Robot Books.com would like to thank Hans Moravec for taking the time to answer our
questions about his work. Mr. Moravec gave us his responses to the following
questions on November 28, 1998.
Interview with Hans Moravec
In your 1988 book Mind
Children, you predicted that robot intelligence would reach the level of human
intelligence in forty years. Now, ten years later, you predict that human-level
intelligence in robots is still forty years away. Is this going to be like "Fusion
Power" which always seems to be 30 years in the future?
Patience. Fusion now seems only 10 years away! Since 1995 conditions in inertial
confinement Z-pinch machines
have risen from 1% to over 30% of the requirements for net power output. A surprising
major reason is exponential computer progress for modeling the fusion plasmas. Magnetic
confinement fusion has also made recent strides for similar reasons. Don't abandon your
chickens before they hatch!
The story for intelligence is even clearer. Small-scale controlled fusion is not observed
in nature, so for all we know it might be astronomically difficult or impossible. Each of
us, however, is direct evidence that intelligence can arise in a few Kilos of matter
powered by a few Calories of energy, in about a billion small evolutionary steps.
Technological evolution is roughly retracing biology's journey. Fifty years ago our most
complex machines were behaviorally simpler than bacteria. Since then some machine behavior
has passed through worm-like complexity to approach the lower rungs of the vertebrates, a
trail that took biology over 500 million years to blaze. Technology is racing along it 10
million times faster. The path continues through reptiles, mammals, primates and humans,
which stretch took another few hundred million biological years to build, and should take
a few technological decades to retrace. I'm doing my best to mark and calibrate this
trail. We've moved quite far along it in the ten years since "Mind Children,"
and I've used the new data to recalibrate a little. The estimate of the computer power
needed is, in "Robot," at the upper bound of the range in "Mind
Children." The consequent increase in the expected time of arrival of humanlike
robots is more than canceled by a recent acceleration in the pace of computer power
growth. Also, I lowered the threshold price of a commercially viable robot controller from
$10,000 to $1,000. Despite this, if you contrast the curves on page 64 of "Mind
Children" and page 60 of "Robot" you will note the arrival time estimate
for sufficient computer power for practical human-level AI has actually come closer, from
2030 in "Mind Children" to about 2025 in "Robot." I've become
conservative in my old age, however, and I trace a rather leisurely development for the
four stages of universal robot, giving ten years for each step from lizardlike to
mouselike to monkeylike to humanlike. As I note in the Preface, though, that pace probably
overestimates the time for the latter steps. In the evolutionary record, each happened
about twice as fast as the previous one. So probably, left to me, it would take another 40
years to make humanlike robots. But vigorous young developers may get there in less than
30 years. We'll see. Either way we're close. I plan to recalibrate again in another ten
Is it possible that there is an essential "something" about human
intelligence that computers will simply not be able to capture, no matter how powerful
I have no reason to suspect that. The best evidence is that our brains were fashioned by
the trial and error process of evolution, exploring a small range of electrochemical
signaling options derived from the earliest cells, and a larger range of possible nervous
system interconnections. I know of no credible hint that the essential behavior of any
possible neural arrangement cannot be emulated by sufficiently powerful universal
computers with the right input and output equipment running the right programs. Of course,
specialized circuitry, like the digital video and audio processors found in most PCs or
the chess chips in Deep Blue, may allow particular performances to be achieved faster or
You seem quite comfortable with your prediction that robots will displace humans from
essential roles, and eventually they could displace us from existence. Aren't we talking
about the end of the world here? If the world goes on and nobody is here to see it, won't
we have failed as a race?
"Failed as a race" implies some external criterion for success. Who knew there
was going to be a test!? Actually, I don't even consider the majority of organisms that
became extinct without leaving any offspring to have "failed as a race." They
also serve who only live and die, and thus map out the unviable portions of the
evolutionary landscape, and provide a background for the evolution of others. I think the
evolution of life is the most interesting thing around, and want to see it continue with
utmost vigor. Producing artificial offspring that transcend us to the max is the most
exalted role I can imagine the human race playing, and would be the grandest success
possible. On the other hand, if we chose to stagnate, indefinitely circumscribed by our
present limitations, then by my standards we would be tragic failures, having turned our
back on our potential.
You call these future machines our progeny, our "mind children". When all
traces of biology are gone from the beings that will eventually dominate the Earth, won't
we, for all intents and purposes, be extinct? Our mind children may be superintelligent,
and physically superior to us, but could they really be considered alive, and somehow
descended from us?
Unlike biological offspring, which are made by chemical processes just like those that
make bacteria, robots will be consciously shaped by our uniquely human minds, by thoughts
more representative of who we are than the unconscious biochemistry in our cells. The
first generations of robots will start with our values, skills and dreams and take them
much further than our old form possibly could. They will certainly be descended from us
(from who else?). Sure, the medium which carries information from generation to generation
will have changed from DNA to something more versatile. But, as in the case of the
transition from vinyl records to music CDs, it is the tune that matters, not the platter.
Do you think that when this time comes, we will "bow out when we can no longer
contribute", or is it more likely that mankind would rather fight than quit? In other
words, what is the likelihood that we will eventually have a War Against the Robots?
Machines have long been an essential part of our society, and will become more integral as
they become more capable. Competition between enterprises will ensure that improved
machines beat out older ones, thus that machine evolution continues. In the book, I
suggest how we can use our founder's position to direct this evolution so the intelligent
machines among us are "tame," meaning that their most basic motivation is to
support us. Like dutiful children caring for aging parents, these machines could provide a
long, luxurious retirement for biological humanity. Few people will object to being
offered steadily greater wealth for steadily less labor. Some that do may choose to leave,
and the great wealth will give them the means. But a war against the machines would be a
war against society. There are some who try to fight such quixotic wars today, but they
can rarely do more than make a minor nuisance of themselves.
Should we put limits on the development of robots or would our best course of action be
to go "full steam ahead"? Is it already too late?
Competition between individuals, companies and countries virtually assures that machine
evolution will continue as fast as reasonably possible. It think we are far better off to
roll with the punch and steer it rather than to try to stop it. Stopping it became
practically impossible 10,000 years ago, when the first agricultural civilizations began
to compete on the scale of city-states. It is really just evolution in a new medium, and
evolution is a vastly resourceful and relentless process.
Is it possible that the "Mind Fire" you predict in your new book has already
happened, and has in fact, nursed our race into existence? In other words, could we
actually be the "mind children" of robots?
I'm sure reality is much more interesting than it seems to us now. I touch on many
different possibilities in the last two chapters.
Which of your robot projects are you most exited about now?
I think we are about five years away from mass-market
utility robots, and I'm orienting my next decade of life to help bring them about.
What are you reading now?
I've recently discovered Greg
Egan's science fiction. He develops in fictional form the same unexpected transcendent
logical consequences of mixing materialism with subjective experience that the idea of
conscious robots forced upon me, and you will find in Chapter 7. Very nice.
Thanks, Carlo Bertocchini
You're welcome! -- Hans Moravec