The short version
Machine learning is programming with examples instead of rules: you show a system thousands of inputs and answers, and it finds the pattern itself. Everything from spam filters to ChatGPT is this one idea at different scales.
What it actually is
Classic software encodes rules a human wrote. Machine learning flips it: you provide data and the desired outputs, and an algorithm searches for a function that maps one to the other. The model is that discovered function, millions of tuned numbers rather than written logic.
Why it took over
Some problems have patterns too subtle or too numerous for handwritten rules: recognizing faces, translating languages, predicting churn. Once data and compute got cheap, learning the rules from examples started beating writing them, and it has not stopped.
How to think about it
Three ingredients decide everything: the data (what examples it saw), the objective (what it was rewarded for), and the architecture (what patterns it can express). Most ML failures in the news are data or objective problems wearing a technology costume.
