Robots: Experts at spotting art forgery, still don’t get art

Johannes Vermeer, the 17th century Dutch painter, produced 34 paintings over the course of what is now regarded as an illustrious career. Or maybe it was 35. The precise number isn’t entirely clear and has changed over time.

From 1937 until 1945, “Supper at Emmaus,” a painting of Jesus Christ and his disciples, was attributed to Vermeer. In fact, the Vermeer expert Abraham Bredius called it “the masterpiece of Johannes Vermeer of Delft.” It was not, however, the masterpiece of Johannes Vermeer; it was the masterpiece of a forger named Han van Meegeren. In 1942, while the Netherlands was under Nazi occupation, van Meegeren sold “Supper at Emmaus” to a Nazi banker who, in turn, sold it to Reichsmarschall Herman Göring. 

The joke was on the art world. 

Van Meegeren might never have claimed his forgery, but his hand was forced. Upon World War II’s conclusion, the Dutch government decided to prosecute citizens who had sold the nation’s treasures to the Nazis. This included Han van Meegeren, ostensibly the one-time owner of “Supper at Emmaus.” Faced with the threat of going to jail, van Meegeren came clean. To prove that the experts had been wrong about “Supper at Emmaus,” he painted another forgery in court. The joke was on the art world.

What if this scenario was slightly altered and it turned out that the famous painting had in fact been produced—and possibly forged—by a robot? Would that be any different?

This, in broad (robotic) strokes, is the situation Nicholas St. Fleur lays out in The Atlantic. He cites the computer scientist Lior Shamir, who “believes that computers will eventually be able to create artwork indistinguishable from a person-made painting.” Computers can now analyze paintings and distinguish between originals and forgeries. Indeed, an algorithm Shamir developed can distinguish between genuine and fake Jackson Pollock paintings with remarkable accuracy. If a computer can figure out precisely what a real painting looks like, fabricating one would be a lesser challenge. Thus, St. Fleur writes, “Digitally, a computer program can create a piece of artwork that can pass for an original.”

That does not mean that a robot is currently positioned to step into van Meegeren’s shoes. “Just because a computer can recognize something,” St. Fleur writes, “doesn’t mean it can reproduce that thing.” Indeed, the pattern recognition skills that make algorithms good at recognizing forgeries do not make for good art. They only make for art that could pass the computer’s decidedly un-artistic test. “The programs would generate artworks based on a single signature,” St. Fleur writes, “but might otherwise miss the overarching aesthetic of an authentic Pollock.” This is the artistic equivalent of missing the forest for the trees.

For the sake of argument, however, let us briefly assume that some combination of algorithms and robots could create respectable art. How might that affect the art market? As of yet, the authentication of a painting depends on its provenance—the record of its ownership over time—more than a computer algorithm. Art forgery has therefore historically been a crime that exploited human vulnerabilities. To wit, van Meegeren exploited the desires of experts like Bredius to identify definitive works and of Nazi collectors who wanted to buy art that reflected their moral worldview. The switch to algorithmic authentication might well just replace playing to a human test with playing to a computerized test. The scenario St. Fleur envisions involves the same amount of pandering as in the status quo, that pandering would just be to our robotic overlords.

The irony in all of this talk of art fraud is that Vermeer may have benefited from technological assistance in creating the originals Han van Meegeren sought to emulate. Tim’s Vermeer, a documentary directed by Teller (of Penn and Teller fame), suggests that Vermeer may have used optical devices when making his paintings. This, allegedly, allowed the painter to achieve a remarkable level of photorealism. If that theory is true, it is proof that technology can help produce art, but that the process remains fundamentally human. Until robots gain compositional awareness, and so long as art is consumed by human eyes, technology alone cannot replace the artist. 

H/T The Atlantic