Researchers: Computers Develop Pack Mentality

Researchers: Computers Develop Pack Mentality

NewsMax.com Wires
Wednesday, Feb. 7, 2001

UNIVERSITY PARK, Pa. (UPI) – Researchers at Penn State University say their computers learned to talk to one another in a self-developed language as they cooperated to "hunt" computerized prey.

Experts say the innovation could have profound implications for the future of robotics and artificial intelligence.

"I could see this having military applications or applications in the space program," said Dr. Lee Giles, a professor of information technology at Penn State. "I think you could see practical applications in hostile environments in the next three to five years, just because we will have so much computer power available to us by then."

Giles and study co-author Kam-Chuen Jim essentially asked a group of computer programs to play hide and seek. These self-directed programs (called "autonomous agents" because of their ability to actively sense and respond to external conditions) were able to "talk" to one another via a message board.

In the course of tracking a computerized "prey," the autonomous agents hunted as a group, developing a common language that enabled them to work together more effectively than has any similar group in the past.

While autonomous agents have been linked together before, they typically have communicated through a preset language created by their programmers. In this case the computers generated their own communications protocol – a unique language that was specific to the task at hand.

That specificity could be the key to enhanced performance in the field, Giles said, "because such a language would be more representative of the environment they were in."

With their own language, he suggested, robots driven by autonomous-agent programs could take the predator-prey scenario out of the lab and into the real world.

Say, for example, a space probe went astray. "You want to find it, and you know it is somewhere, but you don't know exactly where," said Giles. Given sufficient time, it is possible that a team of autonomous agents could find it.

In a military context, "you could see these predators being robots that are trying to capture another 'prey' robot," he said. "That is quite possible."

Giles' team has not yet partnered with any government agency or corporation that might bring those dreams to fruition, and some experts in the field express skepticism about the possible applications of Giles' findings.

"This work was done in the lab, in a controlled environment, and there is a real question as to whether that kind of thing can be scaled up" to handle the demands of a larger or more complex system, said Stan Franklin, a professor in the math sciences department at the University of Memphis and author of the book "Artificial Minds."

Franklin nonetheless said the findings could bode well for the long-term future of robotics.

"A major issue with multi-agent systems is who does what, and if they could talk to one another in their own [situation-specific] language, they could probably do that a little better," he said. "I can easily imagine that there may be good applications of this work, but there is still a ways to go."