they’re also competing with each other.
So what are these artists to do? Even if the threat posed to their livelihood is no different than the threats posed by any other form of automation, even if it’s a threat created by the market and changes to its existing opportunities and relationships (as opposed to being caused directly by the technology)… it’s still a real threat. Imagine a truck driver being replaced by a self-driving truck, or an overworked vfx artist working on the latest Marvel movie, trying to not be outdone by AI-generated special effects. The solution can’t be to say, “We could replace your job with automation. We don’t have to make you suffer this labor of indignity, but we will, and you’ll thank us for it, because it keeps you fed.” That’s inhumane. It’s a broken window fallacy used to justify unnecessary labor. But the solution also can’t be to cut these people loose with no advancement opportunities. They’re in the business because they’re either very passionate or have no other options, and cutting those jobs won’t make them suddenly have other options.
The problem isn’t in the tools we use, it’s more systemic than that. And systemic problems require systemic solutions.
Man, it just keeps coming back to the fact that all art exists in a marketplace, doesn’t it? All of the threats posed by generative AI are really just the negative externalities we see anytime automation is added to a marketplace. Which isn’t to say that they aren’t serious, because they are. But once again, it seems that the best thing we can do is challenge the system that creates these incentives in the first place. Create a path for artists and other laborers to secure their livelihoods outside of the free market and maybe then it won’t matter whether we have generative AI tools or not.
Because until we do that, we’re never going to escape this.