 |
Vol. 4 Iss. 1 The Chemical Educator © 1999 Springer-Verlag New York, Inc. |
ISSN
1430-4171 |
|
Vol. 4 Iss. 1 The Chemical Educator © 1999 Springer-Verlag New York, Inc. |
ISSN
1430-4171 |
Book Review
Reviewed by
Sally Solomon
Drexel University, Department
of Chemistry, Philadelphia, PA 19104
solomosd@dunx1.ocs.drexel.edu
Applications of Artificial Intelligence in Chemistry,by Hugh M. Cartwright. Oxford University Press: Oxford, England, 1993. $12.95. 92 pages. ISBN 0-19-855736-1.
Hugh Cartwright's Applications of Artificial Intelligence in Chemistry reminds us that computer science is about ideas as well as machines. This well-written text, published in 1993, is still relevant; it provides the reader with an excellent introduction to the meaning of artificial intelligence (AI) and the many areas in chemistry where it can be applied. Tucked into the margins of the book are remarks that enhance the text without interrupting its flow. Although most of the examples are chemical, the text should be suitable for undergraduate students in any science.
The first chapter introduces the enormous potential of AI, "an attempt to reproduce intelligent reasoning using machines." Problems that lend themselves to solution by AI, such as playing chess, are discussed. It is pointed out that although the rules of chess are straightforward, an exhaustive search of all legal moves would be impossible to do fast enough. Similar demanding problems that arise in science are treated. The next three sections of the book describe how AI methods work, with particular emphasis on applications in chemistry.
The next chapter covers artificial neural networks, generalized learning machines based on a simplified model of the brain. The simple perceptron, described in detail, is a decision-making unit with several input connections; adjustment of these allows the perceptron to learn by trial and error. One of the chemical problems described is the use of light sensors to notice a 6-membered ring in a structure. Combining perceptrons into neural networks allows the solution of more complex problems such as the analysis of NIR spectra or the prediction of the secondary structure of a protein. An interesting margin remark tells of a pitfall encountered by a a neural network that accidentally learned how to tell when the sun was out rather than to how to identify tanks because the trial pictures of tanks were all taken on a sunny day.
In the third chapter are the knowledge-based systems (expert systems) that rely on stored facts. The essential components are discussed here: the knowledge base, the reasoning engine with a discussion of various searching strategies, and the human interface. When the knowledge obtained is combined with rules, the system can offer excellent advice on how to interpret data, monitor equipment performance, control instruments, instruct novices, and thus become a significant part of a large laboratory.
The final chapter introduces the genetic algorithm (GA) "to search for the optimum solution hidden in a wealth of poorer ones." The text explains that a GA uses a hill-climbing method that is less likely to be fooled by a local maximum as it relies upon randomized searches. The name of the algorithm is derived from its basis in the concepts of evolution. Solutions to a problem are strings, or chromosomes. Fitness of strings is evaluated and new populations are produced using reproduction, mating or crossover, and mutation operators. One application discussed is in the choice of a suitable synthetic route for a complicated organic molecule.
Cartwright's succinct introduction to AI is an excellent starting point for the reader who should be inspired to explore methods made possible by increasingly powerful computing ability.