AI better than art experts? This is the question posed recently by an artificial intelligence that would have authenticated the Lighton Larking player as an original of the Caravaggio. She also questioned the authenticity of another version of the painting, that of the Wildenstein collection, long considered autograph.
Three versions of the work of the caravan exist: that of the hermitage, unanimously recognized as authentic, that of badminton and that of Wildenstein. The version of Badminton had been studied by Sotheby’s and experts from the Metropolitan Museum of Art during a sale in 1969, and described as a copy, probably made by Carlo Magnone, who had executed in 1642 a replica for Cardinal Francesco Maria del Monte. This painting was sold for 860 euros, before being acquired in 2001 by the art historian and Clovis Whitfield merchant for 80,000 euros, then cataloged as a copy of the Caravage circle.
Caravaggio (1571-1610), The lute playerversion of the hermitage in Saint Petersburg.
The AI model, developed by the Swiss company Art Recognition in collaboration with the University of Liverpool, awarded a probability rate of 85.7 % in favor of authenticity, result deemed “Very high” by its founder Carina Popovici. Trained on more than 200 works of the Caravaggio, the algorithm Analysis brush, palette, composition, technique and chiaroscuro. This expertise is reinforced by a textual source: the biography of Caravaggio by Giovanni Baglione (1642), describes a lute player closer to the version of Badminton than that of Wildenstein. The work of the work was also partly traced, painted for Cardinal Del Monte, then passed through the Barberini collection in the 1620s, then sold to the Duke of Beaufort in the 18th century, before staying several centuries in Badminton House, England.
The lute playerversion of the Wildenstein collection.
Several specialists have expressed themselves. Luthier David Van Edwards, president of the Lute Society, underlines inconsistencies in the instrument represented in the Wildenstein version, absent from the other two paintings. Lawyer and art lover William Audland KC considers his side that“It would be unfair to consider the Wildenstein version as an autograph and that of badminton as a poor quality copy”.
However, these conclusions are discussed. George Gordon, co -president of Sotheby’s Worldwide Old Master Paintings, recalls that the 1969 catalog was notice was “Reflect and complete” And that no recent research has, according to him, decisively modified the consensus of 2001. The very reliability of the models of AI remains in debate: in 2023, two algorithms had delivered contradictory results on the authenticity of a Raphael.
Clovis Whitfield wishes to entrust the table to a British institution, the work is currently exhibited in London. The collector also plans to devote a podcast, an academic article and a documentary about authentication.
