TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Foundations · AGE 15—17 · POSITIONAL · HELD
Generative design is not generative AI
Generative design searches a space you set up, and generative AI predicts from its data.
Our position
Generative design and generative AI are not synonyms, and the Lab keeps them apart. Generative design is an umbrella over several lineages — search and optimisation, shape and graph grammars, agent-based systems — and what they share is that a designer sets up the space: the rules, the goals, the scoring. Every result traces back to that setup. Generative AI predicts plausible results from patterns in its training data, with no goals and no scoring unless you bring them. The shared word hides that difference, and the difference decides how each is checked.
Why we hold it
Clients and juries mix these up constantly, and a designer who can separate them in one sentence can lead the conversation. The two are verified differently. A generative-design result is checked against the goals and rules you wrote. A generative-AI result is checked by looking, against criteria you bring afterwards. A student who confuses the two either distrusts an honest search or over-trusts a likeness.
The strongest objection
The clean split is already leaking from both sides. "Generative design" is not one machine, and some of what sits under it — evolutionary solvers, learned surrogates, agent systems — is as opaque as any model. Meanwhile generative AI is being wrapped in goals, constraints and scoring: a diffusion model inside an optimisation loop is both machines at once. Calling them "opposite" may teach students a boundary the tools no longer respect.
What would make us revise it
When mainstream design tools ship generative AI inside a scored, constrained search as the default rather than the experiment, so that "which machine is this?" has no single answer for most tools a student meets, we rewrite this card around a different question: which parts are searched against goals, which parts are predicted, and how is each checked. We test this every edition against what the tools actually ship.
See it in the studio
A structural layout tool that minimises steel tonnage under your loads is generative design. The tonnage number defends the result. An image model's "efficient structure" is a picture of what efficient structures look like. One is engineering. The other is a summary of how such structures look.
Watch for this
The phrase "the computer optimised it" attached to a generated image. Optimisation needs a measurable goal. Without a goal there is no optimisation, only likeness.
Try it
Write one design goal as a number (daylight hours, steel weight, travel distance). Ask: which of my tools can actually score against this? That question separates the two machines at once.
Prove it
Explain to a junior the difference between a tool that searches your goals and a tool that predicts its data, with one honest use for each.
How it works
Most generative design belongs to the search family — define, search, score, rank — with grammar-based and agent-based cousins beside it. You can usually see every step. Generative AI belongs to the prediction family you have met across this ring. Hybrids are arriving that wrap one inside the other. If you know the parts, you can read the hybrids too.
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.