How do machines learn what age is supposed to look like?
This provocation explores how AI systems both estimate and generate age through two connected experiments. The first examines how an age estimation model predicts age under controlled changes in hairstyle, eyewear, and clothing. The second investigates how image generation models visualise “22-year-old,” “35-year-old,” and “50-year-old” East Asian women wearing glasses.
Across both experiments, age emerges not as a fixed biological category, but as a relational visual construction. In the estimation experiment, predicted age shifts according to facial framing, contour visibility, and the interaction between hairstyle and eyewear. In the generative experiment, age is distributed across interiors, clothing, accessories, and atmosphere, with older women associated with increasingly ordered and middle-class environments.
Rather than simply recognising age, AI systems participate in constructing visual expectations of what age should look like.
