In 1995 Anthony Ayers, an American art lover, discovered what he believed to be an authentic painting by the Renaissance painter Rafael in an antique shop in England and bought it with the help of friends for $30,000.

For decades he struggled to prove the authenticity of “Flaget Madonna,” as the painting is named after the original owner, but experts disagreed.

He eventually commissioned the Zurich start-up “Art Recognition» to decide with the help of Artificial Intelligence.

Her conclusion? 97% of the depicted faces of Jesus and the Virgin Mary were painted by Raphael, while the rest of the painting was probably crafted by one of his assistants.

But most experts disagree. The experts of the London auction houses Sotheby’s and Christie’s consider it to be a painting by a modern imitator of Raphael.

Larry Silver, a Renaissance expert at the University of Pennsylvania, believes that it was created by another painter, but one of Raphael’s inner circle in Florence.

According to the American art conservator Karin Thomasthe AI ​​has various problems, such as with worn paint and aging of construction materials.

The experts of “Art Analysis & Research” see many similarities with Raphael, but do not rule out that the painting was done by other painters.

However, it comes from his atelier, because the pigment Sandarachwas known only to a few Renaissance painters, including Raphael.

Only the Patricia Tratti Couhill, Renaissance art historian at Siena College in New York, attributes the painting to the Renaissance painter. Artificial Intelligence is not yet the final arbiter of works of art, something that may change due to its rapid development. Until then, digital and traditional controls must coincide.

However, the buyer of the painting will never know the definitive answer because he died earlier this year. And the work will remain in a Chicago bank vault, away from the public eye.

The idea for one art authentication system with the help of technology was born by accident.

“But why isn’t there an artificial intelligence program to determine if some works of art are fakes?” Karina Popovici wondered three years ago when a friend of hers, an art historian, pointed out to her that there are many fake works scattered on the art market.

Thus, in 2019, the Romanian physicist decided to establish in Zurich the start-up company “Art Recognition” in collaboration with the University of Tilburg in the Netherlands within the framework of the European program “Horizon 2020”.

More than 400 authenticity opinions have already been made for private individuals, galleries, art galleries and auction houses. The very simple ones cost just 500 Swiss francs (about 498 euros), while the complex ones from 1,000 to 5,000 francs and in demanding cases up to 10,000 francs.

The aspiring physicist explained to the Zurich newspaper Tagesanzeiger the method of proving the authenticity of a work: “For three years we have been feeding the computer program with as many high-resolution photographs of an artist’s original works as possible that are considered genuine.

Then we “train” the AI ​​system for 300 painters already in our archive, especially impressionists and expressionists, i.e. works whose technique, anatomical details, the most important of these strokes, painting style and so on.

The method is called “artificial neural network”. Once known, a data set is produced against which comparisons can be made with works of questionable authenticity. “Training” the system takes about three days, but testing an individual project takes just five minutes.”

Art Recognition penetrates a field that has so far been dominated by experts and private or state institutions, and the works in question are examined on the one hand from an art-historical and stylistic point of view, on the other with technical means such as the radiocarbon method, with X-rays and chemical colors analyses. “Usually checking the authenticity of a painting takes a long time, but Art Recognition’s method ensures speed and reliability of up to over 90%,” Popovici told the Neue Züricher Zeitung (NZZ). He admitted, of course, that “there is no 100% guarantee”, but added that “the program is the first filter for the auction house”.

However, this particular Artificial Intelligence technology already has spectacular results:

Art Recognition concluded that the painting “Sampson and Dalida” by Peter Paul Rubenswho was bought by National Gallery London at a record price in 1980, 91% not genuine.

Popovici had told the Tagesanzeiger last November: “We examined it and concluded that the painting is not by Rubens. A few hours later, a Guardian reporter, completely coincidentally, asked me about this painting and I told her about our finding. The English newspaper article about the “impressive result” made our small business known all over the world in an instant. This brought in new orders.”

In the Spring of 2019 Art Museum (Kunsthaus) Zurich get it “Evening landscape with a couple” by Titian (Tiziano), which caused a sensation since it was the only Swiss museum that owned his work.

In the same year, however, doubts were raised about its authenticity. “Art Recognition” came to the conclusion that “by 80% it is not Titian’s” and informed the Museum, because “it has a responsibility for transparency towards the public” as Popovici told NZZ.

The Museum itself responded to a relevant question from the Zurich newspaper that “it took into account the evaluation in question and the work is now exhibited with the clarification that it is “attributed to Titian”.

Art Recognition’s artificial intelligence system can handle even the most controversial works of art. One such case is Vincent van Gogh’s “Self-Portrait” in the Norwegian National Museum in Oslo. In 2019, a leading expert on the painter asked her for an analysis of his works, including “Self-Portrait”. So, according to the startup, the project can be considered 97% authentic.

In early 2020, the Amsterdam Van Gogh Museum announced its authenticity first “Self Portrait” based on his own research.

“We were unaware of the parallel investigation. We analyzed this chart weeks before her announcement. It’s unfortunate that we didn’t publish our results earlier, but it’s also a great confirmation of the power of our algorithm” was Art Recognition’s reaction.