KU training computer for war
From Lawrence Journal World
By Erwin Seba
KU scientists are creating thinking machines for the military.
It's war. And these days it's hellishly electronic.
Radar-guided missiles and jets are flying, satellites are spying, ground motion detectors are buzzing. Life-or-death decisions are made in milliseconds.
In modern warfare, information flows to command centers faster than humans can handle it and two Kansas University professors are working on a project that will allow computers to decide battle actions previously left to human field commanders.
The Defense Department is betting $1.3 million that KU professors Costas Tsatsoulis and Douglas Niehaus can turn networked military computers into thinking machines that decide and act with little or no reliance on human intermediaries.
The key question is this: Can computers "find a solution that's good enough soon enough," said Tsatsoulis, a professor of electrical engineering and computer science.
"The techniques we are using are from artificial intelligence," he said. "The systems we are building are intelligent."
Not all applications of the work being funded by the Defense Advanced Research Projects Agency (DARPA) are meant to replace human responses to the 21st-century combat zone. In situations where time allows, humans still could participate in the decision-making.
"It depends on the amount of time," a pilot or commander might have, Tsatsoulis said. "If you have milliseconds, no. If you wait that long, you're going to be dead."
The research also could lead to applications in civilian computing, telecommunications and emergency management, he said.
"In emergency situations you want Federal Emergency Management Agency messages to have priority over the (non-emergency) phone lines," Tsatsoulis said.
"In general, it would be useful in situations where you have less resources than possible solutions," he said.
Tsatsoulis and Niehaus, along with nine students and staff are studying how computers negotiate with one another when making decisions. The computers must be able to act under deadlines imposed by the situation.
Tsatsoulis said his primary interest in the research is seeing if theory and laboratory results can be duplicated in the real world.
Battle testing would be "the validation of our ideas in a real-world environment," he said.
The project must be completed within two and a half years. DARPA only funds projects expected to be in use by the armed services within five to 10 years.
This is Tsatsoulis' third computer research project for DARPA. He's also done research for NASA and the National Institutes of Health. He came to KU after earning degrees in electrical engineering at Purdue University.
"I decided to stay here forever so I don't have to work," he joked.
He estimates he puts in 60 to 70 hours a week on research. He also teaches one class a semester.
He doesn't get all the research money. About half goes to the university for overhead costs. The rest pays salaries for students and researchers and buys materials.
Neither he nor Niehaus are getting rich, he said. Their salaries from the project equal, at best, about three months' of university pay, in essence making up for college pay they otherwise wouldn't receive during the summer.
"We do the research because we are intellectually interested in that subject," Tsatsoulis said. "And because it's important for the graduate education of the students.
"Graduate students need to be involved in research and the research has to be relevant."
For more information, contact ITTC.