Preface: If this piece is anything, it is preachy. It is in some ways too easy to criticize the author's use of heavily laden emotional, psychological, or other connotative meaning. I will try to leave such comments to others, though it is burdensome to proceed without stating that words like "clearly" and "obviously" are in extremely poor taste. We all use such words, is true, you are of course foolish to think otherwise."
Braitenberg really makes two progressions thorough the chapters in this book. The surface transition goes from simple to more complex mechanisms, while the underlying transition goes from complex behaviors to unbelievably complex behaviors. Being able to construct such (possibly) unwieldy behaviors is extremely exciting, but at the same time is disturbing. Though _creating_ behavior is easy, _understanding_ behavior is extremely difficult.
Braitenberg brings this point home at the beginning of Chapter Five, when he discusses the challenge of analysis, the induction of form rather than the deduction which follows from mechanism and 'policy'. I disagree with his 'psychological' consequence of overestimating complexity of form--we create models that are 'good enough' to explain behavior to the extent that we understand it, the complexity we attribute is likely to be more cognitively minimal, perhaps not mechanistically minimal. Some 'real' evidence would be greatly appreciated. Starting from a black-box abstraction, the best we can do is infer some model, regardless of how complex, and use this model until observation doesn't fit the model's expectations. Even if simple models are more appealing, there is little help here to create such simple models without 'popping the hood.'
We might hope that by studying these simple mechanisms we will gain some understanding of how to deal with complex systems. Hopefully dealing with these simple systems will make analysis of complex systems easier. Unfortunately, this book provides little evidence of that. In fact, by the later chapters, even deducing the behavior of the complex machines has gotten to be extremely difficult. Having a mechanism to simulate and play with such behaviors (the Braitenberg microworld?) becames important to even deduce the behavior.
But analysis remains a difficult task, and Braitenberg's use of emotionally laden words emphasizes this fact. Though we might have good ways of explaining simple phenomena and mechanism, our ability to describe complex behavior is extremely poor. People who try to precisely specify the behavior of computer systems often observe that complete explanations are prohibitively difficult. Braitenberg's own discussions get increasingly vague to the point that the analysis of the eight vehicle involves believing his arguments 'on faith' than it does about reasoning about them yourself.
One exciting challenge of this work is to create intersting complex things from simple mechanisms. An equally exciting challenge is to find ways to understand complex things, perhaps through the use of combining simple analysis tools.
For me, the significance of "Behavior Construction Kits" was pounded home when reading "Vehicles." What better way to physically capture some of the concepts of Vehicles. It is easy for me to see how programmable lego bricks could be used to construct any of the Vehicles (up to Vehicle 8). In fact, a Cricket with two light sensors and two motors would be Vehicle 1. If the Cricket were placed inside a "black box" and a person (perhaps one of Braitenburg's philosophers) were to observe it's "behavior," he/she might use terms like "fear" and "aggression" to describe its behavior. The fact that such a Vehicle would be physical (touchable) emphasizes the "behavior" of such a creature much more concretely than Braitenburg's paper examples.
I must admit that "Vehicles" made me squirm with some discomfort. In reading each chapter, it was at least conceivable that Braitenburg's conclusions were true. For instance, that Vehicle 1 would seem to be "fearful" or "aggressive" to an external observer or that type 6 Vehicles could be selected out of Darwinian evolution. But when examining the conclusion at the end of chapter 8, I have this distinct feeling of discomfort. I am perhaps one of Braitenburg's skeptical philosophers. Certainly a Vehicle (even a physical Cricket based Vehicle), would appear to encapsulate complex behavior despite having a simple underlying mechanism. But to simply examine behavior without careful examination of mechanism seems to me to be only half of the equation. For instance, the fact that Deep Blue plays chess at the Grand Master level doesn't imply to me that Deep Blue is "intelligent." Yes, it does possess an intelligence of a sort; after all, it can outplay Gary Kasparov. However, wouldn't it be informative to understand Kasparov and Deep Blue "think" differently? The mechanism itself - how Kasparov can play so well despite having a fraction of the searching ability of Deep Blue - is as much the point of study as the behavior itself.
In other words, I think that "Vehicles" makes a strong case for studying behavior independently of mechanism, but I would claim that mechanism is at least equally important. It is interesting and informative to be an external philosopher observing the behavior of a Vehicle (even a human one), but understanding the "why" and the "how" adds to our understanding.