By Bart Kosko
Every macroscopic system is subject to noise, from the random fluctuations of its environment as well as its own internal fluctuations. If, like Bart Kosko, you consider things such as e-mail spam and extraneous signals and objects to be noise, then your immersion seems even more overwhelming. Noise is not a trivial detail, and in general it cannot just be ignored on the grounds that it will somehow average out. In the case of nonlinear systems, far from averaging out, it gives rise to a diverse range of important phenomena, such as stochastic resonance.
In Noise, Kosko takes the topic of noise, using the widest possible definition, and tries to make it accessible and interesting to the general reader. He opens with his central thesis that noise is “an unwanted signal” and develops ideas from information theory of noise in digital signals. The next two chapters cover noise in the everyday sense of unwanted acoustic disturbances. There is anecdotal material about unneighbourly behaviour, litigation, crematoria, noise from aircraft, the law on trespass, damage to hearing, noise-induced stress, and undersea noise involving the US Navy, dolphins and whales.
Much of the hard science, and its exposition in simple terms, is consigned to the (long) fourth chapter. Kosko covers a remarkably wide range of topics, from cosmology and photosynthesis to white and coloured noise, and the central-limit theorem. This is followed by a chapter including Fourier techniques and spread-spectrum encryption. The book ends with a discussion of stochastic resonance, an interesting phenomenon in which noise plays a creative role, rather than its usual destructive one. Here the addition of noise to a nonlinear system can cause amplification and enhance the signal-to-noise ratio of a signal passing through the system.
Each chapter is prefaced with ten or so intriguing and provocative quotations that occupy several pages. There are also more than 60 pages of notes at the end of the book, arranged by chapter, expanding on the material in the main text. Many of them are useful and interesting, but in the end I got tired of continually turning pages to find them; they would have been far more convenient as footnotes.
Has Kosko succeeded in his aims? I believe he has, to a large extent, although in one or two places he illustrates the G. K. Chesterton dictum that “He who simplifies simply lies” by conveying a misleading impression. For example, few readers will appreciate that stochastic resonance can do nothing to enhance the signal-to-noise ratio of a given signal. It can certainly ameliorate the information loss that otherwise occurs when a signal passes through a nonlinear system, but I very much doubt whether readers will understand this fact from Kosko's discussion. I suspect, though, that schoolboy howlers such as “The brain consumes about 20 watts of power each day” are not from the author's pen.
I am surprised that Kosko omits all mention of optimal fluctuational paths, given their conceptual simplicity and the nice way they link together many of the other ideas he presents. Most of the interesting and important events in noisy systems involve such optimal paths, including chemical reactions, mutations in genes, and failures of lasers and electronic devices. In all these cases, the system fluctuates near an attractor of some kind for a long time, and then travels along an optimal path to a different attractor. Remarkably, despite the noisy driving force, these paths are deterministic in character.
While accepting the author's broad view of what constitutes noise, I also feel that some of his writing introduces its own noise, through the unnecessary introduction of a multiplicity of distracting ideas that are tangential (or irrelevant) to the point being discussed. What is certain, however, is that every reader of this book will end up learning something new and interesting.