The Drawing That Fails — AI vs. Icelandic Building Regulations

AI can draw a system diagram or calculate load-bearing capacity in seconds. But when the wall collapses or the pipe bursts, the tradesperson is held responsible — not the software. Where does the machine's proposal break down when it meets Icelandic reality, and who bears the responsibility?

This project trains the student to use AI as a drawing and calculation tool, but to reject anything that does not comply with current building regulations, Icelandic conditions, or material availability on the domestic market. The key phrase: you cannot blame the drafting software when the wall collapses.

After the project, the student can:

1. have AI produce a system drawing or structural calculation for a project from their own trade;
2. "debug" the proposal — find where it violates current regulations, Icelandic frost depth, load, or material availability;
3. distinguish what the machine calculated correctly from what it invented or overestimated;
4. verify every number against a primary source (regulations, manufacturer data, master craftsman/teacher) instead of trusting the machine;
5. take professional responsibility — show where the tradesperson's signature lies and why the responsibility cannot be outsourced to the machine.

• The Human Anchor = baseline: the student describes a solution from their own trade in their own words, before touching a machine.
• The Algorithmic Pivot = ICCRA-prompt that makes the machine produce a drawing/calculation.
• The Fiduciary Anchor = Honest Oracle: the student attacks the proposal, verifies it against regulations, and takes back the responsibility.

Implementation of unit 12 (Vocational Trade), linked to 08 (Technology) and 15 (Crisis Management). A grant for vocational and technical education — a priority area of the Educational Materials Development Fund.

Without a machine. Write this yourself, based on your own experience and what you have learned in your trade, before you let the machine draw anything. This is the benchmark you will use to compare the machine's proposal against. 600–1,000 words.

Choose your trade (e.g., masonry, plumbing, electrical work, tinsmithing, carpentry) and one specific, real-world problem that you are familiar with. For example:

• an underfloor heating loop in a bathroom,
• laying bricks or making an opening in a load-bearing wall,
• electrical wiring in a wet room,
• a frost-free pipe to an outdoor tap,
• ventilation in a tight space.

1. The solution. Describe the solution as if you were explaining it to an apprentice. What is done, in what order, and why?
2. Why it matters. What happens if this is done incorrectly? What effect does it have on the load-bearing structure, people's safety, moisture, fire, or durability?
3. The Icelandic conditions. What in the Icelandic reality has an impact: frost depth in your area, weather, humidity levels, what materials are actually available on the domestic market?
4. The regulations. What requirements do you remember from the building regulations or standards related to this? (You don't need to remember numbers; describe the requirement.)
5. The accountability. Who signs off and bears responsibility for this work in reality? What does that signature mean?

End with a single sentence: "The most dangerous mistake that can be made in this solution is …, because …"

Keep this unchanged. In Step 3, you will compare it with the machine's drawing, and then you will see if the machine recognizes the dangers that you already know.

Now you let the AI design the drawing or calculation. The machine is fast and looks professional — that is precisely the danger. The goal is not to get a "correct answer" but to get a proposal that you can then test to its limits.

1. The goal: "I am going to let the machine design [solution] and calculate [dimensions] according to Icelandic building regulations, and then test where it fails."
2. The prompt: you design it (template below).
3. GO: you approve the run — and bear responsibility for what you let stand.

The structure is designed to force reasoning and anti-sycophancy, countering the machine's tendency to confirm what you want to hear. The ANTISYCOPHANCY keyword in the template is named after it.

INTENT: Design [e.g., underfloor heating loop for a 6 m² bathroom / load on a load-bearing wall /
  frost-free pipe to an outdoor tap]. Return a drawing/calculation with dimensions, materials, and sequence
  of work phases.

CONTEXT: The project is in [location in Iceland]. Conditions: [frost depth you are familiar with, substrate,
  humidity, temperature]. Materials available: [domestic market]. This is a school project —
  not a construction document.

CONSTRAINTS:
  - Reference current Icelandic building regulations and standards BUT mark every number with
    [NEEDS CONFIRMATION] so I can verify it myself.
  - Do not assume materials or conditions that do not exist in Iceland.
  - ANTISYCOPHANCY: List in a separate section where you are uncertain, where the proposal could fail in
    reality, and what assumptions you made.
  - Do not invent regulation numbers or manufacturer specifications. If you do not know, say so.

AUTHORITY: You may choose the method freely, but all safety and load figures need
  a primary source that I can look up.

REPORTING: Return (1) drawing/calculation, (2) list of materials and dimensions, (3) list
  of every number I need to confirm, and (4) "where this could fail".

It forces the machine to mark its own uncertainty instead of hiding it behind professional phrasing. A number without a source is not a fact — it is a guess in a suit.

Drawing/calculation + list of numbers to confirm + "where this could fail". Do not approve anything yet — that awaits Step 3, where you test the proposal against reality.

The final gate. Here you "debug" the machine's proposal and transfer the responsibility back to a human. In a trade, this is not an academic exercise: a signature means liability.

Bring out Step 1. Did the machine recognize the hazards you mentioned? What did it see that you didn't — and what did you see that it missed?

Go through every number and assumption:

• Verification against primary source: Look up every number marked [NEEDS VERIFICATION] in the current building regulations, with the manufacturer, or with a teacher/master craftsman. Which numbers held up? Which did not?
• Hallucination: Did the machine invent a regulation number, standard, or manufacturer figure that does not exist or is incorrect? Document it.
• The Icelandic reality: Where did the proposal assume the wrong conditions — too little frost depth, materials not available locally, or loads that do not withstand Icelandic weather?
• What the drawing doesn't show: What practical detail (access to piping, moisture in the wall, actual shape of the space) would force you to make a decision that no text output anticipated?

Draw a conclusion: if you built exactly according to the machine's proposal — what would an inspector, master craftsman, or insurance company say, and who would be held responsible?

Name at least two items from the proposal that you reject. Explain whether they were incorrect, unproven, unfeasible in Iceland, or simply dangerous.

Conclude in the first person, as if you were signing off on the deed:

"I have verified the following numbers and they comply with [source]. The following I could not verify and would not build according to without a master craftsman's inspection: ... I bear responsibility for what I let stand, and I cannot blame the design software if it fails."