The painter, not the pipeline

A robot on a beach at dusk. It watches the sun go down, then paints what it saw. That is the scene Alex Tabarrok set on Tuesday to referee a quarrel between Pope Leo XIV and Paul Graham over whether a machine can make art. The Pope's side was a single post asserting

Graham, a painter himself, answered that this is about to change. Robots wired into the world will perceive it for themselves, and a machine learning from its own sensors isn't borrowing anyone's images. Tabarrok's verdict: once the line between borrowing and perceiving fades, nothing makes the Pope's distinction "more than a statement of faith."

Tabarrok wins the argument the Pope made and misses the one he meant. The sentence carries two claims. "An ontological difference, even before an aesthetic one" is a claim about what the thing is. "Statistical calculation based on millions of images created by others" is a claim about how it was made. Graham's robot defeats the second. It never touches the first. And the second was the weaker reason from the start.

Learning from other people's pictures isn't what separates a painter from a model. It's what they share. Painters have learned by copying the masters for as long as there have been masters; that is what the copyists' easels in the Louvre are for. If training on millions of images made by others disqualifies a work as art, it disqualifies most of art history. The provenance objection was always going to lose, and Graham has only named the date. Sensor-fed robots, then models that learn from their own perception and their own output: each step takes a little more ground out from under "images created by others," and each one will be read as a retreat.

Christians have watched this move before. It is the god-of-the-gaps argument: put the sacred in whatever the science of the moment can't yet explain, and give up ground every time it does. Dietrich Bonhoeffer, writing from prison in 1944, warned against using God as a stop-gap for the incompleteness of our knowledge, because that frontier only moves one way. The Pope's sentence does the same thing with personhood. It hangs a claim about what a person is on a fact about the current state of machine learning, and the current state of machine learning has a short shelf life.

The provenance objection was always going to lose, and Graham has only named the date.

Tabarrok's strongest point deserves its full weight. If a gallery visitor can't tell the robot's sunset from a person's, a difference that shows up nowhere on the canvas looks idle, a private conviction rather than a fact about the object. But believers and skeptics alike already act on exactly that kind of difference. In 1937 Abraham Bredius, the leading Vermeer scholar of his day, hailed The Supper at Emmaus as a Vermeer masterpiece. In 1945 Han van Meegeren, charged with collaboration for selling a "Vermeer" to Hermann Göring, admitted he had painted that one and the Emmaus too, and proved it by painting another under watch.

Nothing on the canvas changed. The work did. Who made it is part of what it is. That isn't a quirk of the art market. It is the same reason a handwritten note from a friend and a mail merge with the same words are different things, and nobody needs a theology to feel it.

So the question the robot on the beach raises isn't faith against fact. It is whether anyone is there: whether the machine is a maker or a process. That is a claim about persons on both sides. "A robot that perceives is an artist" is no more an empirical finding than "it isn't," and Tabarrok's closing question applies to his own answer as much as to the Pope's. The Pope is right that the difference is ontological. He should stop pinning it to the training set.

The robot will see the sunset. The argument was never about whether it could see. It was about whether anyone was looking.