From: Max More (maxmore@primenet.com)
Date: Mon Jan 06 1997 - 10:43:19 MST
I hadn't heard of this neural network system before. Sounds intriguing. Is
it hype?
Max More
more@extropy.org
Computers that can think are on the way
------------------------------------------------------------------------
Copyright © 1997 Nando.net
Copyright © 1997 Toronto Globe and Mail
TORONTO (Jan 6, 1997 00:24 a.m. EST) -- In a cramped office overlooking Bay
Street, with wires snaking across the floor and cardboard boxes stacked on
chairs, John Sutherland seems to possess technology's holy grail: a computer
that thinks like a person.
Dr. Sutherland, president of AND Corp., has created a new type of computing
that he calls holographic neural technology, and some of the people using
the new technology say its implications are staggering, perhaps even
frightening.
What Dr. Sutherland has developed is a software program called HNet, which
is his own unique approach to neural network technology.
Neural networks -- software designed to recognize patterns and to learn from
the input of stimuli -- have been around for many years. They have been the
focus of countless billions of dollars in research and development, most
notably in Japan, but by most standards they have been a commercial failure.
What makes his neural technology different, Dr. Sutherland said, is that it
more accurately mimics the way the human brain works.
His corporate clients say what makes it different is that it is blazingly
fast, able to reduce complex analytical computing chores from taking days to
minutes. That, they say, gives it a mind-boggling potential to be put to
work in countless industries.
"It's a very unique and extremely powerful technology," said Dr. Andrew
Maziarzewski, a researcher at the Clinton, N.J.-based engineering firm
Foster Wheeler Corp. He has worked with the HNet technology at the giant
Swiss engineering firm Asea Brown Boveri Inc.
Dr. Maziarzewski said he has used other neural network programs but none
come close to the speed of HNet.
Dr. Sutherland, who speaks in the carefully measured tones of a scientist,
believes HNet has incredible implications for virtually every aspect of
industry.
He said it can be used to control any manufacturing or chemical process;
predict consumer behavior in any market segment; analyze financial markets'
patterns and predict future market behavior; identify, command and control
military operations; recognize and identify medical conditions; and
recognize faces, voices and handwriting.
"In the next few years ... we'll see systems for voice recognition, for
autonomously piloted planes, and other vehicles. We'll see machine cognition."
Machine cognition? Machines that can actually think, feel and react?
A cognitive system is one that has enormous storage capacity and that will
continually learn and retain all its information, Dr. Sutherland explained,
adding that that's what his technology does. It learns patterns from the
information put into it, almost instantaneously.
But will such a cognitive machine be truly a thinking machine? We tend to
think of cognition -- the act of knowing or perceiving -- as a skill linked
to emotion and used for survival.
"I'm a hard scientist and that type of functioning (having emotions and a
survival instinct) is in fact based on scientific functioning, on the
interaction of information and the storage of information" in biological
cells, Dr. Sutherland said.
That's his way of saying yes, a machine might be able to think independently.
It may be too great a stretch for most people to imagine such a thinking
machine, but Dr. Sutherland has already built one that sees and recognizes
specific faces.
It took him less than two months, using HNet, to create the program that
recognizes individual faces. Researchers at the Massachusetts Institute of
Technology spent millions of dollars and struggled for years to build a
face-recognition system, but it only recognizes faces if they are motionless
and shown head on.
Dr. Sutherland's system can recognize people as they walk by, with their
faces at any angle. Training it to recognize a visiting journalist's face
took a few seconds.
The potential applications are immense.
The U.S. military has expressed an interest in the face-recognition system.
If installed in every major airport in the world and connected to law
enforcement authorities, for example, it could be used to track terrorists
and other criminals in real time.
Dr. Sutherland would rather talk about the medical applications. He said his
system can be used to recognize the various components of blood and tissue,
and render instant diagnoses in case of problems.
HNet has been trained to look for electromagnetic "signatures" from the
Earth, which would enable it to locate mineral deposits or other underground
deposits.
Because it has been able to learn the fundamentals of electromagnetic
interaction, Dr. Sutherland believes his system can learn and simulate
atomic-level interactions, which would enable it to quickly perform the kind
of complex simulations used in creating nanotechnology, or microscopic
structures.
"We don't really have the final word on how good this will be but this looks
very, very promising," said Dr. Ligang Zheng, a research scientist at
Canada's Department of Natural Resources' Energy Diversification Research
Laboratory in Ottawa, which is trying to use HNet to control the mechanical
and chemical processes at a garbage-to-energy incinerator in Burnaby, B.C.
Charles Murphy, a programmer with Oxford Information Inc. of Hagaman, N.Y.,
offers a slightly less glowing but still positive review.
Oxford uses HNet and other neural network software to perform statistical
analyses of markets, and fraud detection for some of its bank customers.
"Sometimes HNet is just what we needed," Murphy said. "It's comparable (to
traditional neural technology), sometimes better; sometimes traditional
(technology) will beat it. It's always worth trying."
In contrast, Steve Verba is one of HNet's and Dr. Sutherland's greatest
fans. As chief information officer at Realty One in Cleveland, one of the
largest U.S. realty brokers, Mr. Verba said he's been working with
traditional neural network technology for more than 10 years, but what Dr.
Sutherland has done is light years ahead.
"This guy may die penniless but this will profoundly influence life in the
21st century, whether he gets credit for it or not," Mr. Verba said.
"I believe he should get the Nobel Prize and I'm not exaggerating," he said.
"I've read the work of guys who have gotten Nobel Prizes in other fields
that weren't even close to the profundity of what he's laid out here."
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 14:43:58 MST