But the work also helped create the computer workstation industry, including such companies as Silicon Graphics and Sun Microsystems.
The agency has "paid back its investment by orders of magnitude," said Paul Saffo, a technology forecaster at the Institute for the Future in Menlo Park.
This year, the agency's 160 program officers will dole out $2.7 billion on more than 200 projects in computing, space weapons, counter-terrorism, drone aircraft and biological defense, plus classified programs.
DARPA rotates program officers out after an average of four years to promote blue-sky thinking, said DARPA's current director, Anthony J. Tether. "You can take inordinate risks that you typically wouldn't take at a place where you think you'll be for 30 years," said Tether, an electrical engineer and former top executive with Ford Aerospace Corp. and Science Applications International Corp.
One project, budgeted at $12 million this year, is to build a "brain-machine interface" that would allow soldiers' thoughts to be "turned into acts performed by a machine," according to a DARPA summary. So far, they've gotten a monkey to move a robotic arm just by thinking.
DARPA is also sponsoring a Los Angeles-to-Las Vegas robot race next March to foster robotic research. The first land-based, driverless, fully autonomous vehicle to navigate the 300 miles of road and desert will earn a hefty prize of $1 million.
DARPA's unlikely triumphs, however, have come at a high cost -- 85% to 90% of its projects fail to accomplish their planned goals, although they sometimes spin off unanticipated technologies, according to Tether.
The list of failures is long and strange.
During the 1970s DARPA studied telepathy and psychokinesis, the psychic manipulation of objects. "The Soviets ... had a woman who was fantastic," Tether said. "She could feel colors."
DARPA probed such methods to see, for example, if anyone could psychically peek around the globe for military advantage. "DARPA spent, for those days, considerable amounts of money because the impact would be tremendous if you could do it" -- and disastrous if the Soviets won the telepathy race, Tether said. Ultimately the agency concluded that parapsychology, if real, could not be used on demand, and killed the project.
Among the agency's greatest fiascos was the decade-long program code-named "AGILE," which spent $264 million on a wide range of social, anthropological and technical research during the Vietnam War.
One project aimed to create a "mechanical elephant" ostensibly capable of traversing on "servo mechanism 'legs' " through a jungle too dense for jeeps.
From the outset there were doubts. AGILE's chief scientist likened the project to "sending a million dollars to chase dimes around a rice paddy," according to the DARPA history.
Nonetheless, the scientist justified it as consistent with Vietnam-era profligacy, according to DARPA's history. "We knew it, but we did it," he said. "ARPA just behaved like the nation did [on Vietnam] and was as effective as what the nation did."
When then-DARPA Director Eberhardt Rechtin found out about the robotic pachyderm, he quashed the project, calling it a "damn fool" idea that would destroy DARPA's credibility if Congress ever found out.
DARPA also threw a team of experts at the perplexing challenge of improving field rations for South Vietnamese soldiers.
"Vietnamese combat units were jumping out of aircraft into battle with live pigs and chickens under their arms because there was no supply system," according to agency history. DARPA worked for months to find a suitable container for Vietnamese nuoc mam -- a popular fermented fish sauce "purported to eat through tin cans," agency history noted. It does not say whether the effort succeeded.
DARPA insiders saw AGILE as a failure. But as Herzfeld later explained in the DARPA history: It was "an abysmal failure; a glorious failure."
Today, DARPA is in the midst of yet another transformation, seeking new tools to fight terrorists, who are often indistinguishable from ordinary people. In this battle, the most powerful weapon is information -- data that must be scooped up by the terabyte on innocents as well as terrorists.
One of its leading programs, called Total Information Awareness, was directed by retired Adm. John M. Poindexter, the former national security advisor under President Reagan who was convicted in 1990 of lying to Congress about his role in the Iran-Contra scandal. The convictions were reversed on appeal.
Widely considered a brilliant iconoclast, Poindexter fit the DARPA culture of visionaries who could find provocative solutions to huge problems.
The Total Information Awareness system seeks to locate terrorists by "connecting the dots" in electronic data, such as driver's licenses, purchases of airline tickets and chemicals, intelligence reports and public records. The system looks for patterns of terrorist activity in the records of foreign citizens and ordinary Americans.