Do we need another textbook on artificial intelligence? The authors of this book argue that we do for two main reasons. Other books either require too much from their readers (background knowledge and commitment to follow the subject in great detail) or offer a limited coverage of the subject matter.
The authors aim to provide a wide coverage by adopting a definition of AI which they hope covers most interpretations of it: "AI is concerned with building machines that can act and react appropriately, adapting their response to the demands of the situation. Such machines should display behaviour comparable with that considered to require intelligence in humans."
Although this definition is satisfactory in covering a non-controversial set of lectures for a short, 12-week module on the subject it leaves most of the exciting new (and not so new) developments in AI beyond the scope of the book.
Does the machine need to have a "body" to be able to relate to its environment in an intelligent way? Does this environment need to be like that experienced by humans if the machine's behaviour is also to be human-like? What kind of infrastructure best supports human-like behaviour?
The authors ignore most such questions and focus on an early notion that intelligence is what goes on "inside the head". As long as we can represent this knowledge in a satisfactory form and discover procedures that operate with it we can then perform most tasks requiring intelligence. Such a view leads to the present book's structure, which is divided into two main sections. Chapters 1-5 provide key techniques such as knowledge representation, planning, reasoning, search, learning and game playing. These techniques are applied, in Chapters 6-10, to various task domains such as vision, language and robotic movement. The hidden narrative of the book is that some general problem solving techniques underlie most aspects of human intelligence.
Unfortunately, this view of intelligence has many critics who consider it a myth about intelligence created in order to talk about our own behaviours. The processes which actually give rise to intelligence may be more primitive and less rational than the classical AI exposed in this book leads us to believe.
The authors attempt to incorporate the neural (or connectionist) approach to AI in a short section in Chapter 11 which deals with the more general issues related to models of the mind. However, the treatment of the neural paradigm is disappointing, considering it poses a fundamental challenge to the whole narrative of classical AI. If you adopt the neural approach then most of the "`techniques" part of the book becomes irrelevant.
The most disappointing is an absence of any reference to the Emergent AI (or A-Life) view where it is argued that life and intelligence only develops inside a "body" which co-evolves and adapts to the "ecosystem" in which the creature is living. This third paradigm has given rise to success where traditional AI has failed to deliver results.
The most worthwhile part of the book for me is Chapter 10 which deals with intelligent software agents. While vision, movement and speech may be better performed by specialist "bodyware" instead of puzzle-solving systems, I feel symbolic approaches to AI are well suited to areas such as information retrieval where the task is itself artificial. Symbolic AI may still find itself a comfortable niche in making search engines more efficient, as it has done in providing expert systems (Chapter 6).
Compared to better known textbooks on AI this book's main virtue is that it is presented in a more compact style and does not demand much from its readers. A student whose only interest in AI is to add it to the list of topics covered in a degree programme would find the book a good source of information without being exposed to much controversy or the latest discoveries in the field.
Masoud Yazdani is founding editor of the journal Artificial Intelligence Review and onJanuary 1 1998 will join theUniversity of the West ofEngland, Bristol as professor of digital media.
An Introduction to Artificial Intelligence
Author - Janet Finlay and Alan Dix
ISBN - 1 85728 399 6
Publisher - UCL Press
Price - £14.95
Pages - 288