Proceedings of the National Academy of Sciences · 24 July 2026 · From issue 11
Context-aware multimodal AI navigates hidden pathways in five centuries of art evolution
In a computer model, a painting's subject matches its art era better than its colors do.
Jin Kim, Byunghwee Lee et al.
A computer program studied 500 years of Western paintings. It learned two things about each painting: how it looks, like colors and shapes, and what it is about, like its subject or meaning. The subject matter matched art history periods and painters better than colors did. The program could also blend future art styles into old paintings, and the results matched real historical style changes.
Integrity screen: passed (3 checks) Checked 26 July 2026. Read the source
The plain-language reading of this paper is in the full issue, for subscribers.
Read the full issue