Browse Books

Reviewer: Michael W. Whitelaw

When a giant like Herbert Simon writes a reflective book on the ontology and epistemology of computing, one can expect a feast of ideas. The book is the combination of two lecture series with an earlier essay and a new chapter, for a total of eight chapters. This is the third edition, so one could anticipate a refined understanding, and indeed Simon has made a number of changes in the existing chapters to reflect the development of his views. In the first chapter, Simon lays out the topic—the science of the artificial. He defines artificial things as being synthesized (designed) by humans, and says they may be characterized in terms of their functions, goals, and adaptation. By distinguishing between the internal structure and the external behavior of an artifact, Simon makes the point that it is not easy (and may not be possible) to predict the behavior from the structure, or vice versa. The behavior is the subject of theorizing and experimentation almost independent of the internals of an artifact. Only when the limitations of the internal structure affect the desired external behavior do we become aware of the structure. Towards the end of the first chapter, Simon makes the assumption that the human mind and brain are in the same symbol-manipulating class as computers, and he introduces the hypothesis that intelligence is the work of symbol systems. In chapter 2, Simon introduces the idea that humans have a rather limited ability to think and that the complexity of human behavior is primarily a reflection of the complexity of the environment. In particular, he argues that economic behavior is the result of the environment rather than being planned; in particular, markets do not need the rational buyer or the rational seller. Simon uses game theory to support the assertion that normally, markets work simply because humans have bounded rationality. When markets do not work, organizations arise to overcome the bounds of the limited rationality of humans in dealing with these situations. Having set up the idea that humans have bounded rationality, Simon expands and quantifies the time and space limitations of human thinking as shown by psychological studies (chapters 3 and 4). Memory is essentially a simple environment for thought. To overcome the limitations of memory, humans store verbal, visual, and linguistic information in a way that is consistent with a hierarchical list. Simon argues that these ways of remembering are artificial constructions, that is, they are constructed for goals and can be understood by examining computer programs that mimic learning and discovery, because these programs can produce results that are similar to human learning and discovery. At this point Simon addresses artificial science—the study of the design of the artificial. One useful technique in this chapter is the specification of a curriculum in the science of the artificial. He notes that the two central topics in his curriculum will be utility theory/statistical decision theory and the techniques for deducing the optimum alternative. He argues that the shape of the design process will be hierarchical and uses examples to show that apparently nonhierarchical methods can be represented as hierarchical. He believes that interdisciplinary cooperation relies on people accepting the idea that knowledge comes from the separation of inner and outer behavior. Chapters 6 and 7 discuss designing a society to maintain standards of living. Simon argues that the goal should be flexibility, because there is difficulty in determining the clients, setting a suitable final goal, getting data for planning, and predicting the future. Indeed, planning should involve an understanding of the effect of chaos on a design. Chapter 8 again takes up the theme that hierarchical systems arise, not from design, but because they are efficient. Examples from social systems, biological and physical systems, human symbolic systems, and problem solving are given. Simon notes that large systems that do not appear to be hierarchical, but (for example) linear, can be seen as hierarchical with a span of one. Simon raises significant questions about the ontology and epistemology of design and, in particular, the design of computer programs. I found this book challenging. However, because it is derived from several diverse sources, not all topics are treated at the same intellectual level; the preface calls the variation a fugue. Again, because of the diverse sources, the author often covers similar ideas in two different chapters, when combining the material would have made it easier to read. Indeed, it is not difficult to find examples of apparent contradiction. Very careful reading (I read the book closely four times) usually removes the contradiction, but not always.