High-tech 'prospector' breaks new ground in computer technology
Star Wars' robot R2D2 with a pick over its shoulder, leading a burro?
That's the fanciful image conjured up by a computer program called PROSPECTOR , which has struck paydirt: It successfully identified a previously unknown ore deposit. It appears to be the first computer program to have achieved such a feat.
Created in the Artificial Intelligence Laboratory at SRI International in Menlo Park, Calif., the program is one of the most advanced of the ''expert systems'' that are on the cutting edge of the computer programming field. These attempt to imitate the thinking process that experts in various fields employ as they analyze specific problems. The program was announced in a paper in a recent issue of the journal Science.
PROSPECTOR may lack the personality of an R2D2, but its capabilities seem impressive. In this case, it located a deposit of molybdenum in the area of Mt. Tolman in eastern Washington state.
''The computer is not as good as a human expert, but it's much better than nothing at all. We are trying to make expertise a little more widely available, '' explains Rene Reboh, project director of SRI's Expert Systems Program. A system like PROSPECTOR may prove useful in third-world firms where geological experts are in short supply or for companies who do not have access to geological experts, he says.
To create PROSPECTOR, SRI brought in experts in specific areas of prospecting. They quizzed and cross-examined each for about 50 hours.
''Essentially, an expert makes connections between certain pieces of information,'' says Dr. Reboh. Through the interviews, the computer experts determined what these connections were and programmed PROSPECTOR to duplicate them. Giving the computer and the expert the same problems, their answers agreed to within 6 or 7 percent on average, Reboh reports. He expects people to begin using programs of this sort within a few years.
PROSPECTOR is a part of a larger system called KAS, for Knowledge Acquisition System. This is used to create expert systems in almost any field. Right now the SRI group is working on similar systems in hydrology and chemical spill response.
The biggest problem with these systems, says Reboh, is the difficulty of dealing with uncertain and incomplete data.
''Human experts,'' he says, ''have ways of combining this kind of information to come up with good answers. We have to find ways to express these uncertainties better.''