30 newyork| november11–24, 2019
in Minority Report; it has never played more than a supporting role
on Black Mirror; and in the Terminator movies, it’s a crucial safety
feature preventing the Terminator from terminating the wrong
people—to full-grown modern worry.
This spring, we found out the FBI’s facial-recognition database
now includes more than 641 million images, which can be searched
anytime without warrant or probable cause. And over the past few
months, spooked lawmakers have banned police use of the technol-
ogy in Oakland; Berkeley; Somerville, Massachusetts; and San
Francisco, of all places, where Orwellian tech products are the
hometown industry. But ever y where else and in all other con-
texts, facial recognition is still legal and almost completely
unregulated—it’s being used on us in city streets, airports,
retail stores, restaurants, hotels, sporting events, churches, and
presumably lots of other places we just don’t know about.
In 2019, we also heard plenty about facial recognition’sunreli-
ability: that misidentifications are common, and that the London
police’s system fails four out of five times, and that Amazon’s soft-
ware once mistook a bunch of seemingly upstanding congresspeo-
ple for criminal suspects. And maybe that meant, we hoped, that
the technology was still only half-baked and that our worries were
premature or that one big lawsuit could make all of it go away.
AndthenweheardaboutLarryGriffin
II.Andif thepolicethoughthisstory would
calmourfears,well,missionnotnecessarily
accomplished.Becauseifthesoftware
workswellenoughtoidentifyonebomb-
hoaxsuspectamong8.4 millionNewYork-
ersinunderanhourusingonlya coupleof
grainy surveillancephotos,maybeourwor-
riesaren’t soprematureafterall.Because
eventhoughfacialrecognitionisn’t fully
reliableyet, andmay neverbe,it’s already
transforminglawenforcement.Andit
probablywon’t needtobeperfectbeforeit
transformsoureverydaylives,too.
INA POSSIBLY RELATEDSTORY,perhaps
youhavenoticedanunusuallyhighdemandforpicturesof yourface
overthepastseveralyears?
Computerizedfacialrecognitionhasbeenindevelopmentsince
the1960s,butprogressspedupwiththerecentarrival ofadvanced
artificialneuralnetworks—i.e.,computersystemsthat canrecog-
nizepatternsbyprocessingexamples.Humanprogrammerscan
feedthesenetworksmany photosof faces,thenstepasidewhilethe
algorithmsteachthemselvestoidentify them.Themore photosan
algorithmhastolearnfrom,themore accurateit becomes.
Whichis whysomecompanieshaveaskedyoutouploadyour
vacationphotosandtagyourselfinthem,andothersaskedfora
selfieinexchangeforanautogeneratedcartoonof youortotellyou
whichcelebrityorRenaissancepaintingyoumost resemble.And
thensomecompaniesgotimpatientwithallthisasking,sothey
acquiredtheothercompaniesthat hadalreadycollectedyourpho-
tos,ortheyscrapedpublicimagesofyoufromsocialmedia,orthey
setuphiddencamerasinpublicspacestotake theirownpictures.
Howmany algorithmshavebeentrainedusingyourface?
Hard tosay, becausethere aren’tany lawsrequiringyourcon-
sent.ButFacebook,Google,Amazon,Apple,Microsoft,IBM,
anddozensofstart-upswithnameslike FaceFirst,FaceX,and
Truefaceallhavetheirownfacial-recognitionalgorithms,and
theyhadtohavelearnedsomewhere.
Some of those algorithms are pretty accurate, but they may be
even more powerful when used in combination with other long-
range biometrics. In China, gait recognition is already identifying
people based on the way they walk with 94 percent accuracy,
according to the marketing of one company offering the service.
And the Pentagon claims to have developed a heartbeat laser—that
is, an infrared beam that can read a person’s unique cardiac signa-
ture through a shirt or jacket, allegedly with 95 percent accuracy,
from a distance of 200 meters.
For a facial-recognition system to recognize you in the wild, it
needs a database with your identifying information in it, and you’re
probably in a database or two. If you’ve ever been tagged in a photo
on Facebook or Instagram, you belong to what is by Facebook’s own
claims the world’s largest facial- recognition database. The FBI’s
database comprises mug shots, the State Department’s entire direc-
tory of visa and passport pictures, and photos from the DMV in at
least 22 states (and counting), which allowed the agency to scan
residents’ driver’s licenses without their consent. (In Utah, Vermont,
and Washington, where undocumented immigrants can legally
drive, DMVs have shared facial-recognition data with Immigration
and Customs Enforcement.)
All of this would be bothersome enough in a world with perfect
data security, or in which you could get a new
face as easily as your bank replaces a stolen
debit card. But we live in neither of those
worlds, and faces have already been hacked.
In June, Customs and Border Protection
announced that hackers had breached the
servers of one of its subcontractors and sto-
len facial data, some of which reportedly
found its way to the dark web.SO FAR, MUCH OF THE WORRY over facial
recognition has been over the potential for
misuse by police. But there will be more quo-
tidian uses—and abuses—too. Commercial
facial recognition has been around for years,
but since there aren’t any laws requiring any-
onetodisclosethatthey’re using it on you, and commercial users
arenotsubject toFreedom of Information Act requests, it’s impos-
sibletosayhowprevalent it is. Which means any camera you pass
couldbediscreetlyrecognizing your face.
There’sa growinguniverse of vendors marketing packages cus-
tomizedtotheneedsof any type of business you can think of. The
TelAviv–basedFace-Six, for one, offers a suite of products including
FA6Retail(topreventshoplifting), FA6 Time and Attendance (to
catchlate-arrivingemployees), FA6 Med (for use in hospitals to
verifypatientidentities), Churchix (for churches that want to track
attendance),andFA6Drone (to identify targets from above).
Amongtheotherways we know facial recognition is already
beingdeployed:Airlines are using it to replace boarding passes, and
theDepartmentofHomeland Security says it plans to use facial
recognitionon 97 percent of airline passengers by 2023. At least
threearenashaveexperimented with the technology, including
MadisonSquareGarden. On stops of Taylor Swift’s recent Reputa-
tiontour, fans’faceswere scanned and searched against a database
ofherhundredsofknown stalkers.
LastyearinNewYork, residents of an apartment complex in
Brownsvilleprotestedwhen they found out their landlord wanted
to deployfacialrecognition to supplement their key fobs—but other
buildingsinthecityhave been known to use it for years, includingWhat ifyou’re not theperson facialrecognitionsays you are?