A forecaster doesn't say 'it will rain' but '70% chance' — reading clues (features) and giving a confidence. AI predictions work the same way, weighing features to output a probability, never a certainty, so its confidence matters.
Turning things into numbers
A model cannot see a photo the way you do — it sees numbers. The measurable clues it uses are called features: the brightness of a pixel, the length of an email, the hour of a purchase. Learning is really finding which features matter and how to weigh them.
Most models do not just answer cat — they answer 92% cat, 8% dog. That number is a probability, an honest measure of how sure the model is. A wise user treats a 55% answer very differently from a 99% one.
Learning by repeated examples is exactly how flashcards work — and how a model is trained. Each pass reinforces the pattern.
- Decide if a fruit is a lemon or a lime using two features: color (yellow vs green) and size.
- Give each feature a weight from 0 to 3 for how much it helps you decide.
- Which feature did you weight higher, and why might color beat size here?
- Now imagine a yellow-green fruit. How confident are you, and why is a probability more honest than a flat yes/no?
What you should see: You chose and weighted features by hand and saw why a model reports confidence rather than false certainty.