Alum leading development of cognitive

Every fan of the long-ago hit movie and television serial M*A*S*H has their favorite character.

Radar O’Reilly is the favorite of Ron Brachman ’71, a trailblazer in artificial intelligence. In June 2002 Dr. Brachman joined the Defense Advanced Research Projects Agency (DARPA) as director of the Information Processing Technology Office (IPTO) to oversee development of “cognitive” computing systems that reason, learn, and anticipate. The goal is a computer that is a helper, not just a tool.

“One of the key challenges is to make something that is not purely reactive but more deliberative, or even reflective,” Dr. Brachman said. “We need machines capable of not just executing prespecified procedures but using common sense to think ahead about what has to be done. When we imagine what we want cognitive computing systems to do, Radar is a very good model.”

Uncanny Radar
Radar, despite his hesitant speech and sheepish nature, had an uncanny ability to complete Colonel Potter’s sentences. He invariably calmed the harried commander by telling him that he had already submitted the paperwork, ordered the supplies, fueled the Jeep, or done whatever other chore the colonel was remembering had to be done.

The modern American military needs all the Radar O’Reilly’s it can get. To provide such a workforce IPTO is exploring how to stretch current technologies into computing systems that, like people, learn from mistakes, reflect on problems, reason with available information, explain their actions, and react intelligently to unforeseen events.

The DARPA research agenda is driven by the goal of the Department of Defense to maintain the technological superiority of the U.S. military. What the military calls “network-centric warfare” was displayed in the post-9/11 wars in Afghanistan and Iraq.

Volumes of data about terrain, weather, analysis of real-time imagery from drones, the tracking of individual vehicles, and small combat teams are collated to bring ferocious power to bear with uncanny precision.

The key to success, Dr. Brachman said, is linking every element from the most rear-echelon supply depot to the most front-line special operations combat team into a coordinated whole. Modern commanders are no longer at the pinnacle of a hierarchy but at the center of a network that inventories resources and evaluates situations, 24 hours a day, under the most stressful conditions.

“As we move toward the transformed military that (Defense Secretary) Donald Rumsfield ’54 advocates, you are going to see more agile, small-team, ad-hoc networking,” Dr. Brachman said. “It is astonishing how much computing and networking has become part and parcel of running the military.”

For every soldier who pulls a trigger, many more soldiers are assigned supply and support duties. The entire effort is organized and coordinated at “operations centers,” such as the well-known combined air operations center at Prince Sultan Air Base in Saudi Arabia, where targets were selected and aircraft, with specified ordinance, were dispatched during Operation Iraq Freedom.

In this video still shot in preparation for the DARPA Tech ’04 Symposium in March, Ron Brachman, left, poses with Gary Burghoff, who played Radar O’Reilly in the television series M*A*S*H.

Labor-intensive
Despite all the technology, the military remains very labor intensive, Dr. Brachman said. The commander needs a large support staff to set up, maintain, and operate the communications and computing systems. In the field, network nodes are constantly being set up, dismantled, moved, and set up again under conditions that vary from austere to under fire. Even highly trained soldiers, exhausted and stressed, are prone to forgetfulness or mistakes precisely when errors can least be afforded.

Computing systems that are self-maintaining, capable of putting pertinent information into context, and anticipating what needs to be done next would be invaluable, Dr. Brachman said.

“What we are looking at down the road is the creation of assistants, if you will, that would relieve the burdens on humans that are made worse by the stress and pace of events,” he said.

While earning his B.S.E., Dr. Brachman honed his leadership skills as coxswain of the heavyweight crew. He was named captain during his senior year. When he graduated from Princeton in 1971, computers were as exotic as sushi.

Dr. Brachman has been intrigued by artificial intelligence since taking his master’s degree and doctorate at Harvard in applied mathematics.

In 1985 he was recruited by AT&T Bell Labs, where he launched a world-class research group that developed applications of artificial intelligence in telecommunications. Upon his retirement from AT&T in February 2002, the company lauded him as “a world authority in the field of artificial intelligence.” He currently is president of the national organization in the field, the American Association for Artificial Intelligence, and has just finished a textbook on knowledge representation and reasoning.

Dr. Brachman described directing IPTO as a unique opportunity because the requirements of the military push research far past where any application-oriented corporate research institution can venture, and the entire national research community can be part of the effort. Since the early 1960s IPTO has been a womb for the Internet, parallel computing, interactive computing, and other beyond-cutting-edge technologies that have reshaped modern life.

He called the problem of constructing cognitive computers as “DARPA-hard,” which means it is possible if current abilities and technology are stretched to their limits.

Feasible idea
Work done by Dr. Brachman and others over the past two or three decades, such as writing algorithms for reasoning and learning, is why cognitive computers seem feasible. Already, computer systems with natural language capabilities routinely answer telephone calls at customer assistance centers. Voice recognition systems that improve the longer they interact with a person are readily available, while planning technology has made a significant impact on military logistics. IPTO’s research agenda is aimed at bringing all those capabilities together.

“Machines capable of reasoning from knowledge are out there,” Dr. Brachman said. “These systems are saving billions of dollars a year for industry, but from the DARPA perspective they are old hat. DARPA is for advanced research, so we have to take calculated risks for potential very large payoffs. That makes the job exciting.”

Dr. Brachman said the cognitive computing research may yield applications, such as much better self-maintenance of systems within a few years, but that machines that pay attention and learn are many years distant.

“We don’t know even in the abstract how to do this,” he said. “At DARPA we look for people to invent new ways of approaching the problem through new architectures and new algorithms, then demonstrate them in prototype systems.”