The old saying goes that four eyes see better than two. The million dollar question that is beginning to hover over the art world is what happens when those “eyes” are those of an artificial intelligence. As AI gains scope and capabilities, so do the cases in which an algorithm breaks into the debate about the authorship of controversial works, sometimes with judgments that clash with those of scholars. Sometimes his vision serves to expand the catalog. Others simply to cut it.
The last example comes from the hand of the very raphael.
Whose work is this? Since the mid-1990s, lovers of Renaissance art have been looking an old wooden panel painted with that question in mind: “Who is its author?” For its owners, its origin is as clear as the content of the work, which shows the Virgin Mary with the baby Jesus on her lap, both accompanied by Elizabeth and her son, John the Baptist: the piece —they maintain— is by Rafael Sanzio, one of the great painters of high renaissance.
The problem is that not everyone shares your opinion. After examining it, David Pollack, a specialist at the prestigious Sotheby’sconcluded that the work actually belongs to Antonio del Ceraiolo, a less appreciated contemporary of Raphael who used a style similar to that of the Urbino master. Another Renaissance art expert who has had the opportunity to examine the piece up close is Larry Silver of the University of Pennsylvania, who has concluded that it was, at the very least, painted in Florence by someone in Raphael’s close circle.
And then came the AI. One of the latest voices that has sounded in this long and heated debate is that of AI, more specifically that of Art Recognitiona Swiss company that uses the machine learning to analyze works. And his sentence is resounding. According to keep it up The Wall Street Journal, has concluded that there is a 97% probability that the two main faces of the composition, those of Jesus and Mary, were painted by Raphael himself. The rest could have been completed by one of his assistants, a relatively frequent practice at the time.
The algorithm was trained on examples of Raphael’s digitized works and forgeries to be able to accurately recognize the artist’s “imprint.” To get an idea of the strength of the result of Art Recognition, it is useful to handle two percentages: less than 10% of the works of clients that have been submitted to their scrutiny have yielded a positive identification with a level of probability that exceeds 95%. The alleged Raphael reached 97%.
And why so much trouble? For a historical, artistic question… and of course economic. Raphael is one of the great masters of the Italian Renaissance, but he was not especially prolific. The creator of ‘The school of Athens’ He left less than 200 works, which explains why the last one to be auctioned, ‘Cabeza de musa’, reached a whopping 48 million dollars at Christie’s. Its owners and the group of investors who support the theory of authorship are so convinced of the relevance of the piece that they have invested hundreds of thousands of dollars in it.
The first to bet on the Raphaelian origin of the panel was anthony ayers, cabinetmaker and amateur artist, now deceased, who bought it in 1995 from an antique shop in the British countryside. So sure was he of her origin that he convinced a group of friends to come up with the $30,000 it cost. Over time, that hunch of his mobilized a group of 40 investors who have raised more than $500,000 to pay for the identification studies.
Is this a unique case? Absolutely. AI has already contributed on other occasions to heating up the debate that revolves around the authorship of certain works, pieces whose attribution is particularly complex and which have been in the midst of an even more complicated academic dispute for several decades. Among the most recent cases there is another table attributed to Raphael: the one known as de Brecy Tondoa circular work, 95 cm in diameter, in which the Virgin Mary and baby Jesus appear.
In 1981 George L. Winward noticed its striking resemblance to the ‘Sistine Madonna’ painted —in this case without a doubt— by the man from Urbino and decided to buy the piece. His case is very similar to that of the Ayers panel. Its owner was sure of its origin. Scholars, not so much. To clear up doubts, experts from the universities of Bradford and Nottingham used a facial recognition tool that concluded that both Mary and Jesus resembled the figures of the ‘Sistine Madonna’ by more than 85%, an almost identical similarity.
Add and continue on the list. They are not the only cases. Not all of them revolve around Rafael. another recent example in which an AI helped us expand the catalog of a master has Art Recognition and Renoir as protagonists. Faced with doubts about whether or not the ‘Gabrielle’ canvas was the work of the French impressionist master, last year the Swiss company was turned to, which subjected it to a process similar to the one it has now used with the alleged Raphael panel. His conclusion is that there is an 80.58% chance that the portrait came from Renoir’s palette.
What AI gives, AI takes away. Not all AI identifications are positive. At times his judgment has helped trim the catalogs of great masters. Occurred a few months ago with the painting ‘Landscape at sunset with a couple’, attributed to Titian and kept in the Kunsthaus Zürich. By putting it to the test of their algorithm, Art Recognition concluded that there is 80% chance that the part has been wrongly assigned.
The same AI has ensured that another work kept in the National Gallery as a possible painting by Peter Paul Rubens, ‘Samson and Delilah’, probably not Baroque artist’s work. His conclusion is of course overwhelming: his algorithm sees a probability of 91% that the work is not authentic.
One more tool or the final judgement? Here is a fundamental question: What credibility should be given to AI? Is it just another tool or should we trust its criteria more than that of scholars? “I worry that people see computers as flawless, but it’s just another tool, not irrefutable proof,” points to The Wall Street Journal Karen Thomas, conservator of paintings. Algorithms can learn from hundreds, thousands of examples, and find similarities; but the expert wonders if they can assess the different nuances related to a work that academics take into account.
One thing is clear: there is a field for study. Only David Pollack, of Sotheby’s, responds each year to more than a thousand of requests from collectors who want to identify anonymous works… and hopefully discover that it is a Raphael.
Cover image: Flaget Madonna
Also in Xataka: Creative artificial intelligences are going to kill art again. It doesn’t matter in the slightest
