Le Monde



LE MONDE / 09 November 1999 / Page 16 (English literal translation from the
(excellent) French newspaper article)

HORIZONS - INTERVIEWS 2000
DEBATES FOR THE CENTURY TO COME

Hugo de Garis, researcher in artificial intelligence


" The XXIst century will be dominated by massively intelligent machines "

This British scientist, who works in the Japanese laboratory ATR, predicts that
gigantic and no doubt immortal "artificial intellects", that he calls
"artilects' risk one day to threaten the human species.


You are building an artificial brain. What is it about ?
- This work is made possible by progress in electronics, described by Moore's
law. Moore is the cofounder of the company Intel, a large manufacturer of
microprocessors. The power of these microprocessors he predicted in 1965 will
double every 18 months. Since then, this law has remained valid and it should
remain valid until the elements of electronic circuits reach the size of
chemical molecules. One will thus attain enormous capacities, which will permit
the placement of a large memory volume into an artificial brain.

" Furthermore, a new technology called "Evolutionary Engineering", allows the
solution of that other large problem which presents itself when building a
brain, namely its complexity. The trick is to build processors capable of
evolving, to adapt themselves to new functionality. Evolutionary engineering is
based on processors developed several years ago, called Field Programmable Gate
Arrays (FPGAs). These are programmable chips, in which a software instruction
can modify a physical circuit.

- How are they integrated into an artificial brain ?
- The idea is to adapt an evolutionary programming technique, that one calls
genetic algorithms, to these circuits. For each task, other circuits measure the
performance of these circuits, and according to their success, send instructions
which allow the evolving circuits to modify themselves to improve their
performance. These techniques allow me to construct a machine which in its
initial version, will include 64000 circuits, each one containing the equivalent
of 1000 artificial cells similar to biological neurons. In total, that
represents roughly 75 million neurons. In 3 years we should reach 1 billion.
Thats still very far from the capacity of a human brain, which contains some 100
billion neurons. But one should already be thinking about the consequences in 50
or 100 years of this evolution in technology.

- Just how far could it go ?
- I envisage what I call "artilects" - artificial intellects -, massively
intelligent machines, which will become far more intelligent than humans. We
will enter the realm of nanotechnology, which operates at the scale of the
nanometer (a millionth of a millimeter). One will be able to put one bit - one
unit of information - onto one atom. If one calculates the number of atom or
molecules in a large asteroid, the answer is 10 to power 40. These numbers are
astronomic. And with these new technologies of the XXIst century, the bits could
change state - a zero becomes a one, a one becomes a zero-, at a speed of 10 to
power 15 times per second. In a large asteroid of 10 to power 40 atoms, one
could therefore have a change of general state at a speed of 10 to power 55, 40
plus 15. Thats much more than the estimated computational capacity of the human
brain, which is of the order of 10 to power 16 state changes per second. The
"writing is on the wall". Its only a question of time.

- It is difficult to not treat your ideas as science fiction.
- Leo Szilard was a nuclear physicist. In 1933, he invented the concept of a
nuclear chain reaction, and from that, the idea of a nuclear bomb. He calculated
the energy that could be derived from this new technology, and concluded that
one bomb could destroy a city. At the time, this idea seemed totally crazy...
Today, there are at least 3 respected researchers who are thinking the same
thing as I am, Hans Moravec and Ray Kurzweil in the US, and Kevin Warwick in the
UK. The idea is beginning to be taken more and more seriously. One needs to be
aware that massively intelligent machines will be appearing in about 50 years.

- What will be the consequences of their emergence ?
- One cannot guess what their preoccupations will be, their interests. But the
dominant question of global politics in the XXIst century, will be, "Humanity,
should it construct "artilects" or not"? I imagaine that the reply will generate
two passionately and violently opposed ideological visions : those for whom
constructing artilects represents the destiny of the human species, who will
have a cosmic vision - I call them "Cosmists"; and another group - I call them
"Terrans" - who will fear that the artilects will decide one day that the human
species is a pest that should be destroyed. With their superior intelligence,
that will be easy for them.

- How will this conflict present itself ?
- In the choice to develop the artilects, the stake is enormous : namely the
destruction of the human species. This stake is so high, that to avoid it, the
"Terrans" will go to any lengths, including killing the Cosmists. The only way
to avoid the risk, is that the artilects are never built. For the Cosmists,
creating the artilects will be the goal to develop a future species with
incredible capacities. They will perceive the artilects as magnificent
creatures, transcendent, and no doubt immortal. Their existence will pose the
question of the destiny of the human species, the destiny of the universe in a
certain sense.

- Can one limit the complexity level of this development?
- To choose to stop an evolutionary development which has been going on for
billions of years? For the Cosmists, stopping that would be as damnable as if
the development of life on Earth, starting with bacteria 3 billion years ago,
had been blocked.

- But the "Cosmists" will include the cultures which are the most advanced in
computers, that is the US, Europe, Japan ?
- Yes, quite. They will recruit especially from amongst the most powerful and
richest people on the planet. In 20 years, in the rich countries, each household
will have its domestic robot, its eductional robots, friendship robots, which
adapt to the personality of the individual. Every year, new, more performant
models will be produced, and millions, billions of people will be able to see
for themselves the rise of artificial intelligence. Everyone will be asking the
question "where does it stop"?

- How will these "artilects" assert their dominance over the human species ?
- I dont know, and, in a manner of speaking, I dont care. The important thing is
that perhaps they can do it. If the acceleration of their intelligence occurs
very rapidly, lets say in two or three years, there will not be enough time for
a political reaction by the human species. Starting from a certain threshold,
the machines will be able to construct other computers which are more complex,
and those in their turn ... and vroooom ! there will be an explosion, a kind of
chain reaction of intelligence. But more probably, considering the difficulty of
creating an artificial brain with the same level as that of a human being, much
time will be necessary, at least 50 years. Thats why I think that the question
will dominate the XXIst century.

- But the wars of our era, from Rwanda to Kosovo, are based on ethnic
differences. Your approach, isnt it far removed from reality ?
- These are little wars. They have nothing in common with the conflict between
the two superpowers, the US and the USSR, whose opposition dominated the second
half of the XXth century. Between the moment when human beings begin to notice
that the primitive artilects become intelligent, and the moment when they become
truly intelligent, there will be about 15 years, a period sufficiently long for
the conflict between the Cosmists and the Terrans to develop in the political
arena. But there is another factor in the Cosmist ideology that will render the
progression of artificial intelligence quasi impossible to stop : the fact that
telecommunications and computing are already the largest industries in the
world. Their momentum are extremely large. Another element supporting the
development of artilects is military : one can predict in the XXIst century a
rivalry between China and the US, that in 20 or 30 years, China will become a
superpower. For the American Minister of Defence, it will be vital to create, to
build, military artilects in order to have a technology superior to that of the
Chinese.

- Bionics, that is, the integration of electronics into the human body, couldnt
it raise humans to the level of "artilects" ?
- Many people think that there will be this 3rd category, the "Cyborgs", i.e.
human beings incorporating components to become hybrids between biological and
mechanical creatures. But for me, an artilect and a cyborg, are almost the same
thing, because with this approach, the humans can acquire, in the same manner as
the artilects, superhuman capacities. From this fact, their interests will be
those of artilects.

- But they will keep their flesh. And that strange thing, that one often calls
the soul and which creates the personality ?
- The human aspect will be completely eclipsed by the artilectual capacities.

- Imagine that one could, in 30 or 40 years, put chips into the brain, would you
want to become a "cyborg", an "artilect" ?
- Probably. But it would no longer be me. What is the identity of a human being
? Once one inserts the "gadget" one becomes something else. In a manner of
speaking, it is the death of that human being, because he would be so swamped by
the capacities of the "artilect" in his head, that he would no longer be a human
being. Perhaps they will discover new phenomena more complex, more magnificent
than consciousness, that I cannot even imagine, because I am too stupid. I am
only a human being. "


INTERVIEW CONDUCTED BY HERVE KEMPF