How AI helps auction houses to verify long lost art works
Yet three weeks ago, the Van Gogh Museum in Amsterdam — the ultimate authority on the Dutch master — declined to authenticate it, leaving those involved bewildered.
A growing contingent of art authenticators argues that such disputes could soon be a thing of the past. They are turning to artificial intelligence to analyse artworks, feeding high-resolution images into systems that assess brushstrokes, composition and pigment composition with unprecedented precision.
• How do you prove your $50 painting is an authentic Van Gogh?
“If you’re looking at a Van Gogh, there are distinct brushstroke styles and compositional elements that AI can detect by analysing thousands of images,” said Hanoch Sheps, a lawyer with the New York-based law firm Mazzola Lindstrom, which specialises in fine art. “It can spot patterns that you and I can’t perceive, helping to determine authenticity.”
Denis Moiseev, founder of Hephaestus Analytical — the first firm to collaborate with an art insurance company to offer third-party guarantees — has worked extensively on Van Goghs, including paintings later authenticated. “There is definitely a distinct difference between how we would approach the Van Gogh case compared to other companies,” he said.
His firm employs traditional authentication methods — carbon dating, pigment analysis and provenance research — but enhances them with AI. “It allows us to use data to support attribution,” he added. “AI approaches this process in a slightly different way, offering another layer of verification.”
Founded in London in 2018, Hephaestus Analytical is expanding rapidly. Last month, it acquired ArtDiscovery, the firm in New York behind the scientific analysis of Salvator Mundi, the Leonardo da Vinci painting that became the most expensive artwork ever sold when it fetched $450 million in 2017.
Dekking left Sotheby’s to found Artory, a company that uses AI valuation and blockchain technology to securely store an artwork’s complete history.
Elizabeth von Habsburg, managing director of America’s leading art appraisal firm, Winston Art Group, believes AI is “an extremely useful tool” but warns that its reliability depends on data integrity. “It has to be used correctly, because if the dataset is flawed, the results will be skewed,” she said.
She pointed to a cautionary tale from June, when AI valued a Pissarro print at $530, far below expert estimates of $30,000. “Sometimes it’s not deliberate. It’s just bias,” she said. “The AI might be trained on inaccurate data or fail to account for an artist’s experimental periods. It can be manipulated, but it can also be misused unknowingly.”
Von Habsburg also sees AI as a powerful tool for forecasting art market trends. “We hear from so many people who want to invest in art,” she said. “AI could be instrumental in identifying trends for specific artists or market sectors.”
In November, a Swiss auction house made history by selling the first painting authenticated solely by AI, without human input. Germann Auctionhouse in Zurich sold a watercolour by the Russian artist Marianne von Werefkin for just under $17,000, almost double its estimate.
Most experts insist, however, that human authentication will remain essential.
James Callahan, an expert in Asian art at the Massachusetts auction house Tremont and a regular appraiser on the British TV programme Antiques Roadshow, warns that AI could become “a perfect venue for fraud … All of a sudden, you’re handing over your entire business to AI. The computers can learn to produce very, very, very convincing fakes — if you give them all the information.”