What you will learn
  • Separate generating possibilities from evaluating them.
  • Request ideas that differ in mechanism rather than wording.
  • Compare options using criteria chosen before scoring.
  • Identify assumptions behind a promising suggestion.
  • Design a small test that could reject or improve an idea.

Before you begin

Read USE03 on task briefs. You need a problem you care about and a realistic list of available time, skills, and materials.

Start with a problem worth solving

A request for ‘ten amazing AI projects’ often produces familiar ideas with impressive names. A better starting point is a small frustration you have observed: students forget to check whether a classroom plant needs attention, or club members cannot find the latest project instructions. The problem gives brainstorming a purpose.

Describe who experiences the problem, when it occurs, and what a useful improvement would look like. Avoid prescribing the solution too early. ‘Help students remember plant care’ leaves room for a paper checklist, a reminder, or a sensor. ‘Build a camera-based AI plant doctor’ assumes a complex approach before establishing the need.

AI suggestions are combinations and transformations of patterns, not evidence that an idea is original, feasible, or desirable. Your observations and later tests provide that evidence. The assistant’s useful role is to widen the options and expose questions you might otherwise miss.

Separate expansion from selection

During expansion, seek several distinct mechanisms. For the plant problem, ask for options based on a routine, a visible reminder, a simple measurement, and a data-driven prediction. These categories produce more meaningful variety than ten names for the same mobile app.

During selection, apply constraints and criteria. Time, cost, available skills, maintenance, and privacy may matter. Choose the criteria before reviewing an attractive favorite so you do not unconsciously design the scoring system to make it win. A criterion should describe something you can inspect or estimate honestly.

Do not treat a numerical score generated by the assistant as a measurement. A four-out-of-five feasibility rating may be a reasoned judgment, but it is not a test result. Ask for the reason, assumptions, and confidence behind each rating. Mark unknowns rather than converting them into invented precision.

A brainstorming conversation with useful variety

Weak prompt: ‘Give me cool AI ideas for school.’ It lacks a problem, resources, and a way to decide among suggestions.

Improved prompt: ‘Our school club wants to help members remember classroom plant care. We have one week, paper supplies, and beginner Python skills, but no sensors yet. Generate six approaches using different mechanisms. Include at least two that do not need AI. For each, state the user action, required resources, main assumption, and smallest test. Do not claim that any idea is new or proven.’

The assistant might suggest a rotating checklist, a calendar reminder, a simple care log, or an AI-assisted explanation of the log. The non-AI options are valuable because the goal is to solve the problem. If a paper routine works, that is evidence about the need; it does not make the exercise a failure.

Follow-up prompt: ‘Compare the checklist, reminder, and care log using setup time, daily effort, and whether students can explain the system. Treat costs and behavior changes as unknown unless our information supports them. Recommend one one-day trial, with a reason.’ Verification: Check resource availability and ask intended users whether the proposed action fits their routine. Do not substitute the model’s prediction of user behavior for observation.

Challenge the favorite constructively

Once an option looks promising, ask what would make it fail. A reminder may be ignored. A care log may collect inconsistent entries. A camera model may confuse lighting changes with plant condition. The purpose is to identify the most important assumption to test, not to dismiss every idea because perfection is impossible.

A useful test is small, observable, and capable of changing your decision. For a care log, have two volunteers record a short entry for one day and note missing fields or confusing instructions. This tests whether the workflow is understandable. It does not yet prove long-term adoption or accurate plant-care advice.

Define a finish condition and a stopping condition. Finish when you have enough evidence to choose the next step. Stop or redesign when the activity requires unavailable resources or produces confusion that defeats its purpose. Google’s prompt design guidance supports using explicit constraints and examples; here they make ideation more relevant and inspectable.

Develop an idea without inflating it

Convert the selected idea into a short brief: problem, intended user, input, behavior, output, and test. Keep optional features in a separate list. A beginner care log does not need accounts, image analysis, push notifications, and a chatbot on its first day.

Ask the assistant to identify the smallest complete version. Complete means it performs the core task from beginning to end, even if it uses sample data or manual steps. A tiny working loop teaches more about feasibility than a large diagram of features that have never been connected.

Save rejected alternatives and the reason you rejected them. They may become useful when constraints change. Also record where the AI helped and where human observation changed the decision. This makes your process repeatable and helps you avoid confusing a polished suggestion with a solved problem.

If originality matters, conduct a separate search for similar products or published work. A model cannot certify that no one has proposed an idea before. For a school learning project, originality may mean adapting a known technique thoughtfully to your own situation, not inventing an entirely new field.

Important terms

Brainstorming
Generating possible approaches before choosing among them.
Constraint
A real boundary on available solutions.
Criterion
A property used to compare alternatives.
Prototype
A limited version built to test an idea.
Assumption test
An observation designed to check an important unverified belief.

Mini project: Run a one-day idea trial

  1. Write one observed problem, the affected users, and available resources.
  2. Request six approaches with different mechanisms, including simple options without AI.
  3. Choose three criteria and compare the options, keeping unknowns visible.
  4. Trial one small behavior for a day using volunteers or your own routine. Finish with an observation, a limitation, and a decision to continue, change, or stop.

Common mistakes and debugging

  • Starting with technology instead of a problem: describe the user need first.
  • Accepting renamed copies as diverse ideas: request different mechanisms.
  • Treating invented scores as data: ask for assumptions and test important claims.
  • Adding features before validating the core behavior: build the smallest complete loop.

Independent challenge

Repeat the plant-care brainstorm with no computer available. Explain which parts of the problem still need solving and what the constraint teaches you about the original idea.

Check your understanding: 10 questions

  1. Why include non-AI options in an AI brainstorming session?

  2. How do expansion and selection differ?

  3. Is an AI-generated feasibility score an experimental result?

  4. What makes a prototype test useful?

  5. Can a model certify an idea’s originality?

  6. In your own words, what does “Brainstorming” mean?

  7. In your own words, what does “Constraint” mean?

  8. In your own words, what does “Criterion” mean?

  9. In your own words, what does “Prototype” mean?

  10. In your own words, what does “Assumption test” mean?

Quiz answers

Reveal all 10 answers after your attempt
  1. They help compare approaches against the real problem and may solve it more effectively with fewer dependencies.
  2. Expansion generates possibilities; selection applies criteria and evidence to choose among them.
  3. No. It is a judgment that should expose its assumptions and be checked where it matters.
  4. It checks a consequential assumption through an observable activity and can change the next decision.
  5. No. Originality claims require separate investigation and cannot be guaranteed by generated suggestions.
  6. Generating possible approaches before choosing among them.
  7. A real boundary on available solutions.
  8. A property used to compare alternatives.
  9. A limited version built to test an idea.
  10. An observation designed to check an important unverified belief.

Summary

Use AI to widen the options, then choose with explicit criteria and small real-world tests. A promising idea becomes useful through evidence and revision.

Continue learning

Continue with USE09 to build a more systematic method for checking AI claims and results.

Choose a connected learning path

Sources and further reading

Prepared 2026-09-18. Draft; conceptual review complete. Project ideas and anticipated outcomes are illustrative, not user-tested claims.

GO DEEPER

Extra reading & source documents

Optional reading alongside the lessons. These sources do not add to your course lesson count.

Explore the AI model guides →