[AI] Motion-sensing phones that predict your every move

Sanjay ilovecold at gmail.com
Tue Apr 20 02:37:50 EDT 2010

          Phones that learn their user's patterns of behaviour can use this
          information to provide a cheaper, more reliable service

by Paul Marks

COULD your cellphone learn to predict what you are going to do before
you've even started doing it?

Communications engineer Arjen Peddemors thinks so, and along with
colleagues at the Technical University of Delft in the Netherlands he
has devised a system that learns users' behaviour patterns to provide
them with an enhanced cellphone service. It could, for example,
prevent the phone starting large downloads such as music tracks or
podcasts when your behaviour suggests you are about to go out of
network range.

Such prediction has become possible because smartphones like the Nokia
N97 and Apple iPhone contain accelerometers that sense motion. They
are normally used to reorient images when the screen is flipped from
vertical to horizontal, or by software that responds to a shake of the
phone. But Peddemors realised that they also generate a data stream
that reflects every move the phone's owner makes.

Routine events such as going to work are likely always to involve
similar sequences of actions: locking the front door, opening the
garage, getting in the car, for instance. The Delft system uses
telltale sequences and timings like this to create an electronic
signature of particular events.

A neural network software app running on the phone is then trained to
predict what happens next and act accordingly. So if your regular
drive to work takes you through a particular phone cell, the "going to
work" signature could trigger the software to negotiate with the
cellphone network to ensure that the cell will have the 3G capacity to
maintain your streaming music channel as you drive through it.

Peddemors says his team's idea of predicting these "mobility events"
(Pervasive and Mobile Computing) 
could prove especially useful in situations when safeguarding against
loss of data is critical, such as in the emerging field of cellphone
transmission of vital physiological data from heart-rate and
blood-pressure sensors. "By predicting the patient's movements, the
upload of that critical data won't be attempted unless their behaviour
says it can be completed," he says.
The upload of critical data won't be attempted unless your behaviour
says it can be completed

The system "makes possible an interesting set of applications", says
Ian Brown, a privacy and security specialist at the Oxford Internet
Institute in the UK. "But to ensure the user benefits from them - and
not, say, behavioural advertisers or law-enforcement personnel - the
data needs to stay firmly under the control of the individual using

More information about the AccessIndia mailing list