{"id":322809,"date":"2026-03-10T17:56:24","date_gmt":"2026-03-10T17:56:24","guid":{"rendered":"https:\/\/www.newsbeep.com\/nz\/322809\/"},"modified":"2026-03-10T17:56:24","modified_gmt":"2026-03-10T17:56:24","slug":"masterpiece-or-cheap-copy-art-historians-and-ai-may-not-agree","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/nz\/322809\/","title":{"rendered":"Masterpiece or cheap copy? Art historians and AI may not agree"},"content":{"rendered":"

\n To the untrained eye, there is very little difference between the three known versions of \u201cThe Lute Player.\u201d Almost identical in composition, the paintings all depict a young, doe-eyed subject in white robes, instrument in hand and turned slightly away from the viewer. Each appears to carry Italian painter Caravaggio\u2019s signature mastery of light and shadow.\n <\/p>\n

\n To art historians, however, there has long been broad agreement: The versions held by Russia\u2019s Hermitage Museum and France\u2019s Wildenstein Collection were created by the Baroque artist, while the one at Britain\u2019s Badminton House is merely a copy.\n <\/p>\n

\n Artificial intelligence begged to differ. In September, Swiss AI firm Art Recognition claimed there is an almost 86% chance that Badminton House\u2019s version is, in fact, authentic. The company\u2019s model, which was trained to recognize markers of Caravaggio\u2019s style, including shapes, color palettes and compositional structures, also declared (albeit with less statistical certainty) that Wildenstein\u2019s version is likely a copy. Its analysis found a \u201csignificant divergence\u201d between the latter painting\u2019s \u201cvisual characteristics\u201d and those of Caravaggio\u2019s other works.\n <\/p>\n

\"Three<\/p>\n

\n This is one of several bold claims made by Art Recognition since it launched seven years ago. In 2021, the company calculated a 91% chance that a painting at London\u2019s National Gallery attributed to Peter Paul Rubens, \u201cSamson and Delilah,\u201d was not produced by the Baroque painter. A long-disputed painting of Vincent van Gogh at the The National Museum in Oslo, meanwhile, had a 97% chance<\/a> of being genuine. The firm\u2019s other analyses have presented more complex results: Rembrandt\u2019s \u201cThe Polish Rider,\u201d for instance, was partly produced by someone else, though some sections carry evidence of the Dutch painter\u2019s hand, ranging in certainty from 69% to 83%<\/a>, according to the AI model.\n <\/p>\n

\n Art Recognition\u2019s declarations have not always contradicted the established scholarship. The Van Gogh attribution, for instance, was subsequently matched by more conventional research<\/a>, including technical analyses and studies of the artist\u2019s letters (museum experts concluded that the portrait\u2019s unusually dampened colors simply reflected Van Gogh\u2019s troubled mental state at the time). Yet, many art experts remain highly skeptical about AI\u2019s ability to supersede, or even complement, the tools traditionally used to authenticate works of art.\n <\/p>\n

\n \u201cI think it\u2019s quite problematic,\u201d said Angelamaria Aceto, a senior researcher at the University of Oxford\u2019s Ashmolean Museum of Art & Archeology. \u201cI\u2019m very open to new technologies; I use technologies all the time that can help you to see what the naked eye can\u2019t \u2014 to go beneath the surface. And I\u2019m sure AI is fantastic at analyzing data and providing data, but connoisseurship is about contextualizing things. It\u2019s about thinking critically.\n <\/p>\n

\n \u201cI may go to a conservation scientist and ask them to analyze a pigment; I may ask a photographer for an infrared image,\u201d she added. \u201cBut thinking AI can substitute the educated, critical eye? That\u2019s a no-no for me.\u201d\n <\/p>\n

\n Combining machine learning, deep neural networks and computer vision algorithms, Art Recognition\u2019s approach can, in theory, be adapted to any painter with a big enough back catalog. To date, the company has produced models for more than 200 artists.\n <\/p>\n

\n In each, the AI is trained on two photographic datasets: A \u201cpositive\u201d one, containing images of undisputed (or widely accepted) paintings by the artist in question, and a \u201cnegative\u201d one comprising similar, but inauthentic, works. The latter group may include known forgeries, copies by students or admirers \u2014 like Caravaggio\u2019s 17th-century stylistic followers, known as the \u201cCaravaggisti\u201d \u2014 and even AI-generated images, created in the artist\u2019s style.\n <\/p>\n

\n Having a \u201chigh degree of similarity\u201d between the two datasets is crucial, said Art Recognition\u2019s co-founder and CEO, Carina Popovici, on a call from Switzerland. \u201cWe really want the AI to learn the difference between Caravaggio and an imitator of Caravaggio \u2014 the difference between a Rubens and an almost-identical painting created in his workshop by an apprentice.\u201d\n <\/p>\n

\"Art<\/p>\n

\n To prevent bias, both training datasets feature a comparable balance of subjects and genres, such as portraits, landscapes, still lifes or religious scenes. Developers include augmented versions of each image, mimicking different lighting conditions and flipping or rotating high-resolution photos of the paintings, to expose the models to different spatial configurations. Images are also divided into small squares, or \u201cpatches,\u201d that force AI to consider the artwork\u2019s characteristics in new ways. Looking at a smaller, quieter part of a painting in isolation may help it learn fine brushstrokes, for instance, while a zoomed-out view could teach it about composition or color.\n <\/p>\n

\n But Popovici admits it is not always clear how the models reach their conclusions. \u201cWe can speculate, but we don\u2019t really, really know for sure,\u201d she said, adding there will be \u201csome types of patterns that AI can see better than humans.\u201d\n <\/p>\n

\n \u201cPeople really want a level of certainty that probably doesn\u2019t exist, whether it\u2019s a machine or the human eye.\u201d\n <\/p>\n

\n Sharon Hecker, art historian and curator\n <\/p>\n

\n It takes up to a week to train the AI on the chosen artist, then another day or two to analyze an individual artwork. However, Popovici said the most time-consuming part is researching and building the datasets. Like any AI model, the output is only good as the data it is trained on (or \u201cgarbage in, garbage out,\u201d as the saying goes).\n <\/p>\n

\n For this reason, Popovici argues that Art Recognition does not sit apart from art history \u2014 it relies on it. The firm\u2019s datasets are, she said, \u201cthe product of scholarly expertise,\u201d built by in-house historians who have studied artists\u2019 biographies, catalogs and academic literature. She hopes AI can be a \u201ctool in the toolbox\u201d for experts, not a replacement for them.\n <\/p>\n

\n \u201cI think it\u2019s very counterproductive to be in a perpetual argument about who\u2019s right, the expert or the AI?\u201d she said. \u201cWe don\u2019t want to be their enemy.\u201d\n <\/p>\n

\n The experts CNN spoke to were, however, less conciliatory about a technology that expresses conclusions without explaining how it reached them.\n <\/p>\n

\n \u201cAuthentication is rarely just about surface style,\u201d said art historian and curator Sharon Hecker. \u201cIt involves a lot of art-historical context. You have to know about the workshop, practices, materials, the condition of the work, the restoration history and how a particular artist\u2019s work evolved over time.\n <\/p>\n

\n \u201cIf you think about patterns of brushstrokes, who\u2019s to say an artist didn\u2019t change styles one day \u2014 they woke up and worked in a different style?\u201d she added. \u201cArtists are just not that predictable, and that unpredictability is part of the beauty of art. So, AI being trained to recognize a consistent style may miss a lot of these nuances.\u201d\n <\/p>\n

\n Among Hecker\u2019s primary concerns about operations like Art Recognition (whose customers frequently include owners looking to sell under-researched paintings inherited from deceased relatives) is their lack of transparency. Without full access to the companies\u2019 commercially sensitive models and datasets, researchers are unable to check AI\u2019s workings, she said. \u201cAny science has to be independently replicable,\u201d Hecker added. \u201cI would have to be able to replicate the study, using the exact same dataset, and come back with the same result.\u201d\n <\/p>\n

\n AI studies may not always even agree with one another. In 2023, a disputed Raphael painting, the \u201cde Br\u00e9cy Tondo Madonna,\u201d was subjected to two analyses that came to very different conclusions. One, by researchers at the UK\u2019s University of Bradford, used AI-assisted facial recognition to compare the painting to Raphael\u2019s \u201cSistine Madonna,\u201d finding both were \u201cundoubtedly<\/a>\u201d by the same artist. Art Recognition\u2019s model meanwhile found an 85% chance that the painting was not by the Renaissance artist.\n <\/p>\n

\"Researchers<\/p>\n

\n In response to the resulting debate, then dubbed \u201cthe battle of the AIs,\u201d Art Recognition argued<\/a> that finding facial similarity between two Renaissance paintings was \u201chardly surprising\u201d and should not constitute an attribution claim. The company then made its Raphael dataset public, saying transparency was \u201cnot a public relations gesture\u201d but a \u201cprerequisite for credibility.\u201d\n <\/p>\n

\n In the art market, credibility is everything. As such, the question facing AI researchers is not just whether their methodologies are sound \u2014 it\u2019s whether anyone is willing to believe, or even listen, to them.\n <\/p>\n

\n Determining if a painting is \u201cauthentic\u201d is often a matter of broad, rather than absolute, consensus. Historical paintings don\u2019t come with certificates; experts are increasingly hesitant to express complete certainty, for fear of litigation or reputational damage should they be proven wrong. For similar reasons, many of the foundations and artists\u2019 estates once considered the ultimate authorities (like the Keith Haring Foundation and the Andy Warhol Foundation for the Visual Arts) have ceased offering authentication services.\n <\/p>\n

\n \u201cI think there\u2019s a problem in the word \u2018authentication,\u2019\u201d said Hecker. \u201cPeople really want a level of certainty that probably doesn\u2019t exist, whether it\u2019s a machine or the human eye.\u201d\n <\/p>\n

\n Instead, buyers must rely on trust \u2014 in museums, auction houses and the market itself. After all, someone\u2019s willingness to spend millions of dollars on a painting can speak volumes, given that a downgraded attribution can knock numerous zeroes off an artwork\u2019s value.\n <\/p>\n

\"During<\/p>\n

\n For Art Recognition, however, a landmark moment arrived in late 2024 when a Swiss auction house used the company\u2019s research to back the sale of three artworks. Among them was a watercolor, ostensibly by Russian artist Marianne von Werefkin, that was otherwise short of evidence supporting the attribution, according to Popovici.\n <\/p>\n

\n \u201cIt was a big moment, because it showed that this is not just an academic tool, but the kind of product that can really make an impact on the market,\u201d she said, calling the auction a \u201cturning point.\u201d\n <\/p>\n

\n Whether AI research will be embraced by bigger auction houses, or used to back more valuable sales (the Von Werefkin watercolor fetched a modest 15,000 Swiss francs, or $19,600), is another matter altogether. So, too is the question of whether museums will ever take the findings seriously. Although Art Recognition has collaborated with institutions like the Kunsthaus Z\u00fcrich, Popovici concedes that galleries have little incentive to embrace technology that might cast doubt over their collections.\n <\/p>\n

\n She said London\u2019s National Gallery \u201cdidn\u2019t want to talk\u201d to her about the firm\u2019s findings on \u201cSamson and Delilah.\u201d In the years since, further evidence questioning the attribution has emerged, though the museum has consistently defended its longstanding attribution to Rubens. In a statement emailed to CNN, the National Gallery spokesperson said that an \u201cextensive study, conducted by our curatorial and scientific teams using the latest imaging and analytical techniques, provides compelling evidence in support of the painting\u2019s authorship,\u201d adding: \u201cNot one single Rubens specialist has doubted that the picture is by Rubens.\u201d\n <\/p>\n

\"London's<\/p>\n

\n Yet, despite experts\u2019 distrust, might AI at least help start conversations, even if it\u2019s not their final word? Popovici expressed hope that her company\u2019s research could help \u201cunlock\u201d paintings that are \u201cotherwise just sitting in a basement somewhere.\u201d And art historian Hecker acknowledged that AI \u201ccould flag an issue\u201d that is then investigated using conventional research, caveating that she \u201cwould be more comfortable using a university-based laboratory that doesn\u2019t have a large commercial interest.\u201d\n <\/p>\n

\n The technology might also be used to expose criminal activity or help vendors weed out fakes, argued Popovici, who was inspired to start Art Recognition after learning about the case of German master forger Wolfgang Beltracchi, who was imprisoned in 2012 for a multi-million-dollar fraud<\/a> that fooled buyers, galleries and auctioneers.\n <\/p>\n

\n \u201cExperts were checking every single one of those paintings, and they all gave the green light,\u201d she said, arguing that the human eye \u201cinherently makes mistakes.\u201d Popovici has since used the company\u2019s models to analyze numerous Beltracchi paintings, which were created in the style of various deceased European artists, to see if they would have successfully exposed his deception.\n <\/p>\n

\n \u201cEverything came out as fake,\u201d she said. \u201cAnd with very high probabilities.\u201d\n <\/p>\n","protected":false},"excerpt":{"rendered":"To the untrained eye, there is very little difference between the three known versions of \u201cThe Lute Player.\u201d…\n","protected":false},"author":2,"featured_media":322810,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[31],"tags":[442,498,499,500,501,156,111,139,69],"class_list":["post-322809","post","type-post","status-publish","format-standard","has-post-thumbnail","category-arts-and-design","tag-arts","tag-arts-and-design","tag-artsanddesign","tag-artsdesign","tag-design","tag-entertainment","tag-new-zealand","tag-newzealand","tag-nz"],"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/322809","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/comments?post=322809"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/posts\/322809\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media\/322810"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/media?parent=322809"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/categories?post=322809"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/nz\/wp-json\/wp\/v2\/tags?post=322809"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}