3 November 2018 | NewScientist | 5For new stories every day, visit newscientist.com/news
What’s in your suitcase? If you
open the suitcase and show me
what is inside, will it confirm that
your answers were true?
These are just two of the
questions that an automated
lie-detection system will ask
travellers during a six-month
pilot starting this month at
four border crossing points in
Hungary, Latvia and Greece with
countries outside the European
Union. It will be coordinated by
the Hungarian National Police.
The lie detector uses artificial
intelligence and is part of a new
tool called iBorderCtrl, developed
by a Europe-wide consortium.
The pilot will involve actual
travellers, who will be invited to
use the system after they have
passed through border control. It
won’t affect their ability to travel.
But the plan is that the system will
eventually be able to grant people
permission to cross a border by
automatically assessing a range
of information, including official
documents, biometric data and
social media activity – as well as
the truthfulness of responses to
security questions.
The web-based tool is intended
to make crossing into the EU
quicker and safer, and identify
anyone wishing to break laws
when entering a country, such
as staying longer than allowed.
Yet several independent experts
contacted by New Scientist
expressed strong reservations
about the idea, questioning
the accuracy of automated
lie-detection systems in general.
In principle, automating the
detection of deceit makes sense,
as people are terrible at spotting
liars, barely better at identifying
them than chance, according to a
meta-study in 2006.
Instead of a human asking
you questions at the border, theinterview stage of iBorderCtrl asks
you questions via a virtual border
guard on your laptop or phone.
When you give your answers, the
device’s camera films your face
This video is then analysed by
AI software that looks at 38 micro-
gestures to spot patterns, such as
slight movements of an eyelid.Some say such gestures are
associated with lying, but the idea
is controversial and has little
evidence, says Bennett Kleinberg
at University College London, who
works on crime prediction and
isn’t associated with the project.
The system scores each
response and if you pass the test –
documents in order, answers
deemed truthful – you will be
given a QR code that you can scan
when you reach the border to be
let through. If it thinks you arelying, the virtual border guard
becomes more stern, changing its
expressions and manner. It will
refer you to a human guard when
you reach the border, who will have
access to your iBorderCtrl report.
The lie-detector component is
adapted from an existing system
created by some members of the
team, and was tested recently in a
small experiment with 30 people.
The participants were asked a
series of questions by the virtual
border guard. Half the group was
asked to tell the truth and the
other half to lie. The software
identified those lying with around
76 per cent accuracy, according
to results the team recently
presented at a conference.
The team acknowledges that
this isn’t high enough and plan to
improve the system’s accuracy by
training it on the much larger data
set that will be obtained during
the pilot. “We’re quite confident
of bringing it up to the 85 per cent
level,” says Keeley Crockett at
Manchester MetropolitanUniversity, UK, one of the team.
However, this is still very low
accuracy. Were the system to be
rolled out across the whole EU,
many millions of people could be
flagged as liars each year.
The key problem with training a
lie-detection algorithm on people
who have been told to lie is that
genuine liars have different tics
than people who are acting, says
Maja Pantic at Imperial College
London. “If you ask people to lie,
they will do it differently and
show very different behavioural
cues than if they truly lie, knowing
that they may go to jail or face
serious consequences if caught,”
she says. “This is a known problem
in psychology.”
Crockett says the team is aware
of the pitfalls and the main point
of the pilot is to identify and
address them. It will also train the
border guards using the tool to
understand its limitations. ■AI to interrogate travellers
Trial of automated lie-detector test set to begin at some EU borders, reports Douglas Heaven
OLI SCARFF/GETTYA virtual border guard could soon
be asking the questions“If the AI system thinks you
are lying, the virtual border
guard becomes more stern,
changing its expressions”