In 1624, a decade on from El Greco’s death, The Baptism of Christ, a towering canvas of Jesus kneeling beneath a radiant God and a swirl of angels, was fixed as the altarpiece inside the church of Hospital Tavera in Toledo. It was among the painter’s last commissions and the interlude, paired with some stylistic irregularities, have long led scholars to believe that the painting was largely the work of El Greco’s son, Jorge Manuel, along with the help of trusted hands from the Spanish Renaissance master’s workshop.
New research that used artificial intelligence to analyze The Baptism of Christ at the microscopic level, however, has suggested that El Greco painted the majority of the painting himself. The findings, led by researchers from Western Reserve University in Ohio, were published in Science Advances on April 17.
(A) Network diagram showing the maximum modularity (Q) partition. Four communities are present. (B) The communities mapped onto the corresponding regions of the painting that were analyzed. Scanned areas of the painting are outlined by dashed white lines. Courtesy of Science Advances.
The team developed a machine-learning model known as Patch that begins by scanning the surface of a painting using high-resolution 3D imaging to capture the peaks and hollows left behind by a painter’s brush. This map of a painting’s texture can then be compared centimeter-by-centimeter. If Patch (short for: pairwise assignment training for classifying heterogeneity) struggles to distinguish sections, they were likely created by the same artist.
El Greco, Christ on the Cross. Photo: Betteman via Getty Images.
After training Patch on 25 paintings by nine student artists, a process in which Patch exhibited “exceptional” classification performance, the researchers turned to El Greco. In addition to The Baptism of Christ, they showed the model Christ on the Cross (1600–10), a haunting depiction of Jesus nailed down and backgrounded by a vortex of furious clouds that is thought to have been painted solely by El Greco’s hand.
Patch passed its first test: asserting that Christ on the Cross was the work of a single artist. But contrary to expectations, when analyzing The Baptism of Christ the model found that there were underlying connections between parts of the painting that were previously thought to have been painted by Jorge Manuel or workshop painters. While there is a region at the bottom of the painting that is undeniably the work of later hands, the differences art historians have long perceived could be explained by El Greco experimenting with style, using different brushes, or his brushwork evidencing the effects of age.
Network diagram showing the maximum modularity (Q) partition. Two communities are identified, although this is the minimum number of communities discoverable using a maximum-modularity partition. (B) The communities mapped onto the corresponding regions of the painting that were analyzed. Courtesy of Science Advances.
Given the ubiquity of Renaissance painters employing large studios of apprentices, Patch has the potential to unpick who painted what in works that most often lack detailed records. Until now, art historians have largely examined brushstroke style and more overt visual clues to determine whether a painting should be attributed to an artist or more broadly to their workshop. Unlike other models, Patch doesn’t require any external training data (i.e. confirmed examples of work by a painter or their assistant), known in A.I. lingo as “ground truth.”
“But we still have work to do to get it to the point where it can conclusively identify the elements of what we’re calling artistic practice regimes,” lead author Andrew Van Horn said over email. “Once we get there, we could figure out what causes the interconnections in The Baptism. We may even be able to, for example, track an artist through their apprenticeship(s) to their own workshop.”