When Mark Zuckerberg addressed Facebook investors last week, he outlined a goal that sounded more like a philosophical quest than a business plan: A major priority for Facebook, he said, was "understanding the world."
No, this wasn't the kick-off to Zuckerberg's journey to self-discovery. But it did mark a new phase in Facebook's journey to discover us in ways that are ever more detailed. Thanks to artificial intelligence, a technology Zuckerberg highlighted as a key focus, Facebook may soon understand not only what we do on the social network, but what we mean. In short, what we tell Facebook is no longer enough. The social network now wants to read between the lines. It's a critical difference that could have Facebook's members -- and advertisers -- using the site in entirely new ways.
Facebook's mission to "understand the world" underscores its growing rivalry with Google and attempt to position itself as a social search engine. With a more thorough grasp of what we're saying about the world, Facebook could also answer more of our questions about the world. And that, in turn, would be a sweet deal for advertisers, who interpret "searching" as "shopping."
In the beginning, Facebook was like a gargantuan junk drawer of memories -- a place where we collected, in any kind of haphazard way, photos, small notes to and from friends, notebooks containing our memories and musings, business cards from restaurants, coasters from favorite bars, and all the other memory knick-knacks. We fed it little bits of our lives -- things we wanted to keep, but no one could be bothered to organize. We could dig up the past, but it would take some rummaging.
In the past year, however, Facebook has been progressively, aggressively cleaning up that drawer. Facebook’s new AI team made 1.2 trillion comments and status updates searchable. Our more than 1 trillion social connections can now also be mined with queries like, “friends of friends who are single in San Francisco.”
On Wednesday, Zuckerberg made clear these kinds of cute searches are just the appetizer. He wants people to be able to "easily ask any question to Facebook and get it answered." (Emphasis added.)
This first requires getting to know us better. Instead of profiling us by what we share explicitly, through our likes, check-ins and tags, Facebook is poised to master some of the messier, more complex information we post to the site, like our comments or our status updates. As Zuckerberg noted, “The goal here is to use new approaches in AI to help make sense of all the content that people share.”
More specifically, Facebook could dig deeper into our lives by harnessing the power of natural language understanding, a sub-field of artificial intelligence that teaches computers to decode the intricacies of human language. Siri, for example, uses natural language understanding to know that "Do I need an umbrella?" refers to the weather, or "What's playing?" to movie times. Deep learning, another area of AI that lets computers find emotions or events that aren't explicitly stated in text, is also receiving concerted attention from Facebook, the company's Chief Technology Officer Mike Schroepfer told the Technology Review earlier this year.
The social network’s current search tool is still a rather crude instrument: We can't find our friends’ comments about “Breaking Bad” unless they included the words “Breaking Bad.” A search on Facebook for “New York restaurants” would skim right over a friend’s rave review of the Manhattan taqueria Dos Toros -- unless she’d specifically written the review with search speak in mind, noting it was one of many delicious “New York restaurants.”
But with a little natural language understanding, Facebook could train itself to "read" its members' billions of posts. Take me, for example -- a Facebook user who’s stingy with her “likes.” Facebook might not know about my obsession with electronic music because I’ve never "checked in" at a club, or “liked” a DJ's page. But with a little clever natural language processing, it might be able to deduce my interest in the genre if my posts have enough references to “the David Guetta show next weekend” or “Avicii’s new single.” Facebook might then know to highlight me to a friend who searches the site for people to join him at a techno concert. A query for "friends who like Avicii" could become "friends who like electronic music" -- and go beyond those who've "liked" some official page.
The motivation for all this harkens back to the golden rule of the web: To know us better is to sell us better.
"Just 'liking' something doesn’t mean we’re interested in it," explains Nick Cassimatis, founder of SkyPhrase, a startup that uses natural language understanding to organize and answer questions about large data sets. “If you can understand what people’s status updates are about, you can build a model of who they are and what they’re into, and target advertising to them."
Zuckerberg took that even further: He promised in the earnings call that Facebook could turn our data points into “the clearest model of everything there is to know in the world.”
Once Facebook knows everything there is to know in the world, it hopes it can quickly become our go-to source when we have any question about a purchase, a job, a passion, or exploring a new city. Being able to “read” our conversations with one another and infer meaning from them allows Facebook to more comprehensively answer a far greater range of queries than it once could. Imagine someone looking for advice on a new backpack to use while camping. A search for “friends of mine who like camping” currently returns anyone who’s added "camping" to their list of interests. But with help from AI, Facebook could use words like “tent,” “national park,” and “campsite” in friends’ status updates to deduce who in your social circle loves the great outdoors, then serve up a more complete list of experts. Searches could go from the general -- “friends who like camping” -- to the specific -- “friends who can recommend good backpacks to buy,” or even “friends who can tell me about REI’s backpacks.”
These more focused queries could be a boon for advertisers, who might be able to better target your interests. "The more precisely they can understand what your query is, the better they can target ads to you," notes Cassimatis.
Facebook's most ambitious plans of all could be in the realm of voice recognition, or training computers to understand what we say. "I’m looking forward to us doing a lot more here," Zuckerberg noted cryptically in the investor call.
This technology could transform Facebook from a passive receptacle for our memories to an active transcriber of our everyday interactions. Zuckerberg's reference to voice recognition calls to mind a 2012 interview with Facebook's then-vice president of advertising, Andrew Bosworth. Bosworth explained that Facebook could one day use a phone's microphone to pick up the sound of music or a person humming -- then deliver ads for concerts, media or content.
We'll be able to ask more of Facebook. But it's asking more of us, too.
BEFORE YOU GO
How to vote
Vote-by-mail ballot request deadline: Varies by state
For the Nov 3 election: States are making it easier for citizens to vote absentee by mail this year due to the coronavirus. Each state has its own rules for mail-in absentee voting. Visit your state election office website to find out if you can vote by mail.Get more informationTrack ballot status
In-person early voting dates: Varies by state
Sometimes circumstances make it hard or impossible for you to vote on Election Day. But your state may let you vote during a designated early voting period. You don't need an excuse to vote early. Visit your state election office website to find out whether they offer early voting.My Election Office
General Election: Nov 3, 2020
Polling hours on Election Day: Varies by state/localityMy Polling Place