Course 2, lesson 15 of 100, Ages 6+
Good guess, bad guess
Why AI makes mistakes
Like I’m 5
When you guess what’s inside a wrapped present, you might be right or wrong. AI is always guessing too, just very cleverly.
The big idea
AI doesn't know answers the way you know your own name. It makes its best guess from patterns, and it often gives a confidence score: 90% sure it's a cat, 10% sure it's a fox.
Good guesses come from good practice, but every guesser can be wrong, especially with unusual examples, like a cat in a fox costume. Knowing that AI guesses helps us double-check important answers.
Examples
- Photo app: It's 95% sure your picture shows a beach.
- Weather app: '70% chance of rain' is a guess based on patterns.
- Tricky pictures: Some muffins look so much like puppy faces that AI gets fooled.
How it works
- AI never knows for sure. It makes its best guess, based on what it has seen before.
- Some guesses come with a score, like ‘90% sure’. Even 90% sure means it’s wrong some of the time.
- When something is new or unusual, AI guesses wrong more often. That’s why people double-check.
Check your understanding
- An AI says it’s 90% sure. What does that mean?
- Options: It can never be wrong; It’s usually right, but not always; It’s guessing at random.
Answer: It’s usually right, but not always. A high score means a likely answer, not a certain one. - An AI says it's 55% sure. What does that mean?
- Options: It isn't very sure, so double-check; It is completely certain; It is broken.
Answer: It isn't very sure, so double-check. Just over half sure is a weak guess. Check before trusting it.
Remember
AI gives its best guess, never a promise.
Talk about it
When did you last make a good guess? How did you decide?
Go deeper
Models output probabilities, not facts. Their confidence can be badly calibrated, and they often fail on inputs unlike their training data. Good systems show their uncertainty and keep a person in the loop for important decisions.