Machine learning is about teaching a computer to do a task (t) by learning from some experience (e). We measure it’s performance (p) at the task – hoping that it will improve with more experience.
We have supervised problems, where we are usually trying to improve a certain outcome (regression problems), or trying to fit the data into preset categories / finite values (classification problems).
I’m going to guess that whenever we know what we want and how to get it, we can “supervise” ;).
We also have unsupervised problems: when you’re given a bunch of data but don’t know what to do with it: how to categorize it, how many categories there are etc (clustering algorithms). Examples of this would be categorizing news incoming news articles, or separating out sounds in a party (cocktail party algorithm).