Book Review - The Sciences of the Artificial

The Sciences of the Artificial-Herbert A. Simon, (Cambridge, Mass.: M.I.T. Press, 1969, 123 pp., $5.00).

This fascinating little book is based on three Karl Taylor Compton lectures given at the Massachusetts Institute of Technology in the spring of 1968, and augmented by a reprint of the author’s The Architecture of Complexity, which appeared in the Proceedings of the American Philosophical Society in 1962. While by no means going into formal detail, this book benefits from the penetrating perspective of a man who has made his mark in the study of economics and business administration, as well as in psychological processes and their simulation. It studies those phenomena that, are “artificial” in that they pertain to systems that are molded to their environments by goals or purposes, and thus might well be other than they actually are–since design involves a choice between many apparently satisfactory alternatives. Thus, the themes of design, psychological activity, and evolution are interwoven, with the notion of hierarchy playing a crucial role in giving us a handle on the complexity that arises when a system must be adapted to a complex environment. Such complexity will clearly be an increasing challenge for the information theorist, even though it will lead him farther and farther away from the comfortable one-dimensional measures in the original approach to information theory that grew from the work of Wiener and Shannon in the late 1940’s.

The first chapter, “Understanding the Natural and the Artificial Worlds,” introduces four indicia that distinguish the artificial from the natural.

Artificial things are synthesized (though not always or usually with full forethought) by man.
Artificial things may imitate appearances in natural things while lacking, in one or many respects, the reality of the latter.
Artificial things can be characterized in terms of functions, goals, and adaptation.
Artificial things are often discussed, particularly when they are designed in terms of imperatives as well as descriptives.

This idea of the artificial does not only apply to the machines or objects designed by man, but also applies to human problem- solving in which in some sense one must psychologically “de- sign” a solution, and also-recalling the oxymoron of “natural selection”-to evolutionary processes, so long as one handles one’s teleology carefully. Especially in chapter 2, “Psychology as a Science of the Artificial,” Simon emphasizes a division of the outer from the inner environment, and is much more concerned with finding possible structures that could subserve the observed input-output behavior of a complex system. than in trying to find the exact internal structure of L the Y system. On page 54 he says:

Our analysis of the artificial leads us to a particular view of the form that the physiological explanation of behavior must take. Neurophysiology is the study of the inner environment of the adaptive system called Homo Sapiens. It is to physiology that we must turn for an explanation of the limits of adaptation. Why is short-term memory limited to seven “chunks?” What is the physiological structure that corresponds to a chunk? What goes on during the five seconds that the chunk is being fixated? How are associational structures realized in the brain?

Such a viewpoint is reached after studying interesting protocols from humans solving problems that can be expressed at the verbal level, or in the manipulation of small sets of symbols as in the game of chess. This, of course, is the sort of work that has been so interestingly simulated on the computer by Newell, Shaw, and Simon in their GPS-the so-called General Problem Solver program-which has not exhibited great intelligence, but has shown that methods of directed search through a “space” of possible solutions to a problem can, if guided by heuristics, yield interesting human-like behavior. While not questioning the value of this work, I would note that it emphasizes a bias that is current among many people today in thinking about the human mind-namely, the author asks questions only about symbol manipulation, and so lists constraints upon the mind which are related to that symbol manipulation. If we teach someone that he must make a different response to each of a small set of stimuli, we may find that he seems to be rather limited in the number of responses he can make. If, however, one of the responses is to actually touch a certain object when it is presented, then, in fact, we will find that without any extra trouble the subject can touch that object no matter where within reach it is located, and if we think of how finely the spatial environment can be subdivided, this immediately gives a far larger measure of information rate than that based on looking at symbol manipulation alone. I believe that when we take into account this “action-oriented” rather than purely symbolic human behavior, we shall find that physiology be- comes far less of a weak boundary condition, and far more a central organizing principle. For instance, the principle of somatotopy is, I believe, the key to understanding the organization of the brain and those consequent features of the human mind that can not be understood by only taking into account the purely verbal stream that we tend to sample in looking at symbolic problem-solving behavior.

In chapter 3, Simon turns to the question of establishing “The Science of Design: Creating the Artificial,” for he notes that engineering, architecture, and business schools, whose central task is that of design, revolted some years ago against a “cook-book” approach to their subject, and started to gain technical mastery and increased respectability by turning over more and more of their courses to the basic sciences. He argues that an unfortunate side effect is that students in such schools now have very little teaching of how to bring together the various techniques they have learned from basic science in the creative solution of design problems. Simon thus argues cogently in this chapter for setting up actual courses in design, and concludes by drawing up a curriculum in design-in the science of the artificial-to take its place by the side of natural science in the whole engineering curriculum. He lists the following topics: the evaluation of designs, the formal logic of design, the search for alternatives, theory of structure and design organization, representation of design problems.

The final chapter, “The Architecture of Complexity,” is the 1962 paper we mentioned in the first paragraph. I strongly urge readers of this review to read at least this last chapter, even if they do not have time to read the whole book. Although it would certainly have been pleasant to see the article updated in the intervening seven years, it nonetheless remains a classic discussion of the way in which hierarchical organization can help us understand extremely complex structures. Rather than give a detailed discussion of this chapter, let me paraphrase the parable that is at the center of the chapter and then briefly mention some of the topics that are built upon the parable. We are told of two watchmakers, each of whom makes watches from a thousand parts. However, one of them makes his watches by assembling subsystems of ten parts, then putting together ten of these subsystems to form a larger subsystem, and then finally taking these ten large subsystems and joining them into the finished watch. Whenever the phone rings and he picks it up, the assembly he is working on disassembles, but since he never had more than ten pieces joined together, making up for lost time is relatively simple. However, the other watchmaker does not use hierarchical design-and every time he puts aside a watch he is working on it disassembles to the smallest components. He completes very few watches. Simon thus argues that whether we are looking at the evolution of animals, or the progress of human problem-solving or design, if useful solutions are to be achieved in reasonable times, then some sort of hierarchical organization seems almost necessary. As a particular formal expression of such hierarchical arrangement, he introduces the idea of a nearly decomposable system, in which we may distinguish subsystems within which there are strong interactions, but between which there are only weak inter- actions. He also discusses the problem of finding simple descriptions of complex systems, and contrasts the state description with the process description. On this basic, he raises the intriguing chicken and egg question: “are we able to under- stand the world because it is hierarchic or does it appear hierarchic because those aspects of it that are not elude our under- standing and observation?”

To those who would represent complex systems in such a way as to achieve understanding, this book can be recommended as a cheering introduction. It is only a beginning.

ChangeLog

2018-07-21 李亮创建