The other sections covered a variety of interesting topics including the use of various expert system shells, several commercial programs presently in use, and one knowledge based system for tutoring. The conference proceedings should provide researchers with a compact resume of the state of the art together with an extensive bibliography for futfire study.
Dr P H Milne CAD Centre, University of Slrathclyde Glasgow I n t e r a c t i o n s in Artificial Intelligence a n d Statistical M e t h o d s Bob Phelps Gower Technical Press Ltd, October 1987, Softback 187 pp, £37.50, ISBN: 0 291 39743 3 This publication is a collection of 13 papers (176 pages in all) given at a Unicorn Seminar in December 1986. The papers are organised into four parts: • Automating Statistical Methods (5 papers) • Integrating AI to Stochastic Modelling (3 papers) • AI approaches to Learning from Data (2 papers) • Statistical Methods in AI (2 papers) The opening paper by D J Hand (The Application of expert systems in statistics) addresses the question of what one might expect of a statistical expert system. He identifies assistance with choosing the right technique, assistance in applying the technique, and interfacing to existing packages as three key application areas. He then describes four examples of the use of expert systems techniques taken from work in his own Institute. In the second paper (Data analysis as search) Lubinsky and Pregibon discuss their TESS system (a Tree based Expert System for Statistics) which tries to incorporate data analytic skills into an expert system. In the third paper (AI and generalised linear modelling: An expert system for GLIM) J Nelder describes GLIMPSE, an expert front-end for GLIM, written in the APES shell. It is an advisory system which lets the user exploit his own special knowledge of his data. The fourth paper examines the feasibility of expert systems being used in generating and testing statistical hypotheses. The paper is entitled "An expert system approach for generating and testing statistical hypotheses" and is by K Wittkowski. By using a pattern matching approach based on a classification of statistical problems in terms of domain knowledge an appropriate choice of a statistical technique can be made. The final paper is the first section is by S Furner (Dialogue management with computer based statistical analysis). He makes a plea for intelligent inferface design to be widened to include issues in human interface design taking into account the practical constraints faced by the consumer.
The first paper in the second section (A computer model with expert rules for the control of African cattle diseases), by G Gettinby describes an epidemiological model of statistical simulations whose use is controlled by expert derived rules. The second paper (AI and stochastic process simulation by R Paul) is concerned with the use of AI techniques in the partial automation of simulation modelling front-ended by a natural language understanding system. The second section concludes with a paper by Reilly and Timberlake on an Intelligent front end to Box Jenkins forecasting. An AI approach is used to assist the designer in choosing appropriate time models. The third section concentrates on the problem of finding rules which describe data in data sets well enough to classify it. A paper by IIolte and MacDonald on "Learning tasks studied in AI" acts as a useful survey paper. Bratko and Kononenko then propose a number of improvements to Quinlan's ID3 algorithm for handling noisy data in a paper entitled "Learning diagnostic rules from incomplete and noisy data". The results of using such improvements are illustrated. Finally Niblett and Clark in their paper "Learning if.. then rules in noisy domains" present a description of CN2, an induction system designed to extract short simple understandable rules from noisy data. The final section consists of two papers. The first "Synthesis of AI and Bayesian methods in medical expert systems" by D Spi~gelhalter examines the treatment of uncertainty in expert systems and stresses the need for a soundly based theory. He proposed a method of propagating probabilities through a causal net. The final paper by B Ripley (An introduction to statistical pattern recognition) discusses the importance and limitations of the statistical approach to pattern recognition. The interesting aspect about this collection of papers is that it represents a junction of two traditions AI and Statistics. There are some interesting areas of overlap, for example probability handling in expert systems and the roles Of machine induction and statistical techniques in learning. Being a report of a Seminar the papers do not usually go into great depth but they do provide some interesting reading for those interested in the overlap area.
J L Ally Turing Institute W h a t E v e r y E n g i n e e r Should K n o w A b o u t A I W A Taylor MIT Press, 1988, £19.95 The title of this book must suggest a work of considerable relevance and interest to all readers of this journal. A preliminary scan through the introductory text and contents merely whets the appetite suggesting a book with a broad technical coverage but presented in a practical way which would suit engineering applications.
Artificial Intelligence in Engineering, 1989, Vol. 4, No. 1 55
Unfortunately, and sadly, these expectations are not satisfied. Despite some useful technical content, the material is presented in a facetious and trivial way, and is riddled with unforgivable generalities and loose statements. It is only a little short of being overwhelmingly patronising. The content is organised in a logical way, moving from introductory material, through an examination of a number of applications, and into the main body which emphasises the LISP language and Logic Programming. The book finishes with a look at sundry topics such as "getting started" and the role of J a p a n ' s Fifth Generation Project. An underlying philosophy in the book is t h a t the main outcome of AI for engineers is new programming styles. However this philosophy
is somewhat overworked to the extent that there is no clear distinction between knowledge representation and reasoning, and programming - surely a mistake if engineers are to adapt to AI techniques. The back cover quotes an industry expert - "As a practising engineer in the fields of AI and automation, I can enthusiastically recommend it for any technical bookshelf". I cannot do the same. I can only hope that we eventually get books "which do the job properly.
K J MacCallum CAD Centre, University of Strathclyde Glasgow
CALL FOR PAPERS Fifth International Conference
AIENG 90 Applications of Artificial Intelligence in Engineering 17 - 20 July 1990 Boston, Massachusetts, USA Organised by: Computational Mechanics Institute, Wessex Institute of Technology, Southampton, UK Application Areas
OBJECTIVES The purpose of this Conference is to provide an
international forum for the presentation of work on the stat~of-the-a~t in the applications of artificial intelligence to engineering problems. It also aims to encourage and enhance the development of this most important area of research. LOCATION The Conference will be held at The Onmi Parker House Hotd, Boston, MA, USA CONFERENCE
CHAIRMAN
Dr R A Adey Computational Mechanics Institute Ashurst Lodge, Ashurst Southampton, SO4 2AA, U K
• Design • Diagnosis • Process Control and Planning • Robotics • Tutoring • Sensing and Interpretation Topics • Representation • Problem Solving • Constraint Reasoning • Learning • Qualitative Models
Telephone: 44 (0) 42129 3223 Telex: 47388 Attn C O M P M E C H
• Tools
Fax: 44 (0) 42129 2853
• User Interfaces
CONFERENCE SECRETARIAT S a n d r a Elliott Computational Mechanics Institute 25 Bridge Street Billerica, MA 01821, USA Telephone: (508) 667 5841 Fax: (508) 667 7582 CONFERENCE THEMES The major themes of this years conference are Design and Manufacturing. However papers are welcome on any applications of AI in Engineering. The following application areas and topics are suggested and other related areas will be considered:
56 Artificial Intelligence in Engineering, 1989, Vol. 4, No. 1
CALL
FOR
PAPERS
Papers are invited on the topics outlined above and other topics which willfit within the general scope of the Conference. Draft papers should be submitted to the Conference Chairman by 13 October 1989 and authors will be notifiedof the preliminary acceptance by 15 December 1989. Final acceptance will be based upon review of the finalpaper, which must be received by 16 March 1990. The Proceedings of the Conference will be published in book form by Computational Mechanics Publications and distributed worldwide. The language of the Conference will be English.