(FORTUNE Magazine) – A COMPUTER CHIP that combines the knowledge of experts and the imprecision of amateurs could bring artificial intelligence down to earth. Developed at AT&T Bell Laboratories, the experimental chip is the first embedded with what computer specialists call an expert system. Unlike conventional expert systems, which are incorporated into software, the Bell system-on-a-chip could be used to guide robots through assembly line tasks or control the temperatures and timing of a chemical production process. Expert systems, the rudimentary form of artificial intelligence, are databases of information and rules of thumb that sum up expertise in a particular field. They help not-so-expert workers analyze and solve problems and are used in such applications as medical diagnosis and oil exploration. By building an expert system on a chip, Bell Labs eliminates the time-consuming process of retrieving instructions from a computer's memory. Those speedier reflexes make it possible for Bell's expert systems to keep up with the fast pace of a robot or machine tool. Designed by two Japanese-born researchers, Masaki Togai, 37, and Hiroyuki Watanabe, 32, Bell's chip gains some of its power from an esoteric branch of mathematics called fuzzy logic, which is a set of rules that deal with imprecise data by mimicking commonsense reasoning. A conventional expert system is based on probability theory and requires hundreds of rules to cope with a problem. An expert system based on fuzzy logic needs fewer rules; it can read between the lines, so to speak, and make inferences about the actions it should take. AT&T illustrates the workings of fuzzy logic with the rules for braking a train. ''Slight'' brake pressure is the rule 5,000 feet from the station, ''moderate'' pressure at 3,000 feet. A conventional expert system would need specific instructions for the proper braking pressure at 4,350 feet; fuzzy logic would infer slight pressure tending toward moderate. Explains Watanabe, ''Because of fuzzy logic, our chip reasons more like a human being than a computer. It can cope with the gray areas of life, not just the black- and-white ones.'' It copes rapidly: the researchers claim a chip speed of 80,000 operations per second, an output that is rated as 80,000 FLIPS, for fuzzy logic inferences per second. That is about 10,000 times faster than conventional expert systems. Sales of expert systems reached $216 million last year. Industry analysts project a $3.5-billion market by 1990; among the possibilities they foresee are chip-based systems in personal computers and copiers that would analyze and correct their own problems. Bell Labs has yet to decide whether it will produce its chip commercially, however. One big uncertainty is whether the body of knowledge in expert systems will be changing too fast in coming years to be locked into hardware. Industry analysts suggest another concern: whether AT&T has the ability to market the chip effectively.

Whatever the outcome, AT&T has given expert systems a boost. Says Larry Geisel, president and chief executive of Carnegie Group, a Pittsburgh developer of artificial intelligence software, ''We've been working toward putting a higher level of capability into hardware to improve performance and realize economies. AT&T is the first to do so. It's a breakthrough.''