Browse Books

Reviewer: Robert L. Glass

While this book is excellent for its intended audience, the question any potential reader must ask is, Am I part of that audience__?__ Before I help you answer that question, let us back up and ask, What is programming by demonstration (PBD)__?__ It is the act of getting a computer to do what you want it to do without explicitly writing a program in the traditional way. Instead, the PBD user does a few example tasks, and an underlying PBD system picks up the pattern<__?__Pub Caret>s in those examples and constructs a program to perform them. Clearly, this technique is of enormous interest to end users, and considerably less interesting to those who already know how to program. Thus a first answer to the question, Am I in that intended audience__?__ would seem to be, Yes, if I am an end user. That answer would be wrong, however. This book was written by researchers in the area primarily for other researchers in the area. In other words, you are not a part of the intended audience for this book unless you are one of those researchers, or you find the topic particularly fascinating and would like to know more about the state of the art and practice of PBD. Another problem arises here, however. Although the book describes 18 examples of PBD systems, they are all research explorations. One author says that “PBD is still waiting for a practical system to show its validity.” So the intended audience does not include people who want to know about the state of PBD practice—none exists. Given that the book is focused on an audience of researchers and those interested in the state of the research, is it a good one__?__ The answer is a resounding “yes.” The book summarizes the state of the art. It is well written and well integrated, and does an excellent job of what it set out to do. Its comparisons of the existing PBD systems nicely put them in a proper context. The book even does a reasonably effective job of presenting the pros and cons of PBD.<__?__Pub Fmt eos-space>Of course, the material is predominately pro (the most important advantage is that PBD supports something one author calls “just-in-time programming,” where the person with a task to perform can learn to “program” while performing the task). PBD has strong disadvantages as well. It is not clear that the method will scale up well from the toy problems to which it has been applied. It does not have a good mechanism for providing those solutions to exceptional cases (one author refers to them as “unforeseen special cases”) that traditional programmers know are an essential part of any programmed problem solution. Also, in many cases, the PBD system must guess the pattern intended by its user, and may guess wrong. One other issue is important here. We in the computing profession have seen the word “programming” used in so many different contexts that it is beginning to lose its meaning. For example, is it programming to use a spreadsheet__?__ Is it programming to write a macro for a spreadsheet__?__ In the context of this book, is it programming to write an example series of (usually fairly simple) steps that a system turns into a program__?__ The terms “programming by demonstration” and “just-in-time programming” tend to imply that this book is about programming. Is it__?__ I would say “no.”