*New Scientist* arranged for early release of the article "Can you catch
a killer before they commit a crime?" which will appear in the August
11th issue.

Here's the article:

IMAGINE the scene. You arrive at New York's JFK airport, tired after a
long flight, and trudge into line at passport control. As you wait, a
battery of lasers, cameras, eye trackers and microphones begin secretly
compiling a dossier of information about your body.

The computer that is processing the data from these hidden sensors is
not searching for explosives, knives, guns or contraband. Instead, it is
working on a much tougher problem: whether you are thinking about
committing a terrorist act, either imminently, or at sometime during
your stay in the US. If the computer decides that might be your
intention, you will be led off for interview with security officers.

The equipment could also screen passengers as they wait to have their
bags checked before boarding, in an attempt to predict when someone is
planning to bomb or hijack a plane.

It sounds far-fetched, but this is the aim of Project Hostile Intent
(PHI), the latest anti-terrorism idea from the US Department of Homeland
Security. According to DHS spokesman Larry Orluskie, the DHS wants to
develop systems that can analyse behaviour remotely to predict which of
the 400 million people who enter the US every year have 'current or
future hostile intentions'.

PHI aims to identify facial expressions, gait, blood pressure, pulse and
perspiration rates that are characteristic of hostility or the desire to
deceive. Then the idea is to develop "real-time, culturally independent,
non-invasive sensors" and software that can detect those behaviours,
says Orluskie. The DHS's Advanced Research Projects Agency (HSARPA)
suggests that these sensors could include heart rate and breathing
sensors, infrared light, laser, video, audio and eye tracking.

PHI got quietly under way on 9 July, when HSARPA issued a "request for
information" in which it asked security companies and US government labs
to suggest technologies that could be used to achieve the project's
aims. It hopes to test them at a handful of airports, borders and ports
as early as 2010 and to deploy the system at all points of entry to the
US by 2012.

But experts in detecting when someone is deliberately hiding something
and training machines to recognise human emotions, say that the DHS
faces huge challenges, and is unlikely to achieve this goal by 2010, if
ever. "I can't imagine they will have any reasonable rates of success
with such a system," says Kerstin Dautenhahn of the University of
Hertfordshire, UK, who specialises in teaching robots to understand
human intentions. "I have serious doubts that it will be successful,"
adds psychologist Paul Ekman of the University of California, San
Francisco, an expert in detecting hidden emotions and intentions from
human facial expressions.

We already know that people betray their true intentions via involuntary
behaviour. In the 1960s Ekman found that even when people are trying to
hide it they often reveal what they are about to do, by showing
fleeting, involuntary facial expressions known as "micro-expressions".
For example, if for a fraction of a second you bare your teeth, lower
your eyebrows and wrinkle your nose, while pretending to smile, you've
just made the micro-expression for disgust.

Since 2003, the US Transportation Security Administration (TSA) has been
using a program called Screening Passengers through Observation
Techniques, which relies on micro-expressions. Under SPOT, dedicated
"behaviour detection officers", who are trained to observe and decipher
micro-expressions, observe people milling around at airports and
discreetly pull aside anyone whose micro-expressions seem suspicious.
After starting a casual conversation, they might then pass them on for
further questioning, depending on their responses. "We have caught a
number of individuals, from drug dealers to money launderers, and a
double murderer in one case," says TSA spokesman Chris White.

A big problem, however, is that SPOT is an expensive, labour-intensive
process and is not something a customs official or baggage screener can
do in addition to their normal work. "Right now, screeners have
typically less than one minute to examine a traveller's documents and
assess whether they are a threat," says Orluskie. Similarly, the
infamous polygraph or lie detector test, used routinely by intelligence
agencies across the world when grilling suspects - despite its
questionable reliability - is time-consuming and requires an officer's
undivided attention, as suspects must be hooked up to electrodes that
measure blood pressure, sweat and pulse.

Enter PHI. With this latest idea, the DHS is hoping to automate the SPOT
program, so that computers, not humans, search for micro-expressions,
and at the same time beef up the range of bodily signs that can be
investigated. Machines will not just look for micro-expressions, they
will also attempt to sense whether someone is hiding something. For this
they might use a remote-controlled, non-contact version of the
polygraph, bouncing lasers or microwaves off a person's skin, as
suggested by the US Department of Defense in 2006. The DHS wants to use
remote sensors so they don't impede the flow of travellers.

Ekman is sceptical of PHI. How do you identify hostile micro-expressions
in a potential terrorist, when they are likely to be highly complex and
you don't know what they are" "Would they show fear of being caught,
contempt for the likely victims, sadness at the prospect of being
killed, joy at the possibility of soon going to paradise" I don't know.
No one knows," he says.

Another unknown is whether you could detect expressions of hostility
hours or even months before the act, when a would-be terrorist first
enters the US. Ekman is currently observing police in order to train
them to recognise an impending violent act seconds before it occurs, via
micro-expressions, "enough time to take protective action," he says.
"What you are looking for at an airport is a far different context."

Gathering the data to work out what the predictors are will be hard,
says Peter McOwan, a computer scientist at Queen Mary, University of
London, who is developing software and sensors that let computers
respond to our moods. "Most prediction techniques are based on broad
statistical data gathered from a wide range of sources on which the
system is trained," he says. A system then learns what is normal
behaviour, and what is hostile, based on hundreds or thousands of
examples of each.

In the case of terrorists planning a hostile act, it is not clear where
you will find the examples of hostile intent necessary to train the
software.

Dautenhahn also notes that airports are high-stress places with people
who are tired, bored, saying tearful goodbyes or carrying upset
children. So anything that looks for signs of stress such as sweat
detection might flag up innocent people. Ekman agrees. "Are we to stop
and question anyone who shows those emotions"" he asks. Orluskie says
the DHS is prepared to risk failure if there is the slightest chance
that national security can be improved. But McOwan thinks that's
unlikely: "It's just like something from Minority Report. They have been
watching too many Tom Cruise movies."