When data scientist Joel Grus wrote an article on using machine learning to solve the "fizzbuzz" problem last year, most people saw it as an exercise in comedy, perhaps with a warning about the inappropriate use of AI. But we saw a deeper lesson. Certainly, you don’t need AI to solve fizzbuzz, so long as someone tells you the algorithm underlying the problem. But suppose you discover a seemingly random pattern like fizzbuzz output in nature? Patterns like that exist throughout real life, and no one gives us the algorithm. Machine learning solves such problems.