DOORS
http://www.sunday-times.co.uk/news/pages/sti/2001/07/08/stidordor03015.html?
Jim Giles finds that by combining the workings of the human brain and
computers, neuromorphic engineering could step in when human senses fail
Chips are all ears and eyes
As anyone who has ever fumed at a computer in frustration will admit,
computers do not think like us. They are great at big number-crunching jobs.
They can do our accounts, solve difficult sums, and beat almost anyone on
the planet at chess. Yet humans and animals have the edge when it comes to
tasks that are harder to define, such as distinguishing between two faces.
Could the two ways of working ever be combined? Is it possible to create a
machine that marries the brute power of silicon chips with the flexibility
and adaptability of animal thinking? According to neuromorphic engineers,
the answer is yes.
Neuromorphic engineering aims to create silicon circuits - the basis of
modern computers - that mimic the way human and animal brains work. Experts
in this field, such as Rahul Sarpeshkar of the Massachusetts Institute of
Technology, are convinced that the future of computing lies in understanding
why humans and animals are so good at particular tasks, and in building
computers that work in the same way. "Stupid-looking organisms are doing
amazing computation," he says. "We need to replicate this in electronics."
The idea of building computers that mimic real brains dates back to the
1940s, but it was not until the 1980s that the technology needed to build
"natural" circuits became available. Progress has been steady since, but the
dramatic improvement in conventional computers over the same period has
overshadowed the field's successes and deterred investors. Now neuromorphic
engineering may be ready to make headlines of its own, as a slew of new
products, from artificial eyes and ears to sensors for intelligent
buildings, approach the market.
Few people realise it, but at least one neuromorphic product is already
available. The Logitech TrackMan mouse is an optical computer mouse - it
contains a simple vision sensor that tracks the movement of the mouse over
the desk and sends this information to the computer. Such visual tasks are
difficult for normal computers, but animals perform them effortlessly. So
when Logitech wanted to develop the mouse, it turned to nature for
inspiration.
The Dutch electrical engineer André van Schaik designs computer circuits,
but he is also an expert in the way flies see. Impressed by the ease with
which the tiny brain of a fly makes quick and accurate decisions - such as
how to avoid an incoming swat - van Schaik designed a circuit that could
track the motion of objects in the way that a fly does. The result was a
cheap, low-power circuit that Logitech found was perfect for use in a device
such as a computer mouse.
A similar approach may soon lead to sensors for tracking the motion of
people around buildings, turning lights on and off as they enter and leave
rooms. Groups at several universities around the world are also working on
developing a neuromorphic retina, an artificial version of the real thing
that could restore sight to people with certain kinds of blindness. Trials
are under way and crude versions have already been implanted into patients.
However, scientists warn that the human retina is complex, and we are still
some years away from being able to build a satisfactory artificial version.
Naturally inspired, artificial versions of the cochlea (the part of the ear
that converts sounds into the electrical signals that the brain uses) may
arrive sooner. Artificial cochleae already exist, and are implanted in
congenitally deaf patients. Although they are crude compared with the real
thing, the results can be impressive. In the best cases, they enable their
wearers to conduct telephone conversations.
Nevertheless, many deaf people are put off by the size of the implants. The
digital processor that converts sound into electrical signals must be worn
externally. The implant is also power hungry, forcing the wearer to sit next
to a recharging device for several hours every few weeks.
In a bid to solve these problems, Chris Toumazou, of Imperial College,
London, has developed a compact, long-lasting device to replace the normal
digital processor. Once again, nature was the inspiration.
The conventional digital processors used in current implants tackle problems
methodically, like following a recipe to bake a cake. Human cochleae take a
less systematic but effective approach. This miraculous ability is the
result of cochleae having being shaped over millions of years of evolution.
Toumazou has borrowed from evolution to create an artificial cochlea that
mimics the processing of a human one. The result is a smaller, lower-power
version of the conventional digital processor; it will fit within the ear
and should need recharging only once a year. Toumazou hopes to begin
clinical trials before the end of the year.
Projects such as these are just the beginning of the potential applications
of neuromorphic circuits. The computer revolution cannot last for ever, and
many researchers predict that the relentless increase in the speed and power
of computers will peak within the next decade.
Neuromorphic circuits may then be the way forward. If computer scientists
cannot rely on newer, faster circuits to crack difficult problems, where
better to look for inspiration than the human brain, the incredible device
that evolution has already designed for us?
Jim Giles is assistant news and features editor at Nature magazine
This archive was generated by hypermail 2b30 : Fri Oct 12 2001 - 14:39:42 MDT