Will the information superhighway turn into a cul-de-sac because of automated filters? (The Wall Street Journal)

Your Results May Vary


Last year Eli Pariser, president of the board of the liberal-activist site MoveOn.org, had a shocking realization. A heavy Facebook user, he had become friends—at least on Facebook—with an assortment of conservative thinkers and pundits. As a serious thinker, he wanted to have his opinions on current events challenged by those with opposing political ideologies.

But it struck Mr. Pariser one day that he hadn’t seen a single status update from any of the loyal opposition in a while. Had his sources of conservative thought stopped posting? Had they unfriended him? No, Facebook had quietly stopped inserting their updates into his news feed on the site. Had the social-networking giant figured out that he was a liberal?

It turned out that Facebook had changed the algorithm for its news feeds, in response to its users’ complaints that they were being overwhelmed by updates from “friends” whom they hardly knew. The 600-million-member social network now filters status updates so that, by default, users see only those from Facebook friends with whom they’ve recently interacted—say, by sending a message or commenting on a friend’s post.

For Mr. Pariser, the algorithm change meant that his news feed was filtered to let him know about only the mostly left-leaning people with whom he bantered, leaving out conservative voices that he simply monitored. Facebook’s algorithm has no political parameters, but for Mr. Pariser it effectively muffled the people he most disagreed with but wanted to hear.

This sifting-out of seemingly dead connections—which might strike many people as a wonderful service—spurred Mr. Pariser to undertake a months-long exploration of the growing trend of personalized content on websites. In “The Filter Bubble,” he recounts what he found. “I was struck by the degree to which personalization is already upon us. Not only on Facebook and Google, but on almost every major site on the Web.”

It’s no secret that Amazon, for example, customizes its pages to suggest products that are most likely to be of interest, based on shoppers’ past purchases. But most Google users don’t realize that, since 2009, their search results have been gradually personalized based on the user’s location, search history and other parameters. By tracking individual Web browsers with cookies, Google has been able to personalize results even for users who don’t create a personal Google account or are not logged into one. Mr. Pariser asked two friends to search for “BP” shortly after the Deepwater Horizon oil spill last year. The two were shown strikingly different pages—one full of news about the disaster, the other mostly investor information about the energy company.

Personalization is meant to make Internet users happy: It shows them information that mathematical calculations indicate is more likely than generalized content to be of interest. Google’s personalized search results track dozens of variables to deliver the links that a user is predicted to be most likely to click on. As a result, Google users click on more of the results that they get. That’s good for Google, good for its advertisers, good for other websites and presumably good for the user.

But Mr. Pariser worries that there’s a dark downside to giving people their own custom version of the Internet. “Personalization isn’t just shaping what we buy,” he writes. “Thirty-six percent of Americans under thirty get their news through social networking sites.” As we become increasingly dependent on the Internet for our view of the world, and as the Internet becomes more and more fine-tuned to show us only what we like, the would-be information superhighway risks becoming a land of cul-de-sacs, with each of its users living in an individualized bubble created by automated filters—of which the user is barely aware.

To Mr. Pariser, these well-intended filters pose a serious threat to democracy by undermining political debate. If partisans on either side of the issues seem uninterested in the opposition’s thinking nowadays, wait until Google’s helpful sorters really step up their game.

Through interviews with influential Internet experts including Google News chief Krishna Bharat, Search Engine Land editor Danny Sullivan and Microsoft researcher Danah Boyd, Mr. Pariser exposes the problem with personalization: It’s hard enough for an army of researchers to create algorithms that can match each of us with things we like. It’s nearly impossible, by contrast, to craft a formula that will show us something we wouldn’t seek out but really ought to read—and will be glad we did. Beyond throwing random links onto a screen, it’s hard to model serendipity on a computer.

And there’s another problem with filters: People like them. The Internet long ago became overwhelming. Filters help make it manageable without our having to do the work of sorting through its content entirely by ourselves.

What to do? Mr. Pariser’s opening argument in “The Filter Bubble” is a powerful indictment of the current system. But his closing chapters fumble around in search of a solution—from individuals, from companies like Google or from government oversight. How do you tell the Internet to back it off a bit on the custom content?

For now, the best Mr. Pariser can hope for is to educate readers who don’t want to live in a solipsistic subset of the Internet, especially regarding political matters. Just knowing that Google and Facebook personalize what you see, and that you can turn it off if you want—on Facebook, click Most Recent instead of Top News atop your feed; for Google, get instructions by searching “deleting Web history”—is a good start. “The Filter Bubble” is well-timed: The threat is real but not yet pandemic. Major news sites are toying with personalization but haven’t rolled it out en masse. And in a test I conducted myself, I enlisted a handful of heavy Google users across America to search for “Bin Laden raid” soon after the event. The search results that came back were all nearly identical. To tell the truth, we were kind of disappointed.

Mr. Boutin writes about Internet technology and culture for MIT Technology Review, Wired and the New York Times.

The Filter Bubble
By Eli Pariser
The Penguin Press, 294 pages, $25.95