Advances in the Evolutionary Synthesis of Intelligent Agents
Mukesh Patel, Vasant Honavar and Karthik Balakrishnan (Editors)MIT PressCambridge, MA2001480 pp.ISBN 0-262-16201-6$52.00
Books on artificial neural networks are still very much in demand. This is a text made up of the contributions of the editors and invited authors. Much therefore depends on how they relate to one another. The problem they encounter from the start is one that challenges all who labour in this area: What needs to be done to design information structures so that they can adapt intelligence agents using evolution and learning? To do this the contributors, presumably under the guidance of the joint editors look in some detail at what we mean by artificial intelligence (AI). What is its real basis? Decades of reports on AI initiated by governments worldwide have not helped us resolve this problem. Governments as opposed to scientific bodies are really interested in whether there will be a return on their financial investment in the research. After years of AI research producing AI systems that have an estimated IQ of 5 does not impress the beaurocratics of this globe. The basic question remains: how can we effectively model human intelligence? The technology of the day limits us to computational devices and the contributors to this book attempt to understand what we mean by AI and then model thought processes using current computer systems. The big disadvantage is that we are really using our current computational devices to model whatever we mean by intelligence when they are still in a comparatively primitive form. Using a computer programming system on today's machines for modelling any processes is difficult enough, but when those processes are cognitive then a real challenge emerges. The editors and invited authors of this book meet the challenge and focus on defining the structure, the functions and the bases of cognitive systems before discussing how computer software can be developed to store, organise, manipulate and then apply the information.
The book does this competently to the limit of our present understanding of AI. It elaborates on the popular methodology in an approach that is wide ranging from symbolic techniques to the later neural nets. There are so many diverse views on the subject that it was pleasing to see that the writers could integrate many and took pains to display any shared values.
With some 480 pages the book provides a great deal of reading matter that attempts to analyse what we know about the concepts of AI in such a way that they can be successfully computationally modelled. In consequence the book is both thought provoking as well as a valuable resource.
D.M. Hutton
