January 22, 2007 (Computerworld) Save for a single manned police
car that has sat in front of the building since 9/11, there is nothing about
this particular office tower to distinguish it from hundreds of others in
Arlington, Va. But inside 3701 N. Fairfax Drive, more than 100 computer
scientists, biologists, materials specialists, microsystems experts,
mathematicians and engineers are hatching ideas around a staggering variety
of new technologies.
It’s the headquarters of the Defense Advanced Research Projects Agency, and
the subjects being studied there include software that can translate and
analyze Arabic TV broadcasts, insects with microcontrollers inside their
bodies and the next generation of supercomputers.
DARPA’s philosophical underpinnings have changed several times over the
years
(see “Shifting Missions”), but its mission remains the same. In 1958, in
the aftershock of the Soviet Union’s Sputnik launch, President Eisenhower
formed what was then known as the Advanced Research Projects Agency. Two
years ago, DARPA Director Tony Tether told a congressional subcommittee,
“Our mission is still to prevent technological surprise, but also to create
technological surprise for our adversaries.”
DARPA focuses on technologies for military use, especially those deemed too
risky for the private sector to tackle on its own. But it has also been a
catalyst for many commercial technologies, including timesharing, networking
and the Internet, workstations, database technology, operating systems,
semiconductors and parallel computing.
Open and Shut
Security is tight at the DARPA offices. Guards are everywhere, and visitors
— mostly vendors looking for a piece of DARPA’s $3 billion budget — must
surrender their cell phones at the front desk if they contain cameras.
At the same time, DARPA is extraordinarily open for a military agency. Its
main Web site, www.darpa.mil,
is packed with detailed accounts of what the agency is up to and where it
hopes to go in the future. DARPA’s six offices have undertaken hundreds of
projects, including the following:
The Information Processing Technology Office (IPTO) is soliciting
proposals for “cognitive” technologies that enable systems to reason, learn
from experience, explain themselves and reflect on their own capabilities.
The Information Exploitation Office is working to develop a handheld,
command-guided “loitering cruise munition,” a tiny guided missile that
soldiers can use to fire at targets that are behind and inside buildings.
The Microsystems Technology Office is sponsoring development of an
atomic clock smaller than a sugar cube.
The Intestinal Fortitude Program in the Defense Sciences Office aims
to use beneficial bacteria in the gut to protect soldiers from enteric
disease.
In its Deep Speak program, the Strategic Technology Office (STO) is
developing techniques that will allow communications signals to penetrate
deep into buildings and underground facilities.
The Tactical Technology Office is sponsoring design of the Oblique
Flying Wing, a supersonic aircraft with no fuselage or tail that flies with
one edge rotated forward and one back.
“When I’m vetting a project, clearly what I look for is its value to the
military,” says David Honey, director of the STO. Commercial applications
are often a byproduct, he says.
For example, the STO is trying to solve a difficult and expensive problem
that U.S. forces face abroad. The spectrum of communications frequencies is
statically allocated by type of use and user, and it varies by country. A
military radio that is usable in the U.S. and Zambia, for example, may be
illegal in Germany and South Korea. So radio frequencies must often be
reconfigured for each local environment.
But any particular slice of spectrum at any given locale may be unused much
of the time. “So you establish a network to operate in those open spaces,”
Honey explains, “and when a legitimate use comes up, you are very agile and
you move the entire network over to another part of the spectrum.”
DARPA has developed prototype software and hardware for just such an agile
radio network. “There’s tremendous commercial interest in this, because it’s
a problem worldwide,” Honey says.
Indeed, a start-up called Shared Spectrum Co. in Vienna, Va., hopes to ride
that commercial wave. “The payoff is going to be when the military
application has a commercial extension,” says Peter Tenhula, a vice
president at the company.
For example, he says, companies with their own private wireless networks,
such as couriers, could boost their bandwidth by 10 to 100 times with the
technology, without losing control of their networks as they would if they
purchased network services from a carrier.
Arpanet Revisited
Much of what Honey’s office does today is aimed at making military networks
robust and self-healing. It’s a goal that goes back to the agency’s creation
of the Arpanet — the embryonic Internet — in the late 1960s.
Packet-switching technology and the TCP/IP protocols were designed to ensure
that the network could survive in the face of multiple failures.
“All of this has its roots right back to the work of Vint Cerf and Bob
Kahn,” who invented TCP/IP while working at DARPA, Honey says.
Now DARPA is taking those concepts to a new level, and pioneers like Cerf
are still contributing. For example, IP-based wireless networks don’t work
well when connectivity is interrupted. So in the Disruption Tolerant
Networking (DTN) program, cheap memory is used to cache data packets at
individual nodes until service can be restored.
Some of the concepts for DTN came from Cerf’s earlier work in delay-tolerant
networking for the Interplanetary Internet project. Recalls Honey, “Vint
came to us and said, ‘We are doing this work. Wouldn’t there be some
military application?’ That’s a very common occurrence — people coming to us
with an idea.”
Charles Holland, director of the IPTO, says his unit focuses on “computing
for human productivity,” and the target users are “warfighters and military
decision-makers.”
For example, the IPTO is developing technology to translate and analyze
voices from Arabic and Chinese television and radio broadcasts. “We’ve been
in this for many years,” he says, “but about three years ago, the real
requirement for this showed up, in Iraq. We had to really speed this up, to
make it happen.”
DARPA’s speech technology has been deployed in nine locations, and it can
translate with about 50% accuracy — “good enough to see if public sentiment
is going a certain way,” Holland says. By 2009, that’s expected to reach
90%, which is as good as human translators.
In parallel, the program is working on “distillation” technology designed to
remove irrelevant and redundant information from masses of translated text.
The goal is to go from 30% to 110% of human ability in the next few years.
One project that has commercialization as a specific goal is DARPA’s High
Productivity Computing Systems (HPCS) supercomputer program. “We asked the
vendors to propose to us systems that would be economically competitive,”
Holland says. “They might not sell many at the very high end [2 PFLOPS
scalable to 4 PFLOPS], but they’d have to have a strategy for marketing that
technology so the government wouldn’t have to pay for it all.”
A similar program in the 1980s resulted in the creation of a number of
innovative, high-performance parallel computing architectures — such as
Thinking Machines Corp.’s Connection Machine. But they never went on to
widespread commercial use, leaving the military to pick up most of the
research, development and support costs.
There’s another crucial difference between this HPCS program and the earlier
program, which focused on peak processing speeds.
“We use the word productivity rather than performance,”
Holland says. “It’s from the time a guy thinks about a problem to the time
the answer comes out of the machine.”
The goal, Holland says, is to improve application productivity by a factor
of 10 through new programming languages and development tools. In November,
DARPA awarded a total of $494 million to IBM and Cray Inc. for the next
phase of the program, which is expected to result in fully functional
systems in four years.
On a more intimate scale, DARPA continues to refine and enhance its PAL, or
Personalized Assistant That Learns, a package of artificial intelligence
technologies that learn user behaviors and preferences by scanning e-mails,
calendar entries, Web activity and so on. A prototype is being readied for
deployment in Iraq, where it will automate the chore of writing up reports
about the situation on the battlefield, which currently takes command center
personnel two hours at the end of their 12-hour shifts.
Later in the program, Holland says, PAL will be able to “automatically watch
a conversation between two people and, using natural-language processing,
figure out what are the tasks they agreed upon.”
At that point, perhaps DARPA’s PAL could be renamed HAL, for Hearing
Assistant That Learns. The original HAL, in the film 2001: A Space
Odyssey, tells the astronauts how it knows they’re plotting to
disconnect it: “Dave, although you took thorough precautions in the pod
against my hearing you, I could see your lips move.”