The Algorithm of Attribution

Why AI Won’t Solve the Art World’s Authenticity Crisis

Why do we care about authenticity anyways?

Artificial intelligence is now entering the high-stakes arena of art attribution, attempting to settle age-old debates with cold, hard data. A recent report in The Guardian details how AI analysis has cast significant doubt on two versions of Saint Francis of Assisi Receiving the Stigmata, long attributed to the Flemish master Jan van Eyck. The Swiss company Art Recognition used brushstroke analysis to determine that the Philadelphia Museum of Art’s version was “91% negative” for Van Eyck’s hand, while the Turin version was “86% negative.”

The Arnolfini Portrait: The Benchmark

At first glance, this feels like a victory for transparency. As someone who has spent years in the art world, I have long been skeptical of the "expert" attributions that grace museum walls. History suggests that many of these labels are more aspirational than factual. This could equalize the playing field, allowing small-time collectors the opportunity to prove their paintings' value. However, my initial optimism has been tempered by a realization: the art world is a closed circuit where wealth and status hold the keys. As long as art is an investment, then attribution equals money, and any tool—even AI—is susceptible to the gravity of vested interests.

The Expert’s Blind Spot

The documentary The Lost Leonardo (2021) serves as a perfect cautionary tale for this phenomenon. It chronicles the "discovery" and $450 million sale of the Salvator Mundi, a painting whose attribution to Leonardo da Vinci remains highly contentious. Watching the film, one sees how easily "truth" becomes secondary to power and profit. From my perspective, in the Salvator Mundi, certain passages, such as the hands, show a master’s touch, while others obviously lack the finesse of a genius. Yet, experts and dealers pushed the narrative until the price tag became too big to fail.

This highlights a fundamental issue: I’ve always believed that to truly identify a master’s work, one must have at least a minimal understanding of how to hold a brush. Art historians are notoriously prone to over-interpreting aesthetic choices. For example, they often attribute a sitter’s expression to "narcissism" or "arrogance," whereas an artist knows that we frequently adjust features simply to make the subject look better. We don’t want to paint ugly people. I have also lost count of how many times I’ve had to explain basic media and process to an art historian. I don’t need a spectrometer to see how an image was built; I use my eye and my experience. This just points out how certain aspects of the art world treat art like a science where they can label and categorize things into neat little boxes.

“I recently suffered an injury that changed the way I hold my arm, which has fundamentally altered my mark-making. Does that mean my new work isn't 'mine'?”

The Flaw in the Machine

Salvator Mundi: A Narrative of Profit

While in this article AI seems to promise objectivity, it is only as reliable as its training data. How do we establish the "standard" by which all other works are judged? Who determines the true authentic Van Eyck? The Guardian article notes that the AI used the Arnolfini Portrait as a benchmark for Van Eyck. But collectors and institutions have been fudging provenance for centuries. If the "standard" image is itself a studio work or has been heavily restored, the AI is simply validating a lie. Furthermore, these algorithms assume that an artist is a static machine. They don’t account for the evolution of style or physical reality. If Van Eyck had a bad year, or a different set of brushes, or a change in health, a rigid AI would likely flag his work as a forgery.

The Battlefield of Creation

This leads to a larger question: Why does the name on the frame matter more than the quality of the work? The answer is simple: investment. If we valued art for its innate qualities—the "shimmering light and supernatural clarity"—then the culture's obsession with attribution would vanish. Maybe this is the most positive thing AI is going to do. When everything can be made by anybody, maybe the “anybody” will become less relevant.

There is also another silver lining. If AI can reduce an artist’s work to a replicable algorithm, it inadvertently defines what art we value as a whole. Maybe we will learn to see that creativity is usually not a standardized process; it is a much messier, like a battlefield. It is the mess of the studio, the constant pivot through mistakes, and the search for a path through the "battlefield" of a canvas.

AI might help experts validate lucrative, inaccurate attributions by providing a veneer of scientific certainty. But for the artist, it serves a different purpose: it proves that the soul of a painting lies in the parts that can’t be calculated—the grit, the struggle, and the human inconsistency that no algorithm can truly map. For me, AI has clarified where the real value is in art and I can easily say it’s not in the easily recognizable patterns of brushstrokes. It’s obvious to most people that the value of art is going to shift in the near future, but it is less clear for most where it will land. Wherever it lands, I have an idea that attribution will not be the winner.

×