Critical thinking with AI

The hallucination problem

A hallucination is when AI states something false as if it were true — fake statistics, made-up book titles, non-existent legal cases, or wrong dates. The response can sound polished and authoritative.

This is not a rare bug. It is a known limitation of how language models work. Your job is to be the quality control layer.

Red flags to watch for

• Overly specific numbers with no source
• Named studies, laws, or URLs you cannot verify in one search
• Absolute claims: “always,” “never,” “guaranteed”
• Advice that seems too convenient or confirms your bias without nuance
• Contradictions within the same answer or across follow-up questions

How to verify without being an expert

You do not need deep technical skills — you need habits:

• Cross-check important facts with a second source (search, official docs, a colleague)
• Ask the AI: “What might be wrong with this answer?” or “What are you uncertain about?”
• Separate draft from decision — use AI to explore, not to authorize
• For high-stakes topics (health, legal, money), treat AI as a starting point only

When to trust AI more

• Rewriting and formatting text you provide
• Explaining general concepts at a high level
• Generating options or frameworks when you will evaluate them
• Tasks where errors are obvious and cheap to fix (brainstorm lists, outlines)

When to trust AI less

• Medical, legal, or financial decisions
• Anything citing specific regulations, dosages, or case law
• Numbers that will go into a board deck or public statement without review
• Sensitive interpersonal situations where tone and history matter deeply

Build a personal checklist

Before you use AI output externally, ask: Would I be embarrassed if this were wrong? Would harm result if this were wrong? If yes — verify or rewrite manually.

Critical thinking with AI is not cynicism. It is using speed without surrendering responsibility.