(Artificial Intelligence Theory Document No. Five of Five)
This paper was originally commissioned by and published in
NOVEMBER Magazine.
Approximately five hundred (500) copies of the Summer 1981 issue
were printed by
the publisher in Seattle WA USA, and about 170 free copies were mailed in
1981 to
contemporary (1980-1981) authors of computer-related articles in BYTE Magazine.
Once we know all there is to know about the workings of the
human brain, we will have a choice of several obvious approaches
to the task of teaching students the essential workings of the
mind. We could teach about the brain-mind in terms of how it
evolved through the eons, or how it develops in the life of the
individual, or how it functions in a mature specimen.
This article presents the author's model of the workings of
the brain-mind, not in terms of sweeping generalizations but on
the ultimate and unambiguous level of the switching-circuit logic
of nerve cells. You are invited to comprehend this mind-model --
to refute it if it is erroneous, or, if it makes sense to you, to
use it in fulfilling the ancient imperative, "Know thyself!"
Either way, you the sovereign mind are offered something to react
against, and possibly a revelation of your inmost mental nature.
Of three obvious approaches to explaining the mind inside
the brain -- evolution, individual development, and static
functioning in maturity -- this author chooses the third route
and seeks to describe your mature mind as you read and comprehend
this article.
The other two approaches -- evolution of the mind in the
species, development of mind in the specimen -- would inherently
contain directions for the starting-place and the order of
presentation of all essential details about the brain-mind. In
both cases, we would simply describe how a single-cell creature
turned into a brain of one hundred billion cells.
But let's take the hundred billion cells and find an obvious
point of departure for describing a model of the organization and
function of that purposive web of cells, the brain. Let us
approach the function of the evolved, mature mind from the
obvious starting-point of sensory inputs into the mind.
This article leads you through a functional model of the
brain-mind. Although the brain is perhaps the most complex
structure on earth, it is no more than a three-dimensional
arrangement of flows of information. The information-flows are
arranged in such a way as to achieve consciousness and thought.
Each flow of information is along one of the dimensions of
the mind. If you are to comprehend this mind-model, you must
understand each dimension and also the very concept of
dimensionality. The dimensions play a double role in this
article: firstly as the building-blocks of the mind for you to
comprehend both one by one and as a grand edifice, and secondly
as the chief arguments to convince you of the validity of the
mind-model.
Dimensionality is the quality of being dimensional, of
having dimensions. The mind is not a seething lump like an
anthill, but a strictly dimensional structure. Although the
brain is curved and convoluted, the mind inside the brain is
rigidly straight (like a taut string or a beam of light) in all
its dimensions, and orthogonal through ninety degrees wherever
the information in one dimension changes its direction of flow
into another dimension.
Although the mind exists within the brain, the mind is not a
material, physical being. The mind is a structure composed
purely of information. The physical structure of the brain
determines the informational structure of the mind, but these two
structures are not identical. Put it this way: The brain holds
information, and information holds the mind. The brain is
organized physically, but the mind is organized logically.
The dimensionality of the mind is crucial to its logical
structure. In some parts of the mind, information must be kept
apart, while in other parts of the mind information must flow
together. The dimensions of the mind serve the purposes of
isolating and combining information.
The first dimensional component of your mind is the straight
and linear record of its sensory input, in parallel with the
straight and linear "keyboard" of its motor output. Please
examine the "mind-diagram" appearing with this article.
A polarity exists between the mind and its environment. An
environment to develop in is just as essential to the mind as a
brain to exist in.
A second polarity exists between our sensory perception of
the environment and our motor manipulation of the environment.
The two polarities -- organism/environment and sensory/motor
-- constitute sufficient logical differentiation for the genesis
of an informational loop.
Your mind sits at one end of the loop and contemplates your
environment at the other end of the loop. Your environment is
the whole cosmos, including your body, brain and mind. Your mind
starts out as tabula rasa, "a clean slate." As your mind
develops and fills with knowledge, it tries to mirror internally
the cosmos which it perceives externally. Who can say which is
the agent -- the cosmos organizing minds, or mind organizing the
cosmos?
Your mind starts out as an empty, but vastly capacious, link
in the loop. Information starts in the environment and flows in
one direction through the loop: through your senses into the
mind, and from your mind out through the motor nerves to the
environment.
It takes a while for your neonatal pathways -- sensory and
motor -- to communicate internally and thus to close the loop
with the environment. The sensory and motor pathways develop in
parallel along the temporal dimension of the mind.
Although your mind is constantly thinking and acting in the
present, its existence stretches off into the past. Every
thought which you think in the present, shapes your mind for the
future. Your mind is the sum of all its past reality.
It is critical to your comprehension of this mind-model that
you think of the sensory and motor pathways as flowing in
parallel, but in opposite directions along the temporal dimension
of the mind. When we go on now to examine in detail the sensory-
input system, you must keep in mind that the sensory and motor
systems develop and operate side by side in lock-step fashion.
A human brain has the five commonly acknowledged senses of
vision, audition (hearing), the tactile sense (touch), gustation
(taste), and olfaction (smell), plus a few other senses such as
the sense of balance and the somesthetic sense.
According to this mind-model, all the senses feed into the
mind in parallel in a flat array like a woven rug. For each
sense, be it vision or audition or smelling the flowers, there is
a flat channel of perception and memory flowing along the time-
dimension of the mind.
The nerves from the sense-receptors travel to the brain.
Inside the brain, the sensory information from vision, and
perhaps other senses, undergoes the pre-processing of feature-
extraction before it enters the mind. In feature-extraction,
basic patterns are discriminated to reduce the work-load and
hasten the operation of the conscious mind. In the brain there
operates a principle of rendering automatic (and subconscious) as
many things as possible.
After the information in any one sensory pathway has reached
the brain and gone through all required feature-extraction, the
information enters the mind by entering the permanent memory
channel for that particular sensory modality. Short-term memory
and permanent memory are identical in terms of physical location,
but they differ with respect to the associative processes which
catalog the memory-traces and control their future accessability
through recall. In other words, short-term memory is not a
function of location but rather of associativity. This assertion
is supported better by the large-scale mind-model than by any
local arguments which may appear in this topical discussion of
memory.
The distinction between preliminary portions of the brain
and the mind itself is based upon a functional demarcation line
beyond which information is free to flow not just along its
original dimension but orthogonally sideways out into other
dimensions of the mind. In other words the mind is circumscribed
and defined by its own dimensionality.
It is important that you now comprehend both a specific
design for memory and a general concept of memory. It is
axiomatic that whatever macroscopic information can be
transmitted can also be recorded. To record information during
transmission, one simply captures samples of the information at a
rate quick enough to catch all instances of significant change in
the information.
The brain-mind records the informational content of each
sensory channel by routing the information through what is both a
transmission channel and an extremely long series of engram-nodes.
Once each sensory information-flow passes the demarcation-line
into the mind, the information in each sensory channel floods the
transmission "fibers" of that permanent memory channel. Each
fiber in the memory channel is like a series of millions of nodes.
Within the particular memory channel for each sense, there are
thousands of the nodal fibers. Your oldest memories were
deposited and permanently, unchangeably fixed in the first nodes
of the lifetime-long memory channels. At each moment of
sensation and perception, all the simultaneously occupied nodes
among all the memory fibers of each memory channel irrevocably
fix their contents. The group of nodes fixed on parallel fibers
at one moment in time is like a "slice" of memory of that moment
in time.
You start out with your sensory nerves and pathways going
through any required feature-extraction and then feeding into
immensely long channels of tabula rasa memory. Your myriad
moments of experience are deposited in densely packed "slices" of
and by simultaneity.
Each sensory (and motor) memory channel is like a flat
ribbon flowing across the logical surface of the mind. The
memory-ribbon is composed of thousands of nodal fibers. The
first experiences go into the first nodal slices. Subsequent
experiences have to travel through all the slices of previous
experience to reach and occupy fresh nodal slices, which will
then be filled and fixed with the experience of the moment,
before serving as a bridge to all future moments.
Although it is critical for you to understand the essential
characteristics of the permanent memory channels in this mind-
model, these essential characteristics are not introduced here
all at once. Advance notice can be given, however, that each
sensory memory channel serves three main purposes, simultaneously
and everywhere along the memory channel: transmission, memory,
and comparison.
Each sensory memory channel is like a pipeline full of nodal
fibers. The nodal fibers are already there, genetically provided
and ready to receive engrams of memory. The pipeline is
gradually filling up with memory slices all through your lifetime.
The memory-slices are so densely packed that you could live to be
over a hundred years old and not run out of fresh, unused, tabula
rasa memory locations. The gradual fixation or consumption of
memory-slices is like a slow burning fuse, so long that it takes
over a hundred years to burn to the end. Even if you did run out
of fresh memory-spaces in your old age, you would still function
as an intelligent mind with full retention of your many decades
of old memories and with the loss of only your ability to
remember each passing moment of the present. You could still
speak, for instance, several languages and do anything else that
you learned to do before your tabula rasa memory ran out. This
assertion is another one which ought to be judged in the light of
the total mind-model.
The flatness of each memory-channel matters to the brain,
but not to the mind. The serial order or arrangement of the
nodal fibers does not matter at all. Note that the information
recorded in a flat slice of memory is certainly not "flat"
information. The flat memory channel for the tactile sense of
touch contains a sensory mapping of the whole surface of the body.
The flat auditory memory channel contains a mapping of a broad
range of frequencies of sound. The flat visual memory channel
contains two-dimensional images in a one-dimensional series of
fiber-nodes. The mind does not know and does not care that the
images are flat. When the mind associatively recalls an image-
slice, the one-dimensional memory-slice springs to life as if it
were the two-dimensional image seen through the eye.
We are really getting into the dimensionality of the mind
when we bring in the idea of associativity. Sensory information
flows into the mind along the time-dimension, but it moves
sideways within the mind along the associative dimension. Every
sensory memory slice is attached to a "concrete associative tag"
that is like a fiber flowing at a right angle to all the fibers
in the flat memory channels of the time-dimension. These
concrete associative tag-fibers are not shown in the mind-diagram,
because they would completely black out the mind portion of the
diagram. They are called "concrete" (as opposed to "abstract")
because they coordinate by simultaneity all the sensory memory-
slices of "concrete" experience. They are called "associative"
because they are the mechanism by which the mind associates a
memory-slice in one sensory modality with memory-slices in all
other sensory modalities and even in the same sensory modality.
For instance, they are the mechanism by which you might associate
the sound with the image of a dog, and vice versa.
A single associative tag governs a whole memory-slice and
associates it with all the rest of the mind. It may look as
though there is a tremendously unworkable ratio of the vast
information that can be contained in the slice to the unitary,
off-or-on information that can flow over the tag, but it will be
argued in this article that the vast information stored in any
sensory memory channel flows sideways to the core of the mind
solely over aggregates of these unitary, off-or-on "concrete
associative tags." In other words, each lifetime-long permanent
sensory memory channel is quite isolated unto itself and does not
flow at its end into some region of further or final processing
of the sensory information. Wherever the sensory memory channel
comes to an end, it just stops. Let us hope that the end of our
tabula rasa memory channels is so remote that we never reach it
in our natural lifetime. (In an artificially intelligent robot
we might recirculate the memory channels by looping around and
erasing the oldest memory-slices just before reaching the end of
the first full loop of the memory-slices.)
Each sensory memory channel is isolated unto itself, except
for the associative tags which lead away at right (orthogonal)
angles from the time-dimension of the memory channel. Over an
associative tag, you can go from one sensory memory channel into
the memory channels of all other senses. For instance, you can
go from vision to audition, or from olfaction to vision. But you
can go only at a right angle; you can not cross directly by
associative tag from a present memory-slice to one laid down
years or even minutes ago. Each associative fiber that
interconnects all the senses is a guarantee of simultaneity. The
associative tags are laid down at each successive moment of the
fleeting present, and they can never after be disconnected or
altered. As the poet says, "The moving finger writes, and having
writ, moves on."
You must have a thorough comprehension of the sensory and
motor plane or "grid" of the mind before you study the two levels
of superstructure by which mankind achieves rational intellect.
You can maintain that thorough comprehension as we examine the
three levels of complexity which are operative at the peak of the
human central nervous system. The three levels to be studied are:
1. The sentient plane of the sensory/motor grid. (The
interface between the external world and the core of the
mind.)
2. The abstract core of the mind. (This core brings a
central nervous system to the level attained by "smart"
mammals, such as dogs.)
3. The linguistic spiral in the abstract core of the
rational mind.
It is important to go level by level so that you see clearly
what the mind is capable of at each level and what is still
lacking. You should be certain to understand the situation at
each lower level before you study a higher level. As with a
ladder of evolution, each level makes sense by itself and without
reference to any higher level.
So far we have discussed the sensory input part of the
sensory/motor grid, which is the flat, two-dimensional substratum
of the mind. It remains only to explain the role played by the
motor-output side of the grid, and then you should have a
sufficient comprehension of the first of the three levels of the
mind.
Let us call this sensory/motor grid at the lowest level of
mind the "sentient grid." If we were to examine an animal or
automaton that had only such a "sentient grid" at the summit of
its central nervous system, that creature would be severely
limited in its capabilities. It would have the power of brute
sensation, and its repertoire of motor behaviors might consist of
many reflex and instinctual actions which it would be able
crudely to link with sensory inputs as triggers for the
initiation (or cessation) of motor activity. Now let us examine
the motor memory channels, in accordance with the mind-diagram.
The motor memory channels are the polar opposite of the
sensory memory channels. The motor memory channels contain
memory slices not of external experience, but rather of internal,
dynamic activation of themselves. This difference is critical
for your understanding of the sentient grid at the bottom level
of the mind. Motor memory is not passive, it is dynamic. If you
make associative access to a motor memory node on a motor memory
fiber, you unavoidably send out a signal to contract a muscle at
the destination of the associated motor nerve.
As you examine the mind-diagram, notice that the sensory
memory fibers flow in parallel with, but never touch, the motor
memory fibers. Yet the sensory side of the mind controls the
motor side of the mind. "Concrete associative tag fibers" flow
between the sensory and the motor sides of the sentient grid. As
was discussed above with reference to the sensory modalities,
concrete associative tags flow at a right angle to all the life-
long memory channels. Just as the memory fibers are all in
parallel, likewise all the associative tags in the flat sentient
grid flow in parallel. By flowing in parallel, the associative
tags preserve the historical record of each successive moment in
time.
If a central nervous system did not have memory as a record
of experience (and as an enabling mechanism for learning), then
its sensory nerves would have to lead directly to its motor
nerves. No variations of behavior would be possible, and the
whole organism would be pre-programmed genetically to respond to
stimuli always in the same way.
When evolution introduces memory channels, it is essential
to buffer or separate the sensory and motor systems so that they
do not fuse together and so that what intercourse occurs can
occur with great discrimination and precision. Therefore, the
sensory and motor channels do not meet head-on, but rather they
attain a close proximity and then flow in parallel. At each
successive moment in time and experience, the sensory and motor
memory channels have the possibility of becoming linked by nodal
fusing at both ends of the particular concrete associative tag
fiber which was provided genetically for that moment in time.
The whole lifelong tapestry of experience has a fresh, new, blank,
concrete associative tag fiber for each moment of experience,
like a corduroy road made out of logs.
But just how do the associative cross-tags link up sensory
experience with motor dynamism? Why do we call it motor "memory,"
when no experience is recorded there?
The motor memory channel is like a giant keyboard of a piano.
The purpose of the motor memory is not to record events, but to
cause them. Or we could say that the purpose of the motor memory
is to cause an event and then remember how to cause it again.
In the infant organism of our sentient being, a mechanism of
"random dynamics" permits various motor nerve cells to fire
spontaneously. When a motor nerve fiber in the motor memory
channel fires, it causes muscle-activation. Then information
starts flowing in the sentient loop. While the infant organism
randomly moves its limbs, it experiences aspects of that motion
through its sensory apparatus leading into its sensory memory
channels. At each moment in time during the random motion, nodal
fixation at both end-regions of a concrete associative tag fiber
is associating passive sensory engrams with dynamic motor engrams.
Before long, control of the motor apparatus ceases to be random
and spontaneous. Instead, associative control passes over to the
sensory side of the sentient grid.
In the mature organism, all motor activation occurs across
associative tag connections laid down in the past, and present
associative tag connections are made solely for the purpose of
re-affirming or updating or strengthening sensory-to-motor
connections made in the past.
This immediately previous statement offers an explanation
why motor-learning time in infancy is crucial to the development
of motor skill. During infancy, the organism has the benefit of
the random and spontaneous firing of its motor control elements.
The sensory side of the sentient grid seizes upon these random
firings and takes control of them. Once a particular pattern of
sensory memory has taken associative control of a particular
pattern of motor memory, all subsequent uses of that control-loop
are recorded and thus re-affirmed by concrete associative tag,
and a habit of routine or skill becomes entrenched.
Note that this mind-model offers an explanation for volition,
although the explanation is different for each of the three
levels of mind. On the level of the sentient grid, and in the
absence of any higher superstructure, volition consists of
automatic response to the stimulus of a sensory pattern. No lee-
way is allowed in the response to a given stimulus, but varying
stimuli are allowed to elicit varying responses.
Notice something general about the information-loop in which
the sensory and motor pathways do not meet but instead launch
into a parallel race into the future. Remember, the interior of
the mind is trying to mirror the exterior of the environment.
Well, just as things are not steadfast and "hardwired" out in the
environment, likewise on the inside the associative sentient grid,
by flowing through time and allowing all manner of novel
associative connections, can be just as varied and changeable
internally as the environment is externally. However,
an organism with no nervous level higher than the sentient grid
is forced to learn unchanging laws from its environment, and such
a sentient being is not free to make its own decisions by letting
logical data freely interact internally. The sentient organism
lacks an abstract core of the mind where the strict bondage of
stimulus-response can be broken down on the one hand and goal-
directedly built back up again on the other hand.
In other words, if you comprehend the associative sentient
grid which is the lowest of the three levels of mind, you are now
ready to proceed to the examination of the second level of mind.
That is the abstract core which further buffers the sensory and
motor memory channels to such a degree that the formerly ironclad
and inviolable principle of simultaneity in stimulus-response is
overruled in one way but kept intact in another.
The second level of mind is roughly on a par with the
central nervous system of dogs or monkeys or horses. Learning
and Pavlovian conditioning are possible. The organism can be so
"smart" as to impress humans and to generate a sense of kinship.
After eons of evolution, when an organism attains the second
level, the sentient grid of the first level is still present and
operative in the now more evolved organism. The sentient grid
neither withers away nor changes significantly in its operation.
Indeed, in the literature about brains you will find a generally
accepted principle to the effect that lower levels of brains are
designed to operate rather independently of higher levels in the
event of successive breakdown or impairment starting from the
topmost levels. The principle is that the higher level dominates
by consistently inhibiting the lower level, so that, if the
higher level is damaged or removed, the lower level is no longer
inhibited and functions in a role perhaps of inadequacy but
certainly of the best coping ability that the impaired brain has
to offer.
The second level of the mind-model is that of the abstract
core of the mind. If this second level seems ridiculously simple
to you, wait until we fashion from it earth's most complex
mechanism on the third level. But you are correct if you deem
simple the innovation worked upon the sentient grid to raise it
to the second level. The innovation is so simple that perhaps
you will now deign to consider how easily evolution (which "does
not make a leap") could have stumbled upon the wonderful
innovation.
In the sentient grid of level one, there are two massive
neuronal flows at right angles to each other. The one massive
flow is that of the permanent memory channels, both sensory and
motor. These memory channels flow along the time-dimension of
the grid. The other massive flow is that of the concrete
associative tag fibers which cover in blanket fashion all the
memory channels so as to provide their only internal avenue of
connection. Every associative tag fiber is at a right angle to
whatever memory fiber it touches. A memory fiber flows through
the time-dimension, but an associative fiber is frozen at, and
indeed represents, a particular, concrete moment in the lifetime
of the organism.
The innovation in the second level -- the tiny step in
evolution -- involves the lifetime-long memory fibers that flow
along the time-dimension. On the merely sentient level, these
fibers are supposed to contain either sensory or motor memory,
because they are connected either to sensory input or to motor
output. In a level-one system, all memory fibers are "dedicated"
-- either to sensation or to motor activation -- and since the
fibers are not free, the level-one organism is not free. If
evolution had never progressed beyond level one, we humans might
still be starfish or barnacles. But the step or stumble among
the dedicated memory fibers was unavoidably, beckoningly easy to
make, and somehow somewhere long ago in the primordial eons the
great escape was made and they got loose! Some of the supposedly
dedicated memory fibers got away from their origin as elongations
of the pathways to the external world. Getting loose from the
external world, they became creatures of the internal world - and
rational mind was on its way.
The brain-mind diagram of this article is actually more
descriptive of level two than of level one or three. Note the
central core of time-dimensional memory fibers which are not
attached and not dedicated to either the sensory or motor side of
the mind-grid. Since these memory fibers at the core of the mind
are unattached and undedicated, we call them "abstract" fibers.
Once evolution stumbled and let loose of a few of the life-
long memory fibers, these formerly dedicated, now abstract fibers
turned around and took over the course of evolution. As the
embodiment of the negentropic principle, they became an "abstract"
vault of the mind and an ordering force. They set about creating
internal order within the mind. On level two they passively
accepted order from without, and next on level three they will
actively impose order from within.
Throughout this article, the term "abstract fiber" refers
only to fibers in the abstract core of the mind. The term
"concrete fiber" refers only to the associative tag fibers which
lie at right angles to the time-dimension of both the abstract
and the experiential fibers. So there are three types of fibers
in this mind-model: experiential (sensory or motor), abstract,
and concrete.
When the abstract fibers got loose from their dedication,
they did not lose their ability to store memories within their
nodes that lie along each fiber like a chain of beads. They lost
neither their orthogonal juxtaposition to the concrete
associative fibers nor their ability to fuse nodes and thus be
tagged by the associative fibers. Since they no longer had any
direct source of memory data, either sensory or motor, the
abstract fibers could henceforth be filled with memory-data only
by receiving inputs sideways from the concrete associative tags,
and that indirect, abstract function is what they fulfill even to
this day. An abstract fiber in the core of the mind serves
associatively as a unifying fiber which crosses time-boundaries
and interconnects potentially all original and re-occurring
instances of the experience of a particular pattern of perception.
A sensory memory fiber is for sensation; an abstract memory fiber
is for perception.
In order to understand how an abstract memory fiber works,
you must keep in mind the two-fold mechanism of original
association and subsequent reaffirmation. The original, neonatal
sensory inputs to level two of the mind flow first directly into
memory nodes in the sensory memory channel and thence indirectly,
associatively, via the concrete associative tags, into memory
nodes in the abstract memory channel. In a newly constructed
organism (such as a baby), the first memory deposits are of a low
level of complexity. The abstract memory channel stands ready to
receive and record whatever inputs are fed to it across the
associative tags. Therefore, in the earliest moments of memory,
identical engrams are formed in the sensory and abstract memory
channels. At its neonatal origin, the abstract memory channel
mirrors the sensory memory channels. Remember, the abstract core
of the mind is trying to mirror the external world, which it must
perceive through the medium of the sensory channels.
However, as time goes by, each abstract fiber becomes
extremely differentiated from its neighbors. The original level
of complexity of the data in the abstract memory channel is on
the order of off-or-on and yes-or-no. This irreducibly simple
logical content is the mirrored reflection of a jumble of data in
the sensory memory channels. The sensory memory channels never
actually become organized internally, but the abstract memory
fibers do become organized. Order develops within the abstract
memory channel through the incessant and potent mechanism of
associative reaffirmation.
Please examine the abstract memory channel from the point of
view of identical contents being held in both the sensory memory
channels and the abstract memory channel. Suppose that through
the eye a particular feature, such as a geometric line, has been
seen and recorded, first in the visual memory channel, and
simultaneously by associative tag in the abstract memory channel.
Every subsequent time that that particular feature is seen again
along the same sensory memory fiber, two important events will
occur. The one rather simple event is that the sensation of that
feature will be recorded one more time within a freshly fixed
node at that point along the sensory memory fiber where the march
of time is presently fixing nodes by simultaneity across a wide,
associative front. Meanwhile, as the signal of the sensed
feature travels along the sensory memory fiber and briefly floods
the fiber at every point, the originally fixed node is faithfully
doing its duty as a comparison device. By simple unitary logic,
it recognizes the (umpteenth) reoccurrence of the signal of the
same sensed feature with which it was originally fixed or written
as an engram.
The sensory memory node, stimulated by the transient signal,
blips out a signal across its associative tag over to the related
node on the related abstract memory fiber. Now in turn the
abstract memory node, stimulated by the transient associative
signal, blips out a signal which travels down the abstract memory
fiber to where unfixed tabula rasa nodes are being fixed by every
data-laden moment of the present. So now we have a mirror
phenomenon occurring in both the sensory memory fiber and the
abstract memory fiber. The associative tag fiber of the present
moment fuses across nodes on both the abstract memory fiber and
the sensory memory fiber. Thus the logical content and the
"dedication" of the abstract memory fiber are reaffirmed by
simultaneity in the present moment of perception.
The concrete associative fiber of the present moment of
perception will fuse with sensory and abstract nodes wherever two
or more signals are present orthogonally. Suppose that the eye
of the organism is seeing an image or pattern composed of many
features. Each extracted feature floods its own sensory memory
fiber within the visual memory channel. The concrete associative
fiber of the present, which is activated by an internal clock of
the brain, fuses nodes with any feature-fiber that is momentarily
being activated by the total sensation of the image or pattern.
Therefore, this concrete associative fiber is henceforth
irrevocably linked to the group of features which comprise the
seen image. Henceforth this associative fiber can either recall
the image internally or recognize the image seen again externally.
The concrete associative fiber is now an associative "tag"
attached to the image.
Although the associative tag may connect to many fibers in
the sensory memory channel, it can connect to as few as one
single fiber in the abstract memory channel. Thus a single fiber
in the abstract memory channel can come to represent a whole
class of fibers in the sensory memory channel and lo, an abstract
concept is born.
If you pause to think, you may see how it makes sense that
often multiple fibers will be activated in the sensory memory
channel while only one or a few fibers are activated in the
abstract memory channel. In the neonatal period, there may be a
releasing mechanism which lets loose of only a few abstract
fibers at a time. Or the abstract fibers may compete to be the
first abstract fiber to be reaffirmed by the associative tag over
to a bundle of sensory fibers comprising a pattern. The main
thing is, each abstract memory fiber can serve as a reaffirmative
collection-point for associations to a whole class of similar
sensory patterns. Voila -- pattern recognition occurs. The
abstract fiber is not in the thick of sensation; it stands aside
and is abstract.
An abstract memory fiber (spoken of in the singular here,
although a gang of thousands of logically fused fibers is meant)
can become the physical and logical seat of a concept in the mind.
For instance, a dog that knows and recognizes its master will
have at least one abstract memory fiber which serves as the
ultimate, concentrated association-point for memory-information
related to the dog's master. This assertion is so serious and so
evocative of hasty disbelief that it is now time to invoke the
force of the dimensionality of the mind.
The level-two mind has two dimensions, the lifelong time-
dimension and the simultaneity-dimension. Within the level-two
mind (and the level-three mind), memory fibers flow in parallel
and only along the time-dimension.
You know from experience that your mind has held a concept
of something or other, such as a concept of the sun around which
our earth orbits. All your knowledge of the sun is tied to that
concept, and that concept is tied to the word "sun." Of course,
your conceptual knowledge of the sun could be broken down into
ingredient concepts, such as the concepts of warmth or light or
chariots. But it seems as if you have one unitary point within
your mind where all the constituent concepts are subsumed under
the operative concept of "sun." So the dimensionality of your
concept of "sun" is punctiform. If your concept of sun were
triangular or circular, you would not be able to focus your mind
upon the same pinnacle of conceptuality each time that you think
about the sun.
But your concept of the sun is not only unitary, it is also
quite constant over time. Just as a point extended through space
becomes a line, likewise a unitary concept held constant over
time can best be represented, both physically and logically, as a
unitary fiber (or its logical equivalent, a gang of fused fibers)
flowing along the time-dimension of the mind. The dimensionality
of a concept is double: it is punctiformly unitary and it is
chronologically linear.
Does it seem ridiculous that this mind-model claims that
perhaps a single gang of fibers in your brain holds your concept
of a thing such as the sun, or of your pet dog, or of yourself --
your concept of ego? But think: the concept-fiber is operative
not by itself, but by virtue of the myriad associative tags
leading from it. Many concepts are interrelated and they
contribute to the composition of one another. Conceptual fibers
are associated not just to sensory data, but also to one another
within the abstract memory channel. Therefore a slice of your
abstract memory channel is like a conceptual topography. The
maze of concepts is like a stick-forest of interrelated points of
knowledge. Concepts are neighbors or relatives of one another
not by physical proximity, but by logical proximity.
Your pet dog has a stick-forest of concepts, but, alas! he
has no words (or symbols) attached to them and therefore he can
not manipulate them in a rational way. Even though your dog may
hear words quite often, he does not develop the use of words.
Your baby, however, quickly develops the use of hundreds of words.
How is the level-three mind of your baby different from the level-
two mind of your dog?
On the third level of mind, rational intellect springs into
being in a process whereby rigid informational structures arise
amid the hodgepodge informational milieu which was level two.
These new structures arise as the means to express relationships
among concepts. They are to some degree logical structures and
to a larger degree linguistic structures. The structures remove
the mind from the bondage of immediate, concrete experience and
allow the genesis of abstract thought.
We can first examine the existence in the mind of a
vocabulary of words solely with respect to level two, and then we
can describe the level-three structures which govern these words
in linguistic thought.
Let us discuss the relationship between word-memories in the
auditory memory channel and image-memories in the visual memory
channel. Let us confine our discussion to concrete nouns which
are readily linked to concrete images.
First of all, the association between the two memory
channels is a two-way street. Activation of the image can evoke
the word in auditory memory, and activation of the word can evoke
many images in visual memory. Note that "word" here is singular,
but "images" is plural. This difference obtains because a single
word can serve as a control-symbol for a whole class of images.
For instance, if you see any one of many varieties of dog,
the word "dog" can come to mind in your auditory memory channel.
If many people listening to a story hear the word "dog," they
will probably summon up quite varying images of dog to
instantiate the concept of "dog."
Humans with words as control-symbols have an extreme
advantage over the level-two minds of animals. The word attached
to a concept makes that concept utterly and fluidly manipulable
within the ratiocinative structures of the mind. Even though the
word is an extended string of phonemes, it behaves logically as
if it were a unitary point.
Indeed, in the level-three mind, each word is attached to a
unitary point, namely an abstract conceptual fibergang associated
with the word in the abstract memory channel.
In a level-one mind that contained words, there would be a
direct associative link between an image and a word. In the
level-three mind, concrete associative tags do not flow directly
between images and words. Instead, from the sensory memory
channels the associative tags make contact with the abstract
conceptual fiber, which is the focal embodiment of a particular
concept and which serves as a unifying point for the development
and linguistic activity of the concept. If a linguistic
structure is going to control a vocabulary of words, each word
must have a sort of "handle" upon it, by which the word, as a
symbol, can be controlled. That handle is an abstract conceptual
fiber.
The abstract memory channel is a set of all abstract memory
fibers. An abstract conceptual fiber is an abstract memory fiber
which happens to hold a concept (by gathering up the associative
tags of a concept). Therefore the set of all abstract conceptual
fibers is a subset of the set of all abstract memory fibers.
Thus far in our discussion, a concept has a tripartite
existence within the brain-mind. Firstly, the word exists as a
short string of sounds within the auditory memory channel. Note
that no word will exist at only one memory location within the
auditory memory channel, but rather each word will be recorded
there in hundreds or thousands of historical instances, depending
upon how frequently the word is used. Furthermore, be very aware
that, since each instance of the word is the same string of
sounds (phonemes), all instances of a word within the auditory
memory channel are logically equivalent. Since the auditory
memory channel is not just a transmission-channel, and not just
a memory-channel, but also a comparison-channel, any one instance
of a word can quickly be compared with all other instances of the
same or even a similar word, so that a word existing in thousands
of spots within the auditory memory channel functions as if all
the spots were interconnected, as indeed they are. To illustrate
this point, think of the word "dog" and how you can conjure up
many different images of "dog."
The second part of the tripartite existence of a concept is
at the abstract conceptual fiber for the concept. The abstract
conceptual fiber is the main and focal seat of the concept within
the mind. From the abstract conceptual fiber, thousands of
concrete associative tags flow across the sentient mind-grid to
make reference to and control word-engrams in the auditory memory
channel. If you hear (or think) a particular word through your
auditory memory channel, that word instantly gains access, across
at least one of the concrete associative tags, to the abstract
conceptual fiber for that word so that your understanding of that
concept is activated within your mind. Likewise, if, in the
interplay of concepts within your mind, that particular concept
fiber is activated, the following scenario takes place.
From the activated concept-fiber, thousands of concrete
associative tags flowing in parallel are activated in parallel.
Only one of them has to reach the word-engram in your auditory
memory channel for that word to be activated and flow through the
channel to the present-most end of the consumed portion of the
tabula rasa channel. In all likelihood, many of the tags will
gain access to the word, but, since it is the same word in all
instances, your mind will hear just one standard production of
the constituent sounds of the word. Note, however, that the
parallel activation of thousands of concrete associative tags
serves, by sheer redundancy, to make for an extremely reliable
mechanism for the internal recall of words during verbal thought.
Note also that your auditory memory channel is a self-perceiving
channel. Although word-engrams are controlled en masse by the
abstract conceptual fiber outside of the auditory memory channel,
we are consciously aware of the words only as they flow within
the auditory memory channel.
The third part or area of the tripartite existence of a
concept within the mind is spread out over all the sensory memory
channels which are associatively connected to the abstract
conceptual fiber of the concept. If the concept is evocative of
images (or sounds or smells or feels or tastes), then from the
abstract conceptual fiber many concrete associative tags will
flow out orthogonally over to the sensory memory data which the
unitary concept represents. An abstract conceptual fiber may be
associatively connected to many visual images, not all of which
are necessarily identical or even similar to one another.
Remember, a word is always the same, but most images will have at
least minor differences. Such a state of affairs is fit and
proper, because a word is an unchanging symbol, while an image is
just a variable slice of the rich pageantry of experience.
An abstract conceptual fiber reigns supreme as the unitary
point under which or toward which all the constituent information
of a concept is subsumed. The abstract concept develops or grows
by the accretion of concrete associative tags over time. The
abstract conceptual fiber is not itself a symbol, but it is often
attached quite fixedly to a symbol, namely a word in the auditory
memory channel.
The abstract conceptual fiber governs both the word attached
to aconcept and also the sensory data associated with the concept.
An abstract conceptual fiber can have concrete associations not
only to sensory engrams, but to other abstract conceptual fibers.
This ability of a concept to exist within a network of related
concepts allows the genesis of such truly abstract and intangible
concepts as our notions of "honesty" and "courage."
Remember that all the abstract conceptual fibers flow in
parallel in a flat plane along the temporal dimension of the mind.
The logical relationships among abstract conceptual fibers are
determined not by physical position, such as contiguity or
proximity, but solely by interconnection over concrete
associative tags. Thus, although the fibers lie in a flat plane
across the surface of the brain-mind, their associative inter-
connections can generate the analog of superstructures or
hierarchies among the abstract conceptual fibers.
To discuss the psycholinguistic nature of language, we must
for the first time in this article introduce the notion of the
control of one abstract conceptual fiber over one or more (i.e.,
thousands) of other abstract conceptual fibers. Up until now we
have discussed how one fiber might influence another fiber, but
not how one fiber would dominate another.
The ability of a neuron to require the summation of multiple
inputs, before firing, permits some fibers to control others. In
that portion of the abstract memory channel which we may hence-
forth call the "linguistic cable," some abstract fibers gradually
take on the role of governing and dominating whole classes of
other fibers. For purposes of simplicity and clarity, we discuss
here only two linguistic classes of words: nouns and verbs.
As an infant learns nouns, he or she also subconsciously
assigns abstract fibers in the linguistic cable to the control of
the whole class of nouns. As each new noun is learned, concrete
associative tags are bonded from general noun-control fibers over
to the abstract conceptual fiber of the particular noun. From
the noun-fiber in turn a concrete associative tag goes to the
engram of the word in the auditory memory channel. Gradually the
noun-control fiber latches on to a burgeoning "family" of nouns,
all segregated conveniently as a class so that they will remain
distinct when other parts of speech are learned.
Suppose that the infant, seeing and recognizing an object,
wants to name that object in a blurt of speech. The "wanting" is
actually the build-up of logical tension within the abstract
memory channel. The general noun-control fibergang is activated
by the confluence of all the logical tension stemming both from
the perceived object and from the internal state of the infant.
This general noun-control fiber-gang sends a blanket semi-
activation signal to all the nouns in the vocabulary of the baby.
In a way, all the noun-fibers are being invited to activate their
word-engrams in the auditory memory channel. But, because of the
multiple-input requirement, no noun-fiber can fire solely on the
basis of the blanket semi-activation signal going out to all
nouns as a class. Only that noun-fiber can fire which is already
or simultaneously semiactivated, so that the two semi-activations
cause full activation, and a recall-signal is fired over to the
word-engram in the auditory memory channel.
Remember, the baby is seeing an object out in the real world.
The perception of that object causes associative links to filter
through and semi-activate the one noun-fiber within the whole
class of nouns. The desire to speak a word causes the general
noun-control fiber to send the blanket signal to all noun fibers.
Two semi-activation signals -- the blanket one and the specific
one -- meet in the appropriate noun-fiber and cause it to fire a
recall-signal over to a word-engram stored in the auditory memory
channel. In this system, if the infant has not yet learned the
most appropriate word for the perceived object, he or she will
blurt out some nearly appropriate word which bears the closest
associative relationship to the perceived object. The word
chosen by the baby may sound funny to adults, but it makes sense
within the mind of the infant.
In like manner, an abstract control-fiber for each part of
speech governs all the members within the class of that part of
speech. When the infant goes on from learning nouns to learning
verbs, likewise general verb-control fibers govern all available
verbs.
Once we clearly make the point here that one abstract gang
of control-fibers for a particular part of speech can govern all
the members of the class of that part of speech, we have finished
the fundamental description of level three of the mind and we
have described the part-of-speech building-blocks which make up
the sentence-structures in natural human languages.
If we describe a particular human language, we move from the
internal domain of genetically provided, universal deep features
of the level-three mind out to the external field of cultural
tradition. We see the innate ability of the mind to segregate or
classify various parts of speech, and we see the cultural ability
of the mind to concatenate part-of-speech control-fibers into
sentence structures. The combinatorial power of the linguistic
portion of the abstract memory channel allows many influences to
affect and determine the dynamic operation of sentence structures.
These influences can include considerations of number, logic,
time or tense, emotion, and so on. Any semantic consideration
that can be conceptualized (preferably subconsciously) can be
represented as a control-fiber which figures in the composition
of sentence structures within a natural language.
This paper does not attempt to formalize the representation
of natural language within a machine mind. We avoid such
formalization by means of utter simplification, and then we leave
the elaborate formalizations to the expert professional linguists.
Our utter simplification of human language consists here in
treating language as if it had only two parts of speech: nouns
and verbs. We want to simplify language so utterly that readers
will, on the one hand, grant that noun-plus-verb is the essential
core of human language, and, on the other hand, comprehend how
this design for a mind generates utterances consisting of noun-
plus-verb.
Therefore, instead of formalizing an elaborate design for
one of the natural languages, we ask the following common-sense
questions. Is it not clear, that a mind, which can grasp the
concept of the doer of some action and then link that concept,
expressed as a noun, with another concept (that of the action
itself expressed as a verb) has performed the basic linguistic
feat which is both representative and definitive of human
linguistic achievement? Is not everything else refinement and
enhancement?
This design does not beg the question by declaring an easy
system of syntax and by ignoring semantics. The foregoing bulk
of this article has laid the semantic groundwork for proposing
that part-of-speech control-fibers are the semantic building-
blocks which the mind concatenates into the sentence-structures,
or syntax, of a human language. This informal simplification of
language is meant as a common meeting-ground for a view of
language and a view of the brain-mind.
Each abstract-memory control-fiber gang for a part of speech
becomes a node on a sentence-structure of concatenated nodes.
The nodes are concatenated by a spiral of linguistic habituation.
Just as an associative tag fetches a word stored in the auditory
memory channel, another associative tag, attached to the end of
the stored word, sends a signal back to the sentence-structure,
reporting that the task of one node is complete and that now the
next node should go into operation. Thus dynamic control of the
semantically driven process of sentence-generation shifts back
and forth between the abstract memory channel where the syntax is
stored and the auditory memory channel where the words are stored.
This shifting back and forth, although it happens in the flat
plane of the mind grid, is extended over time and is logically
complex enough to be the flat analog of a spiral winding through
time.
Each use of a sentence-structure reaffirms the habituation
of the sentence-structure. Any typical node in the sentence-
structure can be added or deleted by the habituational device of
practice. The associative tags which operate under the (short-
term) domination of a sentence-structure exercise their own
(long-term) domination over the sentence-structure by reaffirming
and habituating it. Change is caused from without, but then each
subsequently identical loop of the spiral takes hold of what was
initially change and habituates it into a long-term structure.
The concatenated nodes of sentence-structures within the
abstract memory channel reach over, so to speak, via associative
tags and string together words and morphemes within the auditory
memory channel. We hear our own verbal thought within our
auditory memory channel.
When this system of generating sentences is worked in
reverse, it comprehends sentences by decoding the associations
among concepts conveyed by the linguistic sentence-structure. In
the comprehension of a sentence, new associative links are formed
among the abstract conceptual fibers in the abstract memory
channel of the receiving mind. The sentence is recorded both as
an episode in experiential memory and as a slight rearrangement
of the associative links among abstract conceptual fibers in the
abstract memory channel.
In this system, an incoming sentence does not have to be
believed. The entrenched, pre-existing associative links in the
receiving mind can withstand and overwhelm the links asserted by
the linguistic structure of an incoming sentence.
This design seeks to explain how a multi-lingual speaker can
keep his or her languages apart and avoid running them together
while speaking. Since the vocabulary items are all segregated
down at the deep levels, they remain segregated at the highest
level, that of the particular language.
If you build an artificial mind, do not try to program it
like a computer. Build it, turn it on, and commence teaching it.
Association for Computing Machinery (ACM)
Special Interest Group on Programming Languages (SIGPLAN)
Frenger, Paul -- ACM SIGPLAN Notices (1998 and 2004)
Mind.Forth: Thoughts on Artificial Intelligence and Forth
Ask your local science museum if they have an AI Mind exhibit.
If they do not, volunteer to install
MindForth as an AI exhibit.
See which museum has the oldest living artificial intelligence.
AI4U: Mind-1.1 Programmer's Manual
describes the software implementation of the above theory of mind
and is to be found at prestigious universities in the library collection:
