Book Review: "Cybernetics: Second Edition: Or the Control and Communication in the Animal and the Machine" by Norbert Wiener

People don't often know the origin of Cybernetics, but it started as a branch of control and communication engineering. It's foundational revelation is the principle that, when you reach for a pencil, your hand isn't carrying out a predetermined set of instructions, as a standard program might. Instead, you have a goal-directed action, your brain measures how far away from your goal, and keeps adjusting the course of your hand and fingers in the direction it thinks will bring it closer to picking up the pencil, until the goal is achieved. The evidence for this is particularly interesting, as quoted from the book:

"A patient comes into a neurological clinic. ... he walks with a peculiar uncertain gait, with eyes downcast on the ground and on his legs. He starts each step with a kick, throwing each leg in succession in front of him. If blindfolded, he cannot stand up, and totters to the ground... Another patient comes in. While he sits at rest in his chair, there seems to be nothing wrong with him. However, offer him a cigarette and he will swing his hand past it in trying to pick it up. This will be followed by an equally futile swing in the other direction, and this by still a third swing back, until his motion becomes nothing but a futile and violent oscillation. Give him a glass of water, and he will empty it in three swings before he is able to bring it to his mouth."

Both of these patients are inhibited because the feedback loops are not working properly. The first patient had the sensory nerves carrying information from his legs eroded by late stage syphilis, removing the necessary senses provided by the legs and feet regarding their position. He can walk as long as he can substitute sight for the missing senses, but it is still clumsy. The second patient suffered damage to the cerebellum, and as a consequence his feedback mechanism would constantly overcorrect until the attempt becomes useless. Both issues are a form of ataxia, the second one is also called purpose tremor.
Cybernetics involves applying this concept to machines, which are called servomechanism. A simplified version of this is a thermostat, which has a goal (a specific temperature) in mind, measures the temperature and when the current temperature differs from the desired temperature uses either the AC or heater in order to approach the desired goal. Other examples include guided missiles, and self-steering ships. In fact, Clerk Maxwell wrote an article in 1868 outlining what he called a 'governor', as a proposed machine that tracks a compass compared to a desired direction and adjusts the rudder if the ship starts to go off course. Cybernetics is named after the governors in Maxwell's paper, from the greek word χυβερνήτης‚ for steersman. When I first learned this, I was a bit sore that a word representing such a beautiful concept about human motor perception and applying it to technology, would later come to mean 'guy with robot arm.' However, the core developments of cybernetics seem to have been appropriated in the study of artificial intelligence and machine learning and it was speculated early in the book that one of the primary applications of cybernetics would be in advanced prosthetics, so it's not entirely unfair that the word has its current meaning.

All of the above information was readily available in the preface and introduction of the book. The first chapter, "Newtonian and Bergsonian Time" talks about two different ways of looking at time that are relevant in different fields of science. Astronomy gives a good example of newtonian time, where if you were to reverse it, the planets would all be orbiting the sun at the same speed, but in the opposite direction. Had time been reversed from the beginning, no one would notice, they would just think the planets were supposed to go in that direction. The same could be said for most objects moving according to newtonian mechanics, but if you look at another field, say, meteorology, to reverse time would produce absolutely bizarre behaviour. Bolts of ligtning would split apart into the sky and ground, as opposed to coming together. Clouds would shrink and disappear where they would normally form and grow. In these fields, time behaves more like a vector than a two-dimensional line. In Bergsonian time, future and past are more distinct, and can often be less predictable. Thermodynamics also follows Bergsonian time, and by definition, fields dependent on it must do so as well. This distinction becomes more relevant in future chapters, as the ability to compare past and predict the future will be necessary in cybernetics.
The second chapter 'Groups and Statistical Mechanics' introduces concepts in that field and some of the math for it. However, the math in this book is very heavy on symbols that are not explained. Most of the math in this book went over my head, and so this chapter, the third chapter: 'Time Series, Information and Communication' and the fourth chapter: 'Feedback and Oscillation' are mostly in mathematical form. The overall direction of the three chapters is to introduce the equations and analysis associated with statistical mechanics, apply it to time series (so your machine can look into the past and determine if there's a trend where it's going off course), and avoid the oscillations that occur from overcorrection. Within the fourth chapter, there is an analysis to look at the problem of Maxwell's governor, to see what are the physical limitations of a system that controls the rudder (can the ship get so far off course that it can't sufficiently correct? Is there a threshold so small that any attempt to fix it will result in an overcorrection and get the ship further off course?). The idea is interesting if you can follow it.

The fifth chapter 'Computing Machines and the nervous system' looks at the possibility of building machines capable of learning information and figuring out solutions to problems without being explicitly programmed. The fact that this book was first written in 1946 becomes highly relevant in this chapter, as most of the technology is speculative and has been vastly surpassed many decades ago. More modern texts on machine learning would likely be more useful, but if you wanted to see the subject from a historical perspective then this chapter may be amusing. Chapter six, 'Gestalt and Universals' deals with the possibility of a machine capable of human sight. It is equally anachronistic, but the scanning method it describes involving cathode ray tubes leads me to wonder if some of the relevant hardware is still employed in electron microscopes.

Chapter seven: 'Cybernetics and Psychopathology' applies our knowledge of computing machines to the nervous system to see if there is any enlightenment that can be drawn from the knowledge of our physical hardware and again, the fact that it was written in the 1940's is relevant. It's interesting to see an endorsement of electroshock therapy on the grounds that some machines can suffer errors in programming or bad data that consumes all of the systems resources and prevents it from functioning properly, and the capacity for electroshock therapy to remove some local memories may be a way to shift a human out of a similar state. The chapter also looks at the neurons that make up the human brain relative to other animals: ours are shorter in length, allowing for more networking on local levels and more folds within the brain at the expense of communication across hemispheres, and a speculation that as a result, bilateral functions (like being right-handed or left-handed) are far more pronounced in humans than in the 'lower' animals, and damage to the dominant hemisphere in an adult human is far more debilitating than the same damage applied to the dominant hemisphere of, say, a cat. It was still a very fascinating chapter to read in this day and age.

The final chapter of the first edition 'Information, language and Society' dealt more from the author's perspective of politics and economics, and the limitations of humans based on the centralized means of communication. I was not as impressed with this chapter, but it was still not a bad read. According to the author the chapter was included to dissuade his colleagues in thinking that Cybernetics could revolutionize the way human society functions.

The second edition of the book was written in 1961, and consists of two chapters appended to the end of the first edition. The first of them 'On Learning and Self-Reproducing Machines' takes a second look at the subject of Machine Learning. The following chapter 'Brain Waves and Self Organizing Systems' gives a good description and explanation of electroencephalography, and also discusses possible implications it has on the human perception of time (suggesting, for instance that the brain runs at cycles of roughly 9-10Hz). All in all, i think the fundamental premise of the book is accurately captured both in this review, and is also in any book you may find on machine learning. Nevertheless, it was a good read, and likely would be a far better one if you were familiar with statistical mechanics calculations, and/or the math associated with information and networking theory.

Get the book here!