PG NewsPG delivery
Pittsburgh Post-Gazette Home Page
PG News: Nation and World, Region and State, Neighborhoods, Business, Sports, Health and Science, Magazine, Forum
Sports: Headlines, Steelers, Pirates, Penguins, Collegiate, Scholastic
Lifestyle: Columnists, Food, Homes, Restaurants, Gardening, Travel, SEEN, Consumer, Pets
Arts and Entertainment: Movies, TV, Music, Books, Crossword, Lottery
Photo Journal: Post-Gazette photos
AP Wire: News and sports from the Associated Press
Business: Business: Business and Technology News, Personal Business, Consumer, Interact, Stock Quotes, PG Benchmarks, PG on Wheels
Classifieds: Jobs, Real Estate, Automotive, Celebrations and other Post-Gazette Classifieds
Web Extras: Marketplace, Bridal, Headlines by Email, Postcards
Weather: AccuWeather Forecast, Conditions, National Weather, Almanac
Health & Science: Health, Science and Environment
Search: Search by keyword or date
PG Store: Pittsburgh Post-Gazette merchandise
PG Delivery: Home Delivery, Back Copies, Mail Subscriptions


Headlines by E-mail

Headlines Region & State Neighborhoods Business
Sports Health & Science Magazine Forum

Software agents can guide officials in complex crises

Sunday, September 30, 2001

By Byron Spice, Science Editor, Post-Gazette

The scenario seems simple compared with events of past weeks: Terrorists have exploded bombs in Kuwait City, prompting the evacuation of U.S. civilians attending an international conference.

It doesn't seem so simple to the U.S. ambassador, who is trying to pick the safest, quickest evacuation route. But in this crisis, simulated by software researchers at Carnegie Mellon University, computer programs called intelligent software agents help the ambassador sort through changing and sometimes conflicting information.

"In these types of crises, people become flustered, they forget things," said Katia Sycara, director of the Robotics Institute's Advanced Agent Technology Laboratory. Software agents, a sort of Internet search engine with brains, always keep their cool, rapidly and automatically scanning computer networks, looking for information that human decision makers will need.

In the Kuwait crisis simulation, developed three years ago for the Defense Advanced Research Projects Agency, a variety of cyberagents gather information on weather, news reports, military and civilian airplane schedules, road information and police reports. The agents help the ambassador alter evacuation routes and, when the city airport is deemed too dangerous, schedule a military airlift.

Sycara is convinced that intelligent software agents could have helped overwhelmed crisis managers during the terrorist attacks of Sept. 11.

It's not hard to imagine how.

Could cyberagents have reminded Federal Aviation Administration officials to alert population centers along the path of United Flight 93? Could a search of the design documents for the World Trade Center resulted in a warning of the Twin Towers' imminent collapse? Could a software agent have recognized the attack and advised the FAA to shut down air traffic sooner?

An Internet search engine typically compiles thousands of links to Web pages that might contain relevant information; the computer user then must look through as many pages as necessary to find the information. An intelligent software agent, by contrast, not only compiles the links but examines the pages as well, pulling off the information it is seeking and then summarizing it for the user.

The potential applications are almost limitless. Sycara's group, for instance, has proposed using them to devise alternate routes to help drivers avoid traffic jams. But, thus far, commercial applications such as personal travel agents and "shopbot" purchasing agents have been a bust.

Military and other governmental officials, however, are showing increasing interest in the use of cyberagents.

"You're getting an incredible amount of information [over computer networks] and what you need is relevant information," said Daniel Daskiewich, a program manager in the Air Force Research Laboratory's Information Directorate in Rome, N.Y. Software agents are one way of filtering that information.

An agent-based scheduling system will be part of a five-year upgrade of the Air Force Strategic Command's command-and-control system that begins next month, Daskiewich said. Cyberagents that sift through reconnaissance photos and intelligence reports to rapidly locate and identify targets for attack will be field-tested next year as part of the U.S. Joint Forces Command's Millennium Challenge, a combination of technologies designed to create "knowledge dominance" in military operations.

NASA has begun to explore the use of software agents in range operations at the Kennedy Space Center in Florida, where the space agency is looking for ways to increase the launch rate for both manned and unmanned vehicles without disrupting civil aviation or Air Force operations.

"We haven't pinned down what we want to do," said Mike Shafto, project manager for human-centered computing at NASA's Ames Research Center in Mountain View, Calif. Though automated systems control the final seconds of a space shuttle launch, he predicted wider use of software agents was still five to 20 years away.

"At the moment, this is a technology of the future," said Behzad Kamgar-Parsi, program manager for Intelligent Systems at the Office of Naval Research in Arlington, Va. "These are fairly experimental and small scale." Most agents now can reliably handle information from only a small number of sources, he maintained, while military commanders need agents that can sift through thousands of sources available on military networks.

But Joseph Giampapa, project manager for the Robotics Institute's softagents group, contended that the technology was already quite advanced. It took the Carnegie Mellon group three months to put together the system used in the Kuwaiti demonstration, he said.

In that DARPA demonstration, a number of cyberagents are used. One keeps track of weather. Another monitors news sites. Another monitors the voice conversations of the ambassador's staff so that the needs of decision makers can be anticipated. These snippets are passed on to other agents that search for the information. Some communicate with existing computer systems, such as an automated flight scheduler used by Transportation Command at Scott Air Force Base, St. Louis.

Some of the agents are so-called middle agents, several of which were designed by Carnegie Mellon, that coordinate tasks or help agents communicate with each other.

The number and type of agents can be changed depending on how the system will be used, Giampapa said.

Software agents might enjoy some advantages on military networks that they don't when scanning the civilian Internet.

Most information on the World Wide Web, for instance, is represented using Hypertext Markup Language -- the ".html" suffix on some Internet addresses. The HTML language makes it possible for people to view information and graphics using a Web browser, but it doesn't allow information to be presented in a way that's easy for machines to decipher. Giampapa said agents could be taught how to find information on particular pages, but must be retrained whenever the page is reformatted.

Alternatives to HTML, including the Extensible Markup Language, or XML, and the DARPA Agent Markup Language, or DAML, are designed to be read by software agents, such as by reducing the ambiguities inherent in English words with multiple meanings. Civilian Web designers could use one of these alternative languages, but they aren't much interested, Giampapa said. Civilian Web pages often depend on advertising to make money, so they have little motivation to make it easy for people to avoid looking at ads by using software agents to glean information from the pages.

Military networks, on the other hand, might be enhanced by use of the machine-readable languages.

Kamgar-Parsi is confident that software agents will eventually prove capable of handling the high volumes of information that now must be processed by military planners and operations personnel.

"In the future," perhaps in two to five years, "the agents will be able to do the same work faster. Not smarter, but faster," he said.

bottom navigation bar Terms of Use  Privacy Policy