Matt Gemmell, esteemed independent fiction author and former technologist and pundit, offers a seemingly hardline stance against using new tools that never existed before for serious creative work. He makes the point that many of us already know that we are wary of claiming we “painted” or “wrote” something when we sort of only cobbled it, or built it, or stumbled onto it, or tweaked our way towards it.
- Authorship by Matt Gemmell
He has my number. When I produce artwork using these novel tools, I call it modding or salvaging and I call myself an “ainter” as in an “AI painter” and “I ain’t a painter.” I produce images or aintings as a hobby because it is enjoyable, and I don’t give a damn if I don’t have other serious grown-up approval. But I am on the same side as Matt on this probably because I never grew up in a world where these tools just simply always existed, and for obvious reasons I will never pretend I can produce an entire 9k digital painting—at the current level of quality and rapidity that I produce them—without the aid of AI tools. No one who knows me would believe me anyway, without evidence that I could do it laboriously by traditional means.
As an aside, I have actually developed my eye and my painting ability significantly while working with these tools, because sometimes traditional means of digital painting are just much more rapid ways of communicating what I want to the machine. And the machine screws things up a lot. A lot. Also, I am certainly not hiring models and working from life sketches or photographs to produce these images. But that doesn’t mean I cannot or have not done this. I think I have significant skills in this area (based on being at or near the front of the class in school in a life drawing class) and the assumption that someone who uses an AI art tool probably cannot draw is folly. (Mr. Gemmell does not claim this, but it seems sort of heavily implied by calling use of the tools “automated plagiarism.” It is nothing if not meant to belittle and warn away from developing skills at using these powerful new tools.)
I Know I Didn’t Produce the Entire Image Myself
But so what? Many artists cannot draw something from their mind’s eye with no reference, or at least we know that the results using reference are ten times better. Some artists can, but most cobble together reference into a new image, as a matter of course. It’s how they work. They may do a web Image Search and carefully compose and remix, and it is not currently considered plagiarism, especially when the artist puts together enough pieces to create something “new.” But what are they doing if they cheat and draw pictures of the sea stacks of Iceland if they haven’t been to Iceland themselves? Is it really plagiarism to avail themselves of such a powerful reference tool as web Image Search? And why aren’t the hardliners speaking out against this borderline plagiarism? Is it because we have had this one particular tool for a few years and it is boring to talk about? Or is it because there is room for nuance in discussions about creativity, because web Image Search can still be used both for plagiarism, and for novel creative work?
So now we live in an era when a machine can do the remixing very quickly for us, using billions of reference images, and can produce handfuls of infinitely varying, synthesized reference images for us, and we see one and say “that’s what I was looking for!" Now suddenly, when our eyes are filled with an image that no human has ever seen before, it is suddenly plagiarism? I think that makes no sense.
Every book is a quotation, and every house is a quotation out of all forests and mines and stone quarries, and every man is a quotation from all his ancestors. — Ralph Waldo Emerson.
Somehow human beings know there are billions of us on this earth, yet we still think we are special and that each creative work is some new new thing, when deep down every human knows they are not an island, and every thriller writer is not a plagiarist just because they didn’t invent an entire new genre, and their own work must be cobbled together from many other things, even if they can tell you their influences. Somehow humans keep repeatedly forgetting that we have known for thousands of years that “there is nothing new under the sun,” (Ecclesiastes) that “everything is a remix.” (Kirby Ferguson)
Gemmell’s article uses this hesitancy to underline my own wariness at any claims to the entire final work. I have been stewing over this, wondering sometimes about creating a lot of really interesting work that I have miraculously made seemingly much too easily, but actually still even laboriously (why am I pouring hours into this again?), but I don’t think it is entirely an accurate approach to the question.
No Old Answers to New Questions
AI art tools are genuinely new. And many people (students) will likely try to turn in work that was knocked off with no effort. The teacher is assigning a pointless essay to be written at home instead of monitoring students in a proctored test environment. Plagiarism is and already was a problem to be dealt with. Cheating and impersonation has been here longer than AI. And AI will not help make this better. But teachers could invent new ways to get students to want to write, by picking topics that matter more to students. (Or perhaps there will be ways to use AI tools to generate unique tests for each student, so that cheating might become impossible, with the difficulty level precisely calibrated by AI. Thus AI tools may be the solution to the AI tool problem.)
We know all this. But the handwringing and doom and gloom about new and exciting and scary tools overshadow any possible benefit, to some. Honestly, I am shocked that people are not more excited about the possibilities. I am shocked that otherwise technically capable people who love technology and consider themselves technophiles haven’t seen the creative ways the tools remix endless pieces of ideas in genuinely astonishing and creative and beautiful ways, instead relegating it to some parlor trick, some passing fad.
I would not have pegged Mr. Gemmell as a Luddite and I don’t think he is even entirely wrong. But I think many people are taking sides on the uses of these new AI art and AI writing tools without having seriously attempted to use them. Anyone who actually tries will quickly learn that the results tend to be unreliable at best. People who have spent months (probably no one has spent more than about 18 months using these tools seriously as of April 2023) getting good at this stuff may be good at making it look easy. But reliably finding and improving great images is not as automatic as many non-practitioners might think. My process to creating high-quality high-resolution poster-sized images is involved and convoluted, refined from months of experimentation. And the novel styles created by some AI art practitioners who plumb the tools’ depths lack any reference to any living human artist. These people are homesteading new frontiers.
I think the framing that most people who use these tools casually can only create new things by “automated plagiarism” is not quite the right framing, but I would not hesitate to call it “automated fan art” or “automated remixing.” I think remixing enough influences (not just aping one angle), and finding and refining genuine gems that are coined from random input noise is not actually as trivial as plagiarism. In other words, what makes the work into art is not just the effort, and it’s not just the result, and it’s not just luck. It is some combination of the three. No matter what the tools. But when a skilled practitioner is able to go beyond luck and reliably produce high quality results with significant effort, who are we to say it cannot be called art? Why write yet another damn fiction novel in a world with 129 million ISBNs in 50 years? Because writers can’t not write.
Indeed, the drive to make art is nearly universal. Better tools have always made this easier. I’m sure someone has told students, “it’s not real oil painting if you don’t crush your own pigments.” But many oil painters would call this pointless snobbery. It’s only slightly different from any other kind of gatekeeping of creative work based on just the tools. Instead I would not advocate gatekeeping someone else’s art. Were I to allow it, my gatekeeping would be based on (a) results, (b) the urge to create, (c) the spark and joy of discovery, the muse, even the Fortunes and the Fates, the random noise, the contingency and unpredictability of it all, and (d) the effort, the polishing, the tweaking, the curating, the iteration, the shoulder to the grindstone. How is any such process not just good old-fashioned capital-A Art?
Art has always been about the grind, and then stepping back and sharing just the best. It’s about vision and purpose, about having something to say. It’s editing, it’s inspiration, it’s iteration. And then another round of selection. I think the curating of one’s own work by a prolific photographer (including polishing and tweaking what came out of that automated picture-making machine) is actually not that different to curating a few gems from hundreds of fantastic AI creation finds. I think the creative impulse and the keeping-at-it process using new AI tools is not as different to what a traditional artist could be spending their time doing. Those who proclaim otherwise seem to me to project their ignorance, fear and snobbery.
Working Through a Slow Human Team vs. Working Through a High-Speed Robot Team
Finally I also think that a simple rewording of his argument shows how it can fall apart. We will take the example of the late work of architect Frank Lloyd Wright. He worked through a team of students and understudies and is still credited with his later buildings (students may have received some credit; they executed all the drawings). But according to Mr. Gemmell’s narrow definition of authorship, and taking the structure of his argument:
When people invoke [a team of underlings] to generate something, they often still use the language of endeavour: here’s what I created, or made, or built with this [team]. Those words reveal the truth, as words invariably do.
To [direct a team] to generate something for you is not an act of creativity or engineering, because such acts are in the execution, not the idea. On the contrary, it’s automated plagiarism. It doesn’t matter that the originator is a piece of software instead of a person, [or a team of perople]; what matters is that it wasn’t you.
When the end result is built from the works of others, and when the building is also done by an agent other than oneself, there is no legitimate claim of authorship. This is elementary.
This is bullshit. Directors of films and art directors of large teams share credit, but they are not committing plagiarism. I know I am in danger of putting words in Mr. Gemmell’s mouth, but I think the strucutre of his argument is flawed and it is clear that Frank Lloyd Wright working through his students can still be considered an author, an auteur and not a plagiarist.
(“You can’t compare yourself to Frank Lloyd Wright.”)
You don’t have to consider an AI art tool to be an art department, but I don’t think there is anything structurally different about it besides that it’s made of metal and copper and silicon instead of meat and bones and neurons. I think the main arguments against using these novel tools is that they are too new, too scary, too fast, too cheap, and too good at remixing too much reference. They make creativity too easy, so they must be bad.
I disagree. I think poorly written essays or factual inaccuracies parroted by AI chatbots, or poorly realized AI paintings should of course be considered a nuisance, or simply poor art. But I’m not naive enough to think awesome words or awesome images cannot be considered high quality art simply because it was too easy to produce them, or because I used a GPU instead of a room full of underpaid, overworked concept artists working in the game industry in Los Angeles.
They Are Coming for Me Too
I wonder if those who complain the loudest are simply the most surprised that their area of work is the one most recently automated. I’m a programmer, and AI tools are already coming for me. It’s honestly not something I saw coming two years ago, in 2021. I thought I had a decade or more. But I say, if a younger worker can use the tools to accomplish something at a higher speed and higher quality than I can, then I why do I get to name call to preserve my paycheck? Who cares what I think? Maybe I deserve to be replaced.