For more than a decade, Christopher Kulendran Thomas has trained his own neural networks, building them from scratch before “AI art” was a widely used. He feeds images of Sri Lankan painters across generations into his models, then uses them to produce new compositions that he and his Berlin studio hand-paint onto canvas. The titles are simply the auto-generated file names of the source PNGs.

“The network analyzes the aggregated patterns behind their work,” the artist told me, slowly, so I could understand, at Gagosian’s Upper East Side location. “It learns a way of seeing.”

The result is painting that metabolizes colonial art history—the very visual grammar that redefined Sri Lanka after his Tamil family fled the civil war. “These are the only works that are historically specific,” he said, describe them as “the only ones where I’m explicit about what they are painted off.”