But there’s a part of high culture where AI is taking on a starring role as hero, no displacing the standard protagonists—art experts and conservators—but adding a strong, powerful weapon to their arsenal when it comes to fighting forgeries and misattributions. Based solely on an analysis of a digital photograph of a decoration, AI is already very adept at recognizing and verifying an artist’s work.
AI’s goal research has thrown a screw into this traditional hierarchy. Where do the veteran art historians, whose reputations have been based on their personal expertise, go if an algorithm is use statistical probability to determine the authorship of an artwork? In fact, AI will not change connoisseurs, only as the use of x-rays and graphite dating years ago did not. It is merely the most recent addition to a range of advanced identification tools.
A great AI may be “fed” a customized database by human craft historians to build up its understanding of an artist’s design, and individual art historians may interpret the results. In November of this year, a leading AI company, Art Recognition, published its research of Rembrandt’s The Polish Rider, which had a reputation for confounding scientists and sparked debates over how much, if any, of it had really been painted by Rembrandt himself. The AI exactly matched what most aficionados had posited about which pieces of the artwork were by the king, which were by individuals of his, and which involved the hands of over-enthusiastic painters. When the scientific method supports the expert opinion, it is especially compelling.
We humans find hard scientific data more compelling than personal opinion, even when that opinion comes from someone who seems to be an expert. The so-called” CSI effect” describes how jurors interpret DNA evidence as being more persuasive than eyewitness testimony. But when expert opinion ( the eyewitnesses ), provenance, and scientific tests ( the CSI) all agree on the same conclusion? That’s about as close as one can get to a certain conclusion.
But what happens when the owner of a work that, at first glance, looks totally inauthentic to the point of being laughable, recruits a slick firm with the task of gathering forensic evidence to support a preferable attribution?
Lost and Found
Back in 2016, an oil painting surfaced at a flea market in Minnesota and was bought for less than$ 50. Its owners are now claiming that it could be a lost Van Gogh and be worth millions of dollars. ( One estimate suggests$ 15 million. ) A resounding “nah” was the answer, at least for those who had functioning eyes and were somewhat familiar with art history. The painting is stiff, clumsy, utterly lacking the feverish impasto and rhythmic brushwork that define the Dutch artist’s oeuvre. Even worse, it lacked Elimar as its signature. And yet, this dubious painting has become the center of a high-stakes battle for authenticity, one in which scientific analysis, market forces, and wishful thinking collide.
The” Elimar Van Gogh,” as it has become cliched in art circles, is now run by LMI International, an art consulting firm. They are investing heavily in getting experts to say what they want to hear: that it is, in fact, a genuine Van Gogh. This is where things start to get murky. The world of art authentication is not a straightforward affair. In contrast to the hard sciences, art history deals with connoisseurship, conflicting expert judgments, and probabilities. It is also, crucially, an industry driven by financial incentives. Its value skyrocketes if the painting is deemed authentic. If it’s deemed a fake, or rather in this case a derivative work by someone named Elimar who daubed a bit on canvas, distantly inspired by Van Gogh perhaps, but with none of his talents, it’s virtually worthless—about as valuable as you might expect to find at a flea market in Minnesota for under 50 bucks. This conflict in the stakes has resulted in a risky practice: hiring experts to verify the authenticity rather than confirm it.
Horton’s attempts at authentication clashed with the traditional, often opaque, art world. Due to the absence of any evidence of a connection to Pollock’s studio, the majority of Pollock experts disputed the painting’s legitimacy. In response, Horton turned to forensic science. She employed controversial fingerprint expert Peter Paul Biro, who claimed to have discovered a fingerprint on a painting that matched one on a paint can in Pollock’s studio. This was not enough to sway the art establishment. In contrast, David Grann’s” Mark of a Masterpiece,” a New Yorker film from 2010 that completely discredited Biro and his authentication work, was released in full. Biro then sued for libel and lost. No museum or auction house, despite diligent efforts, accepted the piece of art as authentic, leaving it undisclosed and its status limbo. This is a stark illustration of how challenging ( and subjective ) authentication can be when money, reputations, and scientific claims collide. That’s what happens when such an approach goes wrong. Now that new technology is available, will the Elimar owners have better luck finding new ways to authenticate art?
Real Talk
One of three methods has traditionally been used to authenticate art: connoisseurship, provenance research, and forensic testing. Connoisseurship is the oldest and remains the default method, it relies on the opinion of self-proclaimed experts who physically examine the object in question and tell you what they think. Provenance research examines an object’s documented history and can help determine whether it is actually documented as having ever existed ( mentioned in archival materials, letters, catalogues raisonnés, or gallery listings ), or whether a chain of ownership and attribution is accounted for to indicate that the work was not stolen and that it has been regarded as authentic for an extended period of time. Forensic testing involves conservators examining the object with methods like carbon dating, x-rays, and infrared spectroscopy to see if there are anachronistic elements or if all appears as it should be for an artwork of the era and authorship that the experts suppose it to have. The Elimar painting’s owners are hoping that one or more of these three techniques will convince people that their van Gogh-inspired flea market piece is indeed a Van Gogh.
Enter LMI Group International, a data science firm that bought the” Elimar” from the original owner and assembled a 458-page document that claims to authenticate the painting. The phrasing is crucial because they don’t need to determine whether something is done by Van Gogh, but rather to find a way to confirm that. This distinction is everything. The owners, experts, and auction houses in question stand to gain a ton if the painting is found to be genuine. If it is declared inauthentic, the only winners are the people who weren’t taken in by the charade in the first place. Skepticism should be the default response in these dynamics.
Other experts agreed. Next up: A newly discovered Michelangelo, signed’Steve’, posted on social media by acclaimed American artist Jerry Saltz.
So if it’s not a Van Gogh, what is it? Some art historians speculate that Henning Elimar, a less well-known Danish artist, may be the author of” Elimar.” This theory is supported by similarities in signature and style between the painting in question and known works by Henning Elimar.
Hold on. An Elimar painting signed” Elimar” could actually be by Elimar? That undoubtedly sounds more plausible.
So connoisseurs are not helping out the owners of the former flea-market painting. What does the case study’s provenance say about it? Typically, a documented chain of ownership is crucial in establishing a work’s authenticity. Before its 2016 appearance, there was no such documentation in the case of” Elimar,” which makes it difficult to attribute the painting to Van Gogh.
Additionally, comprehensive catalogues raisonnés, like those created by Jacob Baart de la Faille and Jan Hulsker, do not include anything resembling” Elimar” among Van Gogh’s most well-known works. These catalogues aim to document all recognized artworks by the artist, and the absence of something that could fit the description of” Elimar” further discredits claims of its authenticity.
Forensics, what about it?
Authentication Factor
A new authentication technique that combines AI and machine learning has been developed since the” Lost Jackson Pollock” era.
The aforementioned Art Recognition has already run its own AI analysis on the painting—and the results are damning. Carina Popovici, the company’s founder, claims that a painting’s authenticity is 99 percent certain based on the closed AI algorithm used by Art Recognition. This is a significant result and one that, until now, has not been made public.
Zurich-based Art Recognition has used 834 verified authentic Van Gogh artworks and 1, 785 inauthentic images to teach it what to avoid being deceived by. The program can analyze brushstroke patterns, color composition, and other features imperceptible to the human eye. ( Full disclosure: I’ve advised Art Recognition in my capacity as an author and researcher on forgery and art theft, but my opinion on the” Elimar” was formed long before my involvement with the business. )
The fact that Art Recognition’s AI has come back with such a strong negative result should, in theory, settle the matter. LMI Group International, whose mission seems less about discovery and more about justification, is still supporting the painting. LMI’s website still bears the headline:” Elimar Returns: A Newly Identified Work by Vincent van Gogh. Read the entire report that establishes the identity of Vincent van Gogh’s painting. Despite this, no one of importance seems to be convinced. According to legend, LMI paid about$ 1 million for the painting’s authenticity. That is an astounding sum—authenticating a painting should cost far less than that. ( Art Recognition charges only four figures per test via its AI system. )
This is not an isolated case. Potential conflicts of interest in authentication have always been a problem for the art market. The stakes are too high, and there is often more financial incentive in declaring a work real than in dismissing it as a fake or misattributed work of little value.
The art world takes pride in having a master’s eye through years of research. The saga of the” Elimar Van Gogh” serves as a case study in how art authentication can go astray when it is led by hypothesis bias. In a nutshell, we want this to be by Van Gogh; make a case to confirm it. Authentication is a process that, at its best, should be about discovery and scholarly rigor. However, too frequently it is motivated by the desire for profit, leading to situations where scientific findings are disregarded in favor of more convenient conclusions, or where data is produced or interpreted in a way that supports an hypothesis rather than allowing for the most logical, objective conclusion.
In cases like this one, AI can provide a crucial check on the excesses of the market. It is objective, non-biased, and free of any financial pretensions unlike human experts. And in this instance, it has done what should have been obvious from the start: It has called out the painting as not by the famous artist in question.
There is hope for a more transparent future as AI authentication gains popularity and the art world becomes more aware of the conflicts of interest inherent in traditional methods. Importantly, AI is a forensic tool that can test a work in question using a digital image alone, thereby obviating the need to send a fragile painting on an expensive and risky trip to a lab. This effectively makes it an effective complement to, but not a substitute for, human scientists, conservators, and researchers from all over the world.
For now, one thing is certain: If a painting doesn’t look like a Van Gogh, isn’t signed by Van Gogh, and is confirmed by AI not to be a Van Gogh, then—despite what the market might wish—it simply isn’t a Van Gogh.
