AI for Game Developers [Paperback]

Oh dear. This is a really poor book, and I'd advise you to save your money and buy something else.

Bourg's first book on game development (on Phyiscs) suffered poor reviews for being shallow, not presenting real-world techniques, and glossing over huge gaping problems in his approach. This book has all the same problems, and even more strongly in places.

Firstly he makes a big point of saying that the book is aimed at novice developers and he will cover only a few techniques designed to get you up and running. In his list, however, he has genetic algorithms, neural networks and production systems, all of which large AI teams for major developers find it very hard to get working usefully (e.g. finding an application suited for a GA in a game is just about impossible). His approaches are in places laughable.

Secondly he then lauds completely inappropriate algorithms. His steering system is based on potential functions, for example, and he proposes this because he says it allows you to mix several different movement concerns. This is completely disingenuous to his readers. The technique is very simple and useful only when several concerns are NOT blended together. When you mix things you ALWAYS get wells-of-attraction and your characters get stuck hopelessly. Just about ANY other technique would improve on this (and there's no use just saying 'do pathfinding' either: there's much more mileage in steering algorithms).

And the final thing that annoyed me was his approach all the way along of saying 'I'm only going to show you a few techniques'. Trying to make the selling point of the book that it doesn't cover anywhere near enough ground to get your AI up and running!

Just about every chapter had significant problems in it. His knowledge of the subject wouldn't pass a sophomore AI exam (I've taught AI courses at University: I'm not exaggerating). It is full of innaccuracies (e.g. he says 'new AI' is non-deterministic, while 'old AI' is deterministic - a moronic comment).

If you are a novice trying to get going in AI, this book will seriously harm you development. Try John Funge's new book, or the new Morgan Kauffman book (you'll have to wait a few months). Even reaching through the two AI Programming Wisdom will see you far better prepared for real development, despite the odd lemon of an article in them.

Terrible and useless even for a book on AI for budding game developers. The theory and explanations in this book are sometimes decent but more often than not quite lacking. (es: in one of the first chapter the author uses Bresenham algorithm without taking the time to explain it). The use of tile based examples introduce unnecessary overhead, and the continuos attempts to introduce physics related code and references to the author's other book on game physics are just plain annoying. The range of subjects covered is very broad (chasing and evading, pathfinding, emergent behaviours, rule based reasoning, bayesian networks, neural networks, fuzzy logic, finite state machines, genetic algorithms), definitely too broad to treat each of these subject in decent depth and with clarity. Example code is of low quality and just superficially object-oriented. If you are looking for a decent introduction to game AI I recommend Matt Buckland "Programming Game AI by Example" and "AI Techniques for Game Programming".

AI for Game Developers is a book aimed at beginner game programmers that want to apply AI techniques in their games. The range of topics covered is vast, chasing, flocking, pathfinding, state machines, bayesian networks, neutal networks, genetic algorithms, fuzzy logic and more.

The authors spend time to explain these AI concepts from the ground up, with numerous code samples and accompanied by full programs including GUIs & simulations, downloadable form the book's website.

So far all good, but not really so if you look a little closer. For a programmer with more experience than the complete novice, many problems become apparent quickly:

The code listings ("examples", as the authors uncommonly call them) are lacking in many aspects. The code is of low quality. C++-- some call it, and it's an appropriate name for the code of this book (take a big program with a bunch of globals and void foo(void) operating on them, wrap it all (public) in a class, and call it Object Oriented). The principal author is a scientific programmer, which shows well since the code looks sometimes like a port from Fortran. The code is also poorly typeset, in a font almost identical to the text, which makes samples hard to spot and follow. The authors also paste huge amounts of code in the book that sometimes fill whole pages, without any apparent reason. Another little deficiency is the typesetting / overall formatting quality of the book - some diagrams are simply missing, fonts are uncomfortable (especially the mix of code & text)...

It seems that most of this book's problems are about the code, though. On the conceptual level, the book is not bad. As far as I can judge, the explanations are clear and even novices should have no trouble understanding how the algorithms work. Just look away from the code, please !!