Pre-Crime Facial Recognition System Developed
By PNW Staff June 14, 2016
Share this article:
One day next year, you try to board a plane but are stopped by
security. The automated facial recognition system has informed the
officials that you look like a terrorist and might pose a security risk.
A few days later, you walk into a bank to apply for a
mortgage, but are turned away before you can explain because once again
the automated facial analysis software the bank uses has determined you
have features associated with gamblers.
You
have no criminal record and your credit rating is fine, but that doesn't
matter to the software behind the computer system known as Faception.
Based
on an image, it matches facial features across its database to predict
behavior before it occurs and label you with such tags as pedophile,
terrorist, gambler or thief based solely on the geometry of your face.
The
Israeli startup Faception claims to already be in talks with the
Department of Homeland Security in the United States about employing its
system for which it has, according to marketing material, "built 15
different classifiers, including extrovert, genius, academic researcher,
professional poker players, bingo player, brand promoter, white collar
offender, pedophile and terrorist."
Faception
claims to have identified 9 of the 11 terrorists involved in the Paris
attacks when its criminal detection algorithms were applied to photos
after the attacks and the overall accuracy is estimated at 80% by the
company.
The system, it must be remembered,
does not perform facial recognition in the traditional sense, not
bothering to attach a name to a face.
Using
facial features to predict traits, it instead lies much closer to 19th
century theories of phrenology, a now debunked pseudoscience that used
the shape of the skull to predict a host of factors about a person's
mind and behavior.
While few would argue the
power of body language and facial expression to indicate our mood and
attention, alerting a security guard to a nervous commuter for example,
the notion that inherent criminality can be found in the shape of one's
eyes or the position of a nose is both deeply troubling and doubted by
most experts.
But just like no-fly lists that prevent toddlers who share
names with terrorists, let's not rest assured that truth, racism or
effectives will be an impediment to implementing Faception.
Yet
true facial recognition carries its own dangers as well. In the past
century, the average person enjoyed a measure of privacy through
obscurity. Walk down the street of a major city and there was little
chance of being recognized among thousands of anonymous faces.
That
is all changing now. As facial recognition systems continue to be
refined and produced for smartphones, instant recognition of any face is
in the hands of the public at large not to mention tens of thousands of
CCTV cameras.
One such application is called
FindFace, an app developed by two Russian programmers that allows anyone
to use facial recognition software and their smartphone camera to
identify faces on the street and link each to a real name with 70%
accuracy.
The system cross references facial
geometry with photos found on social networks to produce name matches
for anyone passing on the street, sitting on a bus or entering a store.
It
is not hard to imagine uses for such a system for those who would like
to identify protestors, find the name of their next stalking victim or
perhaps ID a witness or an adult film actress.
If
the potential for abuse with FindFace and its link to social networks,
then consider the power of the UK site Facewatch, an online site that
acts as a public watch list for known face profiles and which has now
been integrated with facial recognition software, NEC's NeoFace, and
private surveillance cameras.
As reported
first by Ars Technica in 2015, the intention is for a shop owners to
place a thief or unruly customer's face on the public watch list which
will allow every security camera across the world to flag that
individual and deny him entry into other restaurants and stores.
What
could possibly go wrong with a crowd sourced black list for automated
social exclusion? Certainly a waiter who gets stiffed on a tip wouldn't
add your face to the list.
With any of the
proposed systems, there exists the danger of not only tearing away the
last layers of public privacy that remain but also of placing an
inordinate level of trust in the accuracy of those systems through a
phenomenon known as automation bias by which we tend to trust automatic
systems more than humans now.
Aside from the
dubious claims of predicting behavior simply on one's facial features, a
premise that appears on the surface to be little more than automating a
centuries old process of racial profiling, the rapid advances of facial
recognition could spell an end to any semblance of privacy in public.
Soon,
these systems promise, our faces will be automatically entered into
government, commercial and crowd sourced databases open to advanced
recognition algorithms.
No comments:
Post a Comment