59
electrode-studdedheadsetcalledan
electroencephalogram,totranslatehis
neuralactivity intowavesrisingandfalling
on agraph.Thetest’s inventor,DrLarry
Farwell, claimsitcandetectknowledge of
acrimehidden inasuspect’sbrain bypicking
up aneural response tophrases orpictures
relating tothe crimethatonlytheperpetrator
wouldrecognise.
Thedevelopmentofothermethods ofbrain-
based lie-detectionwassteppedup after the
9/11terrorist attacksin 2001, whentheUS
government–longasponsorofdeception
science–startedfunding research through
Darpa,theDefenseAdvancedResearch
ProjectsAgency.Now,anew frontieris
emerging.Anincreasingnumber ofprojects
useAItocombinemultiple sourcesofevidence
intoasinglemeasure. Machinelearning is
acceleratingdeception research.Scientists in
Maryland,forexample,havedevelopedsoftwarethey claimcan
detectdeceptionfromcourtroomfootagewith88%accuracy.
Thealgorithmsbehind such toolsare designedtoimproveover
time,and mayultimately endup basingtheirdeterminations of
innocenceandguiltonfactorsthateven thehumanswho
programmedthemdon’tunderstand.Thesetestsarebeingtrialled
in jobinterviews,at border crossingsand inpolice interviews,but
astheybecome morewidespread,civilrightsgroupsandscientists
are growingconcernedaboutthedangerstheycould unleash.
Nothingprovidesaclearer
warningaboutthethreats
ofthenew generation of
lie-detectionthanthe history
oftheworld’smostwidelyused
deceptiontest.Almostacentury
old,thepolygraphstilldominatesour view ofliedetection, with
millions oftestsconductedaround theworldeveryyear.In1921,
29-year-oldJohnLarsonwasarookiepoliceofficer inBerkeley,
California.Havingstudied physiology andcriminology,he was
alsoworkingpart-timein aUniversity ofCalifornialab,where
he builtadevicethattookcontinuousmeasurementsof blood
pressure and breathingrate,andscratchedtheresultsontoa
rollingpaper cylinder.Hethendevised aninterview-based exam
thatcomparedasubject’sphysiologicalresponse when answering
“yes” or“no”questions relatingtoacrimewith thesubject’s
answerstocontrolquestions,suchas“IsyournameJaneDoe?”
Fromthe late1920s,thepopularityof Larson’sinventiontook
off–notleastwith the USgovernment, whichbecametheworld’s
largestuseroftheexam.Duringthe“red scare”ofthe 1950s,
thousandsofemployees were subjected topolygraphs designed
to root outcommunists. Formuch of the last century, many US
corporations alsoran polygraph tests toquizemployeesover such
issues as drug useandtheft.
The only problemwas thatthe polygraph did not work.History
is litteredwithexamples of criminalswho evaded detectionby
cheatingthe test:common “countermeasures”, which workby
exaggeratingthe body’s response to control questions, include
thinking aboutafrightening experience, or simplyclenchingthe
anus. Thepolygraphmachineisnot andnever was an effective
liedetector.Thereisnoway foranexaminertoknowwhether
arise inbloodpressure isdueto fear of getting caughtinalie,
or anxiety about being wrongly accused. Aslong agoas1965,
the year Larson died,the USCommittee on Government
Operations issuedadamning verdictonthe polygraph.“People
have beendeceivedby amyth thatametal boxinthe hands of
an investigator candetect truth or falsehood,” itconcluded.
The polygraphremainedpopular though–not becauseitwas
effective,butbecausepeoplethoughtit was.The threatof being
outedbythe machine was enough to coerce someintoconfession.
One examiner inCincinnati in1 975 leftthe
interrogationroomandwatched, bemused,
throughatwo-waymirrorastheaccused
tore1.8metresofpapercharts offthe machine
and atethem.(Youdidn’tevenhavetohave
theright machine: inthe1980s,policeofficers
in Detroitextracted confessionsbyplacinga
suspect’s hand onaphotocopier that spatout
sheetsofpaperwiththe phrase“He’s Lying!”
pre-printed onthem.)Larsonhimself
recognisedthecoercivepotential ofhis
machine,describingit shortlybeforehis
deathas“aFrankenstein’smonster”.Thesearchfor atrulyeffective lie detector
gainednewurgencyafter9/11.Several ofthe
hijackershadmanaged toentertheUS after
successfullydeceivingborder agents.Suddenly,
intelligenceand borderserviceswanted tools
thatactuallyworked.Aflood ofnew
governmentfunding madeliedetectionbig
businessagain. Morerecently,the need toidentifyEuropean
terrorists returning fromreceivingtrainingabroadhasproduced
asimilareffecton theborders oftheEU.In2014,travellersflying
intoBucharestwereinterrogated byavirtualborderagentcal led
Avatar,an on-screen figure withblueeyes,whichhasa
microphone,aninfra-redeye-trackingcameraand asensorto
measure body movement.But its “secret sauce”,say its makers,
isin thesoftware, which usesanalgorithmtocombineallofthese
typesof data.Avatar’s accuracyratesareclaimed tobe over 80%
in preliminary studies.Newtechnologiesmaybeharder
thanpolygraphsforunscrupu-
lousexaminersto manipulate,
butthatdoesnotmeantheywill
be fair.Like their predecessors,
AI-poweredlie detectorspreyon the tendency ofbothindividuals
andgovernmentstoputfaithin science’s supposedlyall-seeing
eye.Buthistorytellsus thatthey mayget aimed atsociety’smost
vulnerable–suspecteddissidents andhomosexuals inthe 195 0s
and1 96 0s,benefitclaimantsinthe 2000 s,and asylum seekers
andmigrants today.Oneday,improvements inAIcould findareliablepatternfor
deceptionby scouring multiple sourcesof evidence,or more
detailedscanning technologiescoulddiscoveranunambiguous
sign lurkingin thebrain. Inthe real world,however,practised
falsehoods–the stories wetell ourselvesaboutourselves,the lies
thatformthecoreof ouridentity–complicatematters.“We
have this tremendouscapacitytobelieve our ownlies,” said Dan
Ariely,arenowned behaviouralpsychologistatDukeUniversity.
“And once we believeourownlies, of course we don’tprovide
anysignal of wrongdoing.”In his1995 science-fiction novelTheTruth Machine,James L.
Halperin imaginedaworldinwhichsomeone succeedsinbuilding
aperfect liedetector. Theinventionhelps unitethe warring
nations intoaworldgovernment,andaccelerates the searchfor
acancercure. Butevidence from thelasthundred yearssuggests
that it probablywouldn’t playoutlikethat. The scientist Daniel
Langleben toldmethatone ofthegovernmentagencies he was
approached by wasn’tint erestedin theaccuracyrates of hislie
det ector, which uses functional magneticresonanceimaging,or
fMRI.AnfMRI machine cannotbepackedintoasuitcaseor
broughtinto apoliceinterrogationroom.The investigator cannot
manipulatethe test resultstoapply pressuretoanuncooperative
suspect. Theagencyjust wantedtoknowwhetherit couldbeused
totrain agents tobeat the polygraphtests of others.“Truth is not
reallyacommodity,” Langlebenreflected. “Nobody wantsit.”Alonger versionofthis articleappear ed in The Guardian.
©Guardian News&Medi aLtd 2019.The last word2November 2019 THE WEEKAvery:underwentbrainfingerprinting“Duringthe‘redscare’ofthe1950s,thousands
of employees were subjected to polygraph tests.
The only problem was that they did not work”