Pentagon Program Is at Cutting Edge of Artificial Intelligence : Scientists Seek Battle Computer That Can Outthink Enemy

Pierro Bonissone placed a naval destroyer in the placid waters of Lake George with a good-natured chuckle.

“Now, we have two radar detection ships here and here,” he said, jabbing a finger at a computer terminal, on which Lake George, a peaceful oasis of water skiers and pleasure boats in upstate New York, has been been turned into an electronic battlefield with a destroyer prowling its waters.

“Say we want to get our destroyer into this harbor at Bolton Landing,” he continued, pointing out the quiet middle-class town that appears as a dot on the screen map. “The computer will be able to tell us the most advantageous way to skirt radar detection.”

It may seem like a game, but it’s not. Bonissone is part of a team of research computer scientists at the General Electric Research and Development Center hired by the Defense Department to create a computer that can think on the field of battle--to make decisions on how best to deploy troops and warships when fed intelligence information that may be incomplete or even wrong.

Human commanders have been doing the same thing for centuries. However, scientists argue that a computer can retain millions of bits of detailed information that a human would need months, if not years, to slog through.

In the case of the mythical destroyer slipping through the waters of Lake George, its computer would conceivably be able to hold within its memory the exact distance to Bolton Landing, fuel allocation, the number of men available for battle, the mathematical probability of being detected by specific types of radar, the probability of the weather’s hampering detection, the geology of the lake floor and thousands of other factors considered in combat decisions.

“We are doing what the historians do, but in a matter of minutes rather than years,” said Bonissone, who has code-named the GE artificial intelligence program Lotto. “We are able to consider almost everything that will have a bearing in a battle.”

Lotto is one of dozens of artificial intelligence projects that make up a five-year, $600-million strategic computing program funded by the Defense Advanced Research and Projects Agency, which has backed research into defense technology for 25 years.

“The possibilities are quite startling and suggest a new generation of computing could fundamentally change the nature of future conflicts,” the agency said in a report on the program.

The defense research agency’s strategic computing program, in turn, is one of several artificial computer intelligence research programs funded by the Defense Department, which hopes some day to install a global network of “thinking” computers to monitor and deploy “Star Wars” weapons and satellites, manage battle plans and even help with military procurement of supplies--in short, computers to play a role in every facet of the military.

10 Million Programs

Critics--and there are many--say that computer technology lags far behind the department’s goals. They estimate that it will take as many as 10 million computer programs to support such a network and predict that it will never be feasible.

“We’re talking about an enormous amount of hardware to knit together,” said David Morrison of the Center For Defense Information, a defense watchdog group based in Washington. “Sure, it’s attractive, but it’s not very realistic, especially considering the amount (of money) being poured into it.”

Feasible or not, the Defense Advanced Research and Projects Agency strategic computer program is on the cutting edge of artificial intelligence research. Each project, conducted at half a dozen universities and private research centers, is more mind-boggling than the last.

GE’s extensive research complex, almost hidden in a woods in upstate New York, is so large that it has its own test nuclear power plant, and its 1,600 scientists develop, on average, a new patent every working day.

With floor-to-ceiling glass windows and carpeted halls, the center looks more like corporate America than a scientists’ lair. Bonissone, pudgy and bearded, looks more like a good-natured graduate student than a defense weaponry expert.

In his office, the size of a generous closet, he explained that the concept for the Defense Advanced Research and Projects Agency war computer was derived from a computer used for locomotives.

Several years ago, Dave Smith, who knew everything about the locomotives built by a GE subsidiary company, announced that he was retiring. For two years, researchers transferred Smith’s lifetime of knowledge into a computer and then created a program that would enable it to trouble-shoot. The mechanics told the computer what was going wrong with the locomotive and the computer told them what was probably broken and how to fix it.

Using the same base, the researchers are creating a computer for the Defense Department. The computer is expected to solve problems by drawing on the millions of bits of information and calculated probabilities programmed into it, the institutional knowledge of hundreds of military personnel.

To prepare the test system, Bonissone and colleagues traveled to the U.S. military base at Pearl Harbor earlier this year and gathered information about U.S. and foreign ships and weapons. Back in Schenectady, they gathered information about the terrain and towns surrounding nearby Lake George.

After programming both sets of information into the computer, they had a complete test battlefield and the players on it. Now, the scientists must create programs that will allow the computer to reach reasonable conclusions given certain circumstances.

The researchers’ most difficult task is to program the computer to consider and reconsider its options, based on a continuous stream of incoming information.

Bonissone explains it this way:

“You come to a party in a bus and I tell you I have come in a car.

“You think: ‘Aha, I will be able to get a ride home with him.’

“I say: ‘My car broke down.’

“You think: ‘No, I won’t be able to get a ride home with him.’

“I say: ‘It will be fixed in an hour.’

“You think: ‘I will be able to get a ride home with him, after all.’

“I say: ‘Six people came with me.’

“You think: ‘No way, I won’t be able to get a ride home with him.’ ”

The computer’s “thinking” is based on pre-programmed odds. The odds of riding in a broken car are nil, the odds of seven persons’ fitting in a car are not good.