AI for Games Developers

Authors: David M. Bourg and Glenn Seemann



Publisher: O'Reilly Associates

Reviewed by: Simon Wistow

AI is a funny discipline. In the concept of computer graphics there's a term called 'The Uncanny Valley' which describes the odd phenomenom observed when creating photorealistic humans. The graph of quality of graphics versus perceived realism rises smoothly and proportionately until just before 100% realism at which point it suddenly dips dramatically and then continues rising. The reason is that at that point your brain starts noticing the things that are wrong rather than all the things that are right.

AI is just like that.

There are some fantastic academic AIs which would never be used for games because games, like movies, are about suspension of disbelief and not about realism. It's more important for a games AI to seem real rather than actually *being* real. This also has the added advantage that faked AI is often a lot cheaper, in terms of time, cost and processor cycles, than 'real' AI.

This book is about the fake 'AI'.

The big problem this book has it that the state of the art moves incredibly fast and is well beyond the scope of a 360 odd page book. Instead, and quite correctly, what it does is produce a primer of the basic techniques - their pros, their cons - and doesn't try and pretend that at the end of it you'll be able to get a job as senior AI developer at Bungie working on Halo 3.

You might spot some of the techniques they're using though. Which is pretty neat.

So what *does* it cover then?

The book starts gently - the first seven chapters are based on very simple techniques such line of sight chasing, flocking, A* path finding, - all basic techniques that can be easily illustrated using ye olde worlde 2D tile based game worlds and then scaled up to fully 3D, cortex busting, immersively shiny 3D environments.

Then there's a slightly odd (for me at least) chapter technique on scripted AI encounters. It's this chapter which, in my mind, demonstrates the most serious problem with this book. In many ways it's sufficent just to say that there are scripted AI events and these are the advantages and disadvantages. Dedicating a whole chapter to it - nearly the same number of pages as the ones on rules based AI, probability in games and fuzzy logic, seems a bit, well, imbalanced.

I'm not going to dive into details about the rest of the chapters suffice to say that they cover the aformentioned fuzzy logic, probability and rule-based AI as well as Finite State Machines (also called Discrete Finite Automata) and Genetic Algorithms , Bayseian techniques and an introduction to Neural Networks.

To be honest I quite enjoyed this book but had problems imaging who it was actually for. If you know anything about AI then it's really all a bit too simplified and if you don't then, well, it's a bit too simplified for you really learning anything. Although it will point you in the direction of the right words to look up when you actually want to go a bit more hardcore. It's sort of a pop-ai book - a lightweight easy read primer and when I stop being snobby about it and admit that I like reading pop books on psychology, economics and the like then I'll concede that it's not a bad read if this sort of thing interests you.