Crunching a great deal of data about some very small things, AI can provide a stream of information to help detect multi-million-dollar forgeries. For example, every artist’s brush strokes differ but only minutely:

In a new paper, researchers from Rutgers University and the Atelier for Restoration & Research of Paintings in the Netherlands document how their system broke down almost 300 line drawings by Picasso, Matisse, Modigliani, and other famous artists into 80,000 individual strokes. Then a deep recurrent neural network (RNN) learned what features in the strokes were important to identify the artist.

Jackie Snow, “This AI Can Spot Art Forgeries by Looking at One Brushstroke” at Technology Review

The paper is open access.

Many older methods of detecting art forgery yield results. Most of them are high-tech (microscopy, X-ray, mass spectrometry, infrared reflectography). But the first step is rather low-tech: provenance:

Provenance investigators, then, unravel the history of an artwork using public and private records, archives, and other art historical research methods. At times, through looking at correspondence, catalogues, sales receipts, and even the artwork itself they discover it was very unlikely or even impossible for an artist to create a particular work. At other times, they might uncover a previously unknown legitimate record of the artwork’s existence which provides strong evidence that it is real.

University of Glasgow, “Using science to detect art forgeries” at FutureLearn

If a disputed work passes the “Is it possible?” test, high-tech methods can increase the probability to beyond a reasonable doubt. But, of course, no one can ever say for certain. The problem is more along the lines of knowing when and why to suspect forgery because life is messy:

Leaving straight forgeries aside, any discussion about the “authenticity” of an artwork opens suddenly, like a trapdoor, into the murk of semantics. On the sliding scale of attribution that art historians use – painted by; hand of; studio of; circle of; style of; copy of – each step takes the artist farther from the painting. These variations, often subtle, are compounded by the unease about overpainting; Salvator Mundi had been worked over so many times and so heavily, critics argued, that it was less by Da Vinci than by his restorers. Deliberate fakes, misattributions and poor restorations all encroach into the realm of the authentic. In two decades at the Met in New York, Thomas Hoving, the museum’s director until 1977, must have examined at least 50,000 objects, he wrote in his book False Impressions. “I almost believe that there are as many bogus works as genuine ones.”

Samanth Subramanian, “How to spot a perfect fake: the world’s top art forgery detective” at The Guardian

Is a perfect forgery possible?

As a thought experiment, it is possible to envision the immaculate forgery – the one that defeats scientist and connoisseur alike. Our villain is a talented copyist, well practised in the style and the themes of his chosen artist. He is also a resourceful procurer of materials, able to rustle up every kind of age-appropriate canvas and frame, pigment and binder. He fits his forgery neatly into a chain of provenance – giving it the title of a now-lost work, or providing false documents to claim that it had been part of a well known private collection.

In theory, if each of these steps is perfectly performed, there should be no way to expose the painting as fake. It will be a work of art in every way save one. But the world of today, the world in which the forgery is being created, is likely to fix itself in some form within the painting – as radioactive dust, perhaps, or as cat hair, or a stray polypropylene fibre. When that happens, only the scientist can hope to nab it.

Samanth Subramanian, “How to spot a perfect fake: the world’s top art forgery detective” at The Guardian

Or perhaps the forger neglected to get hold of and use the newly developed AI to perfectly replicate the brushstrokes…

The most significant use of AI in art may be the much less sensational role of sorting out controversies over who painted what in situations where forgery is not in question. Art history is more complex than we think. As Subramanian notes, great artists might paint the same theme several times, students or other artists might fill in paintings, and lesser artists might copy the work, maybe changing it. Machine learning systems, trained on hundreds of images, can spot tiny similarities that identify a style and thus tell a story of composition:

The computer, says Honig, can pick up “so many more details, so much more easily”. Take windmills: hundreds of pictures featuring them fill her Brueghel database. The algorithm has picked up identical images of the structures in multiple paintings. It can even show when a replica has been flipped. And it has helped to pinpoint exact copies of lions, dogs and other figures. The workshops of many Renaissance artists were co-working spaces, so the computer technique helps Honig to piece together how different artists, in the family or not, might collaborate. “Rubens comes in and does some figures, and then Jan Breughel comes in and does the horses, the dog and the lion, because he’s ‘Mister Animal’,” Honig says. “And so they fit the things together.”

Many art historians surmised, on the basis of records and close observation, that this is what happened with numerous paintings by the younger Brueghels. The computer helps to prove it. Hong says: “It addresses a lot of questions about the process of production.”

David Adam, “From Brueghel to Warhol: AI enters the attribution fray” at Nature

By the very nature of collaborative work, if simple fraud for financial gain is not in question, many much harder questions remain, questions where AI might provide some information that enables better judgment. But hardly the stuff of “AI Is Taking Over.”

* Notes on the illustration: Portrait of a Woman, 18th-19th centuries. Oil on canvas, 62.23 cm. x 48.26 cm. Fogg Art Museum, Bequest of Grenville L. Winthrop, 1943. This portrait of Maria Isabella de Bourbon, infanta of Spain (1741-1763), thought to be painted by Goya was bequeathed to the Fogg Art Museum in 1943. Although the canvas was old, and the paint bore the crackle marks of age, several scholars came to doubt the painting’s authenticity. In 1954, X-ray images were taken of the painting and conservators were surprised to find an earlier portrait of a different woman beneath the surface. X-ray diffraction analysis revealed the presence of zinc white paint, which was invented after Goya’s death. Thorough cleaning of the painting confirmed that the surface paint was relatively modern and had been applied so as not to obscure the craquelure of the original. Curators also discovered extensive damage to the base portrait; leading some to speculate that the forger attempted to scrape off the earlier face. Upon completing the analysis, the conservators left the work as you see it above (with portions of the original painting visable, on the left, and the newer forgery on the right), to illustrate the intricacies of art forgery, and the inherent difficulty of detecting it. (It is thought that the base painting is a provincial Spanish work dating to the 1790s). More.

Further reading on AI and art:

AI as the artful dodger Watch what happens when I train a neural network on portraits of 56 famous scientists, starting the process with a right eye. (Computer science prof Robert J. Marks tries his hand on portraits of famous scientists.)

and

Software engineer Brendan Dixon offers a three-part series on AI and creativity:

Part I: Why AI appears to create things: When AlphaGo made a winning move, it exhibited no more creative insight than when it played pedestrian moves

Part II: Why AI fails to actually create things Only one of the traits du Sautoy suggests is an essential part of creativity

Part III: The first question posed to me as an artist was, “What are you trying to say?” Du Sautoy believes that AI will “in the distant future” achieve consciousness. For that, we have no evidence. It is a statement of religious faith akin to that of Anthony Levandowski’s AI Church.

Also by Brendan Dixon: Does the Butterfly Effect sharply limit AI’s power?