In today's streaming world, few people use physical disks to watch movies. Most people don't even know that Netflix still offers the red-envelope service, where you wait for your disk to arrive in the mail.
When you plop yourself down on your couch, you're not prepared to wait. You aren't sure what you want to watch, but you want it now.
To meet your need for instant gratification, Netflix spends a lot of effort customizing your home page.
It wants to present you an appetizing smorgasbord. At the same time, it mustn't inundate you with thousands of choices from Netflix's vast library.
On your home page, you see a tidy shelf of a few rows of titles. It takes you a minute or two to glance through them. Most people will go no further.
Within this attention span lies Netflix's chance to engage you. Will you come quickly to what Netflix calls your "moment of truth," settling on something to start streaming, or will you lurch hither and thither, frittering away your attention until it is exhausted, or your phone distracts you?
While you're watching Netflix, Netflix is watching you.
Above all, Netflix wants to prevent "subscriber churn"— viewers getting bored and dropping out.
While you're watching Netflix, Netflix is watching you, recording your behavior. And not just which titles you streamed, either. Did you pause to consider a particular title before moving on? Did you look through its repeating trailer? Perhaps played a few minutes before abandoning it?
As you watch more and more shows, Netflix slowly gets to know your preferences and can make better predictions.
Your preferences are not fixed; they depend somewhat on context. Are you relaxing in front of your TV on a weekend, or are you watching on a small phone or iPad during a break? Are you watching together with your family?
Netflix can glean some of this information from observing your behavior. For example, it can tell whether you are connecting from a small screen or a big screen. But some of it needs guesswork. Why did you play the same movie fifteen times? If it was on your iPad, then maybe it was your five-year-old playing the same episode of Blue's Clues one more time.
Each of these bits of data might indicate your preferences, or it might need to be discarded. Deciding which data to use is not a well-defined, mathematical problem. It depends on a lot of factors and assumptions.
Pattern-matching problems of this type need computer science techniques commonly called machine learning or "ML."
Machine Learning is the type of narrow AI whose star burns the brightest right now.
The term "narrow AI" refers to programs that are a by-product of research in Artificial Intelligence.
See Being Intelligent About AI for why they are called "narrow AI."
Machine Learning programs match patterns, classify images, and generate images or text. They're called "learning" because they improve their performance over time, as they see more and more examples.
These kinds of programs have become popular in the past decade with startups and their sponsors. Nowadays, whenever a news item says "AI," you can be pretty sure it is talking about a program that performs Machine Learning tasks.
Netflix's recommendation system uses several Machine Learning programs. The aim is to build the best possible home page for you.
Your entire home page is customized for you. The rows of genres are selected based on your preferences, sure. But even within each, which titles to show you, and which exact image to use for an episodic TV show, are all specially designed for you.
Netflix's recommender system doesn't just need to answer a question. It needs to be your information agent that will steer you toward what you are looking for, quickly.
The end result is that you stay on the home page for as short a time as possible. Just long enough to find your moment of truth.