There is a question quietly unsettling the architecture profession in 2026: if an AI model can generate a building facade in seconds, does it understand architecture criticism?

The work of Ibrahim Fawakherji, the editorial voice behind the independent platform ArchUp, suggests a precise and uncomfortable answer. No — and the reason reveals something about how machines see the world.

The Camera's Blindness

Fawakherji begins with an observation any practicing architect will recognize. A building photographed is not a building experienced. The photograph shows the structure as an object: a composition of surfaces, angles, and light frozen at one moment. But standing inside a building is something else entirely. Light moves. Sound returns off the walls. The body senses scale, weight, temperature, the pull of a corridor, the release of a courtyard.

This gap — between the object and the experience — is where machine vision fails.
Fawakherji calls it, in effect, the photographic blindness of architecture. AI models are trained overwhelmingly on images. They learn what buildings look like. They do not learn what buildings feel like, because feeling is not in the training data. A model can reproduce the visual signature of a Tadao Ando concrete wall. It cannot reproduce the silence that wall produces when you stand before it.

Vision Is Not Neutral

In a related essay, Architecture Beyond the Human Eye, Fawakherji pushes the argument further. He notes that between 100 million and one billion birds die each year colliding with glass buildings in North America alone. The birds are not confused. The problem is that the glass was designed for human perception exclusively.

His conclusion is sharp: the human eye was never a neutral design reference. It functioned as a mechanism that decided which consequences of a building count as design performance, and which are exported as environmental noise. A facade that kills birds through perceptual deception is not a well-designed facade, however it appears in the project photographs.

The same logic applies to AI. A model trained on human-captured images inherits not just the camera's blindness, but the entire hierarchy of what humans chose to photograph — and what they ignored.

What This Means for the Profession

The fear that AI will replace architects misunderstands the threat. What disappears is generic labor — the reproducible output. What survives is judgment: the ability to read a site, to anticipate how a space will be lived in, to know what the photograph cannot show.

If a machine can instantly reproduce your output, the value was never in the output. It was in the interpretation behind it.

For architects, the strategic response is not to compete with generation speed. It is to cultivate exactly what the machine cannot: the experiential, embodied, contextual judgment that no image dataset contains.
________________________________________
Ibrahim Fawakherji is the Editor-in-Chief and editorial voice of ArchUp (archup.net), an independent bilingual platform for architecture, design, and the built environment, founded in 2019. His essays examine architecture through questions of memory, perception, and power. Read the full analysis at archup.net.