The Big Idea: Powerful, But Not Perfect
Modern AI systems can write essays, answer questions, translate languages and describe images — often so smoothly that they feel all-knowing. They are not. Every AI system has real, predictable weaknesses: it can invent facts with total confidence (hallucination), it can repeat unfair patterns hidden in its training data (bias), it can fail at simple common-sense reasoning, and its knowledge freezes at a cutoff date. None of these are rare glitches — they follow directly from how the technology works.
In this lesson you will learn why these failures happen and, most importantly, how to protect yourself: treat every AI answer as a first draft, and verify anything important before you rely on it.
Learning Goals
- Explain what a hallucination is and why language models produce confident but false statements.
- Describe how bias enters AI systems through training data, with real examples from hiring and face recognition.
- Recognize why AI lacks common sense and why its knowledge becomes outdated after the training cutoff.
- Understand the overreliance risk: why AI answers are drafts, not authoritative sources.
- Apply a practical trust-but-verify workflow: cross-search, check official sources, and ask the AI for sources — then actually check them.
Limitation 1 — Hallucination: Confident Inventions
A hallucination is when an AI states something false as if it were a proven fact. Typical hallucinations include:
- Invented facts — a "famous quote" nobody ever said, an event that never happened, a feature a product does not have.