textfiles/groups/CDC/cDc-0289.txt

                                    _
                                   | \
                                   |  \
                                   | | \
                            __     | |\ \             __
      _____________       _/_/     | | \ \          _/_/     _____________
     |  ___________     _/_/       | |  \ \       _/_/       ___________  |
     | |              _/_/_____    | |   > >    _/_/_____               | |
     | |             /________/    | |  / /    /________/               | |
     | |                           | | / /                              | |
     | |                           | |/ /                               | |
     | |                           | | /                                | |
     | |                           |  /                                 | |
     | |                           |_/                                  | |
     | |                                                                | |
     | |      c   o   m   m   u   n   i   c   a   t   i   o   n   s     | |
     | |________________________________________________________________| |
     |____________________________________________________________________|

  ...presents...     Can There Be Artificial Intelligence?
                                                         by Tequila Willy

                      >>> a cDc publication.......1994 <<<
                        -cDc- CULT OF THE DEAD COW -cDc-
  ____       _     ____       _       ____       _     ____       _       ____
 |____digital_media____digital_culture____digital_media____digital_culture____|

     Since the dawn of history, men have dreamed of other forms of intelligent
life.  There is something within the nature of mankind to reach out, to become
like gods.  Today, in our technologically advanced society, the potential is
right around the corner.  Even in this modern world of technological marvels,
there are many hurdles to overcome.  If the technological boundaries are
overcome, there are still those who believe that no man-made device will think
like a human being.  The doubts of the unbelievers should fade from the
memories of the human race as the first machines begin to think.

     Philosophers and scientists alike have been questing for artificial
intelligence.  It has only been in this century that the goal has been at least
feasible.  When looking for artificial intelligence, a researcher must look
inward before starting anything else.  The human mind and soul are two things
that we have very little knowledge of.  The way our brains work, and why we are
able to think are some of the most important things in our society.  When we
are born, we are self-aware, and it is a general belief that self-awareness is
only possible in humans because of the way we are born.  In fact, most people
never pay any attention to what happens during the gestation period.  It has
been shown that once the brain is developed, it immediately starts to process
information.  It cannot be proven in either direction that there is self-
awareness at that state.  It could turn out, in the end, that computers
designed for thought will have to go through a gestation process, and learn
just like a child.  The opposition to the theory of artificial intelligence,
and the arguments against AI have helped to move research ahead by pointing out
flaws in AI theories.

     There is a lot of technical material presented below, so there are some
terms that should be explained before continuing.  The first term is serial
computer, which is a computer that is distinguished by its capability to handle
only one operation at a time.  The second term is parallel processing, which is
a method of computing where more than one operation can be handled at a time
(Churchland and Churchland 35).  For example, give a task to two computers, one
parallel processing, and one serial processing; the serial computer attacks the
problem one step at a time, taking a large amount of time while the computer
that is capable of parallel processing breaks the task down into simpler
operations and then executes the task concurrently with other nodes of the
processor.  The result is that the parallel processor, being able to do many
things at a time, finishes the task in a fraction of the time.  The third term
that is used often is a computing architecture known as neural networks.  A
neural network is a system of processors, or nodes of a processor, linked to
other nodes in the way neurons in the brain are connected (35).  The way that
the neural network works is that the strength of the connections made from the
input of the network to the output of the network allows a more humanlike
ability to operate in a more than binary basis (35).  To clarify, a human
neuron is capable of firing its electrical charge at many different levels,
with each level signifying a different thing (36).  This allows a greater
variety in the amount and variety of information that the computer can pass
along.  The next term used in this paper is classical artificial intelligence
(or AI for short).  Classical artificial intelligence was the school of AI that
felt that given a powerful enough computer and the properly crafted programs,
you could get a machine that would be able to think (34).

     John R. Searle, in his essay from the _Scientific American_ from January
1990, writes that machines, no matter what the power, or their internal
architecture will not be able to think.  His main argument is what he calls his
Chinese room experiment (Searle 26).  The experiment goes like this: first you
lock some person in a room where there is a door with two mail slots, one in
and one out.  Into the room now and then come a pile of Chinese symbols in
through the slot.  The person inside the room also has a rule book that
explains, in a language he understands, what he should do with the symbols
coming into the room, and having used the rule book to manipulate the symbols,
he drops the rearranged symbols down through the out slot (26).  His point,
using this example, is that if he does not understand Chinese because of
running a computer program for understanding Chinese, then another computer
wouldn't either.  This means that simply manipulating symbols isn't enough to
create cognition, or thinking, therefore, according to him, making it
impossible for a computer to think (26).

     He then breaks down his arguments into axioms that he draws his
conclusions from.  The first axiom is this: "Computer programs are formal
(syntactic)" (27).  Syntactic means purely formal.  He explains the axiom
further by an example, "A computer processes information by first encoding it
into the symbolism that the computer uses and then manipulating the symbols
through a set of precisely stated rules.  These rules constitute the program"
(27).  Before introducing his second axiom he points out that symbols and
computer programs are abstract entities.  In computers the symbols can stand
for anything the programmer wants.  So, according to Searle, the program has
syntax, yet it doesn't have semantics.  This leads to his next axiom, which is:
"Human minds have mental contents (semantics)" (27).  His third axiom is this:
"Syntax by itself is neither constitutive of nor sufficient for minds" (27).
His explanation of that axiom is quite simple.  He says that merely
manipulating symbols is not enough to guarantee knowledge of what they mean.
Later in his paper he poses another axiom, "Brains cause minds" (29).  In other
words that thought is dependent on the biological processes of the human brain.

     The first conclusion that he draws from his axioms is "Programs are
neither constitutive of nor sufficient for minds" (27).  This conclusion is
pretty clear, saying that computers are incapable of having minds.  The second
conclusion is: "Any other system capable of causing minds would have to have
causal powers equivalent to those of brains" (29).  His example of the
conclusion states that for an electrical engine to drive a car as fast as a gas
engine the electrical engine must produce an energy output at least as high as
a gas engine (29).  His third conclusion is that "Any artifact that produced
mental phenomena, any artificial brain, would have to be able to duplicate the
specific causal powers of brains, and it could not do that by simply running a
program" (29).  The fourth conclusion that he draws from his axioms is this:
"The way that human brains actually produce mental phenomena cannot be solely
by virtue of running a computer program" (29).

     The argument presented by John M. Searle is quite formidable, with his
Chinese room example, and then the arguments that he goes on to present.  Some
of the conclusions and axioms, however, although they look sound at first, are
deceptively untrue.  An analysis of the arguments will show that they are
faulty.

     First, Searle's Chinese room example only applies to symbol-manipulating
computers.  In S-M machines the prospect of one ever being able to think is
highly doubtful, only because their architecture is incomparable to human brain
structure.  The human brain is the only thing we know to definitely possess
intelligence.  The problem with Searle's Chinese room example, at least in
reference to parallel processing and neural networked machines is that they
don't work the way that S-M machines work.  They use a method of processing
called vector processing (Churchland and Churchland 36).  The way that it works
is that when you send a combination of neural activations on one level of the
net, it will pass through the network on certain vectors caused by the
activation pattern and then output in another unique pattern (36).  This
process is much like the way that the human brain is believed to work.  This
type of processing is such that symbols are never manipulated in the fashion
that is presented in the Chinese room argument.  Symbol manipulation in a
vector-processing system may or may not be one of the cognitive skills that it
may display as a characteristic (36).  Therefore, the Chinese room is
non-applicable to the argument.  Searle argues against parallel processing by
presenting what he calls a Chinese gymnasium (Searle 28).  The gist of the
example is instead of the one man in the room, the room is full of men in a
parallel architecture.  He explains that none of them understands Chinese, and
the only thing accomplished by the extra men is that it would output faster,
without any comprehension (28).  The problem with this argument is that it is
unnecessary that the individual men need to know Chinese, as a single neuron
doesn't know any language either, but the whole thing probably does (Churchland
and Churchland 37).  For his Chinese gymnasium example to be fair there would
have to be the entire populations of 10,000 Earths in the gym (37).  There is
no way to prove there is no comprehension of Chinese in a network of that
magnitude.  Essentially what you would have in a room that size, with that many
people, is a gigantic, slow brain.  Mr. Searle argues against this view by
saying that it really doesn't matter, if nobody understands Chinese, neither
will the entire system (Searle 29).  The answer to that objection is that it is
possible, with the right architecture, to teach a computer Chinese.  If the
computer's structure was brainlike, the computer would be no different from a
Chinese child learning to communicate.

     Searle's arguments for not believing that computers are capable of human
thought are based on several simple axioms that he believes are true in all
types of computers.  The axioms he presents are sound.  All, except the last
one, which was, "Brains cause minds" (29).  In that axiom he declares that
minds are only capable of existing in brains, because brains are a biological
organ, with neurotransmitters, etc... (29).  This premise is not necessarily
true.  For example, in the Churchland article, they present an example of how
that axiom is not true.  Carver A. Mead, a researcher at the California
Institute of Technology, and his colleagues used analog VLSI (Very Large Scale
Integration) techniques to build an artificial retina (Churchland and
Churchland 37).  The machine is not a computer simulation of a retina, but an
actual real-time information processing unit that responds to light (37).  The
circuitry is based on the actual organ in a cat, and the output is incredibly
similar to the actual output of the cat's retina (37).  The process that is
used is completely without neurochemicals, so there really is no need for them,
hence the supposition that a mind can only exist in a brain is absurd.

     The conclusions that he draws from those axioms are not without flaws.
His first conclusion is that "Programs are neither constitutive nor sufficient
for minds" (Searle 29).  In a standard sense, it is probably the correct
conclusion, at least for the classical AI.  The new artificial intelligence,
however, is a merging of hardware and software in a synergistic relationship,
so programs will not solely handle the challenge of intelligence, but the
software will play a significant part in it.  If you look at the rest of his
conclusions, you will find that they are really only applicable to formal
programs alone, not software/hardware synergies, so they must be irrelevant to
the argument.  With his second conclusion, he essentially agrees that there is
a very real possibility of an artificial intelligence, as long as its causal
powers are at least that of the brain.  Modeling computers after the human
brain makes it probable that it can be done.

     It is improbable that there will be any thinking machines for many years.
The future holds many keys to this process.  It is necessary there be a greater
understanding of the mechanics of thought and memory before this end is
possible.  Classical artificial intelligence is obviously not going to work,
for the reasons stated earlier in the paper.  The answer obviously lies in the
realm of parallel processing and neural networks.  It has been proven that very
complicated and fast matrices of electronics can replicate biological
functioning, as in the example of the artificial retinas (Churchland and
Churchland 37).  Where the possibility lies is in the realm of combining the
processing abilities of complex computer architectures and the increasingly
sophisticated software needed to harness this power.

     We may find a solution within the psychology of childhood development.
When a child is born it is a blank slate.  In essence, they do not have any
real formed concepts, like those of syntax and semantics.  This is the way that
we should perceive a newly made computer of the kind that represents the human.
Everything must start from scratch, therefore it is necessary to teach the
computer as you would a baby.  This process is harder than teaching a newborn
child, since they are born with cognizance, but with time and knowledge of what
a computer needs to learn to become self-aware, it is possible.  There are
currently experiments going on where a doctor and an army of assistants are
building a base of language, and entering it, with referents to what they mean,
into the computer.  They are essentially teaching the computer manually what is
normally experienced by a child.  For example, a single word can have immense
amounts of referents, such as: what it is, what it can be compared to, and what
connotations are generally associated with them.  A word like "duck" for
example, could take weeks of compiling information, since you have to not only
put the concept of "duck" together, but also that of a bird, of colors, of
feathers, the basics of anatomy, and popular notions associated with the word
"duck."  With each layer of explanations you encounter you find a whole new
level of terms to define.  It is well-known that even the least intelligent
human being carries around a simply astonishing amount of information.  The
hardest things to define are on the simplest level of understanding, the
general hope of researchers is that with enough of the complex composite
concepts, the computer will be able to use the whole of its knowledge to puzzle
out the simple pieces.  This idea seems entirely logical, since it is something
that human beings try to every single day.  Humans are the same in that
respect, if we knew these simple truths, all philosophers and other scientists
would be simply unnecessary, as we would know all those things.  To date, the
scientists trying this experiment have succeeded in inputting almost all the
knowledge that an average 3 year old child has.  The strange thing is that in a
system like this, the computer seems to have a curious nature.  This would lead
one to think that the machine were cognizant, although in reality it most
probably is not the case.  The programs that compose this machine are simply
calling for more input to make it run more efficiently.  Although this is not
real thought yet one would suppose that this will be possible when the
computer's electronic architecture is sufficient to begin to change its own
programs.  That means that it would be working enough like a brain to revise
its beliefs, since beliefs are nothing less than knowledge in itself.

     The brain is a gigantic scale information processing machine, which is
simply a biological form of computer.  The implications of this call for a
rational person to assume if it is possible for a biological machine to think,
it would follow there would be a machine of a non-biological (ie. electronic)
nature that would be able to think, at least it would be if the electronic
brain was built to the equivalent of a human brain.

     Technology has increased exponentially in the last thirty years, but we
are still many years away from the first truly cognizant machines.  Because of
the arguments brought up, it is really impossible to prove there will be
cognizant machines, at least in a deductive sense.  In an inductive sense it
could be said there is a strong probability there will be a day when there will
be an intelligent machine.  It has been proven that the answer definitely does
not lie in the realm of computer programs in the manner of classical artificial
intelligence, since the computer architecture that is necessary for thought is
simply impossible in the traditional symbol-manipulating machine.  That part of
the argument is not in doubt, it is when you come into the hardware/software
synergy arena that the battle becomes heated.  Mr. Searle presents some very
strong arguments against the possibility, but these arguments are not
sufficient to destroy the possibility of computer thought.  In a case of
predicting the future there can be no definite proof, but if science and
technology can raise to the challenge of replicating the function of a human
brain, there will be, eventually, a computer that can think.


                                  Works Cited:

Churchland, Paul and Churchland, Patricia. "Could A Machine Think?"
     _Scientific American_ Jan. 1990: 32-37.

Searle, John M.  "Is A Brain's Mind a Computer Program?"  _Scientific American_
     Jan. 1990: 26-31.
 _______  __________________________________________________________________
/ _   _ \|Demon Roach Undrgrnd.806/794-4362|Kingdom of Shit.....806/794-1842|
 ((___)) |Cool Beans!..........415/648-PUNK|Polka AE {PW:KILL}..806/794-4362|
 [ x x ] |Metalland Southwest..713/579-2276|ATDT East...........617/350-STIF|
  \   /  |The Works............617/861-8976|Ripco ][............312/528-5020|
  (' ')  |            Save yourself!  Go outside!  DO SOMETHING!            |
   (U)   |==================================================================|
  .ooM   |Copyright (c) 1994 cDc communications and Tequila Willy.          |
\_______/|All Rights Reserved.                               11/01/1994-#289|

<EFBFBD><EFBFBD>