TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Judgment · AGE 15—17 · POSITIONAL · HELD
Heritage is not a prompt keyword
A keyword carries the look and none of the reasons the building was built.
Our position
Type "Chettinad house" or "Indo-Saracenic" into an image tool and you get a look: the colours, the columns, the arches, arranged the way the photographs usually arrange them. To the model a style is a pattern of pixels. The real building was several other things first. A Chettinad house was a way of building — lime plaster, Burma teak, a sequence of courtyards cut into a deep plan; an answer to a climate — the open court, the thick wall, the deep thinnai; a trade network — Nattukottai Chettiar merchants built with money made in Burma, Ceylon and Malaya and shipped the materials back along the same routes, teak, Italian marble, Belgian glass, so the house is a map of the Indian Ocean; and a social and ritual order — courtyards for weddings and funerals, a plan that ranks who enters where, a town grid and water tanks that belong to the same settlement. An Indo-Saracenic building was a colonial government's programme — British architects assembling Mughal and Hindu motifs onto a European plan for courthouses, railway stations and colleges, so that imperial rule would look native. The style exists because of that politics. A keyword carries none of it. Heritage is those reasons. The style is only what the reasons looked like on the day the photograph was taken.
Why we hold it
A student who learns heritage as a keyword will design heritage as a costume: a jali stuck on a glass box, a Chettinad colour scheme on a flat with no court. The costume does nothing about the sun, which is exactly what the original was built to handle. A student who learns the reasons can use them in a new building that looks nothing like the old one, and it will still work.
The strongest objection
Style is a legitimate analytical abstraction. Historians need categories, designers need shorthand, and search systems need labels. "Hoysala," "Art Deco," "Chettinad," "Indo-Saracenic" each carry real formal, temporal, geographic and political information, and a student who cannot use them cannot read a history book or a city. The error this card describes is not in the label. It is in treating the label as sufficient evidence of understanding — and a prompt keyword is no worse at that than a textbook chapter heading, only faster. Refuse the label and you lose the handle by which the reasons are found; a fifteen-year-old who has seen fifty generated "Chettinad" images may be more curious to visit Karaikudi, not less.
What would make us revise it
The objection is right about labels, and the card keeps them: the position is against the label as evidence, not against the label. We would revise the position itself if image tools began to return the logic with the look — a section through the court with the air path drawn, the tile named, the timber span given, a note on who paid and why — so that the keyword started to carry reasons; the card would then soften to "check what the keyword brought with it." We would also revise if students who began from generated styles went on to study the real buildings more often than students who began from books. We check both every edition.
Try it
Generate "a Chettinad house", or any heritage style from your own region. Then find one photograph of a real one, and a plan if you can. Write down three things the real building does that the image does not show: where the rain goes, where the air moves, who paid for it and why.
Take it to crit
When a student puts a heritage style on their board, ask them to name one reason the original was built that way — climate, construction or patronage — and to point to where that reason lives in their own scheme. If they can only name the look, the style is a keyword.
How it works
This card is the heritage version of PATTERN ≠ PRECEDENT on Foundations. There, a pattern of courtyards is not a studied building; here, a pattern of a style is not a tradition. An image model learned each style from photographs, and photographs record surfaces. What held the roof up, where the water went and who commissioned the house are not in the pixels, so they are not in the model. A 2023 study tested generative models on architectural styles and found them accurate on the much-photographed styles and prone to invention on the rest. Many Indian regional traditions are in the rest (UNEVEN STYLES on this map). The argument that a region's architecture is a response to climate, material and culture, not a set of motifs, is called critical regionalism. The Lab keeps its own teaching note on it, and Frampton's essay is the reading behind that note.
What this idea builds on
What this idea opens up
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.