A CONVERSATION: With Dr.
Marvin
Minsky; Why Isn't Artificial
Intelligence More
Like the Real Thing?
By CLAUDIA DREIFUS
DR. MARVIN MINSKY, 71, a
rumpled-looking man who wears shirts
mended with masking tape,
is Toshiba professor of media arts and sciences
and professor of electrical
engineering and computer sciences at the
Massachusetts Institute
of Technology, and one of the world's leading
theorists of artificial
intelligence. In the late 1950's, Dr. Minsky and John
McCarthy, a professor of
computer science at Stanford University, founded
a research program that
would evolve into the MIT Artificial Intelligence
Laboratory. In addition
to inventing and building thinking machines, Dr.
Minsky wrote the classic
''The Society of Mind,'' (Simon & Schuster, 1986)
in which he tries to show
how intelligence works ''by the particular way the
agents in the brain have
evolved to interact.'' ''The Turing Option,'' (Warner
Books, 1992) a novel by
Dr. Minsky and Harry Harrison, is about
superintelligence in not-too-far-off
2023.
Q. In the 1960's, science
students, particularly those at M.I.T., talked of
artificial intelligence,
or A.I., as if it would create world revolution. Were
they too optimistic?
A. Well, it got stuck. A.I.
was able to produce all kinds of wonderful things
. . . programs that did
better than the average stockbroker or portfolio
manager, programs that could
fix some piece of equipment. Around 1980,
progress stopped in some
ways and people went off in a number of other
directions to try to find
some way to get back. It stopped because we'd done
the easy things. In the
eye of eternity, it got stuck for a moment.
A good example is, in 1964
or 1965, one of our students, Daniel Bobrow
(now a vice president at
the Xerox Corporation) wrote a program that could
read a question from a high
school algebra book, and sometimes, solve the
problem. So it could figure
out a little bit of language and algebra. It didn't
get most of the problems
because it couldn't understand the words. What
people tried to do then
is get a program that would read a story from a first-
or second-grade children's
book. But what happened was this: For any
particular story, we could
build into the program the knowledge necessary
to read that story. We didn't
have much trouble with the grammar. As soon
as something was mentioned
that the program didn't know about . . . (the
system broke down). One
M.I.T. student had a story where some person's
daughter was kidnapped by
the Mafia and they demanded a ransom. So he
asked the program ''What
should we do?'' The program couldn't understand.
Finally, it asked, ''Why
would he pay MONEY to get his daughter back?''
It could figure out a little
bit of language, a little bit of algebra. It didn't get
most of the problems because
it couldn't understand the words. As far as I
know, nobody has been able
to get a machine to solve real problems that
are informally expressed,
the way somebody would normally express them.
Q. How do you define common sense?
A. Common sense is knowing
maybe 30 or 50 million things about the
world and having them represented
so that when something happens, you
can make analogies with
others. If you have common sense, you don't
classify the things literally;
you store them by what they are useful for or
what they remind us of.
For instance, I can see that suitcase (over there in a
corner) as something to
stand on to change a light bulb as opposed to
something to carry things
in.
Q. Could you get machines
to the point where they can deal with the
intangibles of humanness?
A. It's very tangible, what
I'm talking about. For example, you can push
something with a stick,
but you can't pull it. You can pull something with a
string, but you can't push
it. That's common sense. And no computer knows
it. Right now, I'm writing
a book, a sequel to ''The Society of Mind,'' and I
am looking at some of this.
What is pain? What is common sense? What is
falling in love?
Q. What is love?
A. Well, what are emotions?
Emotions are big switches, and there are
hundreds of these. . . .
If you look at a book about the brain, the brain just
looks like switches. . .
. You can think of the brain as a big supermarket of
goodies that you can use
for different purposes. Falling in love is turning on
some 20 or 30 or these and
turning a lot of the others off. It's some
particular arrangement.
To understand it, one has to get some theory of what
are the resources in the
brain, what kind of arrangements are compatible
and what happens when you
turn several on and they get into conflict. Being
angry is another collection
of switches. In this book, I'm trying to give
examples of how these things
work.
Q. In the 1968 Stanley Kubrick
film ''2001: A Space Odyssey,'' a computer
named Hal developed a lethal
jealousy of his space companion, a human
astronaut. How far are we
away from a jealous machine?
A. We could be five minutes
from it, but it would be so stupid that we
couldn't tell. Though Hal
is fiction, why shouldn't he be jealous? There's an
argument between my friend
John McCarthy and me because he thinks you
could make smart machines
that don't have any humanlike emotions. But I
think you're going to have
to go to great lengths to prevent them from having
some acquisitiveness and
the need to control things. Because to solve a
problem, you have to have
the resources and if there are limited resources .
. .
Q. Where were Stanley Kubrick
and his co-author, Arthur C. Clarke, right
with their ''2001: Space
Odyssey'' predictions?
A. On just about everything
except for the date. It's quite a remarkable
piece.
Q. Do you believe the National
Aeronautics and Space Administration
wastes money by insisting
on humans for space exploration?
A. It's not that they waste money. It's that they waste ALL the money.
Q. If you were heading NASA, how would you run it?
A. I would have a space station,
but it would be unmanned. And we would
throw some robots up there
that are not intelligent, but just controlled
through teleoperators and
you could sort of feel what's doing. Then, we
could build telescopes and
all sorts of things and perhaps explore the moon
and Mars by remote control.
Nobody's thought of much use for space. The
clearest use is building
enormous telescopes to see the rest of the universe.
Q. Why are manned shots a NASA priority?
A. Because NASA's people
are basically oriented toward keeping
themselves alive. They are
a big organization. And the biggest part of it is
Houston and that has to
be fed, and what Houston is good at is putting men
in space. The Jet Propulsion
Lab is much smaller and has a smaller staff
and is good at doing everything
else. So, I think, in order to support that,
they get into this vicious
circle where you have to convince yourself that's
what the public wants. Now,
I think, the public is more excited by
Sojourner than by astronauts.
Q. When you go to the movies, what do you see?
A. ''Terminator,'' ''Total
Recall,'' which had ideas about implanted memory.
Pretty clumsy, but I loved
the engineering. I don't like movies exactly. One
of my rules is not to think
of the whole thing as having any unity. The idea of
liking a whole movie is
. . . People have this idea that they have to like
something or not.
Q. What do you read?
A. Science fiction.
Q. Do you read science fiction
in the same way spies read spy novels -- for
ideas?
A. Yes. There are a dozen
very, very rich sources of ideas out there.
Gregory Benford of U.C.
Irvine, David Brin, Larry Niven are the best
writers of our period. When
they write a book, there's some big new idea
about something. I've also
gotten a lot of good ideas from old-timers like
Robert Heimlein and the
late Isaac Asimov.
Q. Where was Mary Shelley
right and where was she wrong with her
''Frankenstein'' last century?
A. She certainly was right
in predicting how people would not understand
the poor thing. That's SUCH
a sad story! By the way, I've gone through that
book very carefully to see
if she left any hints explaining how the robot
worked. But alas, no clues
and the funny part is when you read it, you don't
mind.