AI Fails Again
March 19 2025
Picture: The Art Newspaper
Posted by Adam Busiakiewicz:
The Art Newspaper has run another curious AI art authentication story, this time in relation to a version of Rubens' Diana discovered by Actaeon (pictured) which was unveiled by the Zurich-based Art Recognition at the Art Business Conference at Tefaf Maastricht (do get in touch if any readers were in attendance). The story is particularly complicated to the survival of a badly damaged and reduced 'fragment' in the Museum Boijmans van Beuningen in Rotterdam, which may or may not have been the original from which many copies were produced. The presentation was delivered by Dr Carina Popovici, who you might remember was behind the reoccuring claims about the NG's Rubens back in 2021.
Most intriguingly:
After their latest investigation, Popovici and Art Recognition concluded that the painting, while not the original The Bath of Diana, could be by Rubens and his studio. “It was an authenticity evaluation not a confirmation,” Popovici says. “We concluded that it is partially by Rubens. Our AI cannot know who did the rest but one possible interpretation would be the [artist’s] workshop contribution.”
Click into the story to read the further claims, which include the ways AI is now hoping to show which parts are and are not by the hand of an artist such as Rubens.
____________
I find it rather mysterious that AI is not happy with the National Gallery's Samson and Delilah, yet is perfectly happy (it seems) with duds like this.
Update - Bendor adds: Yes. Oh dear.
