Facebook Graph Search

» see facebook.com as it lived, on the wayback machine

~ the obit ~

graph search never spoke turkish. that is not a figure of speech about cultural distance, it is the literal product boundary: facebook shipped it for us english, added uk english later, and never localized it into anything else in six years of life. set your account language to türkçe and the box at the top of the page stayed an ordinary keyword search. so for most of the world, including where i was sitting, this thing existed only as tech news about a feature other people had.

it was announced on 15 january 2013 at a press event in menlo park, with zuckerberg calling it a third pillar of the site next to the news feed and the timeline. the idea was that facebook already held a structured map of who knew whom and who liked what, and that a search box reading that map could answer questions no keyword engine could. photos of my friends taken in london. single men in san francisco who like star wars, which was one of the examples facebook gave itself.

underneath was an index called unicorn that facebook had been building since 2011, and the people running it were both ex-google: lars rasmussen, who had co-created google maps, and tom stocky. rasmussen said afterward that the whole thing started on a walk with zuckerberg in spring 2011 and was running internally by that summer. for anything the graph could not answer, results came from bing under a partnership that lasted until december 2014.

the technology worked. that turned out to be the problem, because a search engine over a social graph does not distinguish between a query you would be happy to be found by and one you would not, and facebook had spent nine years persuading hundreds of millions of people to mark things public without ever showing them what public would eventually mean.

~ the rap sheet ~

~ the tumblr ~

eight days after the beta went out, tom scott put up a tumblr called actual facebook graph searches. it was screenshots, nothing more, of queries that were possible right then against real accounts. mothers of catholics from italy who like durex. married people who like prostitutes. employers of people who like racism. islamic men interested in men who live in tehran, in a country where that answer can get somebody killed.

it took a quarter of a million visitors in a day and was in the press within forty-eight hours. scott's own point was the sharp one and he made it repeatedly: every person in those screenshots had set that information to public themselves. nothing had been leaked and nothing had been broken into. the data had always been there and it had simply never been this easy to ask for. a privacy setting that depends on your answers being hard to collect is not a privacy setting, it is an accident that lasted a while.

~ the long quiet removal ~

facebook's response was to take things out rather than to put things back. in august 2013, when graph search finished rolling out across us english, the company removed the old setting that let you keep your timeline out of search by name, on the reasoning that it no longer protected anything given how much else was findable. that trade was made in the direction you would expect.

in december 2014 the bing deal ended and graph search stopped being visible in the interface, though the queries still ran if you built the url by hand. that is the state it lived in for four and a half years: officially gone, practically alive, and increasingly the private tool of people who knew the url format. by then that group was mostly investigative journalists and human rights researchers, plus every skip tracer and background-check firm in america.

~ the day it stopped ~

on thursday 6 june 2019 the hand-built urls stopped returning anything. the tools that had been built on top of them, the ones journalists and investigators had been quietly relying on for years, broke together within a day. facebook made no announcement and confirmed it only when reporters asked, saying that the vast majority of people search using keywords, which led them to pause some aspects of graph search.

the timing was not subtle. this was fifteen months after cambridge analytica and six weeks after facebook told its investors it was facing a record ftc penalty over privacy, and the thing being switched off was a tool whose entire function was making the graph legible to outsiders. the useful reading is that graph search was never really killed by low usage. it was killed by working exactly as designed, at a moment when a company under that much scrutiny could not defend a feature that answered questions about its users better than its users understood.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »