My favorite movie, Days of Heaven, is at the top of my recommendations list on Netflix. But I've never actually watched it on Netflix, so how did they know I like it?
"Every time a Netflix member streams a title from us, we learn a little bit more about what's interesting to them," says John Ciancutti, vice president of engineering at Netflix. Netflix has algorithms that mix customer information with other algorithms that group shows together. So if I enjoy Mad Men, it guesses that I might like other dramas with complicated male leads, like Breaking Bad — and it's right.
If the Industrial Revolution was about extending the power of human muscle with inventions like the car, then the computer revolution is about extending the power of the human mind — and algorithms are the key to its success. These formulas find search results, pick the top story on the news feed of your Facebook page, and determine things like your credit score and trades on Wall Street.
Every day Netflix has dozens of engineers improving its algorithms. A huge whiteboard in the hallway of Netflix headquarters displays numbers in a grid, part of a contest to see who can come up with the best algorithms.
"Just in the last couple of months, we've run tests where we've improved overall streaming hours for members with a new algorithm that's just a little bit better at making recommendations, which means it's so powerful as far as delighting members that they are more likely to stay with the service versus not," Ciancutti says.
For Netflix, it's about keeping customers. But algorithms tailored to figure out individual tastes and interests are also being applied to the political arena. When Mitt Romney is on local TV in Ohio, it's no surprise that he'd talk about local interests, like manufacturing jobs. But it would be even better if he could target his message directly to people who lost those jobs.
In this coming election, that's exactly what Democrats and Republicans will be able to do.
"There's been this explosion of data available over the last decade, frankly, at the individual level from voter files, from consumer sources, from other sources," says Tom Bonier, co-founder of Clarity Campaign Labs, a company that uses algorithms to help Democrats target voters. "You can send one piece of mail about choice to one household, and to their neighbors you can send a piece of mail about the environment."
In other words, there are more computers hoarding more data about us than ever before.
Algorithm As Crystal Ball
Sean Gourley is co-founder and CTO of Quid, a company that gets hired by governments and business to create algorithms. He says, "From an algorithms perspective, this is a great time to be alive. Algorithms are just frolicking in the mountains of data that they can play with."
Gourley uses algorithms to predict insurgencies in Iraq and Afghanistan and help banks map developing markets for new technologies. As an experiment for NPR, he maps the development of the Occupy Wall Street movement. He uses algorithms to sort through about 40,000 blogs and articles written since the movement began and group similar ideas together.
"You can't read all this as a human or even necessarily get what's going on, so you start to apply algorithms to help kind of cluster, sort, put topics around them and ultimately visualize [it]," he says. What Gourley's algorithm helps visualize is how ideas from the initial Occupy Wall Street rallies in New York spread to other groups and other parts of the country.
Gurley points to a computer screen. It's filled with dots that are grouped into color clusters and connected by thin lines. One cluster represents politicians talking about taxing the wealthy. It's far from the cluster that represents protesters.
"When the politicians talk about Occupy Wall Street, they're not really talking in Occupy Wall Street," Gourley says. "They're not talking within the cluster — they're talking separately. And so when they talk about taxes and they talk about inequality it's not really resonating [in the Occupy Wall Street cluster], or at least the language is quite different."
Gourley clicks on another smallish cluster that represents "Bank Transfer Day," an activist initiative that calls for people to move their accounts out of larger banks and into credit unions — you can see that people aren't saying much about it anymore. A more recent cluster represents the issue of the courts using Twitter feeds as evidence in cases against protesters, and yet another tracks the conversation around Occupy Oakland, which is dominated by talk of police violence.
Proceed With Caution
Gourley imagines that information like this might be useful to politicians, police and political activists.
"We can start to think about where it's come from," he says. "We can start to think about how it's evolved and we can start to think about where we want it to go and if we want to change the direction."
But Gourley has concerns about other possible uses of algorithms. Netflix is just trying to keep its customers watching movies, which isn't so bad, but he wonders what would happen if Google Maps knew that you were looking for a new car. Maybe when you looked up directions to a party, it would suggest a route that passes right past a dealership.
"[Has] it got your interest at heart or has it got making money from ads at heart?" Gourley says.
What worries him most is that we humans haven't yet evolved to be as wary of algorithms as we are of used car salesmen.
Copyright 2012 National Public Radio. To see more, visit http://www.npr.org/.