The AI artwork "Attention Without Understanding" has been selected for the CVPR 2026 Art Gallery, a premier conference on computer vision and pattern recognition. The work will be presented on-site in Denver and in the online art gallery.
The triptych asks what a vision model sees in a single one-line drawing. Self-attention maps from DINOv2 (ViT-B/14) are read across twelve independent heads and rendered as three panels — The Form, The Conflict, The Gaze. The model attends with mathematical precision, isolating edges and curvature, yet never reaches the meaning a line carries. Like the inhabitants of Plato's Cave, it perceives only the shadows of geometry, separated from the Form by a veil of its own making.
A true delight to have our work selected, and an exciting moment to bring conceptual minimalism into a computer vision venue, shown alongside so many artists exploring the edges of machine perception.
More here: CVPR 2026 Art Gallery