Computers Are Precise - That's Their Problem
THE trouble with computers, say Daniel McNeill and Paul Freiberger in their engaging new book "Fuzzy Logic," is that they're programmed as yes-no, on-off, black-and-white machines in a complex world where nothing is so clear-cut. What computers need, the authors say, is a healthy dose of fuzziness.
Fuzzy logic sprang into being 29 years ago when Lotfi Zadeh, an Azerbaijani-born professor at the University of California at Berkeley, discovered it in a flash one evening. Fuzzy logic confounds conventional logic because it deals with set theory in a new way, the authors say.
If you had New Math, you may remember sets. Imagine a set of men. Jim is 6 feet, 6 inches, Tom is 5 feet, 11 inches, and Bob is 5 feet, 9 inches. Draw a circle around the men who make up the tall men set. Clearly Jim belongs in the set. Does Tom?
Conventional logic forces you to create a definition of tall. If it's 5 feet, 10 inches, then Tom is in the set. If it's 6 feet, he's not.
The problem with this logic is that its precision is artificial, the authors argue. The line that separates tall men from others is not so clear in the real world. If Tom is tall, then Bob, who is only two inches shorter, is pretty close to tall, too.
Even the ancient Greeks questioned conventional logic with their famous paradox of the heap. Remove a grain of sand from a heap, and it's still a heap. Remove another grain and another, and the heap remains. Eventually, though, there is only one grain left. Is it a heap? If not, at what point did it stop being a heap?
Zadeh said objects could be partial members of a set. At some point, the Grecian sand pile becomes a 0.5 member of the set of heaps. Tom is maybe a 0.6 member of the set of tall men. Even Bob, at 5 feet, 9 inches, retains 0.4 membership, while Jim is almost a full member at 0.95.
Such ideas contradict a long line of logicians and mathematicians who claim that A can't be both B and not-B (Aristotle's Law of Contradiction) and that A has to be either B or not-B (Aristotle's Law of Bivalence). Making up mid-points, say critics of fuzzy logic, lacks scientific rigor.
McNeill and Freiberger, who write about computers, quote these critics at length and try to answer them in a forceful way. They largely succeed. But one can't help feeling at certain points in the book that maybe the critics are onto something. Some of the examples of fuzzy theory the authors cite sound a little loose.
Opposed to this, however, is fuzzy logic's success. Japan has already moved five years ahead of the United States in designing fuzzy products. These range from fuzzy rice cookers to subway trains to auto-focus cameras.
In February 1990, Matsushita started selling a fuzzy washer - a one-button appliance that senses the clothes' dirtiness, the type of dirt, and the volume of the clothes, then evaluates a series of fuzzy IF-THEN rules. (An example might be, "If the water is very dark, then increase the water temperature.") The controller picks one of 600 procedures that vary water volume, strength of flow, and washing time.
Some of these products are more marketing hype than real. But supporters foresee more use of fuzziness and the linkup with another cutting-edge technology, called neural networks, to create systems far more advanced than artificial-intelligence machines.
US companies are just starting to evaluate fuzzy logic. This book suggests that these companies don't have much time to catch up with Japan.