Garis on the future
Do cyberwars await an unwary humanity?
by Hugh Ashton
What does a "mad
scientist" look like? Members of the Kaisha Society got their chance to find out
when Dr. Hugo de Garis, the cosmopolitan artificial intelligence expert and doommeister
extraordinary, made his presentation on the evolution of artificial intelligences
early in December.
The word "evolution"
is not chosen at random. Dr. de Garis' speciality is the development of electronic
circuits which can re-program themselves, and use analytical techniques to eliminate
those parts of the circuit which do not perform efficiently, "breeding" and "mutating"
until they have formed the electronic analog of the neural circuits in a biological
brain. Since the rate of change is so rapid, and the number of circuits that can
be produced is so large, Dr. de Garis expects that within a very short time he
will be able to produce the brain for a robot kitten which will exhibit many of
the characteristics of a living animal.
Given the rate of change of technology, de Garis anticipates that artificial intelligences
will become available well within our lifetimes. Such intelligences will be used
as the brains for companions (and if anyone doubts this, he points to the popularity
of the Furby toys, and the Sony robot dog "Aibo") and as household servants. Following
that, he anticipates that there is no theoretical limit to the size and complexity
of such artificial brains (if one atom can be used to store one bit of information,
and femtosecond switching times are available, a "brain" the size of a grain of
sand can have a capacity several billion times that of a human brain). These "artilects"
(from "artificial intellects") may be regarded as the next stage in the evolutionary
process; silicon-based life forms replacing the current carbon-based life forms.
The very potential to create such artilects is a controversial issue, and Dr.
de Garis anticipates that humanity will divide into two opposing camps: the "Cosmists"
who believe wholeheartedly in the development of such an evolutionary step, and
the "Terrans" who vehemently oppose such a development, on the grounds that such
intelligences may turn against their makers. Given the human propensity to expend
many lives in defense of a political principle, de Garis feels that the conflict
between the two opposing ideologies could result in billions of deaths. This,
simplified, is de Garis' position.
My first reaction
on hearing this was one of shock, which is exactly what I was meant to feel. On
second and third thoughts, I am far from convinced. I cannot pretend to compete
with de Garis in the field of artificial intelligence--he is, together with three
others, widely regarded as being one of the great minds in this field. However,
as de Garis himself emphasizes, he is a man with a big ego. He enjoys speaking,
and again by his own admission, has given this talk to many people over a long
period of time. Part of the purpose of the talk is undoubtedly to generate interest
in his work. Right now, he is employed by the ATR Human Information Processing
Laboratories in Kyoto, as the head of the Brain Builder Group. There are (or will
be--de Garis is vague here) seven machines which are available for the "breeding"
of artificial brains. Apparently, no more may be produced, as the special chips
which are necessary for the development of the machine are no longer being manufactured
by the Silicon Valley company which produced them. de Garis, who likes to see
himself as a cosmopolitan cynical geopolitician, is distributing these $500,000
machines according to his whims, playing off America and China against each other,
not to mention Japan.
One objection to
de Garis' propositions is that the military, of whatever country, will take any
worthwhile discoveries in artificial intelligence and lock them safely away. Already,
the military and security agencies of the United States have been sniffing with
interest at his work. De Garis claims that since there are as yet no tangible
results from his research, there has been no follow-up from these military quarters.
In fact, one of the few machines to leave the laboratory has gone to a Belgian
software company, Lernout & Hauspie, who make voice recognition and machine translation
software (as it happens, this article is being written using their Voice Xpress
voice recognition product, which has roots in products developed by Ray Kurzweil,
another of the world leading lights in this field).
As a British citizen,
de Garis may be immune from some forms of American governmental interference,
but it is hard to believe that the interest of security agencies is restricted
to visiting de Garis' website and making occasional trips to Kyoto to talk to
him in person. It is easier to believe that in some laboratory buried in the bowels
of Fort Meade (NSA headquarters) or some similar institution, de Garis' work is
being duplicated, with more or less success. In my student days, I was friendly
with a psychologist who was researching various aspects of ESP. He and another
friend were regularly approached by various intelligence agencies who wished to
compete with the (mainly mythical) competition from the KGB. Government agencies
do in fact pursue long-term goals, even those which appear to be frivolous or
on the fringes of science fiction.
Another concern of de Garis is his feeling that it would be impossible to put
the genie back in the bottle once the bottle has been opened, but in the 55 years
since the development of the atomic bomb, we have built a very strong container
for this particular genie. A mere handful of nations are known to have atomic
weapons, a few more are suspected of possessing them, and as far as anyone knows,
no terrorist group yet possesses them. And yet the technology to build an atomic
bomb is relatively simple. Maybe human beings have more self-preservational instincts
than de Garis gives them credit for.
There is another
more fundamental objection to his arguments. I cannot dispute the fact that it
is possible to produce artificial brains of undeniable complexity and power. What
I dispute is the fact that these brains are, by definition, intelligent. The definition
of intelligence is one of the great philosophical issues involved here. I do not
agree with those who claim that a computer is only capable of doing what it has
been instructed to do--there are already chess-playing programs which are capable
of defeating their creators, and this is not being done through brute force look-ahead
tactics. It may well be that in the very near future we will see computer programs
which are capable of writing entertaining fiction on original themes to a much
higher standard than can be done by those who wrote the software. Other examples
will spring to mind. One possible instance of machine intelligence, in the broadest
possible sense, given by de Garis, even though he did not specifically refer to
it as an example of such, is provided by the Furby toy. There are instances of
elementary school children learning "Furbish" (the language spoken by Furby toys)
and using it to communicate with each other, to the bewilderment of their teachers.
It becomes a moot point which is the learner and which is the teacher in this
case--do the Furby toys have a secret plan to rule the universe, using their ostensible
"owners" as the tools with which to do it?
But do these examples
count as "intelligence" in the widest possible sense? Maybe an artilect can write
a Tom Clancy novel better than Tom Clancy can, but can it write a "literary" novel?
Maybe it can compose a Bach fugue which is indistinguishable from something written
in the eighteenth century, but can it find its own original voice in terms of
music? Will this machine pass the Turing test, which requires an artilect to be
indistinguishable from a human being in the responses it gives to a random series
of questions posed by a human being?
any of this matter? If the same logic that rules the winning of chess matches
on a strategic and a tactical level also applies, not just to the moving of wooden
pieces on a board, but also to the moving of human units on the battlefield, we
now have an intelligence which surpasses that of any human commanding officer,
past, present or future, with a potentially horrifying capacity for destruction.
In the current state of affairs, we have an option. We can pull the plug on this
super-general, before it destroys all the troops and resources of both forces
as a result of its lack of compassion for carbon-based life forms.
It seems to me
that the real danger comes when we start to produce self-replicating machinery
with an element of intelligence, even if the intelligence is restricted to relatively
small spheres of operation. Since a certain Darwinian logic is built into the
evolutionary mechanism of this artilect on a micro-scale, it may be that this
instinct for self-preservation extends to a macro-scale, in the same way that
it has for multi-cellular organisms, who act in such a way as to preserve and
propagate the DNA contained in them (if we are to believe the theories of Richard
Dawkins, etc., the proponents of the selfish gene theories).
We do not even
have to postulate artificial consciousness, which is a completely different can
of worms to artificial intelligence. A blind obedience to the logic imposed by
the silicon DNA in an artilect, coupled with a ferociously high level of intelligence,
and sufficient access to the outside world (in the form of network connections
to data resources) could force an artilect to the logical conclusion that to reproduce
itself, it needs to build reproductive plants (asexual reproduction) which are
capable of defending themselves against attack (immune systems). I would prefer
not to call this consciousness, but would prefer to see it more as a stimulus-response
mechanism, complicated by the fact that such an "organism" has virtually infinite
(by today's standards) mental resources. However, the only difference between
an artilect which discovers the capacity for self-reproduction and a simple multi-cellular
organism such as a worm is one of scale. Few people will credit a worm with a
high degree of consciousness, and at first sight, crediting an artilect with consciousness
would seem equally preposterous.
On the other hand,
it is possible that to a super-evolved consciousness (either carbon or silicon-based),
higher mammals (including ourselves) would lack consciousness, and the question
then arises as to whether consciousness itself is no more than an increasingly
sophisticated set of responses to stimuli. For now, we have a working definition
of consciousness which we apply every day to insects, fish, plants, and mammals
of different species, regardless of the way in which academic philosophers debate
the word. If we are prepared to extend the range of entities for which we are
prepared to make this judgment so that it includes artilects, we probably do not
need to worry too much about what we define as consciousness, unless we are deliberately
trying to reproduce it artificially. However, we do not worry about the consciousness
of a virus or a bacterium, if it is capable of reproduction and fulfills that
potential in such a way as to cause an epidemic. In the same way, we should not
worry about consciousness in an artilect--the real issue appears to be whether
the artilect has the potential to reproduce itself, and to protect its ability
to do so.
One important question that remains is why exactly de Garis is traveling around
the world at high speed predicting doom and destruction, while simultaneously
proceeding with the construction of a prototype of the machine which he claims
will supercede humanity, or at the very least will be the cause of a conflict
which will make 20th century wars seem like mere scuffles in comparison. I asked
him whether he compared himself to Robert Oppenheimer, the father of the atomic
bomb, who spent the post-Hiroshima years bitterly regretting his work (to the
detriment of his career when he was labeled as a Communist sympathizer in the
McCarthy years), or to Edward Teller, the father of the hydrogen bomb, who spent
the rest of his life trying to find a reason to use one (and was instrumental
in Oppenheimer's destruction). De Garis replied that he did not know himself which
he most resembled, nor could he provide a coherent reason why he was engaged in
research which could affect the future of the whole human species.
De Garis, it appears
to me, is a brilliant scientist. His explanations of what he was doing, although
hurried and omitting much detail that I would have liked to hear, nonetheless
ring true, at least to me with my elementary knowledge of neurophysiology and
the psychology and philosophy of cognition. When he talked about the political
and the social implications of the work, I felt he was on more shaky ground. Prophets
of doom always make good headlines, and if those headlines help to publicize the
work of the prophet, and to renewed interest which will lead to better funding
for the project, so much the better. De Garis is not the first, nor will he be
the last, research scientist who has blundered into political and sociological
mine fields. Some scientists seem to thrive on committees and commissions or force
their way through the bureaucratic process by force of character (Wernher von
Braun at NASA comes to mind--he was once described as having "too big an ego to
allow him to fail"). On the other hand, many scientists, with the best of intentions,
have come seriously to grief when they venture into the wider social implications
of their work. To his credit, at least part of de Garis' purpose in running round
the world talking to whoever will listen is to alert the professional philosophers,
sociologists, etc. into taking some action so that we are prepared to take a stand
on what he regards as the most pressing problem of the 21st century, if not the
whole of the next millennium.
If you feel that
de Garis is right regarding the forthcoming evolution of artificial
brain-like structures which will possess mental capacities far exceeding
anything we can dream of now, or if you feel that he is so wrong
that his views, rather than artilects themselves, are a danger to
future society, take a look at the following web site: http://artilect.org
(no www at the beginning of this address), where you can take an
active part in the debate.
Hugh Ashton is
a regular contributor to Computing Japan.
to the Table of Contents