Can AI Have Good Taste?
Of all the ways I expected artificial intelligence to impact my work, this was not one of them
Marcel Duchamp and I have a disagreement.
Duchamp was a revolutionary who expanded the boundaries of art by moving away from painting and sculpture toward the conceptual, to idea as art.
His “readymades” elevated everyday objects to the status of Art simply because the artist said so. There was a bicycle wheel and a hat rack and a urinal.
Duchamp’s intent, he said, was to “completely eliminate the existence of taste, bad or good or indifferent,” because taste is “the greatest enemy of art.”
I can’t quite wrap my head around this. It seems to me that all an artist has is their taste. A photographer friend of mine insists that clients only hire us for our taste.
So how, if Duchamp is to be believed, can it be a good idea to eliminate my taste from my art?
A cynic might argue that conceptual art is simply a trick to allow anyone to be an artist irrespective of their talent. If any idea can be art, anyone can be an artist.1
If anything can be art, then, does my taste have any value at all? Duchamp would argue no and say that’s as it should be. But I need my taste. It’s how I make my living.
If you make art, or commercial photographs, or oil paintings, or poems, your taste is what makes you you. Isn’t our taste what distinguishes us from everyone else? I think it is. Which is what Duchamp didn’t like about it.2
So imagine my chagrin when earlier today my taste was challenged by a machine.
After delivering proofs from an executive’s portrait session, I received a surprising reply. Such messages typically include words to the effect of, “I like number 7. Can you fix my hair?” Often they ask for photos I wouldn’t choose, but as the saying goes, the customer is always right in matters of taste. So when my clients choose a picture or send a brief that isn’t my favorite, I shrug and think of another saying: “It’s your nickel.”3
But today’s email said something I’d never seen before, something I’d never even considered.
“Dear Bill, Thank you for getting these to me so quickly. I’ve selected four photos. To assist in finalizing these selections, I utilized an AI photo analyzer to evaluate each image and provide editing recommendations. I’ve attached the corresponding reports for your reference.”
Gobsmacked, I was.
The reports contained platitudes about composition, brightness, depth of field, and a belief that an upturned mouth and squinted eyes combine to form the perfect expression.
I sent it to a colleague. He texted back immediately.
“An actual client sent you this?” he wrote. “This is just the same old garden variety tackiness and lack of good taste, repackaged for the AI hype cycle. It just shows a real lack of both tact and taste.”
There it is, our word of the day. I texted another friend.
“It’s like asking AI what I should have for dinner,” he said.
I composed myself, comforted in my shared befuddlement. I had a job to do. So I also began composing a reply. I aimed for pleasant helpfulness.
“Thanks so much for sending these,” I wrote. “But I want to make sure they’re your selections, not something a computer chose. I don’t believe there is a formula for determining which image is ‘best’ or which one you should prefer. What the machine values and what you value are likely quite different, and I want to make sure you’re happy.”
He said he understood, and narrowed down to his favorite. Suspiciously, it was the same image the machine ranked “best.” A colleague to whom I showed the options chose a different one—the clear winner, and the same one I had selected. It was of course lowest ranked by the machine. Because it has bad taste. Or, perhaps more accurately, it has no taste at all.
Along with his favorite, my client included retouching instructions—more platitudes pasted from the app.
“The photo isn't perfectly symmetrical, which is good, but it needs some cropping using a 3x3 grid. Neither of the eyes is on a dividing line.”
There was more, but it was all like this. Nothing about reducing shine or smoothing blemishes, much less choosing an expression that communicates relaxed confidence or a sparkle in the eyes.
I was offended that this person thought a free AI app knew more about portraiture than I.4 But I know he didn’t mean to offend. He is a lovely guy but, obviously, a bit of a left-brain thinker, certain that for any given challenge there is an equation to determine the “correct” answer. Photography, as with any artistic endeavor, cannot be itemized in any meaningful way to arrive at an answer to the question, “What is best?”
Because, as you know, there is no accounting for taste.
My taste is better than that machine’s. In fact, my client’s taste would be better than the machine’s, contrary to the evidence. At least it would be an image he responds to positively even if he doesn’t know why.
The world is worried about AI taking over the fun jobs, the creative things we humans love to do. As the meme says, “I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do laundry and dishes.”
But has anybody considered this insidious application of AI? Subjugating our taste and opinions to the all but arbitrary rulings of a machine in a windowless warehouse.
Can you imagine the day when you report to work to discover your boss has been replaced by the Optimizer 9000, an AI-powered desk lamp that delivers a mild electric shock at random intervals to keep you on task? Maybe I shouldn’t write that. I don’t want to give them any ideas.5
What if it’s not that the AI takes over the making of art, and simply becomes the modern arbiter of taste? Much like the dead internet theory, this would commence the march toward dead culture.
To test the AI’s taste, I decided to run some images through the same image analyzer. But instead of my own work, I wanted to see how a few all-time great photographers might fare.
I started with one of the most iconic portraits ever, Dorothea Lange’s Migrant Mother. It received a score of 40/100. “It’s not awful,” the app responded, “but there are plenty of actionable tips below.”
I cropped it slightly and re-uploaded it. 52/100. Progress! Though the app suggested the Migrant Mother should smile more (you can’t make this up) and that “a smile can do wonders for people’s perception of your photo.”
At this point I realized the app was optimizing for LinkedIn. So I chose a classic headshot: Yousuf Karsh’s Ernest Hemingway.
46/100.
“Oh dear,” the app wrote. “This is going to need some work. We’re finding ourselves a tiny bit uncomfortable being this close to your face. Zoom out, show us those shoulders!”
Just like Robert Capa famously said, if your pictures aren’t good enough, you’re not zoomed out enough so we can see your shoulders.
I decided to try a more modern shot, something undeniably striking and beautiful. How about Annie Leibovitz’s headline-grabbing portrait of a pregnant Demi Moore.
58/100. Not bad! But, clearly, not good enough. My own executive’s headshot had scored in the 70s. Take that, Annie.
“You’re doing pretty well on brightness,” it said. “You could increase brightness just a tad for a better result.”
What would the AI think of something outside the box, like an iconic Andy Warhol Marilyn Monroe portrait?
71/100.
“You have a passable smile,” the machine told one of the most famously beautiful women of the 20th century. “Try showing some more teeth. Raise the corners of your mouth more.”
Okay this is silly. If all of these are so bad, what does the machine think is good?
I input a test shot of a mannequin head and worked on the brightness and cropping to bring the score up as much as possible. The straight shot garnered 65/100.
Not enough smile, not enough teeth, okay on brightness. So I used Photoshop to dial up the brightness and liquify the face into a more “pleasing” expression. Got the score up three points. But it wanted teeth, so I cloned some from another portrait, pushed the smile and the brightness even further, added warmth and tilted the frame a bit. Iteration after iteration, incremental improvement after incremental improvement, until the scores began to decline.
I had done it. I had reached the ideal image, the apex of photographic portraiture technique—at least as far as the machines are concerned. This modified mannequin image scored even higher—notably so—than all of the aforementioned 20th century masters (and me). With a score of 78/100, the maximum score I was able to extract from the AI image evaluator, I present to you the ideal LinkedIn portrait as selected by a box full of wires.
10/10. No notes.
If Duchamp thinks we should eliminate taste from the equation, perhaps he would have been an outspoken proponent of AI art. It is demonstrably devoid of taste, though of course in skilled hands it can create beautiful, tasteful images—at least in theory. But would it know those images were beautiful and tasteful? I don’t think so. Because I don’t think AI can have taste. It can only know what we tell it in hopes of approximating the ones and zeros of human thought. But there’s no binary good or bad w/r/t taste.
And isn’t that all there is, anyhow? All we have is our taste. And there’s no way to prove, once and for all, whose is best. We simply own our own and hope it jives with a few people who want to purchase our poems or buy our buildings or procure our prints. There is no accounting for taste. And it turns out that’s a good thing. Because it means the machines can’t ever replace us.
I tend to love idea as art. Some of my favorite experiences have been provided by conceptual art. Michael C. McMillen’s Central Meridian, for instance, or Janet Cardiff’s Taking Pictures. But can the same be said for Maurizio Cattelan’s Comedian (better known as the banana taped to the wall) or, frankly, anything ever done by the impervious Richard Prince?
Pablo Picasso said something similarly dismissive, but his was more an attack on the prevailing thought of the time and how it hindered appreciation for unconventional work such as his own. That’s hardly the same as wanting to eliminate his taste from the work. In fact, I’d argue it’s evidence of the importance of an artist’s individual taste. Without it, Picasso wouldn’t be Picasso. Even Duchamp wouldn’t be known as the father of conceptual art. Taste is all we’ve got.
Of course I offer good counsel, politely suggesting that perhaps they consider this image, or this approach, or this different way of tackling the problem at hand. But in the end, if they’re footing the bill, I believe they get to tell me what they want. If you want me to overcook your steak and smother it in ketchup, who am I to judge?
A real live expert, it could be claimed. I have degrees and experience measured in decades and a history of teaching the subject. But yes, please, what does the app say?
I, for one, welcome our new AI desk lamp overlords.
Brilliant piece, thank you. I am laughing at this. Gah – the AI verdict of those world famous portraits! But seriously, the implications are grim.
I loved this. I appreciate how you responded with curiosity in creating these experiments. Some great lessons here, plus it was really funny. Thank you!