Wear Plaid and Hang Out at Coffee Shops? In the Future, Facebook May Know You’re a Hipster.

Artificial intelligence is coming for your social media.

National Journal
Brian Resnick
Add to Briefcase
See more stories about...
Brian Resnick
Dec. 16, 2013, midnight

So you grow some scruff, start wear­ing tight jeans, oc­ca­sion­ally throw on sus­pend­ers, and reg­u­larly don plaid. You start hanging out at that ar­tis­an­al kom­bucha spot down the street — you know, the one with all the wick­er? And then one morn­ing — be­fore you wax that 360-de­gree handle­bar — when you log onto Face­book, you no­tice something dif­fer­ent.

There’s one ad for sock sus­pend­ers, and an­oth­er one for Amish-made beard con­di­tion­er*.

At parties and on the streets, it’s ob­vi­ous — you’re a total hip­ster. And now Face­book knows it too, just by the pho­tos you’ve up­loaded.

While this dysto­pi­an scen­ario is still set in an un­deter­mined fu­ture, the idea that so­cial me­dia can scan your pho­tos for con­tex­tu­al clues about life­style has already been proven. In his lab at Uni­versity of Cali­for­nia San Diego, Serge J. Be­longie and his stu­dents have cre­ated a com­puter al­gorithm that can guess what “urb­an tribe” (i.e., sub­cul­ture) you be­long to.

Hu­mans can do this al­most without think­ing. We hold ste­reo­types of what the word “hip­ster” im­plies, but we know that not one of these vari­ables com­pletely defines the term. A “hip­ster” is the sum of many parts, like a plaid shirt or tight jeans.

This concept — the abil­ity to use con­tex­tu­al clues to define an ob­ject — is what artists and psy­cho­lo­gists call gestalt. But that type of cog­ni­tion is not easy for a com­puter, which op­er­ates on hard rules.

“That’s the theme in my re­search,” Be­longie says. “Whenev­er I stumble across a pat­tern that people can see but is not im­me­di­ately evid­ent from a quant­it­at­ive per­spect­ive, I’d like to probe that to fig­ure out what’s caus­ing that.”

Be­longie and his team com­pile ref­er­ence pho­tos for each urb­an tribe and then “teach” the com­puter to re­cog­nize the pat­terns in col­ors, tex­tures, and geo­met­ric ar­range­ments. The com­puter then “combs through the data­base and looks for ex­cep­tion­al ex­amples of each of those cat­egor­ies.” In oth­er words, Be­longie is try­ing to teach the com­puter the gestalt it needs to nav­ig­ate a high school cafet­er­ia.

Once the com­puter gets a feel for these pat­terns, it can start guess­ing. Be­longie’s pro­gram isn’t per­fect. The ma­chine op­er­ates at just 39 per­cent bet­ter than chance, Be­longie says. “This is something that we’re just crack­ing open a new re­search area.”

A Seis­mic Shift

Be­longie’s work goes bey­ond hip­sters and goths. This was just an off­shoot of his more sober work — teach­ing com­puters to re­cog­nize can­cer in a biopsy slide or dead cor­al from satel­lite pho­tos.

And yet, he says his re­search can’t ig­nore the pull of so­cial me­dia. “In the old days,” he says, “It was NSF and NIH and DARPA grants — that’s what was kept this field rolling.” But now com­pan­ies like Face­book and Google are in­vest­ing heav­ily in ar­ti­fi­cial in­tel­li­gence, and pick­ing off his stu­dents when they gradu­ate.

The de­fin­ing story of the dec­ade con­tin­ues to be the ap­plic­a­tions of big data. Yes, the Snowden leaks were the biggest story of the year, but they speak to the lar­ger trend of what huge in­sti­tu­tions, such as the U.S. gov­ern­ment, can do when they can pro­cess tera­bytes of data quickly in­to us­able in­form­a­tion. But here’s where that story is start­ing to shift. Big data is go­ing to gain (great­er) con­scious­ness. Or at least that’s where the re­search money is head­ing.

Just re­cently, Face­book an­nounced it is open­ing a gi­ant ar­ti­fi­cial in­tel­li­gence lab to be pi­loted by Yann LeCun, a re­search­er in the field from New York Uni­versity. “The set of tech­no­lo­gies that we’ll be work­ing on is es­sen­tially any­thing that can make ma­chines more in­tel­li­gent,” LeCun tells WIRED about the ini­ti­at­ive.

Google, too, re­cently hired Ray Kur­z­weil, a fu­tur­ist who has pre­dicted that com­puter in­tel­li­gence will ex­ceed hu­man in­tel­li­gence by 2045, to work on ma­chine learn­ing pro­jects. “I en­vi­sion, some years from now, that the ma­jor­ity of search quer­ies will be answered without you ac­tu­ally ask­ing,” he told Sin­gu­lar­ity Hub after he was hired. “It’ll just know this is something that you’re go­ing to want to see.”

Which is kind of creepy, right?

Let’s put it in­to this con­text: Cur­rently Face­book does a lot with its user data, such as run­ning mass so­cial ex­per­i­ments on Elec­tion Day, and re­search­ing at which rates wo­men take the names of their hus­bands when they get mar­ried (the so­cial net­work makes all these stud­ies pub­lic). But if that’s creepy, what’s to come is still creepi­er.

“There is a huge, seis­mic shift that is hap­pen­ing, that every com­pany wants ar­ti­fi­cial in­tel­li­gence in some way,” Be­longie ex­plains. “They are just drown­ing in big data.”

And an enorm­ous amount of that data is in im­age form, which up to now, hasn’t fed back mean­ing­ful in­form­a­tion about users. “These im­ages are just sit­ting in these serv­ers like dark mat­ter,” he says.

Be­longie says it’s part of the “nat­ur­al pro­gres­sion,” that these com­pan­ies will use pro­grams like his own “for good or for ill — but most likely just to serve up more rel­ev­ant ads.”

I asked Be­longie if he thought his own work was “creepy.”

“I’ve lost my ob­jectiv­ity,” he laughs, “be­cause I find these prob­lems so in­ter­est­ing from a sci­entif­ic per­spect­ive. I don’t find it creepy, but I think people need to be aware of what they are shar­ing.”

*Ed­it­or’s Note: A won­der­ful product.

COR­REC­TION: This art­icle orig­nally un­der­stated the ac­cur­acy of the Urb­an Tribe pro­gram. The sys­tem guesses cor­rectly 48 per­cent of the time.

What We're Following See More »
Bill Murray Crashes White House Briefing Room
3 hours ago

In town to receive the Mark Twain Prize for American Humor at the Kennedy Center, Bill Murray casually strolled into the White House Briefing Room this afternoon. A spokesman said he was at the executive mansion for a chat with President Obama, his fellow Chicagoan.

CFPB Decision May Reverberate to Other Agencies
6 hours ago

"A federal appeals court's decision that declared the Consumer Financial Protection Bureau an arm of the White House relies on a novel interpretation of the constitution's separation of powers clause that could have broader effects on how other regulators" like the Office of the Comptroller of the Currency and the Federal Housing Finance Agency.

Morning Consult Poll: Clinton Decisively Won Debate
6 hours ago

"According to a new POLITICO/Morning Consult poll, the first national post-debate survey, 43 percent of registered voters said the Democratic candidate won, compared with 26 percent who opted for the Republican Party’s standard bearer. Her 6-point lead over Trump among likely voters is unchanged from our previous survey: Clinton still leads Trump 42 percent to 36 percent in the race for the White House, with Libertarian nominee Gary Johnson taking 9 percent of the vote."

Twitter Bots Dominated First Debate
7 hours ago

Twitter bots, "automated social media accounts that interact with other users," accounted for a large part of the online discussion during the first presidential debate. Bots made up 22 percent of conversation about Hillary Clinton on the social media platform, and a whopping one third of Twitter conversation about Donald Trump.

Center for Public Integrity to Spin Off Journalism Arm
7 hours ago

The International Consortium of Investigative Journalists, the nonprofit that published the Panama Papers earlier this year, is being spun off from its parent organization, the Center for Public Integrity. According to a statement, "CPI’s Board of Directors has decided that enabling the ICIJ to chart its own course will help both journalistic teams build on the massive impact they have had as one organization."


Welcome to National Journal!

You are currently accessing National Journal from IP access. Please login to access this feature. If you have any questions, please contact your Dedicated Advisor.