Chapter 5: Kurzweil's Turing Fallacy
Reverse-engineering the human brain is doomed to failure because of the "Turing fallacy" -- a nonbiological computation system could never precisely copy the complex neural, structural, and chemical functions of a brain or achieve the required level of reliability, says Thomas Ray, who proposes evolution of "non-Turing" AIs as an alternative
Originally published in print June 18, 2002 in Are
We Spiritual Machines? Ray Kurzweil vs. the Critics of Strong AI
by the Discovery
Institute. Published on KurzweilAI.net on June 18, 2002.
There are numerous directions from which to criticize Kurzweil’s
proposal for strong AI. In this essay I will focus on his failure
to consider the unique nature of the digital medium when discussing
artificial intelligence. But before elaborating on this point, I
would like briefly to call attention to some other issues.
Kurzweil’s interpretation of quantum mechanics leads him to
the conclusion that “consciousness, matter, and energy are
inextricably linked.” While this is true in the sense that
consciousness arises from the interactions of matter and energy,
it is not true in the sense that Kurzweil intends it: that quantum
ambiguities are not resolved until they are forced to do so by a
conscious observer.
Kurzweil’s error is most glaringly apparent in his description
of the paper output from a quantum computer: “So the page with
the answer is ambiguous, undetermined—until and unless a conscious
entity looks at it. Then instantly all the ambiguity is retroactively
resolved, and the answer is there on the page. The implication is
that the answer is not there until we look at it.” He makes
the same error in describing the evolution of the universe: “From
one perspective of quantum mechanics—we could say that any
Universe that fails to evolve conscious life to apprehend its existence
never existed in the first place.”
Kurzweil does not understand that it is the act of measurement
that causes the collapse of the wave function, not conscious observation
of the measurement. In practice, the collapse is (probably always)
caused by a completely unconscious measuring device. Printing of
the result on a paper could be such a measuring device. Subsequent
conscious observation of the measurement is irrelevant.
This psychic quantum mechanics did not originate with Kurzweil.
It has been around for decades, apparently as a way to deal with
Schrödinger’s cat. Thus, Kurzweil may be able to point
to physicists who hold this view. Similarly, I could point to biologists
who believe in the biblical story of creation rather than evolution.
The existence of experts who believe a doctrine, however, is no
argument for the truth of the doctrine.
Colloquial Chaos
Kurzweil’s suggestion that in a process, the time interval
between salient events expands or contracts along with the amount
of chaos (“the law of time and chaos”), is quite interesting.
Yet, the definitions of “salient events” and “chaos”
are quite subjective, making the “law” difficult to support.
Technically, it would probably be more appropriate to use the word
“entropy” in place of “chaos,” but for consistency,
I will also use “chaos” in this discussion.
Most striking is the apparently inconsistent use of chaos. He states
that in an evolutionary process order increases, and he says: “Evolution
draws upon the chaos in the larger system in which it takes place
for its options for diversity.” Yet he states that in the development
of an individual organism chaos increases, and he says: “The
development of an organism from conception as a single cell through
maturation is a process moving toward greater diversity and thus
greater disorder.” Kurzweil suggests that in evolution, diversity
implies order, while in development, diversity implies disorder.
Through evolution, the diversity of species on Earth has increased,
and through development, the diversity of cell types increases.
I would characterize both as processes that generate order. Why
does Kurzweil think that development generates chaos? His apparent
reason is to make his law of time and chaos consistent with our
perception of time: Our subjective unit of time grows with our age.
I believe that the scientific community would generally agree that
the developmental process up to the period of reproduction is a
process of increasing order. In humans, who live well beyond their
reproductive years, the condition of the body begins to deteriorate
after the reproductive years, and this senescence would generally
be considered a process of increasing chaos.
In an effort to fit development seamlessly into his law of time
and chaos, Kurzweil presents the whole life cycle from conception
to death, as unidirectional, towards increasing chaos. This position
is indefensible. The developmental process directly contradicts
the law of time and chaos. Development is a process in which the
time between salient events increases with order.
He attempts to be clear and concrete in his use of the term chaos:
“If we’re dealing with the process of evolution of life-forms,
then chaos represents the unpredictable events encountered by organisms,
and the random mutations that are introduced in the genetic code.”
He explains: “Evolution draws upon the great chaos in its midst—the
ever increasing entropy governed by the flip side of the Law of
Time and Chaos—for its options for innovation.” This implies
that unpredictable events and mutations are becoming more frequent,
a position that would be difficult to defend. His argument is that
increasing rates of mutations and unpredictable events are, in part,
driving the increasing frequency of “salient events” in
evolution. He does not provide any support for this highly questionable
argument.
Despite his attempt to be precise, his use of “chaos”
is vernacular: “When the entire Universe was just a ‘naked’
singularity . . . there was no chaos.” “As the Universe
grew in size, chaos increased exponentially.” “Now with
billions of galaxies sprawled out over trillions of light-years
of space, the Universe contains vast reaches of chaos . . .”
“We start out as a single fertilized cell, so there’s
only rather limited chaos there. Ending up with trillions of cells,
chaos greatly expands.” It seems that he associates chaos with
size, a very unconventional use of the term.
His completely false interpretation of quantum mechanics, his vague
and inconsistent use of terms such as “chaos” and “salient
events,” and his failure to understand the thermodynamics of
development represent errors in the basic science from which he
constructs his view of the world. These misunderstandings of basic
science seriously undermine the credibility of his arguments.
I am not comfortable with the equation of technological development
and evolution. I think that most evolutionary biologists would consider
these to be quite separate processes, yet, their equation represents
a point of view consistent with Kurzweil’s arguments and also
consistent with the concept of “meme” developed by the
evolutionary biologist Richard Dawkins.
The primary criticism that I wish to make of Kurzweil’s book,
however, is that he proposes to create intelligent machines by copying
human brains into computers. We might call this the Turing Fallacy.
The Turing Test suggests that we can know that machines have become
intelligent when we cannot distinguish them from human, in free
conversation over a teletype. The Turing Test is one of the biggest
red-herrings in science.
It reminds me of early cinema when we set a camera in front of
a stage and filmed a play. Because the cinema medium was new, we
really didn’t understand what it is and what we can do with
it. At that point we completely misunderstood the nature of the
medium of cinema. We are in almost the same position today with
respect to the digital medium.
Over and over again, in a variety of ways, we are shaping cyberspace
in the form of the 3D material space that we inhabit. But cyberspace
is not a material space and it is not inherently 3D. The idea of
downloading the human mind into a computer is yet another example
of failing to understand and work with the properties of the medium.
Let me give some other examples and then come back to this.
I have heard it said that cyberspace is a place for the mind, yet
we feel compelled to take our bodies with us. 3D virtual worlds
and avatars are manifestations of this. I have seen virtual worlds
where you walk down streets lined by buildings. In one I saw a Tower
Records store, whose front looked like the real thing. You approached
the door, opened it, entered, and saw rows of CDs on racks and an
escalator to take you to the next floor. Just Like The Real Thing!
I saw a demo of Alpha World, built by hundreds of thousands of
mostly teenagers. It was the day after Princess Diana died, and
there were many memorials to her, bouquets of flowers by fountains,
photos of Diana with messages. It looked Just Like The Real memorials
to Diana.
I wondered, why do these worlds look and function as much as possible
like the real thing? This is cyberspace, where we can do anything.
We can move from point A to point B instantly without passing through
the space in between. So why are we forcing ourselves to walk down
streets and halls and to open doors?
Cyberspace is not a 3D Euclidean space. It is not a material world.
We are not constrained by the same laws of physics, unless we impose
them upon ourselves. We need to liberate our minds from what we
are familiar with before we can use the full potential of cyberspace.
Why should we compute collision avoidance for avatars in virtual
worlds when we have the alternative to find out how many avatars
can dance on the head of a pin?
The WWW is a good counter-example, because it recognizes that in
cyberspace it doesn’t matter where something is physically
located. Amazon.com is a good alternative to the mindlessly familiar
3D Tower Record store.
Let me come back to Kurzweil’s ideas on AI. Kurzweil states
that it is “ultimately feasible” to:
. . . scan someone’s brain to map the locations,
interconnections, and contents of the somas, axons, dendrites, presynaptic
vesicles, and other neural components. Its entire organization could
then be re-created on a neural computer of sufficient capacity,
including the contents of its memory . . . we need only to literally
copy it, connection by connection, synapse by synapse, neurotransmitter
by neurotransmitter.
This passage most clearly illustrates Kurzweil’s version of
the Turing Fallacy. It is not only infeasible to “copy”
a complex organic organ into silicon without losing its function,
but it is the least imaginative approach to creating an AI. How
do we copy a seratonin molecule or a presynaptic vesicle into silicon?
This passage of the book does not explicitly state whether he is
proposing a software simulation from the molecular level up, of
a copy of the brain, or if he is proposing the construction of actual
silicon neurons, vesicles, neurotransmitters, and their wiring together
into an exact copy of a particular brain. Yet in the context of
the preceding discussion, it appears that he is proposing the latter.
Such a proposal is doomed to failure. It would be a fantastic task
to map the entire physical, chemical, and dynamic structure of a
brain. Even if this could be accomplished, there would be no method
for building a copy. There is no known technology for building complexly
differentiated microscopic structures on such a large scale. If
a re-construction method existed, we might expect that a copy made
of the same materials, carbon chemistry, if somehow jump-started
into the proper dynamic activity, would have the same function (though
such a copied brain would require a body to support it). But a copy
made of metallic materials could not possibly have the same function.
It would be a fantastically complex and intricate dynamic sculpture,
whose function would bear no relation to a human brain. And what
of the body and its essential sensory integration with the brain?
In order for the metallic “copy” to have the same function,
we would have to abstract the functional properties out of the organic
neural elements, and find structures and processes in the new metallic
medium that provide identical functions. This abstraction and functional-structural
translation from the organic into the metallic medium would require
a deep understanding of the natural neural processes, combined with
the invention of many computing devices and processes which do not
yet exist.
However, Kurzweil has stated that one advantage of the brain-copy
approach is that “we don’t need to understand all of it;
we need only to literally copy it.” Yet he is ambivalent on
this critical point, adding: “To do this right, we do need
to understand what the salient information-processing mechanisms
are. Much of a neuron’s elaborate structure exists to support
its own structural integrity and life processes and does not directly
contribute to its handling of information.”
The structure and function of the brain or its components cannot
be separated. The circulatory system provides life support for the
brain, but it also delivers hormones that are an integral part of
the chemical information processing function of the brain. The membrane
of a neuron is a structural feature defining the limits and integrity
of a neuron, but it is also the surface along which depolarization
propagates signals. The structural and life-support functions cannot
be separated from the handling of information.
The brain is a chemical organ, with a broad spectrum of chemical
communication mechanisms ranging from microscopic packets of neurotransmitters
precisely delivered at target synapses, to nitrogen oxide gas and
hormones spread through the circulatory system or diffusing through
the intercellular medium of the brain. There also exist a wide range
of chemical communications systems with intermediate degrees of
specificity of delivery. The brain has evolved its exquisitely subtle
and complex functionality based on the properties of these chemical
systems. A metallic computation system operates on fundamentally
different dynamic properties and could never precisely and exactly
“copy” the function of a brain.
The materials of which computers are constructed have fundamentally
different physical, chemical, and electrical properties than the
materials from which the brain is constructed. It is impossible
to create a “copy” of an organic brain out of the materials
of computation. This applies not only to the proposition of copying
an individual human brain with such accuracy as to replicate a human
mind along with its memories, but also to the somewhat less extreme
proposition of creating an artificial intelligence by reverse engineering
the human brain.
Structures and processes suitable for information processing in
the organic medium are fundamentally different from those of the
metallic computational medium. Intelligent information processing
in the computational medium must be based on fundamentally different
structures and processes, and thus cannot be copied from organic
brains.
I see three separate processes which are sometimes confounded.
Machines having:
1) computing power equal to the level of human intelligence
2) computing performance equal to the level of human intelligence
3) computing like human intelligence
A large portion of Kurzweil’s book establishes the first process
by extrapolating Moore’s Law into the future until individual
machines can perform the same number of computations per second
as is estimated for the human brain (~2020 A.D.).
I accept that this level of computing power is likely to be reached,
someday. But no amount of raw computer power will be intelligent
in the relevant sense unless it is properly organized. This is a
software problem, not a hardware problem. The organizational complexity
of software does not march forward according to Moore’s Law.
While I can accept that computing power will inevitably reach human
levels, I am not confident that computing performance will certainly
follow. The exponential increase of computing power is driven by
higher densities and greater numbers of components on chips, not
by exponentially more complex chip designs.
The most complex of artifacts designed and built by humans are
much less complex that living organisms. Yet the most complex of
our creations are showing alarming failure rates. Orbiting satellites
and telescopes, space shuttles, interplanetary probes, the Pentium
chip, computer operating systems, all seem to be pushing the limits
of what we can effectively design and build through conventional
approaches.
It is not certain that our most complex artifacts will be able
to increase in complexity by an additional one, two or more orders
of magnitude, in pace with computing power. Our most complex software
(operating systems and telecommunications control systems) already
contains tens of millions of lines of code. At present it seems
unlikely that we can produce and manage software with hundreds of
millions or billions of lines of code. In fact there is no evidence
that we will ever be able to design and build intelligent software.
This leads to the next distinction, which is central to my argument,
and requires some explanation:
2) computing performance equal to the level of human intelligence
3) computing like human intelligence
A machine might exhibit an intelligence identical to and indistinguishable
from humans, a Turing AI, or a machine might exhibit a fundamentally
different kind of intelligence, like some science fiction alien
intelligence. I expect that intelligences which emerge from the
digital and organic media will be as different as their respective
media, even if they have comparable computing performance.
Everything we know about life is based on one example of life,
namely, life on earth. Everything we know about intelligence is
based on one example of intelligence, namely, human intelligence.
This limited experience burdens us with preconceptions and limits
our imaginations.
Consider this thought experiment:
We are all robots. Our bodies are made of metal and our brains
of silicon chips. We have no experience or knowledge of carbon-based
life, not even in our science fiction. Now one of us robots comes
to an AI discussion with a flask of methane, ammonia, hydrogen,
water, and some dissolved minerals. The robot asks: “Do you
suppose we could build a computer from this stuff?”
The engineers among us might propose nano-molecular devices with
fullerene switches, or even DNA-like computers. But I am sure they
would never think of neurons. Neurons are astronomically large structures
compared to the molecules we are starting with.
Faced with the raw medium of carbon chemistry, and no knowledge
of organic life, we would never think of brains built of neurons,
supported by circulatory and digestive systems, in bodies with limbs
for mobility, bodies which can only exist in the context of the
ecological community that feeds them.
We are in a similar position today as we face the raw medium of
digital computation and communications. The preconceptions and limited
imagination deriving from our organic-only experience of life and
intelligence make it difficult for us to understand the nature of
this new medium, and the forms of life and intelligence that might
inhabit it.
How can we go beyond our conceptual limits, find the natural form
of intelligent processes in the digital medium, and work with the
medium to bring it to its full capacity, rather than just imposing
the world we know upon it by forcing it to run a simulation of our
physics, chemistry, and biology?
In the carbon medium it was evolution that explored the possibilities
inherent in the medium, and created the human mind. Evolution listens
to the technology that it is embedded in. It has the advantage of
being mindless, and therefore devoid of preconceptions, and not
limited by imagination.
I propose the creation of a digital nature. A system of wildlife
reserves in cyberspace, in the interstices between human colonizations,
feeding off of unused CPU-cycles (and permitted a share of our bandwidth).
This would be a place where evolution can spontaneously generate
complex information processes, free of the demands of human engineers
and market analysts telling it what the target applications are.
Digital naturalists can then explore this cyber-nature in search
of applications for the products of digital evolution in the same
way that our ancestors found applications among the products of
organic nature such as: rice, wheat, corn, chickens, cows, pharmaceuticals,
silk, mahogany. But, of course, the applications that we might find
in the living digital world would not be material; they would be
information processes.
It is possible that out of this digital nature there might emerge
a digital intelligence, truly rooted in the nature of the medium,
rather than brutishly copied and downloaded from organic nature.
It would be a fundamentally alien intelligence, but one which would
complement rather than duplicate our talents and abilities.
I think it would be fair to say that the main point of Kurzweil’s
book is that artificial entities with intelligence equal to and
greater than humans will inevitably arise, in the near future. While
his detailed explanation of how this might happen focuses on what
I consider to be the Turing Fallacy, that is, that it will initially
take a human form, Kurzweil would probably be content with any route
to these higher intelligences, Turing or non-Turing.
While I feel that AIs must certainly be non-Turing—unlike
human intelligences—I feel ambivalent about whether they will
emerge at all. It is not the certainty that Kurzweil paints, like
the inexorable march of Moore’s Law. Raw computing power is
not intelligence. Our ability ever to create information processes
of a complexity comparable to the human mind is completely unproven
and absolutely uncertain.
I have suggested evolution as an alternate approach to producing
intelligent information processes. These evolved AIs would certainly
be non-Turing AIs. Yet evolution in the digital medium remains a
process with a very limited record of accomplishments. We have been
able to establish active evolutionary processes, by both natural
and artificial selection in the digital medium. But the evolving
entities have always contained at most several thousand bits of
genetic information.
We do not yet have a measure on the potential of evolution in this
medium. If we were to realize a potential within several orders
of magnitude of that of organic evolution, it would be a spectacular
success. But if the potential of digital evolution falls ten orders
of magnitude below organic evolution, then digital evolution will
lose its luster. There is as yet no evidence to suggest which outcome
is more likely.
The hope for evolution as a route to AI is not only that it would
produce an intelligence rooted in and natural to the medium, but
that evolution in the digital medium is capable of generating levels
of complexity comparable to what it has produced in the organic
medium. Evolution is the only process that is proven to be able
to generate such levels of complexity. That proof, however, is in
the organic rather than the digital medium. Like an artist who can
express his creativity in oil paint but not stone sculpture, evolution
may be capable of magnificent creations in the organic medium but
not the digital.
Yet the vision of the digital evolution of vast complexity is still
out there, waiting for realization or disproof. It should encourage
us, although we are at the most rudimentary level of our experience
with evolution in the digital medium. Nevertheless, the possibilities
are great enough to merit a serious and sustained effort.
Copyright ' 2002 by the Discovery
Institute. Used with permission.
| 
