TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Foundations · AGE 15—17 · FACTUAL · DURABLE
How a model is made
A model is built in four steps, and each step has its own failure.
The idea
A model is built in steps. Gather data. Train on it. Test what was learned. Put it to work. Each step has its own failure: gathering can bake in gaps, training can memorise instead of learn, testing can miss what the test never asked, deployment can meet a world the data never showed. When a tool misbehaves, the useful question is: which step is showing?
Why it matters
"The AI is wrong" is a dead end. "The gathering step missed our building type" is a diagnosis, and sometimes a fix.
See it in the studio
A plan-reading tool fails on your hand-drawn survey. It was probably never shown one at the gathering step. That is not stupidity. That is its diet.
Watch for this
Test scores quoted without asking what the test contained. A model can pass its own exam and fail in your studio, because its makers set the exam.
Try it
Pick one tool you use. Try to find one fact about each step: what data, trained by whom, tested how, deployed where. The steps you cannot find out about are worth noticing too.
Prove it
Walk through the four steps for any real system and name one way each step can fail.
How it works
Professionals split the steps further — cleaning, labelling, validation splits, fine-tuning, monitoring — but the four-step skeleton holds. Later ideas on this map attach a failure to each step: bias at gathering, memorisation at training, benchmark traps at testing, drift at deployment.
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.