A fresh breath of Oxygen-MIT researchers are working to make computers more intuitive

”Begin the presentation,” says Rodney Brooks, and just like that the room lights
wink out, the curtains close out the bright sunshine of a Cambridge summer day,
and a video projector begins flashing PowerPoint slides against the wall.

By Hiawatha Bray, Globe Staff,

Three words, spoken
aloud, was all it took.
Brooks had no human
helpers, just a computer
that understood and
obeyed.

Feel free to try this at
home. It won’t work
unless you’re exceedingly
clever.

Brooks qualifies on that
score: He’s director of the
Artificial Intelligence Lab
at MIT’s Laboratory for Computer Science. And even he needed help from dozens
of colleagues and students to put his system together.

Far from an exercise in self-gratification, Brooks’s automated office is part of a
major MIT research project called Oxygen, a project aimed at finally creating the
kind of interaction between computers and people that we see every summer in
Hollywood blockbusters, but never in real life.

Nowadays when we use computers, ”we go into the computer’s world,” Brooks
said. We have to present information in ways that computers can understand, using
tools like keyboards that seem unnatural to most people.

”I wanted to pull the computer out into our world,” said Brooks – to give it a new
capacity for human-like interaction with humans. That means building machines that
look you in the eye and recognize you, machines that let you ask questions in
casual, conversational English, and answer them the same way, machines that know
how to ask for help from other machines, without having to be told.

It’s a ferociously complex task, but if they pull it off, the Oxygen team’s work could
create a vast new market for computer technology, while flinging a lifeline to a
moribund computer industry. The computing boom of the 1990s is decisively over,
with companies like Apple Computer Inc. and Intel Corp. cutting jobs in the face
of dwindling profits. Even when the economy recovers, the saturated market for
traditional computers will never grow as it once did. By breaking down the wall
between people and computers, Oxygen-based systems could consign today’s
relatively crude digital products to the scrapyard, while spawning new demand
among millions of people who have never touched a mouse or a keyboard.

On the other hand, there are still years of work ahead before Oxygen-based
systems are ready for consumers. Even Brooks’s simple demonstration of a
voice-controlled office leaves something to be desired. The computer still has
trouble distinguishing between commands and noise. ”Sometimes people laugh, and
the curtains open,” Brooks said.

Just as important, it’s not clear that people will pay for automated draperies, or
refrigerators that can tell when it’s time to buy more milk. These are activities that
may be difficult for computers, but they’re trivial for humans. Even if MIT scientists
perfect the underlying technologies, they’ll have to find applications that consumers
will want to buy.

In that effort, they’ll have plenty of help. Some of the world’s leading high-tech
companies are bankrolling the project: Hewlett-Packard Co., Royal Philips
Electronics of the Netherlands, and Taiwan’s giant computer maker Acer Inc.,
among others. They regard Oxygen as an investment in a future in which people will
relate to computers in much the same way they communicate with people. ”We see
this as a very strong opportunity for growth,” said John Ankhorn, a research
scientist at Hewlett-Packard Labs in Palo Alto, Calif.

The features that make computers easier to use – speech recognition, for instance –
require lots of computing power. That’s good news for an industry whose current
products are actually better than they need to be. Because most people still use
computers for simple tasks like Web surfing and word processing, they aren’t
tapping the full potential of the latest high-end processors.

Roger Kay, computer industry analyst with International Data Corp. in
Framingham, says this is one reason for the present slump in computer sales.
Consumers ”basically say, `I bought one in 1999,”’ Kay contends. ”`I bought it to
get on the Internet. I can still get on the Internet. … I don’t see a reason to get
another one.”’ A torrent of new Oxygenized devices could help cure this digital
malaise.

The very name of the project underscores its ambitious goals. Oxygen is vital to
nearly every living thing on earth, but it’s free and it’s everywhere. The late Michael
Dertouzos, director of MIT’s Laboratory for Computer Science, foresaw a future
in which computing power was as cheap, plentiful, and necessary as the air we
breathe. For Dertouzos, that meant that MIT should lead the way in making these
computes as easy to use as it is to breathe. This human-centric vision of computing
led him to launch the Oxygen project in 1999, with seed funding from the US
Defense Advanced Research Projects Agency.

Dertouzos died of a heart attack last August, but his successor, Victor Zue, shares
the vision of a computer-saturated world. ”In five to 10 years, in developed
countries, computing and communications are essentially going to be free,
pervasive, everywhere,” said Zue. ”It’s going to be in your walls, in your cars, on
your body.”

But nobody’s going to use a mouse, monitor, and keyboard to access the computer
in your refrigerator or light switch. What’s needed are user interfaces that are at
once simpler and more sophisticated.

Human voices, for instance. Today’s computers can transcribe human speech, and
carry out a limited number of spoken commands. But for best results, the user must
”train” the system, reading words into the computer to help it learn his speech
patterns. And commands must be phrased in exactly the right words. A computer
might respond to ”What time is it?” but not to ”Tell me the time,” even though the
two phrases mean the same thing.

Zue, an expert in speech recognition, is helping design Oxygen systems that will
recognize anybody’s voice, without a training period. These systems will also
understand that people can say the same thing in different ways. For instance, the
Oxygen team runs an experimental speech-activated telephone system that
provides weather and traffic information. A user can say, ”What’s the
temperature?” or ”How hot is it?” Either way, the system provides the right answer.
The system can even be programmed to recognize multiple languages; in one
demonstration, it answers questions in English or Japanese, depending on the
language used by the caller.

But people don’t just speak to each other. We recognize the faces of friends and
colleagues. We notice their hand gestures, facial expressions, the angle of the
speaker’s head. These things clue us into the meanings of words, or even whether
someone is talking to us or the other fellow. Computers must learn to do this if
they’re to become easy to use.

”We’ve just recently gotten to the point where we can recognize pointing gestures,”
says Oxygen researcher Trevor Darrell, an expert on computer vision systems.
Darrell shows off a system that uses a stereoscopic TV camera with three lenses,
connected to a PC. As one of Darrell’s students points a finger at the screen, a
green dot appears. As his arm moves, so does the dot. It seems trivial at first
glance. But it took nine months to write the software, and the full power of a
late-model desktop computer to run it.

Making computers that understand people is just one of the challenges the Oxygen
team has embraced. They’re also training computers to understand forces of nature,
like gravity. Randall Davis, associate director of the MIT Artificial Intelligence lab,
is working on a ”sketch understanding system” which lets the user draw objects on
a screen, which then interact as they would if the objects existed in the real world.
A user can draw a diagonal line, and add the image of a four-wheeled vehicle on
top of the line. At the touch of a button, the computer interprets the line as a hill,
and shows the vehicle rolling toward the bottom. Change the shape of a wheel, and
the vehicle wobbles as it rolls.

It’s more than a parlor trick. The goal is an engineering design tool that will let users
try out ideas by sketching them, and getting instant feedback on whether the idea
will work.

Oxygen-based computers must also understand each other, instantly sharing
information in order to serve their human masters. Consider Zue’s penchant for
physical fitness. ”I stand in front of my bathroom scale. Every day I measure
myself,” he says. ”Then I religiously go to the PC, go into the spreadsheet and type
those numbers into that. That’s stupid.” What Zue wants is a scale that instantly
transmits his weight to his PC. That means building wireless networking capability
into millions of ordinary devices.

The scientists working on Oxygen have transformed their offices into testbeds for
their innovations. Many feature small battery-powered circuit boards called
”Crickets.” These are wireless transceivers that detect and identify other Crickets
attached to handheld computers. The result is a system that can identify a person
the moment he steps into a room, and tailor the room’s electronics to his particular
needs.

For instance, Ken Steele totes a heavily modified iPaq handheld computer,
featuring a Cricket and a standard wireless networking card. The handheld displays
a live video image of grad student Jason Waterman, who’s located elsewhere in the
building. But when Steele crosses the threshold of his office, Waterman’s image
instantly appears on the monitor of his desktop computer.

Steele didn’t issue any commands or press any buttons. He didn’t have to. The
system knew what to do. A Cricket on the ceiling detected his presence in the
room, and notified the Oxygen network. The network ”knew” Steele was viewing
video on his handheld, and that he’d get a better picture on his desktop monitor. So
it routed the image there, without being asked. If Steele needed to print out a
document, the system would know enough to use the printer in his office, not the
one down the hall.

It’s impressive, but ”it wouldn’t be commercially viable at the moment,” Steele says.
Each Cricket costs about $60 to build. Besides, an iPaq equipped with all the
necessary accessories is about the size of a brick – far too clumsy for use by a
consumer.

In addition, there are still plenty of bugs. Steele’s iPaq also has a small camera, and
software that’s supposed to log him onto his office network by a combination of
voice and face recognition. But it doesn’t work. Time and again, an electronic voice
emerges from the computer: ”You are not Ken Steele. Please go away.”

The glitch is a salutary reminder that Oxygen technology is years away from
appearing in consumer products. The industry would have to adopt major new
initiatives in computer networking in order to make it all work. For instance, the
present Internet Protocol system, called IP version 4, doesn’t have nearly enough
network addresses to support billions of digital devices. A newer, grander version
called IPv6 exists on paper, but has yet to be implemented by the computer
industry.

Still, Zue coyly says that a few Oxygen technologies have been scooped up by the
project’s corporate R&D labs, in hopes of turning them into commercial products.
”I think there will be low-hanging fruit that are going to happen soon,” he says.

It’s likely that the successful Oxygen products will gradually seep into our daily lives
without fanfare – a speech-activated dishwasher, perhaps, or a car with wireless
networking that automatically pays for its own gasoline. Eventually, we’ll be
surrounded with these devices, scarcely able to live without the stuff. Just like
Oxygen.

Hiawatha Bray can be reached at [email protected].

© Copyright 2002 Globe Newspaper Company.

http://www.boston.com/dailyglobe2/224/business/A_fresh_breath_of_Oxygen+.shtml