Book Title: “AI Techniques for Game Programming”
Author: Mat Buckland (www.ai-junkie.com )
Publisher: Premier Press
ISBN 10: 1-931841-08-X
Companion Site: N/A
My Rating: 4/10 (Can be skipped!)

“AI Techniques for Game Programmers” reads like a book for high school students, and though it tries really hard, it barely manages to touch on real-world game applications of the soft-computing methods it introduces to the reader. 

The book starts with an introduction to windows programming; a hello-world program, keyboard input, basic Windows GDI and graphics, etc. Using up the first quarter of this 480 pages book to teach basics of programming is a poor choice on the authors part, considering that most readers of this book will be game AI department programmers, who probably know this stuff, and don’t really need it anyways. Besides, there are better books on windows programming, and game AI programmers mostly use the game interface itself anyways, when they need graphics. Having gone through (or skipped) the first part, the books starts to get a tad more interesting.

The second part is devoted to Genetic Algorithms (GA), and after a very basic, but lengthy introduction to the Genetic Algorithm the reader can check out how GA manages to provide a good solution to the Travelling Salesman Problem (TSP). Again, we get a lot of information on basic C/C++ programming which is really unnecessary. Revisiting variations of genetic algorithm operators, Mat barely mentions that the TSP can be applied to the problem of real-world path-finding, before leaping into another huge chapter of vector graphics! and using s simple genetic algorithm to automatically play a Lunar Lander, and that’s it! Strangely there is no mention of how to use GAs for general path-finding, creating artificial life, co-evolving bots, choosing best driving paths, evolving Finite State Machines, Genetic Programming, and loads of other interesting applications for real games.   

What he did in the second part for GA, mat does in the third part for Neural Networks. Again, simple basics, and a very basic example of what can be done with NN’s. This time he implements a mine-sweeping bot, though I admit this time the example is a bit more interesting, at least in its application to rover/bot creation for games. The second example involves recognizing what gesture the player makes with his mouse (Myst tablets spring to mind) and is quiet useful.