Artificial Intelligence Hasn't Peaked (Yet)

By KATIE HAFNER - New York Times

December 28, 2000

Just what constitutes artificial intelligence has always
been a matter of some dispute. And the terms of the argument
change with each new advance in computer science.

Seen one way, as the effort to produce machines whose output
cannot be distinguished from that of a human, artificial
intelligence, or A.I., is still very far away.

But from another perspective, it is all around us.

Thirty years ago, for instance, speech recognition was an
artificial-intelligence problem of the first order. Today it
is commonplace, a fact that is evident to anyone who has
called the United Airlines flight information line or has
used speech transcription software.

"These things are considered A.I. before you do them," said
Dr. Danny Hillis, who has been working in the field for
years. "And after you do it, they're considered

Other fruits of artificial intelligence research abound as
well. Whether you are struggling to beat your Palm organizer
at chess, watching your word processing program correct your
spelling or playing a video game, you are witnessing the
ways in which artificial intelligence has insinuated itself
into daily life.

"A.I. is becoming more important as it has become less
conspicuous, and it's less conspicuous because it's
everywhere, but often under the surface," said Dr. Patrick
Winston, a professor at the Massachusetts Institute of
Technology who was the director of the Artificial
Intelligence Lab there for 25 years.

Since the time when the first work was being done by Dr.
Marvin Minsky, Dr. John McCarthy, Dr. Winston and others at
M.I.T., in the 1950's and 1960's, computer scientists have
generally agreed that artificial intelligence would arrive

"We're engineering A.I. one piece at a time," said Dr.
Hillis, a former student of Dr. Minsky's and chairman of
Applied Minds, a start-up in Glendale, Calif.

Dr. Hillis and others said that the machine intelligence
currently in evidence fell along a spectrum.

At the less intelligent end are things like smart washing
machines and coffeepots - appliances that can figure out how
dirty a load of clothes is or when to turn off a coffee
warmer. Experts generally agree that such appliances are the
product of rather sophisticated microprocessors and sensors,
not evidence of artificial intelligence.

At the other end are machines whose output is genuinely
difficult to distinguish from a human's, like I.B.M.'s
chess-playing computer, Deep Blue, and Aaron, a robotic
artist that produces paintings that could easily pass for
human work.

And somewhere in the middle are speech recognition programs,
used in lieu of word processors; collaborative filtering
software, like that used by to make purchase
recommendations; and search engines that respond to
questions phrased in full sentences, not just search terms.

One reason for the proliferation of machine intelligence in
the commercial world is the seeding of the computer industry
with artificial-intelligence researchers who have moved
beyond academia and taken jobs at high-tech companies.

The Microsoft Corporation, for instance, employs about 80
artificial-intelligence researchers, many of whom came from
universities. For several years, Microsoft has sold its
Office software with various embedded intelligence features,
like the automatic correction of frequently misspelled words
and the Answer Wizard, which anticipates the needs of users
who look up topics in the electronic documentation.

An infamous piece of Microsoft software that includes
components of artificial intelligence is the Paper Clip help
wizard, which pops up on the screen to offer advice. Many
people say Paper Clip pops up too often with unwanted

In defense of Paper Clip, Dr. Winston said: "It's less
annoying than it would have been without A.I. It does try to
zero in on what kinds of information you're most interested
in, and that sort of thing will get better and better as
time goes by."

Microsoft's next big step into the marketplace with a
product that incorporates artificial intelligence will be
its Outlook Mobile Manager, a system that scrutinizes each
incoming e-mail message, does an automatic synopsis, throws
away extraneous words and abbreviates others, then sends the
message to the user's mobile device. The product is
scheduled to be released next year.

"It's what a great secretary would do," said Craig Mundie,
Microsoft's senior vice president for advanced strategies.

Researchers in artificial intelligence at Microsoft are also
working on a more general effort called the Attentional User
Interfaces and Systems Project, which includes a project for
continually monitoring streams of data like e-mail, voice
mail, Internet news alerts and instant messages. The system
will gauge what the computer user is doing, assign
priorities to the messages and decide whether and when to

Other graduates of university-based artificial intelligence
programs have started companies of their own. In 1983, Dr.
Hillis co-founded the Thinking Machines Corporation, a
supercomputer company that was bought by other companies in
the 1990's. In 1986, Dr. Winston and three colleagues at
M.I.T. started Ascent Technology in Cambridge, Mass., to
apply the research they had been doing to help airports
solve scheduling and allocation problems like gate
assignments for aircraft.

Dr. Winston said the first commercialization efforts of
artificial intelligence, in the 1980's, had made an obvious
mistake. "We blundered about what we thought A.I. was going
to be good for, which was replacing people," he said. "What
we discovered was that's not the commercial appeal of A.I.
It's about making things possible that weren't possible with
people alone."

As examples, Dr. Winston pointed to a project at the
Artificial Intelligence Lab for giving brain surgeons a kind
of X-ray vision by coupling video images with M.R.I. images.
He also pointed to the Mars Rover, which navigates terrain

Another M.I.T. spinoff is the iRobot Corporation, started 10
years ago by Dr. Rodney Brooks, the current director of the
Artificial Intelligence Lab. The company developed an
interactive doll with Hasbro called My Real Baby and in
February will begin selling the iRobot-LE, a self-navigating
home robot that will be equipped with sonar and a camera and
will be controlled via the Web.

Of course, there are those who disagree that pieces of the
artificial intelligence puzzle are falling into place
incrementally. "It's very much like a country that's
declaring a war that it's losing to be won and then
withdrawing," said Dr. Douglas Lenat, an artificial
intelligence researcher who is president of Cycorp in
Austin, Tex.

By way of example, Dr. Lenat described the shortcomings of
speech recognition programs currently on the market. "They
are just the palest shadows of what we can, should and soon
will have with real A.I.," he said. "You have to speak the
punctuation marks, and that's pretty pathetic. And they
don't recognize the simplest inflections for things like
italics and commas.

"There is still this tremendously important problem, which
is to get computers to know enough about the world that they
can do the final few percentage points of speech

Dr. Lenat's criticism of speech recognition raises the
larger question of what constitutes intelligence.

Dr. Hillis addressed that question this way: "Intelligence
is just a whole lot of little things, thousands of them. And
what will happen is we'll learn about each one one at a
time, and as we do it, machines will be more and more like
people. It will be a gradual process, and that's been

Dr. Ray Kurzweil, an artificial intelligence researcher who
created the Kurzweil VoiceReport, a speech recognition
program, agreed that "machines still do not have the
subtlety, depth, range and richness of human intelligence
because it is still a million times simpler than the human

"That gap," Dr. Kurzweil continued, "is going to go away,
and when it does, then machines can combine the subtlety and
pattern-recognition strengths with the other natural
advantages they already have, and that will be a very
formidable combination."

Perhaps the flight schedule information line that
understands words like "Chicago" and "today" helps take
machines a step closer to duplicating the outward signs of a
person's intelligence. But the artificial intelligence field
remains far short of modeling human consciousness and the
inner mind.

"A.I. has done a lot of little things that are very
powerful," Dr. Winston said. "On the other hand, on the
science side, where we try to understand what makes humans
work, we're still a long way from that prize, and we need to
work hard on it if we want to understand our intelligence
the same way molecular biologists understand our genes."

The world of artificial intelligence would not be the same
without a robotic lawnmower, available for about $800 from
Friendly Robotics, which has its United States headquarters
in Irving, Tex.

Dr. Brooks, of M.I.T., has used the device, called the
Robomower, on his own yard in suburban Boston. He said that
although it did a respectable job on a patch of lawn, the
family gardener's reaction reinforced Dr. Winston's point.
"He looked at that third of the lawn and said, `I guess I'm
not out of a job soon,' " Dr. Brooks said.

This file was provided to you by the WordWeaver

[email protected]

End Of File