Handbook of Artificial Intelligence, Vol. 2

This is the second volume of a 3-volume set summarizing what was known at the time in the field of artificial intelligence (A.I.) Most of the content of this book is now called GOFAI, for "good-ole fashioned artificial intelligence", but many of the algorithms and concepts discussed in this book are still used today, making this book still useful. This volume is centered more on applications of A.I. to the sciences, robotics, and language translation. In the introduction to the first volume one can find the following statement, which was true back then and dramatically more so at the present time: "There is every indication that useful A.I. programs will play an important part in the evolving role of computers in our lives-a role that has changed, in our lifetimes, from remote to commonplace and that, if current expectations about computing cost and power are correct, is likely to evolve further from useful to essential." One can at the present time project this statement out into the near future with confidence, due to the accelerating advances in A.I. that have taken place since this volume was written. Indeed, there is a collection of A.I. researchers that believe that intelligent machines will surpass human capabilities by many orders of magnitude by the end of the next thirty years. This is an extremely optimistic prediction but it looks hopeful. The prospect that intelligent, autonomous machines will be living among us very soon is indeed an exciting one. Chapter six discusses the programming languages used in A.I. research. The language LISP was the predominant one used at the time of publication, but now PROLOG has made significant inroads, but is only briefly discussed in the book. LISP and a few other programming languages are mentioned in the book, and a comparison of their strengths and abilities to do certain tasks is done. I have only used LISP and Prolog so I cannot speak of the capabilities of the other languages discussed. The power of LISP to do recursion is one of its strongest points, and this chapter emphasizes this, along with its ability to parallel control and data structures, and its ability to think of programs as data. There is also an interesting discussion in this chapter on "dependency records", which are used to keep track of the steps taken in a reasoning program. An intelligent program or machine must be able to explain their conclusions in terms of the information given. In chapter seven one finds a fairly extensive discussion of how A.I. has been applied to scientific research. My first project in A.I. thirteen years ago was to build an expert tutor for physics students and this chapter assisted me to a large degree in accomplishing this. The examples given in the chapter are of course at the present time way out of date, since an explosion of A.I. applications to science and medicine have occurred in the last decade. For example, the chapter mentions MACSYMA as being a package to do symbolic integration. The

This is the first volume of a 3-volume set published in the early 1980's and thus could be thought of as a summary of what was known at the time in the field of artificial intelligence (A.I.). Now sometimes referred to as "GOFAI" for "good ole-fashioned artificial intelligence", this set of books can still be referred to profitably by anyone curious about the applications of artificial intelligence. Indeed, many of the algorithms discussed in this volume are still being used, and very robustly, in current implementations of artificial intelligence. A lot has happened since this volume was published, especially in the area of chess playing and logic programming, but there are many sections of the book that are still up-to-date. After a brief introduction to A.I. in chapter one, chapter two overviews the use of search algorithms for intelligent problem solving. The emphasis initially is on the problem representations that form the basis of search techniques, such as state-space and problem-reduction representations. Game tree representations are also discussed. The algorithms that implement the problem representations are then treated. If the search space is viewed merely syntactically, these are called "blind search" algorithms, which are distinguished from "heuristic" methods, which exploit various structural information about the problem in order to limit the search. Examples of blind search methods that are discussed include breadth-first, uniform-cost, depth-first, and bidirectional search. Examples of heuristic methods discussed are ordered state-space, bidirectional, and the famous A*-algorithm, the latter of which is still finding considerable use in new applications of A.I. Examples of game tree search that are covered include the minimax procedure, the negmax formalism, and alpha-beta pruning. There is discussion on the use of heuristics in game tree search, but this part is out-of-date due to the advances made in chess playing, checkers, etc, since this volume was published. Chapter three is an overview of knowledge representation in A.I. The author takes a pragmatic approach to the nature of knowledge and intelligence, and defines the "representation of knowledge" as a combination of data structures and interpretive procedures that will lead to what he calls "knowledgeable" behavior. A book needs a reader before it could be considered knowledge, argues the author. He calls this whole enterprise "experimental epistemology" , which endeavors to create programs that exhibit intelligent behavior. The chapter gives an overview of the knowledge representation schemes used in A.I. and discusses their uses and shortcomings. Also, the tension between the advocates of declarative versus procedural knowledge representations is discussed. Declarative systems are more logical/mathematically based, and were exemplified by theorem-provers based on logical resolution. The procedural approach emphasized a more directed approach to the problem of infer