TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Studio Practice · AGE 19—21 · FACTUAL · EVOLVING

The structured building

An image model never knows a door from a gap, and structured systems arrive later.

The idea

A building information model holds walls, doors, levels and quantities as objects with properties and relationships, not as pixels. It is one substrate of practice, beside contracts, specifications and surveys, and the one an image generator lacks. An image model conditioned on a depth or edge map is not blind to structure, but it carries no object semantics: it does not know a door from a gap. Systems that connect a language model to the IFC data read the objects directly. That work is slower than the demos: the data is messy, proprietary and shaped differently in every office.

Why it matters

A tool that handles the skin well and cannot read the body is a visualiser, not a design assistant. Knowing which of the two a tool works on tells you what it can really do.

See it in the studio

An "AI plan generator" demo produces a handsome apartment layout in seconds. The studio's live model of a hospital wing holds tens of thousands of elements with fire ratings, levels and phases attached. The demo cannot open it. The two are not the same kind of thing, and the demo knows only one of them.

Watch for this

"AI-powered BIM" on a product page. Ask two questions: what building data did the learning part train on, and what can it write back into the model? If the answers are "not disclosed" and "nothing", it is a viewer with a chatbot.

Try it

Open any building model. Pick one element — a door — and list every property it carries. Then sort the list: which of these could a learned system have filled from pattern, and which needed a person who knew the project? The second pile is the argument of this card.

Prove it

Explain the difference between a system that sees pixels and one that reads structured objects, with one example of what each can do that the other cannot.

How it works

Two different kinds of system meet the building. A pixel system — a diffusion model, with or without conditioning — learns appearance; conditioning on an edge map or a depth map holds the structure it is shown, but it never knows what the structure is. A structured system — a language model given the IFC file, or a rules engine over the model's objects — reads elements, properties and the relationships between them, and can be asked which doors on level three lack a fire rating. The first kind is trained on billions of photographs. The second is starved: a recent open dataset of residential plans as vector geometry with room-connectivity graphs and metric coordinates holds seventeen thousand, and it is large for its kind. Structured building data is scarce, proprietary and inconsistent, so the systems that read it are narrower and arrive later. The UK practice survey of 2025 showed the same shape — AI use most common in early visualisation and specification writing, with simulation and performance modelling less established; UK evidence, read as such. EVOLVING because the datasets are growing, the IFC-reading systems are new, and the vendors are moving; the gap between the two kinds of system is real today and is reviewed every edition.

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.

Age grows from 11 at the centre to 22 at the edge, and six sectors show the learning strands. Tab into the map and the arrow keys step from idea to idea, following the links where there is one. Enter opens the idea under the cursor, and E reads out its links and the reason recorded on each. Press slash for Search, question mark for the full key list, and Escape to leave. Open Ideas for the complete readable list, including what each idea builds on and what it opens up.

LOGIKA · RBDS AI LAB INDIA
ON-RAMP · AGE 11 · FIRST ENCOUNTERS, NOT GATES — IDEAS · — DEPENDENCIES
DONE
OPENS NEXT
SOLID — STANDS ON