TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Ethics & Provenance · AGE 18—22 · POSITIONAL · HELD
Documenting is representing
A faithful record made with consent makes it possible for a model to imagine the place.
The dilemma
Your thesis is on a wada neighbourhood that no image model can picture. You have a month of site time. You can spend it on your own design, or on a faithful, well-captioned record of the place — plans, sections, materials, the way light enters — that will outlive your thesis. Which is the work?
The choices
Design only, and let the place stay invisible to the next model. Document as a by-product — a few photographs for your own sheets. Or document deliberately: measured, captioned, named, placed where others can find it, with the residents' consent and credit, knowing it may be scraped.
The consequence
The first route leaves the tilt exactly where it was. The second adds a few more uncaptioned images to a feed. The third is slower, and it is the only one that makes it possible for a future model to imagine the place — possible, not certain: whether the record enters a dataset, survives filtering and carries any weight is another chain altogether. It also carries a real risk: what you publish well may be taken without asking, and some of what you could record may be better left unrecorded.
The case
A studio in Hubli documents twenty courtyard houses over a semester: orientation, roof, verandah depth, material, photographed and drawn with the families' permission and names. Two years later a student briefs a model with that record as reference and gets a courtyard that behaves like a Deccan courtyard. The record did that. The model only repeated it.
Our position
When a community, region or vernacular is under-photographed, documenting it well is more than craft. It is representation. Data dignity begins with who bothers to record their own context, how faithfully, with whose consent — and with the right of the people in it to decide that some of it stays unrecorded.
Why we hold it
The tilt in the mirror will not be fixed from inside the model; more of the same web carries the same lean. The only thing that can balance it is better records of the places that are poorly recorded, made by people who know what they are looking at. Architecture students are exactly those people.
The strongest objection
Why should being legible to a machine be the measure of dignity at all? A community may rationally prefer not to be indexed. Sacred knowledge, domestic life, ritual, a vulnerable settlement, an undocumented practice — each can hold its value partly because it is not available to everyone. Framing under-representation as a documentation deficit turns "data dignity" into a duty to make yourself machine-readable, and that is the platform's interest wearing the community's name. Absence from a dataset is sometimes a boundary, not a gap — and even where the record is wanted, representation without control is extraction by another route: the next scrape takes the wada, and the families get a prettier default and nothing more.
What would make us revise it
Evidence that open documentation of under-represented vernacular mainly enriched platforms and not the communities; the arrival of consent-respecting data commons that make the trade-off disappear; or a community the Lab works with saying no — at which point, for that place, the position is the no. Any of these would change how, where, and whether this card tells you to publish. Reviewed every edition.
Prove it
Document one under-photographed place or practice from your own context faithfully enough to correct a model's default, and explain why careful local documentation is an act of representation, not just archiving.
Take it to crit
Does the student see recording their own under-photographed context as representation the world's data is missing — and do they take the responsibility that carries, including consent and credit for the people in it?
How it works
The phrase data dignity comes from Jaron Lanier and Glen Weyl's argument that people should be recognised and paid as the source of the data that makes machines valuable. The Lab uses it one step earlier. Before you can be paid for your data, you have to be in the record at all, accurately, on your own terms. Consent from the people you photograph, and their names where they want them, is where those terms begin.
What this idea builds on
What this idea opens up
- Nothing yet names this as a foundation.
Sources
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.