Customer Reviews

Behavioral Mathematics for Game AI

5 star:		(10)
4 star:		(1)
3 star:		(0)
2 star:		(0)
1 star:		(1)

 

 

I *love* this book. I simply do not understand the negative comments left by another reviewer. Because of my interest in politics, I want to write code that mimics human behavior, but until Dave Mark's book could not find anything that really "fit." Once upon a time, I took advanced math courses--even excelled in the later ones, but that was nearly 20 years ago, and I haven't really needed that type of math since. Mark explains all the math in easy to understand English. He also provides excel spreadsheets and C++ source that can be downloaded at http://www.courseptr.com/downloads. I don't remember when I've read a book as interesting as _Behavioral Mathematics for Game AI_, and that's saying something since, although I have an advanced degree in computer science, I've never really been a lover of math.

Behavioral Mathematics is a book that I'm sure will continue to prove its worth for many future generations of Game AI Developers. I can make that strong assertion because Dave chose, instead of writing a large tomb about a possible AI architecture, to write a book that tackles (as the title implies) the mathematics behind Game AI.

Dave gives the reader a very strong toolkit for building Game AI, and by toolkit I don't mean a large library of code (although there is plenty of code) but fundamental knowledge, such as the concept of utility and how to apply it in a very practical way. Starting from simple example decisions he builds up to much more complicated cases, constantly tying everything back into practical applications of all the concepts he introduces.

Now, I don't mean to imply that this is some dry math book. Dave constantly sprinkles in his very unique brand of humor and ties in family stories that help cement any topic he's trying to get across. I would highly recommend this book to any professional or aspiring AI Programmer, or even just a player who wants a better understanding of whats going on under the hood in his favorite game.

Truth in advertising: I was the technical editor on this book. However I don't have any further financial interest, so I have no conflict of interest in writing a review. I'm also a professional game developer with 6 titles under my belt, as well as a lecturer on the subject at several prominent universities.

There are only a handful of books on AI for Games, and the techniques used in games are quite a bit different than those used in academic AI. Most of the existing books just look at the standard architectures and topics - behavior trees, state machines, scripting, path planning, etc.

Dave takes a novel approach, and looks at techniques for building heuristics that evaluate the game situation and rate the various options. This is a critical part of decision making, it is something which most games with even moderately complex AI need to do, but it is something that up to now wasn't covered very well (if at all) in the literature.

I heartily recommend this book for anybody who's interested in learning more about how to build decision makers - whether for use in games or elsewhere. The material that is here is not something that you'll find somewhere else.

"Behavioral Mathematics for Game AI" offers a unique examination of AI in a fun but technical and specific coverage of the subject from a mathematical perspective. While maintaining a conversational tone, the book manages to cover in depth several interesting topics and techniques.

Several ideas only touched on in other texts are given a comprehensive examination. When do rational agents fail to yield interesting behavior, and how do we deal with irrational opponents? How to we prioritize the tasks we want to accomplish, and how do we keep our agents from being erratic in their choices? Finally, how to we make sure our agents have enough variation to make them feel real?

Mr. Mark's book does not focus on some of the basic constructs common in game AI today such as state machines and behavior trees; instead, it focuses on the nuts and bolts of the math we can use to tie our agents together, augment them, and give them variety. These techniques make them more deliberate, less predictable, and pose more "interesting questions", as the author is so fond of mentioning: in short, these techniques bring AI agents to life and make them more fun to play with and against.

Dave has done a remarkable job of putting together a lot of complicated mathematical concepts in an easy to understand, pleasant to read package, and I recommend it to anyone with an interest in putting together interesting AI!

One of the best things about Dave Mark's book is that he tries to get the reader to approach AI problems from multiple angles using the algorithms that he presents, as well as how to evaluate the complex problems posed by behavioral AI. Written in a very down-to-earth and easy to comprehend manner, the book was an easy read, which can't be said for all AI books.

For my own AI work, Behavioral Mathematics for Game AI gave me a host of new ideas and mathematical tools to apply and test, and I highly recommend it for anyone who is working on game AI.

I met Dave Mark at GDC and he brought this book to my attention after seeing me with Brian Schwab's book, and I'm very glad he did. This book is great at introducing many useful AI concepts (Go Utility!). It also shows several great approaches to how you should approach designing AI, particularly in the area of making AI seem human rather than being 'perfect'. All the source files for the AI in the book are still online and available for download, and several spreadsheets are even available to see some of the math behind the AI in graph form. A very enjoyable read, and easy to understand.

Using classical game theory, Behavioral Mathematics for Game AI offers game programmers the logic and applications of math, offering modeling techniques A1 designers and programmers can use to replicate human decisions in a game structure. Chapters offer ways of using statistics, formulas and more to create believable scenarios and dynamic video games. Any computer library catering to advanced AI programmers needs this.

A lot what I think of a book comes from my expectations. The style, the content and the useful information that it provides me all factor into my opinion.

For starters, I do programming for a living. After 10 years of .Net Web sites and databases, I'm looking to expand my skill set. Not games, necessarily, but AI for sure. I tried a couple math books on Game theory, but they were still a few levels away from practical interpretation. This book brought it down to a level I could use.

In a nutshell, it's about how to program dice for an RPG. The scope is a little broader than that, but this is the main idea. In nontechnical language, the author discusses what the goals of AI are, how rational decisions are made, and how to use probability to generate irrational decisions to model an unpredictable world.

Math wise algebra would be a good idea, and maybe some basic understanding of calculus and statistics, but he gives sufficient background so that your understanding would not be hindered. The book doesn't really teach programming per se, so it really doesn't matter what language you use (although the examples are done in C).Specific algorithms, such was swarming and flocking, are also absent. The material here would be used after you have those ideas down...for example, you have your agents flocking, so now how does the flock decide what to do, and when?

This brings me to the one real draw back to the book. I could tolerate his jokes and stories about his kids, but I suspect some UMLs would have done wonders to make the material clearer. Use cases were almost there in his examples, but it just missed the mark. Maybe this was an attempt to keep the thinking less technical and open it to a wider market, but it is something he should have touched on.

I really didn't learn anything new, but rather, how to take a bunch concepts I was already familiar with and bring them together into a cohesive unit. It's got me thinking and planning and trying to figure out how to put it in to use. This is good. This is what the writer wanted.

So, I would recommend this book if you're at the point where you have your AI agents working, but you want them to be more flexible and unpredictable in their actions, or react with more independently to their environment. IF you check the price of this book on the secondary market, you can see its doing pretty well as a used book, which says a lot for anything in the technology field.

I'm not a game programmer, per se, but a programmer of several years and part of my job involves designing computer simulations of events. This book very clearly presents the concepts of non-player character AI to such an extent, that afterwards you'll see a little math in everyday personal communications. I happen to be in a course on Risk Analysis, "how to make decisions", and the overlap with this material surprised me (and my professor). There are many behaviors we're all involved with intuitively in life, an explicit understanding of which would help us specify software-controlled agents. If that is an area you'd like to dig deeper into, I whole-heartedly recommend this book.

I have been working in the game industry for almost 2 year now and I feel that this book is destined to become one the "must-have" books for AI developers. It has certainly made my list. Dave not only goes into how to create interesting dynamic AI behavior, but also why you should do so and he does it in a way that keeps you engaged.

This book does not talk about AI staples such as state machines or behavior trees. Instead, he goes into detail about how to make an AI make decisions based upon its surroundings and situation and do it in a way that mimics what a human player might do. This is in stark contrast to many AI systems that simply create behaviors based on what a player *should* do. There is a huge difference, and Dave covers this subject well.

Highly recommended for anyone interested in game AI or interested in improving the behavior of their current AI system.