"AI" is used for everything from spell checkers to chatbots. A useful working definition is software that performs a task by learning patterns from data rather than following only rules a person wrote out step by step.
Rules versus learned patterns
A traditional program for filtering spam might contain a list of suspicious phrases. A learned system is shown thousands of examples labeled "spam" and "not spam" and works out which features tend to separate them. Nobody typed in the final rules; they emerged from the examples.
Three questions to ask
- What was it trained on? The data shapes what the system can and cannot do.
- What is it predicting? Most AI systems estimate something: the next word, the category of a photo, the likelihood of a click.
- How do we know it works? Good systems are tested on examples they did not see during training.
Narrow, not magical
Today's AI tools are strong at specific tasks and surprisingly brittle outside them. A model that writes fluent paragraphs may still miscalculate a simple sum. Thinking of AI as a very capable pattern-matcher, not a person with understanding, sets more realistic expectations.
Why this framing helps
Once you ask what data a system learned from and what it is trying to predict, much of the mystery disappears. You can judge where it is likely to help, where it may fail, and when a human should check the result.
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.