Science / Medicine : Fuzzy Logic : Computers Help Machines Think Like Humans, in Shades of Gray
The age of intelligent machines is beginning to look a little fuzzy.
Take video camcorders. Selecting one used to be a fairly straightforward decision: VHS or 8-millimeter format; hand-held or palm-sized. Now, there’s a new alternative: fuzzy or non-fuzzy.
If the world’s consumer electronics industries continue to follow the Japanese lead, shoppers will soon be weighing the advantages of fuzzy washing machines, still cameras, vacuum cleaners, air conditioners, microwave ovens, water heaters, television sets and portable computers.
All of these items, which are already on the market in Japan, are controlled by fuzzy logic, a new computer technique that operates in a more flexible, subjective mode.
Fuzzy decision and diagnostic systems are analyzing medical information, trading stock and helping golfers improve their game. In Japan, fuzzy logic runs some elevators and subway trains. Soon, industrial and manufacturing processes, nuclear power plants and robots will have fuzzy controls. Automobiles and, perhaps, aircraft will depend on fuzzy circuits. Years from now, astronauts maneuvering the space shuttle for a rendezvous with the space station may get a fuzzy assist while docking.
The success of fuzzy systems is linked to their ability to combine the human attribute of common sense with the computer’s virtues of speed and consistency.
“Fuzzy logic is an attempt to bring human beings and machines closer, by explaining human concepts in a way useful to machines,” said Ronald R. Yager, director of the Machine Intelligence Institute at Iona College in New Rochelle, N.Y. “It’s a way to appease those people who are sort of anti-technology, by allowing subjective considerations . . . to be included.”
According to Camerone Welch, a consultant with the Gartner Group Inc. of Stamford, Conn., an information-service company specializing in technology:
“Why not make computers reason the way we do? I think fuzzy logic is a revelation for computers and science.”
The theoretical underpinnings of fuzzy logic were proposed in 1965 by Lotfi A. Zadeh, a professor of computer science at UC Berkeley. Zadeh said his theory grew out of his work analyzing and characterizing different types of systems. He found that there were certain systems that “didn’t lend themselves to precise characterization.”
The best way to handle such systems, Zadeh reasoned, was with a new kind of set theory. Set theory is a mathematical term for the study of groups of elements.
In conventional set theory, elements are either members or non-members. What Zadeh proposed was a new set theory in which membership in a set was a matter of degree. Specifically, set elements would have membership grades between 0 and 1, with a 1 indicating member and 0 indicating nonmember. Thus, in the set of luxurious cars, for example, a Rolls-Royce might have a membership value of 0.99, while a Hyundai might have a membership value of 0.15.
Although seemingly straightforward and innocuous, Zadeh’s proposal was resisted and even ridiculed by many of his peers. In 1975, William Kahan, a close associate of Zadeh at Berkeley, said: “Fuzzy theory is wrong; wrong and pernicious. . . . The danger of fuzzy theory is that it will encourage the sort of imprecise thinking that has brought us so much trouble.”
Zadeh believes that part of the problem then--and part of the reason that fuzzy logic still has not gained wide commercial acceptance in the United States and Europe--is that his theory flies in the face of 2,300 years of Aristotelian mathematics. Another factor commonly cited is the black or white way in which the world is generally viewed in the West. For example, people are labeled a success or a failure; a woman is either beautiful or she’s not; and moves are either good or bad (thumbs up or down).
In Japan, however, the technology has met no such resistance.
“Use of fuzzy logic fits in well with the Japanese culture,” concluded a recent State Department report prepared by the U.S. Embassy in Tokyo. “Japanese religions, philosophies, use of language and decision-making processes, for example, are attuned to ambiguity and lack of precision. The Japanese culture is not so deeply rooted in scientific rationalism as are the North American and European cultures.”
In the 1980s, nonetheless, many computer scientists and engineers in both Japan and the United States realized that for some applications, controllers (electronic devices designed to operate larger computer systems) based on fuzzy principles were simpler and more effective than conventional controllers. Moreover, fuzzy logic could be used to control complex systems for which a precise mathematical model is unavailable.
Another important advantage is that the fuzzy controllers work in the way a human might when controlling a system: by taking into account several factors simultaneously and applying common sense.
For example, a fuzzy controller of a commuter train might take three variables into consideration: the train’s speed, acceleration, and distance from the next station. The values of these variables would be used to determine two operating commands: the power level of the train’s motor and the extent to which the train’s brakes are applied.
While the train is running, the variables are repeatedly applied to a group of fuzzy sets, to determine their degree of membership. For instance, if the train is traveling at a constant 65 m.p.h. is about a mile from the station, the speed, 65 m.p.h., may have a membership value of 0.7 in the “speed is high” set and 0.1 in the “speed is low” set. The fact that the speed value is a member of contradictory sets is unacceptable in most conventional forms of logic but is an important feature of fuzzy logic. The distance, one mile, might have a membership of 0.6 in “train is not far from the next station” and 0.2 in “train is entering the next station.”
The heart of the fuzzy system is a series of perhaps a dozen or so “if-then” rules, such as: “if the speed is high and the train is not far from the next station, then apply brakes gently” and “if the speed is high and the train is entering the next station, then apply brakes sharply.” This type of rule is commonly used in computer programming, but fuzzy if-then rules are less specific, so relatively few of them are needed--about 10 to 100 times fewer than in a conventional control program.
In the train example, both of the “if” conditions are true, so both rules will be acted upon (or “fired,” in fuzzy jargon). However, each of the rules calls for a different action: One calls for the brakes to be applied gently, the other for them to be applied sharply. And since the brakes can only be applied with one degree of force, a process of “defuzzification” is necessary, in which a weighted average of the two responses is computed. Thus the brakes will be applied with a force somewhere in between “gently” and “sharply,” but closer to “gently.”
Such approximations make fuzzy logic a simpler, less expensive alternative to conventional controllers for many applications.
“The point is, you can trade off a little specificity and still make conclusions that are useful, and correct enough,” Yager said. “Fuzzy sets allow you to manage this trade-off.”
The same basic principles, moreover, can be applied to a wide variety of practical problems. A fuzzy washing machine by Matsushita Electric Industrial Co., one of the leaders in commercial applications of fuzzy logic, senses the type and amount of dirt in the wash water, and accordingly adjusts the wash time, agitation speed and detergent used. Fuzzy camcorders interpret focus and exposure information from a number of points within a framed image to choose optimal settings. The television set, a top of the line Sony model, monitors 248 reference points on the screen and continually adjusts the brightness, contrast or color of portions of the picture. Sony also has a hand-held computer that uses fuzzy logic to recognize the user’s handwriting.
A number of Japanese auto makers are using or planning to use fuzzy logic circuits to control automatic transmissions and even engines themselves (to reduce pollutant emissions), according to Erik Horstkotte, chief software engineer at Togai Infralogic in Irvine, a leading U.S. producer of fuzzy software and hardware. One early automotive example is the Subaru Justy, in which a fuzzy controller continually adjusts the car’s transmission-pulley ratio to optimize performance in any driving situation.
American auto makers, like U.S. industry in general, are less enthusiastic than their Japanese counterparts.
“I firmly believe that until a major U.S. corporation, like an IBM or a DEC (Digital Equipment Corp.) announces the use of fuzzy logic in a product, other companies will be reluctant to use it,” said Welch, of the Gartner Group.
However, more research on fuzzy logic may be under way in the United States than is generally acknowledged. A Fortune 500 firm, which Welch declined to identify, spends about $500,000 annually on fuzzy logic research and development, she said, adding that other large companies are also doing significant work in private.
Indeed the U.S. vanguard, if there is one, is in Southern California. Besides Togai, Rockwell International Corp.’s Science Center in Thousand Oaks is considered a leading developer of fuzzy systems. The company has developed controllers for manufacturing lines and industrial processes and is working on a fuzzy system to control the flight surfaces of an experimental aircraft, called the Advance Tactical Wing wind tunnel model. Computer simulations show that the aircraft would fly in a more stable way and be able to roll into turns more quickly with the fuzzy controller, according to Allen Firstenberg, director of information sciences at the center.
Other Southern California institutions nurturing the technology include Fuzzy Ventura Partners in Santa Monica, which plans to invest $70 million in start-up companies working with fuzzy logic. At USC, research is being led by Bart Kosko, an assistant professor of electrical engineering.
Even with its many advantages, fuzzy logic is considered a supplement, rather than a replacement, for more conventional control systems. In fact, extensive research is aimed at finding ways to merge fuzzy logic systems with either more conventional computer techniques or with expert systems, a class of computer applications that have much in common with fuzzy systems.
“One of the problems facing us in technology is that some parts of the world we understand very well mathematically. But other things, we can’t build a mathematical model for, because they’re too complicated,” Yager said. “With fuzzy sets, we’re allowed to combine these things: We can take the things we understand mathematically and combine them with the things we understand heuristically, or intuitively. And that’s a very, very big step.”
