lesson
40 min

When Not to Build

A written test for deciding whether a model should exist, and how to say no with evidence.

Objective

By the end of this lesson, you can apply a five-question test to a model idea and write a short, evidence-based recommendation to build, change, or stop it.

Saying no is engineering

Many harmful systems were built by capable people who never stopped to ask whether they should. The most valuable thing an engineer can do is sometimes to say, "This should not be a model," and explain why.

The five-question test

Answer each in writing:

  1. Purpose. What decision does this support, and who makes it?
  2. People. Who is affected? Were they asked?
  3. Mistakes. What happens to a person when it is wrong? Can they find out and appeal?
  4. Data. Does the data fairly represent the people affected? Does any label or proxy carry past unfairness?
  5. Alternatives. Would a rule, a checklist, or more staff time do the job with less risk?

If you cannot answer one, stop and find out. If the honest answer to question 3 is "serious harm, no appeal," do not build it.

A national example

Practice with an AI reviewer

An AI assistant can act as a tough reviewer. It does not get the final say. You and the people affected do.

Prompt Lab

Get a hard critique of a model idea using the five-question test.

Who should the AI act as?

What exactly should it do?

What does it need to know?

What shape should the answer take?

Your prompt

Act as a skeptical ethics reviewer for a community nonprofit.

Your task: Apply the five-question test (purpose, people, mistakes, data, alternatives) to my model idea and tell me the weakest answer

Format: Five short sections, then one line: build, change, or stop, with the reason

Do not paste private details (full name, address, ID numbers) into any AI tool.

Reflection

Private. Saved only in this browser. No one else can see it.

Ask. Answer. Build together.