Thirty Paintings, Zero Humans

A landmark study proved we prefer human-made art. Then it revealed nothing in it was human-made.

Let me tell you about the most quietly devastating experiment I’ve ever read, and then let me ruin your week with what it actually means.

A team of researchers at Duke, led by a PhD student called Lucas Bellaiche, wanted to know something simple. Do people really prefer art made by humans over art made by machines, and if so, why? So they built a test. They gathered thirty paintings. They showed them to people, one at a time, and asked them to rate each one on four things. How much they liked it. How beautiful it was. How profound. How much it was worth. Every painting carried a little label telling you who made it. Human, or AI.

The results were exactly what you’d hoped for as people rated the human work higher. Not by a hair, either. A real, meaningful swing towards warmth, towards depth, towards worth. The human stuff moved people. The machine stuff left them colder. Score one for the species. We can still tell. There’s something in us that knows.

Except.

Every single painting in that experiment was made by a machine. All thirty. The “human” label and the “AI” label were shuffled and slapped on at random, like name tags at a party where nobody is who they say they are. There was not one human brushstroke in the entire study. The only human thing in the room was the lie on the label.

Sit with that for a second, because it’s doing something sneaky to everything you believe about your own taste.

People didn’t rate the art. They rated the story. Tell someone a person made this, and they find depth in it. Tell them a machine made the exact same image, and the depth evaporates. The pixels didn’t change. The thing that changed was what the viewer believed about the hours behind the image. The late nights they imagined. The ink-stained fingers they pictured. The lived-in human they assumed was on the other end of it, paying for this in some small way.

We don’t respond to the object. We respond to the imagined cost of making it.

And here is where it stops being a cute psychology result and starts being the most important sentence in this whole argument. The researchers cite an older study, one I can’t stop laughing at in a bleak sort of way. Give people Coca-Cola with the label on, and they enjoy it more than the identical Coke poured from an unlabelled cup. Same liquid. Same sugar. The brand does the tasting for you.

So that’s where we are. The thing we tell ourselves is sacred about art, the soul of it, the part no machine can touch, behaves in a lab like the logo on a fizzy drink. A context cue. A story we swallow before the first sip.

Now. The instinct is good. The loyalty is real. People want to value human effort, reach for it, defend it on pure gut feeling before they can even explain why. That subliminal pull towards human creation is the best thing we’ve got going for us. It might be the only firewall left between working artists and a future that has decided it can do without them.

But read the study the way the machine-builders are reading it. Because they are. Those “strong implications for marketing and branding” the authors mention? That’s not a footnote to them, that’s a roadmap. If preference lives in the label and not the canvas, then you don’t need to make better art. You need to manage the label. You need a generation that stops checking it. A generation raised on the feed, that never waited months for an artist’s next issue, never learned to recognise a particular penciller’s hand across decades, never built the loyalty in the first place because nobody ever taught them there was something there worth being loyal to.

If people stop expecting craft, they stop valuing it. And the study is the proof of concept, sitting right there in a peer-reviewed journal, that the valuing was always at least partly a story we could be told. Which means it’s a story we can be told out of.

That’s the part that should keep you up. Not that the machine can paint. We know it can paint. The frightening thing is that the experiment found the soul of art hiding not in the artist’s hand, but in your willingness to believe a hand was there at all. And belief, unlike a brushstroke, can be edited.

So the next time something stops you in your tracks, a page, a panel, a cover that dares you not to look away, check the label. Then check yourself. Ask whether you’d feel the same if the label flipped. Because somewhere, right now, someone is working very hard to make sure that one day you won’t bother looking, and won’t feel the difference when it’s gone.

The machine doesn’t dream. It replicates. The terrifying discovery is how much of the dreaming was happening on our side of the page all along.

Drawn to Extinction is officially out on 1st June 2026, with a foreword by Pat Mills. The Bellaiche study is real; the citation is below. If this rearranged something in your head, that’s the book doing its job, and the best thing you can do is hand it to someone who’s stopped checking the label.

Lucas Bellaiche et al., “Humans versus AI: whether and why we prefer human-created compared to AI-created artwork,” Cognitive Research: Principles and Implications, 8:42 (2023).