This document reviews Margaret Boden's book "Artificial Intelligence and Natural Man". The review provides the following key points:
1) The book provides a comprehensive overview of AI research up until its publication, though it lacks technical details that could have helped explain some concepts.
2) Boden focuses on discussing individual AI programs to illustrate concepts, choosing Colby's neurotic belief system to represent early work.
3) The book's discussion of vision systems is disappointing due to the exclusion of important recent developments in the field.
4) Boden's work provides a useful overview of AI concepts for students and others without a technical background, though some technical details could have further aided understanding.
2. 268 BOOK REVIEWS
only one program in considerable detail. The choice of the distinguished program
must have posed real difficulties: it needs to be on a topic of considerable interest
to her audience, interesting from an AI stance, and yet understandable. She chose
Colby's early system modelling neurotic belief and builds the second part of the
book, "the personal dimension", around it and Abelson's ideology machine.
Predictably, the author immediately has to apologize for the fact that the
neurotic program isn't very smart---certainly not as intelligent as most of the
programs discussed later in'the book. Indeed in chapter 14, she accepts Dreyfus'
denunciation of the exaggerated claims made about systems of the neurotic
program's vintage, but rejects it when applied to later ones. As these are not
discussed in detail, it is to be hoped that she does not overburden her audience's
credibility by this ploy.
The section on language and understanding consists of three chapters, respec-
tively describing the pattern matching systems Eliza and Parry, Winograd's
SHRDLU, and the Wilks and Schank systems. The author scorns the linguistic
abilities of Eliza and Parry, arguing that they "respond to", rather than understand
language. SHRDLU is presented as a theorem prover, because of its representation
of meaning by Planner theorems. A great deal is made of the systems heterarchical
organization, where heterarehy is rather vaguely defined as a network of procedures
and contrasted with a tree. A beginning AI student would probably be thankful
for the discussion of the kinds of knowledge embodied in SHRDLU, each with its
own representation, and the system's various inadequacies, among which are its
inability to paraphrase and its insistence on correct syntax. The chapter entitled
"sense and semantics" outlines the rather different approaches of Wilks and
Schank. There is little discussion of the issue of semantic primitives despite the
fact that it has been a hotly debated subject for several years.
Here, as in other parts of the book, it is difficult to estimate how much. a reader
with no previous familiarity with the material could understand from this account.
The section on vision is very disappointing, although to be fair to the author,
a number of exciting developments came too late to be included. In particular,
the discussion of the representation and deployment of global knowledge, such
as continuity of surfaces and a single light source, perhaps would have been
less oriented towards Grape's system. The first chapter is on line drawings of
polyhedra, and was referred to above. The chapter entitled "glimpses of the real
world" only barely lives up to its name, especially as low level vision is excluded.
Part five, entitled "new thoughts from old", discusses learning, creativity, and
problem solving. By and large the chapters are eminently readable and discuss
programs as diverse as Winston's, Evan's, Sussman's HACKER, Fahlman's
BUILD, and STRIPS. There is an interesting attempt to explain how computa-
tional ideas about intelligence are internalized into programming languages, and
why the development of AI proceeds hand in hand with the development of pro-
gramming languages for AI.
3. BOOK REVIEWS 269
Before discussing the final section of the bgok, it is interesting to contrast
Boden's approach with Winston's, whose book was published at about the same
time. Boden seems to equate AI with the set of AI programs, and presumably
hopes that her audience can gather the threads of ideas about intelligence scattered
throughout the book into a coherent unified statement. Winston's book on the
other hand is mainly notable for proposing such a coherent framework.
The book's final section on the relevance of AI to Psychology, Philosophy, and
Society, is perhaps the mosCinteresting for an AI reader. It overlaps her previous
book "purposive explanation in Psychology" and hopefully will make that book
more widely known to the AI community. One of her major concerns is to address
the apparent division between humanist accounts of man and reductionist accounts
such as behaviourism and physiology. Boden argues that a mechanistic account is
not inevitablyreductionist or demeaning; the idea that it is reflects an impoverished
notion of mechanism, which is based on nineteenth century machines. Computers
provide a vastly enriched idea of mechanism. This notwithstanding, one is tempted
to suggest that rather than pleading to humanists that we are really ok, it would be
more convincingto provide them with compellingcomputationaltheories. Ullman's
motion correspondence, and Marr and Poggio's theory of human stereo vision are
more likely to convince people of AI's pedigree.
Chapter 14 is a spirited counterblast to AI's critics Dreyfus, Polyani and
Weizenbaum. The final chapter outlines some of the more nauseous potential
misapplications of AI, and urges AI programmers to accept social responsibility
for what they are doing. The latter topic is introduced with the delightful observa-
tion that "members of the AI community bear an ominous resemblance to Mickey
Mouse", but you'll have to read the book to find out why!
J. M. BRADY
University of Essex, Department of Computer Science
Wivenhoe Park, Colchester CO4 3SQ, U.K.
A. Morecki and K. Kedzior, Eds., Theory and Practice of Robots and Manipu-
lators, Proceedings of the Second International CISM-IFToMM Symposium,
Warsaw, Poland, September 14-17, 1976,Elsevier ScientificPublishing Company,
Amsterdam, 500 pages.
Few workers in artificial intelligence are likely to take easily to this book. The
contents illustrate the wide disparity between A.I. views of robot manipulators
and those of the mechanical and control engineer. With few exceptions we are
into the world of geometric transforms, simultaneous non-linear differential
equations and real time computer control. In short the theoretical base on which
any real piece of machinery must stand, It is not an easy theory to master and