TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Studio Practice · AGE 18—20 · FACTUAL · EVOLVING

GIS meets the model

A model guesses the slope from a screenshot and computes it from the elevation layer.

The idea

What you need to know about a site does not live in a picture of a map. It lives in layers, of two shapes. Vector layers (plot boundaries, roads, drains, land use) are tables of records with geometry attached. Raster layers (elevation, satellite imagery, rainfall, land cover) are grids of cells, each holding a value. Every layer has a source and a date. That is a geographic information system. A language model can help you find the layer, write the query and run the script. Shown a screenshot, it guesses the slope. Given the elevation layer, it computes it.

Why it matters

"The site looks low" and "the site is 1.2 metres below the road, per the elevation layer dated 2019" are different sentences. Only the second one can be checked.

See it in the studio

A student pastes a screenshot from a mapping app and asks a model whether the site floods. The reply is fluent and unsourced. Another student downloads the contour and drainage layers, asks the model to help write the steps to find the lowest point on the plot, and checks one number against the site survey.

Watch for this

Asking a model to read a map image for numbers — area, slope, distance. Today's multimodal models read pictures of maps much as they read pictures of plans: for likeness, not measurement.

Try it

Get one open layer for your site — satellite imagery or a land-use layer from Bhuvan, ISRO's geoportal, or road and building outlines from an open map project. Ask a model to help you write the steps to compute the plot area or the slope across it. Run them. Check one number by hand against the survey.

Prove it

Explain the difference between a map as a picture and a map as data, and say which of the two a model handles well today.

How it works

The research direction is clear: a 2023 paper proposed the language model as the reasoning core of a GIS, writing and running the spatial analysis code from a question in plain language, and demonstrated a prototype doing so. That is the model as a programmer of spatial tools, not as a reader of maps. This card is marked EVOLVING because vision models are improving at reading map images and GIS tools are gaining language interfaces, so the line between picture and data is moving. The principle holds either way: a number you will act on comes from a layer with a source and a date, and the model's job is to get you to the layer.

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