TALAMANA · THE AI LITERACY MAP FOR ARCHITECTURE AND DESIGN · Foundations · AGE 15—17 · FACTUAL · DURABLE
Learned, not programmed
Nobody wrote this behaviour line by line, so there is no rulebook to read.
The idea
Older design software follows rules a person wrote by hand. It can search, solve and simulate, but somewhere a programmer can point to the line that did it. The systems on this map are different. Nobody wrote their behaviour line by line. They were shown enormous numbers of examples and kept the patterns. That is why they can surprise you, and why nobody, including their makers, can list exactly what they will do.
Why it matters
There is no rulebook to look up. To work with these tools you test them, because there is no manual to read.
See it in the studio
CAD snaps a line to a grid because a programmer wrote that rule. An image model puts a shadow where shadows usually go in photographs. One follows a rule. The other follows a habit.
Watch for this
Asking "why did it do that?" and expecting a clean answer. Pattern-machines often cannot show their reasons the way rule-machines can.
Try it
Give the same simple request to a calculator and to a chatbot, five times each. One never varies. One drifts. That difference is the whole idea.
Prove it
Explain the difference between rule-following software and pattern-learning software, with one example of each from your own toolbox.
How it works
Training adjusts millions of internal numbers until the machine's outputs match its examples well. What is stored is not sentences of knowledge but weights: numbers that say how strongly each part connects to the next. The behaviour lives in the weights. It is real, but it is not written down anywhere.
Lineage
Hand-written rules ran computing for fifty years, and still run most of it. Learning took over wherever there were too many rules to write: seeing, speaking, and making pictures of the world.
What this idea builds on
What this idea opens up
Sources
- Mitchell, Machine Learning, 1997
- understanding-ai
- TEC.U
- Google, ML Crash Course
- Goodfellow, Bengio & Courville, Deep Learning
Open this idea on the map · The complete map · Logika · RBDS AI Lab, India · revised every edition.