Artificial intelligence can help combat forgery in art. This is demonstrated by Art Recognition, a company specialized in this technology that has found more than 40 counterfeit paintings, including an alleged work by Monet and another by Renoir, which were trying to be sold on eBay.
Its manager, Carina Popovici, explained to EFE that, “despite the perception that AI represents a threat to art connoisseurs, that is far from the truth, and on the contrary, it is dedicated to complementing art experts. art and traditional methods of authentication.
Popovici created Art Recognition in Adliswil, outside Zurich, in 2019 and already works with art galleries, auction houses and collectors. “We use artificial intelligence based on a convolutional neural network, which can learn the main characteristics of an artist from authentic photographs and subsequently recognize them in a new work of art not previously seen,” explains Popovici, also a designer of the software.
The AI ??first has to go through a training phase, in which it examines, between one and three days, photographs of all the artist’s known works and negative examples of forgeries if any. One of the main characteristics that helps the software verify whether or not the work is by the author in question are the brushstrokes used on the canvas, but other elements are also analyzed, such as the color palette used, or the level of composition.
“Traditionally, art authentication has depended on the experience and judgment of experts, who are still the dominant authority in the sector, but this approach is subjective, and opens the possibility of human error,” emphasizes Popovici, who acknowledges that The firm also has its limitations, such as the inability to verify some works, such as one by Vermeer, given that only about 36 exist, and it cannot effectively distinguish highly restored works, such as the Salvator Mundi.
It is not easy either to verify copies of Jackson Pollock’s splatter-based works, or of Modigliani’s originals. In any case, Popovici offers the work of his company to “contribute to a more transparent art market.” “We train our algorithm to differentiate authentic pieces from AI-generated fakes, also feeding the program with digital fakes,” he emphasizes.
On the other hand, he adds, this technology can not only help in the fight against imitations but also already participates in the art world through interactive experiences in museums and galleries, or in data analysis of market trends. “In addition, generative AI models are revolutionizing the way artists explore creativity, and I am sure we will see even more spectacular AI-assisted works in the future.”
Art Recognition has been tested on paintings in which there are extensive debates about their authorship, such as a Samson and Delilah exhibited in the National Gallery in London and attributed by some experts to Rubens. The software designed by Popovici considered that the work was not by the Flemish master, while in a similar test with a Van Gogh self-portrait treasured by the National Museum of Oslo it did declare its authenticity.
