

CONCEPTUAL PROGRAM 001
How do human aesthetic preferences train machine perception—and how do algorithmic reward models quietly re-condition what human eyes expect to see?
01 / RESEARCH THESIS
Modern image generation and camera computation rely on preference-optimization loops: thousands of human annotators rank pairs of images, teaching a reward model which compositions, skin textures, lighting ratios, and facial symmetries feel "pleasing."
Once deployed at planetary scale, this statistical mean ceases to be descriptive and becomes prescriptive. Shadows that conceal too much are lifted; postures that feel strained are straightened; grain that interrupts commercial legibility is denoised.
001 / TRAINING BEAUTY isolates the moment where human visual instinct collides with algorithmic normalization—asking visitors to examine their own perceptual habits across three diagnostic axes.
02 / INTERACTIVE PERCEPTUAL PROBE
Examine how an unretouched human photographic study and an autonomous machine artwork negotiate the same aesthetic territory—and test which perceptual signal you prioritize.



