Just Because You Can Doesn’t Mean You Should
On AI and the Creative Process
Generally speaking, I am opposed to the use of AI-generated imagery.1 I find it ethically dubious at best—and how often is it at its best?—lazy, and, most of all, boring.2
That being said, one of the use cases for AI generation I am most intrigued by and sympathetic to is its use as an ideation and “sketching” tool for visual creative projects. I see the potential there to augment the human creative process. But I am also beginning to see problems with this approach.
I don’t know how to write this without sounding pretentious or like a gatekeeper, but here goes nothing: I believe that when a non-creative professional—lets call them “client” for the sake of simplicity—generates a visual idea with artificial intelligence, an integral part of the creative process is often skipped: the interrogation of the idea.
Before I continue, let me lay some personal philosophical groundwork:
I think everyone can be creative. Creativity is a way of thinking that can be trained. It is not innate, nor is it exclusive to a select lucky few.3 How do I know this? I was not “born creative,” nor am I a natural artist. Everything you see from me has been learned. Creativity is also separate, but related to, the physical skills associated with creativity—that can also be trained, by the way—like drawing, painting, or a particular software.
Through this training, creative professionals and designers have not only developed tangible skills, but, ideally, a level of discernment about and detachment from ideas. To be a successful creative professional making work on behalf of clients for any length of time, you cannot become too attached to any singular idea.4 This discernment comes from making lots and lots of stuff, and, importantly, critique.
I do not believe that AI-image generation “democratizes creativity.” It is a shortcut to results without process, and that process is where creativity lies. While there are certainly exceptions, AI-image generation is inherently less thoughtful or discerning—particularly in the case of being used by the untrained client.5 Everyone can be creative, but using AI disincentivizes creative thought.
When a “client” uses an AI tool to generate a visual idea they then bring to a designer, there is a good chance that this discernment about creative ideas is missing. The tires, so to speak, need to be kicked before going on a drive. Will this concept work across every potential deliverable? Does it unwittingly reference some aspect of art and design history that undermines its effectiveness? Is it trendy, cliché, or obvious? Does this align with accepted principles of design, and if it doesn’t, is it for a conceptual reason that works? Are there important details overlooked? Does it simply make sense?
I am reminded of the line from Jurassic Park: “Your scientists were so preoccupied with whether or not they could, they didn’t stop to think if they should.”
Because the idea is suddenly manifested with a few strokes of the keyboard, it is easy to fall in love with it—or at least get used to it—like a film director becoming too accustomed to the temp track scoring an early cut of their movie. The idea is real and it is here and the client, who has not been trained to accept or expect critique, knows it like a friend before the designer even enters the picture. They become attached because they have not learned to iterate, test, and not fall in love with the first idea they have. And of course they do! They are human, and they have brought an idea forth into the world, and that is intoxicating.
This can and does happen to designers, too. If an idea is executed before it has been fleshed out and iterated upon, it is far too easy to become attached or falter to come up with new, better ideas. Hopefully, through experience, the designer knows better.
Never fall in love with an idea. They’re whores: If the one you’re with isn’t doing the job, there’s always, always, always another.
—Chip Kidd, The Cheese Monkeys
Because the client has become attached to their idea—so easily brought forth by artificial intelligence—and has hired a designer to realize their vision in a more polished manner, the designer is cornered by this infatuation with an idea that has not been interrogated and they have, in effect, become nothing but a hammer for the client’s nail. If the client is inflexible to the designer’s ideas, the final work will inevitably be mediocre and, in my opinion, the client has wasted their money and the designer’s time.
To be fair, this doesn’t just happen with AI. It also happens when a client makes something in Canva or with rudimentary Photoshop skills. But with AI-generation, it has become easier and faster than ever to bypass early, integral stages of thoughtful creation. It has become easier than ever to rocket toward mediocre creative work. This isn’t really an AI problem. It’s a human problem, now accelerated by AI.
I do not mean to denigrate “clients,” or people who do not work in creative fields, and I do not mean to imply that designers are the only ones capable of having good, thoughtful, creative ideas. “Client” and “professional” are not monoliths. I have many doubts about my ideas; I have had many stinkers; I have benefitted immensely from incorporating ideas from people not paid to create things every day. In fact, I’d call the latter integral to my own creative process.
But I am wary. I am wary of the would-be client that does not see a professional as a collaborator, but a tool or a means to an end. I am wary of the person who uses AI to jumpstart a visual project and does not accept critique or new ideas because of their attachment to what the machine has regurgitated at their command.
I’m hesitant to share this. Like I said, I don’t know how to write any of this without sounding pretentious or like a gatekeeper.
Reading it back, this might even offend folks and turn them off from even working with a designer. It’s just going to preach to the choir and won’t change any hearts or minds. I’m standing still as the wave of AI slop comes crashing down on my head with excited cries of “Luddite!” as it tries to drown me.
If you have made it this far, I want to reiterate something to those that believe that through AI they now have access to previously unattainable creativity:
What you are missing is not some innate creative aptitude that only some are blessed with. What you are missing is the time and work and training with which to harness the creativity available to all of us, and, critically, the discernment required to know what creative idea to use and how it will work.
Maybe AI tools are the future. Just don’t let them think for you.
Imagery. Not art.
ChatGPT has supercharged online visual trends in which everyone just copies each other. See: The Studio Ghibli trend, the action figure trend, and whatever crap comes next. I’m even bored by the artists turning themselves into action figures using “real” art.
Copying a trend propagated by AI imagery is sad and ironic. This technology will accelerate the trend cycle like never before, leaving visual artists scrambling to keep up.
It’s the Force, Luke.
Though ironically, to do any interesting work, you must be attached to ideas enough to a) generate lots of them and b) fight for their existence, while also being detached enough to survive the inevitable rejection of these ideas.
I’d be willing to hear debate about this style of AI usage from someone who has this creative training and discernment regarding visual ideas.
It’s not pretentious to call out the skills required to do effective design work. I think some of the possible misconceptions people can have about the utility of AI is equating design with tasks like making a mood board. Pulling together references is a fraction of the problem sets we have as designers. But it’s possible that many people think that a quick picture saves hours of work.
Most of my design sketching takes seconds or maybe minutes to produce but the thought processes and conversations that come out of those sketches are so much more meaningful.
Hi. "I’d be willing to hear debate about this style of AI usage from someone who has this creative training and discernment regarding visual ideas." Self taught over the course of 30 some odd years, ended up being an author/editor because it got the ideas out faster, but, it me.
Every single time I have used AI to generate an image it has gone through the mental or paper thumbnail process first, and beyond that, sometimes it's impossible to (historically, I was aiming to be an illustrator, I have bad eyesight though, so I've always been really good at using negative space and edge detection, because until I got glasses, way later in life than I should have, it was how I navigated a very, very blurry world) get within a mile of the idea or concept that you want, if only because there are inherent limits on AI image generation at the level of prompt adherence, even for things that if I were to grab my sketchbook, I'm gonna have no problem with. I'd just as soon use regular old reference photos or fall back on my body of knowledge. Like, lately I've been doing simple collage work in GIMP to go with one particular story cycle. There is no way on god's green earth that my visual ideas for how a 1024x5k long pixel collage is going to read visually if given over to the concept of AI image generation. It is currently impossible unless you take multiple images out of whatever you're using to generate them, and cut them up by hand, because the collage has an emergent theme and elemental repetition that I've baked into it from the conceptual start. And it may not be my best work, but it was 2 in the morning and somehow I'm STILL dodging writing the piece I wanted to finish last night.
The issue I'd say with using AI to generate ideas as sketch etc. is, there will be a flattening effect as the model regresses to the mean of its own guardrails (this can be used to create some neat stuff, but they usually shut it down or change the rules before you can really get to the point of making things that are truly interesting.)
That being said, they're already incorporating it into corporate design workflows and teaching it at design schools. I assume the use of it, and its use cases, will evolve from our current point (early adoption and panic.)
I also dislike discounting AI generated imagery as a WHOLE, because IT IS AN ABLEIST POSITION. If I have a severe tremor, or other disability, I can have a full training in design, graphic art, I can master ALL the theory in the world, but there's still a barrier to entry. (I have a very slight tremor most of the time. Unless I try to overdose on Xanax (literally nearly impossible by the way, unless you add in a secondary cns depressant) So, if I'm on my iPad in Procreate, yeah, I'm using the line smoothing function turned up to about 70% on my ink and line work. Now, this is a very limited scope single asset use case, but keep in mind, THAT IS ALSO A FORM OF AI.
As an artist and fan of certain forms of gatekeeping (I am an editor.... of fiction....) I sparkle with a LOT of what you say, but I also see these sort of gaping blind spots that professionals tend to overlook because, well, they're professionals, with professional mindsets and a professional's toolbox to grab from, essentially critiqueing the tech from a professional standpoint (like if I was to start railing about Sudowrite right now, which I have been known to do under certain circumstances and when the air pressure is just right.)