UX Research
Art is an instinct
Why does art hold people who can’t make it? Intent from the caves to AI, and what designers can do about slop.
Company
Self-initiated research
Role
Researcher · solo
Duration
2026
Toolkit
Desk research, counter-evidence search, synthesis, research plan
At a glance
The problem
Feeds are fuller than ever of polished images that don’t make anyone stop, and nobody could say what’s missing from them.
What I made
A thesis that survived the evidence against it, a five-point intent checklist for designers, and a test plan to check it with real people.
67,800 years of intent
60.6% in a blind AI-art test
5 checks for designers
Context
I’m a UI/UX and product designer, and over the last year my feed changed. There are more images than ever, most of them polished, and fewer of them make me stop.
I’m not the only one who noticed. In December 2025, Merriam-Webster made “slop” its word of the year: low-quality digital content “produced usually in quantity by means of artificial intelligence.”
At the same time, I kept seeing the opposite. People who have never drawn anything stand in front of a cave painting or a Caravaggio and go quiet. Something in those works holds people who couldn’t make them. I wanted to know what that something is, and whether its absence explains the slop.
My role
A solo, self-initiated project. I wrote the question, did the desk research across archaeology, psychology, neuroscience and recent AI studies, synthesised it and planned the visuals. With no client or team, nobody was going to push back on my thesis except me, so I built the pushback into the method: for every era, I went looking for evidence that I was wrong.
The question
Why does art hold the attention of people who don’t make it?
Has that pull stayed the same from the caves to now?
If the pull comes from intent, what happens when the intent goes missing?
What should someone making a website, a video edit or a poster do about it?
My starting thesis: art is the physical form of intent. People put purpose and feeling into it, other people feel that, and nobody has to tell us to do it. AI has no instinct and no intent. It repeats patterns from work people already made, so it can’t trigger our instinct to look.
Success meant two things. The thesis had to survive the strongest evidence against it, and it had to end in something a designer can use the next time they open Figma.
How I approached it
Desk research through three historical lenses plus the present day: the caves, before anyone could have required art; the Renaissance, when makers got names; modern art, when the decision started to matter more than the object; and the AI era, the test of everything above.
For each I asked the same three questions. Was there intent in the work? Could viewers detect it? What happened when they found out it was missing? Then I went looking for counter-evidence: studies where people try to tell human art from AI art blind.
Limits. I haven’t run participants of my own yet, and most of the studies I rely on use online, mostly Western samples. The test at the end is my plan to fix that.
The record
c. 73,000 years ago: nine crosshatched ochre lines on a stone flake, Blombos Cave, South Africa (disputed)
67,800+ years ago: a hand stencil with narrowed fingers, Liang Metanduno, Muna island
1550: Vasari’s Lives gives artists names and biographies
1917: Duchamp submits Fountain
1937: a museum buys a “Vermeer” painted by Han van Meegeren
1958: Fontana’s first cut canvases
2024: Comedian sells for $6.24M; 11,000+ people take an AI art Turing test
2025: “slop” is word of the year
2026: the US Supreme Court declines Thaler v. Perlmutter; copyright still needs a human author
Nobody needed it. Someone wanted it to look a certain way.
1. The caves: nobody needed it
In January 2026, a team led by Adhi Agus Oktaviana published a hand stencil from Liang Metanduno, a cave on Muna island in Sulawesi, at least 67,800 years old, the oldest dated cave art found so far. Someone blew pigment around a hand, then did something extra: the fingers of the outline were narrowed, so it looks more like a claw. A plain handprint says “I was here.” A reshaped one says “I wanted it to look like this.”
None of this helped anyone eat or stay warm. Pigment had to be found and ground, light carried into the cave, time taken from finding food. There’s no sign anyone needed the work to exist. People made marks because they wanted to.
What it tells me: the oldest art we have is already about intent, and nobody told them to.
From Vasari on, the question came with the work: who made this?
2. The Renaissance: the maker gets a name
Medieval art was mostly made by people whose names nobody wrote down. In 1550, Giorgio Vasari’s Lives of the Artists treated artists as individuals with biographies. From then on, “who made this?” came with the work.
The idea lasted. In 1937, the Museum Boijmans in Rotterdam bought The Supper at Emmaus, praised by the leading Vermeer expert of the day as a masterpiece. It had been painted a year or two earlier by the forger Han van Meegeren, who later proved it by painting another fake in front of court-appointed witnesses. The paint never changed. What changed was what people knew about the mind behind it, and the painting’s value collapsed with it.
What it tells me: we pay for the mind as much as the paint.
Chose it. Cut it. Taped it. By the modern era, the decision was the work.
3. Modern art: the decision becomes the work
In 1917, Marcel Duchamp bought a urinal, signed it “R. Mutt” and submitted it as Fountain. He didn’t make the object; he added a choice. In 2004, 500 artists, curators, critics and dealers voted it the most influential work of modern art.
In 1958, Lucio Fontana started slashing his canvases. A 2012 EEG study of 14 people found the real cuts triggered activity linked to simulating the movement, while plain lines of the same size didn’t. Viewers’ brains seemed to re-enact the gesture. It’s a small study, so I treat it as a hint, not proof.
In 2024, a banana duct-taped to a wall, Cattelan’s Comedian, sold for $6.24 million. The buyer ate the banana. What he bought was a certificate and installation instructions. The banana can be replaced. The decision can’t.
What it tells me: by the modern era, the intent was all that was left, and people still paid for it, argued about it and felt it.
4. Now: the label changes the picture
Bellaiche et al., 2023. AI-made paintings randomly labelled “human” or “AI”. The human label raised ratings for liking, beauty, profundity and worth, most of all profundity and worth.
Kruger et al., 2004. People rated the same painting higher when told it took 26 hours instead of 4.
That’s the van Meegeren effect, measured in a lab. People respond to the mind they believe is behind the work.
5. The evidence against me
In late 2024, Scott Alexander ran an AI art Turing test on Astral Codex Ten. More than 11,000 people looked at 50 images and guessed which were human. The average score was 60.6%, not much better than a coin flip. Six of the ten most-liked images were AI-made, and even people who said they loathed AI art preferred the AI pieces when they didn’t know.
At first this looked like it broke my thesis. If intent shows in the work, why can’t people tell when it’s missing? Looking closer, AI can copy what intent looks like: it’s trained on millions of works people made with intent. And two details matter. Alexander hand-picked the 50 images to be hard to call, so the judgment in them was his. And in the label study, the AI label hurt profundity and worth the most. The surface can fool us. What we value most is the intent.
A hundred perfect sunsets, and the one drawing someone made by hand.
What it adds up to
I sorted every finding by what it said about intent: where it lives, how people detect it, what happens without it.
Intent is instinct. The choices are deliberate, but the urge to make someone feel something isn’t taught or required. Because everyone carries it, everyone recognises it in other people’s work. That’s why non-artists are moved.
We feel it because someone felt it first. A photograph, a film or a painting moves us as much as it does because we know a person is behind it. Someone thought of that frame, that cut, that colour, and felt something while making it. Looking at it, we feel a version of what they felt. The work is a bridge between two people. It’s why one handwritten letter means more than hundreds of DMs and emails: someone sat down, chose every word and pressed it into paper by hand, and you can feel that they did. That is what makes art human, and it’s what separates it from AI art: with AI, there is nobody on the other side of the bridge.
AI can copy the look of intent, not the intent. Blind tests show the surface can be copied. Label studies show meaning and worth drop away once people know.
AI repeats. A model drifts toward the most probable output. Doshi and Hauser (2024) found AI-assisted stories were rated more creative, and were also more alike. Each piece got more polished; the whole pile got more uniform.
The environment is the cause, and AI speeds it up. Stock photos and templates existed long before AI. Feeds reward volume, and AI made finished-looking work almost free. The share of work with real decisions in it shrinks.
The thesis, after the stress-test: intent is instinct. For at least 67,800 years, nobody has had to tell us to make marks that make other people feel something. We feel art because we know someone felt it first. AI can copy how art looks, but not the thing that has held our attention since the caves: a person on the other side. Because AI can copy how intent looks, the intent in your work has to come from you.
The intent checklist
Name one decision a template wouldn’t have made. If you can’t, the template made the work.
For every default you kept, know why. The font, the colour and the transition should each have a reason you could say out loud.
Make it for one person. AI aims at everyone, which means it aims at the average.
If you use AI, the intent has to come from you. Know what you want people to feel before you write the prompt, and throw out everything that doesn’t do it.
Leave something only you would do. A crop, a cut, a line of copy, a colour that came from you and not the prompt.
On a website, decide what the visitor should feel in the first three seconds before reaching for “headline, subhead, two buttons, gradient.” In a video edit, every cut needs a reason. On a poster, break the grid once, on purpose, where it matters most.
Impact
It’s too early to claim any. I’ll judge it by saves on the companion carousel (a save means someone wants to use the checklist), by comments that push back on “AI can’t have intent”, and by the test below.
What’s next
A test with 12 to 16 people, half who make things and half who don’t. Twelve images in four groups: human-made slowly, human-made quickly, AI curated from 20+ generations, and raw AI. Round one has no labels (“which ones made you stop?”, “how much do you think the maker cared?”). Round two reveals the origin. If raw AI scores lower on “cared” even blind, insight three holds. If curated AI beats raw AI, people are responding to the human choice behind it.
Sources
Oktaviana et al., Nature, 2026 (coverage); Henshilwood et al., Nature, 2018 (CNRS)
Han van Meegeren (Essential Vermeer); Duchamp’s Fountain (SFMOMA)
Umiltà et al., Frontiers in Human Neuroscience, 2012 (PDF); Comedian at Sotheby’s (NBC News)
Bellaiche et al., 2023 (open access); Kruger et al., 2004, the effort heuristic
Scott Alexander, AI art Turing test, 2024 (post); Doshi and Hauser, Science Advances, 2024 (PMC)
Thaler v. Perlmutter, cert. denied 2026 (Baker Donelson); Merriam-Webster word of the year 2025 (ABC7)
Denis Dutton, The Art Instinct (2009); Giorgio Vasari, Lives of the Artists (1550)
Looking back
What sharpened the argument: the blind tests. They looked like they broke my thesis and ended up explaining how slop can look fine and still be empty.
What I’d change: run the primary test first, not last. Right now everything leans on other people’s samples.
Weak spots I know about: the Fontana study had 14 people, the Blombos lines are disputed, parts of the Dürer and van Meegeren stories come from secondary sources, and nobody can ask the Muna artist what they meant.
The hardest part: I wanted my thesis to be true. The best finding came from looking for evidence against it on purpose.
What is Earth without art, if not a rock floating in the sky?
Next case study