Learning patterns instead of writing rules
Traditional software is built from explicit rules: a programmer writes 'if the email contains these 50 spam words, flag it,' and the computer just follows that logic. Machine learning flips this around. Instead of a human writing the rules, you show the computer thousands of examples of spam and non-spam emails, and an algorithm searches for patterns in those examples that separate the two categories. The 'program' that results isn't a list of if-statements a person wrote, it's a set of numerical parameters the algorithm discovered on its own.
This is why ML is so useful for problems that are hard to describe as explicit rules, like recognizing a face in a photo or predicting whether a customer will cancel their subscription. Nobody could write down every rule that defines 'this photo contains a cat,' but a model can learn the pattern by seeing millions of labeled cat and non-cat photos.