AI Application Programming - Book Review
I’m spending a lot of my time lately reading about AI, both game related, business application related and academia related. Having recently completed “AI Application Programming” by M. Tim Jones (Charles River Media, 2003, First Edition), I have decided to review it in the context of my current situation.
I am new to the field of AI, and my previous reading has been limited to sporadic Wikipedia research on concepts as I come across them. This book covers a lot of material that I haven’t read about before, but it also covers some material I have. This allows me to evaluate the book as both someone with prior knowledge, and someone with no knowledge.
Firstly, the book was a fairly broad overview on integrating aspects of artificial intelligence into application development. Therefore it was no surprise that a majority of the AI concepts discussed were optimization oriented. The book provided an overview of the history of AI, then it covered simulated annealing, adaptive resonance theory, ant algorithms/ant colony optimization, neural networks, genetic algorithms, a chapter on using some of the techniques described to run artificial life simulations, rules based systems, fuzzy logic, bigrams/Markov chains, agent based software and a wrap up the future of AI.
Each chapter is fairly formulaic. It starts with an overview of the theory of the subject for that chapter, and the presentation of any relevant formula. Then a basic example is covered, without code, in order to show how the theory being covered can be applied. Then a specific example with C source is covered, with a detailed explanation of what’s going on. I largely glossed over the source, as I don’t really have much familiarity with C, so cannot comment on the quality of the source.
Overall I have mixed feelings about the book. A lot of the topics have domain specific language, such as referring to parts of algorithms as genes and candidate algorithms as chromosomes in genetic algorithms, and I feel the author sticks to these to the detriment of comprehension of the topic. For example, the author was discussing “energy” and “temperature” with relation to simulated annealing, without outlining that it’s goal was to narrow the search space for a solution to a given problem.
Another detracting factor was the numerous errors in diagrams, equations and the text. These errors can be attributed to a variety of sources, from poor copying jobs on some of the diagrams to bad editing. One particularly annoying error was the diagram showing how a NOT logic gate could be constructed using a single neuron neural network – the diagram showed an input bias of 1, with a weight of 1 on the input, claiming that the output will always be the inverse of the input, by the equation output = input * input weight + bias.
An input of 0 into this equation will indeed output a 1, however an input 1 will yield 2, which is incorrect. The correct diagram should have a input weight of -1, which will yield a correct answer with both the 0 and 1 inputs. These errors hindered the understanding of the topic.
However, it’s not all bad news. From my personal reading, I had always struggled with how the back propagation algorithm in neural networks worked, and this book managed to successfully explain it. A lot of the other topics, such as fuzzy logic and the rules based system were also well presented and relatively straight forward.
From reading this book, I feel I have a solid enough understanding of the basics of the topics covered that I could successfully go out and perform my own, more in depth research and succeed fairly well, as well as integrate a few of the more rudimentary features of these approaches into an application.