TABLE OF CONTENTS
Back to top
|Chapter One - The Biological
Models And Myths
The Slate, The Ghost And The Savage
The First Cognitive Revolution
The Second Cognitive Revolution
Impact On Our Daily Lives
How The Mind Really Works
Plan Of The Three Books
|Chapter Two - Pattern Classification
Pattern Classification In The Wild
Examples Of Human Pattern Classification
Machine Pattern Classification
Training A Machine Pattern Classifier
|Chapter Three - Neural Networks
Why Are Neural Networks Interesting?
A Model Of Biological Neurons
Pattern Classification With Neural Networks
Training Of Neural Networks
Feedback In Layered Networks
Modeling Biological Brains
|Chapter Four - Representation
In Animals And Humans
|Chapter Five - Modeling Primary
Back to top
like to comment on what got me started writing this book, give some
explanation of my background, and discuss my rationale and perspective.
At a symposium on cognition and the humanities that I attended in the
mid-1980s, a heated argument broke out after the screening of an independent
science fiction film. The film’s plot was about removing a volunteer’s
brain to a vat where it would be connected via wires and electronic
communications to his head. This drastic surgical procedure was required
to prevent the volunteer’s brain from being destroyed by radioactivity
when he descended into the ocean depths to retrieve a damaged nuclear
device that threatened to kill all life on our planet. In the movie,
the brave volunteer (played by the mind science philosopher Dan Dennet)
first underwent a test where he could switch over from his own brain
in a vat to a back-up external computer brain. By switching back and
forth, he could determine whether the external computer brain functioned
as well as his own biological brain.
appeared at the symposium wearing the brain switch on his belt and frequently
“switched over,” to the great amusement of many people in
the audience. However, a number of other people in the audience were
very upset by the concept of an equivalence of a machine brain and a
biological brain. My impression of the loud and acrimonious debate that
followed the screening of the film was that the humanists and the scientists
were unable to communicate with each other. It’s not that they
simply disagreed but that neither side could clearly articulate what
the disagreement was actually about.
battle cry was essentially, “I am not a machine!” while
the scientists’ response was, “Yes, you are!” But
that was about as deep as the discussion got. I felt that both sides
of the debate needed to deeply rethink and more clearly explain their
ideas so that a superficial argument could become a fruitful debate.
I wrote these books to fill that need.
I am aware
that at first glance this issue of whether the brain is computational
or not may appear to be an exotic scholarly debate of little interest
to a broadly educated person. To counter this view, I first point out
that the people who get involved in this discussion become very emotional
about it very quickly. Something very deeply ingrained in our natures
is troubled by the issues raised by the computational mind debate. Moreover,
the issue has deep consequences for how we view our own intelligence,
how we manage our lives and how we educate ourselves.
I Am Not a
Machine addresses the deep issues that provoked the arguments I witnessed
almost twenty years ago and continue to witness now in a variety of
settings. Many people have very strong gut reactions against computational
models of the mind. It could conceivably be the case that the humanist’s
objections to these models are no more founded than the humanist’s
initial objections to Galileo’s theory of a sun-centered solar
system or Darwin’s theory of the evolution of man. However, the
jury cannot decide this issue based on their gut reactions or based
on the track record of the defendants. We must dig deeper into the issues,
and my book does exactly that.
negative reaction that people have to being compared to computational
machines is related to some complex issues. There is something deeply
wrong with the way that both the public and many scientists think about
such diverse issues as truth, DNA codes, and the human mind. It’s
more than the desire for simple answers to complex questions. It’s
a belief in binary yes/no dichotomies supported by a computational interpretation
of how the mind works.
early symposium, many other researchers have also taken great strides
to meet the need to clarify the debate. One recent contribution is a
well-argued book by Gary Marcus entitled The Algebraic Mind: Integrating
Connectionism and Cognitive Science. In it, Marcus argues persuasively
for a view that the mind is fundamentally computational—that is,
it operates via symbol-manipulating algorithmic processes.
The view that Marcus
argues against is often called connectionism because its essential idea
is that the mind is best modeled as a complex interconnected network
of neuronlike elements that operate without symbols or algorithmic rules.
Researchers who have argued with conviction for this connectionist view
include Jeff Elman, James McClelland, David Rumelhart and William Bechtel.
afraid that I do not accept the arguments of either camp. First, let
me say that they all make very clear, intelligent, and plausible arguments.
In fact, what either side says could be true. However, I think both
camps suffer from too much computer science and not enough biology.
Both approaches also suffer a huge “can’t get there from
here” problem because neither approach offers an evolutionary
path from simple organisms to more complex non-human primates and then
to humans who use grammatical language. Unfortunately, it will be a
long time before we have detailed enough empirical data to conclusively
resolve the issue. In trying to understand how the mind works we all
do what Steven Pinker calls “reverse engineering.” We all
bring to that task our own education and experiences that inform our
understanding of how to design things.
of the ideas in this book may be new, even to readers of the traditional
cognitive science literature, I have taken great pains to start at the
beginning in all areas. Thus, I hope that humanists, scientists, and
the ideologically open-minded and curious reader will find my book engrossing
model also bridges the evolutionary gap from animals to the early hominids
that began the journey towards human spoken language. Human natural
language is enormously more complex than the body language and simple
calls found in most animals. Human language is syntactic; that is, it
is composed of many sub elements that combine to form larger units.
We combine a repertoire of syllables to form words and we combine words
to form sentences. This complexity has compelled most cognitive scientists
to a model of the human mind that is symbolically driven. They simply
could not conceive of another alternative. As implausible as a computational
model was to many scientists, they had to embrace it or face eternal
puzzlement––a disastrous state of mind for most people but
especially for a scientist. This computational notion of mind is not
new. Although sophisticated computers are new, deep thinkers such as
Descartes conceived of an “engine of reason” and even some
Greeks thought in terms of idealistic mechanistic models of the world.
I realize that these old views are subtly entrenched in many people’s
minds and that I have a tough row to hoe if I am to entice them to consider
an alternative view of the mind, but that’s the challenge I have
like to remind you that science has two phases, a hypothesis generation
phase and a hypothesis validation phase. When we validate, we need to
make our argument linear and clear. Checks with other concepts and checks
with the real world must be formal; that is, they should be set out
in a structure so that others can check the reasoning and the data.
That’s the hard-work part of science and
I am afraid I must leave it to the hard workers. I’ve taken on
the easy job of generating some new ideas, that is, the job of hypothesis
generation. When I look at the old ideas I do not think they are misguided
because I see a small chink in the armor; in fact, viewed each step
at a time, the computational argument appears very solid. However, I
believe the computational argument is wrong because the whole pattern
does not make sense. My presentation is something of the opposite. My
argument has not yet been reviewed by hundreds of scholars and therefore
the steps of my argument are not as solidly made as others will someday
make them, but I do think that the whole model of mind I present is
very solid. This book is not intended as the end of a discussion but
rather as part of the conversation.
THE BIOLOGICAL MIND
The essential difference between
a human mind and a nonhuman primate mind is language. To set the stage
for unraveling that statement we introduce ideas associated with language,
computational machines, metaphors, models and the first and second revolutions
in cognitive science. We also discuss some practical applications of
Can animals think? Do they have
language? Can an Orca whale sing to its pod, “See you guys in
the spring off San Juan Island?” I have found that many people
have very strong, yet divergent, views on such questions even though
they apparently have not read about or reflected on the questions before.
One of the things that make such discussions difficult is that people
often don’t have a foundation of concepts for understanding and
describing human or animal thinking.
My focus has almost always been
with human thinking but I found that I needed to understand animal thinking
first. No problem, I thought. A week’s digression is all that
I needed. No such luck. I ended up writing this little book. My approach
to understanding animal thinking is to ask not what animals think but
rather how animals think. In particular, how do animals think without
using words? Many people including myself, but by no means all people,
believe that humans also can think without using words. Visual artists,
architects, and musicians may head that list but athletes and car mechanics
are not far behind. Consequently, characterizing thought without words
is important for understanding animal and human cognition.
This is the first book in a series
of three books about thinking and the biological processes that make
it possible. Thinking about thinking is a tricky business not because
of the confusing play on words or some vague notion that study of the
self is impossible, but because, like riding a bicycle, thinking is
so automatic and effortless that it’s very difficult to tease
apart its component processes.
It’s also unnatural to think
about thinking. It’s natural for boys to think about girls and
vice versa. It’s also natural to think about baseball, politics,
and the neighbor’s new car. However, thinking itself is something
of a mystery for most of us and there is sometimes an underlying sentiment
that we should just let the mystery be. Well, I can’t. I’m
hooked on trying to better understand the human mind. I’m not
the first and I’m sure I’m not the last person with this
obsession. In fact, in some small circles, and not just academic circles
either, thinking about thinking is almost considered cool.
For better or worse, I am not part
of a well-defined circle of people that thinks about thinking. Consequently
my ideas may be a bit different from what you’ve been exposed
to. I have no way to evaluate my own ideas. No one does. We rely on
our peers to do that and I’m going to be relying on you.
I have found it extraordinarily
difficult to give a one-minute summary of my ideas—what some people
call an “elevator speech.” The ideas that I find tie things
together don’t seem to have much of an impact on people who have
been kind enough to listen to me. Sometimes a consequence of an idea
is what catches people’s attention—as for example Einstein’s
claim that light always travels at the same velocity and that nothing
could travel faster. Now that’s catchy because you can say it
in one sentence and it’s very counterintuitive so it makes you
wonder. I can’t come close to matching that but here is my attempt
to catch your attention.
The essential difference between
a human mind and a nonhuman primate mind is language (French, Finnish
or Farsi, whatever). Human intelligence is based on language. Language
is not merely a means of communication as asserted by most scientists
who study thinking. By language, I mean natural syntactic language,
that is, a language made of simple pieces that combine into complex
pieces. In most human languages, letters combine to make words and words
combine to make sentences. The communicative acts of bees, dolphins,
and cats are not based on this combinatorial process and consequently
the complexity of their communication and thinking is severely limited.
For the purposes of our discussions these animal communicative acts
are not considered to be language.
I suspect we all agree that nonhuman
primates are very intelligent and fortunately we humans are blessed
with that form of intelligence as well. I call this non–language-based
intelligence primary cognition. By contrast, I call language-based cognition
secondary cognition. I think you might now be ready for an example.
I claim that in an afternoon you cannot teach a chimp to efficiently
parallel-park a car in a tight parking spot or teach a chimp to make
Coq au Vin. Please ignore some practical issues such as how the chimp
touches the gas pedal or holds a knife so that we can go to the heart
of the matter. To quickly parallel-park a car you need to follow a simple
procedure. A possible set
of instructions is “Pull up beside the car in front of the parking
space, leaving about one foot of space between the cars; back up straight
until your rear wheel is even with rear of the front car, then turn
your steering wheel hard right; continue backing until your front wheel
is aligned with the rear wheel of the front car, then turn your steering
wheel hard left; finally, straighten your steering wheel when you are
parallel to the curb.” If you are feeling gutsy or rude, ask someone
struggling to parallel-park a car what procedure he is following and
I am sure that you will be given the answer, “Huh? What procedure?”
Some people just avoid parking in
tight spaces and others laboriously go back and forth endless times
as they get into a parking space. Many of them will not even know there
is a simple procedure or for one reason or another don’t like
cluttering their minds with procedures. It’s cooler to wing it.
A chimp might be able to wing it as well if you could get him to understand
the goal of the exercise. The point is that the procedure can be expressed
as just a few words and those words code a set of things to do with
your body. You can learn to do a complex set of body motions by carefully
watching but it is very difficult and takes a long time. Language makes
almost impossible tasks easy.
Making a complex recipe such as
Coq au Vin can also conceivably be learned by a creature without any
language by just watching someone do it but it would probably take a
lifetime or two. Each step of the process requires specialized skills
represented in complex body experiences. By using language, a creature
makes the steps easily labeled and recalled when needed. Spoken language
helps a great deal but written language is even better.
Creatures that can parallel-park
a car or make Coq au Vin appear to be intelligent. It has been tempting
for scientists to define intelligence in terms of the kinds of tasks
a creature can do. The popular SAT test we give college-bound high school
students and even IQ tests tend to define intelligence
in this way, that is, by a list of things a creature can do. Although
an amorphous fundamental capability may conceivably underlie the list,
such test-based definitions of intelligence are inherently circular.
We say a human is more intelligent than a chimp because he can make
Coq au Vin but then in our next breath say the reason he can make Coq
au Vin is because he is more intelligent. This is analogous to sports
announcers who occasionally say that such and such a player is a better
athlete because he
has more athleticism.
Please note that I am doing two
things here. I am listing the underlying capabilities that make a human
more intelligent than a chimp and I am limiting that list to just one
thing, natural syntactic language. I have no illusions that this is
an easy sell, for if it were I wouldn’t be writing these books.
The point of putting this outrageous claim upfront is to let you see
just how different my ideas are from what you are used to.
There is virtually no such thing
as an isolated fact in science or in any unified body of knowledge.
Consequently my little claim is not a little claim at all. It has ripples
that spread throughout all aspects of theories of the human mind. That’s
why we will have to deal with some fundamental issues in the science
of the mind in order to understand and evaluate my simple, outrageous
claim. That, by the way, is where the fun lies. In trying to understand
the mind from a new perspective we will have to reexamine many cherished
views of what it means to be a human and what it means to be an animal.
This is not an activity that the fainthearted would enjoy but I suspect
that you are up to the challenge, else you wouldn’t have started
to read this book in the first place. I’d like to say something
else about my claim that the significant cognitive difference between
humans and chimps is language. Obviously there are lots of differences
between them such as chimps have more hair. So my claim is that these
other differences are not necessary to explain the differences between
the minds of chimps and humans. I want to identify just those differences
that are absolutely necessary and that are sufficient to explain the
crucial aspects of chimp and human thinking. Many of the physical differences
between chimps and humans are important and greatly contribute to our
understanding of ourselves and our place in the world. Our dexterous
hands, for example, allow us to do many things and this contributes
to our so-called intelligent behavior. Our dexterous hands also have
a decisive role to play in how language is processed in our brains.
However, a creature with dexterous hands but without syntactic language
would not be able to create modern civilization.
You might well be thinking that
you get the gist of what I believe but now you want to know why I believe
it. That’s a fair simple question but I can’t give you a
fair simple answer. I have developed a new perspective on human and
animal cognition and I need to explain and justify that entire perspective.
That is what these three books are about. I realize that when we teach
traditional concepts in any domain we do so in a style reminiscent of
a religious catechism. However, that pedagogical style only works for
well-known and accepted ideas. The answers given to simple questions
are not really answers at all but placeholders for more complex answers
that take considerable effort to develop and understand.
What I will do now is mention some
of the other cherished beliefs that I will attempt to dissuade you of.
Many readers will have heard that natural language such as English has
a surface structure and a deep structure. The surface structure is the
“i before e, except after c” variety we were taught in grade
school and is often called the grammar of a language. The deep structure
is often called the formal syntax of a language and is what linguists
study. Most linguists believe that this syntax is a formal system, much
like a computer’s formal programming languages. The human mind
is thus also seen as a formal computational system. I claim that this
is not true. I believe that the mind is inherently an analog system,
more like a steam engine than a computer. Things do things to things
rather than things following a procedure specified by rules as in a
game of cards.
Another arena that we will reevaluate
surrounds Skinner’s stimulus response theories. It’s clear
to virtually all of us that Skinner’s theories are naive when
applied to humans, yet in applying them to nonhuman animals we need
to be more cautious. Although many scientists have said that animals
are too intelligent for stimulus response theories to be true, no one
has made a proposal delineating what thinking mechanisms might be operating
for the trillions of trillions of the earth’s simpler creatures.
I will attempt to persuade you that stimulus response processes cover
a great deal of animal cognition but we will ultimately see that they
are not sufficient. I will need to posit two additional cognitive features.
One feature has to do with how a creature represents what is out there
and the second has to do with the ability to “plan and evaluate”
the very next motor action a creature might make. I put “plan
and evaluate” in quotes because how animals do it is significantly
different from how you and I do it using language. The simple process
I will present to you eliminates the need to endorse a theory of animal
thinking based on formal computational processes. Understanding how
animals think turns out to be absolutely essential to understanding
how humans think and my theory fills that need. This first book in the
series of three books addresses the cognition of creatures without language
and that part of human cognition not supported by a natural language
such as French, Finnish or Farsi.
Moving forward at warp speed, I
want to say something about Mr. Ithink- therefore-I-am. Descartes wrote
obscurely about how the mind and the body are based on separate “substances.”
Today most scientists find Descartes’ discussion of the soul and
“mind” to be arcane and perhaps even silly. We all know
that the physical brain is where it’s at and most of us avoid
spiritualist theories. However, I am going to suggest an alternative
interpretation of Descartes’ theories. Descartes and I share an
intuition about how human cognition has two fundamentally different
modes. I am going to align Descartes’ idea of the body with the
eye-blink and searchfor- sex-and-food aspect of human and animal cognition
and I am going to align Descartes’ idea of the mind with the build-airplanes
and write-novels aspect of human language-based cognition. I’m
not going to cite Descartes as an authority but rather as a friend who
has been deeply misunderstood. The role of language in human cognition
is addressed in the second of the three books.
As you know, associations are mental connections between sensations,
ideas, or memories. Associations are also an important form of pattern
classification. You might think that pattern classification only refers
to visual forms we see in space or on a piece of paper, but I’d
like you to think otherwise.
Associations are based on our experiences
and much of our non– language-based knowledge of the world is
built upon them. The classic example of association is Pavlov’s
experiments with dogs. In Pavlov’s famous experiment, he rang
a bell when the dog was being fed. As time went by, the dog salivated
and drooled when it heard the bell. The dog had come to associate the
sound of the bell with the food the dog was about to be served.
Animals make similar associations
when something they eat makes them sick. The animal associates that
food with its sickness and will avoid it in the future. The animal has
no way of knowing for sure that eating that particular food was the
cause of his getting sick. If an experimenter induces sickness by means
of a drug that the animal is unaware of and timed with the eating of
the unfamiliar food then the animal will erroneously avoid that food
in the future. Some of us cringe when we see a flash of bright lightning
because we have learned to associate it with the loud thunder that usually
follows it. Associations are not
always helpful or pleasant. I have heard examples of people who have
received chemotherapy for the treatment of cancer. These patients have
sometimes reacted strongly, even involuntarily vomiting, when they approached
the building in which they received treatment or upon seeing a stranger
who they momentarily thought was their doctor.
think of pattern classification as operating in just one sensory domain.
We recognize the visual pattern of a set of stripes on an animal as
being those of a zebra or we recognize a sequence of acoustic inputs
as the sound of the dinner bell. Associations, however, often cross
We see and then eat the food; later we feel a pain in our stomach. We
see the lightning, then we hear the thunder. Learning the pattern, “lightning-thunder,”
is what we mean by associating the sense input from the lightning with
the sense input from the thunder. Associations often occur over time,
linking something that occurs in the present with something that is
likely to occur in the future. Associations provide a heads-up signal
that allows one to prepare for what is coming next. If I hear the lion’s
growl, I associate that with the big aggressive lions I have seen, and
I seek cover as if I had actually seen the lion.
We can think of associations as
classifying patterns that exist in two different sensory modalities
at two different times. I learn the pattern of the sound of the rattlers
on the snake and the later pain of being bitten by the snake. I therefore
think of associations as a kind of pattern classification.
Humans use associations to make
sense of the world. We learn that our typing gets better if we practice.
We notice that the brownies we make are better if we don’t bake
them too long. We often have explanations for why things work this way:
“Practice makes perfect,” or “Baking things too long
dries them out.” Such explanations often come after we have made
the associations. We use mnemonic devices to help us remember our associations
and to communicate them to other people. In other words, we often use
experience, not reason, to arrive at insight. Our gut reactions mirror
the primary cognitive processes of non–language-using animals.
The secondary or language aspect of our cognition often kicks in after
the fact. Of course, the experience that creates associations is not
our only teacher. We do have language based models and theories and
these are an entirely different aspect of human cognition. But for now
we limit our attention to the primary cognitive processes that we share
with many non-human creatures.
Another example of the importance
of associations is how coincidence plays such a strong role in our beliefs.
If I turn a light switch on at the same time my neighbor turns on her
radio, I am momentarily convinced that I caused her radio to turn on.
That belief is just there, I don’t have to ponder it. In fact,
I have to ponder it to dispel the belief. It is, after all, just an
unlikely coincidence. This same mechanism is what convinces people that
if a stranger was in town when a crime was committed then the stranger
must have committed the crime. Isn’t this part of where our religious
and racial prejudices come from? Many of our personal beliefs are based
on the perception of patterns that in fact are only partially suggestive.
Not too long ago before the rise of general education many people really
did believe that bad luck would follow if a black cat crossed their
path. Superstitions are often based on people’s observation of
patterns. However, it sometimes only takes two crossings of a black
cat and a person’s finding something bad in his life to make a
connection, especially when his friends hold the same belief.
I’d now like to consider
some mental behavior that may not at first seem to be based on the ability
to make associations. How much is four times five? Did you calculate
it or did you just know? Most of us memorized the multiplication table
in elementary school. But what goes on in your brain when you recall
the answer to a simple arithmetic question? If you are familiar with
simple computer procedures you might have said, “Well, maybe,
it’s a table lookup.” With a table lookup procedure you
do not compute the answer but find an answer that was previously computed.
A table is a way of representing, presenting, or storing information.
Your grade school multiplication table had rows and columns. The row
starting with a “4” and the column starting with a “5”,
intersected at a cell of the table with a “20” in it. A
table could also be a long list with an entry beside each item in the
list—like a telephone book. You list the names in alphabetical
order and place the person’s telephone number next to his or her
name. That is how computers store data. Each place on the list (person’s
name) corresponds to a storage register in the machine, and the related
item (telephone number) is the contents of the storage register. Such
a computational scheme is also a possible model of how the brain stores
However, there is little evidence
to support the table lookup model. The simplistic computer metaphor
breaks down here. Believing in that simple metaphor is tantamount to
expecting to find gears inside your
arm because that is how we would build a robotic arm. It’s easy
for a machinist to make gears out of metal, but it is very hard for
a biologic system to grow gears. More importantly, we just haven’t
found any gears in any body’s arms. Looking for storage registers
in the neural tissue of a brain is a much more difficult matter. We
can’t imagine how millions of registers would be organized, how
they would be addressed, or how their contents would be encoded. Furthermore,
we haven’t found any biological brain structures that are possible
candidates. Consequently, models of the human body, brain, or mind should
not be restricted to mechanisms or procedures that engineers use to
build physical devices. For now at least, let’s reject the idea
of a computerlike table lookup in the brain as a model for how we know
our multiplication tables.
As an alternative, I suggest that
we regard the process of retrieving an entry to the multiplication table
as a straightforward pattern-classifiPATTERN cation task. One’s
brain instantaneously recognizes the sequence “4 x 5” and
associates it with the number “20.” In fact, we can imagine
a cognitive process that recognizes the following arithmetic problems
as all belonging to the same class: 4 x 5, 5 x 4, 2 x 10, and 10 x 2.
When this class of number patterns is recognized, the process triggers
the associated “answer”: 20. Our brains handle much language-based
and non–language-based cognition in the same basic manner that
it handles the classification and association of patterns.
Another area we might not immediately
think of as involving pattern recognition is body motor control, such
as wiggling your ears or learning to shoot a basketball through a hoop.
In both cases, we tend to learn by trial and error. We try to repeat
a movement that appears to work. Part of this repetition involves recognition
of our own muscle and body movement and trying to duplicate the motion,
perhaps with a slight variation. Even a simple motion such as walking
down stairs involves keeping our balance by classifying inputs from
our body’s position sensors. Thus, body awareness is a form of
pattern classification. If you have ever tried to learn to walk down
stairs using crutches, you might recall how much attention and care
that took the first few times. In addition, watch a toddler during
his first month of toddling and you may gain an appreciation of the
complexity of the task of walking.
Pattern classification and association
of sensory experiences are fundamental processes involved in animal
and human cognition. This mode of cognition is quite distinct from high-level
cognition involving spoken or written language. To learn more about
pattern classification and association as models of human and animal
cognition we need to examine some details of how classification works.
The best way to do this is to study how an abstract machine might accomplish
the task—not because the brain would do it the same way but because
the nature of the task is the same and easier to understand with simple
Almost all creatures have the apparently
bland ability to recognize objects. Even ants with their tiny hundred-thousand
(105)-neuron brains apparently do it. (An ant’s brain is tiny
relative to a human’s brain. The human brain, with about 1012
neurons, is ten million times bigger than an ant’s brain.) Sensory
perception deals with how the eye and the visual cortex conspire to
make objects more salient in the visual field. While this is an interesting
topic, I wish to focus our attention on the higher mental processes
that work with the recognized objects
As we have discussed previously,
mental representation92 is key to understanding cognition and I propose
that object representation is the cornerstone that enables animal cognition.
Derek Bickerton in his great book, Language and Species, refers
to this ubiquitous mental capability as the primary representation system
(PRS). He writes:
It follows that what is presented
to any species, not excluding our own, by its senses is not ‘reality’
but a species-specific view of reality—not
‘what is out there,’ but what is useful for the species
to know about what is out there.93
Object Cognition Is a Big Deal
The insight that object cognition
is a big deal comes from engineers who try to make computer image processing
systems that can understand twodimensional images.94 We see trees, trucks,
people, and sky in a picture but the computer only sees splotches of
brown, green and blue. It has taken thousands of engineers, each working
tens of thousands of hours to devise partially successful computer algorithms
to make sense of common visual scenes. Your brain’s only input
about the visual world is through your eyes. The retina of your eye
has the same splotches of brown, green and blue that a simple photograph
or TV image has. We do not need to understand the details of the algorithms
that engineers use to process visual scenes or the neural processing
in the brain. We only need to note that the brain does this totally
amazing job of image interpretation effortlessly and fast. By the way,
not all biological creatures parse their images into objects. Very simple
creatures can be attracted or repelled by certain smells, sounds, or
light. In their simple environment, the specific sensation may be due
to some object but the simple sensory cue is sufficient to allow them
to make life-enhancing decisions.
Some creatures may have very fragmentary
representation of objects. For example, researchers have induced bees
to attempt to pollinate from a two-dimensional unstructured yellow disk
with a cup of nectar in its center. Apparently bees do see just a vaguely
compact shape of color and are not fussy about what sort of thing they
interpret as being a flower. Another example is the various contraptions
people place in their gardens to keep the birds from eating their seeds
and young plants. A shirt stuffed with straw, called, of course, a scarecrow,
often scares crows, and pie tins tied to poles move in the breeze and
create light and sound that ward off many birds. These last examples
illustrate that mentally representing objects admits a continuum of
specificity. I don’t know if the crow responds to the shape of
the scarecrow, its size, its array of colors, or to all of these features.
Let’s try to imagine a second
world that does not have objects on the scale easily observed by human
beings. Imagine an exploding galaxy where there are random formations
of colored and luminescent gas. In this imaginary world, objects, as
we normally think of them, do not exist. The visible patterns of gas
continually change just as fog and clouds in our 138 I Am Not a Machine
world continually change shape. Sometimes in our world, we see a cloud
whose shape resembles a whale or a human face but then we blink and
the cloud looks different. In such an imaginary objectless environment,
a device—be it a brain or a computer—designed to recognize
patterns and identify objects would be useless. The patterns would be
shifting and they would not be important to survival. A person in our
real world who is trying to predict the weather or fly an airplane through
clouds doesn’t care if the cloud looks like a whale or a face.
Shapes are only important for reasonably stable tangible objects.
Back on planet earth, we have things
like tigers that want to eat us and trout that we may want to eat. It
is very important that the animal brain recognizes and deals with these
objects. These physical objects in the real world are the source of
those splotches of color on our retina. Over millions of years, brains
have evolved structures for parsing retinal images, that is, breaking
them into parts that correspond to those external objects. We don’t
see splotches; we see trees and houses. By comparison, very simple biological
organisms, such as bacteria that respond to temperature and chemical
variations in their environment, do not have the ability to represent
Let’s push this idea a bit
further. Imagine walking through a jungle for the first time. You may
not recognize many objects. You may not even notice that the object
in front of you is not a tree limb but a snake poised to strike. There
are lots of strange sounds and smells as well. This is the sensory confusion
we can imagine a newborn infant experiencing. Eventually we learn to
associate a certain smell with the sight of a dead animal, and a certain
rattle and hiss with a deadly kind of snake. Our ability to parse scenes
also works for sounds and smells as well. Furthermore, and this is astounding,
a smell, sound, and image can become associated with a single object.
We do not say to ourselves after some intellectual exercise that a certain
rank smell, a barking sound, and the image of a shaggy four-legged medium-sized
creature must all be from the same source; no, we just know it’s
our dog that needs a bath and wants dinner. Our inputs from several
senses are combined and automatically attributed to a single source.
No big deal? Yes, big deal.
Engineers who have been pulling
their hair out trying to get machines to do this
signal combining effectively call the process signal fusion. Signal
fusion simply means seamlessly merging signal streams from different
sensors to produce a unified understanding of the source of the signals.
Signal fusion is really a kind of association process. We learn with
experience that the lightning we see is associated with the thunder
we hear a few seconds later. We attribute both sensory inputs to a single
electrical phenomenon, a single source. We internally represent the
source of both sensory signal streams with a single “virtual object.”
I am using the word virtual to convey the extended use of the word object.
Our brains have evolved to represent objects and they do it very well.
They do it particularly well in the natural world where they were designed
to work and they did it millions of years before they evolved to handle
that we have begun to understand the recognition of unique objects and
classes of objects, as well as the parsing and fusion of sensory signals,
let’s hypothesize that objects are represented as discrete entities,
that is, individually distinct items in our brains. To highlight this
idea of a virtual mental object I am now going to give it a name. I
call these cognitive objects cogjects and present them in the same spirit
that we have come to accept gravity and electromagnetic radiation as
useful constructs. We know how they work, but we don’t know what
they really are. In a sense that I will clarify later, the cogject is
more than a “percept” (an impression of an object obtained
by the senses) but is less than a “concept” (something conceived
in the mind or an abstract or generic idea generalized from particular
instances95). The intent of this
discussion is to show how cogjects work in various cognitive processes
and how they can be used to model cognition in general. By giving something
a name we can focus our attention on it and learn some of its attributes.
Think of an abstract concept such as inflation, which is now a household
word that we know a great deal about. Simply giving a name to something
does not in itself solve any problems but it can start a learning process
that leads to new ideas and solutions.
I Annotated Bibliography I For
Pet Owners I Thought You'd Never Ask I
Buy the Book...
Click on a vendor below to purchase
Barnes & Noble