TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Foundations · AGE 14—16 · FACTUAL · DURABLE
Same input, same output?
A rule is checked by reading it, and a draw by looking.
The idea
Write a rule out completely — every step fixed — and ten people following it produce ten copies of the same drawing. That is deterministic generation: same input, same output, every time. Loosen it into a rule system and it defines a space of valid outputs: one pulli grid admits many valid kolams. A generative AI model is a third thing. Give it the same prompt ten times and you get ten cousins, each a draw from a range of likely answers: probabilistic generation. A rule is checked by reading it. A draw is checked by looking at what came out.
Why it matters
Most confusion about AI in design comes from treating a draw as if it were a rule. Knowing which machine you are holding tells you how to check its work.
See it in the studio
A Grasshopper definition rebuilds the same façade every time you open it. An image model given "a brick façade with deep windows" gives you a new one each run. One is a recipe. The other is a roll of the dice. You keep the recipe. You take several rolls and compare them.
Watch for this
"The AI decided on this design." A draw is not a decision. And the reverse trap: expecting a rule system to surprise you. It cannot, and that is its strength.
Try it
Run one fully written-out rule three times on paper — a kolam rule tightened until only one drawing fits. Then run one image prompt three times. Lay the six results in two rows. Write one sentence on what stayed the same in each row, and why.
Prove it
Explain, with one hand-drawn example and one generated example, the difference between same-input-same-output and same-input-different-output, and which of the two you would trust without looking.
How it works
Deterministic systems — a fully specified kolam procedure, Sulba constructions, talamana tables, CAD, parametric scripts — carry their result in their rules. Run them twice and nothing changes unless a rule changes. Rule systems such as the kolam grammars sit between: they fix what counts as valid and leave the choice among valid patterns to the hand. Probabilistic systems hold a spread of possible outputs with a likelihood for each, and a sampler picks one. Settings can narrow the spread, and some tools offer a fixed "seed" that makes a run repeatable, but the machine is still choosing from a distribution rather than following a recipe. This idea sits under THE DRAW and the two "generatives" elsewhere on this map.
Lineage
Three Indian rule systems open this strand on purpose — the kolam, the Sulba constructions, the talamana tables. They show that rules making form is old, familiar and done by hand, so that the probabilistic machine can be met as a new kind of machine rather than as magic.
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.