Intelligent Systems: Architecture, Design, and Control

Citation

Andrew, A.M. (2002), "Intelligent Systems: Architecture, Design, and Control", Kybernetes, Vol. 31 No. 6. https://doi.org/10.1108/k.2002.06731fae.006

Intelligent Systems: Architecture, Design, and Control

Alexander M. Meystel and James R. AlbusWiley-InterscienceNew York2001xxi + 696 pp.ISBN 0-471-19374-7hardback, £55.95

Wiley Series on Intelligent Systems

This is a comprehensive review, and in some ways a rethink, of the nature of intelligence in both natural and artificial systems. Questions of definition are treated, as well as the origins of the idea that intelligence can be a feature of living systems and artefacts alike. The authors are well qualified to expound on intelligent machines as they have played a leading part over many years in developing industrial robots and, even more significantly, autonomous land vehicles that are among the most convincing demonstrations of machine intelligence.

The book could very well be the basis of a taught course, and at the end of each chapter there are quizzes and test questions. The treatment has a particular slant, which certainly has to be acknowledged as productive in view of the results achieved by the authors. Emphasis is placed on the ''multiresolutional'' nature of intelligent behaviour, and the cover illustration is an artwork called ''Multiresolutional Thinker'' by the first-named author. This looks similar to Rodin's ''Thinker'' but with a nesting of smaller thoughtful heads within the main one.

The treatment allows the topics of perception, knowledge representation, and planning to be viewed in separate compartments, and in this sense it amounts to ''good old-fashioned AI''. In this it differs from the recent ''Cambrian intelligence'' approach due to Rodney Brooks, and other studies of simple robots in which sensory input is coupled rather directly to motor output, and these studies are clearly felt to be of limited value. (It is of course true that animals have spinal reflexes, but these account for only a small part of overall behaviour and are in any case subject to a degree of control from higher levels.)

The emphasis on multiresolution is maintained throughout. It can be seen to operate in a simple way in the analysis of sensory input, where for example the input from the sensory elements of a retina is of very high resolution but is whittled down in stages to a description in terms of a relatively small number of significant objects and features. Conversely, behaviour generation proceeds in stages from a relatively simple setting of goals to a plan of action that is enormously more complex. The successive stages in these processes are levels of resolution.

There is very much more to the theory than this thumbnail might suggest, and processes of Grouping, Filtering and Search, as well as Abstraction, Aggregation and Generalisation, are introduced as components. The structure of stored knowledge is argued to be necessarily multiresolutional, as is intelligence itself. There is a chapter on Motivations, Goals and Value Judgement, and one on Learning.

The need to organise and store knowledge makes the approach essentially symbolic, in contrast to ''Cambrian intelligence'', and an appropriate assignment of symbols is assumed to be made by a learning process. There is reference here to the discussion of ''signs'' by Peirce, and particularly to a recent development of the theory of semiotics due to the Russian worker Dmitri Pospelov. A project that is discussed is the making of a baby-robot that would simulate early cognitive development, with the possibility of comparison with human infants. Interest in such a project demonstrates that the aim is very deep understanding of cognition. Semiotics is defined as the art and science of constructing models for interpretation and thus for meaning extraction. An intriguing relationship between semiotics and mathematics is introduced on page 653, with mathematics seen as differing in being free of the need to be grounded in reality at each step but nevertheless serving as a laboratory for exploration of structures and relationships that may subsequently be amenable to grounding and to becoming part of semiotics.

In the final chapter of the book, very general issues concerning the nature of intelligence are treated, with the suggestion that criteria of intelligent behaviour might be found that do not depend on reference to the biological paradigm. Systems to understand text and for example to summarise and categorise documents do not at present exist but it is claimed that rapid progress is being made towards ''this Frankenstein of the twenty-first century''. There is much thought-provoking discussion of underlying principles, of which the above observation on a connection between mathematics and semiotics is an example. The title of the chapter is: ''Intelligent Systems: Precursor of the New Paradigm in Science and Engineering''.

The development of the theory in the earlier part of the book has an essentially practical basis, and examples of multiresolution are found in such elementary contexts as the automatic variation of step size within a range of numerical integration, and similar variation of granularity in finite element methods. A number of the illustrations show records of such things as exploratory activity of robots, and other practical results. An application to factory planning is described, and also an autonomous vehicle able to pick its way across a cluttered piece of terrain and then to return much more quickly using stored information. Another project involves the collaborative operation of a platoon of autonomous vehicles.

The book is clearly written, and there are numerous helpful figures. At the same time, it is a big book and the wealth of diverse arguments that are brought to bear at every stage means that getting to grips with it is a daunting task. The effort will certainly be found to be well worth while, both for developers of advanced robotic devices and workers in AI generally, and for psychologists and other biologists. The book offers a new perspective on many basic aspects of cybernetics and is admirably multidisciplinary in its linking of esoteric considerations of semiotics to down-to-earth robotics and biology.

Alex M. Andrew