Fuzzy Thinking: The New Science of Fuzzy Logic
by Bart Kosko

Jul 2001

Fuzzy vs Concrete

Before going out on my own, I worked in an office where the music was always too loud or too soft. The problem was the volume control mechanism embedded in the wall. At setting 1, the music was audible but too low to hear. At setting 2, the music was loud enough to disrupt concentration and made you jam the phone against your ear. I longed for a round knob control that offered a continuum, but alas, had to be satisfied with 1 or 2.

This throwaway analogy illustrates Kosko's frustration with "bivalence." We do not live in a black and white world; many things are in between, and science is silly to pretend otherwise. It is not just the engineers and physicists, of course. Economists also suffer greatly from this delusion (largely because much of economics is sociology dressed up in fancy math). Reality is messy, complex, and does not fit into a box. A simple statement at heart, but one with profound implications that Kosko does a decent job of exploring.

With the statement "precision up, fuzz up," Kosko reiterates a truth known since biblical times. I like T.S. Eliot's version, "All our knowledge only brings us closer to our ignorance." The thrust is that excess information often obscures as much as it reveals. Relevant facts are lost in the wash of noise, like needles in the proverbial haystack. And even the relevant facts are subject to interpretation, counterweighting, and levels of degree.

I felt that the systems Kosko described were not "fuzzy" as much as "flexible." They still make black and white decisions, they just make a higher number of decisions and thus have a more averaged or "fuzzy" result. A machine that utilizes fifteen data points will be much more versatile and flexible than one that only uses two, but it is still a binary system at heart. Maybe our brains are binary as well, except with gazillions of data points to access. But that is a subject for another book (one by Penrose perhaps).

Can computers optimize a set of rules, like those that apply to washing machines or traffic lights? Yes. Can they predict the future outcome of the stock market? No.

The actions of a chaotic/reflexive system can never be predicted with true accuracy, because the stream of variables is too fast and furious to follow. Even for a supercomputer, it would be like drinking from a fire hose. Thousands of opinions converge and diverge every second. Unforeseen events jar the market like random electrical shocks. Feedback amplifies itself, the way a sound looped from speaker back to microphone becomes an ear splitting shriek. Feedback diffuses itself, like soft spring rain on a pond. The market measures itself and reacts to itself a thousand times a day, a living embodiment of the Heisenberg principle. The puzzle has a trillion pieces that spontaneously rearrange at will. You can stump a computer with the turbulence in a glass of water, and yet it's supposed to decipher mass interaction? I'm not holding my breath.

As the book progresses, Kosko gets more and more wildly optimistic in his predictions of what fuzzy logic might be capable of, what fuzzy machines could do. He goes out on a limb and overshoots his premise by a country mile by the time you reach the last page, filling the last half of the book with rabbit trails, but that is okay by me. Fuzzy logic is his baby, his contribution to the world. So it is completely natural that he would bubble over with possibilities and offshoots of his original thesis. Excitement gets your brain going. Separate the wheat from the chaff and enjoy.