Machine learning replaces hand-written instructions with a loop: make a prediction, measure the error, adjust, and repeat. Understanding that loop explains most of what AI can and cannot do.
Learning from examples
Suppose you want a system to estimate house prices. You provide many houses with known prices. The model starts with poor guesses, compares them to the real prices, and nudges its internal numbers to shrink the error. After many rounds, it captures patterns such as "more floor area usually means a higher price."
Training and testing are different
If you only measure a model on the examples it studied, you learn whether it memorized them, not whether it learned something useful. That is why data is split: one part for learning, another held back to test on unseen cases. A large gap between the two scores is a warning sign called overfitting.
Why examples matter so much
A model can only learn patterns that appear in its data. If an important group or situation is missing, the model will perform poorly there. Biased or messy examples produce biased or messy behavior.
What to take away
- Learning is repeated adjustment guided by error.
- A fair test uses data the model has not seen.
- Better data often helps more than a bigger model.
Continue with the book
Explore the full treatment, worked examples, and exercises in AI Made Simple.
Adapted from AI Made Simple, published by Choral Software. Article text and original examples © 2026 Choral Software. All rights reserved.