 The magazine of the Melbourne PC User Group Artificial Intelligence Major Keary |
 |
Media speculation about what artificial intelligence (AI) could and would achieve, and the failure of the scientific community to define what is meant by 'intelligent machine', has given AI a bad public image. Even some academics are dismissive, as we saw in a recent issue of PC Update.
Just because a computer can perform complex computations infinitely more quickly than a human does not make it intelligent, whatever definition one may give to the word. As the author of Blondie24 says, "a machine doesn't become a thinking machine, an intelligent machine, just because it can fool you into thinking that its thinking". However, it is possible to develop a program capable of learning and applying its new knowledge.
Blondie
A book that anyone should be able to read and understand, Blondie24: Playing at the Edge of AI, is a deceptively small title that is unlikely to catch one's eye, which is a pity. It deserves to be widely read.
Two concepts applied to the Blondie24 project are neural networks and evolutionary computing. Neural networks is not new, it was suggested by Frank Rosenblatt in 1958; and evolutionary computing had its beginnings even earlier: at least ten people came up with much the same concept between 1953 and 1970 - an Australian, Alex Fraser, published in 1957.
The author of Blondie24 is CEO of "a company that addresses complex problems in industry, medicine, and defence by applying such techniques as evolutionary computation, neural networks, fuzzy systems, knowledge-based systems, and stochastic processes". In this book he describes how some of those concepts were applied to a computer program devised to play checkers, which we call draughts.
What is interesting about this project is that the machine played human participants in web-based competitions without its non-human nature being declared. Readers will be amused by accounts of exchanges between unsuspecting male players and the program.
The hardware was quite modest, a 400 MHz Pentium II PC, so this is not an account of how some super-computer was programmed. It is the story of how the idea was conceived and translated into a successful application.
When a machine, Deep Blue, defeated Garry Kasparov at chess it was not a computer, or its program, that won; it was an army of programmers. Deep Blue was given an immense library of possibilities, but was not able to build on that library as a result of its own experience. Blondie24, on the other hand, acquired 'her' own 'knowledge' of checkers from games played. The seed information was provided by the human programmers, but from there on Blondie had to find 'her' own way.
The first part of the book discusses AI in language that is meaningful to lay readers, and which offers fascinating insights to teachers and students of computer science. Draughts aficionados will find the discussions illuminating and useful. However, it is the author's engaging explanations of evolutionary computing and neural networks, and how they have been used for a real-world application, that will capture the interest of lay readers with an interest in AI.
Science and maths teachers in the secondary school system should find this a useful resource and worth recommending to their students. It deserves to be included in general reading lists for tertiary courses that include or touch on AI-related subjects.
| David Fogel: Blondie24: Playing at the Edge of AI ISBN 1-55860-783-8 Published by Morgan Kaufmann, 404 pp., RRP $72.50 incl. GST |
 |
Swarm Intelligence
This is a book that deserves the description, 'a good read', even though it won't be everybody's idea of what
to take on a long flight. For any "intelligent layperson ... with a driving curiosity and interest in the
current state of science" this is an important text. Even though not designed primarily for a general
readership, it is an exceptionally well-written narrative that deserves a much wider audience than
"researchers and graduate students in social and computer science" and computer professionals.
The authors argue that "intelligent human cognition derives from the interactions of individuals in a social
world" and that view can "be effectively applied to computationally intelligent systems". They present an
account of "some exciting research that you may not have heard about - since it covers recent findings in
both psychology and computer science, [the authors] expect most readers will find something here that is new
to them" [and those] "interested in trying out some of these ideas . will find enough information to get
started or . where to go for the information".
The book is in two parts, Foundations and The Particle Swarm and Collective
Intelligence. The first part is a foundation for understanding the second part; it includes an excellent
overview of the fashions of psychology and how they have been changed by the work of great thinkers such as
Noam Chomsky. The development of computational science has both influenced and been influenced by that
work.
The second part of the book examines the authors' propositions about swarm intelligence and describes some of
the experimental work that has been carried out with a new algorithm, particle swarm optimisation
(PSO) that is "a simple computer program [that] can be written in a few dozen lines of code [and can be] used
to solve tough math problems". A companion web site has a Java PSO applet and source code available for
download with a genetic algorithm application and other useful material.
Anyone interested in communications aspects of computer science will find a good discussion of Shannon's
information theory experiments and lucid descriptions of soft computing, fuzzy logic, genetic algorithms, and
evolutionary programming.
For those with an insatiable curiosity about the way in which the boundaries of various disciplines blur and
cross to create new insights and new ways to achieve solutions, this is essential - and enjoyable -
reading.
| Kennedy and Eberhart: Swarm Intelligence ISBN 1-55860-595-9 Published by Morgan Kaufmann, 512 pp., hc, RRP $152.90 incl. GST |
 |
Intelligent Machines
Intelligent Machines: Myths and Realities is a collection of lectures
published in book form "intended for both [a] technical and non-technical readership". The editor says, "In
that context, an extra attempt has been made to present complex technical issues in simple and qualitative
terms". What impresses me is the scope of discussion: the problem of defining intelligence, concepts that
underpin 'machine intelligence', an overview of the latest research, and descriptions of recent developments
and applications.
The contributors are not simply concerned to explain technical matters, but include discussion of social,
economic, and environmental considerations. It is a real-world look at intelligent machines and their
development. As one of the contributors says of IBM's Deep Blue-II win over Gary Kasparov, "It was not the
contest between man and machine, but between man (Kasparov) and machine assisted by a team of experts in the
development of complex algorithms. . It was the collective intelligence of the programmers, with the enormous
speed and memory of the computer, that won the match".
Chapter headings indicate the scope of the text: What Makes a Machine Intelligent?; Information, Knowledge,
and Machines; Machines and the Elusive Wisdom; Research and Advances in Intelligent Machines; Soft Computing
Tools for Intelligent Machines; Intelligent Production Machines: Benefiting from Synergy Among
Modeling,Sensing, and Learning; Intelligent Control of Machines; and Using an Intelligent Machine to Modify
or Adapt Human Behaviour.
Soft computing, a recent development, encompasses fuzzy logic, genetic algorithms, and artificial neural
networks, whereas conventional computer operations demand crisp logic: a definite yes or no, true/false. In
an analog world things are not set in digital concrete and artificial intelligence systems must be capable of
dealing with variables that may exist between binary zero and binary one. This text provides an overview of
soft computing and explains fuzzy logic, genetic algorithms, and artificial neural networks in language that
should not set the interested lay reader's head swimming.
The application of soft computing to a particular problem is illustrated by an expert system for image
enhancement; SmartPhotoLabc, developed using Visual C++, applies expert systems to "detect unwanted features
in an image and enhance the quality ... in terms of brightness, contrast, colour, and tint". The 'case study'
is also interesting in that the intelligent machine is in fact the program running on a desktop computer.
Intelligent Machines is not a popular account of the technology, and does not gloss over the difficult
bits. It is a text that will satisfy the general public as well as engineers, academics, researchers, and
students. If you have an interest in the technology, either for its own sake or for the potential social
impact of such developments, Intelligent Machines is well worth reading. If it is not in your local library,
suggest that it be put on the acquisition list. The price may seem daunting, but this title is an important
contribution to the literature; it deals with many issues that don't find a place in the conventional texts,
and caters for an intelligent lay readership. It deserves a place in secondary school libraries, especially
if students are involved in robotics, and should be on the reading list for any computer science
course.
| Clarence de Silva (ed.): Intelligent Machines: Myths and Realities ISBN 0-8493-0330-3 Published by CRC Press, 326 pp., hc, with index and references. List price (as at May 2002) $196.05. Available through http://www.dadirect.com.au. |
 |
Reprinted from the July 2002 issue of PC Update, the magazine of Melbourne PC User Group, Australia
   |