Science & Technology
SOFTWARE THAT CAN DETHRONE `COMPUTER TYRANNY'
For a time, it looked as if fuzzy logic would pass America by--to the detriment of U.S. competitiveness. The Japanese have spent five years cranking out innovative "smart" products, from subway trains to camcorders, that rely on this arcane branch of mathematics. With new "fuzzy" automatic transmissions, Japanese cars shift with unrivaled smoothness. Fuzzy elevators in Tokyo whisk people up and down more quickly. Japanese brokers even use fuzzy logic in stock-trading programs that, according to Wall Street rumors, have outperformed American software. Most U.S. companies, meanwhile, have scoffed at this technique, known formally as fuzzy-set theory.
No more. Now, fuzzy logicis about to appear in scoresof American-made products. Some, such as an elevator-dispatch system coming this fall from Otis Elevator Co., are a response to stiff Japanese competition. Others, such as Allen-Bradley Co.'s new controllers for shop-floor manufacturing, stem from the discovery that fuzzy logic can sometimes produce better accuracy and reproducibility--and therefore quality--than traditional techniques. "We may be five years behind," says Piero P. Bonissone, a computer scientist at General Electric Co.'s research-and-development center, "but at least we've gotten started."
`FREEWARE.' To spur the trend, chipmaker Motorola Inc. has just begun selling a software-development system, created by Aptronix Inc. (box), to help engineers design fuzzy-logic products. In June, Motorola will roll out a computer-based education program to teach customers to write fuzzy-logic software. And it is working on special microchips that have fuzzy logic embedded in the circuitry, similar to the pioneering chips introduced three years ago by Togai InfraLogic Inc., an Irvine (Calif.) startup with many Japanese customers.
Motorola's passion for fuzziness also takes a cue from Japan: The company discovered that Sony Corp. and Canon Inc. were using its standard chips, plus fuzzy logic, to add superior auto-focus features to camcorders. So Motorola decided to test the waters in the U.S. It wrote a simple fuzzy-logic program and, last November, listed it on the company's electronic bulletin board, which contains dozens of programs that customers can download for the price of a phone call. The fuzzy one quickly became Motorola's most popular "freeware" ever.
Steven C. Marsh, director of strategic operations at Motorola's microcontroller division in Austin, Tex., interprets this as proof that U.S. engineers are feeling the heat of Japan's innovations. "Sometimes you have to hit Americans on the head to get their attention," he says. "But now, it's clear to me that by 1995 half of everything we ship worldwide will use fuzzy logic."
EASY OVERLAP. Despite its name, fuzzy logic uses crisp mathematical methods to deal with approximations--such as most, many, few, and slightly--by assigning specific numbers to clusters of overlapping gradations, or sets. It was the 1965 brainchild of Russian-born Lotfi A. Zadeh, who retired last year as a professor of computer science at the University of California at Berkeley. "Classical logic represents an attempt to formalize human reasoning and make it more precise," Zadeh says. "Fuzzy logic accepts the fact that 99.9% of human reasoning is not precise, and it tries to conform to it." Fuzzy logic thus promises to end "computer tyranny." Rather than forcing people to adapt to the way computers work, it imbues products with what Zadeh terms "machine IQ" so they can better accommodate human foibles.
For the Japanese, this has proved invaluable for getting products out faster. Software, typically the main bottleneck in any new-product project, no longer requires programmers to foresee all eventualities in excruciating detail and encode the proper responses in a maze of equations. Instead, problems can be solved with a few rules written in everyday terms.
Example: Who is tall? To answer that with the yes/no, or binary, logic of digital computers, each height must be either tall or not. If you decide that tall begins at 72 inches, you exclude 71.9 inches. But in real life, tallness is relative. Among basketball players, 72 inches is short. With fuzzy logic, the set of tall people can overlap the not-tall. Thus, 72 inches can be tall in one situation but not in another. Binary logic can cope with such anomalies, too--but not easily.
Now, U.S. companies are seeing the wisdom of this approach. Ford Motor Co., for instance, challenged an engineering team to apply fuzzy logic to controls that keep a car's engine idling smoothly under changing conditions, such as when the air-conditioning kicks in at a stoplight. A month later, the program was ready. With binary logic, another team took three months.
AT SEA. Last year, Southwestern Bell Corp. spent many months trying to improve an information-retrieval system that searches for keywords in data bases. The programming problem was to avoid calling up unwanted citations. "We found that marine biologists were getting lots of articles about the Miami Dolphins," notes Jackson Tung, Southwestern's director of information-services technology. Finally, Tung hired a fuzzy expert, who modified the software so it would consider not just keywords but also their context.
United Technologies Corp.'s Otis Elevator unit had a more urgent problem. It needed to match Hitachi, Mitsubishi, and Toshiba, which already have elevators that reduce waiting time. They use fuzzy-logic controls that respond on the fly to changing demands. "Our motivation was primarily competitive," says Siddiq A. Sattar, senior vice-president. "But once we looked at it, we concluded that, for dealing with uncertainties, fuzzy logic offers significant benefits."
GE is pushing to use fuzzy logic to regulate jet-aircraft engines, save water in home appliances, and drive 200-ton rollers in steel mills. Appliance controls will be first, says GE's Bonissone, because "it takes a shorter time to implement them, and there's a faster payoff."
DISCOURAGING WORD. U.S. converts probably won't brag about their fuzzy features. "In Japan, `fuzzy' has achieved a fad status that helps sales," says Gerald A. Eisenbrandt, a staff engineer at Whirlpool Corp., which uses fuzzy logic to regulate defrost cycles in its latest refrigerator. That's partly because in Japanese the word "fuzzy" doesn't carry a disparaging connotation. There's no contradiction in fuzzy lenses that improve the focus of camcorders.
In the U.S., however, marketing executives worry that fuzzy logic might be confused with fuzzy thinking--especially since some computer scientists insist that the two are synonymous. Ford is therefore reluctant to jeopardize the image of a new cruise control that the company hopes will use fuzzy logic not only to maintain a car's speed more accurately but also to brake automatically should the car get too close to another.
Despite the remaining pockets of skepticism, fuzzy logic's newfound popularity is a sweet victory for Zadeh. Like other leading American thinkers, notably quality gurus W. Edwards Deming and J. M. Juran, Zadeh is accustomed to being revered in Japan but ignored at home. That began to change in mid-March, when San Diego hosted the first fuzzy-logic conference organized by the prestigious Institute of Electrical & Electronics Engineers. Zadeh was the keynote speaker and found himself besieged by TV reporters as well as by engineers and other representatives of Corporate America. Even engineers, it seems, are sufficiently tired of computer tyranny to snuggle up with a new idea.Larry Armstrong in San Diego