Course 4, lesson 38 of 100, Ages 10+

Thinking in chances

Probability and confidence

Like I’m 5

When you toss a coin, it might land heads or tails: one chance in two. AI thinks in chances too. It says how likely each answer is.

The big idea

Probability measures how likely something is, from 0 (impossible) to 1 (certain). Many AI models output probabilities: 80% cat, 15% dog, 5% fox.

A good model's confidence should match reality: when it says 80%, it should be right about 8 times out of 10. Knowing the confidence helps people decide when to trust the AI and when to double-check.

Examples

  • Dice: Rolling a six has a one-in-six chance.
  • Spam score: '97% likely spam' is a probability.
  • Medical tools: A low-confidence result is sent to a doctor to review.

How it works

  1. The model looks at the input.
  2. It gives a probability for each possible answer.
  3. People use that confidence to decide whether to trust or check it.

Check your understanding

An AI says 80% cat, 15% dog, 5% fox. What's its best guess?
Options: Cat; Dog; Fox.
Answer: Cat. Cat has the highest probability.
If a model says 80% often, how often should it be right?
Options: About 8 times out of 10; Every single time; Never.
Answer: About 8 times out of 10. Well-calibrated confidence matches how often it's actually right.

Remember

AI gives answers with chances. Good confidence matches how often it's actually right.

Talk about it

What's the chance of rain today, and would you take an umbrella?

Go deeper

Matching confidence to accuracy is called calibration. Classifiers often use a softmax layer to turn scores into probabilities, and calibration can be checked with reliability diagrams.