Artificial intelligence is increasingly being used to help determine whether valuable works of art are real or fake. According to Guardianthe start-up Art Recognition from Zurich has just assigned, thanks to a technology it developed, a work entitled Portrait of a Woman (Gabrielle) to the artist Pierre-Auguste Renoir. The painting, which belongs to a Swiss collector, was denied authorship by the Wildenstein Plattner Institute – one of two institutes that publish a catalog raisonné of known works by Renoir.
To do this, Art Recognition used photographic reproductions of 206 authenticated paintings by the French Impressionist to teach the algorithm the painter’s style, characterized by broken brushstrokes and bold combinations of complementary colors. To increase accuracy, the company also broke the images into small pieces and showed them to the algorithm, while training it on a selection of paintings by similarly styled artists active around the same time as Renoir. Based on this assessment, the algorithm concluded that there was an 80.58% chance that Portrait of a Woman (Gabrielle) it was painted by Renoir.
Dr. Carina Popovici, CEO of Art Recognition, believes this ability to quantify certainty is important. Speaking at a meeting on the use of technology in the art business at the Art Loss Register in London, she said: “Art owners are often told by connoisseurs that it is their ‘feeling’ or ‘intuition’ that determines whether a painting is authentic or not, which can be very frustrating. They really appreciate us being more specific”🇧🇷
Encouraged by this result, the painting’s owner approached another group of specialists, Dauberville & Archives Bernheim-Jeune, which publishes its own raisonné catalog of Renoir’s works. After requesting a scientific analysis of the painting’s pigments, they too concluded that it was an authentic Renoir.
But experts point out that the accuracy of the algorithm depends a lot on the quality of the paints it was trained with. In the presence of a forgery or copy, or if the areas contain retouching, the degree of certainty would be much lower. British art historian Bendor Grosvenor, known for his rediscoveries of master paintings, worries that these technologies devalue the contribution of experts in assessing the authenticity of a work of art. “The technology is particularly weak due to its inability to take into account the condition of an ink. Many old master paintings are damaged and disfigured by layers of dirt and paint that make it difficult to distinguish between what is original and what is not.he also added.
Portrait of a Woman (Gabrielle)attributed with an 80% chance to Pierre-Auguste Renoir by an artificial intelligence developed by Art Recognition.
© Art Recognition
Carina Popovici agrees that the quality of the training data is important, ensuring that her company only uses photographs of authentic works of art.
“We sincerely want to give the possibility [aux connaisseurs] use this system to help them make a decision, perhaps in cases where they are not so sure. But for that to happen, they need to be open to this technology.” she explained. So far, Art Recognition has trained its AI to recognize 300 artists, mostly French Impressionists and Old Masters. Already a month ago, the start-up announced that it had determined that the only Titian should be made in Switzerland – a work entitled Night landscape with couple (1518-1520), belonging to the Kunsthaus in Zurich – probably not painted by the Venetian artist.
Julian Radcliffe, president of the Art Loss Register, which manages the world’s largest private database of stolen art, antiques and collectibles, agrees: “Artificial intelligence […] it must be combined with the experience of the artist’s expert connoisseurs, with a science of pigment analysis and with the determination of the origin of the work. Its advantage lies in the ability to give yes or no answers. […] and constantly improve, but your work must be interpreted by a human who must ask the right question.”🇧🇷