In order
to learn something about the physical structure of the brain, and how we think,
look at the thinking part of the brain of a young animal, its cerebral cortex.
It looks like a mass of jelly
containing a large number of cells. Arriving over part of the brain's surface
are nerves bringing in electrical signals from the animal's sensors (eyes,
ears, nose, taste, touch). Somewhere else on the brain's surface are nerves carrying instructions to the
various body muscles. It is obvious that the brain controls the muscles
depending on the combination of input signals. To do this you will see there
are connections between the brain cells. Some have only a few
inter-connections; others many.
Now if you look at the brain of an
adult animal which for some
reason has
had a "deprived childhood" you will notice that there is not much
difference in the number of interconnections compared to those in a young
animal. Look now at the brain of a normal adult and you will find a vastly
greater density of interconnections. It is apparent that the memory, the life
experience of the adult animal, resides in this tangle of interconnections.
As with animals, so with us.
Our brain is basically an adaptive
switchboard between data inputs and action outputs. As such it is quite
different from the construction of present-day computers with its specialized
memory unit, holding data and program, and the separated calculating unit. The
memory of our brain is indistinguishable from the program - both reside in the
way the brain cells have been wired up. (Strangely enough, the cells in our
brain are quite similar to the "logic elements" in a digital
computer. They receive many or few impulses from other brain-cells and if these
pulses arrive at the same time it will generate a pulse which goes on to other
cells. I am simplifying here, this is only one possible mode of operation but
the point I am making is that these cells are relatively simple and we can
quite easily imitate them electronically). A completely un-programmed brain
could be considered as a uniformly high resistance jelly, with no connections
between input and outputs. But we are born with a lot of cells in our brain
already wired-up so as to make pathways between certain combinations of inputs
and outputs. These "hard-wired" pathways have been inherited from our
ape past. They give us the basic drives of self-preservation, sex and the
instinct of the herd, and some simple abilities that enable us, without previous
experience, to react to certain standard situations that might occur.
The process of "learning"
consists of modifying our brain to make other pathways from inputs to outputs.
Think of learning to ride a bike. First
attempts are uncertain. Urgent signals from the ear-canals (which detect
motion) signal "about to fall over to left" - action: "put left
foot on ground". Ear-canal signals "about to fall over to right"
action: "put right foot on ground". After a lot of this you will
suddenly notice a vital fact - turning the front wheel in the same direction as
you are about to fall will cancel the fall. You are now rapidly joining up a
lot of brain cells. Some of these will have to be modified when you learn that
at higher speeds you must turn less into the fall than at lower speeds, but
gradually you are getting the "feel" of riding a bike and once learnt
it is there for life. You have wired-up a part of your brain which, when joined
to other wired-up sections (such as the bits giving brake sensitivity, effect
of gear change, and a large section concerning how to behave in traffic (wired
up when you were learning to drive a car ten years ago), will enable you to
zoom along a country lane, admiring the country-side, chatting to your partner
and oblivious to the fact that you are doing something which can by no stretch
of the imagination be called "natural".
The action of learning something new is
therefore that of
joining up
cells that were not already joined up and then joining these new networks to
already joined up networks. The proof that these are complex networks and not
single pathways is shown by the fact that people sometimes have had accidents
where sections of their brains have been removed or damaged but they have still
retained their memories, albeit a bit vague in parts.
Insight
It often happens that we study some
subject and thereby produce a little network of wired up neurones in one part
of our brain. Sometime later we work on another problem and another little
network is constructed. These two networks represent two separate pieces of
knowledge or ability and as there is no obvious connection between them, they
are used separately for maybe years. And then suddenly a way is found to
connect them up into one network. This may be in the form of an analogy, eg. If
you have a network wired up when you did military service and now you are
medical doctor, you may suddenly see infection of the body in military terms
with the defending forces being thinly spread but with a good early-warning
system. The enemy has to move as quickly as possible to establish a beach-head
before the defenders can arrive in strength. You can find a new application for
"first strike capability, bringing up reserves, counter-attack,
consolidation, mopping-up, intelligence, espionage ...". It will give you
a whole new viewpoint of a familiar situation and you may start asking yourself
questions such as "Do microbes make feint attacks on the body?" And
then start looking to see if they do.
Another way this link-up may occur is
as in an argument I heard once in a pub in Germany. It was around the time of
the Falklands War and Helmut was telling Joe what a waste it was that so many
people had died to liberate an island mostly filled with sheep. Joe (whose brother
had apparently been there) was going on drunkenly about the 2000 islanders and
how they had been saved from the overwhelming forces of the military dictator
by the fantastic fighting qualities of the British soldier. Helmut was getting
more and more annoyed. "Fantastic fighting abilities garbage," he
snorted (I'm translating freely) "The Argentinians were `ausgehungert'
(starved) by the British blockade!" (Germany was blockaded in two world
wars by the Royal Navy and a blockade is now looked on as a standard component
of any British military activity).
And then the master stroke by Joe.
"Ausgehungert?" he said triumphantly, banging his beer-mug down on
the table "I thought you just said the island was filled with sheep. Were
the Argentinians all bloody vegetarians then?!"
It was almost comical to watch the
expressions chasing each other across Helmut's face as this piece of
information, which he already knew in the context of making the island not
worth dying for, sprang out, now in the context of nullifying any naval
blockade.
As networks are joined up physically in
our brains, so mentally we experience the same effect. We always try to relate
something new to something we know already. Everything that we have memorised
is "hooked" to something else. When we look into our memory we find
that one memory seems to "pull out" another. Unlike our present-day
computers, we cannot remember an isolated piece of information - what we
remember is always joined up to other related pieces of information, like when
and where we heard that piece of information or how we felt about it etc.
The
associative memory
Yes, we have an “associative memory.”
Think of anything and it is apparently joined by little hooks of association to
many other pieces of stored data. This is a very useful method of filing and
recovering information, and the people studying Artificial Intelligence have it
high on their list. Our associative memory is responsible for our love of
metaphors and analogies, the ease with which we invent things by bringing
distantly related concepts together and our irritation when someone else
invents something and we look in our heads and find we had the same information
(“why didn't I think of that?!”) It gives us our love of classifying things
(He's one of those types who ...) which leads up to the general aim of Science
(our proudest creation) which is to bring together (associate) all knowledge
and hang it on as few hooks (axioms) as possible. We try to find rules which
compress whole blocks of phenomena into neat equations. And then we are
constantly trying to find equations which combine whole sets of other
equations. To construct “A Grand Unified Theory.”
It all comes from having an associative
memory. (Or has our associative memory developed because it is possible to compare lots of things
and this is a neat way to file them away because all data is ultimately hung on
a few hooks anyway..? Perhaps we are looking for the same associative patterns
in Nature that we find in our brain.)
This way we have of filing data by
"hooking" it to other pieces of data already on file (here I am using
an analogy!) explains the way we always try to relate new knowledge to old
knowledge. The analogy of electric current as water flowing through a pipe with
water pressure = voltage and litres per second = current, is a way of hanging
the new idea of electric current on the old idea of water flow. You may want to
shuffle the filing around a bit later when you find the analogy is not perfect,
but it is an excellent way of assimilating new knowledge, and is the basis of
all teaching.
Here are two tricks to exploit the
associative character of our memory:
1. In order to acquire new information
"cold", manufacture an artificial hook to hang it on. For instance, a
Mr. Berwell was seen to like eating so remember his name by thinking of him as
Mr Burpwell (Yes, the more repulsive the better. But keep the easily remembered
version for internal consumption only!)
2. If you are doing something inventive
like writing and you are in the full flood of creation, you are pulling out
ideas and the hooks on these ideas are pulling out other ideas. You can't write
it down fast enough. If you pull out the last idea, put it all down, finish the
scene and go to bed, you will come next morning to a " writer’s
block". You will find a finished piece of work with all loose ends neatly
tucked in and nothing to grasp to tie on your next scene. So when you take a
break, make sure it is when you are in the full spate of creation, with lots of
ideas on the hooks of association, only partially pulled out of the pool of
your subconscious. And you will restart with no difficulty.
Filters in
the brain
Imagine someone says "Here is an
old school photo. Can you recognize Bill Jones?" You take the photo and
prepare to recognize Bill. You immediately block off most of your brain,
leaving open only the networks that correspond to Bill's appearance (or the
last you saw of him). You make a few corrections, Bill now has a moustache,
there won't be one on the photo, he has dark hair - he's only 25 now so that
will be the same ... You build up a picture of how Bill may have appeared 8
years ago. Notice no data has come in yet (you have not looked at the photo)
but you have prepared a series of paths in your brain ready to receive the most
probable image of Bill. You can say you have built a filter matched to Bill.
(see Tutorial). When you see the group photo you now "slide" your
filter over one face after another. You will get some output from your filter
immediately (you go down part of the recognition network immediately) - they
are all 17 year old schoolboys, mostly dark haired. "Ah," (slightly
greater signal) .."no,"...carry on ... and then suddenly
"plop," 3rd from the right, you get a strong output from your filter.
You quickly scan the other faces to check but no, 3rd from the right it is.
Looking closer you now see features which differ slightly from your matched
filter, he was fatter then, for instance.
This construction of filters (or
preconceived notions) is all part of our drive to make sense out of the vast
quantity of data that floods in on us.
In this example you had a fairly good
idea what to look for and could construct a detailed matched filter. This means
the filter will strongly discriminate against other images but when the right
one arrives it will pass it with very little attenuation. You have used the
filter to improve the signal to noise ratio.
Noise into
the filter
If on the other hand you are shown a
group photo and asked "Do you recognize anyone here?" you would have
had to construct a rather sloppy matched filter, the only clue being that
presumably someone you knew at school is in the picture. On sliding this filter
over the faces you would get all sorts of false outputs "Could that be
John? No, perhaps not .." etc. Finally you would find Bill, a little more
definite than the others, but by no means sure. (You might be able to sharpen
it up a little .." God, that's Bill alright! He always had his top pocket
filled with pens")
Yes, we can construct filters in our
head. Look in the embers of a fire, and you will see pictures. These pictures
are the output of a matched filter you have just built, filtering out the
"head of a dog" from all the other possible images in the noiselike
input. This is the reason for pictures in the sky, pictures in ink blots.
Patterns in tea leaves, in animal entrails, in a handfull of thrown twigs. At
New Years Eve the Germans melt lead in a spoon and pour it into cold water. The
frequently obscene-shaped lumps are supposed to foretell your future during the
coming year.
What we actually see in all these cases
depends on what choice we have available in our mental library of images and
what our mood is at the time of choice. In engineering terms, perception of an
image is the equivalent of holding up a lot of internally stored images to the
incoming data and seeing which is the best fit. But if there is no pattern in
the data, holding up any image will give some correlation. This fact is much
used in Abstract Art (see Art) and by
psychiatrists.
Think of someone who, perhaps because
of his own personality, tends to sympathize with the views of a political
party. He constructs a matched filter in his head accordingly. The media feeds
him a mass of information - against, for and neutral. His filter selects out
only that data to which it is tuned and he starts to pay attention to the
Party. He begins to fine tune his filter, increasing its output. Now he only
listens to arguments supporting the Party line, which confirm his beliefs even
more. His filter become so finely tuned that he can finally ignore anything
against the Party - which he now sees as the repository of all wisdom. He has
developed "tunnel vision."
This can be a very strong effect as you
will know if you have ever tried to argue anyone out of belief in a
superstition (particularly a religion). "My mind is made up, don't confuse
me with the facts," is no joke, that's really how people are. Before you
enter into such an argument you should always ask them: "Can you conceive
that there are any arguments or facts that would make you change your
mind?" If the answer is "no", then save your breath. They cannot
be reached through reason.
As a real life example of a false
pattern being detected: - I had just arrived to work at a French organization
which represented a number of American companies. I was immediately shown
several Telexes which had been interpreted by the over-subtle French as a
Machiavellian manoeuvre on the part of one American company to discredit them
with their customers by deliberately delivering late, by not answering
technical questions etc. and so causing sales to fall. This would supposedly
give the Americans the excuse they needed to change their representation in
France. My boss suggested I call the Americans - the theory in France being
that the only person who can understand an "Anglo-Saxon" is another
"Anglo-Saxon". I phoned the Americans (with two French listening in
on those extra earphones they have). As I suspected, the Americans had been
just a little too casual over delivery dates, the questions had not been
answered because the responsible engineer had been on holiday but (pause) the
answers were in the post (ironic gestures from the two French). In reply to my
blunt enquiry "Are you trying to get rid of us?" (gasps of horror
from the two eavesdroppers) "Christ, no. We're quite pleased really. And
as for changing the rep - if you think I'm going to go through signing one of
those goddam French contracts again - do you know we had five versions before
...?" I removed the phone from my ear because I had heard Americans get
hysterical on this subject before.
Well, to general surprise, the letter
arrived with the morning mail and the French (rather reluctantly, I thought)
climbed down from war footing.
Suggestion
So far we have discussed how we
construct matched filters in our head in order to make ourselves more discriminating
(improve signal to noise ratio). We have seen the strange results that can
appear if these self-built filters do not have the input for which they were
built, but only noise signals.
Now consider what would happen if you
were told Bill was in the picture but were shown a rather fuzzy, out-of-focus
group photo, not in fact containing Bill at all. If you really believed Bill's
face was there you would probably choose one of the faces and say "This is
Bill." Initially it would be the
one that looked most like Bill, but if you continued looking at it (and
remember you believe Bill is there) you will
gradually modify your "Bill" filter, its output would increase
and you would finally think you had actually found Bill.
Yes, it is possible for someone else to
build these matched filters in your head for you. "Look at those clouds.
Can you see that monkey, just above the black bit?" "Oh, yes! I can
see it now!" A harmless example here, but if you have ever seen a demonstration
of hypnotism you would realize how incredibly realistic these matched filters
must be for the person "under the influence". Note that imaginative, “suggestible", intelligent people are the most easily
hypnotized.
Association
of memories with other senses
Around the age of 9 a doctor and nurse
visited my school in order to inoculate us all against something. He set up his
sinister instruments in a junior class-room and one by one, class by class, we
were called in to be "done". It took about 3 days to do us all and
the parents came to comfort their children. I can still remember the horrid
rattle of the needles as they were brought out of the boiling water of the
sterilizer. The nurse would then scrub our arms with a piece of cotton-wool
soaked in methylated spirit (rubbing alcohol) before the needle was plunged in,
and even now the smell of meths immediately recalls this scene (and others of a
similar nature).
Evidence that I am not the only one so
affected was given at a Halloween ball I attended recently, dressed as a doctor
and accompanied by a girl dressed as a nurse. We were supposed to be 21st
Century vampires. We installed ourselves by the door and all arriving guests
were requested to sit in our chair. The nurse would then hand me an evilly
glinting hypodermic syringe, empty, which I would apply to the
"patient's" jugular vein and then hand back apparently filled with
blood (actually tomato juice). The nurse would then transfer the blood to a
glass, add vodka and after some wine talk about a presumptuous little Group O,
or whatever, would offer the guest a "proper" Bloody Mary. I would
then complete the operation by putting a Band-aid on the guest’s neck. It was
really just a way of giving the guests a welcoming drink.
In spite of the blood-soaked swabs in
the white metal bowl etc., most of the visiting ghosts and goblins would take
the chair good-humouredly enough until I remembered I had brought a small
bottle of meths as another medical prop. So I started rubbing the jugular with
a piece of cotton-wool soaked in meths before doing my act with the hypodermic.
The effect was immediate and electric. Some people went quite white and one
girl fainted! I had to stop using the meths.
Yes, the sense of smell is different
from the other senses. I can quite easily remember sights and sounds and how I
felt during certain events, but I cannot remember smells. On the other
hand certain sights and sounds evoke
memories of events, but nothing does it with the peculiar intensity of smell.
Consciousness-
"Aware, awake to one's surroundings and identity"
For some reason we attach great
importance to awareness of our environment (consciousness) and especially
awareness of ourselves, or self-consciousness. I am powerfully aware of my
position in this room and what I am doing in it. I see my hands reaching out to
press the computer keys, and hear their clicks. I feel my bottom pressing on
the chair, I can still smell the toast from breakfast. It is, and is meant to
be, a unique feeling, and whole schools of philosophy have been built on it.
Can a machine have consciousness? Can a machine be self-conscious?
First, we're never really going to find
out. It's difficult enough to convince ourselves that another human being has
consciousness. He reacts as though he has, he is another human being like you
and so he should have - but you can never be quite sure. You can only really
check it by climbing into him yourself.
Then dogs. They sure look alert, are
able to run and catch a ball in the air, can look eager, unhappy, etc. Yes,
they must have consciousness too, but perhaps not as refined as ours. Ah, now
we're in to measuring degrees of consciousness which would please Lord Kelvin,
who said you don't know much about anything unless you can measure it.
Then down the scale a lot to
mosquitoes. Do they have consciousness? They have to be able to detect the heat
from their moving prey and fly in varying wind currents to contact it. But do
they have consciousness? The damn things certainly seem to be aware of me as I
sit on a balcony on a hot evening. Yes, I guess they must have, in a primitive
way.
Now let's examine the doors of the
Transit Lounge. As I approach, the doors slide invitingly apart. Go forward and
pass through and they close with a sigh
of relief behind me. If I stay on the threshold, the doors stay open but jitter
impatiently. Step back and they close again with a hiss of irritation. There is
no doubt that the thing is aware of me. But has it consciousness? If we want to
be consistent I think you must say it has.
With a little reflection we can
simplify the above example even more. The Transit Lounge door opener has two
components: the part that detects me (an ultrasonic radar), and the actual
mechanism which then slides the door open. But only the part that reacts to me
and can be said to be aware of me, to be conscious of me, is the radar. The
actual door opener tells me the mechanism is aware of me but is not necessary
for consciousness. A lamp connected to the radar proximity detector that lights
when I approach would be enough to tell me it is working. But I don't need to
know that the thing is conscious of me for it to be conscious of me. Burglar
alarms are in fact specially made to be aware of a burglar without warning him
that he has been detected. But burglars don't have to be detected with
ultrasonic radars. A simple switch in a foot-mat operated by his weight, or a
fine wire broken by his passage can also be said to be "conscious".
So the word "conscious" really just means "reacts to".
Litmus paper turns red in an acid and must be said to be conscious of the acid.
So there is nothing mystical about
"consciousness". Even the thermostat in my apartment must be said to
be conscious, clicking on and off as the set temperature is crossed.
But what about
"self-consciousness"? There
is a lump in the seat which hurts my bottom so I wriggle in my chair to find a
more comfortable position. I feel cold so I turn on the heating and feel warm.
I am hungry so I eat and am no longer hungry. The radio is too loud so I turn
down the volume. To generalize, I measure a signal from outside me, do
something active to modify it and measure that it has been changed. I am able
to modify the signal that I am aware of. In this way I am conscious of myself.
Self-consciousness requires feedback. (If someone has his body covered with
cotton wool, his ears blocked, and blind-folded is suspended (with an air
supply, of course) in a warm swimming bath, he starts to hallucinate because of
"sensory deprivation".)
The mosquito, by flying in the
direction of the strongest signal can also be said to be modifying the signal
that it is aware of, and so must also have self-consciousness.
The thermostat in my apartment is
conscious that the temperature has dropped below the value I set. It now
switches on the heating. The temperature rises, all the time monitored by the
thermostat, until the set temperature is reached. At this point the thermostat
switches off the heating. It is therefore modifying the signal that it is aware
of and if you want to be consistent you must say it is self-conscious.
So self-consciousness requires a minimum
of one sensor
connected
to one actuator which can modify the sensor input through a feedback loop. Self
consciousness is therefore measured by the number of feedback loops.
Using this criteria, most mechanisms
have consciousness, and some have self-consciousness although the feedback loop
may not be so obvious as the ultrasonic detector/hydraulic actuator of the
door. A 1 second pendulum, for instance, always swings at 1 second,
independently of how far you pull the bob back to start it. You can consider
that the bob "detects" how high you displace it before you release
it, and then is forced to swing down (by the laws of physics) at a speed just
sufficient to climb up the other side and return to its original position in 1
second. The further you displace the bob, the faster it moves but the time
taken for the back and forth trip is always 1 second.
Reflection "Conscious thought, meditation".
Now as so far described, the brain is a
network of connections between body sensors and body actuators. But if there
are no data pulses coming from the sensors, you would expect the brain to be
dormant. But we know this not to be so - we can "think" by calling up
memories, thinking of new ideas, making plans. There is thus a sort of
"internal self-consciousness" produced by internal feedback in the
brain (between calling up a memory and receiving it).
But where do these "calling
up" or internal pulses come from?
Well, no one has been able to give me a
clear answer to this question. Apparently the brain produces pulses on its own
and we are able to direct these to the part of the brain we want to use or
examine. The supply is variable, sometimes we are feverishly inventive, at
others sleepily indifferent. If we don't direct these pulses they just drift
around in clouds anyway, producing the loosely-coupled memories of free
association. And even when we direct them, they are variable. The variable
component is probably the basic "noise level" of the brain. Provided
there are not too many undirected as compared to directed (not too low a signal
to noise ratio), I assume they would not have a great effect on the general
operation of the brain apart from introducing a slight element of variability
in our thoughts and actions, a pleasing unpredictability, novelty,
unreliability (choose your own word). Look around your acquaintances and you
will find great variation in ability to "free-associate".
Dreaming
There is a theory that when we dream we
are filing away new data received during the previous day. Children certainly
dream more than adults. If a vivid dream wakes you up, you can quite often
associate it with something that happened that day. I remember a dream where I
was looking at a bike that had fallen on its side and the back wheel was still
spinning. The dream woke me up for some reason and I seemed to feel the dream
was "about" this spinning wheel. Thinking back into the events of the
day I then immediately "just knew" that the spinning wheel was
connected with an old Hitchcock spy
film I had seen that evening where a mysteriously rotating wind-mill was used
to send signals to the enemy.
The taking of hallucinatory drugs such
as LSD is supposed to trigger this "filing" process while the person
is fully conscious. Normal incoming data (such as the sight of a chair) appears
to have exaggerated importance and may be filed away associated with strong but
irrelevant emotional feelings.
How to use
your brain and be creative.
Here are some well-known tricks that you
can use to manipulate your brain, to prod it into giving you unusual
combinations of ideas. But remember - however much you stir up your brain, the
combinations that come out are only made from the data that were already there.
So to make these tricks work best you should cultivate an open, curious
attitude to life. Read a lot, listen to and discuss new ideas, be continually
learning new things, stuff your mind with facts and data, even if you can't see
an immediate application.
Brainstorming
Brainstorming is a way of pulling
combinations of concepts (or "ideas") out of the memories of a group
of people. They first need to select a team leader - he keeps in touch with
reality, sets the time limit and stops things getting too wild. They sit round
and just let ideas float to the surface. Positive feedback increases the gain
of the group, raising the noise level. Eventually the noise from one group
member modifies a matched filter in another which produces an output above
thresh-hold and he has a sudden idea. The process continues with the leader
noting everything down. Evaluation is done later by another group. Occasionally
a useful new idea appears.
The positive feedback of the
brain-storming effect can occur at any time. You can come up with an ingenious
twist during a conversation at a party and make everyone laugh - including
yourself. I suppose everyone experiences this now and then. It's an odd feeling
- you feel as though someone had just told you to say that.
Attribute
listing
This is mostly used in the advertising
industry where they are trying to give new life to an old product. The first
thing is to make a list of all the "attributes" of a product - eg.
low power consumption, light, cheap, can be used to beat carpets, can be folded
up in a small volume, etc. The idea is then to see if these attributes can be
modified, extended or combined in any way to improve the product or service.
SCAMPER
technique
This is
also used in the advertising industry.
S-ubstitution
- Can an unfamiliar or unconventional substitution be made?
C-ombine -
Can two unusual elements be brought together?
(economy/luxury;
strength/lightness)
A-dapt -
Can we adapt a new idea from the competition? Can this
be used as that?
M-agnify
or m-inify - Can this be overstated or understated?
P-ut to
other uses - Can the familiar be shown in an unfamiliar light? (eg. Can a clothes line be used as a tow rope?)
E-liminate
- Can more emphasis be given by elimination? (The ad showed an airline
passenger flying alone in his own seat.)
R-everse
- Can some element or function be
reversed?
These last two techniques produce new
ideas by modifying (multiplying) the old idea (filter) by some new concept and
feeding noise into it. Perhaps a new product (signal) will appear. Positive
feedback could be added by doing it in a group as for brain-storming.
Puzzle
solving
If someone gives you a piece of
machinery and asks you its purpose, you can start by examining all the
individual components. "Ah, here is a motor, but not a very powerful one.
It drives this sharp-edged wheel so it must be to cut something. Something
fairly soft otherwise it would be Carborundum tipped. And this screen operates
a switch which stops the motor unless it’s pulled down. And the screen is about
5 inches wide so whatever is to be cut is about this size .." This is the
"bottom up" or "deductive" approach - drawing conclusions
from facts. The Artificial Intelligence people call it "forward
chaining". The principal is to simply collect information and to see where
it leads you. Like Sherlock Holmes.
The other approach is to make a guess
as to what the machine is for and then try to find confirming details in the
mechanism. "I guess it is a bread-slicer. See, here is a rotating blade
and this must be to hold the loaf steady." This is the "top
down" or "inductive" approach. "Backward chaining" in
the Artificial Intelligence world.
In practice we rarely use one method
alone but usually switch back and forth between the two.
Artificial
Intelligence (AI)
With the
development of electronics and the ability to construct complicated analog and
digital devices came the study of "system engineering" - how to build
sophisticated systems to perform complicated tasks, like aiming an
anti-aircraft gun to shoot at a fast-weaving bomber, or playing chess.
An engineer confronted with the task of
constructing a device to solve a difficult problem, like speech recognition,
tries to find out how he does it himself. Once he knows how his brain does it,
it is relatively easy to imitate it.
And so an inevitable interest in how
the brain is constructed and how it works. Models have been built to imitate
certain facets of the brain's behaviour, the underlying assumption being that
if the model behaves like the brain then the model must work like the brain and
we have learnt something.
The target (although you won't get anyone to admit it in public)
is to produce an artificial brain. We're not there, yet. But a lot of
interesting progress has been made.
Expert
Systems
The ability to hand on knowledge from
one generation to another has arguably been the reason for human progress.
First it was passed on by word of mouth and then by writing. But there are a
number of skills that are not so easy to write down. How for instance do you
pass on the skill of Fred, a successful heart doctor, or Peter who is a skilled
motor mechanic?
Well, to be exact, we want to be able
to pass it on of course, but what we really want to do is to be able to find
out how they do it and possibly be able to program a computer to behave like
them, which would also be a much neater way of storing and disseminating their
lifelong accumulated knowledge.
Fred and Peter are Experts and a
computer program that imitates an expert is called an Expert System.
The ones that I have seen so far are conceptually
simple - the real problem is getting the information out of the expert.
Assuming he is willing to part with his expertise, he may not know himself
exactly how he performs his apparent miracles. The job of the "knowledge
engineer" is to find the rules he uses. And not only the rules but the
weighting factors he gives to the various data. For instance, to find out if
it's going to rain, John, a Weather Expert reveals how he interprets the data:
Data Certainty of rain
today
Air
pressure falling
0.1
It rained
yesterday -0.2
Moon in
first phase certainly won't
rain
Red sky
and humidity >30% 0.4
He then
integrates these factors and comes up with an overall certainty figure
"supporting the hypothesis that it will rain today".
All this is used to program a computer
which can be used by anyone to forecast the weather as well as Fred.
Note that an Expert System is a passive
program imitating an expert.
Neural
Networks
Present-day
digital computers are roughly split into two sections. There is a calculating
unit with a fairly large fast memory where all the work is done. There is also
a very large slower memory which contains the program plus the data to be
operated on.
The first step in using a computer is
to load in the program. This can be on a CDROM and typically takes a few
seconds. Your computer is now ready to beat you at chess, bridge, monopoly etc.
Alternatively (if you are in a chemical manufacturing plant) you load in a
different program and the computer will take the various sensor outputs
situated around the plant, compare them with each other and certain stored data
and then send out signals to control various valves in the plant to keep it
operating at maximum efficiency.
But there is another type of computer
available, albeit in experimental form. It is called a "Neural
Network" and works on the same principle as our brain. There are inputs
connected to outputs via a complex network of "Neurones". These are
logic devices which have several inputs and give an output depending on the
state of these several inputs. Like in the brain, they are much interconnected.
The exact method of interconnection has been done by "training" the
computer to do some specific task and is done by "growing" wires
electrolytically (in one model). Imagine one being trained to steer a car down
a motorway. To start with it just "observes". The outputs of the car controls, and sensors
which tell it where the car is on the road and where other cars are, are sent
into the neural network computer.
After a few times up and down the
motorway, paths gradually form in the computer and it starts to uncertainly
make a contribution to the steering wheel and gas pedal. A few more trips and
the human driver is gradually able to take his hands and feet off the car
controls as the computer begins to take over. Finally, as in training a human
driver, the "instructor" is able to let the trained computer take
over completely. (It is interesting to note that the above experiment has in
fact been performed, and the computer faithfully imitates the driving habits of
its instructor.)
Presumably such computers could be
taught all sorts of difficult tasks such as to translate French into English.
They are much more complicated than an Expert System and allow a higher level
of expertise to be imitated in a computer, and therefore not lost to posterity.
But there are two big disadvantages to
neural networks:
1.The programming of a neural network
computer, by its very nature, is not "documented" - there is no
description by the programmer of how the program works. Assume we have built a
neural network computer and Michel has trained it to translate English into
French. The computer works, but what have we learnt about how the trick was performed? Nothing. We can
look in the programmed computer afterwards but all we see is a mass of complicated pathways. We didn't program the
computer - we can't understand the results. "All" we have really done
is copy the neural networks of part of
Michels's brain into a computer.
In order to discover the
"algorithms" (the thought processes) that Michel used, we must find a
way of analysing the pathways in the computer, and I suspect this may be as
difficult as analysing Michel himself. About the only advantage we have is that
we can take more time over analysing the computer than in analysing Michel who,
being French, is notoriously impatient and may even be dead.
2.The program and the memory of the
neural network computer is the interconnecting network, eg. the electroplated
wires. It is not easy to see how this network can be copied and used to wire up
or program another neural network computer. Each neural network computer may
have to be individually "trained". This is quite unlike digital
computers where a program written on one can easily be copied and the copy used
to program other computers.
There is a partial solution to these difficulties
in that it is possible to “simulate” a neural network on a digital computer (in
fact that is how most of the research on such networks is performed.) A big
computer is required and it is not efficient but at least the program can be
copied and transferred to another digital computer.
And understanding the undocumented
program of a digital computer, which is modelling a neural network, might
conceivably be marginally easier than tracing out the connections of a physical
neural network.
Neural networks may well end up as merely a different way of
programming digital computers.
It will have the advantage that instead
of the programmer painfully converting his wishes into a computer
"language" (and today the quicker you want the computer to perform,
the further this language is away from English), he will in effect say to it
"Watch and Learn." When it has learnt the task correctly, the
programmer will make a copy of the software and it can be transferred to other
digital computers. A "Watch and Learn" programmed computer will
initially be much bigger and run much slower than a computer conventionally
programmed, but as computers become bigger faster and cheaper, the difference
will decrease.
Our brains have the same advantages and
disadvantages as neural network computers. As the program and the data is
"distributed", it is very resistant to damage. On the other hand,
each brain has to be individually trained. It is a great pity that we cannot
just slip in a piece of software which confers on us the ability to be
instantly (eg) a mathematical prodigy or speak Italian.
The
artificial associative memory
I read a
science fiction story once where Earth comes into possession of an Alien
artefact, a computer. Immediately it was conveyed to a secret Government
laboratory under the tightest imaginable security as it was realized that
Mankind had had a most fantastic stroke of luck and must not flunk it. If we
could find out the operating principles of this incredible computer, so far in
advance of Earthly science, we would be able to make an enormous spring
foreword. It would revolutionize our whole science - we would catch up with the
Aliens!
Well, on the last count we have about 4
billion of these computers on Earth at
the moment, and about 250 new ones being delivered every second. We can study
them at leisure, either individually or how they network. Moreover, each
computer scientist has been provided with one for his personal use. To further
aid their researches, they have each been given a direct link (Direct Memory
Access) into their (literally) personal computers, by-passing the usual slow
and cumbersome keyboard and screen. In fact, the researchers can be considered
as actually sitting inside their computers. From this unique stand-point they
can quickly flip through the memory banks and observe the various algorithms
and sub-routines. They can write new software programs to check performance and
work out the computer architecture in the most minute and intimate detail. In
no time at all they will be able to uncover its fundamental secrets, the fruits
of millennia of Alien research, ours for the plucking!
Why are they taking so long?
Yes, it does seem rather surprising.
Especially the construction of an
associative memory. This item has long been recognized as the heart of any
"thinking" computer. We build large memories for our digital
computers, but they usually store individual facts in specific storage
pigeonholes or "addresses". To learn the fact you have to go to that
address. A primitive form of "content addressable" memory can be
found in databases that assign storage locations based on the contents of the
data to be stored. But we have a long way to go yet.
Intelligence
"The
capacity to acquire and use knowledge. Quickness of understanding, mental
power."
That there is such an attribute as
"intelligence" and that we are all different is proved by the large
number of words to describe it - in all languages.
"Intelligent" is a positive
characteristic – synonyms being: acute, alert, bright, clever, sharp,
well-informed.
(But aren't "cunning" people
just as capable of solving difficult mental problems? Synonyms: artful, crafty,
foxy, guileful, sharp, tricky, wily, shrewd, subtle. Do I detect a moral
judgement here? How come we don't have "Guile Tests"?)
I would like to explore these
differences to find out why they occur and what, if anything, can be done to
remove them. In an attempt to reduce the emotional content, as no one likes to
be thought unintelligent, (doltish, dull, foolish, inane, obtuse, senseless,
simple, slow, sluggish, stolid, witless) I will use the analogy of physical
strength and let the reader translate the various conclusions back into the
mental equivalent.
Paul and Ian enrol at a fitness center and take the same course of
muscle development. The graph below shows what happens.
At the
beginning they both start to get stronger, at approximately the same rate. But
after a while Paul seems to hit a block - however much he exercises he doesn't
get any stronger whereas Ian gets stronger and stronger until he too hits a
block. At this point Ian is three times as strong as Paul. However much extra
exercise they take, nothing changes - neither of them become any stronger and
Ian remains 3 times stronger than Paul.
The graph shows that the answer to the
Nature/Nurture debate is more complicated than the question. At point A both
boys have had the same training and there is no difference in their strength.
Nurture is the same for both and Nature has yet made no difference. But from
point A onwards Ian starts to get stronger than Paul, even though the Nurture
is still the same for both. The increasing strength of Ian can therefore only
be due to his Nature.
So what can Ian do to redress the
balance? He must learn that raw muscle power is not everything, and that what
the muscles are trained to do is more important. He must concentrate on
building up certain combinations of muscles which enable him to perform a
useful well-paid task.
To conclude by translating muscles back
into grey matter (for those not intelligent enough to have followed the
analogy) - you must exploit your feeble brain to the utmost:
- Be more confident. This applies particularly
to females. Management courses are filled with advice on how to get over the
"Inhibition Barrier". Learn to drive a car, to do public speaking.
History is filled with people who were academically mediocre but a great
success in the real world.
- Take an interest in ideas, clever
solutions to problems, gadgets, tricks.
Read and discuss them. These will be all stored in your brain and enable you
suddenly pop out an interesting solution to some problem.
- Become a specialist in something -
something which can earn you lots of money - something that lots of people
think is very difficult. A good example is to learn about computers.
The future.
Technology
is changing so quickly that it is only possible to say something about tomorrow
and the relatively far future.
The far
future
Science
fiction has gone out of fashion now - it has lost the "fiction" part
and either has come true or looks very much like coming true. And one of the
persistent themes of science fiction is the "Giant Brain". Well,
human history shows that if something, however far-out or repugnant, has ever
been possible, somewhere it has been done. And the Giant Brain is one legend
that could really come true for the following reasons:
- It has started to grow already. Large
sections of our world would stop if all the computers were switched off. We are
dependent on them even now.
- There is no visible upper-limit to
the size of a computer nor the complexity of the program that can run on it.
Already we are networking computers world-wide.
- Like a lever, the computer enables a
brain to be constructed having a greater ability than any of the brains
programming it – and computer science attracts some of humanity's finest
brains. At the moment the computer can only do things better than things that
we know how to do. For instance, there is a program which makes a computer play
to chess Grand Master standard. We don't know how we think creatively yet. But
we will, and immediately program our computers to think. And they will think
creatively better than us.
- Computers, software and hardware, are
getting more and more sophisticated from month to month, but the rate of change
is not fast enough, nor well enough understood, to be alarming to the people who
are in charge of us - the politicians. The politicians' power to affect us by
emotional speeches and public relation tricks (and our willingness to listen to
them) has not yet been challenged. But one day there will be a "Computer
Party".
- Most of humanity is mentally lazy and
would only be too glad to have someone do their thinking for them. The Giant
Brain would also be a substitute parent or God, which most of us seem to need.
I have often been horrified at how willing many people are to prostrate
themselves before a computer and take its rapid neat printout as holy writ. We
are only too ready to be colonized by the Giant Brain.
So what can we do to prepare ourselves
for the day when the Great Brain has taken over the world? Well, my advice ties
in rather neatly with that given above. As in any colony, your value will be
proportional to how well you can understand and collaborate with the Colonial
Power.
You must learn to become computer
literate so you can
communicate with the Great Brain.