A review of the Program UltraScience which aired 11/20/96 on the Discovery
Channel. This program was produced by Andrew Waterworth, through Beyond
Productions, Australia.
ANDROIDS, Science Fact or Science Fiction?
Lee Kent Hempfling
Chairman, CEO
Neutronics Technologies Corporation
http://www.neutronicstechcorp.com
Errors in spelling or grammer are profoundly regretted. Fast placement of
this material required a lack of detailed proofing.
PREFACE:
It is by no coincidence that the contents of this program have been so
interesting and at the same time so depressing. I will state right up front
that my personal interest in the topic of the program is quite intense. The
interest of this Corporation is also quite intense. I will not hide the
obvious. This company is in the business of solving the problems presented
by this episode of UltraScience. We have presented bits and pieces of our
technology to a small community of people for four years now. The result has
been both stimulating and frustrating. The entire contents of the program
about to be reviewed addresses the individual elements of what this company
is based on. I can not fault the producer of the program for not knowing
about it. I can only fault myself for not protraying it loud enough. But
there is a special consideration.
Almost all scientific progress is a slow and connected evolution of ideas,
each one elaborated by subsequent researchers and theorists. It makes for a
connected undertanding. It is no wonder the program's contents are as they
are. Each researcher is using terms to fit his own perspective of things
intelligent and things biological and in so doing is spawning more like him.
Each will take the slight advance made in accomplishing the original
intention and move the pieces ever so closer together.
What can not be tolerated in that form of evolutionary science is the
presentation of the whole. Major breakthroughs in science are indeed rare.
But not as rare as might be considered. For it is not the breakthrough that
is the main point. It is the acquisition of acceptance by peers.
Everything is indeed relevant. Rodney Brooks was touted as being the bad boy
of robotics in an article published a couple of years ago. Professor Brooks
is not a bad boy. In fact, I consider him to be quite a good man, regardless
of his methods. He was just considered to be different in his thinking. So
the image of a rogue or lone wolf was placed on him. He does not deserve it.
Yet he was and still is within the acceptance potential.
In the normal progression of scientific endeavors little threat is posed to
previous researchers by elaborating on their work. A great deal of threat is
implied by presenting something no one else has ever worked on. It turns the
review of the material from an examination of the science to a proliferation
of the status quo.
Renegades as a rule do not tend to care about such acceptance. But in the
real world there is more to acceptance of ideas than simply the presentation
of them. Affiliation has much to do with it.
If you were a newspaper reporter and a press release from MIT graced your
desk exclaiming the accomplishment of an intelligent machine you would sit
up and take notice. But if the same press release was presented from an
unknown source, even worse an unknown corporate source, you would not
believe it. You might even consider it a ruse. A scam. After all, to most
people, science is a thing of University. Industry just carries out what
professors have concocted.
Sure, there have been great advances from industry. Bell Labs made its name
with great advances. But the corporate laboratory has its own set of
criteria for belief. Universities are where thinking occurs. Government
grants is where money comes from to turn idea into reality. Corporate
sponsorship ensures first crack at a new advance. The true lone sheep can
not possibly think in the same league with the system of organized and
government funded big business science.
Sometimes there is the exception. Big research business seems to forget
where they came from. Universities seem to forget their roots. Things as
they are today are what is compared to in evaluating the credibility of an
idea. If the idea is totally different than accepted status quo it is wrong.
That leaves the burden of proof where it belongs. With the presenter of the
idea.
As a company we have struggled to keep presenting ideas to people who can
not possibly relate to them. If I were to tell you that everything you know
is wrong you would not agree. If I were to tell you that everything observed
is really upside down you would not agree. It could not be otherwise. The
only way for any person to comprehend a major change in thinking is to
discover it on their own.
UltraScience: Androids:
Duplicating a human.
And that is the problem. The program speculates that such an endeavor would
take an orchestrated and combined effort greater than any space program.
Since most of the perspective of the presentation duplicates the perspective
of the researchers depicted within it, it is not unusual that such a
prediction would lead the insight.
The human being is seen as a complex entity. It is normal to try to break
that complexity down to its parts and then by duplicating those parts
acquire a complex duplication of a human being. The perspective being from
the outside. What a human does. How a human is seen to do what a human does,
all the time keeping the perspective of the desired attributes.
And those desired attributes are simple. Stronger. Faster. Smarter. More
power. More of everything. Quantity is an intrinsic requirement of
complexity. But complexity would not make singularity if the real underlying
requirement was not efficiency.
So the program begins. We are invited to take a look at the state of the art
of robotics engineering. The state of the art of computer programming. The
state of the art of the effort to duplicate the human. What we are actually
presented is ancient history. One advance being improved to be called a new
advance. The parts of the human being duplicated to DO what the human does.
Not do it HOW the human does it.
Afterall, the goal in robotics is to simulate the abilities, the outcomes,
the things people DO. And when the simulator can be made to look human the
illusion is ever stronger.
Any rubber band can be made to launch a pebble. Only in science would the
rubber band be given credit for it. Every child knows the child launched the
pebble. The rubber band was only the tool.
Machines around the world:
Part 1: The Plastic Covered Head.
The admission is right up front. The purpose is to impersonate a human. What
does the human do? If we make a machine that does it too can we call it a
duplication? No!
In Japan the plastic covered collection of servo motors looks human. It
moves artificial eyelids. Its plastic covering face shows expression. The
illusion begins.
Part 2: The Walking Monster.
The state of the art of walking technology is presented from Tokyo. It is a
cumbersom counterbalanced monster. Loaded with motors and hydraulic
actuators it looks like two metallic legs on a massive square hip. It walks.
It looses its balance, but it walks. People walk. So walking is a part of
the whole of a person. But people do it by the efficiency of muscle
structure and joint arrangement. The HOW is not addressed in this device.
The WHAT is. Having a machine accomplish what its program tells it to do is
fine engineering. But the walking of a human being is not complexity. It is
efficiency. And it all hinges on the correct muscle structure, the correct
joints, the correct motor control and the correct mental addressing of motor
function. All of that is accomplish in a human by a complex, yet simple
architecture. Nothing at all liek the behemoth presented. But it walks. It
falls because balance is internal. It is not because we have arms to swing,
as the program contends. You can walk without swinging your arms because
arms are for other uses. People have just noticed that swinging arms can act
like a counter balance and if heavy enough and programmed well enough such
counterbalance can take the place of internal efficiency. The machine falls
and it can't get up. Its design is to stand upright and walk. Humans are not
so constructed. The balance is system wide.
Part 3: MIT's Leg Laboratory.
On a computer screen single leg mutant kangaroos are followed by a hopping
two legged dual pogo stick with a UFO hat. Its how scientists, it is said,
study how to make a more graceful version of two legged locomotion. It would
make a nice Saturday morning cartoon. People walk. People walk gracefully.
High technology affords the scientist the ability to model his idea for
graceful walking in a computer program. Fluid movements are simple. People
make fluid movements. Another part of the whole is in sight. But what about
the causes of the fluid movements?
The mechanical version of the pogo sticked space cap is displayed bouncing
on a tread mill. Nice touch. But what possible application could be derived
by a bouncing set of pogo sticks with a program controlling the movements
have to do with replicating walking ?
Part 4: Arms and Hands
A new arm design is presented depicting multiple joints and motor
controllers. It is more flexible than the average industrial robot and is a
nice partial step toward another part of the human. The arm. But its design
is no different than any other robotic arm. It does not duplicate the human
arm. To do so would take the right joint design, the right muscle design and
the right understanding that efficiency is the point. The arm is far too
heavy for an android's use and its "hand" is just a gripping tripod.
How things grip is the next part of the human. People have hands that grab
and hold and control. To build a hand that closely models a human the first
requirement is for it to look like one. Half of the potential objection to
believing the illusion is the appearance. From the human hand's 25 muscles
comes the design of 32 kevlar and dacron tendons controlled by 288 pulleys.
Once again at MIT. Efficiency is not the issue here. Accomplishing a machine
that grabs (with three fingers) and has an opposing thumb..... IS. So there
is one. But it is not at all like HOW the human hand does what it does. The
human had also uses the same joint design and the same muscle design to
efficiently, in a miminum of space, accomplish the most finite controller
ever observed.
Part 5: Muscle.
"What almost all of these robot wonders till lack, is muscle," says the
announcer. "Our muscles are deceptively simple, long thin fibers that
contract to produce motion." But they are not. They are made up of muscle
cells that contract within a framework of many other muscle cells that give
the illusion of a long thin fiber. A case in point is back at MIT:
"Artificial muscle would be a robot designer's dream." Proclaimed the
announcer. Indeed it would. We have it pending at the US patent and
trademark office.
At MIT polyacrilicacid is being shown as a potential revolution for robotic
muscle. When chemicals are applied to it, it shrinks thereby pulling up a
very light weight plastic frame similar to an arm bone. Here we have
simplicity ignoring the complexity. Muscle cells are controlled in groups to
respond to mental simulus. The entire muscle does not work on one impulse.
It is more complex than the simple polyacrilicacid displays. And in a human,
as well as all other biological muscles the brain's instructions are
transmitted to be converted to a chemical. It does not start out that way.
Part 6: Senses.
I found it quite interesting that sight was proclaimed to perhaps be the
most important of all senses. Why? Sight is what science relys on for
observing things. The tree falling in the forest can not make sound if it is
not observed by sight. All of these parts being duplicated are observed by
sight. Not a single internal causative action can be observed by sight. Not
so unusual then that most scientists ignore them.
Sight:
The announcer correctly describes the process of incoming vision. Light
strikes the retina in each eye in a slightly different perspective. Both
eyes see the same thing, from a different perspective. But the eyes are made
up of their parts too. Each part (rod-cone) sees the same thing from a
slightly different pespective within the same eye.
Robotics has found the camera to be a nice way to duplicate sight. Cameras
take in light and by assigning X-Y coordinates to the pixels of the camera's
"vision" many things are possible. By putting two cameras together to
overlap each other a sense of peripheral vision can be accomplished.
Computers can calculate the pixel and manipulate a machine to give the
illusion of intelligence and the illusion of the machine being more than the
tool.
We can judge an object's distance, motion and speed. We do it (those of us
who are humans) by blending the continuous and synchronous holographic image
of light presented from each different incoming receptor (part of the eye)
and by further blending them for both eyes we acuire depth perception and
momentum. All of that occurs at the same time. A computer, unless it is
connected to many more, carrys out those functions in serial and combines
them faster than our eyes can comprehend to give the illusion of actual
visual representation.
Peripheral vision is not at all what Dr. Yasuo Kuniyoshi explains it be. His
design of a panning visual dual camera device is not at all similar to HOW
the human eye acquires peripheral vision. It IS a nice way of doing it for a
computer. Dr. Kuniyoshi has built a panning platform where the cameras are
telephoto as well as with fisheye lenses. It is true that peripheral vision
of the human eye is out of focus from the central focal point but it is not
distorted. The eye design that exactly duplicates the human eye is contained
in our patent application.
Watching Kuniyoshi's "eyes" follow a light bulb reminded me of a kitten.
Input controls behaviour. His computer program has been set to instruct one
camera to keep the light in its cross hair sight. The other eye has no such
instruction and always looses the light. What does this show? It does not
show intelligence. The program is just as much a tool as the eye apparatus
is. If it showed intelligence it would do the function without a program.
The intelligence is a product of the architecture, not the instructions.
This too is contained in the patent application.
Hand Eye Coordination: The Whole Arm Manipulator at (you guessed it, MIT)
This machine is perhaps the most deceptive of them all. Sitting in a padded
area is a metallic arm on a large pedastal. The arm catches a specially
colored ball. It throws it back. But is it doing it? No! It is carrying out
instructions given to it by a computer program that is following the
specially colored ball through two , two color only cameras which provide an
X-Y coordinate for the location of the ball ad instructs the "arm" to meet
it and upon pressure to close its metal grips around it. It looks
impressive. It is said to be impressive. It is an illusion. A tool doing the
job another tool told it to do. Exactly as it was told to do it. The
controller of the arm is actually the person who programmed the computer in
the first place. He (or she) has simply put the program to the task instead
of manipulating the machine personally. It is not intelligent. It is not
even very interesting. But it is showmanship. A product that most MIT
projects seem to be devoted to. 'Look at what we made today.'
"It's programmers have even given it, make that "her" the illusion of
personality." Said the announcer. That is not all the illusion is.
THE MAJOR MISCONCEPTION
What is the difference between "top down" and "bottom up"?
Part 7: COG!
The program depicts TOP DOWN as when someone instructs you how to do
something. BOTTOM UP is defined as when you learn by experience.
In reality, TOP DOWN is when a program is made to instruct something to do
something by defining a rule. There is no understanding of the instruction.
Only the outcome. BOTTOM UP is when the machine has no instruction and
acquires, through experience, the subtle differences that make up what will
become to be understood as rules.
In the Artificial Intelligence community TOP DOWN is programming based.
BOTTOM UP is hardware based. The TOP DOWN approach assumes that knowlege
makes intelligence. The BOTTOM UP approach assumes that intellgence permits
knowledge.
But that is not how it is now being treated.
Cog (MIT again) is designed from a collection of sensors, motors and
computers and is claimed to be a bottom up design. It can not be. The
computers provide rules and can not operate with understanding of the
relationship of one rule to another other than what has been given them by
their programmers. As COG accepts incoming data it has to be conforming to
the programming of the computers feeding it. It itself is nothing more than
a human like appearance that carrys out the computer program driven
instructions. The programs may adapt to the incoming stimuli but they do so
within the confines of the program. So COG is nothing but a front "man".
Rodney Brooks plans to add some 200 odd more computers to COG over time once
again underlying the more is better ideal. There is no concern for
efficiency here. Just for the most illusionary presentation of what the
collection of the parts seems to be visibly able to duplicate. The desire to
have it learn as a small child (given by Brooks as the reason it was
designed to look somewhat human) only adds to the illusion.
Cog is built to sense, react and remember the world around it. Isn't that
what a child does? Humans sense. Humans react. Humans remember. All parts of
the whole. But no concern has been paid to HOW humans sense and what they do
with the sense input when it enters the brain to be evaluated with previous
sensory input. No concern has been paid to HOW humans react to their
environment just that they do. And humans remember is true. But how a human
remembers is not at all considered by ANY of the machines depicted in this
program.
How humans sense, react and remember is FULLY contained in the patent
application and all three have been proven to be correct, without computers
and without programming in a device built to display just a part of the
patent application's claims.
CYC:
Douglas Lenat of Austin's CyCorp is 'priming the pump', as he said in the
program by giving a computer a massive amount knowledge. The program uses a
sort of Thesauras method to cross relate one idea to another so file
searches for one type of word (such as the word, FUN, used in the program)
will bring up references to pictures and ideas of other things coded with
similar ideas. This TOP DOWN knowledge base approach is hope by Lenat to
someday emerge with a mind of its own, to spawn some intelligence to use all
that knowledge. It might be able to, someday. If Lenat programs it to do so.
Knowledge is not intelligence. It is the result of it. Humans are not memory
storage devices. Knowledge we have is a result of a reference to most all
previous similar knowledge and , this is where CYC comes close in Lenat's
approach. But human knowledge is a compounded thing. Each stimulus will
support a greater understanding. The more practice the better the result. It
is not just because one concept can be compared to another. It is the
comparison of dissimilar concepts that permits us to compare a ball (any
color, not just red) to a pizza because both are round. And to compare an
idea of something new to what we already know. Resulting in declaring the
new idea can not be real since it has not already been known. Which of
course would not make it new.
CYC is claimed to still have millions of rules and facts to absorb before
coming to its own mind about things. Perhaps it is the scientist who needs
the facts.
Part 8:
Language:
It is said to lie at the core of our sociability. Can we ever expect an
android to master the subtleties of human language? (The question presented
by the program).
Back again to MIT, (you see, proof that great ideas only happen at
Universities) where Galaxy is being displayed as the program that
understands human language. Let me state it clearly. NO IT DOES NOT.
The program recognizes speech. This is not understanding it. The program
results in speech. This is not understanding it. It is an illusion. Well
programmed to give the illusion of an understanding computer. But the
computer is just a tool. The program is just a tool. The programmer's
understanding of language is depicted in its use of those tools. The result
is a program that will continue on the same subject when presented with
another question that may relate to the same subject. Asking it what the
Italian restaurants are in Cambridge results in a list and response. Asking
it for which one is closest to MIT is a reference to Italian restaurants. So
the result is the closest to MIT. The appearance is that the program somehow
understood the concept of close. Close would be nearest. Nearest would be
what the programmer gave each Italian restaurant entry as a location in
relation to the other locations contained in the data base.
Go ahead. Ask it for how close that one Italian restaurant is in relation to
how far it used to be. Will it understand the concept of time in relation to
distance? It will display what the programmer gave it the ability to display
and nothing else. As it does not understand anything. It just looks like it
does.
It was asked , how long it would take to walk to there. It responded with
the preprogrammed time it would take to walk there. But did it understand
that it could have taken longer in bad weather? Did it understand what time
is or did it recite the time it was inputted to recall? It recalled its
input in relation to the context of the question. Context is not
intelligence. It is another rule which whem mixed with a broad comparison of
knowledge will give an even greater illusion of understanding. But it still
does not understand.
Part 9: Emotion.
"Can emotion be programmed into a mechanical being?" Asked the announcer.
Now there is an oxymoron. Emotion/programmed.
Nayoko Tassa (spelling probably incorrect) has designed a virtual baby. The
graphic depiction of a baby is programmed to respond with visual movements
and facial expressions and actions to the tone of voice of its "mother". One
CAN NOT program emotions. Let me put it bluntly.
One can program reactions to specific stimuli but an emotion is NOT a
reaction to a specific stimuli. All of us respond to different things
differently. Emotions are the result of the expectation we place on a given
topic or concept. When the expectation is met we are either excited or
pleased. When it is not met we are either sad or despondent. All of the
range of emotions are in between. Emotions are the result of the human
ability to project potential from input or memory "now" based circumstances.
The ability to do that is finitely depicted in the patent application. It is
only a proper funding away from also being physically proven. Without
programming and without computers.
Neural Baby's program (remember a virtual depiction is NOT the program doing
it) converts sound to wave patterns and compares sound to a bank of data
waves. Each wave in the bank is linked to a different emotion display and a
different reaction. This is not emotion. This is the illusion of the
outcomes of emotion.
The program referred to the term "robot emotional response" as an elusive
thing. It is not elusive. It is quite possible and is in fact already
designed. What is impossible is programming it. Emotions have to be built.
Emotions are the result of many different acquired knowledge bases that
relate to other acquired knowledge bases. One can not make a machine have
emotions. One must LET a machine have emotions.
Rodney Brooks ended the program with this thought:
"It may be the case that we are led to building robots that have more of
these animal and maybe even human characteristics, so it may offer us a way
of getting to the ultimate goal of being able to build systems that act like
animals and even ... humans."
And there is the rub. The goal is to build something that acts like
something else. Not build something that does something like something else.
Only the outcome is desired. Almost like many other things humans try to do.
A young person wants a career when it takes a job first. We all want to run
faster but few remember it took crawling first. We all want to jump to the
conclusion of the book but we would only miss the point of the story.
The announcer: "Already , the android has captured the imagination of
forward looking artists with fascinating results. For the rest of us though:
the question of what androids will look like pales beside the question of
how radically they will change our world, and our lives."
There will be no radical change in our world or our lives by simply
concentrating on what a human does. There is no understanding in that. But
there is understanding HOW a human does something. If we understand how the
brain functions we can stop treating it like a black box and offering cures
for its symptoms. If we understand HOW the body accomplishes movement we can
start to replace the parts lost by ones that will respond to the brain and
not to some computer. If we understand HOW the eyes see we can understand
how to improve the visual stimuli. If we understand how the brain processes
the visual and aural and other sensory information we can understand how to
treat children who need help in overcoming imbalance of processing and stop
calling it disease or mental illness.
The real desire is to build a machine in our own image. If we are only what
we do and not HOW we do it, only what we say and not HOW and WHY we say it,
only what we see and not what causes what we see, only what we sense and not
HOW we process it then we do not deserve to build anything in our own image.
Other than, perhaps, an empty shell.
It is up to each and every one of us to learn on our own. To look at things
that are different and to question the things we have been taught to be
fact. If we do not, we are only clones, ourselves. And the android of the
computer and the motor and fancy programming need not be built. We are not
worth copying.
I find a higher reason for what we are doing. I see it as helping to
understand the human and all life. If we do not use the technology available
to us to probe deeper into causes, to find the reasons for things, to do
more than just duplicate the outcomes we are lost as a civilization.
It is not the robot or android that threatens society. It is us.
CONCLUSION:
This company possesses the technology to build a fully intelligent,
conscious human form android within three years of the beginning of the
project. Our patent application details it all. We have only been able to
venture into a small bit of the technology so far.
If the time ever does come that someone does take the effort to understand;
programs such as UltraScience will be devoted to imparting that
understanding and the so often used "it may be" will cease to be the whole
topic of science programmes.
Will things observed from the perspective of the observer forever rule
science? I don't think so.
The Author:
Lee Kent Hempfling is Chairman, CEO of Neutronics Technologies Corporation.
The company world wide web site contains access availability for over 50
scientific papers on this topic and many others having to do with the
replication of HOW we do things. The address is:
http://www.neutronicstechcorp.com
