How AI Works
Lesson 3 of 3 7 min +45 XP

AI in the World

Bias, limits, and being a smart user.

What you'll learn

  • Recognize where AI can fail
  • Explain why human oversight matters
  • Use AI tools responsibly
Garbage in, bias out

Train a hiring AI on years of biased decisions and it faithfully learns the bias. AI mirrors the data it's fed, so being a smart user means knowing its limits — it's a powerful pattern-matcher, not an all-knowing oracle.

Powerful, but not magic

AI now recommends videos, filters spam, helps doctors read scans, and writes text. But it can be confidently wrong. Because it learns from human-made data, it can absorb human biases; because it finds patterns, it can mistake coincidence for cause. It has no understanding of truth — only patterns.

Keep a human in the loop

For decisions that affect people — loans, medical care, who gets hired — a person should review the AI's suggestion. AI is a powerful assistant, not a final judge.

  • Check sources: AI text can state falsehoods fluently and convincingly.
  • Ask about the data: who and what was left out of the examples?
  • Protect privacy: do not paste secrets into tools you do not control.
Lab · Spot the failure
  1. Think of a hiring AI trained on a company's past hires, most of whom shared one background.
  2. Predict who this model might unfairly favor, and why.
  3. Propose one change to the data or process that would reduce the bias.
  4. Explain who should have the final say on a hire, and why.

What you should see: You traced how biased data produces biased decisions and argued for human oversight in high-stakes uses of AI.

Knowledge Check

+10 XP / correct

1. A good reason to keep a human reviewing AI decisions is that AI…

2. AI text that sounds fluent and certain…