The Human Premium: When Making Is Free, Provenance Is the Product

In June, two researchers at East Carolina University, Andy Bowman and Eric Kisling, reported a result that sounds like a contradiction. People in their study had similar emotional reactions to art made by humans and art made by AI. They still rated the human work as more valuable, and saw it as more complex.

Same feeling. Different worth.

The BBC has run an analysis under a headline that asks the obvious next question: AI is creating a “human premium” for art made by people, but will it hold? That is the honest place to start, because nobody knows yet. A premium built on a preference disappears the moment the preference does, and this one has not been watched long enough to tell a lasting market from a passing mood.

What we can do is ask where the human premium has a reason to exist at all. That question has a sharper answer.

One illustrator, two kinds of work

Take a freelance illustrator with an ordinary mixed book of clients. Some hire her because of her. Others hire her because they need a picture of something.

The portraits come from couples, authors, and people who will frame the result. The headers and icons come from teams who need a blog post to have a picture on top of it. Both kinds of work paid her the same day rate not long ago.

The bottom panel did not get worse. Her icons are as good as they ever were. What disappeared was the buyer’s reason to pay for them, because that buyer was never paying for her. They were paying for an icon, and an icon is now nearly free.

The top panel is untouched, and the reason is not quality. A model can draw in a style that looks like hers. It cannot make the portrait be by her, and being by her is the entire thing the couple is buying.

The model: provenance is the product

Here is the model the rest of this piece uses, in one sentence: the human premium attaches to provenance, and provenance only matters to a buyer who is paying for who made the thing rather than for what the thing does.

Most real work sits somewhere along that line rather than at either end. A novel is bought partly for its story and partly for its author. A wedding photo is a function (proof the day happened) and a relationship (this photographer saw us). The model does not sort work into two bins. It predicts which part of the price survives when making becomes free: the part that was always about the maker.

The rest of this piece tests that model against four findings.

Finding 1: people feel the same and value differently

The East Carolina result is the cleanest version of the split, and it is not alone.

In 2023, Lucas Bellaiche and colleagues, writing in Cognitive Research: Principles and Implications, took paintings that were all made by AI, labeled some “human-created” at random, and asked people to judge them. The ones labeled human scored higher on liking, beauty, profundity and worth. Same pixels. The only thing that changed was the story of who made them.

That is the model working exactly as stated. If the buyer were paying for the image, the label would not move the number. The label moves the number.

Finding 2: buyers pay for the making, not the object

If the premium is not in the pixels, where is it? Three older findings answer that, and none of them is about AI.

Justin Kruger and colleagues named the first one the effort heuristic in the Journal of Experimental Social Psychology in 2004: people judge work as better when they believe it took longer to make. George Newman and Paul Bloom found the second in the Journal of Experimental Psychology: General in 2012: an original is worth more than a perfect duplicate partly through contagion, the belief that the artist’s touch left something in the object. Bowman put the third plainly when describing his own study: the question is whether art’s value comes from the creator’s intent or from the final image itself.

Notice what the three share. None is a property of the object. You cannot inspect a poem and see eighteen hours in it. That is exactly why generation cannot compete here. It can match the output perfectly and still have nothing to offer on the axis the buyer is pricing.

Finding 3: where function is the job, the premium collapses

There is a full, finished precedent for what happens when a machine wins the function outright. It is on your wrist, or it used to be.

Seiko released the Astron, the first quartz wristwatch, in December 1969. Over the next fifteen years Swiss watch employment fell from roughly 90,000 to roughly 30,000, depending on whose count and which year you take. The machine did not need to be loved. It only needed to tell time better and cost less, and it did both.

Then the part people forget. The Swiss industry that survived is built on mechanical watches, which today account for most of the value of Swiss watch exports, according to figures from the Federation of the Swiss Watch Industry. A mechanical watch keeps worse time than a cheap quartz one. Nobody buys it for the time. They buy it for the movement someone assembled, the name on the dial, the hand-finished parts they will never see.

That is both halves of the model in one industry. Wherever the watch was a function, the premium died. Wherever it was a provenance, the craft survived, smaller and pricier, valued for the opposite reason it was once bought.

Finding 4: provenance has to be provable

The premium has a structural weakness. If you cannot tell human work from generated work by looking, the buyer has to take someone’s word for it.

Music shows how quickly the eye, or ear, stops being useful. In a survey Deezer ran with Ipsos in late 2025, 9,000 people across eight countries heard three tracks, two AI-generated and one human, and 97 percent failed to identify the human one. When the object cannot carry its own provenance, something else has to. The BBC piece describes “AI-free” labels already appearing on some books, films and albums, and academics working on a human-made mark modeled on Fairtrade.

US law has landed in the same place from a different direction. In Thaler v. Perlmutter, the D.C. Circuit held in March 2025 that the Copyright Act requires a human author, so a work generated entirely by an AI system cannot be registered. The Supreme Court declined to hear the appeal in March 2026. Purely generated work is, legally, nobody’s.

Put those together and the premium has an address. It does not live in the finished file. It lives in the evidence of the making: the sketches, the session recordings, the signature, the name a buyer can check.

What this gets wrong

The model is clean at the two ends of the line and shaky in the middle, which is where most work actually lives.

First, almost every finding above measures what people say, not what they pay. A later replication effort, published as “The Effort Heuristic Revisited,” found the effort effect on liking and perceived quality but not on monetary value. A premium that shows up in ratings and vanishes at checkout is not a premium. The watch market is the strongest evidence here precisely because it is money, not a survey.

Second, the preference is young. The BBC’s own question, whether it will hold, is the right one. People may currently value human work partly because generated work is new and slightly threatening. A generation raised on it may not care, and there is no long-run data either way.

Third, the line between human and generated work is already blurry and will get blurrier. The US Copyright Office has said that human contributions to AI-assisted work can still be protected, which concedes that most work will be mixed. If an illustrator sketches by hand and a model fills the background, which column is that? The model says the premium attaches to the human part, but buyers have no way to see the proportions.

Fourth, every label is a claim, and claims get faked. A human-made mark is worth exactly as much as the process that checks it. Nobody has built that process yet.

What to do with this

The practical move is the same whatever happens to the market: stop hiding the making.

For most of the last century, the record of how something was made was waste. The sketches went in the bin, the drafts got overwritten, the hours were invisible by design, because a client wanted a clean finished thing. If provenance is now the product, that waste is the evidence. Keep it, show it, sign it. Let the buyer see the eighteen hours.

Then be honest about the other column. Work bought for its function is going to be priced like a function, and arguing for a human premium there is arguing with the quartz watch. Price it as a commodity and make it cheaply, or move it toward the side of the line where someone wants it specifically from you.

The illustrator’s icons are not coming back. Her portraits were never in danger. The question for everyone who makes things is which of those two their work actually is, and whether the buyer can see the difference.