11.3 Hallucinations and Incorrect Outputs
AI can make things up - fake facts, fake quotes, fake sources - and sound completely confident doing it.
Chapter 11: Ethics, Risks, and Challenges
A hallucination is when an AI produces information that sounds right but is false. It can invent statistics, quotes, book titles or academic papers that don't exist.
Why does it happen? Language models predict likely-sounding text. They're built to sound fluent, not to check facts. If they don't know, they might still produce something that sounds like an answer.
There's a real-world case where lawyers submitted court filings with cases that a chatbot had made up - and got sanctioned. The fix is simple: verify anything important, especially numbers, quotes and sources.
Rule of thumb
The more it matters (school, money, health, law), the more you check.
Key takeaways
Hallucination = confident but false output
Happens because models optimise for fluent text
Always verify important facts and sources