Machine learning is a computer's way of learning from examples, and it is one of the most useful tools we have for the construction of artificially intelligent systems.
It starts with the effort of incredibly talented mathematicians and scientists who design algorithms (a fancy word for mathematical recipes) that take in data and improve themselves to better interact with that data. The algorithms effectively "learn" how to be better at their jobs.
Consider the spam filter working in the background to block your junk email. Since it has "studied" a large set of sample spam emails, it can come to mathematically "learn" what spam email looks like and accurately identify new spam before it leaks into your inbox.
An excellent documentary called "The Smartest Machine On Earth" tells the story of Watson, IBM's famous Jeopardy-winning supercomputer, and delves into how IBM used machine learning to make its creation into a game show champion.
The film introduces viewers to the concept by asking how to best teach a computer to identify the various types of letter "A"s out there — uppercase, lowercase, those written in unusual fonts, and so forth — and it turns out that there is no way to programatically instruct a computer to reliably identify a letter. Enter machine learning, in which the computer studies thousands of examples so that it can build its own mathematical model and eventually have no problem identifying our beloved first letter of the alphabet.
Stepping this example up, IBM had Watson process thousands of actual Jeopardy questions and their correct responses, which effectively taught Watson how to play the game. This "understanding" of Jeopardy combined with Watson's vast storehouse of data — encyclopedias, bibles, and the entire Internet Movie Database, just to name a few — is exactly the trick that enabled a cold, non-living computer to beat two thinking, breathing humans at an especially human game.
The relevant portion of the documentary is embedded below at the appropriate starting point. Just click the play button:
People are building businesses on top of this as well. Heyzap, a mobile app discovery service, figures out what kinds of apps its users like in order to offer them customized recommendations, and machine learning lives at the core of its recommendation engine.
Jude Gomila, founder of Heyzap, told us it works: "Every time Heyzap recommends an app to a user, we ping back to our machine learning engine to work out what to show. We take into account a multitude of contextual data, including looking at how the impact of the filesize of an app relates to the probability of installation given a particular mobile connection speed that the user is on. Through the billions of recommendations and data points we are collecting, this allows us to build correlations in a giant data set and for our algorithm to learn over time what it should be recommending. Machine learning is one of the most efficient technical ways that we have found to get the right apps to the right users."
It doesn't stop there — Wired notes that Amazon recommends loads of products on nearly every page, Pandora builds playlists around the sensibilities of a designated song or artist, and Google makes smart predictions on when you should leave your house to make a meeting on time, and it all happens via machine learning.
It is firmly established as a useful method in making computers more "intelligent" and likely represents the most effective tool we have for one day seeing truly artificial intelligent systems that can adapt and learn with us. It's simply a matter of how we apply it, whether it's to find your next favorite app, identify a letter, or to crush former Jeopardy champions.