How to Use AI for Problem-Solving

AI is great at asking questions, seeing a problem from new angles, and generating options fast. It does not know your situation as well as you do, so the best results come from working with it step by step.

A problem well defined is half solved, and AI is great at helping you define it.

Help me define this problem clearly. Ask me up to seven questions, one at a time, about what is happening, since when, who is affected, and what success looks like. Then write a one-paragraph problem statement.

• Ask “why?” five times to dig below symptoms
• List all possible causes, then rank them by evidence
• Separate facts from assumptions
• Ask AI which causes are most likely and how to test them

Big problems feel impossible. Ask AI to split the problem into smaller parts you can solve one at a time, and to show which parts depend on others.

Give me 15 different ways to solve this problem, from safe to bold. Include at least three unusual ideas. For each, give one benefit and one risk.

Quantity first, judgement later.

First principles

Break it down to basic truths and rebuild

Inversion

Ask how to make the problem worse, then avoid that

Second-order thinking

Ask what happens after the first effect

Constraints

Ask what you would do with half the time or money

Ask AI to apply one model at a time to your problem.

Imagine we chose this solution and it failed badly six months from now. List the ten most likely reasons, and how we could prevent each one.

• Compare the top options in a table: cost, time, risk, impact
• Decide what matters most before scoring
• Ask AI to argue against your favourite option
• Make the decision yourself, and write down why

• Turn the solution into a small experiment
• Decide in advance what result counts as success
• Review results and adjust

• AI may sound confident when it is guessing
• It can agree with you too easily, so ask it to challenge you
• It lacks your full context, so give it facts, not just feelings
• Check any facts it brings in

Define

What exactly is the problem?

Diagnose

Why is it happening?

Generate

What could we do?

Decide

What will we do, and why?

Test

How will we know it worked?