WHERE'S HAL?

TEN YEARS TO 2001, AND THE ULTIMATE COMPUTER IS NOWHERE IN SIGHT

Hal felt the greatest enthusiasm for the Discovery mission. In fact, Hal had a
whole range of emotions and thought processes, right down to psychosis. It
could simultaneously hold intelligent conversations, maintain radio contact with
Earth, beat its crewmates at complex games, and control every circuit aboard the
Jupiter-bound spaceship-no small feat for any computer we know.

With A.D. 2001 only ten years away, we don't have anything close to Hal 9000,
the silicon villain/technological wonder of Arthur C. Clarke's 2001: A Space
Odyssey. It's not because we don't need such a machine. "I imagine that the
scientific community is looking forward to the first prototypes," says Clarke.
According to many scientists, however, there's only a slim chance that we'll
make the deadline or even the sequels-although other artificial intelligence
experts are considerably more optimistic.

AI has already missed one milestone. In the 1968 book, Eighties scientists were
"growing" artificial brains using self-replicating neural networks, a
breakthrough that led directly to Hal, or heuristically programmed algorithmic
computer. Researchers in the Nineties, however, have taken only the first few
steps. "Artificial intelligence is good at specific areas of knowledge, at doing
one task at a time," says Marvin Minsky, a computer science professor at the
Massachusetts Institute of Technology (MIT) and one of the physicists credited
in Clarke's book with the breakthrough. An expert system, for example, performs
one chore, such as giving medical diagnoses. The closest AI has come to
machines with Hal-like abilities are control systems, which can fly, navigate,
and land a Boeing 767.

Hal would be able to do multiple chores at the same time. "When you consider
what would go into a Hal, you think of knowledge, vision, advanced language,
learning-all working together," says Ron Brachman, head of AI principles
research at AT&T Bell Labs in New Jersey, who raised the topic at an American
Association for Artificial Intelligence meeting earlier this year. "We either
have or are working on many of the pieces, but no one has tied them all
together."

Yet Hal's prospects aren't all gloomy. For instance, AI may soon possess common
sense, "the one single component that has been holding back progress," says
Douglas Lenat, principal scientist and director of AI at MCC, a research
consortium in Texas.

Six years ago Lenat began programming common sense and reasoning into a
knowledge base. "You have to put in information about daily human activities:
If you don't drink water for several days you will get thirsty; if you go to
work by car, you typically get home by car," he says. "There are also the
subtleties of human language: If I say, 'I saw a bicycle through the window and
wanted it,' you'd know it meant the bicycle, not the window."

Lenat expects that his work, which will take several more years of inputting,
could lead to a Hal prototype before the deadline. "By the mid-Nineties, natural
language systems will be doing well enough that the information will be input
verbally or by reading. We will be teaching the computer rather than
programming it [a feature of the Hal 9000]." He also hopes his machine will
understand emotions. "Computers don't have to have emotions, but they can't
understand the world unless they know, for example, why people get angry."

Another Hal hopeful is the Soar project, started in 1982 by Professor Allen
Newell and two graduate students at CarnegieMellon University (CMU). "Soar is
the first general problem solver that learned as it worked," says John Laird, a
CMU graduate and an assistant professor in the University of Michigan AI lab. A
similar project, called Prodigy, is advancing under CMU professor Jaime
Carbonell.

In spite of these advances, Laird doesn't hold out much hope for a Hal for at
least 50 years: "We don't understand human intelligence enough to make a
complete artificial intelligence." Yet all the researchers believe we will
someday have a Hal. Once we do, Clarke hopes it will be more than one small
step for man. "You won't see it doing something as mundane as, say, being a bank
teller or working in a fastfood restaurant," he says "Once, after the book came
out David Frost asked me whether they could build a computer to do his job. I
said they would not waste a computer on that."