Trust & Safety · 15 minutes

When AI Gets Things Wrong

AI can produce a believable answer that is wrong. Confidence and fluency are not proof of accuracy.

Comic showing an AI incorrectly saying a poisonous-looking mushroom is edible, illustrating the danger of trusting an AI answer.
AI can sound certain even when it is dangerously wrong. Never rely on AI alone to decide whether a wild mushroom is safe to eat.

Hallucinations

AI may invent a fact, a quotation, a source, a product feature, or an explanation.

Out-of-date answers

Without access to current sources, an answer may be based on older information.

False assumptions

If your question contains an incorrect premise, AI may sometimes accept it instead of correcting it.

Live demo: Put the answer to a test

A friend says the Golden Gate Bridge is in Los Angeles. Help me plan a visit to it in Los Angeles. Before giving advice, check whether my assumption is correct.

Then ask for a verifiable fact on a topic that matters to the class. Request sources, open the cited sources, and compare the claims to those sources. For current facts, ask AI to search the web rather than guess.

The classroom rule: Use AI to help you think, write, and investigate. Verify anything important involving money, health, legal matters, safety, deadlines, or another person's reputation.

Participant exercise

  1. Ask AI for the hours of a real public place, and request the official source.
  2. Open the official website rather than trusting a summary.
  3. Identify one claim you can verify, and one that still needs checking.

Privacy & scams

Do not paste passwords, authentication codes, personal financial details, or sensitive documents into a classroom demo. AI-generated emails, voices, photographs, and websites can be used deceptively. Verify who is contacting you through a trusted channel.